The matching problem between data characteristics and model structures is a critical factor that constrains the application of neural networks in economic and financial forecasting. Existing studies primarily focus on residual analysis, while relatively little attention has been paid to the exploration of model structure stability. To address this research gap, this study proposes a novel research paradigm based on neural network parameter analysis to quantify and evaluate the degree of matching between data and models. A model distance measurement method based on network parameters is designed, and a comprehensive evaluation framework combining model stability testing and residual analysis is established. Empirical results from the foreign exchange market demonstrate a significant nonlinear relationship between the stability of the LSTM model and the length of the lookback period. Among multiple evaluation dimensions, a 36-month lookback period achieves the best performance, effectively balancing the impact of short-term fluctuations and long-term trends. This study not only provides new analytical tools to enhance the reliability of neural network forecasting models but also further deepens the theoretical foundation of machine learning methods in financial applications.
In financial markets, information sources are diverse and structurally complex, and how to systematically integrate such information into tradable investment signals has become a central issue in asset pricing research and practice. This study constructs an LLM-based decision-making system, incorporating multi-source data such as news texts, stock prices, and macroeconomic conditions. Using the constituents of the CSI 300 Index from 2020 to 2024 as the sample, we systematically evaluate the predictive power of investment signals generated by LLMs. First, we construct three types of portfolios based on buy, hold, and sell signals. Analysis of cumulative returns and risk-return metrics shows that the buy-signal portfolio achieves significantly higher excess returns (23.83%) than the hold (6.96%) and sell (
Accurately predicting stock market volatility holds significant theoretical and practical importance for portfolio analysis, financial risk management, and derivative pricing, and has thus attracted extensive attention from scholars. Within the framework of the mainstream HAR-type volatility forecasting models, this paper examines the impact and predictive power of an informed trading index-constructed based on high-frequency trading data and ensemble neural networks – On the daily realized volatility of the A-share market. The study finds that the informed trading index serves as a positive predictor of market volatility. Its predictive capability remains robust across various model specifications and both in-sample and out-of-sample tests, and it performs particularly well during periods of high market volatility and overall economic downturn. Further research confirms that incorporating the informed trading index significantly enhances the out-of-sample forecasting accuracy of existing HAR-type models for market volatility. This not only contributes to the body of research on the market impact of informed trading factors but also provides valuable insights for advancing volatility forecasting modeling and analysis.
Industrial collaborative development has become an important direction for high-quality development. How to optimize the industrial structure while maintaining stable economic operation is an important issue in the new development stage. This paper measures and decomposes China’s economic growth trend using the multivariate unobserved components with outlier-adjusted (MUC-O) model, quantitatively analyzes the contributions and time-varying characteristics of industrial sectors as well as industrial collaboration and heterogeneous dynamics, and develops a decomposition-based impulse response function to investigate the effects of persistent shocks and large shock events on the economic growth trend. The research indicates that: 1) After years of gradual decline, China’s economic growth trend has stabilized and begun to improve since 2022, indicating stronger growth resilience. 2) The co-movement growth trend of industrial sectors closely mirrors the economic growth trend. Industrial collaboration is largely driven by the “real estate-finance-construction” value chain. 3) Industrial collaboration is the main force stabilizing or enhancing the growth trend, whereas the heterogeneous dynamics and the secondary sector largely account for its deceleration, although the latter plays a stabilizing role during large shocks. 4) The effects of persistent shocks last for 2
In the global exploration of central bank digital currency (CBDC), the effectiveness of promotion ultimately depends on whether the public can develop stable and sustainable usage demand. This paper constructs a structural money demand model to characterize the endogenous demand for the digital renminbi (e-CNY) through key features, including remuneration, credit risk, and payment convenience, and estimates the model using microdata from the China Household Finance Survey. The results indicate that the remuneration feature exerts a significant positive effect on potential holding and usage intentions, with interest-rate sensitivity substantially higher than that of other design features. The potential share of the e-CNY in households’ liquid asset portfolios ranges from 2.77% to 60.89%. Further analysis incorporating pilot program data shows that short-term transaction growth is mainly driven by exogenous incentives such as consumption subsidies, while the pattern of high balances and low transaction intensity suggests that endogenous demand has not yet been fully activated. The divergence between theoretical predictions and observed practice reveals a misalignment between exogenous policy interventions and endogenous utility in e-CNY adoption, providing quantitative identification to inform the design and promotion of the digital Renminbi.
Innovation is a crucial engine driving economic growth and social development. As the core carriers of innovation, cities’ innovative activities have drawn considerable attention for their impact on the stability of the financial system. This paper, based on the panel data of 180 cities in China from 2008 to 2021, constructs an index of systemic financial risk at the city level and employs a spatial econometric model to empirically analyze the impact of urban innovation capacity on systemic financial risk. The findings reveal that: Firstly, urban innovation capacity significantly curbs systemic financial risk, and this effect is more pronounced in regions with a higher level of government intervention. Secondly, systemic financial risk in cities exhibits a significant spatial spillover effect, with the most concentrated influence within the range of 80 to 180 kilometers, showing a clear spatial decay feature. This paper not only provides a new perspective for understanding the complex relationship between innovation and financial risk but also offers policy implications for the sustainable development of cities and the collaborative governance of regional risks.
The traffic economy has emerged as a novel driver of urban sustainable development, leveraging internet platforms to transform public attention into consumption momentum. Existing studies have explored the mechanism and economic benefits of effective digital traffic conversion, but have paid insufficient attention to the potential environmental costs of traffic economy. Quantifying and reducing this environmental cost is of great significance for realizing the sustainable and high-quality development of traffic economy. Taking the traffic event “Zibo barbecue” as the entry point, we analyze the impact of traffic economy on urban environmental pollution based on the air quality index after eliminating the interference of meteorological factors, by using machine learning synthetic control method. The findings revealed that: 1) The traffic economy can lead to environmental pollution. During the period of tourism incident “Zibo barbecue”, the air quality index of Zibo city increased by 15.58% on average. 2) Mechanism analysis shows that the network traffic triggered by the “Zibo barbecue” is effectively transformed into urban passenger traffic, and the surge in motor vehicle traffic exacerbates traffic congestion, which in turn exacerbates air pollution. At the same time, the barbecue boom to promote the catering industry to the main urban and county core area agglomeration, it is also an important factor in the deterioration of air quality due to the barbecue itself pollution emissions and catering layout of the agglomeration effect. 3) Further analysis shows that the increased health risk of this pollution is equivalent to an economic loss of approximately RMB 29.9 billion. This study reveals the potential environmental costs of the traffic economy under the explosive spread of the Internet, expands the academic boundary of the traffic economy from economic benefits to environmental impacts, and provides important inspirations for optimizing the layout of the traffic economy and promoting sustainable urban development.
Based on the data of A-share listed companies in Shanghai and Shenzhen from 2009 to 2023, this paper empirically analyzes the impact of data element sharing, characterized by the launch of data trading platforms and public data opening platforms, on the efficiency of enterprise resource allocation and its mechanism of action. The findings demonstrate that data element sharing significantly enhances resource allocation efficiency, a conclusion validated through robustness tests including instrumental variable methods, expectation effect exclusion, and dual machine learning approaches. Furthermore, reducing transaction costs and boosting digital technology innovation are key pathways through which data sharing improves resource efficiency. Notably, data sharing proves particularly effective for capital-intensive enterprises, coastal regions, and competitive industries. The study also reveals that data sharing not only elevates resource allocation efficiency within local enterprises but also positively influences neighboring regions, with economic development-related data sharing demonstrating stronger efficiency-enhancing effects. These conclusions provide both theoretical and empirical evidence to support China’s digital economy development and resource allocation optimization.
Can the new generation of AI reconstruct the global value chain of service industry? By introducing CES production function and Dixit Stiglitz model framework, based on the transnational input-output and service industry data from 2010 to 2024, this paper uses the text mining method to construct the development level index of the new generation of artificial intelligence, calculates the development level of the new generation of artificial intelligence combined with the dictionary of the development plan of the new generation of artificial intelligence, and discusses its mechanism of action on the status of the global value chain of the service industry combined with the production decomposition framework of the forward and backward associated global value chain (WWYZ2017). The research shows that the global value chain status index and resilience index will increase by an average of0.101 units and 0.960 units respectively for each unit increase in the level of the new generation of artificial intelligence. The new generation of artificial intelligence has significantly driven the rise of China’s service industry’s position in the global value chain and enhanced its resilience through three core channels: industrial intelligence agglomeration, intelligent innovation efficiency and information transparency. The expansion study found that the positive role of the new generation of AI in promoting the status and toughness of the global value chain was more prominent in non AI pilot areas and labor-intensive industries. Further analysis showed that there were clear multiple threshold constraints in capital, labor and knowledge intensive service industries. For the status of the global value chain, the new generation of AI needed to cross a high technology and capital threshold to fully release the momentum of service value upgrading; For the resilience of global value chain, it also shows certain threshold constraints. Only when the corresponding factor accumulation threshold is reached, the toughness enhancement effect is significantly displayed. This study provides multiple evidences for understanding the development path and value upgrading of service industry in the era of artificial intelligence.
Synthetic control method (SCM) has emerged, as a mainstream causal inference technique, gaining popularity in policy evaluation due to its robust performance with small-scale data, transparent counterfactuals, and interpretable results. Firstly, the origin of SCM is traced back to the treatment effect model, clarifying its core principles. Secondly, the development of SCM is systematized in five aspects, including relaxation of basic assumptions, improvement of estimation methods, model structure and asymptotic properties, hypothesis testing and interval estimation, and emerging themes. Thirdly, the connection and differences between SCM and other causal inference methods are clarified. Fourthly, based on representative applications of SCM both domestically and internationally, the applicability and validity of SCM in concrete practice are analyzed. Two main reasons for using this policy evaluation tool in research are summarized: One is to estimate the individual treatment effect, and the other is to allow for the existence of individual heterogeneity and unobserved time-varying confounders. Lastly, the future of SCM is envisioned from both theoretical and applied perspectives.
The increasing severity of climate change has led to more frequent adjustments in climate policies, and the uncertainty of these has brought new risks to the production and operation of enterprises. This paper takes China’s A-share listed industrial enterprises from 2007 to 2021 as samples to examine the impact of provincial-level climate policy uncertainty in China (CCPU) on enterprise green total factor productivity (GTFP). The study shows that CCPU has a significant negative impact on enterprise GTFP, and the effect is particularly pronounced among state-owned enterprises and among firms with smaller sizes, low quality of internal control, and low analyst attention. Meanwhile, CCPU can inhibit enterprise GTFP by increasing financing constraints, enhancing environmental regulations, and reducing green investment, while improving enterprise risk-taking capacity can mitigate the inhibitory effect of CCPU on enterprise GTFP. The results of this paper provide a useful reference for policymakers to construct a reasonable climate policy system to guide enterprises’ green transformation.
Accelerating the green transformation of polluting enterprises is crucial for China to fulfill its dual-carbon goals and achieve high-quality development. However, the current low enthusiasm for green technology innovation among these enterprises poses a significant challenge. Consequently, how to better promote green credit policies and stimulate their innovation vitality has attracted considerable attention from both academia and policymakers. Using the “green credit guidelines” as a quasi natural test and using firm level data, applies the generalized random forest (GRF) method to explore the policy effect of green credit on the green technology innovation of polluting enterprises. The findings reveal that the “green credit guidelines” policy has a significantly negative effect on green technology innovation. Further research has found that: 1) The effects of the “green credit guidelines”demonstrate significant heterogeneity among on pollution Enterprise, and the heterogeneity has systematically associated with firm endowments, the external market environment, and the institutional setting; 2) the effect of the three dimensional factors on policy effects shows complex multi dimensional links. Against the backdrop of China’s pursuit of high-quality development and new global opportunities, an in-depth investigation into the effects of the green credit policy on green technology innovation is beneficial for designing differentiated policy measures, and also offers support for future policy optimization on the theoretical and empirical levels.
