Data & Code | Krisna 'imed' Gupta

Reviewing the National Economic Council Paper on the Economic Impact of Free Nutritious Meals

Mon, 24 Mar 2025 21:06:59 +0700

It was quite surprising yet interesting when I watched the by the National Economic Council (DEN) some time ago. The reason was that Luhut Binsar Panjaitan (LBP), the DEN chair, gave the spotlight to Professor Arif Anshory Yusuf (AAY) to present a teaser of his ex-ante model results on the impact of the Free Nutritious Meals program (MBG) on the economy. According to AAY, MBG has the potential to create 1.9 million jobs and reduce poverty by up to 5.8%. MBG, he said, is equivalent to giving IDR 600,000 per family if the average number of children in poor families is 3. This figure is 3x larger than PKH! He also added that to realize the maximum impact of MBG, “leakage” of food demand “that does not need to be imported” should be minimized by prioritizing domestic products. Those numbers sound off.

Then on the evening of March 23, 2025 Jakarta time, I watched Professor Arief Anshory Yusuf (AAY) present his latest paper written for the National Economic Council (DEN). The event was organized by Doctrine UK, an Indonesian student association in the United Kingdom. There, he presented a paper on the macroeconomic impact analysis of the Free Nutritious Meals program (MBG). The paper itself was shared in the event chat. The recording has been uploaded to YouTube and can be viewed .

After reviewing both the paper and the presentation, I felt compelled to write some notes about it. The main purpose is actually for myself, since this model is quite fun to explore and replicate. Additionally, at the event, AAY also represented DEN, which means the policies emerging from this paper can be considered as coming from the government and will likely become public policy. I’m also writing this for friends who aren’t very familiar with CGE¹ but want to learn. Lastly, it would be great if this post were read by AAY and DEN, but I’m sure someone of his caliber already receives plenty of input from people far more capable than me, so it’s fine if this goal isn’t achieved. Mostly, this post is for learning and sharing.

I greatly respect him as an academic because AAY is one of the few Indonesian economists who is very vocal about economic issues in Indonesia, especially on Higher Education and pro-poor and pro-job programs. He does not hesitate to criticize the government. Even his paper, which serves as the ex-ante justification for MBG, was shared with everyone, and he regularly presents and seeks opinions from many people across various circles. This level of transparency is something I haven’t found in other government policies.

Having said that, I have some notes that I believe are very important and worth writing about on this blog. I will focus on 2 caveats that he mentioned in his presentation. For other parts, that can be an opportunity for other friends who want to contribute comments. For example, commentary is also very interesting in my opinion, and easier to read than this post haha.

I will start by briefly summarizing the points I agree with, then move on to things I’m less sure about or disagree with.

Key points in my view

In his paper, AAY summarizes various studies about programs similar to MBG in other countries. In other countries, according to him, MBG has been quite successful. He also says that MBG is a very progressive program. I think it’s hard to disagree that MBG, as a concept, is a progressive program.

The paper also explains some details about MBG that are consistent with explanations from other sources (e.g., CISDI). He mentions a budget of 171 trillion rupiah, one of the largest welfare programs in the world, targeting approximately 82.9 million people.

Equally important: AAY mentions several official objectives of MBG: better nutrition, job creation, supporting local food systems, and improving student attendance and health. Nevertheless, he says that MBG should not be designed to pursue too many objectives at once, and I agree.

Finally, AAY uses input-output (IO) analysis and microsimulation to show the economic impacts in terms of increased output, value added, and employment, as well as poverty reduction from this program. Various simulations with different scenarios were conducted, but the results are consistently positive overall.

Important caveats

However, in his presentation, AAY stated two caveats that are very heroic and automatically dismiss much of the discussion about MBG:

This model is an ex-ante scenario assuming ideal implementation. All criticisms regarding operational aspects cannot be answered with this paper.
There is no general equilibrium aspect (supply, demand, market, and optimization, at least), so things like price changes and international trade are not reflected in this model.

In my personal opinion, these two things are arguably the most influential factors that could potentially derail this program. At the Doctrine UK event the other day, perhaps the majority of questions and discussions from participants were precisely about best practices and institutional readiness. I add some additional thoughts here beyond what was already raised by friends at the event, which can be read directly at the .

But this post will hyperfocus on point 2, something more relevant to AAY’s paper. I was quite surprised to hear AAY argue that IO analysis $\neq$ Computable General Equilibrium (CGE), and that IO does not account for price changes. In my opinion, that statement is not technically correct. I will also provide 3 arguments for why the results from the IO scenario may not be suitable for making claims as confident as what AAY showed at the DEN press conference above.

I will nerd out a bit about CGE to address caveat no. 2. This section will be too geeky. Read at your own risk. You’ve been warned.

The General Equilibrium Aspect

What is Input-Output Analysis (IO)?

In his paper, AAY explains IO in section 3.1.1., which is on page 4 of the paper.

There are 3 important terms AAY uses when explaining IO analysis: output, value added, and employment. Output is essentially the total production of an industry. Value added is total production minus inputs/raw materials. So value added consists of factors of production, mainly labor and capital. And labor equals employment.

The IO structure in AAY’s paper is exactly the same as a typical IO model, which is more or less like this:

$$ \begin{align*} x&=Ax+y \newline (I-A)x&=y \newline x&=(I-A)^{-1}y \end{align*} $$

where the resulting vector x represents input demand for sector $i$. That is, input demand from vector $i$ depends on the shift-share $A$ of each industry in $x$ for each industry in $x$, and final demand $y$.

In an Input-Output table, we can show output and value added if you open an actual IO table from BPS.

wew	s1	s2 … sj	y	sum
s1	$q_{1,1}$	$q_{1,2} \dots q_{1,j}$	$y_1$	$x_1$
s2	$q_{2,1}$	$q_{2,2} \dots q_{2,j}$	$y_2$	$x_2$
$\vdots$	$\ddots$	$\ddots$	$\vdots$	$\vdots$
si	$q_{i,1}$	$q_{i,2} \dots q_{i,j}$	$y_i$
va	$va_1$	$va_2 \dots va_j$	-	GDP
sum	$x_1$	$ x_2 \dots x_j$	GDP	-

Here, $q_{i,j}$ can be formed into a matrix equivalent to $Ax$ by setting $q_{i,j}=a_{i,j}x_i$. Hopefully that’s clear.

Usually, matrix $Ax$ is set through some optimization (you know, lagrange and stuff). Well, typically, matrix A in IO analysis is exogenous. In other words, the elements of matrix A are fixed and only expand along with output $x$. Therefore, all exogenous changes in the economy (in this case in $y$) are assumed to only expand according to matrix A.

Actually, IO is also CGE. It’s just that IO is a special case of CGE, where the output demand function has a perfect complement property. Perfect complement is a very useful assumption because the optimization becomes linear. Computationally, it becomes much easier since we only need linear algebra without any nonlinear optimization.

Additionally, Leontief is convenient because the optimization only expands according to the optimal share of each component.

For example, an input function $U=min(a_1x_1,a_2x_2)$ will have an optimal point at $a_1x_1=a_2x_2$ so the demand function for $x_1$ will be $x_1=\frac{a_2}{a_1}x_2$ and $\frac{a_2}{a_1}$ will be constant and fixed throughout changes in the consumption composite.

This method is relatively straightforward to implement in Python, as long as the IO matrix is available. I tried to replicate using a lite version of the IO table, . Of all the scenarios, I will try to replicate scenario A1 in the paper. From the 17-sector IO table, the lite version, I converted it into matrix A, x and y, and the data for the matrices and vectors can be downloaded . Note that the y matrix there still contains import values. That’s why in my simulation I construct y myself, not using y from there, so the imports are removed. Actually, you can also subtract the imports from final demand in the Excel file yourself.

import numpy as np
A=np.loadtxt('A.csv',delimiter=',') # matrix A
#y=np.loadtxt('y.csv',delimiter=',') ini final demand yang ada final importnya
u=np.loadtxt('x.csv',delimiter=',') # output
y=u-np.dot(A,u) # final demand tanpa final import
ia=np.identity(17)-A
invA=np.linalg.inv(ia)
np.savetxt('invA.csv', invA,delimiter=",") # save the change matrix result to a new file
x=np.dot(invA,y) ## Make sure replikasi final output di tabel.
x-u

array([ 0.00000000e+00, 4.76837158e-07, 9.53674316e-07, 1.19209290e-07,
-1.49011612e-08, -4.76837158e-07, 4.76837158e-07, 0.00000000e+00,
2.38418579e-07, 0.00000000e+00, 0.00000000e+00, -1.19209290e-07,
0.00000000e+00, -1.19209290e-07, 1.19209290e-07, -1.19209290e-07,
-1.19209290e-07])

## Simulasi MBG A1 di paper AAY
yy=y+np.array([0,0,0,0,0,0,0,0,171e6,0,0,0,0,-171e6,0,0,0]) ## + MBG shock - budget shock
xx=np.dot(invA,yy)
print(f'perubahan %output sektoral setelah +MBG-penghematan:\n {(xx-x)/x*100}') ## Percent change of sectoral output

perubahan %output sektoral setelah +MBG-penghematan:
[ 1.08864224 0.03824485 0.57615184 -0.79322713 -0.71996072
-0.27533884 0.34350708 -0.80514386 13.29429074 -1.13615204
0.0553673 -0.13483742 -1.60608706 -22.73385632 -0.14187647
-0.18583608 -0.23295487]

print(f'pertumbuhan output nasional adalah {(np.sum(xx)-np.sum(x))/np.sum(x)*100} persen') ## Percent change of total output

pertumbuhan output nasional adalah 0.046445748306372714 persen

## Put everything into a dataframe
import pandas as pd
sec=["Pertanian, Kehutanan, dan Perikanan", "Pertambangan dan Penggalian", "Industri Pengolahan", "Pengadaan Listrik dan Gas", "Pengadaan Air, Pengelolaan Sampah, Limbah dan Daur Ulang", "Konstruksi", "Perdagangan Besar dan Eceran; Reparasi Mobil dan Sepeda Motor", "Transportasi dan Pergudangan", "Penyediaan Akomodasi dan Makan Minum", "Informasi dan Komunikasi", "Jasa Keuangan dan Asuransi", "Real Estate", "Jasa Perusahaan", "Administrasi Pemerintahan, Pertahanan dan Jaminan Sosial Wajib", "Jasa Pendidikan", "Jasa Kesehatan dan Kegiatan Sosial", "Jasa lainnya"]
delta=(xx-x)/x*100
## Create pandas of the result
df=pd.DataFrame({'sektor':sec,'delta':delta})
#concat the national results
df2=pd.DataFrame({'sektor':'Total','delta':(np.sum(xx)-np.sum(x))/np.sum(x)*100},index=[17])
df=pd.concat([df,df2])
df

	sektor	delta
0	Pertanian, Kehutanan, dan Perikanan	1.088642
1	Pertambangan dan Penggalian	0.038245
2	Industri Pengolahan	0.576152
3	Pengadaan Listrik dan Gas	-0.793227
4	Pengadaan Air, Pengelolaan Sampah, Limbah dan ...	-0.719961
5	Konstruksi	-0.275339
6	Perdagangan Besar dan Eceran; Reparasi Mobil d...	0.343507
7	Transportasi dan Pergudangan	-0.805144
8	Penyediaan Akomodasi dan Makan Minum	13.294291
9	Informasi dan Komunikasi	-1.136152
10	Jasa Keuangan dan Asuransi	0.055367
11	Real Estate	-0.134837
12	Jasa Perusahaan	-1.606087
13	Administrasi Pemerintahan, Pertahanan dan Jami...	-22.733856
14	Jasa Pendidikan	-0.141876
15	Jasa Kesehatan dan Kegiatan Sosial	-0.185836
16	Jasa lainnya	-0.232955
17	Total	0.046446

indicator	original	replica
National output	0.06	0.046
Agriculture	1.37	1.08
manufacture	0.12	0.57

The A1 simulation results: the original is from AAY, the replica is mine. There are differences in the numbers, likely due to different tables and certainly different aggregation levels. But they’re relatively close, so I think it’s fairly accurate haha.

Three limitations of this model are immediately apparent.

First is the limitation of the constant shift-share assumption a la Leontief. This is immediately visible from the source of growth. The balanced budget reallocation produces positive additional output because the food service sector has a larger “total multiplier” (in the paper, terms $m_{ij}$ and $m_{ik}$, pp.6) than the government services sector. The problem is that this total multiplier is constant regardless of how much is added or subtracted. Without changes in marginal return, if the goal is output growth, we might as well move ALL output from government services to the food service sector (i.e., corner solution). But in the real world this doesn’t happen, because as a sector loses output, marginal productivity there rises, while a sector that is already bloated will experience diminishing returns, eliminating gains as more resources flow into it. Plus, on the demand side, there is also marginal utility that changes as consumption patterns concentrate on a single service. This is not a problem if the shock is relatively small. But given that the MBG shock raises the food service sector by 13.3% and cuts the government service sector by -22.73%, the constant marginal return assumption is likely to be problematic.

Second, the vector $y$ is treated as final demand without considering exports and imports. Indeed, the $y$ column in the BPS IO table originally contains capital formation and exports, which are aggregated into $y$ here. Meanwhile, in the first column there should be an imported raw materials row. In CGE, we often add a substitution function with Armington elasticity between imported and domestic inputs for each sector. Naturally, the absence of international trade (or more precisely, treating the shift-share of international trade as exogenous/constant) makes it difficult to forecast MBG’s impact on exports and imports. That is why AAY said at something like MBG will have a good impact as long as there is no “leakage,” that “materials that don’t need to be imported” should not be imported. I think this is not the best way to conclude from the model.

More importantly, in AAY’s paper, I could not find information about which table he used for the analysis. This is important because BPS has 2 types of IO tables: and . In the domestic table, only domestic input figures appear in the $Ax$ matrix, and all imports are pooled in row 2000. In the total table, the $Ax$ matrix already includes imports, so row 2000 is empty. There is no information in the paper about whether Armington calibration was performed for domestic vs. imported product choices.

If the domestic transaction table was used, then the $Ax$ matrix does not reflect total inputs (tends to be biased), implicitly treating imports as fixed, hence the “don’t leak” conclusion, which may be consistent with what the model says. But if the total transaction table was used, then the $Ax$ matrix already includes imported goods, and it may well be that “leakage” occurs in the model with no negative consequences whatsoever. Or maybe I’m the one misreading it and the procedure for removing imported input is actually explained somewhere in the paper. But it should be made more explicit.

However, the critique about the exogenous trade balance is actually very much related to the third critique: the absence of prices. The absence of prices becomes super relevant when we use a model with optimization and constraints. Without constraints, production in the MBG-recipient sector can seemingly expand as far as $Ax$. In reality, it is the supply-side constraint that would cause inflation. When there is a (budget) constraint, optimization must be performed, and consequently income effect and substitution effect will emerge (if using assumptions other than Leontief). In particular, this income effect will greatly affect the need for exports/imports, but more importantly, total consumption and poverty calculations.

Let me explain.

Constraint and prices

Under the Leontief assumption, when we shock final demand (in column $y_i$ where $i=$ the shocked sector), output from the shocked sector increases, and input demand from that sector increases, visible in column $s_j$ where $j=$ the sector experiencing the change in $y$. Like $\frac{dy}{dx}$ sort of thing.

For example, if demand in sector $k$ increases to $y_k+1$ so we have a vector $y_{new}$ shaped like this:

$$ y_{new}=y+ \begin{bmatrix} 0 \newline \vdots \newline 1 \newline \vdots \newline 0 \end{bmatrix} $$

let us call that new vector $d$ so that $y_{new}=y+d$. Then we compute $x_{new}=(I-A)^{-1}y_{new}$.

$$ \begin{align*} x_{new}&=(I-A)^{-1}(y+d) \newline x_{new}&=(I-A)^{-1}y+(I-A)^{-1}d \newline x_{new}&=x+(I-A)^{-1}d \end{align*} $$

The change in input demand from sector $k$ is exactly $(I-A)^{-1}$. AAY explains this on page 6 of his paper. Now, that term contains the identity matrix and matrix $A$, which we have made exogenous. This means the change in input demand from that sector depends only on $\Delta y$ and does not depend on price changes at all.

The BPS 2020 IO table is essentially a transaction matrix. Transactions imply current prices, where there is both quantity and price. Now, a simple IO exercise usually treats those matrix values as quantities. So if an element of the matrix (which is actually a value) moves, the movement is treated as a quantity movement.

The problem is that, with constraints, we can show that input demand usually depends on prices!

Consider typical Leontief production:

$$ \begin{align*} \max \min(a_1q_1,a_2q_2) \newline \text{subject to} \ p_1 q_1+p_2 q_2 \leq I \end{align*} $$

Optimization implies $a_1q_1=a_2q_2$, so $q_1=\frac{a_2}{a_1}q_2$. Substituting into the constraint:

$$ \begin{align*} p_1 \frac{a_2}{a_1}q_2+p_2 q_2 = I \newline q_2=\frac{I}{\frac{a_2}{a_2}p_2+\frac{a_2}{a_i}p_1} \end{align*} $$

If we replace income $I$ with the price of output 1 times the quantity of output 1, $p_1x_1$, then demand for output 2 becomes:

$$ q_2=\frac{p_1x_1}{\frac{a_2}{a_1}p_1+\frac{a_2}{a_2}p_2} $$

and we can generalize the equation above for $\max \min(a_1q_1,a_2q_2,\dots,a_Jq_J)$ to:

$$ q^j_i=\frac{p_ix_i}{\sum_{s=1}^S \frac{a^s_i}{a^j_i}p_s} $$

where $q^j_i$ is the material input from industry $j$ to industry $i$, $p_i$ is the output price of industry $i$, $x_i$ is the output of industry $i$. $q^j_i$ is a function of the isocost of industry $i$ (which equals total revenue) and input prices for each industry $s$ including $s=j$.

In the replication , $q^j_i$ is proxied by $a_{i,j}x_i$, where this term only changes because of $x_i$. Yet, as shown above, prices also matter. If $p=1$, then $q$ simply has relative shift share $a$ and final demand $x$, which is exactly what happens in the replication above.

Now, one can argue that price-dependent input demand is not a problem as long as we assume the sector is a price taker. That is, the demanding sector is so small that the market will always provide supply (i.e., there are infinitely many suppliers) without affecting price increases. But this is certainly not always realistic, especially since the sector in question has high demand for agricultural products, a sector that has historically always been plagued by bureaucratic trade governance, miscalculations, perpetually failed self-sufficiency programs, and so on. It seems difficult to expect super elastic supply from this sector.

Of course, we could also model the limitations from the supply composite labor side, where we could model a production nest. Indeed, in the IO table there is actually a separate labor share, which is also assumed to have a constant parameter $e$, just like value added which has a constant parameter $v$. Both assumptions can be relaxed.

But perhaps another time.

Implementation problems

I also intend to add some thoughts on the importance of considering implementation problems that will have a very significant impact on the conclusions about MBG’s impact.

I agree that if ALL schools receive this program, then it could be progressive, assuming the budget sacrificed for MBG comes from a line item that is usually enjoyed by the (hence lower marginal propensity to consume²). But there are several implementation possibilities suggesting this may not happen on the ground.

First, regarding the distribution of recipient schools. As I understand it, not all schools receive this program simultaneously. Some schools get MBG (treatment) and others do not (control). The problem arises if treatment does not occur randomly/exogenously. For example, if in general the schools that receive it first are those that tend to be more prepared in terms of infrastructure. Such schools are typically located in relatively wealthier areas. If this is the case, although AAY provides some anecdotes about poor community groups as beneficiaries, many participants may actually be in areas that are already fairly well-off.

Second is corruption. If this program is corrupted, the corrupt actors are most likely much wealthier than the beneficiaries. This means MBG would be dangerous if it becomes a vehicle for income transfer from the party whose budget was cut to the party that corrupts MBG. This is certainly not pro-poor. Indications of corruption (or at least failure to meet good nutritional standards) were already raised by participants at the Doctrine UK event the other day.

Lastly, the CGE scenario above may also be less effective if it turns out that not all of the 171T is used for food portions. Some of the money could also be used for operations and institution-building. Someone mentioned that the Ministry of Health already has institutions ready for use, so there’s no need to create new institutions like BGN and communal kitchens. A waste of resources.

Conclusion

In closing, I have tried to offer constructive criticism in this post. I focused on the IO model and tried to show that 3 assumptions in the IO model (constant shift-share, absence of international trade, and fixed nominal prices) could be violated. Of course, no model will be exactly the same as the real world. All models are wrong but that’s fine as long as the model is useful for explaining phenomena. It only becomes a problem when the assumptions that don’t hold have very important consequences for the conclusions drawn as a basis for policy.

I also want to once again express my appreciation for AAY for providing a level of transparency that is quite rare in government circles. It is very rare for anyone to document the studies used by the government as a policy guide, so that civil society like me can participate in the debate. As someone still junior, my input may not be that important, but it’s nice to be able to follow the thinking of Indonesia’s policy architects and learn from it. Hopefully this kind of openness will spread to all government sectors.

I’ll stop here for now. Maybe next time we can try modeling that production nest if I have some spare time. It’s gonna be fun! It’s been a long time since I’ve tinkered with CGE. Alright, see ya next post.

Honestly, I’m not that great at CGE either, so please take this post with a grain of salt. ↩︎
BTW, MPC <1 is what makes in-kind assistance like MBG potentially more effective at boosting consumption than cash transfers. Worth a separate discussion. ↩︎

Probability Theory

Tue, 14 Jan 2025 08:00:00 +0700

Goodbye dollar? Not quite yet

Thu, 25 May 2023 06:35:25 +1100

The topic of de-dollarization has been a hot-button issue lately. De-dollarization – the effort by many countries to reduce dependence on the US dollar (USD) – is reportedly being championed especially by BRICS nations (Brazil, Russia, India, China, and South Africa).

Indonesia is no exception. Bank Indonesia has established Local Currency Settlement (LCS) agreements with South Korea, Japan, China, Malaysia, and Thailand. In other words, Indonesia and its partner countries can now use their respective domestic currencies to settle cross-border payments without using the US dollar.

According to this article, LCS is important for minimizing exchange rate risk in bilateral transactions. That may be true if the currency usage focuses solely on trade in goods and services. But these days, currencies are also used for asset transactions. In fact, the use of currencies for asset trading may well dominate their use for goods and services.

Currency as a medium for asset exchange?

To answer this, some technical knowledge of the balance of payments (BoP) is needed. The BoP consists of two main accounts: the current account and the financial account.

In an open economy, we have:

$$ Y=C+I+G+(X-M) $$

Meanwhile, saving is income that is neither consumed nor taxed, so S=Y-C-G (assuming total taxes = G). Therefore:

$$ Y-C-G-I=X-M $$

Replacing Y-C-G with S:

$$ S-I=X-M $$

S-I is the financial account, while X-M is the current account. One balances the other.

Put simply, a current account surplus means you sell goods but get paid with IOUs. Or more precisely, you accumulate savings that you don’t spend right away.

Back to the currency question: for countries with trade surpluses, it means they store the proceeds from their exports in the form of foreign assets or investments. This is where a problem emerges: why would we hold another country’s currency?

Holding foreign currency may be useful if you plan to import from that country. But this logic breaks down when the countries in these partnerships are NEARLY ALL IN SURPLUS! See for yourself:

Above I show the trade balance deficit (as % of GDP) for BRICS, Indonesia, Japan, and South Korea. As you can see, nearly all of them run surpluses. In other words, they generally supply more goods to the world than they buy, and the proceeds are stored as assets rather than spent.

Undervaluation

This may be because these countries largely follow a mercantilist philosophy – an economic doctrine that prioritizes export promotion and dislikes imports to boost industrial competitiveness. Export-oriented countries typically prefer a weak currency, because a weak currency makes exports cheap and imports expensive, pushing X-M into surplus.

For these countries, currency intervention is fully justified to keep their currencies weak, even though they should arguably be valued higher. We can check for this by looking at undervaluation – comparing the nominal exchange rate with the “true” value. One approach is to use the Big Mac Index from The Economist.

#import wbdata as wb

import datetime

## Read data from the economist

url='https://raw.githubusercontent.com/TheEconomist/big-mac-data/master/output-data/big-mac-raw-index.csv'

bmi=pd.read_csv(url,parse_dates=['date'])

## Create the 'real' exchange rate measures

bmi['bmi']=bmi['dollar_ex']+bmi['dollar_ex']*bmi['USD']

bmi.iso_a3.unique() # Showing unique iso_a3 codes for the next cell

array(['ARG', 'AUS', 'BRA', 'GBR', 'CAN', 'CHL', 'CHN', 'CZE', 'DNK',
'EUZ', 'HKG', 'HUN', 'IDN', 'ISR', 'JPN', 'MYS', 'MEX', 'NZL',
'POL', 'RUS', 'SGP', 'ZAF', 'KOR', 'SWE', 'CHE', 'TWN', 'THA',
'USA', 'PHL', 'NOR', 'PER', 'TUR', 'VEN', 'EGY', 'COL', 'CRI',
'PAK', 'SAU', 'LKA', 'UKR', 'URY', 'ARE', 'IND', 'VNM', 'AZE',
'BHR', 'HRV', 'GTM', 'HND', 'JOR', 'KWT', 'LBN', 'MDA', 'NIC',
'OMN', 'QAT', 'ROU'], dtype=object)

b=[]

ctr=('CHN','IDN','IND','ZAF','RUS','BRA','JPN','KOR')

for i in ctr:

 bmii=bmi.loc[(bmi['iso_a3'] == i)]

 b.append(bmii)

b=pd.concat(b)

sns.lineplot(data=b,x='date',y='USD',hue='name',palette='tab10')

plt.axhline(0,color='grey')

<matplotlib.lines.Line2D at 0x2b17256fa48>

As we can see, except perhaps for Brazil, BRICS countries, Indonesia, Japan, and South Korea all have currencies that are undervalued relative to the USD. In other words, there are indications that these countries deliberately suppress their currencies to keep their products “competitive” in global markets.

Using the Big Mac Index isn’t exactly rigorous, but if you’re interested in diving deeper into currency misalignment, check out my colleague Wishnu Mahraddika’s publication here.

In other words, these countries appear to actively pursue trade surpluses! But not everyone can be in surplus at the same time, can they? This becomes even more stark when we compare trade balances in USD terms:

Who can absorb China’s surplus??

Uncle Sam to the rescue

So far, the surpluses of these net-exporter countries have been absorbed by none other than the United States.

As you can see, it is the US that has been absorbing these countries’ surpluses. How does the US government manage this? Through debt, of course! Everyone wants to sell to the US because it has deep pockets, and because everyone sells to the US, everyone holds US dollars, which makes the dollar the world’s primary reserve currency – a self-fulfilling network effect!

On top of that, the US has a relatively open financial account. It’s very easy for anyone to trade assets in the US. You can easily buy stocks and other financial products there. Capital controls are lax, so everyone feels safe parking their money in the US. The US is also where many creative and promising startups emerge.

Does de-dollarization reduce exchange rate risk?

The problem is that the countries in these partnerships (including Malaysia and Thailand) are almost all surplus countries. They maintain their surpluses by exchanging goods for foreign assets. The most popular foreign assets? Those of the country with the largest deficit in the world – the United States. Consequently, even if bilateral trade is settled in local currencies, those currencies end up being exchanged for dollars anyway to purchase US assets. Or perhaps bilateral trade is only a fraction of total trade with the US.

Again, as long as the US remains the world’s number one buyer of goods and services, it will be hard to avoid holding US dollars. And since we’ll ultimately transact with the US (whether buying assets or selling goods), the exchange rate relative to the USD will remain relevant.

That said, de-dollarization pressures may actually be coming more from the US side itself! US sanctions against Russia have made the entire world rethink the safety of USD-denominated assets. Additionally, the possibility of a US government bond default due to the debt ceiling may also erode global confidence in US assets.

And beyond exchange rates, the changing global order going forward could have even bigger implications for our economy. If BRICS and many countries do push through with de-dollarization, the giant surplus countries will need to rethink their domestic industries because they won’t be able to export like before (since the main buyer for many countries is the US). The impact could be far greater than “mere” exchange rate effects.

That said, I think LCS options like these are still attractive. The more payment systems available, the better, and efforts like LCS deserve our appreciation.

That’s it for this post – hope it’s useful.

Recent FDI developments: Indonesia and the region

Sat, 07 Jan 2023 18:29:28 +0700

Indonesia was surprised on a holiday by the issuance of the as a replacement for the law currently being debated by the Constitutional Court (MK). This appears to have been driven by the government’s concerns about . The argument is that the Job Creation Law being contested at the Constitutional Court creates uncertainty for investors. This is compounded by the approaching election year, which makes the future of the law – currently in the government’s court – even more uncertain.

The government’s concerns seem well-founded. The global economy amid raging inflation in developed countries is not the best time for investment. Many investors are between staying in China or moving elsewhere, and those looking to diversify . All of this amid our pride in growing foreign investment in nickel smelters that were painstakingly attracted with an export ban that , backing out of its investment promise in the new capital, and of course , whose commitment has yet to materialize.

Looking at data, Indonesia appears to have attracted fairly high FDI. As shown in Figure 1, Indonesia’s FDI is second only to India.

However, we need to account for the size of the economy. While Indonesia’s FDI is fairly high, we should look at FDI per worker. After all, the capital built through investment will be used by workers. Think of it this way: two companies might each buy 10 computers, but that investment is more valuable in the company with 10 employees than the one with 20. In the first company, each employee gets a computer; in the second, employees have to share. That’s if they even get a computer – some might have to bring their own laptop (this is a personal gripe; please ignore).

On a per-capita basis, Indonesia clearly needs more FDI. By the way, Figure 1 doesn’t include China because adding China would make all other countries look tiny in comparison.

Looking at the trend in both figures, FDI growth in Indonesia hasn’t outpaced neighboring countries. Even the impact of the Job Creation Law isn’t clearly visible in the trend.

Lastly, as a country with a large domestic economy, it’s quite natural for Indonesia to depend more on domestic investment than foreign investment. shows that foreign investment is indeed relatively small compared to domestic investment.

Table 1. Domestic and foreign investment realization in million USD

Investment	2016	2017	2018	2019	2020	2021
Domestic	216,230.8	262,350.5	328,604.9	386,498.4	413,535.5	447,063.6
Foreign	28,964.1	32,239.8	29,307.9	28,208.8	28,666.3	31,093.1

On the other hand, this may also indicate that foreign investors haven’t been very interested in Indonesia from the start. While domestic investment keeps growing, FDI has stagnated. Can the Perppu change this? Let’s see!

The code to reproduce Figures 1 and 2 is below.

library(WDI)
library(tidyverse)

indi<-c(
 "fdi"="BX.KLT.DINV.CD.WD",
 "lab"="SL.TLF.TOTL.IN",
 "pop"="SP.POP.TOTL"
)

ctr<-c('BGD','CHN','IDN','MYS','IND','THA','VNM')

dat<-as_tibble(WDI(
 indicator = indi,
 country = ctr,
 start=2017
 ))
dat$weks<-dat$fdi/dat$pop
dat$wuks<-dat$fdi/dat$lab

dat |>
 subset(country!="China") |>
 ggplot(aes(x=year,y=fdi,color=country))+geom_line(linewidth=1.1)+theme_classic()+
 labs(title = "Total FDI, BoP, current USD (minus China)", y="Total FDI, current USD")+
 scale_y_continuous(labels = scales::label_number_si())
ggsave("pic1.svg")
ggplot(data=dat,aes(x=year,y=weks,color=country))+geom_line(linewidth=1.1)+theme_classic()+
 labs(title = "Total FDI per capita, current USD (with China)", y="FDI per capita, current USD")+
 scale_y_continuous(labels = scales::label_number_si())
ggsave("pic2.svg")
ggplot(data=dat,aes(x=year,y=wuks,color=country))+geom_line(linewidth=1.1)+theme_classic()+
 labs(title = "Total FDI per worker, current USD (with China)", y="FDI per worker, current USD")+
 scale_y_continuous(labels = scales::label_number_si())
ggsave("pic3.svg")

A Simple Model of Local Content Requirements

Tue, 29 Mar 2022 14:09:58 +0700

Discussions about increasing the use of domestic products and anti-import sentiments have long echoed among public officials. Lately, these conversations seem to be growing ever more frequent. Of course, using domestically produced goods is not inherently a bad thing. The problem arises when it is forced. Like any other kind of love, love for domestic products is no fun when it is coerced. One of the instruments Indonesia is currently pushing hard is TKDN.

What is TKDN?

, or Tingkat Kandungan Dalam Negeri (Local Content Requirement), has become increasingly popular in Indonesia¹. When mandated for a given industry, TKDN requires that a minimum share of a final product contains domestic content. This content can include labour, machinery, and raw materials. For example, , as of April 2022, must meet a 35% TKDN threshold, meaning all 5G devices on the market must have at least 35% local content (or, equivalently, at most 65% imported components).

TKDN effectively prevents firms from simply selling imported goods. Products must be manufactured domestically. Imported components are allowed but capped. The idea is that TKDN will encourage producers to set up factories in Indonesia. I wrote a bit about this effect in . The resulting investment would create jobs. In a country with high informality like Indonesia, formal employment can be a good avenue for improving welfare². TKDN is also expected to build production capacity and technological capability among domestic suppliers of intermediate goods.

TKDN is popular with governments partly because it is “free” (unlike subsidies, it does not require budget outlays). Moreover, anti-import policies tend to be popular with the public, potentially boosting politicians’ approval.

At first glance, raising TKDN does seem beneficial. If investment flows in, there should be additional value added. However, it is important to understand that the concept of value is neither static nor exogenous. The value of a product can rise or fall depending on its inputs. For instance, does a 5G device with high domestic content perform as well as a fully imported one? Can a high-TKDN product compete in global markets? Does building a smelter really capture more value ?

The answer is not necessarily yes. Several theoretical papers dissect how value and productivity shift between intermediate-input producers and final-good producers³ ⁴ ⁵. There is indeed a trade-off, and TKDN does not automatically produce outcomes as optimal as they may appear, especially if TKDN fails to raise productivity in the sector it aims to protect.

In this post, I try to illustrate this trade-off using a very simple model.

The Model

Suppose there is a final-good producer that often needs imported inputs. These inputs could in principle be substituted domestically, but for the firm, imports are cheaper and offer higher value added. Real-world examples might include or .

If we focus solely on the input-sourcing decision, the firm has the production function:

$$F(M,N) = \alpha M + N$$

where $M$ is the quantity of imported inputs and $N$ is the quantity of domestically sourced inputs. Meanwhile, $\alpha$ represents the quality of imported raw materials relative to domestic ones. If $\alpha > 1$, then imported goods are of higher quality. We also assume imported inputs are cheaper ($P_M

The firm is required to purchase a minimum share $k$ of domestic raw materials:

$$ P_NN \geq kP(\alpha M + N) $$

$$ \text{where } k \in (0,1) $$

The firm wants to meet demand $F(M,N)=Q$ at the lowest possible cost. That is, the firm solves:

$$ \min P_MM+P_NN $$

$$ s.t. \ \alpha M+ N \geq Q $$

$$ P_NN \geq kP(\alpha M + N) $$

$$ k \in (0,1) $$

Since the problem is linear, we can use linprog. I will use the following parameters:

$$ P_M=1 $$

$$ P_N=3 $$

$$ α=1.5 $$

$$ k=0 $$

$$ P=9 $$

$$ Q=30 $$

First, let us define the function:

from scipy.optimize import linprog
def TKDN(Pm=1,Pn=3,α=1.5,δ=1,k=0,P=9,Q=30):
 c_ex1 = np.array([Pm, Pn])
 A_ex1 = np.array([[-α,-1],[k*P*α, k*P-Pn]])
 b_ex1 = np.array([-δ*Q,0])
 res_ex1 = linprog(c_ex1, A_ub=A_ex1, b_ub=b_ex1, method='revised simplex')
 names=[f'imported inputs = {res_ex1["x"][0]}',
 f'domestic inputs = {res_ex1["x"][1]}',
 f'production cost = {Pm*res_ex1["x"][0]+Pn*res_ex1["x"][1]}',
 f'firm profit = {Q*P-Pm*res_ex1["x"][0]-Pn*res_ex1["x"][1]}']
 for i in names:
 print(i)

The default numbers above should not be taken literally – $P_M=1$ is simply a numeraire, or base price. Because the problem is linear (and therefore invariant to affine transformations), $P_N=3$ can be interpreted as meaning domestic raw materials are three times more expensive than imported ones.

Winners and Losers

Now let us use the function we just defined. Remember the linprog dependency from scipy. For the no-TKDN case, we set $k=0$:

TKDN()

imported inputs = 20.0
domestic inputs = 0.0
production cost = 20.0
firm profit = 250.0

What about a TKDN where domestic content must be at least 20% of total final-product value?

TKDN(k=0.2)

imported inputs = 8.0
domestic inputs = 18.0
production cost = 62.0
firm profit = 208.0

As we can see, TKDN causes the final-good producer’s profit to fall and total cost to rise. However, the domestic input supplier gains additional revenue of $18 \times P_N$. In other words, TKDN tends to benefit domestic raw material sellers but hurts final-good producers.

When TKDN is set to 40%, the model crashes – the final-good producer is better off not producing at all.

If we tabulate TKDN from 0 to 30% (in 10% increments):

k	M	N	C	$\pi$
0	20	0	20	250
10%	14	9	41	229
20%	8	18	62	208
30%	2	27	83	187

What about total domestic value added? Although TKDN raises value added for the input producer, it reduces value added for the final-good producer. With TKDN=0, the input producer’s value added is 0, while the final-good producer’s profit is 250. At TKDN=10%, the input supplier gains 9 but the final-good producer loses $250-229=21$. So total value added actually declines.

What about the government? Assuming duty-free imports, government revenue comes from corporate income tax. With TKDN=0, the input supplier pays no tax. The final-good producer pays corporate income tax of $t \pi$ where $t$ is the tax rate. If we assume the domestic input supplier’s cost is 0, then the supplier’s income is $P_N \times N$. Total government revenue becomes:

$$GR=t(P_N-c)N+t \pi$$

where c is the marginal cost, linear in output. I use c=2 so the domestic firm needs to charge $P_N=3$ to earn a margin of 1, the same as the importer. If domestic costs were low, they could sell at a lower price and win orders from the final-good producer.

Let us calculate government revenue at a 20% tax rate:

t=0.2 # corporate income tax
Pn=3 # domestic input price
c=2 # so firms can't sell below 3
brp=('0%','10%','20%','30%')
taxs=((Pn-c)*0*t+250*t,
(Pn-c)*9*t+229*t,
(Pn-c)*18*t+208*t,
(Pn-c)*27*t+187*t)
for x,y in zip(brp,taxs):
 print(f'At TKDN={x}, total GR = {y}')

At TKDN=0%, total GR = 50.0
At TKDN=10%, total GR = 47.6
At TKDN=20%, total GR = 45.2
At TKDN=30%, total GR = 42.8

The higher the TKDN, the lower the total income tax from the two firms. Although tax from the domestic firm rises, tax paid by the final-good producer falls because profits shrink. So even though TKDN appears not to require fiscal expenditure like subsidies, the inefficiency it generates carries a hidden cost to the state budget as well.

This is on top of the budget spent on enforcing TKDN rules. For instance, to verify that a firm complies with TKDN, it must be surveyed for a TKDN certificate. The first two product certificates are paid for by the government, funded from the state budget. This does not include the systems and human resources the government must allocate for TKDN enforcement. Moreover, for the third product onward, the firm must bear the certification cost itself, further reducing profits and consequently the corporate income tax paid.

So Is TKDN Good or Bad?

The issues above illustrate potential problems created by TKDN. This linear model, while somewhat arbitrary, does a reasonable job of explaining TKDN’s short-run effects. More representative (but more complex) models can be found in the footnotes.

One important caveat is that we cannot simply assume $Q$ and $P, P_M, P_N$ remain constant under TKDN. It is possible that instead of becoming more efficient, domestic input producers become complacent under protection and show no improvement at all. Rather than improving, they may even raise prices because the final-good producer has no alternative as long as TKDN binds.

Furthermore, TKDN makes domestic final-good producers less competitive against foreign ones. Why? Because foreign producers can access cheap, high-quality inputs without TKDN constraints. As a result, domestic final-good producers cannot compete in global markets and are forced to rely on the domestic market. Unable to compete with imported finished goods either, they will demand protection too.

This is why IMEI blocking exists. Domestic phone manufacturers do not want to compete with imported phones that are not subject to TKDN. Has the public ever been told how much of the state budget is spent on implementing IMEI blocking?

Finally, once TKDN is in place, it is very hard to roll back. Firms that have already made large investments to comply with TKDN would suffer significant losses if the policy were suddenly removed and firms that did not invest could enter the previously protected market. Therefore, firms that have already invested to meet TKDN have a strong incentive to lobby the government to maintain and enforce TKDN rules.

TKDN can be considered successful if domestic input producers actually learn and improve. For example, if their industry grows and becomes efficient enough to match import prices ($P_N \leq P_M$), or improves quality to surpass imports ($\alpha \in (0,1)$). If this happens, $k$ is no longer binding and TKDN can be declared a success, depending on how large and how fast the domestic industry’s improvement is.

Therefore, policymakers who have already adopted TKDN should continuously monitor $\alpha$ and $P_N$ (or more precisely $\frac{P_N}{P_M}$). If there is no progress, it is hard to expect domestic industry to become efficient and competitive, and the desired advantages will not materialise. Coercion alone is unlikely to succeed. The government may need to complement TKDN with other policies that promote efficiency. So yes, the regulatory burden just keeps growing.

Alternatives to TKDN

There are several alternatives to TKDN⁶:

Improve the business climate. This one seems obvious. But nothing beats this approach. It is hard for industry to advance when upstream sectors are protected, and both importing and exporting carry high administrative costs. Customs officials still engage in .
Leverage CSR. Ask multinational companies whether they can help develop local firms as suppliers.
Provide training for workers and component suppliers. Training is essential for improving the quality of local businesses and workers. A government that understands industry should be able to identify what training is needed and deliver it. This also costs money, but it is far better spent than on enforcing TKDN.
Improve logistics processes and physical infrastructure. This too seems obvious.
If all else fails, using tariffs and subsidies, though not ideal, is still better than TKDN. Of course, the main problem with tariffs is trade agreements and retaliation. Tariffs are easier to quantify than TKDN, making them more likely to face challenges at the WTO. Subsidies burden the budget directly, unlike TKDN which burdens the budget and the economy indirectly. Such is life when your policymakers and politicians cannot calculate hidden costs.

Update

A new book on TKDN has just been released for those who want to learn more. It is an edited volume by Lili Yan Ing and Grossman⁷. One particularly interesting chapter is by Yessi Vadila⁸.

I will discuss it further some other time. Mention me on if you have comments, criticisms, or suggestions. Thanks.

Fernando, O., & Ing, L. Y. (2022). Indonesia’s Local Content Requirements: An Assessment on Consistency with Free Trade Agreement Commitments. ERIA Discussion Paper Series, 420. ↩︎
Gupta, K., & Mardjono, E. (2021). Indonesia Emas Berkelanjutan 2045: Kumpulan Pemikiran Pelajar Indonesia Sedunia Seri 01 Ekonomi. LIPI Press. ↩︎
Grossman, G. M. (1981). The Theory of Domestic Content Protection and Content Preference. The Quarterly Journal of Economics, 96(4), 583-603. ↩︎
Veloso, F. M. (2006). Understanding Local Content Decisions: Economic Analysis and an Application to the Automotive Industry. Journal of Regional Science, 46(4), 747-772. ↩︎
Lin, S.-H., & Weng, Y. (2020). Can strengthening the local content requirements meet a government’s need to raise industrial productivity and production? Journal of Applied Economics, 23(1), 316-328. ↩︎
Hufbauer, G., Schott, J. J., Wada, E., Cimino, C., & Vieiroand, M. (2013). Local Content Requirements : A Global Problem. Peterson Institute for International Economics. ↩︎
Ing, L. Y., & Grossman, G. M. (2023). Local content requirements: promises and pitfalls. Routledge. ↩︎
Vadila, Y., & Christian, D. (2023). Effects of Local Content Requirements on Trade. In B. Hoekman & K. Saggi (Eds.), The Political Economy of Trade Policy: Theory, Evidence and Applications. Routledge. ↩︎

Trade in Value Added: Indonesia's Position in the Global Value Chain

Mon, 13 Dec 2021 09:54:16 +1100

Global Value Chain (GVC) is a buzzword that frequently comes up among academics and consultants when discussing international trade and manufacturing growth. The essence of GVC is splitting the production of a finished good into several links in a value chain, with each link produced in a different country. Take bicycles as an example: to assemble one bicycle, a manufacturer sources components from various countries.

Source: World Development Report 2020, World Bank.

The image above is from the World Development Report 2020, which I consider a fairly comprehensive read on GVC. I once illustrated (in a very simple way) in one of my posts how GVC helps produce gado-gado sauce that successfully enters export markets. I also wrote a short article on the role of GVC during COVID-19. The longer version is here.

GVC is often invoked by government officials, who argue that Indonesia’s participation in GVC can drive the growth of its manufacturing sector. GVC has been one of the key strategies for ASEAN and East Asian countries in producing competitive manufactured goods. Indonesia is frequently said to underutilise GVC compared to its regional peers. Is this true?

One indicator that can be used to assess Indonesia’s GVC participation is Trade in Value Added (TiVA). TiVA is an indicator based on trade data that estimates how much of a country’s final goods consists of domestic versus imported content. A country that relies heavily on GVC will have a high share of imported content in its export value. In other words, using the bicycle example, TiVA shows how large the total imports of gears, chains, tyres, and so on are relative to the value of bicycle exports. The TiVA data used in this post is sourced from here. As usual, I will use Python. Readers can replicate the data with this code.

tiva=pd.read_excel('tiva.xlsx',sheet_name='pake')
# source: https://stats.oecd.org/Index.aspx?DataSetCode=TIVA_2021_C1
tiva=pd.melt(tiva, id_vars=['year'])
tiva2=pd.read_excel('tiva.xlsx',sheet_name='pake2')
# source: https://stats.oecd.org/Index.aspx?DataSetCode=TIVA_2021_C1
tiva2=pd.melt(tiva2, id_vars=['year'])

tiva.variable.unique() # check countries

array(['OECD', 'AUS', 'AUT', 'BEL', 'CAN', 'CHL', 'COL', 'CRI', 'CZE',
'DNK', 'EST', 'FIN', 'FRA', 'DEU', 'GRC', 'HUN', 'ISL', 'IRL',
'ISR', 'ITA', 'JPN', 'KOR', 'LVA', 'LTU', 'LUX', 'MEX', 'NLD',
'NZL', 'NOR', 'POL', 'PRT', 'SVK', 'SVN', 'ESP', 'SWE', 'CHE',
'TUR', 'GBR', 'USA', 'NON', 'ARG', 'BRA', 'BRN', 'BGR', 'KHM',
'CHN', 'HRV', 'CYP', 'IND', 'IDN', 'HKG', 'KAZ', 'LAO', 'MYS',
'MLT', 'MAR', 'MMR', 'PER', 'PHL', 'ROU', 'RUS', 'SAU', 'SGP',
'ZAF', 'TWN', 'THA', 'TUN', 'VNM', 'ROW', 'APEC', 'ASEAN', 'EASIA',
'EU27', 'EU28', 'EU15', 'EU13', 'EA19', 'G20', 'ZEUR', 'ZASI',
'ZNAM', 'ZSCA', 'ZOTH'], dtype=object)

The chart below shows the percentage of foreign content in the exports of selected countries. In other words, the y-axis represents $\frac{\text{foreign content}}{\text{total exports}}$ in aggregate. So we are counting not just bicycles, but all of Indonesia’s exports. For foreign content, we are counting not just gears, tyres, etc., but all intermediate imports such as chips and so on. This measure does not reflect individual industries but the economy as a whole.

b=[]
ctr=('MYS','THA','AUS','CHN','IDN','VNM')
for i in ctr:
 ttt=tiva.loc[(tiva['variable'] == i)]
 b.append(ttt)
b=pd.concat(b)
sns.lineplot(data=b,x='year',y='value',hue='variable')
plt.legend(title="",bbox_to_anchor=(0.8,-.12),ncol=3)
plt.ylabel("%")

Text(0, 0.5, '%')

As we can see, Indonesia (IDN) has a very small share of imported content relative to its total exports, especially compared to other ASEAN countries. This is actually quite understandable, given that Indonesia’s exports are dominated by raw commodities. Raw commodities do not require intermediate inputs from other countries to produce. By this indicator, Indonesia’s participation is even higher than Australia’s, a country that essentially has no manufacturing industry.

But perhaps the most striking case is Vietnam. Its foreign content share has been rising steadily.

The chart below shows total imports of raw materials and intermediate goods for each country. The data also comes from the TiVA database. By this metric, China is far ahead of the others. However, Indonesia remains the lowest. This is notable because Indonesia’s economy is relatively larger than its neighbours’. Again, look at Vietnam.

b2=[]
for i in ctr:
 t2=tiva2.loc[(tiva2['variable'] == i)]
 b2.append(t2)
b2=pd.concat(b2)
sns.lineplot(data=b2,x='year',y='value',hue='variable')
plt.legend(title="",bbox_to_anchor=(0.8,-.12),ncol=3)

<matplotlib.legend.Legend at 0x133fa61e190>

Vietnam’s economy is tiny. In 2019, Vietnam’s GDP was only 261.9 billion USD, less than a quarter of Indonesia’s 1,119 trillion USD in the same year. How can such a small economy import so many raw materials and intermediates as shown in the chart above, and keep growing? Who consumes the finished goods? Indonesia’s large imports make sense given its large population and big market. But Vietnam? Why import so much?

The answer: to export. Let us look at high-technology exports below. This data comes from the World Bank.

import wbdata as wb
import datetime
tanggal=(datetime.datetime(1995,1,1), datetime.datetime(2019,1,1))
a=wb.get_dataframe({"TX.VAL.TECH.CD" : "hitek"}, country=["IDN","VNM","THA","MYS"],
 data_date=tanggal, convert_date=True, keep_levels=True)
a=a.reset_index()
sns.lineplot(data=a,x='date',y='hitek',hue='country')
plt.legend(title="",bbox_to_anchor=(0.8,-.12),ncol=4)
plt.ylabel("USD")
plt.xlabel("year")

<matplotlib.legend.Legend at 0x133fa6be820>

You can clearly see the growth in high-technology exports. Vietnam is not a country with a strong research tradition compared to its regional neighbours. They do not do their own R&D. They simply assemble. By importing raw materials and intermediates like chips, they become producers of high-technology goods. By the way, I did not include China in the chart above because if China were included, the others would look tiny.

The chart below shows scientific journal publications, also sourced from the World Bank. This serves as an illustration of each country’s research capacity. Compare Vietnam with the rest.

import wbdata as wb
import datetime
tanggal=(datetime.datetime(1995,1,1), datetime.datetime(2019,1,1))
a=wb.get_dataframe({"IP.JRN.ARTC.SC" : "Journal"}, country=["IDN","VNM","THA","MYS"],
 data_date=tanggal, convert_date=True, keep_levels=True)
a=a.reset_index()
sns.lineplot(data=a,x='date',y='Journal',hue='country')
plt.legend(title="",bbox_to_anchor=(0.8,-.12),ncol=3)

<matplotlib.legend.Legend at 0x133fa7e5820>

Speaking of research, Indonesia’s publication count suddenly surged from 2016 onward. Impressive! How did that happen? What is the secret?

Who published in "predatory" journals (2015-2017)?
Source: https://t.co/j5TQBWgrbr pic.twitter.com/PLYXDx9QYX
— Arief Anshory Yusuf (@anshory72) February 8, 2021

What is the takeaway? I will leave that to the reader. That is all for today’s post. Hopefully it is useful! Mention me on twitter for feedback, suggestions, or discussion. Thanks!

On District/City Minimum Wages: Too High or Too Low?

Wed, 08 Dec 2021 00:12:12 +1100

Some time ago, Twitter was abuzz about minimum wages. The trigger seemed to be news about Minister of Manpower Ida Fauziyah commenting on how high minimum wages are in Indonesia. One widely shared report was tweeted by CNN Indonesia below.

Menaker Sebut Upah Minimum di Indonesia Terlalu Tinggi https://t.co/QcN7DKj8bx
— CNN Indonesia (@CNNIndonesia) November 18, 2021

As you can see, the tweet received about 1,000 replies, far above CNN Indonesia’s usual engagement level.

But what is even more interesting is this tweet from a highly influential account:

Menteri Tenaga Kerja Ida Fauziyah mengatakan tak layak buruh menuntut kenaikan upah karena upah minimum di Indonesia sudah terlalu tinggi. Benarkah terlalu tinggi? Menurut Global Wage Report 2020-2021: upah minimum RI ada di papan bawah negara2 Asean, hanya menang dari Myanmar. pic.twitter.com/16PDZcAB7M
— Farid Gaban (@faridgaban) November 17, 2021

The account holder is someone quite respected in the Indonesian Twittersphere who frequently offers opinions that I find very critical, sharp, and thought-provoking. The chart he tweeted was taken from the published by the International Labour Organization (ILO).

The chart shows minimum wages in several countries in both US dollars and PPP dollars. It shows that Indonesia’s minimum wage is very low, only slightly above Myanmar. This seems to prove that Indonesia’s minimum wage does not match the Minister’s claim that it is “too high.”

So what was the Minister’s basis for saying that? Is there a missing link? Looking at replies and quote-tweets, we find several clues. Here is one:

UMR $111 itu UMR Yogya bukan Indonesia. Jabar sudah Rp4.7 jt ($337). Lebih tinggi dari Cina, Thailand, Malaysia.. https://t.co/MIhUJfgSOJ
— ARK (@ngabdul) November 18, 2021

It appears the Global Wage Report used the lowest minimum wage in each country. However, Indonesia’s minimum wage regulations exhibit substantial variation. Some districts/cities have very low minimum wages, while others are quite high. What the Minister likely meant by “expensive” were minimum wages in industrial hubs, not the very lowest. But as you might expect, the screenshot, with the context provided, .

So are minimum wages in Indonesia actually high or low? Well, it probably depends on the benchmark. High or low compared to what?

Sampling UMK on Java Island, 2019

I tried to do a quick fact-check, at least for Java. Following the chart tweeted by @fgaban, I will use 2019 figures and PPP dollars as the unit. According to the chart, Indonesia’s minimum wage is 331 PPP dollars. I looked up 2019 UMK regulations for various districts/cities on Java. The PPP conversion factor comes from the , while UMK data comes from:

Banten https://indolabourdatabase.files.wordpress.com/2018/12/SK-UMK-Banten-2019.pdf
East Java https://surabaya.tribunnews.com/2https://indolabourdatabase.files.wordpress.com/2018/12/SK-UMK-Banten-2019.pdf018/11/16/berikut-ini-nilai-umk-2019-di-jawa-timur-surabaya-tertinggi
DIY https://maucash.id/umr-yogyakarta
Central Java https://jatengprov.go.id/publik/umk-35-kabupaten-kota-ditetapkan-semarang-tertinggi-banjarnegara-terendah/
West Java https://databoks.katadata.co.id/datapublish/2019/11/29/daftar-umk-di-jawa-barat-karawang-cetak-tertinggi
DKI https://www.kompas.com/tren/read/2019/11/22/161505465/membandingkan-ump-dki-jakarta-dengan-umk-daerah-sekitarnya?page=all

umk19=pd.read_csv('gabung.csv').query('year == 2019')
umk19=pd.melt(umk19, id_vars=['provname'], value_vars=['meanP','minP','maxP'])
sns.barplot(data=umk19,x="provname",y='value',hue='variable')
plt.axhline(331,color='orange',linewidth=2)
plt.xticks(rotation=30)

(array([0, 1, 2, 3, 4, 5]),
[Text(0, 0, 'DKI Jakarta'),
Text(1, 0, 'Jawa Barat'),
Text(2, 0, 'Jawa Tengah'),
Text(3, 0, 'DI Yogyakarta'),
Text(4, 0, 'Jawa Timur'),
Text(5, 0, 'Banten')])

The orange line above represents 331, the figure for Indonesia’s minimum wage according to the Global Wage Report 2020-2021. Indeed, the lowest minimum wage in Indonesia is in DIY (Yogyakarta), as several people who replied to @fgaban’s tweet pointed out. However, the average UMK on Java appears quite high. The closest to DIY are Central Java and East Java. But West Java and Banten have much higher average UMKs. Is this the effect of being near Jakarta?

By the way, Jakarta’s average, maximum, and minimum are all the same. Obviously, because it only has a provincial minimum wage (UMP). For other provinces, the UMK varies by district/city. The variation is quite different across provinces.

Some additional facts about the 2019 UMK (in Rupiah):

Province	Highest UMK	Amount	Lowest UMK	Amount
Banten	Kota Cilegon	3,913,078	Lebak	2,498,068
West Java	Karawang	4,594,324	Kota Banjar	1,831,884
Central Java	Kota Semarang	2,715,000	Banjarnegara	1,748,000
D.I. Yogyakarta	Kota Yogyakarta	1,846,400	Gunung Kidul	1,571,000
East Java	Kota Surabaya	3,871,052	multiple	1,763,267

The highest UMK is the red bar, the lowest is the green, and blue is the average. In East Java, the districts/cities with the lowest UMK include PAMEKASAN, NGAWI, PONOROGO, SITUBONDO, MADIUN, MAGETAN, PACITAN, SAMPANG, and TRENGGALEK (apologies for the all-caps – the labels came from Stata).

As shown above, the variation between the highest and lowest is remarkable. The biggest gap is probably in West Java, which is reportedly home to . Karawang’s UMK is even higher than DKI Jakarta’s! It is perhaps unsurprising that .

UMK Trends on Java, 2002-2014 (+2019)

Minimum wages are not the focus of my research. Fortunately, I have a colleague whose research centres on minimum wages. She is , whom I usually call Dika. She recently presented on the impact of minimum wages on poverty. She used Java UMK data from 2002 to 2014, and was kind enough to share it with me.

Below I plot UMK trends on Java for 2002-2014 using Dika’s data. I added my own collected 2019 data. I do not have figures for 2015 to 2018, so in the charts below you will see a straight line between 2014 and 2019. I include the same chart in both PPP dollars and Rupiah.

umk=pd.read_csv('gabung.csv')
fig, axs = plt.subplots(figsize=(10,5),ncols=2)
sns.lineplot(data=umk,x='year',y='maxP',hue='provname',
 style='provname',linewidth=2, ax=axs[0],legend=False)
sns.lineplot(data=umk,x='year',y='minP',hue='provname',
 style='provname',linewidth=2, ax=axs[1],legend=False)
axs[0].set_title('Highest UMK in each province')
axs[0].set_ylabel('PPP$')
axs[0].set_xticks([2002,2005,2008,2011,2014,2019])
axs[1].set_title('Lowest UMK in each province')
axs[1].set_ylabel('')
axs[1].set_xticks([2002,2005,2008,2011,2014,2019])
axs[0].set_ylim(100,1000) #a
axs[1].set_ylim(100,1000) #b
labels=['DKI Jakarta','Jawa Barat','Jawa Tengah','DI Yogyakarta','Jawa Timur','Banten']
plt.legend(title='',labels=labels, bbox_to_anchor=(0.7,-0.15), ncol=3)

<matplotlib.legend.Legend at 0x238ae9840d0>

fig, axs = plt.subplots(figsize=(10,5),ncols=2)
sns.lineplot(data=umk,x='year',y='maximum',hue='provname',
 style='provname',linewidth=2, ax=axs[0],legend=False)
sns.lineplot(data=umk,x='year',y='minimum',hue='provname',
 style='provname',linewidth=2, ax=axs[1],legend=False)
axs[0].set_title('Highest UMK in each province')
axs[0].set_ylabel('current IDR')
axs[0].set_xticks([2002,2005,2008,2011,2014,2019])
axs[1].set_title('Lowest UMK in each province')
axs[1].set_ylabel('')
axs[1].set_xticks([2002,2005,2008,2011,2014,2019])
labels=['DKI Jakarta','Jawa Barat','Jawa Tengah','DI Yogyakarta','Jawa Timur','Banten']
plt.legend(title='',labels=labels, bbox_to_anchor=(0.7,-0.15), ncol=3)

<matplotlib.legend.Legend at 0x238ac3f8730>

From both charts, we can see that UMK rose gradually but then jumped sharply around 2012/2013, and increased even more steeply by 2019. Unfortunately, I am not interested enough to fill in the 2015-2018 data, but minimum wages almost certainly did not decline, so the insights from those years would not add much. The trend is definitely upward toward 2019.

In any case, the charts above illustrate how rapidly UMK on Java has been rising, likely well above inflation.

So Are Indonesia’s Minimum Wages High or Low?

A proper investigation would need to go beyond Java and beyond 2019. I strongly recommend that interested readers look at more comprehensive studies or contact Dika to chat. However, I think our exercise here is reasonably valid, especially since both the highest and lowest UMKs are on Java. Java also remains the top destination for manufacturing investors, so I believe Java provides sufficient insight to continue the discussion.

Returning to the question of high or low, I think we can at least look at it from two angles. First, comparing wages between Indonesia and other countries. Second, comparing Indonesia’s current wages with its own past (in other words, the rate of increase).

On the first point, comparing Indonesia’s lowest minimum wage with the lowest minimum wage in other countries using PPP dollars is appropriate. This is the advantage of PPP dollars – inflation and consumer prices in each country are already incorporated, making cross-country comparisons valid.

However, the lowest minimum wage alone does not provide full context. If we look deeper into the report, there is actually a lot of information about the distribution of minimum wages. Figure 7.16 in the report gives some sense of the UMK distribution in Indonesia. Reading this chart alongside the one tweeted by @fgaban immediately gives a better picture of UMK distribution and adds context to the Minister’s comment.

But the most important chart is Figure 9.7 in the same report. This chart, about three pages after Figure 9.3 (which @fgaban posted), shows the lowest and highest minimum wages in several countries.

From the chart above, we can see that the minimum wage in the lowest-paying province in Indonesia is indeed relatively lower than in the Philippines, Vietnam, and Brazil. However, if we look at the highest minimum wage, Karawang’s UMK is higher than the minimum wages in Vietnam and the Philippines, and even higher than Shanghai’s minimum wage!

The second angle is comparing current UMK with historical levels. We can see that on Java, UMK has been rising very rapidly, especially in DKI Jakarta and surrounding areas. This gives businesses a clear signal that in the long term, minimum wages in these areas will keep rising, and at some point production there will no longer be attractive.

There is actually a third angle (sorry): comparing minimum wage regulations with the median and/or average wage. You may remember from high school statistics the difference between the mean and the median. The minimum wage is supposed to be the lowest wage that should be paid in a given area. Since it is the floor, there should naturally be a gap between it and the mean or median. If the minimum wage is too close to the median or mean, it may indicate that the minimum wage is set too high for the business sector.

Fortunately, this can also be seen in the same report.

The chart above shows $\frac{\text{minimum wage}}{\text{average wage}}$ (red bar) and $\frac{\text{minimum wage}}{\text{median wage}}$ (blue bar). 100% means the minimum wage equals the average or median wage. As we can see, Indonesia’s minimum wage is above the global average when measured by the ratios $\frac{\text{minimum wage}}{\text{average wage}}$ and $\frac{\text{minimum wage}}{\text{median wage}}$. By this measure, Indonesia’s minimum wage regulation appears somewhat high relative to the regional wage level (mean or median), compared with Vietnam, China, the Philippines, and Thailand.

In other words, there are many indicators for determining whether Indonesia’s minimum wage is “too high” or “too low.” We do not really know which measure the Minister used when she said minimum wages are too high. There are many layers between the original statement and the Twitter frenzy. The Minister made her statement at a press conference, which most of us did not attend. The press wrote it up, chose a “representative” headline, posted it on Twitter, and it was disseminated by netizens with their own added context. Communication distortion can occur at any level in this process. We do not know whether people who read the tweet also read the article, or the full report, and so on.

According to , the Minister used the Kaitz Index. The Kaitz Index is more appropriately linked to $\frac{\text{minimum wage}}{\text{median wage}}$. Using this measure, the Minister’s statement appears reasonably valid.

Minimum wages that are too high can also be seen through changes in the level of economic informality. The informal economy consists of economic actors (workers and business owners) who are not registered in business databases. Informal economic actors generally do not have tax identification numbers (and do not pay taxes), and rarely have protections such as employment insurance or pensions. Because they are unregistered, informal businesses are also more likely to pay below the minimum wage, and informal workers are generally paid below it.

In other words, the higher the minimum wage, the less incentive businesses have to formalise, because they would have to pay the minimum wage. Likewise, as the number of jobs paying above the minimum wage shrinks, workers are forced to join the informal economy and accept wages below the minimum. In other words, an excessively high minimum wage can lead to a growing “black market” in the labour market. Indonesia is one of the countries with a .

That is all for this post. So are minimum wages too high or too low? Ask Dika, not me. The main point is to always be critical when receiving information, whether from Twitter posts or any other source (including this blog!). Always cross-check and make the effort to read more in-depth studies, or better yet, conduct your own!

n.b.: The 2019 Java UMK data is shown below. For the 2002-2014 data, feel free to contact Dika.

pd.set_option('display.max_rows', None)
umk.query('year==2019')

Global Value Chain dalam bumbu gado-gado

Tue, 31 Aug 2021 23:27:21 +1000

Hari ini saya sarapan gado-gado. Menurut saya, gado-gado (dan bumbu kacang secara umum) adalah sebuah fenomena yang luar biasa. Meskipun kacang pertama kali diperkenalkan oleh Portugis yang dibawa dari Mexico, Masyarakat Indonesialah yang pertama kali membuat bumbu kacang. Bumbu kacang ini sungguh versatile karena bisa dicampur apa aja. Dikasih sayur, bisa. Kasih gorengan, enak. Nasi uduk, hayu. Buat bumbu ayam bakar, ya bisa (alias jadi sate).

Orang bule juga ternyata sangat menyukai bumbu kacang. Seperti kita ketahui, orang bule kalo sarapan suka pake roti yang diolesin selai kacang. Tentunya sungguh wajar jika mereka juga suka bumbu kacang. Daripada salad dressing a la eropa, jelas lah enakan gado-gado. wkwkwk. Kesukaan rang bule akan bumbu kacang (ditambah dengan banyaknya orang Asia dan orang Indonesia di Australia) memberikan skala ekonomi yang cukup untuk importir bumbu kacang.

Tentunya sebagai insan yang ga bisa masak, saya lebih suka pake bumbu kacang yang udah jadi. Praktis. ga perlu ngulek-ngulek. Ntar kalo bikin sendiri mah udahlah nguleknya capek, makan waktu, kagak enak pula! Udahla mending kita beli bumbu jadi aja. Praktis dan udah pasti rasanya oke.

Saya suka beli bumbu ini, yang namanya “Enak Eco”, beli di toko Vietnam deket rumah. Harganya sekitar A$4. Mantap ya. Export quality wkwkw. Diekspor dari mana sih benda ini?

Ternyata bumbu ini adalah karya anak bangsa. Made in Malang, tepatnya diproduksi oleh PT Enak Jaya Makmur. Diimpor oleh Sony Trading di New South Wales.

Cabe lalu gado-gado: peran Global Value Chain

Beberapa waktu lalu, saya sempet bikin postingan tentang impor cabe yang ternyata lumayan banyak peminatnya. Memang ketika itu situasinya adalah harga cabe sedang rendah, sehingga petani merugi. Di postingan tersebut juga saya memperlihatkan bahwa cabe mentah sudah dibatasi impornya, sementara cabe kering impornya bebas.

Katanya sih cabe mentah dikonsumsi oleh masyarakat, sementara cabe kering dikonsumsi oleh industri. Meskipun ternyata ga juga, akan tetapi tidak dapat kita pungkiri bahwa industri berperan besar dalam mengimpor cabe, terutama melihat betapa besarnya pertumbuhan industri makanan dan minuman di Indonesia.

Menggunakan produk impor untuk diolah di dalam negeri menjadi produk ekspor bukanlah hal baru. Sejak zaman di mana banyak negara melakukan perjanjian dagang, adalah hal lumrah bagi perusahaan di suatu negara untuk menggunakan bahan baku dari negara lain. Seperti halnya coklat belgia yang cokelatnya datang dari pantai gading, gado-gado Indonesia pun cabenya datang dari negara lain. Vietnam, negara yang sejatinya berteknologi rendah, dapat menjadi eksportir ponsel pintar besar dunia karena mengimpor komponen. Ekonom memakai istilah “Global Value Chain” (GVC) untuk mendeskripsikan fenomena ini.

Di postingan cabe itu, kita juga jadi tahu bahwa ada negara lain yang sanggup bikin cabe lebih murah daripada Indonesia. Akan tetapi, mereka ga bisa bikin gado-gado seenak kita lho (mungkin). Hahaha. Dengan cabe yang murah, maka bumbu gado-gado juga bisa jadi murah. Akibatnya, produk ini jadi laris tidak hanya di pasar dalam negeri, tetapi juga di pasar luar negeri.

Tanpa berpanjang lebar lagi, mari kita coba cek ‘kandungan internasional’ di dalam bumbu gado-gado impor ini. haha.

Ingredients dengan bantuan UN COMTRADE

Tentunya saya ga begitu paham struktur industri bumbu gado-gado. Karena itu, kita akan pake kira-kira aja. Dari bungkus belakang bumbu ini, dapat dilihat bahwa bumbunya antara lain adalah kacang tanah, cabe (capsicum), coconut sugar, bawang putih, bawang merah, garem, gula, dan pengawet (hiii baru ngeh ada pengawetnya tidaaaak).

Dari list tersebut, saya tinggal googling HS Code, untuk mendapatkan kode HS buat kita tarik datanya dari UN COMTRADE. Di sini kita sekali lagi harus bikin caveat wkwkw. Pertama, Ga ada informasi bahan-bahan ini dibeli oleh PT Enak Jaya Makmur dalam bentuk apa. Misalnya, apa kacang tanahnya murni yg masih ada kulitnya? Atau udah dikupas dan disangrai? Cabe pun demikian. Basah? Kering? Bubuk? Kode HS nya beda-beda. haha! Karena tidak ada informasi ini, maka saya melakukan sedikit spekulasi. Jadi jangan terlalu percaya sama kode HS yang akhirnya saya pake wkwkwkw. Jika anda lebih tahu, kode yg saya pake di postingan ini bisa direplikasi dengan mudah, jadi bisa coba versi anda sendiri. Atau lebih baik lagi, kasih tau saya resep yg benernya melalui twitter wkwkwk.

Seperti halnya postingan cabe, kita tarik data-data yang diperlukan pake API dari UN COMTRADE.

# cari api key-nya di https://comtrade.un.org/Data/

api_key='/api/get?max=502&type=C&freq=A&px=HS&ps=2019&r=360&p=0&rg=2%2C1&cc=200811%2C170290%2C070320%2C070310%2C071220%2C250100%2C291619%2C170111%2C090421%2C170114'

url='https://comtrade.un.org/'+api_key+'&fmt=csv'

gado2=pd.read_csv(url)

gado2=gado2[['Year','Trade Flow','Commodity Code', 'Commodity',

 'Netweight (kg)', 'Trade Value (US$)']]

Setelah kita download datanya, lalu saya coba perlihatkan deskripsi dari komoditas yang saya pake. Itu 4 komoditas pertama harusnya ada 0 di depannya (jadi 070310, 6 digit) tapi saya terlalu malas untuk konversi jadi biar aja jadi 5 digit wkwkw.

pd.set_option('display.max_colwidth', None) # show full content of a column

desc=gado2[['Commodity Code','Commodity']] # ambil cuma kode dan nama barang

desc.drop_duplicates() # print

	kabid	year	umk	prov	kabname	provname	rerata	minimum	maximum	PPPC	inflation	PPP	meanP	minP	maxP
2002	3101	2019	4267349.00	31	KEPULAUAN SERIBU	DKI Jakarta	4267349.0	4267349.0	4267349.0	4751.936	3.031	898.02325	898.02325	898.02325	898.02325
2003	3171	2019	4267349.00	31	KODYA JAKARTA SELATAN	DKI Jakarta	4267349.0	4267349.0	4267349.0	4751.936

	Commodity Code	Commodity
0	70310	Vegetables, alliaceous; onions and shallots, fresh or chilled
2	70320	Vegetables, alliaceous; garlic, fresh or chilled
4	71220	Vegetables; onions, whole, cut, sliced, broken or in powder but not further prepared, dried
6	90421	Spices; fruits of the genus Capsicum or Pimenta, dried, neither crushed nor ground
8	170114	Sugars; cane sugar, raw, in solid form, other than as specified in Subheading Note 2 to this chapter, not containing added flavouring or colouring matter
10	170290	Sugars; n.e.c. in heading no. 1702, including invert sugar and other sugar and sugar syrup blends containing, in the dry state, 50% by weight of fructose
12	200811	Nuts; ground-nuts, whether or not containing added sugar, other sweetening matter or spirit
14	250100	Salt (including table salt and denatured salt); pure sodium chloride whether or not in aqueous solution; sea water
16	291619	Acids; unsaturated acyclic monocarboxylic, cyclic monocarboxylic, their anhydrides, halides, peroxides and peroxyacids; their halogenated, sulphonated, nitrated or nitrosated derivatives, n.e.c. in item no. 2916.1

Seperti dapat dilihat, deskripsinya agak menyebalkan wkwk. Saya sertakan juga bubuk bawang di situ. Untuk cabe, saya pake cabe kering. Saya ga gitu tau Coconut sugar pake HS yg mana, tapi jadinya di sini saya pake 170290. Feel free to disagree with my choices hahaha.

Oh iya satu lagi, saya pake data tahun 2019. 2020 ada tapi takut rusak gara-gara COVID. 2019 harusnya ga jauh beda dengan business as usual.

Okeh, setelah dapat barang-barang yang mau dilirik, mari kita tampilkan ekspor dan impornya bahan-bahan ini.

## Tarik ekspor doang

x1=gado2.query('`Trade Flow`=="Export"').reset_index()

x1=x1[['Commodity Code','Netweight (kg)','Trade Value (US$)']]

x1=x1.rename(columns={"Netweight (kg)": "Ekspor (kg)",

 "Trade Value (US$)": "Ekspor (US$)"})

## Tarik impor doang

m1=gado2.query('`Trade Flow`=="Import"').reset_index()

m1=m1[['Commodity Code','Netweight (kg)','Trade Value (US$)']]

m1=m1.rename(columns={"Netweight (kg)": "Impor (kg)",

 "Trade Value (US$)": "Impor (US$)"})

## Gabung dan bikin tabelnya enak dilihat

gabung=pd.merge(x1,m1,on='Commodity Code',how='outer')

gabung['Product description']=['Onion/Shallot','Garlic','Onion Powder','Dried Chili','Cane Sugar','Coconut Sugar','Ground-nuts','salt',

 'Perservatives']

gabung=gabung[['Commodity Code','Product description','Ekspor (kg)','Impor (kg)','Ekspor (US$)','Impor (US$)']]

## Print dengan format thousand separated by comma

gabung.style.format({"Ekspor (kg)": "{:,.0f}", "Impor (kg)": "{:,.0f}"

 , "Ekspor (US$)": "{:,.0f}", "Impor (US$)": "{:,.0f}"})

	Commodity Code	Product description	Ekspor (kg)	Impor (kg)	Ekspor (US$)	Impor (US$)
0	70310	Onion	8,786,797	111,934,997	10,588,352	56,596,401
1	70320	Garlic	25,804	465,344,332	11,977	529,965,497
2	71220	Onion Powder	6,224	6,098,177	19,052	17,914,797
3	90421	Dried Chili	321,959	39,132,270	934,735	65,280,761
4	170114	Cane Sugar	124,025	3,965,555,215	113,024	1,317,595,639
5	170290	Coconut Sugar	39,810,415	91,207,379	53,547,449	71,438,428
6	200811	Ground-nuts	2,768,870	3,126,465	8,220,574	5,334,185
7	250100	salt	542,077	2,595,397,085	155,069	95,522,423
8	291619	Perservatives	2,229	2,277,642	13,200	11,570,694

Seperti dapat dilihat pada tabel di atas, hampir semua bahan-bahan yang saya gunakan di sini memiliki impor yang jauh lebih tinggi daripada ekspornya. Dengan kata lain, kemungkinan kandungan internasionalnya (setidaknya dari sisi bahan baku) cukup tinggi. Bumbu gado-gado ternyata adalah produk GVC!! Kita beli bahan baku dari negara lain, kita olah, lalu kita jual lagi ke berbagai pasar di belahan dunia. Di konsep GVC, kita bisa bilang PT Eka melakukan “Backward participation”. Ga salah emang kalo Indonesia dibilang surganya makanan!

But at what cost?

Kebijakan pembatasan impor

Diantara bahan-bahan yang saya pake di atas, sebagian mengalami pembatasan impor, dan sebagian lain dapat diimpor secara bebas. Bawang merah, bawang putih, gula, kacang dan garam adalah barang-barang yang impornya diregulasi. Dengan kata lain, pemerintah menerapkan kuota impor, di mana hanya perusahaan tertentu yang ditunjuk oleh negara (biasanya BUMN) yang boleh impor.

Pembatasan ini dirasa perlu karena Pemerintah berupaya melindungi petani, peternak dan nelayan di Indonesia dari kompetisi internasional. Indonesia memang memiliki beragam masalah dalam memajukan sektor pertanian. Solusi paling mudah dan populer adalah dengan melakukan pembatasan impor. Kenapa bisa populer ya? Ntahlah.

Sayangnya, meski memiliki tujuan mulia, tidak semua hal yang diserahkan negara memiliki pengelolaan yang baik. Diantara kasus korupsi kuota impor yang paling terkenal mungkin adalah kasus suap kuota daging sapi yang melibatkan petinggi penting sebuah parpol. Namun kasus korupsi juga tidak jauh dari bumbu gado-gado yang saya beli. Diantaranya kasus suap kuota bawang putih, yang akhirnya pelakunya dipenjara meskipun di bawah tuntutan. Di samping itu, ada juga kasus korupsi gula yang melibatkan eks bos BUMN gula, dan ada juga tersangka pengadaan garam impor. Suap-menyuap ini, menurut KPPU, diakibatkan oleh buruknya transparansi penunjukkan perusahaan yang berhak mendapatkan kuota.

Tidak semuanya kasus korupsi, kadang pengendalian perdagangan juga tidak sempurna dan mengakibatkan adanya barang ilegal atau palsu, seperti di kasus bawang merah.

Salah satu penelitian yang mencoba mendokumentasikan hal-hal ini ditulis oleh kolega saya Felippa Amanta yang bisa dibaca di sini.

Namun pada prinsipnya, kebijakan pembatasan impor sejatinya adalah sebuah keberpihakan. Produsen bahan baku (petani, nelayan dan peternak) adalah pihak yang diuntungkan dari dilakukannya kebijakan seperti ini. Koruptor juga untung sih hahaha. Yang rugi siapa? Tentunya industri manufaktur dan konsumen secara umum. Dengan akses yang sulit terhadap bahan baku, maka akan sulit manufaktur kita berkembang.

Kalau ditanya ke orang ekonomi, maka pastinya jawabannya adalah soal efisiensi. Mana yang lebih memberikan nilai tambah tinggi? proteksi petani untuk konsumsi domestik, atau mengimpor bahan baku dan menghasilkan produk yang bisa diekspor? Biasanya sih keterbukaan perdagangan memberikan hasil yang lebih efisien. Bahkan, kalo kata dosen saya di sini, akan lebih baik jika “loser” dari perdagangan kita kompensasikan dari “gain” yang didapatkan dari perdagangan bebas. Dengan kata lain, kita pajakin eksportir (looking at you CPO dan batubara), lalu uangnya kita gunakan untuk membiayai program-program pemerintah yang beriorientasi pada pelayanan publik, seperti kesehatan, pendidikan dan infrastruktur. Di samping itu, fasilitasi untuk mempermudah perpindahan tenaga kerja dari yang kerja di sektor “loser” untuk pindah ke sektor “gainer” bisa juga jadi opsi. Sebenarnya anak muda yang memilih jadi produsen manufaktur daripada petani adalah sebuah awal menuju perubahan struktur ekonomi yang tinggi agrikultur menuju tinggi manufaktur.

Bisa jadi juga program-program seperti itu lumayan kompleks dan merepotkan. Program-program tersebut juga bisa aja dikorup juga. Di samping itu, pembatasan impor lebih mudah, dan disenangi tidak hanya masyarakat tetapi juga elit politik dan birokrat.

Mungkin juga ada alasan yang lebih fundamental. Lebih ideologis. Apa artinya efisien jika identitas kita sebagai negara agraris terkikis? Jika identitas sebuah bangsa adalah berdasarkan produk yang dia ekspor, apa mau kita berubah dari negara agraris jadi negara CPO? Atau negara pembuat gado-gado? Semua demi efisiensi?

Yha pada akhirnya ini adalah pertanyaan politik, bukan ekonomi. Ekonomi mah cuma sebuah alat aja, untuk membantu para pembuat visi untuk mencapai visi tersebut. Jika masyarakat memilih untuk menjaga identitas sebagai negara agraris dengan mengabaikan efisiensi, maka tentu itu wajar-wajar saja.

Sekian postingan ini! Mari kita selalu optimis menjadi bangsa yang besar, maju, dan berdaya saing! Dan tetap makan gado-gado!

The chili price mystery: low because of imports?

Sat, 28 Aug 2021 23:22:01 +1000

Chili is one of the most important commodities in Indonesia. Indonesians love chili. Even while studying abroad, I still bought sambal because chili is just that good. It’s not only vital in various Indonesian dishes but also contains very high levels of vitamin C!

One morning, I saw a somewhat viral video of a farmer angrily kicking and stomping on his chili plants.

Ngene Iki Lurrrrrr akibaté, nek Rego / Harga si-Lombok terjun bebas, murahnya kebangetan,..petani tekor, modal tanam tekadang pinjaman / kredit Bank, tahu kenapa...! pic.twitter.com/IjTEcvZxO0
— Susilo Samin Blora (@Susilo_Blora) August 26, 2021

According to the account I embedded above, the cause appears to be the collapse of chili prices during harvest. This is obviously heartbreaking. Imagine – months of effort, cost, time, and energy, all lost when output prices plummet.

Another post I found quite interesting. I’ve embedded it below:

lanjutannya. diambil dari sini sepertinya ya https://t.co/Lz4vNibZb6 pic.twitter.com/lthDKPFjvc
— Krisna 'imed' Gupta (@iMedKrisna) August 28, 2021

That post was interesting for two reasons. First, the account linked the phenomenon to the Job Creation Law (UUCK), claiming that now chili imports no longer need to consider domestic farmer production. Second, the account took an image and headline from a Bisnis.com article titled “Chili Supply is in Surplus, So Why Are We Importing 27,851 Tons?”.

Of course, what’s interesting is that the article confirms the imports were dried/powdered chili for the manufacturing industry. Quoting the article:

Bisnis.com, JAKARTA – The Ministry of Agriculture states that the supply of various chilies for consumption in Indonesia is in surplus. Indonesia also imports chili to meet industrial needs.

Director of Horticultural Product Processing and Marketing at the Ministry of Agriculture, Bambang Sugiharto, explained that the 27,851 tons of chili imported in the first semester of 2021 was to meet industrial needs. Chili was imported in the form of dried chili, crushed or ground chili, not fresh chili for consumption.

Based on BPS data, national chili production in 2020 reached 2.77 million tons. This figure represents a 7.11% increase compared to 2019. Additionally, Indonesia exported various chilies valued at USD 25.18 million in 2020, up 69.86% from 2019.

Unfortunately, the account didn’t provide a link to the actual article for its followers to read. Just a screenshot of the headline. Be careful, folks! Always question headline screenshots and always read the full article before sharing with a particular prejudice!

But regardless, I found the narrative quite interesting. The account linked the phenomenon to imports. In other words, chili prices fell because imports happened during the chili harvest season. Second, the account linked these imports to the UUCK.

It’s obviously hard to prove this comprehensively and coherently. But maybe we can at least estimate a few things.

First, for imports to affect domestic prices, we need to at least examine the magnitude of imports relative to local production and exports. According to the article excerpt above, national production is 2.77 million tons. Moreover, Indonesia is actually a chili exporter!

Second, the claim that high chili imports are caused by the UUCK. This means that before the UUCK, imports should have been lower or even insignificant, then risen in 2020 and 2021.

Let’s check with export-import data. For this post, I use UN COMTRADE’s API because the data is quite comprehensive and accessible to anyone. Skip this section if you don’t want to see the code.

I’ll pull total Indonesian export and import data to and from other countries. There are 2 goods I’ll extract data for. First is Fresh Chili with HS 6-digit code 070960, which is alleged to be heavily imported since the UUCK took effect. Second is Dried Chili (non-crushed) with HS 6-digit code 090421, the product that was actually imported according to the article above. See – fresh chili and dried chili have different HS codes. Dangerous to just read headlines.

From these two datasets, let’s see whether we actually imported a lot of Fresh Chili starting from 2020, when the UUCK took effect. Unfortunately, UN COMTRADE doesn’t have 2021 data yet. BPS has monthly data, but since not everyone has access, we’ll hold off on 2021.

Pulling data from UN COMTRADE

Pull via API from UN COMTRADE.

# cari api key-nya di https://comtrade.un.org/Data/
api_key='/api/get?max=502&type=C&freq=A&px=HS&ps=2016%2C2017%2C2018%2C2019%2C2020&r=360&p=0&rg=2%2C1&cc=070960%2C090421'
url='https://comtrade.un.org/'+api_key+'&fmt=csv'
cabe=pd.read_csv(url)
# cek nama variabel
cabe.columns

Index(['Classification', 'Year', 'Period', 'Period Desc.', 'Aggregate Level',
'Is Leaf Code', 'Trade Flow Code', 'Trade Flow', 'Reporter Code',
'Reporter', 'Reporter ISO', 'Partner Code', 'Partner', 'Partner ISO',
'2nd Partner Code', '2nd Partner', '2nd Partner ISO',
'Customs Proc. Code', 'Customs', 'Mode of Transport Code',
'Mode of Transport', 'Commodity Code', 'Commodity', 'Qty Unit Code',
'Qty Unit', 'Qty', 'Alt Qty Unit Code', 'Alt Qty Unit', 'Alt Qty',
'Netweight (kg)', 'Gross weight (kg)', 'Trade Value (US$)',
'CIF Trade Value (US$)', 'FOB Trade Value (US$)', 'Flag'],
dtype='object')

Lots of variables. Let’s keep only the important ones. I’ll make two types of tables, one for Fresh Chili and one for Dried Chili.

# Ambil variabel yg penting aja
cabe=cabe[['Year','Trade Flow','Commodity Code', 'Commodity',
 'Netweight (kg)', 'Trade Value (US$)']]
# Menyiapkan cabe mentah
xcabe_mentah=cabe.query('`Commodity Code` == 70960 & `Trade Flow` == "Export"').sort_values(by=['Year']).reset_index()
xcabe_mentah=xcabe_mentah[['Year','Netweight (kg)','Trade Value (US$)']]
xcabe_mentah=xcabe_mentah.rename(columns={"Netweight (kg)": "Ekspor (kg)",
 "Trade Value (US$)": "Ekspor (US$)"})
mcabe_mentah=cabe.query('`Commodity Code` == 70960 & `Trade Flow` == "Import"').sort_values(by=['Year']).reset_index()
mcabe_mentah=mcabe_mentah[['Year','Netweight (kg)','Trade Value (US$)']]
mcabe_mentah=mcabe_mentah.rename(columns={"Netweight (kg)": "Impor (kg)",
 "Trade Value (US$)": "Impor (US$)"})
#cabe_mentah=cabe_mentah.set_index('Year')
#cabe_mentah['Tahun']=pd.to_datetime(cabe_mentah['Year'],format='%Y')
xcabe_kering=cabe.query('`Commodity Code` == 90421 & `Trade Flow` == "Export"').sort_values(by=['Year']).reset_index()
xcabe_kering=xcabe_kering[['Year','Netweight (kg)','Trade Value (US$)']]
xcabe_kering=xcabe_kering.rename(columns={"Netweight (kg)": "Ekspor (kg)",
 "Trade Value (US$)": "Ekspor (US$)"})
mcabe_kering=cabe.query('`Commodity Code` == 90421 & `Trade Flow` == "Import"').sort_values(by=['Year']).reset_index()
mcabe_kering=mcabe_kering[['Year','Netweight (kg)','Trade Value (US$)']]
mcabe_kering=mcabe_kering.rename(columns={"Netweight (kg)": "Impor (kg)",
 "Trade Value (US$)": "Impor (US$)"})

Fresh Chili

First let’s check the fresh chili export-import table, along with the charts.

cm=pd.merge(xcabe_mentah,mcabe_mentah,on='Year',how='outer')
print(cabe.Commodity.iloc[2])
print(cabe['Commodity Code'].iloc[2])
cm

Vegetables; fruits of the genus capsicum or of the genus pimenta, fresh or chilled
70960

	Year	Ekspor (kg)	Ekspor (US$)	Impor (kg)	Impor (US$)
0	2016	433825	587063	NaN	NaN
1	2017	309464	635331	NaN	NaN
2	2018	152292	387459	120.0	526.0
3	2019	156327	316311	2112.0	4216.0
4	2020	418972	669580	6.0	37.0

fig, axes = plt.subplots(1, 2, figsize=(15, 5))
fig.suptitle('Ekspor dan Impor Cabai Mentah (HS6:070960),Indonesia, 2016-2020 (UN COMTRADE)')
sns.lineplot(ax=axes[0], data=cm ,x='Year', y='Ekspor (kg)',marker='o')
axes[0].set_title('Ekspor (kg)')
axes[0].xaxis.set_ticks([2016,2017,2018,2019,2020])
sns.lineplot(ax=axes[1], data=mcabe_mentah ,x='Year', y='Impor (kg)',marker='o')
axes[1].set_title('Impor (kg)')
axes[1].xaxis.set_ticks([2016,2017,2018,2019,2020])

[<matplotlib.axis.XTick at 0x26ed40c7ca0>,
<matplotlib.axis.XTick at 0x26ed40c7c70>,
<matplotlib.axis.XTick at 0x26ed40f9520>,
<matplotlib.axis.XTick at 0x26ed41123a0>,
<matplotlib.axis.XTick at 0x26ed4126df0>]

As shown in the table and chart above, our fresh chili imports are tiny. In 2020, when the UUCK took effect, imports didn’t increase – they actually fell. UN COMTRADE data says in 2020, Indonesia imported 6 kilograms of chili. This figure is clearly far below the 2.77 million ton production level. UNCOMTRADE has no fresh chili import data for 2016 and 2017, probably because we simply didn’t import fresh chili in those years.

In fact, we’re actually a net exporter of fresh chili. Quite interesting actually. This means our country is fairly competitive in chili production (perhaps?). Far from importing, we’re exporting a lot! If so, why are farmers losing money? Are chili farmers just being dramatic?

Not necessarily! Many factors could be at play. It could be distribution to markets, someone might be hoarding supply, not all farmers may have access to export markets, consumer demand might be weak, or planting seasons might not be coordinated. For instance, during planting season everyone plants and nobody harvests, so chili is scarce at market. During harvest, everyone harvests simultaneously, creating a supply glut. When supply is high, prices crash.

There are many possible issues I don’t fully understand. But clearly, imports are not the problem. Don’t get me wrong – there are many problems caused by import rents (and export rents), but apparently not for chili.

Dried Chili (non-crushed)

One problem we could hypothesize is weak industrial absorption of farmers’ chili. This would be reflected in dried chili imports. The logic: if there’s an oversupply of fresh chili, those chilies should be usable for dried chili production, reducing dried chili imports. Let’s look at how dried chili exports and imports look in the table and chart below.

ck=pd.merge(xcabe_kering,mcabe_kering,on='Year',how='outer')
print(cabe.Commodity.iloc[10])
print(cabe['Commodity Code'].iloc[10])
ck

Spices; fruits of the genus Capsicum or Pimenta, dried, neither crushed nor ground
90421

	Year	Ekspor (kg)	Ekspor (US$)	Impor (kg)	Impor (US$)
0	2016	680078	4497982	24374457	31019978
1	2017	673598	2393926	37490445	45749699
2	2018	223126	725423	34409357	51208284
3	2019	321959	934735	39132270	65280761
4	2020	446247	1095573	25674802	52323399

fig, axes = plt.subplots(1, 2, figsize=(15, 5))
fig.suptitle('Ekspor dan Impor Cabai Kering/Bubuk (HS6:090421),Indonesia, 2016-2020 (UN COMTRADE)')

sns.lineplot(ax=axes[0], data=ck ,x='Year', y='Ekspor (kg)',marker='o')
axes[0].set_title('Ekspor')
axes[0].xaxis.set_ticks([2016,2017,2018,2019,2020])
sns.lineplot(ax=axes[1], data=ck ,x='Year', y='Impor (kg)',marker='o')
axes[1].set_title('Impor')
axes[1].xaxis.set_ticks([2016,2017,2018,2019,2020])

[<matplotlib.axis.XTick at 0x26ed47da910>,
<matplotlib.axis.XTick at 0x26ed47da8e0>,
<matplotlib.axis.XTick at 0x26ed480bd90>,
<matplotlib.axis.XTick at 0x26ed483f400>,
<matplotlib.axis.XTick at 0x26ed482db50>]

Now dried chili is a different story. While there are exports, imports still dominate. The import scale in the chart is so large it had to be expressed in scientific notation (1e7 or $10^7$). Dried chili imports are large. I’m also not sure whether dried chili imports are controlled like fresh chili, but looking at the data, it shouldn’t have too significant an impact on farmer prices. Indonesia has a very large food industry, so it’s quite reasonable that it needs more supply than farmers can provide. Moreover, dried chili has a storage advantage – you can buy in bulk and store it. This means dried chili prices probably don’t need to depend on harvest timing. Advantageous for industry.

The problem, of course, is when consumers also switch to buying dried chili. If the price gap is really that wide, even those who prefer fresh chili might opt for dried. As mentioned above, dried chili imports are large. How much is actually used by industry? Should we ban dried chili imports too? What about industries that use chili (like producers of instant gado-gado sauce) that would have to buy chili at higher prices?

The chili mystery

Regardless, low chili prices for farmers remain a problem. I dug around a bit on this chili issue and found only 1 paper that seems to agree that the problems are many, but not imports.

Muflikh et al. (2021)¹ state that since 2016, Indonesia has completely stopped importing chili because the reference price policy has completely failed to control chili prices that are always volatile each year. This could influence why industry imports so much dried chili: abroad, dried chili producers have free access to cheap fresh chili. Since domestic industry finds it hard to import fresh chili, they just import dried chili instead. As I mentioned earlier, Indonesia’s food manufacturing industry is one of the largest and fastest growing.

Well, if the problem is low international prices, the problems of chili farmers are clearly far more fundamental than imports or the UUCK. Domestic fresh chili, despite farmers complaining about low prices, is still relatively more expensive than abroad. This remains problematic even amid the chili import ban. This shows domestic problems that are more fundamental than blaming imports.

Additionally, Muflikh et al. (2021) state:

The volatility is caused mainly by seasonal production, unorganised market governance, and consumer preference for fresh chillies.

In other words, Indonesia’s chili price problems seem to be largely domestic issues.

The UUCK mystery

Regarding the UUCK: chili price volatility has been happening for a long time, as have fresh chili import regulations and high dried chili imports. Fresh chili has been protected all along, and since 2016 there have been no fresh chili imports at all. Meanwhile, dried chili has never had import restrictions. What’s supposed to become “more freely imported” if it’s already free?

This means it’s hard to blame the UUCK, at least for now. We may need to see 2021 data to clearly blame the UUCK, or examine implementing regulations that specifically target fresh and dried chili imports. The UUCK is indeed a controversial law. There’s much we can criticize, but we should do so with a cool head and on target. Especially sharing headline screenshots without citing sources – that should be avoided.

I’ll stop here for now. I know quite a bit about international trade, but I’m no chili expert. Hopefully this post motivates those who want to advance agriculture, especially chili farming, in Indonesia. As I said, there seems to be room for a rigorous study on chili in Indonesia, which could make for a very interesting thesis topic. Even better if you can do fieldwork. Hehe.

That’s it for this post. Thanks for reading!

Updates

Ban on curly chili imports?

I’ve embedded some Twitter responses about chili in this section. If there’s anything interesting I’ll keep updating.

Hrg cabe dlm negeri lg hancur, hrg hancur sdh berbulan², klo pun dipanen malah petani tambah merugi. Sebagian petani membiarkan cabenya membusuk dilahan.

Masyarakat konsumsi cabe segar jng cabe kering, krn kita kalah bersaing dng hrg cabe kering impor.

Mari kita diskusikan ini. https://t.co/dZD3SwSPr7 pic.twitter.com/nZcyT4QfBe
— Petani Rasional (@PetaniRasional) August 28, 2021

If the price difference is that large, even those who love fresh chili might choose dried. Not just industry – household consumers might also opt for dried chili if fresh chili prices are high.

The solution: ban dried chili imports too? Maybe we’ll discuss that in another post.

Tujuan impor cabe kering yg harusnya untuk kebutuhan industri ((( KONON KATANYA ))), tetapi faktanya bocor sampai beredar dipasaran tradisional, ditambah lg petani dlm negri lg panen raya. Dampaknya harga cabe hancur dalam tempo waktu yg lama, petani menjerit. pic.twitter.com/coalvk1m7L
— Petani Rasional (@PetaniRasional) August 26, 2021

Grow other commodities

The tweet below is also interesting: grow commodities that can be exported. Do your own market research and forget the government. Of course, not all farmers can do this, and it doesn’t mean we should stop criticizing government agricultural programs.

Mas,
Aku tanam Lemon California, walau gak begitu banyak. Di bulang juni-juli kemarin juga kena badai harga. Di petani anjlok sampai 1500-2000 perkilo. Sedang harga dipasaran tetap sama. Kenapa gak dipasarkan langsung? Mana ada waktu dan tenaganya. Belum lagi biayanya untuk itu.
— Calon Ketua RT (@bukanboslagi) August 29, 2021

Subsidy difficulties and forming farmer groups

Poorly targeted subsidies seem to be par for the course in countries with bad bureaucracies. This tweet provides a more detailed account of the mess around fertilizer subsidies, from difficulty accessing them (hard to join farmer groups) to resale at market prices (instead of using the subsidized fertilizer, they resell it like rent seekers).

Pupuk subsidi itu hanya diperoleh kelompok tani. Jgn lupa masih sgt banyak petani yg tdk masuk kelompok tani. Bentuk kelompok tani jg gak mudah di daerah.
Jd distribusi pupuk subsidi gak semulus yg dibayangkan. Kacaunya anggota kel tani jual balik pupuk subsidi dgn harga pasar.
— Andreas S Kurniawan (@ansyaku) August 30, 2021

Muflikh, Y. N., Smith, C., Brown, C., & Aziz, A. A. (2021). Analysing price volatility in agricultural value chains using systems thinking: A case study of the Indonesian chilli value chain. Agricultural Systems, 192, 103179. https://doi.org/https://doi.org/10.1016/j.agsy.2021.103179 ↩︎

Awesome Stata 17 Upgrade: Trying Pystata by using PPML on USITC and WITS database

Sun, 22 Aug 2021 20:39:04 +1000

Finally got my hand on Stata 17. When I got my email about Stata 17, I was very excited on the new PyStata feature. It allows for calling python from Stata and calling Stata from Python. There are many other cool upgrades on Stata 17 especially for Macroeconometrician, but the PyStata is especially interesting for me!

It doesn’t take long for me to try using PyStata on my Jupyter Lab. While using Python on Stata is certainly super interesting, I of course like better using Stata on Jupyter. I write things from Jupyter (like this post), so running things straight from Jupyter would be super awesome. While there are many things we can do with Python, it’s still hard for me to say goodbye to Stata. For one, I use packages like Levinsohn-Petrin algorithm for my research. Moreover, Stata still has plenty of familiarity for my own compared with Pandas. Lastly, most of my colleagues don’t even speak python or r, so Stata remains to be my main tool for collaboration.

So this post is dedicated to myself being super excited to Trying PyStata for the first time! As a trade economist, what’s best but to start with some trade data? In this post, I take data from the internet with Pandas, then transfer it to Stata database (you know, the .dta), and then run a PPML a la Silva and Tenreyro. Yes there is a PPML package for Python called GME (I even blogged about it) but come on I am having fun with PyStata here!

There is of course a better starting place to read if you really want to go on with playing with your PyStata, but I hope this post is fun and useful enuff four you.

Since this post is all about using Stata command in Python environment (and not the other way around), I mainly use Stata Magic. Maybe if I manage to use Mata Magic and PyStata Magic I will blog about it as well but the Stata Magic is what I am super excited about for now.

Okay lets go on to the show!

Installing PyStata

obviously the first thing you should do is to have Stata 17 installed on your computer. Remember where it is installed (it will be used later) but if you just click next like I do then it’s gonna be in C:/Program Files/Stata17/. When we’re done installing Stata 17, then all we need to do is to install PyStata in our env using

pip install --upgrade --user stata_setup

Then we’re good to go!

Collecting the data (in Pandas)

I am using publicly available data for this post so everyone can try this. No, not car, don’t worry.

Since this will be about gravity regression, I will need gravity data. Let’s get it from USITC Dynamic Gravity Dataset for our trial here. Be mindful of the size, this dataset is suuuuuuuuuper huge!

weqs=pd.read_csv('https://www.usitc.gov/data/gravity/dgd_docs/release_2.1_2000_2016.zip')
weqs.columns # checking its variables

Index(['year', 'country_o', 'iso3_o', 'dynamic_code_o', 'country_d', 'iso3_d',
'dynamic_code_d', 'colony_of_destination_ever', 'colony_of_origin_ever',
'colony_ever', 'common_colonizer', 'common_legal_origin', 'contiguity',
'distance', 'member_gatt_o', 'member_wto_o', 'member_eu_o',
'member_gatt_d', 'member_wto_d', 'member_eu_d', 'member_gatt_joint',
'member_wto_joint', 'member_eu_joint', 'lat_o', 'lng_o', 'lat_d',
'lng_d', 'landlocked_o', 'island_o', 'region_o', 'landlocked_d',
'island_d', 'region_d', 'agree_pta_goods', 'agree_pta_services',
'agree_fta', 'agree_eia', 'agree_cu', 'agree_psa', 'agree_fta_eia',
'agree_cu_eia', 'agree_pta', 'capital_const_d', 'capital_const_o',
'capital_cur_d', 'capital_cur_o', 'gdp_pwt_const_d', 'gdp_pwt_const_o',
'gdp_pwt_cur_d', 'gdp_pwt_cur_o', 'pop_d', 'pop_o', 'hostility_level_o',
'hostility_level_d', 'common_language', 'polity_o', 'polity_d',
'sanction_threat', 'sanction_threat_trade', 'sanction_imposition',
'sanction_imposition_trade', 'gdp_wdi_cur_o', 'gdp_wdi_cap_cur_o',
'gdp_wdi_const_o', 'gdp_wdi_cap_const_o', 'gdp_wdi_cur_d',
'gdp_wdi_cap_cur_d', 'gdp_wdi_const_d', 'gdp_wdi_cap_const_d'],
dtype='object')

Since this is a test, I will use just some of the variable. First, I will only use Indonesia as a country of origin. Secondly, I will only use cross-section in the year 2016. Next, only several variable will be used, namely, export partner’s population, GDP and distance, as well as FTA with Indonesia. Obviously there are more stuff needs to be added if you want to do a real gravity analysis but this is just a test come on.

grav=weqs.query('country_o == "Indonesia"') # take only Indonesia as a country of origin
grav=grav.query('year==2016') # take only year 2016
grav=grav[['country_d','pop_d', 'agree_fta', 'gdp_wdi_cur_d', 'distance']] # take some variables only
grav=grav.rename(columns={'country_d':'Partner'})
grav.head() # check content

	Partner	pop_d	gdp_wdi_cur_d	distance
455937	Aruba	0.104822	2.646927e+09	18924.247763
455954	Afghanistan	NaN	1.936264e+10	6158.679225
455971	Angola	28.813463	1.011239e+11	10483.002562
455988	Anguilla	0.014764	NaN	18167.048245
456005	Aland Islands	NaN	NaN	10410.128664

Now that i have the gravity variables, let’s get the trade data. I will need Indonesia’s export value to partners as well as their import tariff. Again, we normally need to do more rigorous check on the export and tariff, but since this is just a test, I will use total export as well as MFN weighted average tariff.

To get from WITS i use world_trade_data package in python.

import world_trade_data as wits

# Grabbing Indonesia's export value to partners in 2016
exports = wits.get_indicator('XPRT-TRD-VL', reporter='idn',partner='all', year='2016',product='Total')

exports=exports.reset_index()
exports=exports[['Partner','Value']] # Take only necessary columns
exports=exports.rename(columns={'Value':'Export_Value'}) # rename to merge later

# Grabbing tariff from Indonesia's partners
tariff=wits.get_indicator('MFN-WGHTD-AVRG',reporter='all',year='2016',datasource='tradestats-tariff',product='Total')
tariff=tariff.reset_index()
tariff=tariff[['Reporter','Value']] # take only necessary columns
tariff=tariff.rename(columns={"Reporter": "Partner","Value":"Tariff"}) # rename to merge later

trade=pd.merge(exports,tariff,how='left',on='Partner') # merge the two dataset

now that I have my trade,tariff and gravity data, lets merge them to get the final data to feed to our Stata PPML.

idntrade=pd.merge(grav,trade,how='left',on='Partner')
idntrade['Export']=idntrade['Export_Value']/1000000
idntrade.head()

	Partner	pop_d	gdp_wdi_cur_d	distance	Export_Value	Tariff	Export
0	Aruba	0.104822	2.646927e+09	18924.247763	878.469	10.158305	0.000878
1	Afghanistan	NaN	1.936264e+10	6158.679225	16220.103	NaN	0.016220
2	Angola	28.813463	1.011239e+11	10483.002562	57713.955	9.379807	0.057714
3	Anguilla	0.014764	NaN	18167.048245	NaN	NaN	NaN
4	Aland Islands	NaN	NaN	10410.128664	NaN	NaN	NaN

Well, obviously we can merge and cook our data ini Stata as well but as you can see it’s kinda faster to do it in Pandas unless I need to generate anything new.

Run a PPML in Stata

Now that the data’s ready, we can begin running PPML in Stata! Now, we need to call Stata to our Python environment with the import command. This is where we need our Stata location.

import stata_setup
stata_setup.config("C:/Program Files/Stata17/", "mp")

 ___ ____ ____ ____ ____ ©
/__ / ____/ / ____/ 17.0
___/ / /___/ / /___/ MP—Parallel Edition
Statistics and Data Science Copyright 1985-2021 StataCorp LLC
StataCorp
4905 Lakeway Drive
College Station, Texas 77845 USA
800-STATA-PC https://www.stata.com
979-696-4600 stata@stata.com

Look at how georgeous Stata logo is.

Next, we load the dataset to Stata! I use Stata magic to do this. there is also a possibility calling Stata with API like this one and this one, but in Jupyter, using magic is somewhat easier for me.

Now lets call the data and describe it.

%%stata -d idntrade -force
describe

Contains data
Observations: 255
Variables: 8
-------------------------------------------------------------------------------
Variable Storage Display Value
name type format label Variable label
-------------------------------------------------------------------------------
Partner str45 %45s
pop_d double %10.0g
agree_fta double %10.0g
gdp_wdi_cur_d double %10.0g
distance double %10.0g
Export_Value double %10.0g
Tariff double %10.0g
Export double %10.0g
-------------------------------------------------------------------------------
Sorted by:
Note: Dataset has changed since last saved.

Wooohoo cool! Let’s try summarize

%%stata
su

 Variable | Obs Mean Std. dev. Min Max
-------------+---------------------------------------------------------
Partner | 0
pop_d | 183 42.87855 150.9889 .005152 1403.5
agree_fta | 255 .0588235 .2357568 0 1
gdp_wdi_cu~d | 206 5.25e+11 2.26e+12 3.65e+07 1.87e+13
distance | 253 10594.37 4663.103 951.1062 19190.28
-------------+---------------------------------------------------------
Export_Value | 192 676920.4 2420380 .101 1.68e+07
Tariff | 119 6.739741 4.004538 0 18.73486
Export | 192 .6769204 2.42038 1.01e-07 16.78559

Awesome!! Now let’s do our ppml command.

%%stata
ppml Export Tariff pop_d agree_fta gdp_wdi_cur_d distance

note: checking the existence of the estimates
WARNING: Tariff has very large values, consider rescaling or recentering
WARNING: pop_d has very large values, consider rescaling or recentering
WARNING: gdp_wdi_cur_d has very large values, consider rescaling or recenterin
> g
WARNING: distance has very large values, consider rescaling or recentering
note: starting ppml estimation
note: Export has noninteger values
Iteration 1: deviance = 118.6375
Iteration 2: deviance = 101.5413
Iteration 3: deviance = 101.0093
Iteration 4: deviance = 101.0082
Iteration 5: deviance = 101.0082
Number of parameters: 6
Number of observations: 114
Number of observations dropped: 0
Pseudo log-likelihood: -102.58858
R-squared: .75786551
------------------------------------------------------------------------------
| Robust
Export | Coefficient std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
Tariff | -.0906943 .0542772 -1.67 0.095 -.1970757 .015687
pop_d | -.0002608 .0004712 -0.55 0.580 -.0011843 .0006626
agree_fta | 2.28588 .5362387 4.26 0.000 1.234871 3.336888
gdp_wdi_cu~d | 2.04e-13 3.58e-14 5.69 0.000 1.33e-13 2.74e-13
distance | -.0000199 .000065 -0.31 0.760 -.0001473 .0001075
_cons | -.488918 .7264721 -0.67 0.501 -1.912777 .9349412
------------------------------------------------------------------------------
Number of regressors dropped to ensure that the estimates exist: 0
Option strict is off

woohoo! The result is also great considering how under-specified the model is! High $R^2$ , Tariff is negative and strong, FTA and GDP are both highly important in determining foreign demand!

Now one thing I can’t quite work out is to extract the information from the regression back to pandas. I usually just go with outreg and do it manually, but it’s no fun if I can’t integrate the regression result in my notebook.

What I can do is to export the result to csv and read it thru Pandas.

%%stata
eststo
esttab using hehe.csv, plain replace wide

. eststo
(est1 stored)
. esttab using hehe.csv, plain replace wide
(output written to hehe.csv)
.

dff=pd.read_csv('hehe.csv')
dff

	Unnamed: 0	est1	Unnamed: 2
0	NaN	b	t
1	Tariff	-.0906943	-1.670947
2	pop_d	-.0002608	-.5535483
3	agree_fta	2.28588	4.262802
4	gdp_wdi_cur_d	2.04e-13	5.688106
5	distance	-.0000199	-.3056529
6	_cons	-.488918	-.6730031
7	N	114	NaN

The result is not the best so this is kinda bummer. But at least I have something. haha. There may be a better way to do this but let me look for it and post it later on. If you have a suggestion please mention me on twitter!

Anyway, that’s it for now. I hope this post is enjoyable!

Question 4 Tutorial 1 ECON3102 2021

Sun, 01 Aug 2021 17:32:43 +1000

About this post

This post is created to quickly demonstrate one of the possible answer of question 4 of tutorial 1 of ECON3102. I chose to use r’s WDI to show it but any other software and any other source of data can be used obviously. The reason why I use WDI is because it is convinient and easy to show some graph. If we use World Development Indicators we don’t even need to download it. Just use its graph function which is super convinient.

some packages:

library('WDI')
library('ggplot2')
library('tidyverse')
## -- Attaching packages --------------------------------------- tidyverse 1.3.1 --
## v tibble 3.1.2 v dplyr 1.0.7
## v tidyr 1.1.3 v stringr 1.4.0
## v readr 1.4.0 v forcats 0.5.1
## v purrr 0.3.4
## -- Conflicts ------------------------------------------ tidyverse_conflicts() --
## x dplyr::filter() masks stats::filter()
## x dplyr::lag() masks stats::lag()
library('ggdark')
library('patchwork')

A bit of aesthetic:

graph<-ggplot()+
dark_theme_bw(base_size=16)+
theme(text = element_text(color='#e0e0e0'),
panel.border = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
axis.line = element_line(colour = "#e0e0e0"))
## Inverted geom defaults of fill and color/colour.
## To change them back, use invert_geom_defaults().

The question itself consist of two parts. Students are to find, download and sketch numbers for two countries: one is developed one while the other is an emerging economy. In this example, I use Indonesia and Australia.

There are two sub-questions in this exercise. The first one is to plot real GDP and show a bit of a business cycle (which I will do by showing GDP growth)

Question One

Real GDP

Quickly browse website of World Development Indicators to get all keys needed for the WDI package.

let’s check out real GDP.

realgdp<-as_tibble(WDI(indicator = 'NY.GDP.MKTP.KD',country = c('IDN','AUS')))
realgdp
## # A tibble: 122 x 4
## iso2c country NY.GDP.MKTP.KD year
## <chr> <chr> <dbl> <int>
## 1 AU Australia 1.45e12 2020
## 2 AU Australia 1.45e12 2019
## 3 AU Australia 1.42e12 2018
## 4 AU Australia 1.38e12 2017
## 5 AU Australia 1.35e12 2016
## 6 AU Australia 1.31e12 2015
## 7 AU Australia 1.28e12 2014
## 8 AU Australia 1.25e12 2013
## 9 AU Australia 1.22e12 2012
## 10 AU Australia 1.17e12 2011
## # ... with 112 more rows

Which only shows top 10 which contain Aussie. If we plot ’em:

graph+geom_line(data=realgdp, aes(x=year, y=NY.GDP.MKTP.KD,color=country),size=1.1)+
labs(title = 'Fig 1. Real GDP, Indonesia and Australia, 1960 - 2020',caption='source: World Development Indicators')+
ylab('2010 USD')+geom_vline(xintercept=c(2009,1998,1983))

Real GDP are going up for both countries. Note that, while generally the graph goes up, but it has some ‘brakes’ for a short while. There are 1980s oil crisis which apparently struck Australia a bit more. There’s also something in 90s that hit Aussies. Asian Financial Crisis in 1998 definitely crushed Indonesian economy. The GFC do affect Australia a bit but not as bad as the northern hemisphere.

Also, Indonesia seems to be able to catch up with Aussie (note the relatively smaller gap in the late 2010s). Is it a converge? It remains to be seen.

Also, f#ck COVID-19.

GDP growth

While I spoke a bit about business cycle in some points, seeing GDP growth give us better picture.

First we get GDP growth and calculate arithmatic mean of them. (for the sake of the graph. You should really use geometric mean if you want to calculate mean growth rate.)

gdpg<-as_tibble(WDI(indicator = 'NY.GDP.MKTP.KD.ZG',country = c('IDN','AUS'),start=1961))
growth<-tibble(
country = c('Australia','Indonesia'),
mean_growth = c(mean(gdpg$NY.GDP.MKTP.KD.ZG[gdpg$country=='Australia']),
mean(gdpg$NY.GDP.MKTP.KD.ZG[gdpg$country=='Indonesia']))
)
growth
## # A tibble: 2 x 2
## country mean_growth
## <chr> <dbl>
## 1 Australia 3.38
## 2 Indonesia 5.13
graph+geom_line(data=gdpg, aes(x=year, y=NY.GDP.MKTP.KD.ZG,color=country),size=1.1)+
labs(title = 'Fig 2. GDP growth, Indonesia and Australia, 1960 - 2020',caption='source: World Development Indicators')+
ylab('%')+
geom_hline(yintercept = 0) +
geom_hline(yintercept = as.numeric(growth[1,2])) +
geom_hline(yintercept = as.numeric(growth[2,2])) +
geom_vline(xintercept = 2009) +
geom_vline(xintercept = 1998)

Fig 2 shows that the growth is not a smooth line: There are ups and downs. You can see how bad 1998 AFC does to Indonesia, while GFC decrease both countries’ growth but not much.

Question 2

Collect the data for nominal GDP (current year) and its four main types of expenditures: household consumption, government consumption, gross fixed capital formation, and net exports. This is a bit annoying because I can’t find nex export in World Development Indicators.

indo<-as_tibble(WDI(indicator = c('NE.CON.PRVT.ZS','NE.CON.GOVT.ZS','NE.GDI.FTOT.ZS',
'NE.EXP.GNFS.CN','NE.IMP.GNFS.CN','NY.GDP.MKTP.CN'),country = 'IDN'))
ausi<-as_tibble(WDI(indicator = c('NE.CON.PRVT.ZS','NE.CON.GOVT.ZS','NE.GDI.FTOT.ZS',
'NE.EXP.GNFS.CN','NE.IMP.GNFS.CN','NY.GDP.MKTP.CN'),country = 'AUS'))
indoo<-tibble(
year=indo$year,
C=indo$NE.CON.PRVT.ZS,
I=indo$NE.GDI.FTOT.ZS,
G=indo$NE.CON.GOVT.ZS,
NX=(indo$NE.EXP.GNFS.CN-indo$NE.IMP.GNFS.CN)/indo$NY.GDP.MKTP.CN*100
)
indooo<-indoo %>% pivot_longer(!year,names_to='variable',values_to='value')
ausii<-tibble(
year=ausi$year,
C=ausi$NE.CON.PRVT.ZS,
I=ausi$NE.GDI.FTOT.ZS,
G=ausi$NE.CON.GOVT.ZS,
NX=(ausi$NE.EXP.GNFS.CN-ausi$NE.IMP.GNFS.CN)/ausi$NY.GDP.MKTP.CN*100
)
ausiii<-ausii %>% pivot_longer(!year,names_to='variable',values_to='value')
gindo<-graph+geom_line(data=indooo,aes(x=year,y=value,color=variable),size=1)+geom_vline(xintercept=2009)+geom_vline(xintercept=1998)+dark_theme_bw(base_size = 12)+labs(title = 'Indonesia')+geom_hline(yintercept=0)
gausi<-graph+geom_line(data=ausiii,aes(x=year,y=value,color=variable),size=1)+geom_vline(xintercept=2009)+geom_vline(xintercept=1998)+dark_theme_bw(base_size=12)+labs(title = 'Australia')+geom_hline(yintercept=0)
gindo+gausi+plot_layout(ncol = 2, guides = "collect")+ plot_annotation(
title = 'Fig 3. Component of GDP by consumption, Indonesia and Australia, 1960-2020',
caption = 'Source: World Development Indicators'
)

There are a few point we can discuss here. First up, while both countries generally export very similar things (natural resources), Australia runs long term trade deficit, while Indonesia mostly has trade surplus. Consequently, trade has higher GDP component in Indonesia.

In the context of developed vs emerging, it is generally true that developed countries have higher G. Make sense given their better capacity to tax and generally richer people with more population sit on the upper-ish tax brackett. In terms of combined private-government consumption, Australia is higher mainly because its government spend more (as a % of GDP).

Lastly, Indonesia’s GDP structure change a bit in the late 60s to late 90s due to its govermental transition. That is, consumption’s fraction of GDP goes down as investment go up. This is also a pivotal moment where Indonesia starts to grow faster (see fig 2).

One thing to note is while some countries may follow similar pattern, country’s economic growth and business cycle can be quite nuance and ungeneralisable. It depends on plethora of things: exposure to international market, political regime, etc. Some general insight is very useful especially if we are talking about small countries with high international exposure, but always be careful, be mindful of local wisdom and take everything with a grain of salt.

Try China vs US, for example. Will be fun.

New update of the recent second wave (or third?) of COVID-19 in Indonesia

Wed, 28 Jul 2021 10:18:28 +1000

It has been a while since i last wrote about COVID-19. Today i’d like to check out Indonesia’s statistic on COVID-19, especially since it seem to get worse these days, unfortunately, and so many people talked about possibility of the government intentionally undertest to push down new cases at the cost of human lives.

Penambahan kasus Covid-19 harian cenderung menurun. Hal ini terjadi seiring dengan turunnya jumlah pemeriksaan secara signifikan. Masih terlalu dini untuk menyimpulkan bahwa gelombang Covid-19 telah terkendali. #Humaniora #AdadiKompas @aik_arif https://t.co/eYvloMmpIC
— Harian Kompas (@hariankompas) July 21, 2021

I rely heavily on Our World in Data ¹ which give free access of COVID-19 data.

Grab the data and show 6 tops

url='https://covid.ourworldindata.org/data/owid-covid-data.csv' # simpan url
df=pd.read_csv(url, parse_dates=['date']) # download dari url. parse_dates untuk menjadikan kolom date jadi tipe waktu
df.head(6) # menampilkan 10 baris paling atas

	iso_code	continent	location	date	total_cases	new_cases	new_cases_smoothed	total_deaths	new_deaths	new_deaths_smoothed	...	extreme_poverty	cardiovasc_death_rate	diabetes_prevalence	female_smokers	male_smokers	handwashing_facilities	hospital_beds_per_thousand	life_expectancy	human_development_index	excess_mortality
0	AFG	Asia	Afghanistan	2020-02-24	1.0	1.0	NaN	NaN	NaN	NaN	...	NaN	597.029	9.59	NaN	NaN	37.746	0.5	64.83	0.511	NaN
1	AFG	Asia	Afghanistan	2020-02-25	1.0	0.0	NaN	NaN	NaN	NaN	...	NaN	597.029	9.59	NaN	NaN	37.746	0.5	64.83	0.511	NaN
2	AFG	Asia	Afghanistan	2020-02-26	1.0	0.0	NaN	NaN	NaN	NaN	...	NaN	597.029	9.59	NaN	NaN	37.746	0.5	64.83	0.511	NaN
3	AFG	Asia	Afghanistan	2020-02-27	1.0	0.0	NaN	NaN	NaN	NaN	...	NaN	597.029	9.59	NaN	NaN	37.746	0.5	64.83	0.511	NaN
4	AFG	Asia	Afghanistan	2020-02-28	1.0	0.0	NaN	NaN	NaN	NaN	...	NaN	597.029	9.59	NaN	NaN	37.746	0.5	64.83	0.511	NaN
5	AFG	Asia	Afghanistan	2020-02-29	1.0	0.0	0.143	NaN	NaN	0.0	...	NaN	597.029	9.59	NaN	NaN	37.746	0.5	64.83	0.511	NaN

6 rows × 60 columns

I am not super familiar with its variable. So let’s check them out with df.columns.

df.columns # untuk panggil list dari nama-nama variabel

Index(['iso_code', 'continent', 'location', 'date', 'total_cases', 'new_cases',
'new_cases_smoothed', 'total_deaths', 'new_deaths',
'new_deaths_smoothed', 'total_cases_per_million',
'new_cases_per_million', 'new_cases_smoothed_per_million',
'total_deaths_per_million', 'new_deaths_per_million',
'new_deaths_smoothed_per_million', 'reproduction_rate', 'icu_patients',
'icu_patients_per_million', 'hosp_patients',
'hosp_patients_per_million', 'weekly_icu_admissions',
'weekly_icu_admissions_per_million', 'weekly_hosp_admissions',
'weekly_hosp_admissions_per_million', 'new_tests', 'total_tests',
'total_tests_per_thousand', 'new_tests_per_thousand',
'new_tests_smoothed', 'new_tests_smoothed_per_thousand',
'positive_rate', 'tests_per_case', 'tests_units', 'total_vaccinations',
'people_vaccinated', 'people_fully_vaccinated', 'new_vaccinations',
'new_vaccinations_smoothed', 'total_vaccinations_per_hundred',
'people_vaccinated_per_hundred', 'people_fully_vaccinated_per_hundred',
'new_vaccinations_smoothed_per_million', 'stringency_index',
'population', 'population_density', 'median_age', 'aged_65_older',
'aged_70_older', 'gdp_per_capita', 'extreme_poverty',
'cardiovasc_death_rate', 'diabetes_prevalence', 'female_smokers',
'male_smokers', 'handwashing_facilities', 'hospital_beds_per_thousand',
'life_expectancy', 'human_development_index', 'excess_mortality'],
dtype='object')

There’s a huge chunk of variable names! Musta been a super hard work collecting all the data. Shout out to Hannah Ritchie et al.

Aight now let’s check new cases! New cases tends to be volatile, especially if there’s seasonality in the data itself. It is quite common to see seasonality on daily data just because of weekends. Thankfully, there’s new_cases_smoothed which I imagine take into account seasonality by plotting 7-day rolling average. I only take Indonesian data for this post.

indo=df[["iso_code","date","new_cases","new_cases_smoothed"]].query('iso_code == "IDN"')

Plot time!

sns.lineplot(data=indo,x='date',y='new_cases')
sns.lineplot(data=indo,x='date',y='new_cases_smoothed')
plt.xticks(rotation=45)

(array([18322., 18383., 18444., 18506., 18567., 18628., 18687., 18748.,
18809.]),
[Text(0, 0, ''),
Text(0, 0, ''),
Text(0, 0, ''),
Text(0, 0, ''),
Text(0, 0, ''),
Text(0, 0, ''),
Text(0, 0, ''),
Text(0, 0, ''),
Text(0, 0, '')])

I try to make my own 7-day rolling average by copying codes from here

indo['cases_7day_ave'] = indo.new_cases.rolling(7).mean().shift(-3)
indo.head(10)

	iso_code	date	new_cases	new_cases_smoothed	cases_7day_ave
44074	IDN	2020-03-02	2.0	NaN	NaN
44075	IDN	2020-03-03	0.0	NaN	NaN
44076	IDN	2020-03-04	0.0	NaN	NaN
44077	IDN	2020-03-05	0.0	NaN	0.857143
44078	IDN	2020-03-06	2.0	NaN	2.428571
44079	IDN	2020-03-07	0.0	0.571	3.571429
44080	IDN	2020-03-08	2.0	0.857	4.571429
44081	IDN	2020-03-09	13.0	2.429	4.571429
44082	IDN	2020-03-10	8.0	3.571	9.285714
44083	IDN	2020-03-11	7.0	4.571	13.142857

Which confirms that new_cases_smoothed is indeed 7-day rolling average.

sns.lineplot(data=indo,x='date',y='new_cases')
sns.lineplot(data=indo,x='date',y='new_cases_smoothed')
sns.lineplot(data=indo,x='date',y='cases_7day_ave')
plt.xticks(rotation=45)
plt.legend(['new cases','new cases smoothed','7-day average bikinan sendiri'])
plt.ylabel('kasus')
plt.xlabel('tanggal')

Text(0.5, 0, 'tanggal')

A year and a half is a bit too long (dear god it’s already a year and a half??), so let’s cut it to just 2021.

indo2=indo.query('date>20210101') # ambil hanya setelah 1 Januari 2021
# lalu kita plot persis seperti di atas
sns.lineplot(data=indo2,x='date',y='new_cases')
sns.lineplot(data=indo2,x='date',y='new_cases_smoothed')
plt.xticks(rotation=45)
plt.legend(['new cases','new cases smoothed','7-day average bikinan sendiri'])
plt.ylabel('kasus')
plt.xlabel('tanggal')

Text(0.5, 0, 'tanggal')

Cases is indeed seem to go down even with the smoothed one. But is this because of undertesting? We can also see it from our dataset. We add positive rate to really make sure.

indo=df[["iso_code","date","new_tests","new_tests_smoothed",
 "new_cases","new_cases_smoothed","positive_rate"]].query('iso_code == "IDN"')
indo2=indo.query('date>20210101')

fig, axes = plt.subplots(1, 2, figsize=(18, 10))
fig.suptitle('Data tes baru dan positive rate Indonesia')
sns.lineplot(ax=axes[0],data=indo2,x='date',y='new_tests')
sns.lineplot(ax=axes[0],data=indo2,x='date',y='new_tests_smoothed')
axes[0].tick_params(labelrotation=45)
axes[0].legend(['new tests','new tests smoothed'])
axes[0].set_ylabel('tes baru')
axes[0].set_xlabel('tanggal')
axes[0].set_title('new cases')
sns.lineplot(ax=axes[1],data=indo2,x='date',y='positive_rate')
plt.xticks(rotation=45)
plt.ylabel('0-1')
plt.xlabel('tanggal')
axes[1].set_title('positive rate')

Text(0.5, 1.0, 'positive rate')

And yes test is indeed goes down. At the same time, positive rate seem to be trending down as well. This will depend on how testing is conducted in terms of selecting who gets to be tested and who’s not. We can be sure if we check hospitalisation and death. Unfortunately Indonesian hospitalisation number is non-existent in this dataset.

df.query('iso_code=="IDN"')[['weekly_icu_admissions','weekly_hosp_admissions']]

	weekly_icu_admissions	weekly_hosp_admissions
44074	NaN	NaN
44075	NaN	NaN
44076	NaN	NaN
44077	NaN	NaN
44078	NaN	NaN
...	...	...
44582	NaN	NaN
44583	NaN	NaN
44584	NaN	NaN
44585	NaN	NaN
44586	NaN	NaN

513 rows × 2 columns

On death (Dear God, bless all the lost souls and those who they left), situation is rather gloom.

indo=df[["iso_code","date","new_deaths","new_deaths_smoothed"]].query('iso_code == "IDN"')
indo2=indo.query('date>20210101')
sns.lineplot(data=indo2,x='date',y='new_deaths')
sns.lineplot(data=indo2,x='date',y='new_deaths_smoothed')
plt.xticks(rotation=45)
plt.legend(['kematian baru','kematian baru rerata bergerak 7 hari'])
plt.ylabel('kasus')
plt.xlabel('tanggal')

Text(0.5, 0, 'tanggal')

Judging from the death data, pandemic still far from over. Note that death may follow new cases, hence have a lag in its trending down. However, if we cannot trust test data, death data is also hard to be trusted. I think with unreliable data, it is hard to react on any news really, whether cases go up or down. It is hard to make a good case for the government, because people’s like: low cases: bad data! bad testing!. High cases: Government is stupid!

So yeah. I guess it is helping if we don’t overreact over the new cases because it might not reveal the true state of Indonesian COVID-19 Pandemic situation.

What about vaccination? Judging from all of our graph up there, new cases and positive rate shot up during June-ish. What happen during that month? Delta entrance? What kind of crowdy events happen during that time? What high mobility event took place during that date? The government might let high mobility events to take place amid vaccination program has started. So let me end this blog by posting vaccination speed between countries, including Indonesia.

Hannah Ritchie, Esteban Ortiz-Ospina, Diana Beltekian, Edouard Mathieu, Joe Hasell, Bobbie Macdonald, Charlie Giattino, Cameron Appel, Lucas Rodés-Guirao and Max Roser (2020) - “Coronavirus Pandemic (COVID-19)”. Published online at OurWorldInData.org. Retrieved from: ‘https://ourworldindata.org/coronavirus' [Online Resource] ↩︎

Why Indonesian Sugar is Expensive? A closer look on Indonesian trade policy (and Neraca Komoditas)

Sun, 27 Jun 2021 21:57:43 +1000

Sugar Prices

Sugar is one of the most important commodity in Indonesia. Sugar consumption in Indonesia is high. In 2018, Indonesian consumption of sugar reached 26.34 kg per capita, much higher than India, Vietnam and The Philippines¹. That is true even when Indonesian domestic sugar price is consistently higher if we compared to International market¹.

I downloaded data from CEIC Indonesia Premium Database² and FAO website³ and use it to plot the different prices of Indonesian sugar price in both traditional and modern market, and compare it with the world price.

url='secret!' # url is deleted because I can't freely distribute data from CEIC sorry
a=pd.read_csv(url) # taking data from google drive
a=pd.melt(a,id_vars='date',value_vars=['tprice','mprice','ip']) # melt it to fit my plot
# making a neat x axis
lol=pd.date_range('2016-07-01','2021-07-01',
 freq='6MS').strftime("%m/%Y").tolist()

# plotting
sns.lineplot(data=a,x='date',y='value',hue='variable',style='variable')
plt.xticks(ticks=lol,rotation=45)
plt.legend(labels=['traditional market','modern market','international market'])
plt.title('Sugar prices in Indonesian market and international market')
plt.xlabel('month/year')
plt.ylabel('USD/Kg')

Text(0, 0.5, 'USD/Kg')

As the graph above show, Indonesian domestic price of sugar is consistently higher than international market by about 3 times. One might wonder if international market price of sugar is so low, why don’t we import more of it? The obvious answer is because trade is restricted.

Sugar Trade

Indonesia import almost 70% of its sugar consumption from international market⁴. Not only Indonesians love its teh manis, but sugar is also play a crucial role in sugar refining industry. This industry is then supplying other industries like sugary drinks.

Previously I wrote that sugar-using industry also need to apply for sugar import to Ministry of Industry and Ministry of Trade. However according to Risetio, a person who once involved in this industry, sugar-using industry don’t import. Instead, they buy domestically from sugar refinery who do the import. Not sure why sugar-using industry don’t import directly. See thread below.

Balesnya di sini aja Med. Jadi kalau baca artikel tersebut, ada 1 kesalahan paling krusial. Perusahaan industri pengguna gula (industri makanan dan minuman) TIDAK meminta rekomendasi impor maupun persetujuan impor gula.
— Risetioz (@risetioz) July 10, 2021

The regulation that restrict trade of sugar is MOT regulations. To be precise, it’s Peraturan Menteri Perdagangan Nomor 14 Tahun 2020 which superseeds Peraturan Menteri Perdagangan Nomor 117 Tahun 2015. This regulation put a quota restriction on sugar. According to the regulation, imports are allowed only to supply domestic industry and control inflation. How much quota to import is determined in a coordination meeting held by Coordinating Ministry of Economy. I feel like this is basically Ministry of Agriculture asking for low quota to protect farmers, while Ministry of Industry asking for high quota to help sugar-using industries. But I might be wrong.

To get how much quoota to import next year, the flow chart looks like this:

sequenceDiagram participant A as Firms participant B as Ministry of Industry participant C as Surveyor participant D as Coordination meeting participant E as Ministry of Trade A->>B: Provide an import plan B->>C: Audit request C->>B: Report back B->>D: Argue for quota D->>E: Quota is settled

In short, the data Ministries are using come from (partially) sugar-using firms. Surveyor is needed to make sure that the asked quota is in-line to firms’ actual production capacity. A small firms asking for a big quota might turn out to be a reseller, a rent-seeker if you will, hence needs to be audited. This process happens every year.

Now that the quota has been settled, sugar refineries need to apply for the actual quota.

sequenceDiagram participant A as Firms participant B as Ministry of Industry participant C as Ministry of Trade participant D as Customs A->>B: Apply for a recommendation B->>A: Provide recommendation letter A->>C: Apply for the import quota C->>A: Provide the right to import A->>D: Use the right to import as a part of import documents

The process is rather tedious, involving many agencies and web applications. Ministries may only grant a fraction of the asked quota. This process is not very transparent, in the sense that you may not always know why your quota is cut, or how much quota is left for the taking¹. Additionally, some firms who get refined sugar intended for industry were also sometimes sold as white sugar in the consumer market which priced higher, creating an illegal trading¹.

The quota to import sugar for household is granted by the ministry only to 3 or 4 State-Owned Enterprises¹. These firms are also prone to badly time their imports and rent-seeking behavior can’t be excluded since these firms will always get the quota even when they don’t realize the quota¹.

If that is the case, let’s improve domestic sugar production! We should not rely on imported sugar, right? Well, that depends. Improving domestic sugar market is also a bit of a problem. Some of the on-farm problems include sufficient manpower, lack of monitoring of subsidized seeds and fertilizers, and inefficient mills¹. On-farm and off-farm is already highly subsidized, but productivity remain low. Not only the government spent a lot of money on improving productivity, it still turns out to be more expensive compared to international production of sugar.

New Solution?

There is this thing called ‘Neraca Komoditas’, which supposed to have an information on how much production and consumption of a commodity, including sugar. The difference between production and consumption on that data will be used to settle on how much quota to import. Setting the quota to import seems to be highly political. It is basically a battle between sugar producer and sugar consumer. It is clear that sugar is used by many industries under the Ministry of Industry which just finished a regulation which says it will guarantee a steady flow of industrial inputs.

Neraca Komoditas wants to cut a process of negotiating how much sugar is imported in the coordination meeting. Remember my impression about sugar producer vs sugar-using industries? The data for sugar production mainly comes from Ministry of Agriculture, while the data for sugar consumption comes from Ministry of Industry (like in the first diagram). When the two clashes, Ministry of Industry might accuse Ministry of Agriculture of inflating their figures, as Industry still have hard times getting sugar. With Neraca Komoditas as the one sole data which rule them all, this dispute is (hopefully?) no more. This is hypothesized to cut a bit of a red-tape as the coordination meeting may not last as long.

It will be much more interesting if Neraca Komoditas actually cuts the need to apply for recommendation letters. It is possible that the web portal SIINAS to actually remember how much quota firms planned for the import, the quota is set per firms instead of in aggregate. Firms can feel safer as ministries can no longer cut their proposed quota. It may reduce rent-seeking activity. It will be even better if ministries actually publish the data, the quota, and who gets how much. It will be easier for firms to check their actual allowance and for public to audit the system.

If the whole recommendation letters are still in place, then the red tape faced by industries seems to be staying. Neraca Komoditas might help bureaucrats in their debates of settling in a quota (remember, Neraca Komoditas is not only about sugar, but also a huge lists of ‘commodities’). But it seems like it won’t affect industries that much.

At the end of the day, it is still going to be the good ol’ quota. Neraca Komoditas might solve the data issue, but the old quota problem seems to not be addressed. If firms granted by the quota is still restricted, it won’t solve the problem. If how much quota left or how fast it is disbursed to the market is not clear, it won’t solve the problem. If industries which import industrial inputs will still mask their refined sugar to white crystal and do shadow trading, it won’t solve the problem.

There will be a material center for small business who can’t import their own inputs. Will it be another SOE? we’ll have to see.

It is easy to be pessimistic, but let’s give the benefit of the doubt to the government. Again.

What do you think? Please comment on my twitter!

Saputri, N. K., & Respatiadi, H. (2018). Policy reform to lower sugar prices in Indonesia. CIPS Indonesia. https://repository.cips-indonesia.org/media/270481-policy-reform-to-lower-sugar-prices-in-i-839081a1.pdf ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
CEIC. “Average Retail Price: Sugar: Local”. CEIC Indonesia Premium Database. CEIC. ↩︎
FAO. “ICE Future US: Sugar”. GIEWS FPMA. https://fpma.apps.fao.org/giews/food-prices/tool/public/#/dataset/international ↩︎
Amanta, F. (2021). The cost of Non-Tariff Measures on Food and Agriculture in Indonesia. CIPS Policy Paper, 36. https://www.cips-indonesia.org/cost-of-ntm ↩︎

The heterogenous impact of tariff and NTM on total factor productivity of Indonesian firms

Fri, 04 Dec 2020 21:12:42 +1100

Artikel ini telah diterima untuk diterbitkan di Bulletin of Indonesian Economic Studies (BIES) dan sedang mengalami proses edit. Artikel versi pre-edit dapat diakses di sini. Artikel ini telah dipresentasikan di Economics PhD Seminar di Crawford School of Public Policy. Slide presentasi dapat diakses di sini

Statistik ekspor impor Indonesia di 2018 dengan negara-negara RCEP

Sat, 28 Nov 2020 13:39:10 +1100

Just a bit of stats on RCEP. To connect it a bit with the Global Value Chain, I am using Stage of Production (SoP) UNCTAD:

	productcode	isgroup	nomenclaturecode	grouptype	productdescription
27	UNCTAD-SoP1	Yes	H0	Stages-Of-Processing	Raw materials
28	UNCTAD-SoP2	Yes	H0	Stages-Of-Processing	Intermediate goods
29	UNCTAD-SoP3	Yes	H0	Stages-Of-Processing	Consumer goods
30	UNCTAD-SoP4	Yes	H0	Stages-Of-Processing	Capital goods

Just a very long code to aggregate RCEP. Sorry too lazy to make it DRY. Skip these if you don’t need it to go straight to the stats.

The stats is:

	Trade Value	All goods	Raw materials	intermediate goods	Consumption goods	Capital Goods
0	export RCEP	$101,578,213.46	$29,583,083.81	$28,876,371.04	$32,692,137.50	$10,003,962.16
1	export World	$180,215,034.44	$45,863,376.29	$49,419,462.61	$68,578,923.17	$15,547,001.92
2	import RCEP	$125,215,957.00	$7,859,326.12	$40,495,720.49	$33,166,747.09	$42,937,992.48
3	import World	$188,711,171.62	$24,893,192.22	$61,575,534.29	$44,976,972.35	$56,497,488.96

This is, by the way, in 2018. There are RCEP countries in the World cuz i am too lazy to substract them. Also, I think it is useful to see how much Indonesians are trading with the rest of the world.

I wrote once on my blog about how important imports are for Indonesian production. It is indeed reflected in this 2018 stat where the fraction of consumption goods in Indonesian import is just mere:

23.83376244573637 %

which actually a decrease.

As for export, the other way around is true. That is, most exports are consumption goods. Even with this simple stat, it is so hard to convince Indonesian government that it is important to keep a free flow of goods not just export but also imports, to participate in the GVC. Indeed, increasing imports to help exports may seem very counter-intuitive for a country so used to exporting natural resources.

But of course, back to RCEP, what’s important is how big is the fraction of RCEP to Indonesia’s total trade.

	importance	All goods	Raw materials	intermediate goods	Consumption goods	Capital Goods
0	export	0.563650	0.645026	0.584312	0.476708	0.643466
1	import	0.663532	0.315722	0.657659	0.737416	0.759998

And as you can see, RCEP countries are indeed very important to Indonesia’s export and import. more than 70% Indonesia’s consumption goods and capital goods came from RCEP countries in 2018. Notably, Indonesian consumption goods export to RCEP countries is rather small, which makes sense because Indonesia’s export destination for consumption goods also consist of plenty of developed nations like US, which reflects the importance of the extended GSP tariff preference from the US. The small importance of raw materials import is probably due to agriculture stuffs such as beef, soybean and wheat which many come from the US and Latin America.

That’s all for this post and I hope it’s useful.

FDI, industry, and trade policy

Wed, 11 Nov 2020 21:17:11 +1100

This post briefly discusses FDI developments in Indonesia.

FDI Trends Since 1970

The chart below shows Indonesia’s net FDI inflows as a percentage of GDP. The data source is the World Bank’s World Development Indicators.

The chart is accompanied by a narrative from Linblad (2015). According to him, there are at least three major phases in Indonesia’s FDI story: before 1985 (the oil boom era), after 1985 through the crisis, and the reform period.

FDI and high-tech industry

Again using World Development Indicator data, I searched for indicators containing the keyword “high-tech,” and WDI returned these four indicators:

id name
----------------- ------------------------------------------------------------------------------------
NV.MNF.TECH.ZS.UN Medium and high-tech Industry (including construction) (% manufacturing value added)
TX.MNF.TECH.ZS.UN Medium and high-tech exports (% manufactured exports)
TX.VAL.TECH.CD High-technology exports (current US$)
TX.VAL.TECH.MF.ZS High-technology exports (% of manufactured exports)

I use the first one for the visualization below.

There doesn’t seem to be much of a link between FDI and high-tech manufacturing growth. In fact, it almost looks inverted: when FDI rises as a share of GDP, high-tech manufacturing actually declines. Of course, manufacturing also includes construction and CPO processing. Besides, FDI also flows into mining and services like e-commerce and banking. FDI may not be going into high-tech manufacturing at all – it may be going into industries with low high-tech value added. This makes sense: FDI flows into the most profitable industries, and when those industries grow, factors of production shift toward them. High-tech manufacturing growth dropped in 2005 and never recovered. The mid-2000s was the peak of the commodity boom and the era of so-called ‘sunset industries.’

Two Non-Tariff Measures (NTMs): TKDN and Export Ban

TKDN stands for Tingkat Komponen Dalam Negeri (Local Content Requirement). One of the most prominent TKDN cases involves smartphones.

The chart below shows realized FDI (source: BKPM) in the electronics, software, and IT industries. TKDN for smartphones became mandatory in 2016. We can see that FDI in these three industries suddenly surged after 2016.

<Figure size 504x360 with 0 Axes>

Below is the mineral export ban. In 2014, mineral exports were banned, though the ban was relaxed in 2017. Investment in base metals increased while mining of metal ores declined. Investors appear to have shifted into mineral processing due to the raw commodity export ban.

Text(2017-01-01 00:00:00, 300000, 'Relaxed')

How about FDI and industrial growth?

This time I use the World Bank Global Economic Monitor (GEM) to get Indonesia’s industrial production data. Since it’s in constant USD, I convert it to growth rates ($\Delta log$), though this is still not ideal.

id name
--------- -----------------------------------------------
IPTOTNSKD Industrial Production, constant US$
IPTOTSAKD Industrial Production, constant US$, seas. adj.
<function matplotlib.pyplot.show(*args, **kw)>

Industrial growth, as we can see above, was briefly negative in 2006. It has since stagnated at around 4-5%. There doesn’t seem to be any relationship with FDI. Again, it’s possible that FDI isn’t going into the manufacturing sector at all.

Indonesian tariff dynamics, 2000-2018

Thu, 05 Nov 2020 22:26:15 +1100

A couple of weeks ago, a colleague of mine, Deasy Pane from Bappenas, invited me for a discussion about Indonesia’s trade and investment situation policy. Indeed, that’s the beauty of zoom right? Collabs know no boundaries! With us was Donny and Chandra, also trade anthusiast at ANU. Some of Bappenas’ employee were also there.

In fact, one of Deasy’s Bappenas colleague asked us about Indonesian tariff dynamics. Indeed, Indonesian manufacturing import was a bit volatile during 2000s and early 2010s, which could be driven by changes in tariff and Indonesian overall protectionism dynamics. Commodity boom might also a factor here. Generally, protectionism in terms of tariff was declining during this time. However, the dynamics on overall protectionism is not so straight-forward if we include other measures, and in fact even in terms of tariff, it rises a little bit during Jokowi’s presidency.

Additionally, since my last post I was showing how to use World Bank’s API with Python,, might as well shows how to use WITS’ API this time around. To install:

pip install world_trade_data --upgrade

To show tariff regulation (instead of how it might change welfare), I use effective applied simple average tariff and MFN simple average tariff. Here’s what they are:

import pandas as pd
import world_trade_data as wits
tariff=wits.get_indicators('tradestats-tariff')
tariff=tariff.filter(['name','definition'])
tariff=tariff.loc[['AHS-SMPL-AVRG','MFN-SMPL-AVRG']]
#pd.set_option('display.max_colwidth', None)
tariff

	name	definition
indicatorcode
AHS-SMPL-AVRG	AHS Simple Average (%)	Effectively Applied Simple Average tariff (%)
MFN-SMPL-AVRG	MFN Simple Average (%)	Most Favored Nation ( (MFN) Simple Average tar...

Using the API

import pandas as pd
import datetime
#pd.set_option('display.max_rows', 20)
import seaborn as sns; sns.set()
import matplotlib.pyplot as plt
pd.options.display.max_rows=999
b=wits.get_products('tradestats-tariff')
b=b.loc[(b.grouptype=='Sector') & (
 b.index!='Total')].productdescription.values

# MFN tariff, weighted average
a=wits.get_indicator(indicator='MFN-SMPL-AVRG',
reporter='IDN',datasource='tradestats-tariff',
 year=['2000','2019'])
a=a.reset_index()
a=a.set_index('ProductCode')
c,d,e=a.loc[b[0:4]],a.loc[b[5:10]],a.loc[b[11:]]
c,d,e=c.reset_index(),d.reset_index(),e.reset_index()
a=a.reset_index()

# Effective applied tariff, weighted average
o=wits.get_indicator(indicator='AHS-SMPL-AVRG',
reporter='IDN',datasource='tradestats-tariff',
 year=['2000','2019'])
o=o.reset_index()
o=o.set_index('ProductCode')
p,q,r=o.loc[b[0:4]],o.loc[b[5:10]],o.loc[b[11:]]
p,q,r=p.reset_index(),q.reset_index(),r.reset_index()
o=o.reset_index()

For some reason, Indonesian tariff is missing in 2014 and 2015. If you see from all of my graph below, you will see a jump from 2013 to 2016. Seems like the API extract the year not as datetime object. It’s a bit of a pain to change it myself cuz then I need to add two NA rows per sectors and I am too lazy to do that.

There are 16 different sectors WITS is giving us, HS-based. I split them into three broader groups so the graph is much more readable. In total, there are 6 graphs in this post. Here’s the sectors:

array(['Animal', 'Vegetable', 'Food Products', 'Minerals', 'Fuels',
'Chemicals', 'Plastic or Rubber', 'Hides and Skins', 'Wood',
'Textiles and Clothing', 'Footwear', 'Stone and Glass', 'Metals',
'Mach and Elec', 'Transportation', 'Miscellaneous'], dtype=object)

Effective Applied Tariff, simple average

Three graphs below are the effective applied tariff. After 2008, Indonesia was very diligent in doing FTAs. Among them are ASEAN FTA driven (AFTA). The AFTA turned to plenty of AFTA+1 in 2010. Among them are AKFTA (with South Korea), ACFTA (China), AANZFTA (Australia and New Zealand), and AIFTA with India. Additionally There’s also IJEPA with Japan, perhaps the only bilateral one Indonesia had before IACEPA comes along.

These FTAs eventually lower effective applied tariff.

# FIGURES FOR AHS
plt.figure(figsize=(9,6))
sns.lineplot(data=p,x='Year',y='Value',hue='ProductCode')
plt.axvline('2016', color='black')
plt.axvline('2008', color='black')

<matplotlib.lines.Line2D at 0x165b71fc790>

Firstly, primary products. You can straight away see a far higher tariff in food producing sector compared to animal, vegetables, and minerals. The GFC clearly increased Indonesian food protectionism (as with other countries during that time). Overall, other sectors aren’t that highly protected by tariff, which hover around 2-4%. Same thing can’t be said for NTMs, however. But that’s for another post.

plt.figure(figsize=(9,6))
sns.lineplot(data=q,x='Year',y='Value',hue='ProductCode')
plt.axvline('2016', color='black')
plt.axvline('2008', color='black')

<matplotlib.lines.Line2D at 0x165bdbb2760>

In some manufacturing, you see a higher tariff. These sectors were highly protected before Asian Financial Crisis (AFC) in 1998. It was lowered after AFC, but also after GFC, interestingly. Indeed, bad times lead to good policy. Another interesting thing is that Indonesia was famed to have a premature deindustrialisation during 2005s if my memory serves. These are times where textiles got more protection, as you can see from the graph.

Lastly, these tariff shot up in 2016 (remember, 2014 and 2015 are missing). What happened? What huge event happened in 2015/2015?

plt.figure(figsize=(9,6))
sns.lineplot(data=r,x='Year',y='Value',hue='ProductCode')
plt.axvline('2016', color='black')
plt.axvline('2008', color='black')

<matplotlib.lines.Line2D at 0x165bdc9ef10>

On this group, it seems like in general tariff are progressing to the lower end. Especially after AFC and GFC, protection given to transportation is stripped up, but also in metals, machinery, electronics, and other equipments. These are also sectors that have large intensity of Global Value Chain (GVC). Make sense to lower the trade cost of these stuff as low as possible. Again, you can see a bit of a jump in 2016.

MFN, simple average

Here’s MFN! In general, we should expect a higher level of tariff using MFN. But the trends are pretty much the same: aside from food products, crisis reduce the overall tariff level. 2016, everything jumped up.

This time, let me hear more from you! Please do comment on this blog or mention me on twitter for a discussion!

# FIGURES FOR AHS
plt.figure(figsize=(9,6))
sns.lineplot(data=c,x='Year',y='Value',hue='ProductCode')
plt.axvline('2016', color='black')
plt.axvline('2008', color='black')

<matplotlib.lines.Line2D at 0x165bdd830a0>

plt.figure(figsize=(9,6))
sns.lineplot(data=d,x='Year',y='Value',hue='ProductCode')
plt.axvline('2016', color='black')
plt.axvline('2008', color='black')

<matplotlib.lines.Line2D at 0x165bdedc280>

plt.figure(figsize=(9,6))
sns.lineplot(data=e,x='Year',y='Value',hue='ProductCode')
plt.axvline('2016', color='black')
plt.axvline('2008', color='black')

<matplotlib.lines.Line2D at 0x165bde1e8b0>

Indonesia's export and import for COVID-19-related goods: July 2020

Sun, 01 Nov 2020 23:10:58 +1100

This is an updated figures on COVID-19-related goods export and import in Indonesia. I was once presenting about it in ANU Indonesia Project’s global seminar series, and I cited Indonesian monthly trade data until March 2020. WCO/WHO classification second edition was used to identify what are considered COVID-19-related goods.

This time around, I’ve got my hands on the trade data until July 2020, and I am using WCO/WHO third revision classification. Illustration is conducted in R codes. So let’s see what happens.

Exports are generally similar to last year’s

First, let’s take a look at export.

# Loading and cleaning the data
library(tidyverse)
library(lubridate)

expor<-read.csv('Book1.csv') # read the data
a<-colnames(expor) # add column names to a for easy rename
expor<-expor %>% # rename variable names
 rename(hs=a[1],deskripsi=a[2],tujuan=a[3],periode=a[4],kg=a[5],usd=a[6])

# make periode into time series
expor$periode<-as.Date(paste0(as.character(expor$periode), '01'),format='%Y%m%d')

# Make aggregation to all countries
exporall<-expor %>% group_by(hs,periode) %>% summarise((expor=sum(usd)))
a<-colnames(exporall)
exporall<-exporall %>% rename(usd=a[3])

# Make aggregation on WCO goods classification
wco<-read.csv('wco.csv') # adding wco classification
wco<-wco %>% distinct(hshs, .keep_all = TRUE) # drop duplicates
wco$hshs<-as.character(wco$hshs) #make hshs a string so i can join them
exporall$hshs<-substr(as.character(exporall$hs),start=1,stop=6) # taking the first 6 digits
final<-inner_join(exporall,wco,by="hshs")
finall<-final %>% group_by(desc,periode) %>% summarise(expor=sum(usd))
a<-colnames(finall)
finall<-finall %>% rename(usd=a[3])
finall$usd6<-finall$usd/1000000

# plot
ggplot(finall,aes(x=periode,y=usd6, color=desc)) +
 geom_line(aes(linetype=desc),size=1.1) +
 scale_x_date(date_labels="%b-%y",date_breaks = "2 month")+
 # theme_ipsum() +
 # theme(axis.text.x=element_text(angle=60, hjust=1)) +
 geom_vline(xintercept=as.Date('20200301',format='%Y%m%d')) +
 labs(x='tahun',y="Million USD",title="Indonesia's export of COVID-19 related goods",subtitle="BPS,Compiled by Pusdatin Kemenperin")

Apart from other consumables and PPE, Indonesia isn’t really comparatively advantageous in producing health equipments in general. PPE jump in February was the infamous skyrocketed export in masks due to demand surge by China and Singapore, the two countries experienced COVID-19 at the early stage. In March, Indonesia consulted to export ban on PPEs and other stuffs, but export was still quite strong amind commitment to buyers.

During the talk, I mentioned how Indonesia can be a powerhouse in making PPEs in a competitive way, provided the inputs. However, export ban seems to prevent this. It is also still possible to say production did increase but it was mostly consumed domestically.

Indonesian export value tends to reduced during the ramadhan season which happened in June in 2019, but May in 2020. The export ban and many other trade measures was revoked in June 2020, so it will be interesting to take a look at the numbers after July 2020.

Testing kits imports were raising, but most imports went down in July

More things can be said on imports.

As you can see, import plays a significant role in Indonesian health service even before COVID-19. While Indonesia can produce surgical gown competitively, capital-intensive PPE such as masks are still mostly imported. Moreover, Indonesia also imports a lot of other COVID-19-related goods quite extensively.

Indonesia’s first case was in early March. And as you can see, the number of imports shot up right after March, especially testing kits and PPEs. The trend weakend on May (ramadhan?) except for the testing kits import. Interestingly, coming July, all imports went down except for other medical equipments. I mentioned about revoked rules (yes, including for easing up imports for private enterprises), so this may reflect that. It could also reflect improvement on domestic production but I am not sure. Sadly domestic production numbers are a bit harder to get.

Notably, Imports are generally decreased in February (take this with a grain of salt but I am thinking Chinese New Year. China is a major source of import).

Despite its ability in improving growth and standard of living, trade has never been championed as the tool to improve production. On the countrary, countries are trying to look inward, including (especially?) Indonesia. I am quite sceptical Indonesian economy can crank-up essential production in a short time alone, but I highly hope I am wrong.

Here’s what covered in WCO Classification.

section	desc	hs	deskripsi
1	testing kit etc	38220010	Plates, sheets, film, foil & strip of plastics impregnated or coated with diagnostic reagents or laboratory
1	testing kit etc	38220020	Paperboard, cellulose wadding & web of cellulose fibres impregnated or coated with diagnostic reagents or laboratory
1	testing kit etc	38220030	Sterilisation indicator strips & tapes
1	testing kit etc	38220090	Diagnostic/laboratory reagents on a backing prepared diagnostic/laboratory reagents Oth than HS 3822.00.10 - 3822.00.30
1	testing kit etc	30021500	Oth blood fractions, whet/not modified/obtain by biotechnological process,put up in measured doses
1	testing kit etc	90278010	Oth instruments&apparatus,exposure meters
1	testing kit etc	90278030	Oth instruments&apparatus oth exposure meters,electricallyoperated
1	testing kit etc	90278040	Oth instruments&apparatus oth exposure meters,not electrically operated
1	testing kit etc	38210010	Prepared culture media for the development of microorganisms
1	testing kit etc	38210090	Other than prepared culture media for the development of microorganisms
2	PPE	48189000	Cellulose wadding or webs of cellulose fibres, oth thn hs code 4818.10.00 - 4818.50.00
2	PPE	63079030	Umbrella covers in pre-cut triangular form
2	PPE	63079040	Surgical masks
2	PPE	63079061	Safety harnesses safety suitable for industrial use
2	PPE	63079069	Safety harnesses suitable not for industrial use
2	PPE	63079070	Fans and handscreens
2	PPE	63079080	Laces for shoes,boots,corsets & the like
2	PPE	63079090	Oth made up articls excl umbrella covers/surgical masks/safety harnesses/fans&handscreens/laces shoes,boots,corsets
2	PPE	90200000	Breathing appliances&gas masks,excl protective masks having neither mechanical part nor replaceable filters
2	PPE	90049010	Corrective spectacles
2	PPE	90049050	Protective goggles
2	PPE	90049090	Spectacles & the like excl corrective, protective googles
2	PPE	39269010	Floats for fishing nets
2	PPE	39269020	Fans and handscreens, frames and handles therefor, and parts thereof
2	PPE	39269032	Plastic moulds with denture imprints
2	PPE	39269039	Hygienic, medical and surgical articles oth plastic moulds with denture imprints
2	PPE	39269041	Police shields
2	PPE	39269042	Protective masks for use in welding and similar work
2	PPE	39269044	Life saving cushions for the protection of persons falling from heights
2	PPE	39269049	Other safety and protective devices
2	PPE	39269053	Transmission or conveyor belts or belting
2	PPE	39269055	Plastic J-hooks or bunch blocks for detonators
2	PPE	39269059	Other articles for industrial uses
2	PPE	39269060	Point of poultry feeders
2	PPE	39269070	Padding for articles of apparel or clothing accessories
2	PPE	39269081	Cards for shoe lasts
2	PPE	39269082	Cards for prayer beads
2	PPE	39269089	Cards for jewellery or other small objects of personal adornment
2	PPE	39269091	Point of a kind used for grain storage
2	PPE	39269092	Empty capsules of a kind suitable for pharmaceutical use
2	PPE	39269099	Other articles of plastics & other materials of headings 39.01 to 39.14. oth than 3926.10.00-3926.90.92
2	PPE	40151100	Gloves, mittens and mitts for surgical, of vulcanised rubber other than hard rubber
2	PPE	40151900	Gloves, mittens and mitts for not surgical, of vulcanised rubber other than hard rubber
2	PPE	61161010	Divers’ gloves,knitted or crocheted, Impregnated,coated or covered with plastics or rubber
2	PPE	61161090	Other gloves, mittens and mitts, knitted or crocheted,impregnated, coated or covered with plastics or rubber
2	PPE	62160010	Protective work gloves, mittens and mitts
2	PPE	62160091	Other gloves, mittens and mitts,of wool or fine animal hair
2	PPE	62160092	Other gloves, mittens and mitts,of cotton
2	PPE	62160099	Other gloves, mittens and mitts,of oth textile material excwool or fine animal hair or cotton
2	PPE	65050010	Hats & oth headgear of a kind used for religious purposes,knitted/crocheted
2	PPE	65050020	Hair-nets of any materials
2	PPE	65050090	Oth hats&oth headgear,not for hair-nets/religious purposes,knitted/crochet
2	PPE	62114210	Garments for fencing or wrestling,of cotton,women’s/girl’s
2	PPE	62114220	Prayer cloaks, of cotton, women’s or girl’s
2	PPE	62114290	Other garments exc for fencing/wrestling/prayer cloaks, ofcotton, womenâ€™s or girl’s
2	PPE	40159010	Lead aprons, of vulcanised rubber other than hard rubber
2	PPE	40159020	Divers’ suits (wet suits), of vulcanised rubber other than hard rubber
2	PPE	40159090	Articles apparel&clothing accessories,oth thn 40151100-40159020 for all purposes, vulcanised rubber oth thn hard rubber
2	PPE	48185000	Articles of apparel and clothing accessories
2	PPE	62101011	Garments used for protection from chemical substances, radiation or fire, of fabrics of heading 56.02 or 56.03
2	PPE	62101019	Other protective work garment (exc used for protection fromchemical substance,radiation/fire),of fabric head 5602/5603
2	PPE	62101090	Garments of fabrics of heading 56.02 or 56.03, other than protective work garments
2	PPE	62104010	Menâ€™s or boysâ€™ garments, used for protection from fire
2	PPE	62104020	Menâ€™s or boysâ€™ garments, used for protection from chemicalsubstances or radiation
2	PPE	62104090	Other menâ€™s or boysâ€™ protective work garments (exc used forprotection from fire/chemical subtances/radiation)
2	PPE	62105010	Women’s or girls’ garments, used for protection from fire
2	PPE	62105090	Other women’s or girls’ protective work garments (exc usedfor protection from fire/chemical subtances/radiation)
2	PPE	40169911	Parts&access kind used for vehicles of Chapter 87 for vehicles of head 8702,8703, 8704 or 8705,oth thn weatherstripping
2	PPE	40169912	Parts and accessories of vulcanised rubber other than hard rubber, of a kind used for vehicles of heading 87.11
2	PPE	40169913	Weatherstripping of vulcanised rubber oth thn hard rubber, of a kind used on motor vehicles of heading 8702,8703 or 8704
2	PPE	40169915	Parts&accessories of vulcanised rubber oth thn hard rubber,of a kind used for vehicles of head 8709, 8713, 8715 or 87.16
2	PPE	40169916	Bicycle mudguards of vulcanised rubber other than hard rubber
2	PPE	40169917	Bicycle parts of vulcanised rubber other than hard rubber
2	PPE	40169918	Other bicycle accessories of vulcanised rubber other than hard rubber
2	PPE	40169919	Parts&accessories of a kind used for vehicles of Chapter 87 of vulcanised rubber exc hard rubber oth thn 4016.99.11-20
2	PPE	40169920	Parts and accessories of rotochutes of heading 88.04 of vulcanised rubber other than hard rubber
2	PPE	40169930	Rubber bands, of vulcanised rubber other than hard rubber
2	PPE	40169940	Wall tiles of vulcanised rubber other than hard rubber
2	PPE	40169951	Rubber rollers of vulcanised rubber other than hard rubber
2	PPE	40169952	Tyre mould bladders of vulcanised rubber other than hard rubber
2	PPE	40169953	Electrical insulator hoods of vulcanised rubber other than hard rubber
2	PPE	40169954	Rubber grommets and rubber covers for automotive wiring harnesses of vulcanised rubber other than hard rubber
2	PPE	40169959	Oth articles a kind used in machinery/mechanical/electric appliances,or for oth technical uses,oth thn 40169951-40169954
2	PPE	40169960	Rail pads of vulcanised rubber other than hard rubber
2	PPE	40169970	Structural bearings including bridge bearings of vulcanised rubber other than hard rubber
2	PPE	40169991	Table coverings of vulcanised rubber other than hard rubber
2	PPE	40169999	Other articles of vulcanised rubber other than hard rubber, n.e.c in 4016
3	sterilization product	22071000	Undenatured ethyl alcohol of an alcoholic strength by volume of 80 % vol or higher
3	sterilization product	22089010	Medicated samsu of an alcoholic strength by volume not exceeding 40 % vol
3	sterilization product	22089099	oth than samsu, arrack and bitters of an alcoholic strength by volume exceeding 114 % vol but not exceed 40 % vol
3	sterilization product	38089410	Disinfectants containing mixtures of coal tar acid & alkalis
3	sterilization product	38089420	Disinfectants other than those containing mixtures of coal tar acid & alkalis
3	sterilization product	38089490	Disinfectants other than HS 3808.94.10 - 3808.94.20
3	sterilization product	84192000	Medical, surgical or laboratory sterilisers
3	sterilization product	28470010	Hydrogen peroxide, whether or not solidified with urea. In liquid form
3	sterilization product	28470090	Hydrogen peroxide, whether or not solidified with urea. Oth than liquid form
3	sterilization product	30049010	Transdermal therapeutic system patches for the treatment of cancer or heart diseases
3	sterilization product	30049020	Closed sterile water for inhalation, pharmaceutical grade
3	sterilization product	30049030	Antiseptics
3	sterilization product	30049041	Anaesthetics,Containing procaine hydrochloride
3	sterilization product	30049049	Anaesthetics,Other than Containing procaine hydrochloride
3	sterilization product	30049051	Analgesics,antipyretic&oth med cont para cetamol/dipyrone,acetylsalicylic,taken o
3	sterilization product	30049052	Analges,antipyretic&ot.medicament for the treat.of coughs/colds,contain/not antihistamine,contain chlrpeniramine maleate
3	sterilization product	30049053	Analges,antipyretic&oth.medicament for the treat.of cough/cold,contain/not antihistamine,contain diclofenac,taken orally
3	sterilization product	30049054	Analgesic,antipyretic&oth medicament for treatment of cough/cold,contain/not antihistamine,contain piroxicam/ibuprofen
3	sterilization product	30049055	Analgesic,antipyretic&oth medicaments for the treatment of cough/cold,contain/not antihistamines,other,in liniment form
3	sterilization product	30049059	Analgesic,antipyretic&oth med cont acety lsalicylic acd,paractaml/dpyron,not oral
3	sterilization product	30049064	Antimalarials, Containing artemisinin other than of subheading 3004.60.10
3	sterilization product	30049065	Antimalarials, Not Containing primaquine & artemisinin other than of subheading 3004.60.10, Herbal medicaments
3	sterilization product	30049069	Antimalarials, Not Containing primaquine & artemisinin other than of subheading 3004.60.10, Herbal medicaments
3	sterilization product	30049071	Anthelmintic, Containing piperazine or mebendazole (INN)
3	sterilization product	30049072	Anthelmintic Not containing piperazine or mebendazole (INN) for Herbal medicaments
3	sterilization product	30049079	Anthelminti containing piperazine or mebendazole (INN) for Herbal medicaments
3	sterilization product	30049082	Other medicaments for the treatment of HIV/AIDS
3	sterilization product	30049089	Oth medicaments for other intractable diseases oth than HIV/AIDS
3	sterilization product	30049091	Other medicaments containing sodium chloride or glucose, for infusion
3	sterilization product	30049092	Other medicaments containing sorbitol or salbutamol, for infusion
3	sterilization product	30049093	Other medicaments containing sorbitol or salbutamol, in other forms
3	sterilization product	30049094	Other medicaments containing cimetidine (INN) or ranitidine (INN) other than for injection
3	sterilization product	30049096	Nasaldrop medicaments containing naphazoline, xylometazoline or oxymetazoline
3	sterilization product	30049098	Herbal medicaments
3	sterilization product	30049099	Other medicaments except HS 3004.10.15 to 3004.90.98
3	sterilization product	29051200	Propan-1-ol (propyl alcohol) & propan-2-ol (isopropyl alcohol)
3	sterilization product	29151100	Formic acid
3	sterilization product	29151200	Salts of formic acid
3	sterilization product	29182100	Salicylic acid & its salts
3	sterilization product	85394900	Ultra-violet/infra-red lamps
3	sterilization product	85395000	Light-emitting diode (LED) lamps
3	sterilization product	85437010	Electric fence energisers
3	sterilization product	85437020	Remote control apparatus,oth radio remote control apparatus
3	sterilization product	85437030	Electrical machines with translation/dictionary functions
3	sterilization product	85437040	Equipment/machine for removal dust particle&curingmaterial by UV light for mfg of printed circuit boards
3	sterilization product	85437090	Oth equipment/machine for removal dust particle&curingmaterial by UV light for mfg of printed circuit boards
4	Oxygen teraphy	90192000	Ozone therapy,oxygen therapy, aerosoltherapy, artificialrespiration/other therapeutic respiration apparatus
4	Oxygen teraphy	90183910	Catheters
4	Oxygen teraphy	90183990	Cannulae and the like
4	Oxygen teraphy	90268010	Oth instruments/apparatus for measuring/checking the flow,level,pressure,electrically operated
4	Oxygen teraphy	90268020	Oth instruments/apparatus for measuring/checking the flow,level,pressure, not electrically operated
4	Oxygen teraphy	90181900	Other electro-diagnostic apparatus used in medical
5	other medical	90221200	Computed tomography apparatus
5	other devices	90181200	Ultrasonic scanning apparatus,used in medical
5	other devices	90181100	Electro-cardiographs,used in medical
5	other devices	90189020	Intravenous administration set
5	other devices	90189030	Electronic instruments and appliances,used in medical
5	other devices	90189090	Other instruments and appliances,used in medical
5	other devices	90251911	Thermometers&pyrometers,elect operated,temperature gauges for motor vehicles
5	other devices	90251919	Thermometers&pyrometers,elect operated,oth temperature gauges for motor vehicles
5	other devices	90251920	Thermometers&pyrometers not electrically operated
5	other devices	84131910	Pump fitted with measuring device,electrically operated,excl for dispensing fuel/lubricant in filling-station/garage
5	other devices	84131920	Pump fitted with measuring device,not-electrically operatedexcl for dispensing fuel/lubricant in filling-station/garage
5	other devices	90282020	Water meters
5	other devices	90282090	Liquid meters other than water meters.
5	other devices	73249010	Other, including parts, flushing water closets or urinals (fixed type)
5	other devices	73249091	Parts of kitchen sinks or bathtubs
5	other devices	73249093	Parts of flushing water closets or urinals (fixed type)
5	other devices	73249099	Other parts of kitchen sinks or bathtubs/flushing water closets or urinals (fixed type)
6	other consumables	28044000	Oxygen
6	other consumables	30059010	Bandages
6	other consumables	30059020	Gauze
6	other consumables	30059090	Wadding & similar articles, except bandage & gauze
6	other consumables	30051010	Adhesive dressings & other articles having an adhesive layer, impregnated or coated with pharmaceutical substances
6	other consumables	30051090	Adhesive dressings&other articles having an adhesive layer,except impregnated or coated with pharmaceutical substances
6	other consumables	34012020	Soap chips
6	other consumables	34012091	Soap in other forms of a kind used for flotation deinking of recycled paper, other than soap chips
6	other consumables	34012099	Soap in other forms, other than soap chips & soap of a kind used for flotation deinking of recycled paper
6	other consumables	34011140	Medicated soap including disinfectant soap
6	other consumables	34011150	Other soap including bath soap
6	other consumables	34011160	Soap&org surf-active prod&prep, of felt/nonwovens,impregnated,coated/covered w/ soap/detergent
6	other consumables	34011190	Soap&org surf-active prod&prep,form of bars,cakes,mould pcs/shapes,&pap.wad,felt etc,for toilet use exc HS3401.11.40-60
6	other consumables	34013000	Organic surfaceactive products&prep for washing the skin,in the form of liq/cream&put up for retSale,containing soap/not
6	other consumables	90183110	Disposable syringes, with or without needles
6	other consumables	90183190	Syringes, with/without needles oth than disposable syringes
6	other consumables	90183200	Tubular metal needles and needles for sutures
6	other consumables	39232910	Aseptic bags of oth plastics,reinforced or not with aluminium foil,widthâ‰¥315mm&lengthâ‰¥410mm,incorporating a sealed gland
6	other consumables	39232990	AsepticBags of oth plastics,reinforced/not w aluminiumfoil (exc retort pouches),widthâ‰¥315mm&lengthâ‰¥410mm,not SealedGland
6	other consumables	30067000	Gel prep designed to be used in hum or vet med as a lubricant for parts of the body for surg operations or physical exam
6	other consumables	37011000	For Xray
6	other consumables	37021000	For Xray
6	other consumables	87059050	Oth special purpose motor vhcl,not design for person & goodStreet cleaning vhcl, cesspool emptiers, mobile clinics
6	other consumables	87059090	Oth special purpose motor vhcl,not design for person & goodother special vhcl
6	other consumables	94029010	Furniture designed for medical,surgical/ veterinarypurposes &parts thereof, oth than chairs
6	other consumables	94029020	Commodes
6	other consumables	94029090	Oth furniture designed for medical, surgical/veterinarypurpose&part thereof, oth than chairs and commodes
6	other consumables	84213920	Air purifiers
6	other consumables	84213990	Filtering/purifying mach & aparatus for gases
6	other consumables	73110023	Containers for compressed/liquefied gas,of iron/steel seamless steel cylinders,capacity>30L for LPG
6	other consumables	73110024	Containers for compressed/liquefied gas,of iron/steel seamless steel cylinders,capacity>30L &<110L for LPG
6	other consumables	73110025	Oth containers for compressed/liquefied gas seamless steel cylinders, for Liquefied Petroleum Gas (LPG)
6	other consumables	73110026	Oth containers for compressed/liquefied gas seamless steel cylinders, capacity<30L
6	other consumables	73110027	Oth containers for compressed/liquefied gas,seamless steel cylinders,capacity>30L&<110L
6	other consumables	73110029	Oth containers for compressed/liquefied gas seamless steel cylinders, capacity>30L
6	other consumables	73110091	Other Containers for compressed or liquefied gas, of iron or steel with capacity < 7,3L
6	other consumables	73110092	Other Containers for compressed or liquefied gas, of iron or steel with capacity > 7,3L &< 30L
6	other consumables	73110094	Other Containers for compressed or liquefied gas, of iron or steel with capacity > 30L &< 110L
6	other consumables	73110099	Other Containers for compressed or liquefied gas, of iron or steel with capacity > 110L
6	other consumables	76130000	Aluminium containers for compressed or liquefied gas.
6	other consumables	63062200	Tents of synthetic fibres
6	other consumables	63062910	Tents of oth textiles of cotton
6	other consumables	63062990	Tents of oth textiles, oth than cotton
6	other consumables	84248910	Hand-operated household sprayers of capacity <= 3 l
6	other consumables	84248920	Spray heads with dip tubes
6	other consumables	84248940	Wet processing equipment;apparatus for the spot applicationof liquid;apparatus for the application of dry film/similar
6	other consumables	84248950	Mechanical appliances, electrically operated, oth than foragricultural/horticultural
6	other consumables	84248990	Mechanical appliances, not electrically operated, oth thanfor agricultural/horticultural
6	other consumables	70171010	Quartz reactor tubes&holders glass,for production of semiconductor wafers
6	other consumables	70171090	Laboratory/pharmaceutical glass of fused quartz/oth fusedsilica; other
6	other consumables	70172000	Laboratory,hygienic/pharmaceutical w/coef< 5x106/kelvin,0-3000c
6	other consumables	70179000	Laboratry,hygienic/pharmaceutical glasswre,nor fused quartz/fused silica

A quick look at Indonesia's current account and balance of trade using World Bank's API in Python

Wed, 21 Oct 2020 21:07:25 +1100

Indonesia has been always famed for its lack of depth and innovation in its financial market. Banks are powerful and aim mostly at consumer banking. Indeed, the role of foreign investment is quite central to Joko ‘Jokowi’ Widodo’s development policy, from building infrastructures and the same-price gasoline policy, both relies on SOE’s corporate bonds, to attracting FDI to provide jobs. The high influx of foreign investment of course leads to a current account deficit, which means increase the surge of imports. This is a basic GDP accounting where current account balance is $=S-I=X-M$.

Interestingly, minimizing Current Account Deficit (CAD) has been something Indonesian government set as their de facto policy targeting. Targeting CAD may be important(?) for stability, but might not be something you would want if you aim for growth. In this blog, I try to have a look at Indonesia’s CAD, what’s causing it, and a bit of insight from looking at it. My visualization relies on Python’s Seaborn and , wbdata. My learning resources are and World Bank’s and .

Using wbdata

install it first using:

pip install -U wbdata

We then import it and see what are the source.

import wbdata as wb

wb.get_source()

 id name
---- --------------------------------------------------------------------
1 Doing Business
2 World Development Indicators
3 Worldwide Governance Indicators
5 Subnational Malnutrition Database
6 International Debt Statistics
11 Africa Development Indicators
12 Education Statistics
13 Enterprise Surveys
14 Gender Statistics
15 Global Economic Monitor
16 Health Nutrition and Population Statistics
18 IDA Results Measurement System
19 Millennium Development Goals
20 Quarterly Public Sector Debt
22 Quarterly External Debt Statistics SDDS
23 Quarterly External Debt Statistics GDDS
24 Poverty and Equity
25 Jobs
27 Global Economic Prospects
28 Global Financial Inclusion
29 The Atlas of Social Protection: Indicators of Resilience and Equity
30 Exporter Dynamics Database – Indicators at Country-Year Level
31 Country Policy and Institutional Assessment
32 Global Financial Development
33 G20 Financial Inclusion Indicators
34 Global Partnership for Education
35 Sustainable Energy for All
36 Statistical Capacity Indicators
37 LAC Equity Lab
38 Subnational Poverty
39 Health Nutrition and Population Statistics by Wealth Quintile
40 Population estimates and projections
41 Country Partnership Strategy for India (FY2013 - 17)
43 Adjusted Net Savings
44 Readiness for Investment in Sustainable Energy
45 Indonesia Database for Policy and Economic Research
46 Sustainable Development Goals
50 Subnational Population
54 Joint External Debt Hub
57 WDI Database Archives
58 Universal Health Coverage
59 Wealth Accounts
60 Economic Fitness
61 PPPs Regulatory Quality
62 International Comparison Program (ICP) 2011
63 Human Capital Index
64 Worldwide Bureaucracy Indicators
65 Health Equity and Financial Protection Indicators
66 Logistics Performance Index
67 PEFA 2011
68 PEFA 2016
69 Global Financial Inclusion and Consumer Protection Survey
70 Economic Fitness 2
71 International Comparison Program (ICP) 2005
72 PEFA_Test
73 Global Financial Inclusion and Consumer Protection Survey (Internal)
75 Environment, Social and Governance (ESG) Data
76 Remittance Prices Worldwide (Sending Countries)
77 Remittance Prices Worldwide (Receiving Countries)
78 ICP 2017
79 PEFA_GRPFM
80 Gender Disaggregated Labor Database (GDLD)
81 International Debt Statistics: DSSI

So many databases, but we will get what we want from World Development Indicators, number 2 on the above list. Now, let’s search current account balance in the database

wb.search_indicators('current account balance',source=2)

id name
----------------- ------------------------------------------
BN.CAB.XOKA.CD Current account balance (BoP, current US$)
BN.CAB.XOKA.GD.ZS Current account balance (% of GDP)

I will use one with % of GDP. I will take the current account balance only for Indonesia and only from 1981 to 2019. Don’t forget to create a datetime tuple to limit the year you’re taking.

import pandas as pd
import datetime
tanggal=(datetime.datetime(1981,1,1), datetime.datetime(2019,1,1))
a=wb.get_dataframe({"BN.CAB.XOKA.GD.zS" : "Current account balance (% of GDP)"}, country=["IDN"], data_date=tanggal, convert_date=True, keep_levels=True)
a=a.reset_index()
a.head()

	country	date	Current account balance (% of GDP)
0	Indonesia	2019-01-01	-2.714101
1	Indonesia	2018-01-01	-2.939161
2	Indonesia	2017-01-01	-1.594657
3	Indonesia	2016-01-01	-1.819151
4	Indonesia	2015-01-01	-2.035042

Let’s visualize it with seaborn

import seaborn as sns; sns.set()
import matplotlib.pyplot as plt
plt.figure(figsize=(13.5,8))
graph1=sns.lineplot(x="date", y="Current account balance (% of GDP)",data=a)
graph1.axhline(0, color='black')
graph1.axvline(datetime.datetime(1998,1,1), color='red')
graph1.text(datetime.datetime(1999,1,1),-6, "Asian Financial Crisis",color='red',size='x-large')
plt.show()

The above graph is Indonesia’s current account balance. I suppose it is clear to us that Indonesia were running a current account deficit since the collapse of oil price in the early 80s. Financial liberalization were then took place around the same period and was getting more progressive up until right before Asian Financial Crisis (AFC) in 1998. Indeed, running a mismanaged CAD is somewhat dangerous when the investment is not allocated efficiently. Those investment were not productive and Indonesia failed to earn enough foreign currency to pay back its debt. The economy was then crash, and perhaps haunt us until now. This is, perhaps, the reason why we hate CAD so much.

Indonesia was having a positive current account since then, but starting 2011, the CAD started to happen. Since then, CAD has been happening consistently up until now. The anti-import sentiment was return, and the government has been rising Non-Tariff Measures to combat it. In fact, according to , NTM shot up by almost 30% in numbers, from 2015 to 2018.

What causes our CAD?

Interestingly, Indonesia’s trade balance (export - import) has always been positive. We have been exporting enough to offset our import. Our deficit was mainly driven by a negative net primary income. The negative net primary income was driven by interest payment from portfolio and FDI. Indonesia’s foreign investment inflow has largely been used to actually pay these interests. That’s one reason why Indonesian policy makers talk a lot on speeding up growth and improving export relative to import, because reducing net primary income deficit is much harder in the short run.

b=wb.get_dataframe({"BN.GSR.FCTY.CD" : "Net Primary Account (Current USD)"}, country=["IDN"], data_date=tanggal, convert_date=True, keep_levels=True)
b=b.reset_index()
b.head()
c=wb.get_dataframe({"BN.GSR.MRCH.CD" : "Net Trade in goods (Current USD)"}, country=["IDN"], data_date=tanggal, convert_date=True, keep_levels=True)
c=c.reset_index()
b["Net Trade in goods (Current USD)"]=c["Net Trade in goods (Current USD)"]/1000000
b["Net Primary Account (Current USD)"]=b["Net Primary Account (Current USD)"]/1000000
b=b.set_index('date')
del b['country']
plt.figure(figsize=(13.5,8))
graph2=sns.lineplot(data=b)
graph2.axhline(color='black')
graph2.set(ylabel='Juta USD',xlabel='Tahun')
plt.show()

Increasing export or reducing import?

If we take a look at the graph below, Indonesia’s import follows its export closely.

b=wb.get_dataframe({"BX.GSR.MRCH.CD" : "Goods Export (Current USD)"}, country=["IDN"], data_date=tanggal, convert_date=True, keep_levels=True)
b=b.reset_index()
b.head()
c=wb.get_dataframe({"BM.GSR.MRCH.CD" : "Goods Import (Current USD)"}, country=["IDN"], data_date=tanggal, convert_date=True, keep_levels=True)
c=c.reset_index()
b["Goods Export (Current USD)"]=b["Goods Export (Current USD)"]/1000000
b["Goods Import (Current USD)"]=c["Goods Import (Current USD)"]/1000000
b=b.set_index('date')
del b['country']
plt.figure(figsize=(13.5,8))
graph2=sns.lineplot(data=b)
graph2.set(ylabel='Juta USD',xlabel='Tahun')
plt.show()

Improving export in the short run can be challenging, especially given the current global situation. Indonesia’s two main industries mining and plantation, are still performing quite well, but these two commodities are very sensitive to price volatility, and .

Unfortunately, trying to reduce import is also not ideal. Indonesian imports are currently consist of many industrial supplies and capital goods, needed for firms to operate. In fact, for firms operate in the Global Value Chain (GVC), imported inputs are crucial for them to be competitive in the global market. This is perhaps the reason why Indonesian import and export are moving together, because its exporting firms need imports to be able to export. Reducing imports can be bad both in the short run (industrial supplies) and in the long run (capital goods).

# This one's not using WBdata. I use UNCOMTRADE i used for my other project
import pandas as pd
import seaborn as sns; sns.set()
import matplotlib.pyplot as plt
c=pd.read_csv('data.csv')
c=c.rename(columns = {'Trade Value (US$)' : 'Trade Value (Million US$)'})
c['Commodity']=c['Commodity'].replace({'Capital goods (except transport equipment), and parts and accessories thereof':'Capital goods'})
c['Commodity']=c['Commodity'].replace({'Transport equipment, and parts and accessories thereof':'Transport eq & parts'})
c['Trade Value (Million US$)']=c['Trade Value (Million US$)']/1000000
plt.figure(figsize=(13.5,8))
g=sns.lineplot(x="Period", y="Trade Value (Million US$)", hue="Commodity",data=c)
g.legend(loc='upper left', ncol=1)

<matplotlib.legend.Legend at 0x1be0560a148>

Indeed, Indonesia face challenge again during this times. I personally think that reducing import is a bad move. What is important is to make sure its investment pays off and keep .

The heterogeneous impacts of tariffs and NTM on total factor productivity of Indonesian firms

Fri, 02 Oct 2020 11:00:00 +1000

COVID-19 death rate in Indonesia

Sun, 27 Sep 2020 09:37:03 +1000

This morning, as usual, I prepared breakfast, did the dishes, all while listening to a podcast. One of my favorite, Babbage aired the recent development of COVID-19, including the decreasing death rate across the world.

As cases grow, we should expect a still increasing death rate as well. What happens is the contrary. Slavea Chankova, the healthcare-correspondent at The Economist, said that healthcare workers now have the knowledge to better treat COVID-19 patients. They also able to reduce the use of ICU bed, which is a good thing.

Other publications also say similar things. A post from The Conversation says that the decreased death rate is the result of increased proportion of young age. As test capacity increases, the death rate converges to the ’true’ death rate.

Furthermore, The Conversation post shows a graph where albeit increased number of new cases, new deaths are going down. This is in UK btw.

What about Indonesia?

The case in Indonesia is a bit different. I try to re-create the graph for Indonesian context using data from Our World in Data

import pandas as pd
import seaborn as sns; sns.set()
import matplotlib.pyplot as plt
c=pd.read_csv("https://raw.githubusercontent.com/owid/covid-19-data/master/public/data/owid-covid-data.csv")
c['date']=pd.to_datetime(c['date'])

y1=c['new_cases'][c['iso_code']=='IDN']
y2=c['new_deaths'][c['iso_code']=='IDN']
x=c['date'][c['iso_code']=='IDN']
df=pd.DataFrame({'date':x,'new_cases':y1,'new_deaths':y2})
plt.figure(figsize=(15,5))
ax = df.plot(x='date', y='new_cases', legend=False)
ax2 = ax.twinx()
df.plot(x='date', y='new_deaths', ax=ax2, legend=False, color="r")
ax.figure.legend()
plt.show()

#nt=sns.load_dataset("c")
#sns.lineplot(x=x, y=y, data=c)
# sns.lineplot(x=x, y=y4, data=c)

As you can see from the graph above, Indonesia’s new deaths are still on the increase as new cases goes up. New cases in Indonesia is still increasing, with new record breaking number of cases every day. Indeed, Indonesia, albeit its ability to ramp-up testing capacity, is still being criticised for not testing enough.

Death Ratio is converging

But when in comes to death rate, Indonesia is indeed converging to the world. The Jakarta Post, citing the satgas’ spokeperson Wiku Adisasmito, report that Indonesia’s death rate is constantly decreasing since April, although still above the world’s average.This may be at least a silver lining, although in Indonesia there’s also a ruckuss about the government trying to propose change the definition of COVID-19 death.

using $death rate = \frac{total cases}{total deaths} * 100$ , I try to illustrate Indonesia’s death rate and compare them to some countries I feel comparable. Data from Our World in Data.

cn=c[(c['iso_code'] == 'IDN') | (c['iso_code'] == 'IND') | (c['iso_code'] == 'BRA') | (c['iso_code'] == 'USA') | (c['iso_code'] == 'VNM') | (c['iso_code'] == 'MYS')]
cn['rate']=cn['total_deaths']/cn['total_cases']*100
plt.figure(figsize=(15,5))
#nt=sns.load_dataset("c")
#sns.lineplot(x=x, y=y, data=c)
sns.lineplot(x="date", y="rate", hue="iso_code", style="iso_code",data=cn)

Apart from Vietnam, the deathrate among these countries seem to have similar patterns. The explanation might be similar: better knowledge on how to treat patient, more, converging tests, and increased fraction of young population.

So, all is good?

Not necessarily. Remember, Indonesia’s case is still increasing fast, and with it, its new deaths. Hospital capacity and the availability of healthcare workers are increasingly become issues in handling COVID-19. This is still a big concern.

Naively speaking, there are two possible ways to handle this. One is to decrease the viral infection’s sped (so called flattening the curve), and the other is to improve healthcare capacity. Which one’s easier?

Another big recent news in Indonesia is the resignation of Professor Akmal Taher, a respectable doctor, from Indonesia’s COVID-19 task force. His reason was that improving testing and tracing would be easier to be done from outside of the task force. To me, this is very interesting development. What other place with better authority and resource to improve testing and tracing? Resource-wise, there can’t be a better place?

Does this means the top structure of the task force does not put testing and tracing as the main strategy? It is possible that the task force is prioritizing healthcare infrastructure, which is not what Akmal Taher would want. Me, being nobody, have not enough information to assess the best strategy in handling COVID-19, but it does seem that Indonesian government is having a lot of problems in terms of conducting a proper test and trace. Does that mean it is easier to buy hospital beds and recruit doctors and nurses?

Alas, this post provides more question than answer. So what do you think? Testing and tracing or improving healthcare capacity? Or both?

When Instrumental Variable (IV) is problematic

Wed, 16 Sep 2020 18:46:31 +1000

As a tutor for both Australia Awards Scholarship and ANU, I get the privilege to learn many courses. There’s a lot of great content in ANU in terms of courses which I can’t cope with. But as a tutor, I am ‘forced’ to learn many of these courses so that’s definitely something. Haha!

This time, i’d like to write about a demonstration of weak instrument that I learned from a course at CBE.

What’s an IV?

One problem with OLS is endogeneity/reverse causality. This happens when our independent variable is actually depending on something else. This something else is then absorbed by the error term and violate $E[MXM\epsilon] \neq 0$.

We often use IV as one of the treatment to this problem. IV is a third variable $Z$ that should be independent from $Y$ but have high correlation with $X$. IV will establish $X$ causes $Y$ but not $Y$ causes $X$.

Problem IV

However, when the instrument $Z$ is not explaining $X$ very well, IV may lead to even more bias. We try to simulate this using R below.

Setting

Here, we try to use finance setting. Let $X$ be a firm’s cash flow while $Y$ is firm’s borrowing. We try to establish causal relationship between $X$ and $Y$. When firm’s cash flow is negative, it will need to borrow from financial market to survive for the next revenue collection time. However, when firm’s cash flow is positive, it may also borrow if the positive cash flow is a sign of positive shock and the firm needs fund to expand. We can use technological shock $Z$ as an instrument for $X$.

Generate 2,000 observation, where half experienced a flat negative shock $( − 3 + rand())$ while half experience positive shock $X = Z + rand()$.

data <- data.frame(
id=c(1:2000)
)
# generate data observation
data$lnz <- rnorm(2000, mean = 0.5, sd=1)
data$Z <- exp(data$lnz)
data$U <- rnorm(2000,mean = 0, sd=0.1)
data$B1 <- rbinom (2000, 1, 0.5) # to explain about probability 1
data$B2 <- 1-data$B1 # to explain about probability 1
# generate firm cashflow
data$X <- data$B1*data$Z - 3*data$B2 + data$U
# generate firm borrowing
data$E <- rnorm(2000,mean = 0, sd=0.1)
data$R <- 0.5*data$U + 0.5*data$E
data$X1 <- ifelse (data$X>0, 1, 0)
data$X2 <- ifelse (data$X<0, 1, 0)
data$Y <- (data$X*data$X1) - (data$X*data$X2) + data$R
head(data)
## id lnz Z U B1 B2 X E R
## 1 1 0.53446900 1.7065418 -0.02069438 1 0 1.685847 -0.16523761 -0.09296600
## 2 2 -0.25843412 0.7722599 -0.01968836 0 1 -3.019688 -0.09326667 -0.05647752
## 3 3 0.58658233 1.7978335 0.05513357 0 1 -2.944866 -0.10034947 -0.02260795
## 4 4 1.48663543 4.4221917 -0.18736333 0 1 -3.187363 -0.10648732 -0.14692532
## 5 5 1.07463211 2.9289152 0.04403934 0 1 -2.955961 0.03063488 0.03733711
## 6 6 0.05710789 1.0587700 0.03335503 1 0 1.092125 0.07427162 0.05381333
## X1 X2 Y
## 1 1 0 1.592881
## 2 0 1 2.963211
## 3 0 1 2.922258
## 4 0 1 3.040438
## 5 0 1 2.993298
## 6 1 0 1.145938

Positive cash flow $X$ have a 1-on-1 relationship with positive $Z$, but a flat $-3$ with negative $Z$.

Let’s try plot $Z$ and $X$

library(ggplot2)
a<-ggplot(data, aes(Z,X, colour=B1)) +
geom_point()
a

Blue is those experience positive shock while black is those who get negative shock. Positive shock has 45^∘ from $X = 0$ while all the negative ones are flat around $X = -3$. Look at that. What a bad pattern. Horrible for researchers 🤣🤣.

Now let’s try $X$ and $Y$:

b<-ggplot(data,aes(X,Y,colour=X1)) +
geom_point()
b

Firm’s borrowing sudah pasti positif, dan kita buat sedemikian sehingga perusahaan meminjam sesuai cash flow (alias koefisien beta-nya = 1). Nah, ketika $X$ = − 3, mereka minjamnya $Y$ = 3, makanya yang hitam ngumpul di ($X$, $Y$) = ( − 3, 3) semua, sementara yang biru punya tren lurus dengan sudut 45^∘ alias $Y$ = $X$.

Jika kita, ceritanya nggak tau hubungan $X$, $Y$ dan $Z$ di dunia nyata, meregresikan $X$ dan $Y$ dengan OLS biasa, kira-kira seperti apa bentuknya?

b +
geom_smooth(method = "lm")

Keliatan bias banget yah?

Bagaimana dengan koefisiennya? Coba anda bisa nebak gak? ingat bahwa ketika $X$ positif, $X$ naik 1 beriringan dengan $Y$ naik sebanyak 1, sementara ketika $X$ negatif, $X$ turun 1 beriringan dengan $Y$ naik 1 juga. Sementara itu ada 50% $X$ yang negatif, dan 50% $X$ yang positif. Jika kita regresi biasa, kira-kira koefisiennya berapa? Coba tebak!

# Regression
mod <- lm(Y~X, data = data)
summary(mod)
##
## Call:
## lm(formula = Y ~ X, data = data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.0007 -1.7264 0.9776 1.1259 15.2347
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.88313 0.04075 70.76 <2e-16 ***
## X 0.32862 0.01145 28.70 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.82 on 1998 degrees of freedom
## Multiple R-squared: 0.2919, Adjusted R-squared: 0.2916
## F-statistic: 823.7 on 1 and 1998 DF, p-value: < 2.2e-16

Jawabannya (hampir) benar! (hampir) 0.5! Jadi sebenarnya dengan OLS, kita sudah mendapatkan koefisien yang (hampir) sesuai dengan data yang kita regenerasi, yaitu (hampir) 0.5! Yah tentu saja hasilnya masih salah wkwkwk karena 0.32 masih agak jauh dari 0.5 yha, tapi kan ceritanya kita gak tau dunia nyata bentuknya kayak gimana he he he.

Regresi IV malah bikin tambah jelek

Tentu saja OLS belum ngasih kita koefisien yang kita inginkan ya (tentu saja masalah bias masih ada). Yang jadi masalah, regresi IV malah bikin koefisiennya makin menjauh dari 0.5!

Coba kita regresi $Y$ dan $X$ dengan menggunakan $Z$ sebagai instrumen:

library (AER)
modelIV <- ivreg(formula=data$Y~data$X | data$Z)
summary(modelIV)
##
## Call:
## ivreg(formula = data$Y ~ data$X | data$Z)
##
## Residuals:
## Min 1Q Median 3Q Max
## -5.366 -3.175 -2.877 3.252 3.884
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.01206 0.07358 40.94 <2e-16 ***
## data$X 1.08897 0.05915 18.41 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 3.26 on 1998 degrees of freedom
## Multiple R-Squared: -1.271, Adjusted R-squared: -1.272
## Wald test: 339 on 1 and 1998 DF, p-value: < 2.2e-16

Setelah di-treat dengan IV, koefisien X dan Y malah bermasalah. Tapi sebenernya kan memang kita meregresikan X dan Y dengan koefisien 1 (meskipun aslinya 50% doang, 50%-nya lagi koefisien -1) Tentu saja, IV-nya hanya menangkap 50% dari fenomena, dan 50%-nya lagi akhirnya diserahkan ke intercept (dan error). OLS emang masih ada bias, tapi IV malah bikin tambah parah. Inilah salah satu contoh problem menggunakan IV, jika hubungan IV nya sendiri bias.

Cara terbaik? Tentu saja memisahkan regresinya jadi dua. Contoh jika kita regresi hanya shock positif saja:

dataX<-data
dataX<-dataX[!(dataX$B2=="1"),]
modelIV2 <- ivreg(formula=dataX$Y~dataX$X | dataX$Z)
summary(modelIV2)
##
## Call:
## ivreg(formula = dataX$Y ~ dataX$X | dataX$Z)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.228315 -0.047317 0.002166 0.045417 0.201270
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.0020950 0.0028174 0.744 0.457
## dataX$X 0.9997967 0.0006977 1432.905 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.06716 on 999 degrees of freedom
## Multiple R-Squared: 0.9995, Adjusted R-squared: 0.9995
## Wald test: 2.053e+06 on 1 and 999 DF, p-value: < 2.2e-16

Ini benar. Intercept = 0, dan koefisiennya 1, sesuai dengan yang kita buat. Tentu saja yang negatif tidak perlu kita IV karena ketika negatif, $X$ akan otomatis -3. Artinya, untuk yang negatif, $X$ sudah eksogen! Tidak perlu ditreatment apa-apa (meskipun, coba lihat bagaimana saya generate error term?)

Intinya, hati-hati menggunakan IV

Tentu saja di dunia nyata kita tidak tau fungsi asli dari $Y=f(X)$, dan biasanya memang IV kita gunakan untuk memperbaiki kelemahan OLS. Sayangnya IV dapat berbahaya jika instrumennya tidak menjelaskan dengan baik penyebaran dari variabel endogen yang di-treat, atau jika anda tidak tau-tau banget hubungan si instrumen dengan variabel yang di-treat. Bukannya memperbaiki, malah bikin OLS tambah parah. Tentu saja masalah lainnya adalah jika ternyata instrumennya punya korelasi dengan Y. Di contoh kita ini tentu saya Y independen terhadap Z, tapi di dunia nyata, sulit banget cari instrumen yang benar-benar independen. Contoh, jika ternyata shock-nya berpengaruh ke sektor finansial, otomatis si-shock ini akan punya pengaruh ke borrowing. Wkkwkwk.

Biasanya yang kita lakukan di paper adalah menunjukkan hasil regresi OLS dan IV, sehingga bisa dilihat pola-nya apakah IV memperbaiki OLS atau tidak. Tentu saja justifikasi dari publikasi lain soal keabsahan IV juga harus ditulis di paper kita wkwk. Di Crawford sendiri, jika anda menggunakan IV, siap-siaplah ditanya-tanyain soal kekuatan instrumennya. Ha ha ha ha!

IV memang ribet. Moga-moga kita tidak harus berurusan dengan IV. wwkwkw. Ok sekian semoga postingannya berguna.

Daily COVID-19 Testing in Indonesia

Sun, 31 May 2020 00:00:00 +0000

The government seems increasingly eager to usher in the “New Normal” and start reopening the economy. A guidebook on how to handle the “New Normal” from BPOM has already been circulated. News about malls reopening was everywhere (though apparently Summarecon Mal Bekasi did not open after all). The economy was set to bounce back!

According to many sources (for example, this one), the key to a successful economic reopening is a high growth rate of daily testing – in other words, massive testing. But have we actually achieved sufficient massive testing?

import pandas as pd
import seaborn as sns; sns.set()
import matplotlib.pyplot as plt
c=pd.read_csv("https://covid.ourworldindata.org/data/owid-covid-data.csv")
c['date']=pd.to_datetime(c['date'])
y=c['new_tests'][c['iso_code']=='IDN']
y1=c['new_tests_smoothed'][c['iso_code']=='IDN']
x=c['date'][c['iso_code']=='IDN']
plt.figure(figsize=(15,5))
#nt=sns.load_dataset("c")
sns.lineplot(x=x, y=y, data=c)
sns.lineplot(x=x, y=y1, data=c)

The chart above shows the number of “new tests” per day. The data is taken from Our World in Data. It is updated daily based on government reports. The x-axis shows the date, while the y-axis represents the variable “new_tests” from the original dataset. In short, this is the daily test count.

The dots and the thin line show the raw daily test numbers, while the blue line shows the smoothed new tests.

As a brief aside, the code above was run on Python 3.7. The data is read directly from Our World in Data’s online portal, so every time you run it, you get the latest figures. If you have Python on your machine, feel free to copy the code above to see the most current data.

Indonesia’s daily testing clearly has a lot of gaps, and up until 26 May 2020, the daily test count was still very volatile. The ups and downs are somewhat random. Some opinions I have seen on Twitter suggest that daily test counts are hard to predict because they depend not only on the number of swabs conducted but also on the queue of samples at laboratories. So the numbers reported on a given day may actually be the results of swabs taken the day before.

I have not followed the COVID-19 testing debate too closely, to be honest. I had not updated on COVID-19 in a while, but the sudden buzz around reopening the economy piqued my curiosity. Also, Bio Farma reportedly succeeded in producing its own PCR test kits, claiming it can produce 100,000 per day. Assuming smooth logistics to labs and sufficient lab capacity, massive testing should be achievable. Hopefully this upward trend continues and remains sustainable during the New Normal.

By the way, the data comes from the central government, and the central government’s data reflects its own testing plus tests conducted by regional governments (and perhaps private entities). Although we do not have data beyond the central government’s reports, at least regional governments know their own numbers.

Also, other countries have somewhat erratic “new tests” data as well. New Zealand and Malaysia are also volatile, while Japan does not report daily new tests (though cumulative totals are available). Vietnam’s data suddenly drops and disappears midway. So perhaps this is why President Jokowi felt so confident.

import pandas as pd
import seaborn as sns; sns.set()
import matplotlib.pyplot as plt
c=pd.read_csv("https://covid.ourworldindata.org/data/owid-covid-data.csv")
c['date']=pd.to_datetime(c['date'])
plt.figure(figsize=(15,10))
#cn=c.drop(c[(c['iso_code'] != 'IDN')| c[(c['iso_code'] != 'VNM') | c[(c['iso_code'] != 'SGP') | c[(c['iso_code'] != 'AUS') | c[(c['iso_code'] != 'MYS')].index)
cn=c[(c['iso_code'] == 'IDN') | (c['iso_code'] == 'VNM') | (c['iso_code'] == 'MYS') | (c['iso_code'] == 'JPN') | (c['iso_code'] == 'NZL')]
sns.lineplot(x="date", y="new_tests", hue="iso_code", style="iso_code",data=cn)

Here I used raw daily tests, not daily tests per 1,000 population. If I used the per-capita measure, other countries would be so far above that Indonesia would look like a flat line. Anyway, I was trying to use the argument that “we are confident enough to reopen because testing capacity has been ramped up.” Using this argument, Indonesia’s upward trend does look reasonable. Whether it is sufficient or not, of course, is still a matter of debate – and I am sure many others have already weighed in.

In closing, there are of course many reasons behind the decision to reopen the economy, not just testing. I am not trying to discuss those other reasons. This post was simply meant to provide a brief snapshot of how Indonesia’s daily testing has been progressing. I will leave the rest to the experts.

Amiti and Konings 2007

Tue, 25 Feb 2020 00:00:00 +0000

This is a long-overdue blog review of Amiti and Konings’ paper. This paper has been widely cited, and when it came out, the issue was really hot: how tariff reductions could help Indonesia join Global Value Chains (GVCs) and become more productive.

The article begins by summarizing the debate over the role of trade liberalization for industry. Most of the articles they cite show that trade liberalization has a positive impact on domestic industries. They emphasize the importance of examining the role of input import liberalization, which at the time still had a gap in the literature. The article continues with the dynamics of import duties in Indonesia and Indonesia’s involvement in the World Trade Organization (WTO) in 1995.

The most interesting part of the article is of course the methodology. They conduct 2 stages of econometric estimation and one deterministic stage of productivity calculation (TFP). Assuming a Cobb-Douglas production function, they estimate factor return parameters on output.

$$ y_{it} = \beta_0 + \sum_{Z={k,l,m}} \beta_Z Z_{it} + e_{it} \label{1} $$

After obtaining the estimated parameters, they compute log TFP with

$$ tfp_{it}^k = y_{it} - \sum_{Z={k,l,m}} \hat{\beta}_Z Z_{it} \label{2} $$

where k, l, m are capital, labor and materials respectively. After obtaining TFP values for each firm, TFP is then regressed on input tariffs and output tariffs. Simple.

The interesting part is that equation \ref{1} is estimated by accounting for the fact that within $e_{it}$ there is a time-varying productivity shock, which will affect the firm’s input purchasing decisions. This problem is accommodated by the Olley-Pakes method, which they discuss in the appendix. Additionally, because they care about imported materials, they modify the Olley-Pakes method so that import decisions are also accounted for in the firm’s optimization process. It’s fascinating how they connect trade and the optimization process of imported material inputs into TFP without going through the standard gravity equation approach.

This paper really showcases their hard work in processing data. Aggregating and disaggregating tariffs into the Indonesian Statistics (SI) data they used was no easy task. Many parts had to be done manually. Imagine mapping HS 9-digit codes to ISIC, then building an input-output table from SI on your own. Not to mention they had to calculate tariffs for both inputs and outputs.

One more thing: a paper published in AER typically has extensive robustness checks. They performed various robustness checks, such as examining sensitivity to alternative TFP metrics, endogeneity of trade policy shocks, alternative tariff measures, and Non-Tariff Measures.

In conclusion, this is a fascinating article. For the TFP estimation method specifically, if you’re curious you’d still need to look at the other papers they cite. But nonparametric estimation like this is always interesting, as it doesn’t rely entirely on the assumptions required by standard econometrics. A very engaging paper to read, especially if you’re interested in Indonesian trade issues. This paper also serves as an excellent introduction to Lili Yan Ing’s article in the book World Trade Evolution.

Thorbecke 2019: Japan's Exchange Rate and Electronics Exports

Tue, 25 Feb 2020 00:00:00 +0000

For my first paper review post, I will cover a paper by Thorbecke that is certainly interesting. The paper, which examines how Japan’s electronics export competitiveness was affected by exchange rate movements, was recently published. It is fascinating to see how a manufacturing giant like Japan has been displaced by South Korea and Taiwan.

The paper starts with a compelling premise. Japan has long been a powerhouse in exporting electronic components and parts. Competition in components and parts is quite fierce, so Japan has to compete on price with other countries (mainly Taiwan and South Korea). Trouble came when the Global Financial Crisis (GFC) hit. Japan suddenly became a safe haven. Capital flowed in, the Yen appreciated, and export prices became expensive relative to costs denominated in Yen. This caused Japan’s electronic components and parts industry to lose competitiveness against other countries – an effect that persists to this day.

The paper uses a Vector Auto Regressive (VAR) method. The estimation takes roughly the following form:

$$ \Delta p_{jt}^x = \beta_0 + \sum_{i=0}^p \beta_{1i} \Delta e_{jt-i} + \sum_{i=0}^p \beta{2i} \Delta p_{t-i}^f + \sum_{i=0}^p \beta_{3i} \Delta c_{jt-i} + \sum_{i=0}^q \beta_{4i} \Delta y_{t-i}^f + u_t $$

where $p^x$ denotes Japan’s export price to a destination country, $e$ is the exchange rate, $p^f$ is the international market price, $c$ represents production cost, and $y$ is the producer price index in OECD countries. All data in the paper come from the Bank of Japan and the OECD. The model itself is fairly straightforward and could readily be replicated for other country contexts – provided, of course, that comparable data are available as those supplied by the Bank of Japan. The key identification assumption is that the sudden exchange rate change is treated as exogenous.

Upon reading this paper, the first thing I noticed was how convincingly the initial premise was built. The author opens with a simple yet impactful narrative supported by clear findings. The charts used (such as producer price indices and export price indices) strongly support the storytelling in the opening section. The model is not overly complicated either, relying on the assumption that the exchange rate is exogenous. Having a natural experiment like a financial crisis is certainly advantageous for a researcher.

Just kidding. Crises are bad. Nobody wants a crisis. Hopefully we will not see another one.

Anyway, this paper is very concise – only 9 pages. The results are as expected, and with the key variable assumed to be exogenous, there is no need for complex methods to deal with endogeneity bias. It is a fairly accessible paper for someone who is just starting to learn about VAR.