Decomposition of the temperature-driven output losses in India: Plant-level evidence for the climate change adaptation policy

There is growing macro-level evidence about the negative and non-linear impact of temperature on aggregate economic output, with the temperature effects varying widely across geographical regions and extending to both agricultural and non-agricultural sectors (Hsiang, 2010; Dell et al., 2012; Burke et al.,  2015). Less is known about the country-specific micro-level mechanisms behind the temperature-output relationship and their role in adaptation to the warming climate. Addressing this knowledge gap is of considerable importance for designing effective climate change policies, especially in developing countries, which are generally exposed to higher temperatures and have limited capacity to adapt to a changing climate (Somanathan et al., 2021). This paper contributes to the progress on this issue by estimating the impact of temperature on the output of manufacturing plants in India and decomposing it into the effects on TFP and factor inputs.

The paper combines a plant-level panel of detailed production data from the formal manufacturing sector in India over 1998-2007 with high-resolution satellite-based meteorological and pollution datasets, merged at the district level. I use two approaches to construct temperature variable: a standard in the literature binned-variable and seasonal-variable approaches, both used in empirical specifications in contemporaneous and lagged forms. To isolate the role of temperature more clearly, I account for simultaneous variations in temperature and a rich set of weather and pollution controls. I further minimize the estimation biases by including plant-specific and year-by-two-digit-industry fixed effects. Standard errors are clustered at plant and district-year levels to address spatial and serial correlation.

My main findings are two-fold. First, the relationship between temperature and manufacturing output is non-linear. The output losses are especially large during the hottest season and at extreme temperatures, with more substantial losses occurring at low rather than high temperatures. This finding is consistent with the theoretical prediction from Nath (2021). An additional day with a temperature above 33°C decreases output by 0.12% or $3,749 relative to a day in the optimal interval. The comparable estimate for an additional day with a temperature below 8°C is a decrease of 0.27% and $8,435, respectively. Second, the estimated temperature-output relationship is driven by the joint effects of temperature on TFP and capital, contributing roughly 30% and 70%. The response of TFP to temperature closely follows the response of output, while the response of capital mirrors the response of output only to higher temperatures. I further decompose these primary channels to show that temperature affects TFP through its impact on labor productivity, and machinery is the most suitable for adaptation category of capital. I also find suggestive evidence of labor reallocation between seasonal manufacturing industries and between economic sectors.

These findings have important implication for adaptation. Manufacturing sector in India can adapt to changing climate by reducing the sensitivity of labor productivity to temperature and by investing in capital, prioritizing investments in machinery. Labor-related adjustments can contribute to adaptation by offsetting direct productivity losses or facilitating labor reallocation. Patterns of the seasonal responses and timing of the adjustments’ effects should also be taken into account.