I recently read an interesting paper by David Atkin, Amit Khandewal and Adam Osman that looks at not only an intervention to get firms to export, but also the mechanism by which exporting helps the firms learn.
The setting: carpets in Fowa, Egypt. Atkin and co. are evaluating a randomized program to connect producers of carpets with export markets. They work with Aid to Artisans and a local intermediary (Hamis) to identify a set of small firms (
Initially, the order-mobilizing side of the program couldn’t generate these big orders, so the first round of take up was fairly low. And it was limited to one type of rug (the paper has a fascinating explanation, and set of pictures of, different types of carpets). However, after a bit of time the larger orders came through, and a second wave of firms had much higher take up. Every firm that accepted the initial order delivered on it.
So what do they find? Not surprisingly, the treatment vastly increases the odds that firms had ever exported – raising it by 55 percentage points off of a baseline of 13 percent. And this translates into higher profits on the order of 16-26 percent (with the variation depending on how profits are constructed). These are sizable effects when compared to some of our standard business promotion interventions.
So this is neat, but then Atkin and co. take us on a journey to understand where this is coming from. To start with, treatment firms get significantly higher prices per square meter of carpet. But they also produce less per month. Treatment firms are using somewhat more labor (5 percent) but this is with existing workers, not new ones. They aren’t spending more money on new capital, but the length of production runs is increasing.
Overall, the lower level of output and the higher price commanded per meter suggest quality upgrading. And when Atkin and co. measure the quality of rugs on 11 different dimensions, they find that export firms are producing rugs that are of significantly higher quality in 10 of these.
So what’s going on? One possible explanation is that these firms always knew how to produce better quality rugs and, now faced with a market that demands and pays for this, they do so (a movement along the production possibilities frontier (PPF)). Another possible explanation is that the firms are actually learning from the process of exporting (a shift outwards in the production possibilities frontier). Atkin and co. take us through 5 steps to convince us it is the latter.
Step 1. The buyers of export rugs are providing different specifications – in terms of type, thread count, and more difficult designs. So, while unadjusted measures of productivity in export firms fall, once Atkin and co. take the specifications into account, productivity actually rises in treatment (export) firms. As they summarize it “conditional on making similar rugs, treatment firms are making them faster than control firms.”
Step 2. Since the export specifications are indeed different, Atkin and co. set up a “quality lab.” They rent a workshop and ask each firm to come to it and produce a rug of identical specifications -- specifications that would be mid-tier for the domestic market. They pay the firms to produce these rugs. These rugs are then graded by a master artisan and a professor of handicraft science at an Egyptian university. If firms are just moving along the PPF, treatment and control firms’ quality adjusted productivity shouldn’t be different. Lo, treatment firms are producing rugs in about the same time, but they are 0.5 (professor) to 0.85 (master artisan) standard deviations higher in terms of quality. So the export experience is spilling over into higher quality of goods even when they are destined for the domestic market.
Step 3. If firms always had the knowledge on how to produce for export, the productivity changes should be instantaneous. Atkin and co. show us that they aren’t: firms steadily increase quality until they’ve produced about 200 square meters (about 5 months), then they level off.
Step 4. To give us an even more direct sense of the mechanism, Atkin and co. collect data on the information exchange between Hamis (the intermediary) and the firms. It turns out buyer feedback (part of this information) comes in two types: complaints (e.g. my rugs frayed) and advice (e.g. pack them this way). Most of the communication is about the latter type. Atkin and co. can also map this information to specific dimensions of quality that they measure. It turns out that for both complaints and advice, feedback is significantly associated with quality improvements in the relevant dimension.
Step 5. Ruling out any other explanations. Atkin and co. run through some other explanations – and if you still have an alternate hypothesis, I urge you to read the paper.
In sum, this paper is neat not only for the export effects of the intervention, but even more so because it gives us a clear sense of the mechanism that is at play. And to do this, the authors clearly spent a huge amount of time not only collecting data, but also in designing a rather spectacular range of data collection tools. These range from making sure the intermediary kept log books of the timing and content of its interaction with firms, to setting up the “quality lab” where they could get a clear sense of productivity in isolation of other factors to super-detailed (to my non-carpet expert eye) measures of carpet quality. And it clearly paid off in providing some key insights in differentiating among explanations.
The setting: carpets in Fowa, Egypt. Atkin and co. are evaluating a randomized program to connect producers of carpets with export markets. They work with Aid to Artisans and a local intermediary (Hamis) to identify a set of small firms (
Initially, the order-mobilizing side of the program couldn’t generate these big orders, so the first round of take up was fairly low. And it was limited to one type of rug (the paper has a fascinating explanation, and set of pictures of, different types of carpets). However, after a bit of time the larger orders came through, and a second wave of firms had much higher take up. Every firm that accepted the initial order delivered on it.
So what do they find? Not surprisingly, the treatment vastly increases the odds that firms had ever exported – raising it by 55 percentage points off of a baseline of 13 percent. And this translates into higher profits on the order of 16-26 percent (with the variation depending on how profits are constructed). These are sizable effects when compared to some of our standard business promotion interventions.
So this is neat, but then Atkin and co. take us on a journey to understand where this is coming from. To start with, treatment firms get significantly higher prices per square meter of carpet. But they also produce less per month. Treatment firms are using somewhat more labor (5 percent) but this is with existing workers, not new ones. They aren’t spending more money on new capital, but the length of production runs is increasing.
Overall, the lower level of output and the higher price commanded per meter suggest quality upgrading. And when Atkin and co. measure the quality of rugs on 11 different dimensions, they find that export firms are producing rugs that are of significantly higher quality in 10 of these.
So what’s going on? One possible explanation is that these firms always knew how to produce better quality rugs and, now faced with a market that demands and pays for this, they do so (a movement along the production possibilities frontier (PPF)). Another possible explanation is that the firms are actually learning from the process of exporting (a shift outwards in the production possibilities frontier). Atkin and co. take us through 5 steps to convince us it is the latter.
Step 1. The buyers of export rugs are providing different specifications – in terms of type, thread count, and more difficult designs. So, while unadjusted measures of productivity in export firms fall, once Atkin and co. take the specifications into account, productivity actually rises in treatment (export) firms. As they summarize it “conditional on making similar rugs, treatment firms are making them faster than control firms.”
Step 2. Since the export specifications are indeed different, Atkin and co. set up a “quality lab.” They rent a workshop and ask each firm to come to it and produce a rug of identical specifications -- specifications that would be mid-tier for the domestic market. They pay the firms to produce these rugs. These rugs are then graded by a master artisan and a professor of handicraft science at an Egyptian university. If firms are just moving along the PPF, treatment and control firms’ quality adjusted productivity shouldn’t be different. Lo, treatment firms are producing rugs in about the same time, but they are 0.5 (professor) to 0.85 (master artisan) standard deviations higher in terms of quality. So the export experience is spilling over into higher quality of goods even when they are destined for the domestic market.
Step 3. If firms always had the knowledge on how to produce for export, the productivity changes should be instantaneous. Atkin and co. show us that they aren’t: firms steadily increase quality until they’ve produced about 200 square meters (about 5 months), then they level off.
Step 4. To give us an even more direct sense of the mechanism, Atkin and co. collect data on the information exchange between Hamis (the intermediary) and the firms. It turns out buyer feedback (part of this information) comes in two types: complaints (e.g. my rugs frayed) and advice (e.g. pack them this way). Most of the communication is about the latter type. Atkin and co. can also map this information to specific dimensions of quality that they measure. It turns out that for both complaints and advice, feedback is significantly associated with quality improvements in the relevant dimension.
Step 5. Ruling out any other explanations. Atkin and co. run through some other explanations – and if you still have an alternate hypothesis, I urge you to read the paper.
In sum, this paper is neat not only for the export effects of the intervention, but even more so because it gives us a clear sense of the mechanism that is at play. And to do this, the authors clearly spent a huge amount of time not only collecting data, but also in designing a rather spectacular range of data collection tools. These range from making sure the intermediary kept log books of the timing and content of its interaction with firms, to setting up the “quality lab” where they could get a clear sense of productivity in isolation of other factors to super-detailed (to my non-carpet expert eye) measures of carpet quality. And it clearly paid off in providing some key insights in differentiating among explanations.
Join the Conversation