The impact of climate change on economic output across industries in Chile

Econometric model It is well known that temperature affects the dynamics of virtually all chemical, biological and ecological processes [12], while precipitation can affect agriculture [5, 29], especially in Latin America [7]. and also non-agricultural activities if excessive floods disrupt transport and urban connections [11, 12]. Chile, in particular, has been strongly affected in terms of reduced water availability [20] and a decade long mega-drought [25]. Zivin and Neidell (2014) [30] found that warmer temperatures reduce labor supply, while Cachon, Gallino, and Olivares (2012) [31] document that high temperatures decrease productivity and performance. Seasonal temperatures and precipitation can affect productivity both in outdoor activities such as agriculture, fishing and construction [7, 11], but also for non-agricultural activities due to the influence of the weather on workers’ health or urban movement [12, 13]. For this reason our vector Tr,s,t for the measure of the weather variables in region r in season s of year t includes both average temperature and precipitation. There can be other shocks besides the weather (for instance, international shocks such as the Great Financial Crisis or higher demand from commodities due to a higher economic growth in China) that affect the economic growth of each industry i at time t. For this reason our chosen model must account for both time-industry fixed-effects (αt,i) and the dynamic effect of shocks in the previous year by controlling for the lagged growth (Δyr,i,t−1). Further-

more, an adequate model must account for regional heterogeneity in terms of natural resources, weather and industry specialization, therefore our model will include fixed-effects across regions and industries (αr,i) and heterogeneous coefficients (βs,i for the impact of the weather variables Tr,s,t, ρi for the impact of the lagged growth Δyr,i,t−1). Our econometric model therefore follows a panel structure of log GDP growth (yr;i;t ¼ lnðGDPadj r;i;t Þ) as the dependent variable, with explanatory variables including the lagged GDP growth, the average temperature and precipitation of each season plus fixed-effects for region-industry (αr,i) and year-industry (αt,i): Dyr;i;t ¼ P s2Sbs;iTr;s;t þ riDyr;i;t 1 þ ar;i þ at;i þ εr;i;t ð3Þ where s is the season (either every quarter or every month of the calendar year), Tr,s,t is a vector of the weather variables (average temperature, precipitation) affecting the region r in season s of year t. We estimate the models by OLS with robust standard-errors clustered by region and year. In relation to other alternatives such as random-effects, the fixed-effects added in our model help to control for fixed unobservables across time-industry and region-industry with- out imposing any distribution assumption or any correlation assumption with the other observable variables, while the random-effects models assume that the fixed unobservable errors are normal distributed and uncorrelated with the other observable variables [32, 33]. It is also worth noting that several of the previous papers that estimate the impact of climate

change on GDP use fixed-effects rather than random-effects (see [2, 12, 13, 16]). Results Main regressions Since the Chilean GDP series for region-industry are available only at an annual frequency, then it is hard to estimate the impact of each month on the yearly GDP of each region-industry (too many coefficients for a 32 year period). However, using only quarterly averages for the weather can mask strong highs and lows in temperature and rainfall. Both the models with monthly weather (too many coefficients and low precision) and quarterly weather (too little PLOS ONE The impact of climate change on economic output across industries in Chile PLOS ONE | https://doi.org/10.1371/journal.pone.0266811 April 28, 2022 9 / 21