Weather Variations and International Trade

Abstract

In this paper I investigate the effect of weather variations in the exporter and importer countries separately, as well as a the difference between weather variations in both countries, on bilateral trade flows. The analysis is done at the country, sectoral and product levels, worldwide, and over the 1992–2014 period. I find a negative effect of temperature variations in the exporter country and in the difference between exporter and importer countries, on bilateral trade, at the country level. At the product level, both negative and positive effects arise, but the negative effect of temperature dominates. The temperature effects are on the agricultural and manufacturing sectors, especially in the textile and metals sectors. I show that possible channels are the impact of temperature on output and labour productivity. The negative impacts are larger in exporter countries that are closer to the Equator, that have lower quality of institutions, and that export to more remote countries. If countries are able to adapt to climate change, the long term effects of temperature variations should be lower than the contemporaneous effect. Nevertheless, my results on the long term effects analysis do not support this hypothesis, suggesting no or very low adaptation. Moreover, the negative effect of temperature is persistent and cumulative through several years after the temperature shock. Concerning precipitation variation effects, they are found mainly at the product level, with the positive effect dominating for the affected products.

Appendices

Interpretation of $\Delta \mathbf C _{ijt}$

Table 15 simulates a database with 5 observations and with the variables year, temperature of the exporter country, temperature of the importer country, the calculated difference between the countries’ temperatures (which represents $\Delta \mathbf C _{ijt}$), and the calculated change in $\Delta \mathbf C _{ijt}$ between two consecutive years. The interpretation for each point and the expected effect is commented in the bottom of the table.

Table 15 Example of interpretations for the $\Delta \mathbf C _{ijt}$ variable

Table 16 Estimation model comparison and separated weather variations, without zero observations

Table 17 Estimation model comparison and separated weather variations, with zero observations

Table 18 Estimation model comparison and bilateral weather variations, without zero observations

Comparison Between Different Estimators and Samples

To demonstrate the importance of using a Poisson maximum likelihood estimator (PMLE) with robust standard errors, Tables 16 to 19 compare OLS and PMLE estimations. I do four groups of estimations: Tables 16 and 17 present estimations for weather in exporter and importer countries, and Tables 18 and 19 present estimations for the difference between exporter and importer countries’ weather; then the groups of estimations differs in their samples: in Tables 16 and 18 the estimations are presented excluding zero value trade observations, and finally Tables 17 and 19 present estimations for when they are included. The four Tables show 5 estimations: columns(1)–(2) show estimations with OLS and log of trade as dependent variable; and columns (3)–(5) present PMLE estimations. I also present the estimators without and with bilateral fixed effects (estimations in columns (2) and (4)do not have bilateral fixed effects). When bilateral fixed effects are not added, I include several bilateral control variables (distance between trade partners, common official and ethnic languages, contiguity and colonial relationship). All the estimations have clustered standard errors at the bilateral level, and to see how the significance of the coefficients varies with and without robust standard errors, the estimations in column (6) do not include robust standard errors. Overall, the results vary widely between estimators and samples. In the full sample, the effect of temperature in the exporter country goes from 8% with OLS to − 3.2% with PMLE (Table 17). In the importer country, the effects are positive and statistically significant for the OLS estimation and non-significant for PMLE. While the coefficients in the PMLE estimations from different samples are rather similar, in the OLS estimations all the coefficients differ in both value and significance (in line with Tenreyro 2007). In these estimations, the variable of interest, weather, has different signs with the different estimators. The same inconsistency across samples for OLS is found in the results of Tables 18 and 19. This suggests that OLS estimations may be biased and that heteroskedasticity can lead to misinterpretation of the results, and different conclusions depending on the sample chosen. The main difference between the estimators is the process generating the error term. Therefore, the main estimations in the chapter all use the PMLE estimator, with robust standard errors, and accounting for zero observations which results in a balanced panel dataset.

Table 19 Estimation model comparison and bilateral weather variations, with zero observations

Weather Anomalies

The definition of anomalies follows Marchiori et al. (2012):

$$\begin{aligned} {\textit{Weather\ Anomaly}}_{i,t} = \frac{Weather_{i,t}-\mu _{i}^{LR}(Weather)}{\sigma _{i}^{LR}(Weather)} \end{aligned}$$

LR:	Long run period from 1901 to 1990
$WeatherAnomaly_{i,t}$:	Temperature or precipitation anomaly in country i at time t
$Weather_{i,t}-\mu _{i}^{LR}(Weather)$:	Temperature or precipitation long run deviation in country i at time t
$Weather_{i,t}$:	Temperature or precipitation level in country i at time t
$\mu _{i}^{LR}(Weather)$:	Temperature or precipitation long run mean for country i
$\sigma _{i}^{LR}(Weather)$:	Temperature or precipitation long run standard deviation for country i

Table 20 Bilateral trade flow and weather anomalies

Table 21 Bilateral trade flow and weather anomaly: country heterogeneous effect

Table 21 presents the same estimations as Table 4 but with the anomaly measures, instead of weather. The anomalies results are different: colder and warmer or richer and poorer exporter countries have almost the same negative effect of differences in temperature anomaly (columns (2) and (4)). In terms of exporter country location with respect to the equator, temperature anomaly differences have a different effect compared to average temperature differences. Temperature anomaly differences have a positive and significant effect on bilateral trade, for countries both near and far from the equator (column (3)). The institutionally differentiated effect is not conclusive for weather anomalies. Precipitation anomaly differences have the same effect as precipitation differences, the higher the precipitation anomaly in a poor exporter country compared to the importer country, the higher the increase in bilateral trade.

Estimations with Time-Importer Country Fixed Effects

To control for all the determinants of importer countries that vary in time (including weather of the importer country), and that may be correlated with weather in the exporter country, I add fixed effect for time-importer country. Table 22 shows the results, column (1) presents the results of the baseline estimations, and columns (2)–(4) add GDP per capita, GDP per capita and population, and production proxy respectively. The comparison of these results with the previous one (Tables 2 and 3) shows that the before estimated coefficients are not affected.

Table 22 Bilateral trade and weather with importer-time fixed effects

Additional Tables

See Tables 23, 24, 25, 26, 27 and 28.

Table 23 Correlation matrix

Table 24 List of countries in the sample by income level (based on the 2018’ World Bank classification)

Table 25 Bilateral trade flow and weather variations by sector with GDP per capita, first part

Table 26 Bilateral trade flow and weather variations by sector with GDP per capita, second part

Table 27 Bilateral trade flow and weather variations by sector with GDP per capita, third part

Table 28 Bilateral trade and weather variations: controlling for inflation