This is the first in this year’s series of posts by PhD students on the job market.
While everyone will have to deal with climate change, poor countries will be especially affected. We have extensively studied the effects of extreme weather on agriculture in these countries, but are only beginning to understand its effects on non-agricultural firms (which produce 80 percent of sub-Saharan Africa’s GDP, for example). We do know that extreme weather negatively affects them, and that this happens through two channels:
● Demand: Extreme weather means local agriculture suffers, decreasing incomes and demand for non-agricultural firms’ products
● Supply: Extreme weather reduces worker productivity
We don’t know how firms react to these weather shocks, though. That is a problem when designing policy to reduce the impact of climate change, because firm reactions can reinforce or counteract the intended effects of different policy instruments.
My job market paper quantifies non-agricultural firms’ reactions to extreme weather and their implications for policy design. I combine data on firms across Africa and South Asia with global weather data to estimate the causal effect of weather on firms and to measure firms’ reactions. Given a key finding – weather disproportionately affects exporters – I develop an international trade model capturing key firm reactions. Combining the model with machine learning estimates of the impact of climate change, I show that firm reactions substantially alter policy effectiveness: Trade policy becomes considerably stronger at counteracting the negative effects of climate change.
Data
I use World Bank Enterprise Surveys data, which contain detailed information on non-agricultural firms’ operations for the last fiscal year. They allow me to observe not just sales, but also to differentiate between domestic and international sales and between different cost categories. I use data on firms across sub-Saharan Africa and South Asia, with surveys covering fiscal years between 2005 and 2019. The range of firms covered in the Enterprise Surveys is large: The 25th percentile of sales is roughly USD 28,000, while the 75th percentile is USD 500,000, and for the number of employees, the corresponding percentiles range from six to 23. Almost a third of firms are in manufacturing and about 12 percent are exporters.
I link firms’ locations (shown in Figure 1) to daily weather data on temperature and precipitation spanning the period from 1980 – 2020. I also use weather projections from 2015 – 2100, covering different climate change scenarios and climate models.
Figure 1: Locations of firms used
Identification
The first half of my paper comprises several reduced form regressions relating outcomes y(j,t) such as sales, for firm j in year t , to a measure of weather over that year x(j,t) as
Here, z(j,t) are additional firm characteristics, most importantly exporter status; different impacts of weather shocks on exporter and non-exporters are a crucial result in my paper, as I discuss below. The deltas are year fixed effects, to increase precision of the estimates (none of the main results depend on these). The core identification concern with regressing firm outcomes on weather is that all the best performing firms are located in cooler climates. I solve this by grouping firms into clusters n, using cluster fixed effects gamma(n(j)) to isolate random year-to-year weather variation. (The Enterprise Surveys have panel data for some countries and some years, but the geographic range is limited compared to the amazing coverage I get using this approach.) My reduced form estimations use a parsimonious temperature index as x(j,t) combining different measures of heat across the year to capture how extreme a fiscal year’s weather was. This simple measure makes it easier to relate something as complex as weather to firm outcomes.
Weather is primarily a supply shock
I first show that weather shocks are, primarily, supply rather than demand shocks. This is important to understand because firm reactions to each shock differ. I run a simple test to tell the difference, building on two basic open economy intuitions:
● If extreme heat mostly depresses local demand, exporters should do better than non-exporters in response: exporters have access to foreign demand, which isn’t affected by the local shock
● If it mostly makes workers less productive, raising firms’ costs, exporters should do worse: Domestically, firms can pass cost increases on to consumers by raising prices. Internationally, exporters face stiffer competition, so they instead have to reduce sales.
To test this, I interact the temperature index with an indicator for exporter status, as I alluded to above, and check whether exporters see a larger or smaller impact of weather shocks. I find that an 80th percentile weather shock decreases domestic producers’ sales by 3.9 percent, but exporters’ sales by 6.9 percent, with the difference statistically significant at the one percent level. (These effect sizes are comparable to, for example, the effect of ethnic conflict on flower packers’ or mobile phones on fishers’ output.) The key takeaway is that weather shocks have a larger effect on exporters’ sales. This means weather is, on average, a supply shock, leading me to focus on supply-side firm reactions.
To be sure this differential effect is about exporting, rather than being due to some correlate of being an exporter, I run several robustness checks showing that the exporter effect is not due to differences between exporters and non-exporters, such as the sectors they are in, firm size, or the complexity of their production processes.
Another concern here is survival bias: It could be that the least productive domestic firms shut down and disappear from the data, not reporting their dismal sales, while the least productive exporters do not, leading to larger observed impacts on exporters. I cannot rule this out completely, since I do not have long panel data allowing me to observe exit. I do show that extreme weather does not lead to more firms reporting zero sales (which is not the same as shutting down, and rare, but the most direct test I can run) and that extreme weather does not cause me to see a higher share of exporters in the data.
Firms react by reducing productivity
To understand firm reactions, I consider the margins firms can adjust. I study firms’ basic production structure, finding that domestic sales show a similar pattern as total sales: exporters’ domestic sales decrease more in response to weather shocks. This suggests firm operations across different markets are linked, which begs the question: Where does that link come from?
I show it is due to expenditures on productive capability. Productive capability comprises rented or hired equipment, facilities and non-production personnel. For example, it includes rented machinery, rented office space or a sales team. Productive capability improves overall firm performance, by increasing labor productivity or making it easier to sell the firm's output. When higher temperatures reduce labor productivity, firms cut expenditures on productive capability, because those are complementary to labor productivity. Scaling back productive capability in turn reduces productivity even further.
I observe these expenditures in the data and quantify the reaction. In response to an 80th percentile weather shock, domestic producers reduce productive capability expenditures by 2.9 percent, while exporters reduce theirs by 6.7 percent. I also show that controlling for productive capability spending, exporters do not see a larger impact of weather shocks on their domestic sales. This provides strong evidence that productive capability links sales across markets: Accounting for the change in productive capability, exporters’ sales losses are comparable to domestic producers. (I run several robustness checks showing that the exporter effect is not due to other differences between exporters and non-exporters, such as the sectors they are in or the complexity of their production process.)
The upshot: Trade policy matters more
The second half of my paper assesses whether productivity reactions matter for policy design. To do that, I answer two questions:
● What are the implications of productivity reactions for the economy as a whole?
● How does that affect policy effectiveness in a climate change scenario?
To answer the first question, I develop an international trade model allowing firms to hire productive capability. For the second question, I use a causal forest, a widely used machine learning method, to estimate the average drop in sales firms face under different climate change scenarios. (See the paper for details on both; I want to point out that I allow for firm adaptation to climate change in my estimation.)
I then calibrate the model to a climate change scenario matching the causal forest estimate. From that baseline, I estimate what happens under different policy interventions, comparing the results to a standard trade model without the productive capability channel. I find that a ten percent drop in trade costs leads to a 1.7 times larger welfare increase due to firms’ productivity reactions.
What this means for policy design
Climate change reduces firm productivity. Some of the losses, however, are due to firms reducing productive capability expenditures. Importantly, these losses can be partially recouped, for example by connecting firms to large foreign markets, where reaching new customers warrants spending on more office space or a larger logistics team. This could be done by reducing tariff lines and streamlining export and import regulations, for example. On the flip side, rich country policies that could push poor countries out of international trade can cause additional collateral damage.
One might conclude from my results that another option would be for poor countries to use tariffs to reduce import competition, allowing exporters to pass more of their cost increases on to domestic consumers. For most countries, however, their domestic market is so small relative to world markets that even strong protectionist policies would not compensate exporters for the loss of access to foreign markets.
Where to next
This is a first step towards understanding how firms in poor countries deal with extreme weather and climate change. I highlight one reaction and show that it matters for policy design. As we keep working on this, we will uncover more such reactions. Hopefully, tailored policy taking those reactions into account will prevent some of the dire consequences of climate change for poor countries.
Max Huppertz is a PhD candidate in Economics at the University of Michigan
Join the Conversation