Can Export Promotion Reduce Unemployment?

The paper examines the impact of export promotion on aggregate unemployment. We findnd that increases in the share of Export Promotion Agencies' (EPAs) budgets on total exports lead to small decreases in aggregate unemployment. This effect is amplifi ed when export promotion efforts are concentrated in sectors in which the country has a comparative advantage. On the other hand, when EPAs aim at reducing aggregate unemployment by focusing their efforts in sectors with high levels of unemployment, then aggregate unemployment increases. These results suggest that even if EPAs' pri- orities were to shift towards reducing unemployment, this would be better addressed by focusing on sectors in which the country has a comparative advantage rather than sectors with high labor market frictions.


Introduction
There is growing evidence that export promotion activities contribute to export growth (for example Atkin, Khandelwal and Osma, 2017, Broocks and Van Biesebroeck, 2017, and Volpe and Carballo, 2008, 2010a, 2010b, and 2012. While this export growth almost automatically lead to employment growth in exporting firms, trade economists know at least since Brecher (1974) that this does not necessarily leads to reductions in aggregate unemployment. This paper examines the impact that increases in export promotion budgets have on aggregate levels of unemployment in a panel of 52 countries over the period [2005][2006][2007][2008][2009][2010][2011][2012][2013][2014]. Using data on the sectoral expenditure of Export Promotion Agencies (EPAs) we also explore whether export promotion efforts should concentrate in sectors with a comparative advantage or in sectors with high unemployment, when aiming at reducing aggregate unemployment.
We find that a 1 percent increase in the share of EPAs' budgets in total exports leads on average to a 0.32 percent reduction in unemployment (or 0.03 percentage points). This is a qualitatively interesting and statistically significant result, but export promotion does not seem to have a very large impact on aggregate unemployment. However, the effect can be significantly amplified if export promotion efforts are concentrated in sectors with a comparative advantage without any need to increase EPAs' budgets. If EPAs were to fully align their promotion expenditure with each country's comparative advantage, unemployment would decline on average by 2.85 percent (or 0.23 percentage points). On the other hand, if EPAs expenditure were to be fully aligned with sectorial levels of unemployment, then unemployment would increase on average by 6.22 percent (or 0.50 percentage points).
This suggests that the allocation of promotion efforts across sectors matter if EPAs were to target aggregate unemployment.
These results are important for at least three reasons. First, EPAs have proliferated over the last 20 years, and as their effectiveness has been shown in terms of export growth, it is important to understand whether there are unintended consequences on other important policy objectives such as unemployment. With export and income growth being often referred to as being jobless (see Ancharaz, 2011 or Schmitt-Grohé andUribe 2017), it is important to ensure that resources put into export promotion are not contributing to jobless growth, or worse, increases in unemployment. Second, the results highlight that not all export promotion efforts help reduce aggregate unemployment. The strategic choices made by EPAs when deciding where to focus their efforts matter. Perhaps more surprisingly, export promotion efforts that target sectors with high unemployment are more likely to lead to increases in unemployment, whereas those that target sectors with comparative advantage are more likely to reductions in unemployment. Understanding this heterogeneity is crucial to help policymakers trying to disentangle what works and what does not work when it comes to export promotion, and how this may depend on the policy objective associated with the intervention. It also clearly illustrates how focusing efforts in a sector that is slowing down or for which unemployment is increasing can backfire. Last, but not least, we provide evidence suggesting that the reallocation of export promotion efforts, rather than increases in export promotion budgets, are likely to have a larger impact on aggregate unemployment. With more than 50 percent of EPAs' budgets being publicly funded and large increases in government deficits and public debt over recent years, this is an important result for governments facing tighter budget constraints.
The main challenges were data related. First, to examine the impact of the allocation of export promotion expenditures across sectors, one needs data on export promotion expenditures at the sectoral level. This was obtained from three EPAs' survey where EPAs were asked to report their export promotion expenditure by sector. These surveys were undertaken by the World Bank and the European Trade Promotion Organization, and summarized in Olarreaga, Sperlich and Trachsel (2020). Second, because we want to examine how these sectoral expenditures correlate with comparative advantage, we need to compute measures of comparative advantage across countries and sectors. This was undertaken following Costinot, Donaldson and Komunjer (2012) gravity model approach where for every year bilateral sectoral exports are run on a set of exporter*sector, importer*sector and exporter*importer fixed effects. The exporter*sector fixed effect of each year is then used as a proxy for the country's sectoral comparative advantage. Finally, we need sector level unemployment data, which is difficult to observe. We rely on an approach developed by  which shows that aggregate unemployment can be decomposed into sector level, country level and year components that can be then used to compute sector level unemployment for every country and year.
As mentioned earlier there is a large and growing literature showing that export promotion contributes to export growth. But to our knowledge this is the first paper examining the impact of export promotion on aggregate unemployment. The closest paper to ours is Munch and Schaur (2017) which focuses on export promotion by the Danish Trade Council and shows that it leads to a higher level of employment at the firm level, but no evidence is given for the impact at the aggregate level. It is important to note that this cannot be obtained by aggregating across exporting firms as resources are pulled away from other exporting and non-exporting firms which can potentially lead to a decline in aggregate employment. Similarly, several papers have examined what type of export promotion works best. Volpe and Carballo (2010) using data for ProChile show that higher returns are obtained when export promotion focuses on small exporters. Volpe and Carballo (2008) show that efforts should focus on the extensive rather than intensive margin, while Broocks and van Biesebroeck (2017) show that in the case of experienced exporters there could also be positive returns at the intensive margin. However, none of these papers have explored how the sectoral allocation of promotion efforts affects the overall returns, and whether efforts should be aligned with the country's comparative advantage.
The remainder of the paper is organized as follows. Section 2 provides the empirical framework. Section 3 discusses data sources and estimates of comparative advantage and sectoral unemployment, as well as summary statistics. Section 4 discusses the empirical results and section 5 concludes.

Empirical framework
To assess the impact of export promotion and its allocation across sectors on aggregate unemployment, we use the following empirical model: ln U c,t = β 1 ln Budget Exports c,t + β 2 corr Budget s,c,t ; CA s,c,t c,t +β 3 corr Budget s,c,t ; u s,c,t c,t + β 4 corr (CA s,c,t ; u s,c,t ) c,t where U c,t is the unemployment rate at the national level in country c at period t (in logs) and Budget/Exports is the share of the EPAs' budget on total exports (in logs), corr Budget s,c,t ; CA s,c,t c,t is the correlation between the EPAs' budget expenditure across sectors and the countries' comparative advantage (CA s,c,t ), and corr Budget s,c,t ; u s,c,t c,t is the correlation between the EPAs' budget expenditure across sectors and the sectoral unemployment (u s,c,t ). β 1 estimates the impact that increases in the share of EPAs' budget on total export have on aggregate unemployment. Note that we have no a priori on the sign of the coefficient. Increases in export promotion lead to export growth, but this can perfectly be jobless export growth, or worse, it can lead to increases in unemployment if the firms benefitting from export promotion are less labor-intensive than firms in the rest of the economy, or are in sectors with higher levels of sectoral unemployment; β 2 allows to capture the impact of increases in the alignment between export promotion expenditures and comparative advantage at the sectoral level on unemployment. Again, we have no a priori on this coefficient as focusing on sectors with comparative advantage can in principle lead to more or less unemployment depending on labor-intensities or labor market frictions across sectors; β 3 captures the impact of increases in the alignment between export promotion expenditures and sectoral unemployment on aggregate unemployment. Again, there are no a priori here as focusing on sectors with high levels of unemployment increases labor demand in sectors with high levels of unemployment and therefore can help reduce unemployment, but it also shifts resources to sectors where labor market frictions may be stronger leading to an increase in aggregate unemployment due to a composition effect. Importantly, the sign of these last two coefficients will allow us to examine how the strategic choice of EPAs in the allocation of export promotion expenditure across sectors affects aggregate unemployment.
As control variables we have the correlation between comparative advantage and sectoral unemployment, GDP per capita, and population (the last two in logs), as well as country and year fixed effects. We control for the correlation between comparative advantage and sector level unemployment because as shown by  and Carrère, Grujovic and Robert-Nicoud (2020) this matters for aggregate unemployment. We expect β 4 > 0, as a higher correlation between comparative advantage and sector level unemployment leads to higher levels of aggregate unemployment as shown both theoretically and empirically by . Indeed, the labor force is attracted into sectors in which the country has a comparative advantage. If these sectors tend to have higher levels of unemployment, this will result in higher levels of aggregate unemployment.
The share of export promotion budgets in total exports can be endogeneous in equation (1). To address this concern we follow Olarreaga Sperlich and Trachsel (2020) instrumental variable strategy, and use a measure of the broadness of the EPA mandate to instrument for the share of EPA's budget. The variable takes the value 1 if export promotion is the only responsibility of the agency; 2 if it is the top two priority, 3 if it is one of the two top priorities, 4 if it is one of three or more top priorities, and 5 if it is secondary to other priorities. Thus, as the value of the variable increases, the responsibility of the agency in terms of export promotion gets diluted. The more often cited responsibilities cited in the World Bank and International Trade Centre surveys that are given to EPAs apart from export promotion are investment and tourism promotion. We expect a positive correlation between the broadness of the agency's mandate/responsibilities and the size of the budget.
After estimating equation (1) we compute the changes in unemployment that would occur if EPAs were to fully align their promotion efforts with the country's comparative advantage or with the levels of unemployment observed at the sectoral level, illustrating a potential strategic choice for EPAs.
The change in unemployment if EPAs were to fully align their promotion expenditure with their comparative advantage (CA) is given by equation (2). The change in unemployment if EPAs were to fully algin their promotion expenditure with the levels of unemployment at the sector level (u) is given by equation (3). In the results section we compute ∆U (B = CA) and ∆U (B = u) for each country year to see how a change in the allocation of promotion efforts would affect the aggregate level of unemployment without any need for increases in EPAs' budgets.

Data and Variable Construction
Aggregate unemployment rates are borrowed from ILO's Key Indicators of the Labour Market (KILM) indicators, 1 which cover 96 countries over the period 1995-2009. The KILM database provides the raw data reported by each country, and an adjusted series estimated by the ILO. Because we use this data in a regression framework where we allow for measurement error of the left-hand-side variable we prefer to use the raw data. Differences in values between the two series are rather small and therefore when we estimate equation (1)  To estimate equation (1)  linearly project the budget shares in between two surveys. To check that this is not driving the results we will provide a robustness check in which only non-projected data is used.
The second data challenge is that we do not observe comparative advantage. However, several options exists in the literature to estimate it. We adopt the theoretically well-grounded Ricardian measure proposed by Costinot, Donaldson and Komunjer (2012) using the gravity framework. Using CEPII's BACI trade data described above we estimate for every year between 2003 and 2014 the following gravity equation (12 regressions are undertaken): where export c,i,s are exports country c to importer i of goods from sector s; α c,s are exporting country*sector fixed effects that capture the comparative advantage of the exporting country in sector s; α c,i are bilateral fixed effects that capture bilateral trade costs between the two countries; and α i,s captures the comparative disadvantage of the importer i in sector s. As shown by Costinot, Donaldson and Komunjer (2012) one can then construct a theoretically well-grounded measure of comparative advantage using the estimate of α c,s for each year: CA c,s,t = exp( α c,s,t /σ).
where σ is the elasticity of exports with respect to productivity, which is estimated at 6.53 by Costinot, Donaldson and Komunjer's (2012).
The third data challenge is that sector level unemployment rates are not observed neither. To estimate these unemployment rates at the sector level we followed the methodology developed by . The idea is simple. We observe aggregate unemployment from the ILO. We then assume that it can be decomposed into sector level component that is common However, it can be decomposed into the weighted sum of sectoral unemployment, where the weights are given by the labor force in each sector. The labor force by sector is not observed but sectoral employment is observed and using the unemployment rate at the sector level (that we will estimate) is straightforward to see that the labor force looking for jobs in each sector is given by the observed level of employment divided by 1 plus the sector level unemployment. This implies that the sector level unemployment can be estimated using: where u s is the sector-specific component of the sectoral unemployment rate that we estimate using equation (6); c,s,t is the employment share in sector s in country c at time t; γ c are country fixed effects that capture the country-specific component of the sectoral unemployment rate; and γ t are year fixed effects that capture the year-specific component of the sectoral unemployment rate.
To estimate (6) we use employment data at the sector level from INDSTAT compiled by UNIDO 7 and match it to employment shares in each country for the seven sectors in EPAs' surveys. The estimation is run on an unbalanced panel of 847 observations at the country- year level resulting from the merging of ILO and UNIDO indicators for the largest possible span. We use a non-linear estimator than imposes u s to be positive, as well as our estimate of sector level unemployment by country and year which given the additive form in equation (6) is given by u c,s,t =û s +γ c +γ t .

EPAs' strategic choices
We then measure for every country and year the correlation of EPA's sectoral allocation of their export promotion budgets with the estimated sector level unemployment rates and comparative advantage. If the correlation between EPA's budget and comparative advantage is high this suggests that the EPA strategically focuses its promotion efforts in sectors in which the country has a comparative advantage. Similarly, if the correlation between EPA's budget and the estimated sector level unemployment rate is high, then the EPA strategically focuses its promotion efforts in sectors where unemployment is more problematic. We also compute the correlation between sector level unemployment and comparative advantage that we use as a control variable as suggested by .
Summary statistics for all variables are provided in Table 1. The average correlation between export promotion budgets and comparative advantage in our sample is 0.296. However there is a lot of variation across countries and time and the correlation ranges between -0.997 and 0.975. 8 The average correlation between export promotion budgets and sector level unem-7 See https://stat.unido.org/. 8 We also compute a measure of this correlation using a two-year lagged measure of comparative advantage ployment is much lower at 0.066, but again there is a lot of heterogeneity, as it varies between -0.733 and 0.543. However, given the large difference between these two average correlations, one could argue that on average EPAs tend to focus their promotion efforts in sectors with comparative advantage, rather than sectors in which there are high levels of unemployment. 9 European and Central Asian agencies (in blue) are more heterogeneous. These agencies also represent the largest share of our sample (56 percent). Latin American agencies (in red) tend to be in the right panels suggesting that they tend to focus on high unemployment sectors.
They represent 18 percent of the sample. Agencies in Asia and Oceania (in orange) tend to be in the top panels suggesting that they tend to focus on sectors with high comparative advantage. They represent 14 percent of the sample.
as the impacts of export promotion on unemployment may be delayed. 9 Note that on average there is a positive correlation between comparative advantage and sector level unemployment rates. This could suggest that there could be a high correlation between the two other correlations, which would be problematic in our econometric specification. As reported in Table 2 this is not the case. The correlation between ρ(u, B) and ρ(CA, B) is only 0.073. Table 3 presents the results of the estimation of equation (1). In the first column we only introduce the share of the export promotion in total exports on the right-hand-side without controlling for the correlations between budget and unemployment, budget and comparative advantage, and comparative advantage and unemployment. The second to fourth column introduce one by one each of these correlation, and then finally the fifth column runs the regression as in equation (1) with all correlations simultaneously.

Results
Results in all columns suggest that increases in the share of export promotion on total exports leads to reductions in unemployment. The size of the impact is very small as a 1 percent increase in the share of the export promotion budgets seems to lead to a reduction in unemployment somewhere between 0.05 and 0.07 percent. With average unemployment at 8 percent in the sample, this implies a reduction in unemployment of only 0.004 percentage points. Thus increasing export promotion can on average help reduce unemployment, but the impact is not very large.
Interestingly the allocation of export promotion budgets across sectors matters. A higher correlation between export promotion budgets and comparative advantage leads to lower levels of unemployment as shown in columns (2) and (5). On the other hand a higher correlation between export promotion budgets and sector level unemployment leads to higher levels of unemployment, although this effect is not statistically significant.
Control variables suggest that countries with higher GDP per capita tend to have lower levels of unemployment, perhaps capturing a better functioning of the labor market in these countries. Similarly, countries with a larger population also tend to have lower levels of unemployment. The correlation between comparative advantage and sector level unemployment has a positive impact on aggregate unemployment as shown by , but the effect is not statistically significant.
We perform several robustness tests reported in Table 4. We first examine the geographic robustness of the estimates. In column (1) we exclude Asian and Africa countries from the sample. Together they represent 26 percent of the observations in our sample. In column (2) we exclude Latin American countries, which represent 18 percent of the sample. In column (3) we exclude European and Central Asian countries which together represent 56 percent of the sample. In column (4) we run equation (1) using data only for European and Central Asian countries. In column (5) we use observations for which we had data in the EPA survey for the allocation of export promotion budgets across sectors, and without estimating the evolution of these shares in between surveys. Finally in column (6) we use a two-year lagged measure of comparative advantage to capture potential delays on the impact of export promotion budgets on the level of unemployment. The results in all columns are perfectly consistent with the ones reported in Table 3. Increases in the share of export promotion budgets will lead to small reductions in unemployment, and shifting the allocation of export promotion budgets towards sectors with a comparative advantage leads to lower levels of unemployment.
As discussed in section 2, these results could be biased due to endogeneity problems. Table 5 presents the instrumental variable estimates where the broadness of EPAs' mandate is used as an instrument for the share of export promotion budgets in total exports. Columns (1) shows the results of the second stage, whereas column (2) shows the results of the first stage.
Columns (3) and (4) provide the results of the second and first stage respectively, but when the sample is restricted to those observations for which export promotion budgets were not projected in between surveys. Results in column (1) suggest avlarge impact of increases on the share of export promotion budgets on aggregate unemployment, but the size of the impact is still quite modest. A 1 percent increase in the share leads to 0.322 percent reduction in unemployment (or 0.03 percentage points at the sample mean). The impact is slightly smaller in column (3) without the projected allocation of export promotion budgets across sectors.
More interestingly both columns (1) and (3) suggest that increases in the correlation between export promotion budgets and comparative advantage across sectors leads to decreases in aggregate unemployment. Similarly, increases in the correlation between export promotion budgets and sector level unemployment leads to increases in aggregate unemployment. Using the estimates of column (1), and plugging them into equations (2) and (3) we can then compute the change in unemployment that would occur if countries were to perfectly align their allocation of export promotion budgets with their comparative advantage, or the sector level unemployment observed in their country. The results are reported in Figure 2 where the horizontal axis reports changes in unemployment following a full alignment of export promotion budgets with comparative advantage, and the horizontal axis the change in unemployment following a full alignment of export promotion budgets with sector level unemployment.
In most countries a full alignment with comparative advantage leads to a reduction in aggregate unemployment. In only 23 percent of the observations there is an increase in unemployment when EPAs fully align their strategy with the country's comparative advantage.
The average decline in unemployment is 2.85 percent (or 0.23 percentage points). On the other hand, a full alignment with sector level unemployment rates leads to an increase in unemployment in almost all countries (97 percent of the observations). The average increase of unemployment is equal to 6.22 percent (or 0.50 percentage points). These are much larger effects that the change in aggregate unemployment following increases in export promotion budgets. These results suggest that if unemployment were to be targeted by EPAs, their focus should be on sectors with comparative advantage, and that relatively large reductions in unemployment can be reached without increases in EPAs' budgets.
Interestingly if EPAs were to concentrate their export promotion efforts in sectors with low unemployment (instead of high unemployment), the decline in aggregate unemployment is very similar to the decline observed when EPAs align with the country's comparative advantage. The correlation between the two changes is plotted in Figure 3. There is indeed an almost perfect fit between an alignment in sectors with high comparative advantage and an alignment in sectors with low unemployment. The correlation coefficient between the two series is 0.999. This is not because of the fact that on average countries have a compara-tive advantage in sectors with low unemployment. If anything our estimate suggest rather the opposite, with an average correlation of 0.21 between comparative advantage and sector level unemployment. The reason has to do with the fact that the coefficients in front of the correlation between EPAs' budget and comparative advantage, and between EPAs' budget and sector level unemployment estimated using equation (1) and reported in column (1) of Table 5 are of similar size, but different sign. 10 The policy implications of these simulations on EPAs' strategic choices are quite clear. If EPAs' mandate were to change to target unemployment, this could be best achieved by EPAs targeting sectors with comparative advantage. The reduction in aggregate unemployment that will be reached will be similar to the one that would be obtained if EPAs were to target sectors with low unemployment. It is important to understand that this is not because EPAs will allocate export promotion efforts in the same sectors under the two scenarios.
They would putting their efforts in different sectors, but the overall effect on aggregate unemployment would be the same.

Concluding remarks
We examine the impact of export promotion on national unemployment. We consider three different channels through which export promotion can affect unemployment. The first are increases in the share of export promotion budgets in total exports. The second and third channels capture potential changes in the allocation of export promotion budgets towards sectors with a stronger comparative advantage or higher levels of unemployment.
We find that increases in EPA's budget help reduce unemployment, but the size of the impact is very small, suggesting that this is not necessarily the best tool to achieve large reductions in unemployment. However, much larger declines in unemployment can be obtained if EPAs change the strategic allocation of their export promotion efforts across sectors. A sizeable reduction in unemployment can be achieved by focusing export promotion efforts in sectors with a strong comparative advantage and away from sectors with high levels of unemployment.
This is an important result for policymakers as it suggests that important reductions in unemployment can be achieved through export promotion without necessarily increasing export promotion budgets. In a world where most EPAs are mainly publicly funded, and where governments face large and growing fiscal deficits, this could be an effective way for EPAs to contribute not only to export growth, but also to reductions in unemployment, which is arguable a more important policy goal.
Last, but not least, while focusing EPAs' promotion efforts in sectors with comparative advantage is clearly not a strategy that minimizes unemployment, our results show that EPAs could not do better in terms of reductions in aggregate unemployment by focusing their efforts in sectors with low unemployment. In other words, while export promotion efforts would clearly focus on different sectors under the two, the reduction in aggregate unemployment is almost identical. This is important as it suggests that focusing export promotion efforts in sectors with a strong comparative advantage is a strategy that also minimizes aggregate unemployment.   Standard errors in parentheses. * p < 0.10, * * p < 0.05, * * * p < 0.01 Standard errors in parentheses. * p < 0.10, * * p < 0.05, * * * p < 0.01    Eastern and African countries in green, and countries in Asia and Oceania in orange.