Page 134 - AJWEP-22-4

Wang, et al.

(eastern/southern), temperate (northern), and arid the day; if there were multiple stations within the city,
(northwestern) – which represent the wide spectrum the maximum and minimum values were calculated for
of climatic zones emblematic of the different all stations.
characteristics of extreme temperature exposure. We
analyzed 12,908 firm-year observations (2011 – 2022) 3.2. Mold
from 1824 A-share manufacturing firms, whose spatial In this study, we constructed an ordinary least squares
distribution reveals high-density clustering in the (OLS) model to analyze the impact of extreme
Yangtze River Delta (28.7% of firms) and Pearl River temperatures on the digital transformation of enterprises.
Delta (22.1% of firms). The western region, although Exrteme
covering a relatively lower density, all have enterprises Digital ij t,, 0 j t, 1 Extreme j t, 1 1 X i t,
included in the study sample. X WW W (I)
'
2 jt, 0 jt, 1 jt, 1 ij t,,
3.1.2. Company data Where i refers to enterprises, t is time, and j denotes
Shanghai and Shenzhen A-share listed companies from prefecture-level cities. The explanatory variable is the
2011 to 2022 were taken as the research subjects, and extreme temperature (Extreme ); the explanatory
j,t
their company-level data were obtained from the China variable is the degree of enterprise digital transformation
'
Stock Market and Accounting Research (CSMAR) (Digital ); and X and and X are a series of control
i,j,t
i,t
jt ,
database (https://data.csmar.com/) with the following variables at the enterprise level and macroeconomic
processing: first, retaining the data of manufacturing level, respectively, which are the other climate variables
companies; second, excluding samples such as special (including average wind speed, average humidity, and
treatment; and third, shrinking the tail at 1% level for hours of light) at the prefecture level. In the actual
continuous type variables. Municipal-level control regressions, industry-fixed effects as well as time-fixed
variables were obtained from the City Statistical effects were controlled for, and all standard errors were
Yearbook (https://www.stats.gov.cn/). clustered at the city×year level. Given the possible lags
in the effects of weather variables on various economic
3.1.3. Weather data variables, we also included the lagged term sums of
Weather data were obtained from China Meteorological these weather variables in the model.
Administration (CMA), which contains daily
observations of meteorological indicators such as 3.3. Variables
average temperature, maximum temperature, minimum 3.3.1. Explanatory variables
temperature, precipitation, barometric pressure, relative This study measured corporate digital transformation
humidity, sunshine hours, and average wind speed, and through text analysis of annual reports using keyword
37
provides detailed geographic coordinate information frequency metrics (Digital : absolute count; Digital :
2
1
for each weather station. The latitude and longitude frequency ratio). We identified five keyword dimensions
of the weather stations were matched with the cities in (artificial intelligence, blockchain, cloud computing,
the provinces to confirm the geographical zones before big data, and digital technology applications),
retrieving the daily observation data compiled by the collected all A-share listed firms’ annual reports from
domestic weather stations, and Python was used to CSMAR through Python, and extracted text using
process the cleaned data, which were compiled into a Java PDFBox. To ensure accuracy, we rigorously
CSV file on a daily basis, with the latitude and longitude excluded: (i) keywords preceded by negations (e.g.,
and the values of the daily average temperature and “not,” “non-,” “un-,” “lack,” “without,” and “failed to”)
extreme temperature retained as needed. The daily CSV within a ±5-word window through dependency parsing
files were spread and then projected, and the daily data and a predefined negation lexicon, and (ii) contexts
were interpolated using the inverse distance weighting referencing external entities (e.g., shareholders/
method. The data are partitioned and counted and suppliers’ activities) or executive backgrounds, retaining
spliced by administrative divisions, and finally, the day- only firm-owned digital initiatives. Digital aggregates
1
by-day meteorological data such as temperature and validated keyword counts; Digital computes their
2
humidity were obtained for each prefecture-level city proportion to total words, thereby eliminating false
in the country. For a city containing only one station, positives (e.g., “no AI deployment”) while capturing
the maximum and minimum temperatures observed genuine transformation actions (e.g., “launched our
at the station were taken as the extreme temperatures of blockchain system”).

Volume 22 Issue 4 (2025) 126 doi: 10.36922/AJWEP025210166