Will temperature affect the export quality of firms? Evidence from China

Abstract

Purpose

This study aims to investigate whether temperature affects the product quality of exporters and whether the effect is non-linear. More specifically, whether the impact of high temperatures differs from the impact of low temperatures, and whether different types of companies or industries are affected differently.

Design/methodology/approach

The paper uses detailed data covering all Chinese exporters from 2000 to 2016 to estimate the effects of temperature on the product quality of export firms. To clarify the relationship between them, the authors use a semi-parametric regression method, trying to test whether there is a non-linear relationship between temperature and the export quality of firms.

Findings

The increase in the number of high temperature days significantly reduces the quality of exported products, and this negative effect increases as the temperature rises. High temperature has the most significant negative impact on export quality for firms with low technical complexity, private firms and firms with no intermediate imports and located in historical hot cities. Product quality of both labor-intensive and capital-intensive firms will be affected by heat. High temperatures have the greatest negative impact on the export quality of newly entering products, followed by exiting products, with the least negative impact on persisting product.

Originality/value

To the best of the authors’ knowledge, this paper is the first to examine the impact of temperature on the quality of economic development. The findings of this paper again show that the potential economic impacts of global warming are huge. In addition to some potentially devastating impacts in the future, global warming is already causing imperceptible impacts in the present. Public and economic agents need to fully understand the possible adverse impacts of climate change and take corresponding adaptation measures to cope with global warming.

Citation

Zhang, J. and Li, H. (2023), "Will temperature affect the export quality of firms? Evidence from China", International Journal of Climate Change Strategies and Management, Vol. 15 No. 4, pp. 493-514. https://doi.org/10.1108/IJCCSM-05-2022-0066

1. Introduction

Due to global warming, extreme climate events have intensified. As such, since the 21st century, climate change has become a key topic for most countries. All countries and economies have been more or less affected, and climate change has become a common challenge throughout the world. In December 2015, 195 countries and the European Union (EU) joined the Paris Agreement at the Paris Climate Change Conference, demonstrating the importance the international community attaches to climate change and the urgency of addressing the issue.

As far as China is concerned, climate change is an inescapable issue. On the one hand, as a rapidly developing country, China is also a major carbon emitter. As such, it is a top priority for China to strike a balance between development and emission reduction in its response to climate change. On the other hand, China is one of the most vulnerable regions to climate change, due to its vast territory and large population. Since the second half of the 20th century, China’s temperature increase rate has been significantly higher than that of the global corresponding period (Green Paper on Climate Change in China, 2019). Figure 1 shows the changing path of annual days with temperatures greater than or equal to 30°C from 1951 to 2019. The figure clearly shows that the days with extreme temperatures greater than or equal to 30°C exhibit an upward trend, with a faster growth rate after 2000.

What are the economic consequences of global warming? Since the representative studies of Jones and Olken (2010) and Dell et al. (2012), a batch of empirical studies using data from different countries and regions have emerged, and climate economics has gradually become a research hot spot. However, most of the existing literature focuses on the “quantity” relationship between climate and economic development, but few studies have looked into whether climate will also affect the “quality” of economic development. In fact, it is of great practical significance to focus on “quality.” China has experienced rapid economic growth with the help of demographic dividend. However, with the increasingly severe impact of an aging population and rising labor costs, the factor-driven trade growth model cannot be sustained. This is accompanied by China’s emphasis on high-quality development. Whether climate change will be a factor affecting China’s high-quality development is a topic worth exploring. Current studies on the economic impact of climate change are not sufficient or comprehensive. This paper contributes to the existing literature and aims to fill this gap. In the context of China’s increasing emphasis on high-quality development, it is instructive to investigate whether global warming will affect the quality of export firms, and to promote firms and the public to comprehensively understand climate change and take corresponding steps to cope with and curb this change.

Throughout the existing literature, macro data are mostly used to investigate the impact of climate change on economic output and growth (Jones and Olken, 2010; Dell et al., 2012; Deryugina and Hsiang, 2014; Park, 2016; Colacito et al., 2019; Karlsson, 2021; Li et al., 2015). Some studies have used micro-level firm data to investigate the impact of climate change on firm performance (Martin et al., 2011; Cachon et al., 2012; Adhvaryu et al., 2020; Sudarshan and Tewari, 2014; Chen and Yang, 2019; Zhang et al., 2018; Li et al., 2021). However, the macro-economic impact of climate change is rooted in the impact on micro-individuals. Clearly, the current research focusing on micro-level individuals in the field of climate economics is not sufficient.

Compared with the existing literature, the main contributions of this paper are as follows. First, to the best of our knowledge, our paper is the first to examine the impact of climate change on firms’ export quality. As previously mentioned, most of the existing literature has looked into the economic impact of climate change from the “quantity” perspective. However, we extend the study of climate economics to the impact of “quality,” which aims to provide a deeper understanding of the economic impact of climate change. Second, we use a rather unique data set – the Chinese Customs Database, as well as meteorological data from China’s National Meteorological Information Center (NMIC) from 2000 to 2016. The data cover a significant time span, and virtually covers all of China’s export firms. Therefore, the large sample size allows the research conclusions of this paper to be more representative and of greater practical significance. Additionally, as a transitional economy, and the largest developing country in the world, China has its own particularity in economic development. By using Chinese data, we present empirical evidence and facts of China, which is conducive to enriching the research on climate economics. Finally, from the perspective of the research content, this paper further investigates the heterogeneity of the impact of temperature, so as to deepen our understanding of how climate change affects the quality of firms’ export products. We also investigate the role of firms’ export dynamics to make the research more comprehensive. At present, the general public’s understanding of the impact of climate change is still relatively superficial. Comprehensive investigations of the possible impact of global warming on the economic activities of firms are conducive to enhancing firms understanding of the negative impact of global warming. At the same time, encouraging firms to assume social responsibility to deal with global warming, such as investing in clean production and reducing carbon emissions, so as to actively contribute to curbing global warming.

The rest of the paper is organized as follows. Section 2 reviews the most relevant literature and provides theoretical discussions regarding the relationship between temperature, the ability of workers and export quality. Section 3 introduces the econometric model and data. Section 4 provides our main empirical results. Section 5 further discusses the issue of the impact of export diversity and intra-firm product dynamics. Finally, Section 6 summarizes the paper with main conclusions.

2. Theoretical analysis

Product quality upgrade can be reflected from multiple dimensions, one of which is the improvement of objective characteristics of the product, including durability, compatibility, security and supporting services (Garvin, 1984). The upgrading of product quality in this dimension mainly comes from the upgrading of a firm’s technology, as well as the sophistication and optimization of the production process. In this process, the quality and ability of employees play an indispensable role. Workers must strictly abide by the process, and also master the necessary skills and relevant knowledge to ensure the production of high-quality products. If not, the quality control of the firm will be negatively affected.

Studies in human biology have found that the brain’s functioning capacity is highly sensitive to temperature (Schiff and Somjen, 1985). For example, when ambient temperature increases, the brain’s ability to process waste heat is weakened (Heal and Park, 2013), and therefore, the individual’s cognitive ability will decline (Lepori, 2016). Additionally, some scholars have observed the differences in productivity of subjects under different temperatures through modern laboratory experiments. Here, they found that the working efficiency, ability to handle complex tasks, cognitive ability, alertness and mental arithmetic ability of subjects were significantly affected by temperature, with higher temperatures often being a negative factor (Grether, 1973; Seppanen, 2003; Niemelä et al., 2002; Wargocki and Wyon, 2007; Lee et al., 2014). Hence, it seems reasonable to suggest that higher temperatures may increase the probability of workers making mistakes, therefore causing more issues when it comes to implementing delicate and complex processes, thus reducing product quality. However, there is not enough evidence to show that lower temperatures will have a negative impact on human cognitive ability. As such, it is difficult to determine whether lower temperatures will reduce the quality of exported products. In which case, it remains an empirical question.

Additionally, part of the literature has shown that temperature not only impacts labor efficiency, but also capital efficiency (Lilja, 2005; Chen and Yang, 2019; Zhang et al., 2018). For instance, extreme temperatures have the ability to negatively affect the performance of certain machinery and equipment. Thus, temperature may also affect the quality of exports by affecting the efficiency of capital factors. Finally, adverse temperature shocks may affect the supply of domestic intermediate goods (Sudarshan and Tewari, 2014), and its quality may be negatively affected. A large body of literature points to the use of high-quality intermediate goods as an important way for firms to improve product quality (Fan et al., 2015; Bas and Strauss-Kahn, 2015; Manova and Yu, 2017). Therefore, adverse temperature shock could reduce the quality of export products through the intermediate goods channel.

Based on the above discussions, we will discuss in detail the empirical evidence surrounding the relationship between temperature and quality of export products. On this basis, we further examine whether certain firms might be affected more than others, allowing us to gain insight into which types of firms are climate-sensitive and why they might be affected differently.

3. Empirical strategy, variables and data

3.1 Empirical strategy

We refer to Deschênes and Greenstone (2007) and Zhang et al. (2018) to establish a semi-parametric model to investigate the impact of temperature on the export quality of firms. The model is constructed as follows:

(1) Qualikct=α+∑mβmTbinctm+∑nηnPbinctn+δXct+γik+γrt+γkt+εikct

where firms are indexed by i, products are indexed by k, and we define the product according to the harmonized system (HS) two-digit level; cities are indexed by c; years are indexed by t. The dependent variable Qual_ikct represents the quality of product k exported by firm i in city c in year t. Tbinctm is the number of days in year t experienced by prefecture city c with daily temperatures that fall in bin m. Pbinctn is the number of days in year t experienced by prefecture city c with daily precipitation that fall in bin n. X contains a series of other city-level climate-related control variables, including average wind speed (Wind), sunshine duration (Sun), humidity (Wet) and average atmospheric pressure (Atmo). γ_ik represents the firm-product fixed effect to control firm characteristics, product characteristics and firm-product-level characteristics that do not change with year, such as the altitude and slope of a firm’s location; γ_rt represents the province-year fixed effect to control for macroeconomic shocks and economic trends at the provincial level; γ_kt represents product-year fixed effect; ε_ikct represents the error term. Referring to Zhang et al. (2018), we allow standard errors to be correlated within a given firm as well as within city-year, thus clustering the standard errors to the firm and city-year levels (Cameron et al., 2011). Among them, β^m is the coefficient of most interest. In particular, β^m represents how the quality of firms’ exported products change with each additional day in the temperature bin m compared with the omitted category. According to the theoretical analysis, it is expected to be negative, i.e. adverse temperature shock will reduce the quality of export products, in which the negative impact of high temperature shock may be more significant, while the impact of low temperature shock has yet to be tested.

3.2 Variable declarations

3.2.1 Explanatory variables.

The temperature bins Tbinctm are constructed as follows. Referring to Deryugina and Hsiang (2014), we set each bin as 3°C wide, where m = 1 represents the temperature bin less than −12°C; m = 2 represents the temperature bin of [−12,−9)°C; the last bin is temperatures greater than or equal to 30°C. Finally, there are 16 temperature bins. In the regression model, we take [15,18)°C as the omitted category [1]. The precipitation bins Pbinctn are constructed as follows. We construct six precipitation bins, n = 1 represents the precipitation bin 0 mm; n = 2 represents the precipitation bin (0, 10) mm; n = 3 represents the precipitation bin [10,25)mm; n = 4 represents the precipitation bin [25,50)mm; n = 5 represents the precipitation bin [50,100)mm; n = 6 represents the precipitation bin [100,250)mm; n = 7 represents the precipitation bin greater than or equal to 250 mm; and n = 3 is the omitted category [2]. Variables (i.e. Wind, Sun, Wet and Atmo) are strongly correlated with temperature and precipitation, and the omission of these variables may cause estimation bias (Chen and Yang, 2019). These annual climate variables are obtained by averaging the daily data, and the variance of humidity and air pressure is too large, which is smoothed by taking logarithmic values.

3.2.2 Export quality.

The current methods of measuring the quality of export products are roughly divided into four types. The first method uses unit value as a proxy variable of product quality (Schott, 2004). The second method ranks the quality of a particular product according to its characteristics and consumer preferences (Crozet et al., 2011). The third method starts from the demand side perspective, introduces product quality into the utility function and calculates this quality by using a retrospective method (Khandelwal et al., 2013). The fourth method is proposed by Feenstra and Romalis (2014). They combine the demand and supply sides to build a model, thereby incorporating quality decision issues and endogenous product quality. According to the availability of data and our research needs, following the practice of most scholars, the third method has been used to measure the quality of export products. The specific calculation method is as follows.

First, the utility function of consumers in country m is as follows:

(2) Um={∫ω∈Ω[qm(ω)λm(ω)]σ−1σdω)}σσ−1

where U_m is the utility function of country m. q_m(ω) is the quantity of product ω exported to country m. Ω represents a series of products exported by the country. The formula assumes that the utility of consumers in the export destination country depends on both the quantity and quality of products consumed. According to utility maximization, the following demand function can be obtained:

(3) qivmt=λivmtσ−1pivmt−σPmt1−σYmt

where q_ivmt, λ_ivmt and p_ivmt, respectively, represent the quantity, quality and price, respectively, of product v exported by firm i to country m in year t. P_mt, Y_mt is the price index and total output of country m in year t. We then take the natural logarithm on both sides of equation (3). To overcome the endogenous problem of price, we move the price to the left side of the regression equation. Then we get:

(4) ln⁡(qivmt)+σln⁡(pivmt)=γv+γmt+εivmt

We will use equation (4) to measure the quality of export products. quality_ivmt = ln(λ_ivmt) = ε_ivmt/(σ−1), and we set σ = 5. In the later robustness tests, we will take different values for σ . γ_v is a product fixed effect, γ_mt is the export destination country-year fixed effect. We use the Chinese customs database to measure the quality of export products at the firm-destination country-HS6 level according to equation (4). Then, we add the export product quality data at the firm-destination country-HS6-year level to the firm-HS2-year level using the export value share as the weights. The resulting equation is as follows:

(5) Qualikt=∑imvt∈ikt(valueivmt∑imvt∈iktvalueivmt×qualityivmt)

where value_ivmt is the export value of HS6 product v exported by firm i to country m in year t, and k represents HS2 product.

3.3 Data

This article focuses on two large databases. First, day-station level weather data from the NMIC, from 1951 to the latest date, was obtained. Variables in the meteorological database include daily mean temperature and daily maximum temperature, daily minimum temperature, precipitation at 20–20 h, average relative humidity, average wind speed and a series of important meteorological data. In this paper, we use the continuous data of 833 stations from 2000 to 2016, and referred to Chen and Yang (2019) to obtain the daily meteorological data of 312 prefecture-level cities by taking the mean value of the station data of the same city. Finally, the meteorological data at the prefectural city’s annual level are constructed according to the meteorological data at the prefectural city’s daily level. Additionally, we use the administrative district code to uniquely identify the same city. As the administrative district code has been changed in the data range used in this paper, we have unified it into the 2019 version.

Second, we also use Chinese customs data from the China’s General Administration of Customs. The data include import and export information at the firm-product-destination country-year level from 2000 to 2016. We only retain export data and use this data to measure export quality at the firm-product level. The product code in the customs database has been changed many times. As such, we have unified the customs product code to the 2002 version. Each firm is uniquely identified by a ten-digit code in the customs database, with each digit having a specific meaning, among which the first four digits represent a firm’s location. Based on this, we can identify the prefectural city where the firm is located, and finally, merge it with the weather data according to the prefectural city’s administrative division code. We finally obtain a sample size of 10,619,712 for empirical analysis.

4. Empirical results

4.1 Benchmark estimation results

Table 1 shows the benchmark regression results of changing control variables. In Column (1), we control all the fixed effects without adding any control variables. All regression coefficients of temperature bins are negative. Most of the regression coefficients of temperature bins higher than 18°C are significantly negative, while almost all regression coefficients of temperature bins lower than 18°C are not significant. In Column (2), we add all the control variables, and the results show that most coefficients of temperature bins are negative, in which all coefficients of temperature bins higher than 18°C are significant at the 1% level. Regarding the absolute value of the coefficient, with a rise in temperature, the negative impact of export quality is greater.

Compared with the results in Column (1), the negative effects of the temperature bins between 3 and 24°C decrease with the inclusion of a series of temperature-related weather variables. However, the negative effect of the temperature bins greater than 24°C increases and is accompanied by an increase in significance. Overall, the inclusion of control variables does not change the basic conclusions of the paper. Looking at the control variables, there is no non-linear relationship between precipitation and the quality of a firm’s exported products, which is similar to Zhang et al. (2018); unexpectedly, when the number of days with 100–250 mm of precipitation increases, it significantly improves the quality of exported products. Additionally, the regression coefficient of wind speed is significantly negative, indicating that high wind speed is detrimental to production – a finding similar to Zhang et al. (2018).

To be more intuitive, we graphically show the trend of the regression coefficients of the core explanatory variables (Figure 2) [3]. As most of the regression tables are similar in structure to Table 1, to save space, we have followed other related studies by adopting the same way of presenting the main results, such as in Figure 2 – the graphical visualization of the estimation results. In this instance, Figure 2(a) corresponds to the results of Column (1) in Table 1; Figure 2(b) corresponds to Column (2), after controlling all control variables. When the temperature is higher than 18°C, the regression coefficient of all temperature bins is significantly negative, and as the temperature increases, the negative effect increases.

According to the results, the negative influence of temperature bin ≥ 30°C is the largest, and the coefficient in this bin shows that each additional day of temperature greater than or equal to 30°C will reduce the quality of firms’ export products by 0.19%. On average, from 2000 to 2016, there were 7.37 days in the temperature bin ≥ 30°C each year. The existence of 7.37 days in a year would reduce the export quality by 1.4%. Similarly, the regression coefficient of the temperature bin 27–30°C shows that each additional day in the temperature bin 27–30°C in a year will reduce the firm’s export quality by 0.17%. The average number of days in the temperature bin 27–30°C in each year from 2000 to 2016 is 27.18. This would reduce the quality of exports by 4.62%. Overall, the above results show that higher temperatures have a significant negative effect on export quality, while lower temperatures basically have no significant effect, which is consistent with our theoretical analysis.

4.2 Robustness test

First, we change the standard error clustering level. Different from the baseline regression, standard errors are clustered to the level of prefectural cities, allowing correlation of error terms at the level of prefectural cities. The regression results are shown in Figure 3(a). It can be seen that the regression coefficients of high temperature bins are significantly negative, indicating that the increase of high temperature days is indeed not conducive to improving firms’ export quality, which is relatively consistent with the baseline regression conclusion.

Second, we consider the impact of extreme samples. Taking into account that the results of this paper may be influenced by the extreme values of the sample, we will drop the dependent variable Qual_ikct beyond the upper and lower 5%. The regression results are shown in Figure 3(b). Overall, the regression coefficients of all temperature bins higher than 18°C are significantly negative, indicating that our basic conclusion is still valid after extreme samples are excluded.

Third, we aim to determine the influence of lagged temperature variables. In the benchmark regression, we only take into account the impact of current temperature distribution, without considering the influence of the temperature distribution in the previous years. The temperature variables may actually have strong serial correlation. Thus, without the lagged temperature variables in the baseline regression, we may not be able to cleanly identify the economic impact of the current temperature. Additionally, adding the lagged term of temperature variables into the regression also allows us to further investigate whether temperature has a long-term impact on export quality. Several scholars have pointed out that climate variables not only have a short-term impact on economic activity, but their long-term impact is also worthy of concern (Fishman et al., 2019; Kahn et al., 2019). In view of this, the lagged first-order values of all temperature bins are added to the baseline regression, and the regression results are shown in Figure 3(c). The regression coefficients of all temperature bins higher than 18°C in the current values are still significantly negative, and the absolute values of the coefficients are bigger compared to the baseline result, indicating that our basic conclusion is still robust after controlling the lagged terms of temperature. However, the regression coefficients of the lagged items in each temperature bins are basically not significant, indicating that the temperature distribution of the previous year has no significant impact on the export quality of firms this year.

Fourth, in the baseline regression, we take into account weather conditions on both weekdays and weekends. However, most economic activity actually takes place on weekdays. Since the promulgating of the regulations on working hours for employees in 1994, China began to implement the double day-off system, with most production activities taking place from Monday to Friday. Yet, at the same time, we should also note that this institution is usually strictly enforced in state-owned and foreign-owned firms, while many private firms – especially small and medium-sized firms (SMEs) that are paid on a piece-rate basis – usually work extra hours on weekends. Therefore, it seems appropriate to examine whether higher temperature weekends or weekdays drive our results. Specifically, referring to Deryugina and Hsiang (2014), we divide the number of days in each temperature bin into the number of weekdays that fall within the temperature bin and the number of weekend days that fall within the temperature bin, and jointly estimate their effect on export quality. The regression results are shown in Figure 3(d) and (e). The results show that the negative effect of high temperature on the quality of exported products is driven by the temperature distribution on weekdays rather than weekends. Upon closer inspection of the regression coefficients of the weekday temperature variables, we find that the absolute values of the coefficients have increased substantially compared to the baseline regression (Column (2) in Table 1). On the contrary, an increase in the number of hot days on weekends can improve product quality. This result is similar to Deryugina and Hsiang (2014), who used US data to show that an increase in the number of hot weekend days raises US total income per capital. Additionally, an increase of cold weekends significantly inhibits product quality improvement, which is also similar to Deryugina and Hsiang (2014). Overall, the results for the full sample are consistent with those for weekdays due to the greater effect of weekdays.

Finally, we reconstruct the explanatory and explained variables separately. First, we reconstruct the temperature variables by changing the width of the temperature bins. The width of temperature bin changes means the basic assumption of the regression has changed. In the following analysis, we set 6°C is a bin width, and take 12–18°C as an omitted category to perform the regression analysis, which means that we assume that in every 6°C wide temperature bin, the impact of temperature on export quality are the same. Figure 3(f) shows that the regression coefficients of all temperature bins are negative, among which the coefficients of temperature bins higher than 18°C are significantly negative, and the negative impact on product quality increases with the increase of temperature, while the low temperature bins are basically insignificant, which is consistent with the baseline conclusion. Second, we reconstruct the explained variable. In the benchmark regression, the elasticity of substitution in equation (4) is taken as a constant value σ = 5 when measuring product quality. According to Anderson and Van Wincoop (2004), the estimated elasticity of substitution is generally between 5 and 10. Therefore, we take the case where the elasticity of substitution is 10, measure the product quality and then estimate equation (1); again, the result is shown in Figure 3(g). Essentially, the high temperature bins had a significantly negative effect on export product quality, and the negative effect was greater as the temperature increased. Additionally, we consider the different elasticities of substitution for different products and use the price elasticities of different products estimated by Broda and Weinstein (2006) to estimate product quality again; the regression results are shown in Figure 3(h), which is similar to Figure 3(g). The increase in weather greater than 21°C will significantly reduce the quality of firms’ export products.

4.3 Heterogeneity analysis

Different types of firms may have certain responses to climate shocks due to their specific capital capacity and ability to deal with emergencies. In this section, we will investigate the heterogeneity impact of temperature shock.

4.3.1 Export technology complexity.

For firms with low export technology complexity, they often rely on factor price advantage to participate in the global value chain in the way of low-end embedding, and they are flooded with a large amount of low-quality and cheap labor. However, high-tech firms tend to rely more on specific asset investments than “human-sea tactics,” and employees generally have a better office environment. Therefore, the varying complexity of export technology may lead to a different impact of temperature shock on export quality. In view of this, we divide the sample into two groups: high export technology complexity and low export technology complexity. First, we calculate the export technical complexity of firms based on Hausmann et al. (2007) and Xu (2006). The equation is as follows:

(6) Prodyv=∑cxcv/Xc∑c(xcv/Xc)×pcgdpc

where the subscript v represents HS6 code products, and c represents a country or region. x_cv represents the product export value of a country or region. X_c represents the total export value of a country or region. pcgdp_c represents the actual per capita gross domestic product (GDP) of a country and region. Prody_v represents the technical complexity of the product. Then, according to the Chinese customs database, to measure the complexity of export technology at the firm level, the specific calculation formula is as follows:

(7) ESIi=∑v(xivXi)×qcvλ×Prodyv

where i represents firms, ESI represents the export technical complexity of firms and the indicator has been adjusted for product quality (Xu, 2006). X_i represents the export value of a firm, and x_iv represents product value exported by a firm. q_cv represents the product quality of country c. The specific calculation formula is as follows:

(8) qcv=pricecv/∑n(μnv×pricenv)

where price_cv represents the unit value of a product in a country or region. µ_nv indicates the proportion of a country’s or region’s export of a product in the world’s total export of that product. According to the existing literature, λ in equation (7) is set as 0.2. To measure the complexity of firms’ export technology, we need two additional sets of data. The first is the Comtrade Database of the United Nations Statistics Division, from which we can obtain the export data at the country-HS6 level. The second is real GDP per capita data from the World Bank database. After calculating the export technology complexity of firms, the samples are divided into two groups of firms, with high-technology complexity and firms with low-technology complexity according to the annual mean value of export technology complexity. We then estimate equation (1) using the two samples. The results in Figure 4 show that, from the perspective of significance, both kinds of firms’ export quality will be negatively impacted by high temperature. However, the negative impact on firms with low technological complexity are noticeably larger, especially under extremely high temperatures that are greater than 30°C. With each additional such day, the export quality of firms with low technological complexity will decrease by 0.23%, while the export quality of firms with high technological complexity is not significantly affected. As mentioned above, low technical complexity firms are often labor-intensive, and the work environment is usually worse. Therefore, when exposed to higher temperatures, workers are more likely to suffer negative effects in production, thus reducing the export quality of firms. Moreover, it should also be noted that under extremely high temperatures, the export quality of high-tech firms will no longer be significantly affected. The possible reason is that high-technology firms usually pay more attention to human capital and the protection of workers’ rights. Therefore, under extremely high temperatures, high-tech firms with brain-intensive workers are more inclined to adopt adaptive behaviors to cope with higher temperatures.

4.3.2 Firms with import of intermediate goods.

According to our theoretical analysis, we believe that temperature shocks may reduce export product quality through the domestic intermediate goods channel. If this channel does exist, one would expect exporters using domestic intermediates (compared to those using imported intermediates) to experience a greater decline of product quality under temperature shocks. To verify this idea, export firms are divided into two groups according to whether they import intermediate products. We then perform regressions separately using two kinds of samples. Referring to Feng et al. (2016), we identify the intermediate goods in the import database of China Customs, and then combine the import database and export database according to the firm code. With this, we can finally identify which firms import intermediate products. The regression results from different samples in Figure 5 show that for firms without importing intermediate products, their export quality will suffer more decline when high temperature days increase, which is consistent with the expectations mentioned above. Meanwhile, this finding also verifies the validity of the intermediate goods channel effect to a certain extent.

4.3.3 Firm ownership.

Firms with different ownership often face different financing constraints, profit maximization goals and institution implementation (Zhang et al., 2018), which will further influence firms’ behavior. At present, the studies using industrial firms’ database, complied by the National Bureau of Statistics of China, find that firms of different ownership when suffering from temperature shock tend to perform differently. Specifically, the adverse temperature shock will significantly reduce the total factor productivity and output of private firms, while the impact on state- and foreign-owned firms is small (Zhang et al., 2018). In view of this, we may consider that when subject to temperature shock, the export quality of different ownership firms will also be affected differently. We divide the sample into state-owned firms, private firms and foreign-owned firms [4]. We then regress separately using three groups of samples. We identify firms’ ownership according to the sixth digit of the firm code in the customs database. When the sixth digit equals to 1, the firm is state-owned; when it equals to 6, the firm is private; and when it equals to 4, the firm is foreign-owned. The regression results are shown in Figure 6. Similar to the results of Zhang et al. (2018), temperature shock does not have a significant negative impact on the export product quality of state- and foreign-owned firms. In fact, an increase in days with temperatures over 30°C improves the export quality of foreign-owned firms. However, the export quality of private firms is significantly negatively affected by higher temperatures. This finding can be easily explained, because foreign- and state-owned firms tend to have a better labor institution and stronger execution of institutions in China, and thus have a more comfortable working environment for employees. However, private firms often face financial constraints, and employees often work in relatively poor conditions, making them more vulnerable to uncomfortable temperature; hence, the quality of products they produce is likely to decline the most.

4.3.4 Historical temperature.

In the long term, cities may adapt to temperature shocks to reduce the negative effects of temperature. For example, cities that frequently suffer from heat waves may respond to the effects of high temperatures by optimizing the spatial layout of the city, increasing green vegetation cover and improving the heat dissipation capacity of buildings. To examine whether cities have sufficient climate adaptation investments that make hot cities more insensitive to the onslaught of high temperatures, the following regressions are conducted in this paper. Specifically, we determine whether a city is historically hot based on the city’s historical temperature averages over the past 30 years. First, we calculate the average temperature for the past 30 years for all sample cities. Second, based on the mean value of the historical temperature variable, the sample is separately divided into cities with high historical temperature and low historical temperature for the regression. The regression results are shown in Figure 7, which indicate that cities with higher historical temperatures still suffer from a greater number of high temperature shocks, resulting in a more significant decline in the quality of export products [5]. The cities with lower historical temperatures are largely unaffected by the extreme temperature. This finding suggests that historically, hot cities have not invested sufficiently in climate adaptation to mitigate the negative effects of high temperatures, and that overall, hot weather remains more detrimental to hotter regions.

5. Further analysis

5.1 Product category

Different product categories may have different sensitivities to changes in temperature. To further explore exactly which product sectors will lose more competitiveness in a future temperature rise scenario, in this section, we explore the differences in the impact of temperature on different product categories. To make the regression results comparable, we refer to Jones and Olken (2010) to classify the products. First, we matched the HS codes in the customs database with SITC rev2 codes and measured the product quality at the two-digit SITC code level. Second, estimate equation (1) separately for the two-digit SITC category. To conserve space, similar to Jones and Olken (2010), we report coefficients only for products where the effect of temperature bin greater than or equal to 30°C is statistically significant at the 10% level. The results are shown in Table 2.

The regression results are similar to Jones and Olken (2010), where the regression coefficients are significantly negative (20 out of 66 category) for most product types and significantly positive for only a few products (five out of 66 category). Looking at the results in Panel A, we find that the effects of higher temperatures on manufactured products are more widespread than on primary products. Moreover, both labor-intensive (e.g. leather and rubber manufactures, paperboard, textile yarn, fabrics, travel goods, handbags, articles of apparel and clothing accessories and so on) and capital-intensive products (e.g. inorganic chemicals, dyeing, tanning and coloring materials, medicinal and pharmaceutical products, essential oils, fertilizers, manufactured, telecommunications and sound recording apparatus and so on) have been negatively impacted by high temperatures. This result is in line with Zhang et al. (2018) and Karlsson (2021), who find exports of both labor- and capital-intensive industries to be negatively affected by high temperatures. This finding also partially corroborates our research mechanism that temperature shocks affect not only labor productivity, but also capital productivity. It is also worth noting that some technology-intensive products such as power generating machinery and equipment, office machines and automatic data processing equipment, electrical machinery, apparatus and appliances and road vehicles are not sensitive to extreme heat (the effect of the 27–30°C on these industries is also not significant).

5.2 The impact of intra-firm product dynamics

Existing studies have found some differences in the quality, efficiency and profitability of export for firms in different life cycles (Tang and Zhang, 2012). Exiting firms have the worst performance in all aspects, while entering firms tend to outperform exiting firms, thus achieving effective allocation of resources by way of dynamic elimination. Bernard et al. (2010) find that intra-firm product entry/exit contributes more to US manufacturing growth than growth due to firm entry/exit based on US data, and the dynamic intra-firm product switching is an important resource allocation method that is closely related to a range of characteristics, including firm efficiency.

Therefore, do export products in different dynamics behave differently when subjected to temperature shock? Within the firm, we divide the sample into new entry products, exiting products and persistent products, and then regress separately. New entry products are those that do not exist in year t-1 and exist in year t in a firm. Exiting products are those that exist in year t and do not exist in year t + 1 in the firm. Persistent products are those that exist in year t-1 and also exist in year t in a firm. The regression results in Figure 8 show that high temperature has a negative impact on a firm’s export quality, regardless of their intra-firm product dynamics. However, new entry products suffer the most, followed by exiting products, and persistent products are the least affected. Interestingly, low temperatures have a significant negative impact on the export quality of new entry products.

6. Conclusion and policy implications

Based on the daily high-frequency meteorological data from the NMIC and highly segregated firm data from the China Customs Database between 2000 and 2016, this paper uses a semi-parametric approach to examine the effect of temperature shock on the export quality of firms. The study finds that, first, the increase in the number of high temperature days significantly reduces export quality, and the higher the temperature, the greater the negative impact of temperature on quality. This conclusion holds after changing the standard error clustering level, considering the effect of extreme samples, taking into account the effect of lagged temperature variables, factoring the effect of annual days and changing the construction method of temperature bins, indicating the robustness of the conclusion. Second, the analysis based on firm characteristics finds that high temperature has the most significant negative impact on the quality of products exported by low-tech firms, private firms and firms that do not import intermediate goods and located in historical hot cities. Third, we further examine the effect of temperatures above 30°C for different product categories. We found that both labor- and capital-intensive industries are affected by heat shock, but the quality of products in high-tech industries is largely unaffected. Fourth, an examination of intra-firm product dynamics reveals that high temperatures have the greatest negative impact on the export quality of new entering products, followed by exiting products and persisting products.

To the best of our knowledge, this paper is the first to examine the impact of temperature on the quality of economic development and finds that heat shock significantly reduces the quality of firms’ export products. This suggests that focusing only on the “quantitative” impact of climate shocks on economic development as found in other studies underestimates the negative impact of climate shocks. Our findings show that the potential economic impact of global warming is highly significant. As such, public and economic agents must fully understand the possible adverse effects of climate change and take corresponding adaptation measures to cope with global warming. The sensitivity of different firms to temperature varies, with the negative impact on private low-tech firms being of particular concern. Such firms should improve their climate risk management capabilities and make appropriate climate adaptation investments to improve their working conditions, which will contribute to their international competitiveness. In terms of limitations, this paper does not empirically examine how temperature affects output quality by influencing workers’ status in the workplace, as well as the degree of job refinement due to data availability. Further data collection would be required to better validate this in future studies.

Figures

Figure 1.

Trend in the number of days every year with temperatures ≥ 30°C (1951–2019)

Figure 2.

Baseline estimation results that show a non-linear relationship between temperature and firms’ quality of export goods based on Model (1)

Figure 3.

Robustness tests

Figure 4.

Export technology complexity

Figure 5.

Whether a firm is importing intermediate goods

Figure 6.

Firms with different ownership types

Figure 7.

Heterogeneous effects of historical temperatures

Figure 8.

The impact of intra-firm product dynamics

Figure A1.

Firms with different ownership types (expansion of Figure 6). Panels (a), (b) and (c) show the effect of daily average temperature on the export quality of Sino-foreign firms, collective firms and individual businesses, respectively

Table 1.

Baseline regression estimation results

	(1)	(2)
Ind. var.	Qual	Qual
<−12°C	−0.0007 (0.001)	−0.0001 (0.001)
−12∼−9°C	−0.0003 (0.001)	0.0005 (0.001)
−9∼−6°C	−0.0004 (0.001)	0.0001 (0.001)
−6∼−3°C	−0.0007 (0.001)	−0.0002 (0.001)
−3∼0°C	−0.0009 (0.001)	−0.0002 (0.001)
0∼3°C	−0.0010 (0.001)	−0.0003 (0.001)
3∼6°C	−0.0011* (0.001)	−0.0007 (0.001)
6∼9°C	−0.0009 (0.001)	−0.0004 (0.001)
9∼12°C	−0.0015*** (0.001)	−0.0012** (0.000)
12∼15°C	−0.0005 (0.000)	−0.0003 (0.000)
18∼21°C	−0.0012*** (0.000)	−0.0011*** (0.000)
21∼24°C	−0.0013*** (0.000)	−0.0011*** (0.000)
24∼27°C	−0.0015*** (0.000)	−0.0016*** (0.000)
27∼30°C	−0.0012** (0.001)	−0.0017*** (0.001)
>=30°C	−0.0009 (0.001)	−0.0019*** (0.001)
0 mm		0.0001 (0.000)
0–10 mm		−0.0001 (0.000)
25–50 mm		−0.0010* (0.001)
50–100 mm		−0.0007 (0.001)
100–250 mm		0.0094*** (0.003)
>= 250mm		0.0020 (0.010)
Wind		−0.0387*** (0.011)
Sun		−0.0023 (0.007)
Wet		−0.0556 (0.081)
Atmo		2.4134 (2.697)
Constant	0.7253***(0.115)	−26.7565 (30.972)
Firm-by-product FE	Yes	Yes
Province-by-year FE	Yes	Yes
Product -by-year FE	Yes	Yes
Observations	10,619,712	10,619,712
R2	0.674	0.674

Notes:

*,** and ***denote significance at the 10, 5 and 1% levels, respectively. Standard errors in parentheses are clustered by firm and city-year levels

Table 2.

The effect of temperatures above 30°C on the quality of products exported by different products

SITC code	Product category description	Coefficient	St. error	t-stat
Panel A: negative and statistically significant products
00	Live animals chiefly for food	−0.0368**	0.0146	−2.53
05	Vegetables and fruit	−0.0035**	0.0016	−2.11
29	Crude animal and vegetable materials, nes	−0.0046*	0.0026	−1.78
35	Electric current	−0.0193**	0.0076	−2.53
42	Fixed vegetable oils and fats	−0.0143*	0.0076	−1.88
52	Inorganic chemicals	−0.0038**	0.0019	−1.96
53	Dyeing, tanning and coloring materials	−0.0046*	0.0023	−1.97
54	Medicinal and pharmaceutical products	−0.0052***	0.0020	−2.57
55	Oils and perfume materials; toilet and cleansing preparations	−0.0048***	0.0018	−2.63
56	Fertilizers, manufactured	−0.0072*	0.0038	−1.88
58	Artificial resins and plastic materials and cellulose esters etc.	−0.0024*	0.0014	−1.75
61	Leather, leather manufactures, nes, and dressed furskins	−0.0074***	0.0025	−2.96
62	Rubber manufactures, nes	−0.0057***	0.0020	−2.89
64	Paper, paperboard, and articles of pulp, of paper or of paperboard	−0.0021*	0.0012	−1.70
65	Textile yarn, fabrics, made-up articles, nes and related products	−0.0025**	0.0010	−2.41
68	Non-ferrous metals	−0.0041**	0.0016	−2.48
76	Telecommunications, sound recording and reproducing equipment	−0.0046*	0.0024	−1.89
83	Travel goods, handbags and similar containers	−0.0035*	0.0020	−1.71
84	Articles of apparel and clothing accessories	−0.0031**	0.0012	−2.52
89	Miscellaneous manufactured articles, nes	−0.0018*	0.0010	−1.77
Positive and statistically significant products
22	Oil seeds and oleaginous fruit	0.0079*	0.0043	1.84
63	Cork and wood, cork manufactures	0.0058**	0.0029	1.99
73	Metalworking machinery	0.0045*	0.0025	1.78
82	Furniture and parts thereof	0.0035**	0.0015	2.33
94	Animals, live, nes, (including zoo animals, pets, insects, etc.)	0.1341***	0.0320	4.18

Notes:

Panel A shows the SITC codes where the estimated extreme high temperature (higher than 30°C) effect is negative and statistically significant; Panel B shows those SITC codes where the effect is positive and statistically significant

Notes

1. As Chen and Yang (2019) pointed out, the reference bin is not chosen arbitrarily, and when the bin is used as the reference group, the regression coefficients of any other bins should be significantly negative, which indicates that this reference group is the optimal temperature bin and can satisfy the nonlinear hypothesis that temperature and export are convex functions. Additionally, the interval selected in this paper is also close to several existing studies (Burke et al., 2015; Burke and Tanutama, 2019; Deryugina and Hsiang, 2014).

2. The classification of precipitation bins is based on the China Meteorological Administration, who define rainfall of less than 10 mm in 24 h as light rain, 10–25 mm as moderate rain, 25–50 mm as heavy rain, 550–100 mm as a rainstorm, 100–250 mm as a heavy rainstorm and over 250 mm as extremely heavy rain.

3. There are many core explanatory variables in this article. To reflect the influence and trend change more clearly and intuitively of the regression coefficients in different temperature bins, according to the practice of the existing literature, in the following sections, we will use graphs to show the regression results. All the regression results are available upon request.

4. The firms in our sample are divided into seven categories according to their ownership type, including state-owned firms, foreign-owned firms, private firms, Sino-foreign joint ventures, Sino-foreign cooperative enterprises, collective firms and individual entrepreneurs. Among them, state-owned firms, foreign-owned firms and private firms occupied 87% of the sample, and in line with most articles, we focused on these three types of firms. Additionally, based on the reviewer’s suggestion, we regressed the sample on the remaining types of ownership firms. Specifically, we ran regressions on collective firms, Sino-foreign joint ventures or cooperatives and individual businesses, and the regression results are shown in Figure A1 in Appendix. Overall, Sino-foreign joint ventures or cooperative firms will not be affected by temperature shock, while the quality of exports of collective firms will be negatively impacted by high temperatures, and individual businesses are negatively impacted by low temperatures.

5. In Figure 7(a), the low temperature part starts from −6°C. This is because in the sample of cities with high historical temperatures, they have a minimum temperature of −6°C, and temperatures below −6°C have not occurred.