Digital economy as a catalyst for low-carbon transformation in China: new analytical insights

The paper investigates whether and how the digital economy (DE) promotes low-carbon transformation in China by improving carbon total factor productivity (CTFP). DE has rapidly developed and is central to China’s strategy for high-quality, green development. Yet, two main challenges impede credible estimation of DE’s impact on CTFP: (1) measuring DE comprehensively across infrastructure, platforms, technologies, and new business models; and (2) addressing endogeneity arising from reverse causality between DE and CTFP and from correlations between measurement error of CTFP and regression residuals. The study proposes a comprehensive DE indicator and an endogenous stochastic frontier analysis (SFA) framework that jointly measures CTFP and estimates DE’s impact while mitigating endogeneity. Using Chinese prefecture-level city data (2004–2017), the paper finds a positive, significant effect of DE on CTFP and explores mechanisms (efficiency improvements, green innovation, and enhanced environmental regulation) and heterogeneity (public green awareness, resource abundance, and human capital). The study provides policy implications to leverage DE for China’s low-carbon goals and highlights its contributions: a novel one-step endogenous SFA addressing endogeneity, a comprehensive DE indicator, and empirical evidence for carbon reduction potential via balanced DE development.

Research on DE’s green development effects has recently expanded. Prior work documents DE’s benefits for economic growth, firm innovation, organizational optimization, energy efficiency, and total factor productivity. Studies indicate DE and its components (e.g., digital finance, internet, big data) can reduce CO2 emissions, enhance green total factor energy efficiency, and promote urban green development efficiency. However, many rely on partial DE proxies (e.g., internet penetration, digital finance), omitting broader DE technologies and business models. On CTFP measurement, total factor approaches (CTFP) improve on carbon intensity by incorporating inputs (capital, labor, energy) and outputs but DEA-based measures face issues with statistical noise and outliers. SFA offers robustness but many studies adopt two-step procedures that can yield biased estimates due to omitted variables and measurement error. Traditional one-step SFA also risks bias if DE correlates with inefficiency or two-sided errors. Recent methods (e.g., Karakaplan and Kutlu) handle endogeneity in SFA more fully. Gaps remain: unified DE measurement is lacking; endogeneity is insufficiently addressed; and evidence on DE’s impact on CTFP and mechanisms in China is limited. This study addresses these gaps with a comprehensive DE index and endogenous SFA.

Data: Panel data for 223 Chinese prefecture-level cities, 2004–2017. The period is constrained by the unavailability of city-level fixed asset investment data beyond 2017–2018 and limited trade data prior to 2004. Sources include the China City Statistical Yearbook, the Enterprise Search database, and listed companies’ annual reports. CO2 emissions are from the Center for Global Environmental sources noted in the paper. CTFP measurement: Inputs and outputs include capital stock (constructed via perpetual inventory method using fixed asset investment), labor, energy (electricity consumption at city level), GDP (gross industrial/output), and CO2 emissions. CTFP is defined via the carbon Shephard distance function D(K,L,E,Y,CO2) as CTFP = 1/D(.) with the environmental production technology set PT capturing joint production of desirable output and CO2. Model: A translog specification of the carbon Shephard distance function is used. By exploiting homogeneity in CO2, the estimating equation is rearranged to express the negative of log CO2 as a function of logged and mean-deviated inputs/outputs, interactions, and time trends, plus a composite error with a two-sided noise term and a one-sided inefficiency term. Inefficiency u_it is modeled with variance depending on Z (including DE), and endogeneity arises if variables affecting inefficiency are endogenous. The study employs endogenous SFA following Karakaplan and Kutlu (2017, 2019) to address: (a) reverse causality between DE and CTFP; (b) correlation between the two-sided error and one-sided inefficiency; and (c) correlation between measurement error of CTFP and regression residuals. Digital economy (DE) indicator: Constructed to reflect four connotations—(1) digital information (share of information transmission and computer service employees), (2) internet platform (telecom business revenue, number of broadband subscribers, number of mobile phone subscribers), (3) digital technology (penetration of digital technology applications in listed companies), and (4) new economic/business models (digital trading foundation level). Baseline construction uses principal component analysis; robustness uses normalized indicators with arithmetic mean (Bai and Zhang, 2021). Controls and instruments: Controls include socio-economic variables consistent with Table 2 (e.g., tertiary GDP, trade, FDI). To address endogeneity, instruments interact historical communications infrastructure with recent internet usage: (i) number of post offices per 10,000 people in 1984 × lagged (previous year) number of internet users; (ii) number of telephones per person in 1984 × lagged internet users. First-stage regressions include city and year fixed effects and standard controls; F-statistics exceed 10, indicating no weak IV problem. Estimation and robustness: Baseline endogenous SFA with translog frontier; comparison to exogenous treatment (Model EX) illustrates bias when endogeneity is ignored. Robustness checks include alternative IV sets, alternative DE indicator construction, and a Cobb–Douglas (C–D) frontier specification. Mechanism analysis uses proxy outcomes for: efficiency improvement (TFEE), green innovation (shares of green utility model patent applications and grants), and environmental regulation (industrial solid waste disposal rate), instrumenting DE as above. Heterogeneity analyses split cities by public green awareness (CGSS 2013 behavior score), resource abundance (share of mining employees), and human capital (share of college-and-above students in population).

- Digital economy (DE) significantly increases carbon total factor productivity (CTFP) after addressing endogeneity via endogenous SFA. Ignoring endogeneity understates the positive effect and overestimates frontier efficiency levels. - Magnitude: A 10% increase in DE raises CTFP by about 0.25% in the baseline endogenous SFA results (significant at the 1% level). Endogeneity tests (e.g., eta test X≈15.36, p=0.000) reject DE exogeneity. IV first-stage F-statistics exceed 10 (e.g., F≈327), indicating strong instruments. - DE development trend: The DE index rose steadily from 0.22 (2004) to 1.17 (2017), averaging 13.7% annual growth. South China leads, followed by East China; Northwest and Northeast lag. - Mechanisms: DE improves CTFP via three channels—(i) efficiency improvement in production/operations (TFEE: positive and significant), (ii) green innovation (higher shares of green utility model patent applications and grants; positive and significant), and (iii) strengthened environmental regulation and governance (higher industrial solid waste disposal rate; positive and significant). Instrumental-variable Kleibergen–Paap F-statistics exceed critical thresholds (e.g., ~58–61), indicating no weak IV problem. - Heterogeneity: • Public green awareness: In regions with stronger awareness, a 10% increase in DE raises CTFP by about 0.37%; with weaker awareness, the effect is not significant. • Resource abundance: In resource-rich regions, DE’s effect on CTFP is insignificant; in regions with lower resource abundance, a 10% increase in DE raises CTFP by about 0.49% (evidence of a resource curse moderating effect). • Human capital: In higher human capital regions, a 10% increase in DE increases CTFP by about 0.43%; effects are insignificant in lower human capital regions. - Robustness: Results hold when (i) changing IV construction (telephone/post office interactions with lagged internet users), (ii) altering DE indicator construction (normalization and averaging), and (iii) using a Cobb–Douglas frontier instead of a translog. Endogeneity tests consistently support using endogenous SFA. - CO2 reduction potential from balancing regional DE: Under three scenarios of reducing DE disparities (within-province, within-region, nationwide), estimated CO2 reduction potentials (million tons) for China are 564.91, 1973.39, and 3502.86, respectively. As shares of 2020 global CO2 emissions (31.98 Gt), these are about 1.8%, 6.2%, and 11.0%; relative to China’s 2020 emissions (~9893.5 Mt), they are about 5.7%, 20.0%, and 35.4%, respectively. East China shows the largest potential; Southwest the smallest.

The research question—whether and how the digital economy fosters low-carbon transformation by improving CTFP—is addressed through an identification strategy that corrects for endogeneity in frontier estimation. The positive causal effect of DE on CTFP confirms that digitalization contributes to cleaner and more efficient production by enhancing management efficiency, enabling green innovation, and strengthening environmental regulation via digital monitoring and data analytics. The heterogeneity results demonstrate that enabling conditions—public green awareness, lower resource dependence, and higher human capital—amplify DE’s impact on CTFP, suggesting that institutional, structural, and human capital contexts condition the effectiveness of digitalization in driving low-carbon productivity gains. Robustness across alternative IVs, DE measures, and production function specifications reinforces the credibility of the findings. Policy-wise, results imply that closing regional DE gaps can unlock substantial CO2 reduction potential, and that complementary investments in public awareness and human capital are essential to maximize the productivity and environmental benefits of digitalization.

The study proposes and implements a novel endogenous SFA framework that jointly measures CTFP and estimates the causal impact of DE, overcoming biases from reverse causality and error correlation. It constructs a comprehensive city-level DE indicator spanning digital information, platforms, technologies, and new business models. Empirically, DE significantly improves CTFP in Chinese cities, with stronger effects where public green awareness is higher, resource abundance is lower, and human capital is higher. Mechanism analyses confirm that DE operates through efficiency gains, green innovation, and improved environmental regulation. Scenario analyses show large national CO2 reduction potentials from balanced DE development across provinces and regions. Policy recommendations: (1) Continue advancing DE to achieve win–win gains in productivity and emissions reduction; (2) promote energy-efficient technologies and increase incentives for corporate green innovation; (3) deepen integration of DE with green technologies (e.g., big data, cloud computing, 5G) for environmental monitoring and energy management; (4) enhance public green and low-carbon awareness; (5) build multidisciplinary talent linking DE with green technology; and (6) pursue balanced regional DE development—accelerate digital park construction in leading cities while fostering advanced digital enterprises and collaboration with energy-intensive industries in lagging regions.

The analysis adopts a macro, city-level perspective due to data availability, which may mask firm-level heterogeneity in digital adoption and productivity dynamics. The time window (2004–2017) reflects constraints in city-level fixed asset and trade data, potentially limiting generalizability to later years. While the endogenous SFA and IV strategies address key endogeneity concerns, residual unobserved time-varying factors could remain. Future research at the firm level could provide micro-level causal mechanisms and finer policy guidance.