Climate effects on the diversification strategy of export firms: Evidence from China

using a parametric model, the annual average temperature can mask the effects of daily extreme temperatures, significantly reducing the information contained in the estimation results. In the following sections, we conduct robustness tests using a parametric regression model. Specifically, we construct the non-parametric model as follows. Diversityicrt ¼ a þ X m6¼5bm 1 Tbinm crt þ d1Xcrt þ gi þ grt þ εicrt ð11Þ The indices i, c, r, and t represent firms, cities, provinces, and years, respectively. The dependent variable, Diversityicrt, indicates the diversity of products exported by firm i, mea- sured by the number of product categories exported by a firm. The core explanatory variables, Tbinm ct , denote the number of days in city c in year t with a daily average temperature within the temperature bin m. Temperature bin variables are defined with a length of 5˚C, for a total of 9 bins. Where, Tbin1 ct represents days with a daily average temperature below -7˚C. For example, if a city experiences temperatures below -7˚C for 10 days in a year, then the value of Tbin1 ct is 10. Tbin2 ct represents days with a temperature in the bin [-7, -2)˚C. Tbin3 ct repre- sents days with a temperature in the bin [-2, 3)˚C, and so on, with the final bin, Tbin9 ct, indi- cating days with a daily average temperature �28˚C. The regression uses the temperature bin (8, 13˚C (m = 5) as the base group. Given that the annual total of days is fixed, the cumulative days across all temperature bins remains constant. To mitigate multicollinearity in the regres-

sion analysis, one temperature bin must be designated as the baseline, thereby excluding it from the set of variables. This baseline selection is critical as it should ensure that the coeffi- cients associated with all other temperature bins are negative, thereby signifying it as the opti- mal temperature range [4,24]. In accordance with this criterion, this study selects the temperature range of 8, 13˚C as the baseline. The coefficient βm 1 is of primary interest, repre- senting the change in product diversity for each additional day with a daily mean temperature in the mth bin relative to the baseline. We expect βm 1 to be consistently negative, reflecting an inverted U-shaped trend based on our theoretical model. Control variables, denoted by X, include city-level and temperature-related factors, such as precipitation (Prec), represented by the annual mean precipitation in the city, wind speed (Wind), represented by the annual mean wind speed in the city, sunshine duration (Sun), rep- resented by the annual mean sunshine duration in the city, humidity (Wet), represented by the annual mean relative humidity in the city, and atmospheric pressure (Atmo), represented by the annual mean atmospheric pressure in the city. To reduce omitted variable bias, we include firm (γi) and province-year (γrt) fixed effects in the model. ε represents the error term. Stan- dard errors are two-way clustered at the firm and city-year levels [36]. Data description This section outlines the use of two key datasets: meteorological data and firm export data.

The meteorological data, obtained from the National Meteorological Information Center, cov- ers daily observations from 1 January 1951 to the present, including variables such as daily average temperature, maximum and minimum temperatures, 20-20-hours precipitation, aver- age relative humidity, and average wind speed. The article focuses on data from 833 continu- ously operating meteorological stations between the years 2000 and 2016. Using a nearest- neighbor approach based on station coordinates and county-level centers, each county is matched with its closest meteorological station. City-level meteorological data are derived by averaging data from all stations within the same city. Firm export data are mainly taken from the China Customs database (2000–2016). It con- tains import/export information at firm-product-destination-year level. The product codes in the customs database have undergone changes in various years, such as 2002, 2007, and 2011. PLOS ONE Climate effects on the diversification strategy of export firms PLOS ONE | https://doi.org/10.1371/journal.pone.0308391 August 16, 2024 6 / 28