Investigating the heterogeneous effects of temperature on economic growth

This paper investigates the effect of temperature on economic growth on a panel of 156 countries over a 50-year time period. We use random fluctuations in a country's annual average temperature over time as our strategy for identifying the causal effect of temperature on GDP per capita growth. Previous work has found that temperature has a statistically significant relationship with growth when fitting data from all countries with a nonlinear model. We hypothesize that countries with different observable characteristics have differing responses to changes in temperature and we use recursive model partitioning to find data-driven splits in the dataset. We run the model on a number of country-level characteristics and we find that GDP per capita percentile and historical mean temperature are the variables that best fit the data. We divide the countries into four groups: low income - low temperature (20 countries), low income - high temperature (70 countries), high income - low temperature (36 countries), high income - high temperature (33 countries). We obtain different coefficients for the relationship between temperature and growth for each group. We also use agricultural GDP per capita growth as a dependent variable and obtain coefficients for the relationship between temperature and agricultural GDP per capita growth. We compare our results with results obtained from a model that does not have any splits (Burke et. al., 2015). We find that dividing countries into groups using a data-driven method has a significant impact on future projections of GDP per capita growth under different climate change scenarios and how we interpret the effects that changes in average temperatures might have on countries. A model that does not split countries into groups overestimates the gains that low temperature countries might make from rising temperatures and underestimates the ability of high income - high temperature countries to capitalize on the high temperatures that they regularly experience. Our model takes into account adaptations to historical temperatures that countries may have and we find that low temperature countries have lower "optimum temperatures" that help them achieve maximum growth and likewise, high temperature countries have higher "optimum temperatures." We predict that low income - low temperature countries and high income - low temperature countries are likely to face low growth (1.73 % and 0.00879 % respectively) in the year 2100; low income - high temperature countries will face negative growth (-3.19 %) and high income - high temperature countries will face positive growth (3.16 %)