Atkin, Khandelwal, Osman (2017). (Exporting and Firm Performance: Evidence from a Randomized Experiment)

This paper uses a randomized empirical experiment to address how much exporting impacts firm performance. Tens of billions of dollars are spent each year by organizations like the world trade organization in an effort to help developing countries participate in global export markets. One justification for this spending is that people believe that exporting improves firm productivity through learning-by-exporting. When learning-by-exporting happens, trade generates efficiency gains that narrow the productivity gap and further magnify gains from trade.

There are, however, doubts about whether learning-by-exporting happens. There are two main empirical issues that make causal inference unclear:

1. Selection: modern trade theories suggest that more productive firms select into exporting, so any gain in observed productivity seen amongst exporters might be a selection effect.
2. Measurement: revenue based TFP is the easiest to compute, but will be changed whenever there is a change in markups, product mix, or product quality. These are all likely to shift when a firm enters the export market. Quantity based TFP measures would resolve these issues, but requires very detailed data that is not usually available to the economist that is measuring TFP.

To get around both of these issues, the authors of this paper partnered with a non-governmental organization in the US and an intermediary in Egypt to directly connect high income US consumers with rug producers in Egypt. I’ll call the authors of this paper, NGO, an intermediary the trade partnership.

Once orders were known, the authors of this paper surveyed hundreds of small run manufacturers in Fowa, Egypt. A random subsample was given the opportunity to fill the orders. Almost all the firms in the survey had never directly sold to a foreign client.

By bringing orders to the rug producer, the partnership completely eliminated any frictions for the rug maker to find export clients. Furthermore, because the subsample of firms who engaged with the foreign clients was random, the empirical issue of selection was resolved.

To address the measurement concern, the authors recorded detailed, production-line level data for both treatment firms that exported and control firms that didn’t. The data includes cost, quantity, and pricing; but also includes detailed information reported by a third-party expert quality assessor like thread count, how sharp the corners are, how flat is lies on the floor which allowed the authors to carefully track product mix and quality at a level of detail far beyond typical trade datasets.

Because of the randomization procedure, causal effects can be identified by looking at group means across the treatment and control groups. The authors find that treatment firms report 16-26% higher profits relative to control firms. The bulk of this paper investigates the causes of this profit increase. The main findings using only the data are:

• Prices and labor hours rose, but total quantity output fell. This suggests that foreign consumers demanded higher quality products that take more labor hours to produce. The quality measures from the quality assessor confirm this hypothesis.
• Productivity measures that don’t account for product quality would suggest that productivity for treatment firms fell by 24-28% percent. This reinforces the measurement concern from before that product quality is important to consider when measuring TFP.

Simple theoretical framework

The authors note that the increased profitability of firms might be the result of an outward shift of the PPF (learning-by-exporting) or movements along the existing PPF. To identify which of these forces makes up most of the profits, they provide a simple theoretical framework.

I won’t go into detail here, but the main features of the model are that

• There are multiple specifications of rugs
• Firms choose which specification to make as well as capital and labor inputs
• Output is a function of capital and labor, augmented by a multiplicative TFP
• TFP is a function of rug specification and firm a specific TFP parameter
• Good quality is a function of the rug specification and a firm specific quality parameter
• Inverse demand for a unit of a good is affine in the quality of the good

Within this framework movements along the existing PPF happen when the slope term in the inverse demand function changes (higher willingness to pay for a specific quality). Outwards shifts of the PPF happen when there is an increase in either the TFP or quality firm specific parameters.

They are able to estimate this model using their detailed data set and using the estimated output and raw data present five main results in favor of the learning-by-exporting argument:

1. Both quality and productivity rise, as long as productivity is computed in a way that accounts for product specifications. If firms moved only along the PPF then specification-adjusted quality and productivity would remain constant
2. At the end of the experiment all treatment and control firms were asked to make an identical-specification rug suing the same materials and loom. Treatment firms completed the job faster and had higher quality on all collected dimensions.
3. Quality and productivity for treatment firms rose slowly over time, suggesting a gradual outward movement of the PPF (learning new skills). If it was a movement along an existing PPF then adjustment would have been immediate.
4. The intermediary met with producers to talk about specific aspects of the higher quality export goods. Along these dimensions the treatment group showed the highest gains over the control group, suggesting that knowledge transfer was important in the learning process.
5. No firms in the sample invested in new capital and could not pay to receive additional guidance from the intermediary. The growth in productivity was not “financed”

In the end, the evidence put forth by the authors of this paper suggests at least for the rug trade in Egypt (small firms in developing economy), that the learning-by-exporting mechanism may be active and lead to magnified gains from trade. They hypothesis, but do not explore in this paper, that any means of lowering the barriers to entice this type of firm to export (through direct search friction reduction as in this experiment or perhaps through a reduction of tariffs or other costs) would lead to similar outcomes.