Multi-factor Empirical Asset Pricing Based On Firm Linkages

The stock market is an important part of the financial market.Since the reform and opening up,China has made world-renowned development achievements in social,economic,and cultural fields.Along with the rise of China’s national power,China’s stock market has moved from embryonic to mature.As of February 2022,the total market capitalization of the Shanghai and Shenzhen markets reached 86.29 trillion yuan,equivalent to 75.45% of China’s GDP in 2021,making it a cornerstone to support the development of the national economy.It can be seen that China’s stock market is an indispensable and important part of the national economy,and the safety and stability of China’s stock market play a key role in the sustainable development of China’s macroeconomic system.Therefore,whether from the perspective of investment,risk aversion,or risk management,insight into the risk premium of the stock market is conducive to protecting the interests of market investors,maintaining the intrinsic value of enterprises,and providing a solid guarantee for the prosperity of the financial market and the stability of social order.In fact,the study of stock market risk premium is a classic proposition in the field of finance.The analysis of the overall and local volatility of the stock market based on a micro perspective is essentially an asset pricing problem,the core of which is to analyze the key factors affecting the risk premium of the stock market and reveal the micro structure and inner operating mechanism of the market.However,traditional asset pricing models mostly use linear regression models to explain and forecast expected stock returns based on certain important pricing factors.In the face of today’s evolving and complex and changing stock market environment,the effectiveness of traditional asset pricing models for stock market analysis is facing great challenges,specifically:(a)The challenge of a large number of pricing factors: With the mining of factors in academia and industry,hundreds of pricing factors that can provide excess returns have been accumulated in the stock market.In the face of a large number of pricing factors,it is difficult for traditional asset pricing models to incorporate all of them,and there is no effective method to capture the impact of a large number of factors,which affects the ability of traditional asset pricing models to analyze the expected return of assets.(b)Challenges in analyzing the interaction effects of pricing factors: Traditional asset pricing models either treat pricing factors as independent features or use a combination of two factors or multiple factors to construct interaction term indicators when fitting the model.However,in the face of factor interaction analysis,it is difficult to effectively reveal the true effect of factor interaction term construction on asset pricing.In particular,considering the complexity of the real market,the construction method using the factor interaction term approach often brings too much noise information to the traditional asset pricing model and increases the complexity of the model analysis extremely.(c)The challenge of corporate correlations: traditional asset pricing models are mainly from the perspective of the asset itself,explaining the impact of its own risk factors on its asset price and expected return.However,the stock market is a dynamic market with diverse capital,complex composition,and constant changes.There are various correlations among listed companies,such as cooperation and competition,intrinsic value correlation,analysts’ perception and comparison of future development trends,and regulatory needs of supervisory authorities,which constitute an extremely complex dynamic network of linked listed companies.In such a connected network of listed companies,the status of a listed company will undoubtedly have different degrees of impact on the performance of its connected listed companies in the capital market.Therefore,in the field of asset pricing,more and more research results show that the momentum spillover effect based on the dependence of listed companies has become one of the important factors of asset volatility or risk premium.However,few asset pricing models consider the inclusion of linkage relationships due to the difficulty of extracting such essential,integrated and dynamic linkage relationships,which poses a significant obstacle to reasonably quantifying the impact of spillover effects.Therefore,this paper addresses the shortcomings of existing research and is dedicated to using a deep learning framework to intelligently model the real stock market using holistic and continuous rather than single data relationships.Specifically,in order to conduct data-driven empirical asset pricing research,this study first collects,cleans,and constructs a database of major multi-factor databases and company correlations of CSI 300 index constituents from 2009 to 2020 based on literature combing.Then,this study innovatively proposes a multi-factor representation learning algorithm to model asset pricing for the high dimensional and interactive data characteristics of multi-factors.On the basis of this,we further consider the impact of momentum spillover effects on expected returns based on firm dependencies and find the essential and comprehensive linkages among various firm linkages by adaptive multi-graph fusion algorithms.Collectively,the deep learning intelligent computing framework proposed in this study aims to explore the essential factors and intrinsic laws affecting market volatility and risk premiums from a holistic and continuous,rather than single,data relationship,to truly understand the microstructure and intrinsic operating mechanism of the stock market,and to provide a new research paradigm attempt for traditional finance research.Based on the above research work,this study realizes a deep combination of empirical asset pricing research and deep learning technology,finds a new intelligent computational solution for the classical proposition of asset pricing and provides a new perspective and technical means for empirical asset pricing.Therefore,the main innovations of this study are the following three aspects.The first contribution is that this study addresses the asset pricing modeling of high-dimensional,interaction multi-factor data characteristics,and proposes a matrix parametric modeling of interaction feature representation learning algorithm,which maps vectorized high-dimensional market factor characteristics,to a matrixed feature space,and uses the idea of parametric learning to dynamically model complex combinations of factor pairs.The method is able to capture the interaction relationship and strength between factors at different periods.Using this dynamic portrayal of the combined effects can distinguish the change in the importance of the factors’ influence and reasonably capture their impact on the stock market volatility.Meanwhile,the interaction feature representation learning algorithm of this study lays down and extends the application of deep learning in data-driven empirical asset pricing based on data,which helps to improve the prediction effectiveness and profitability of asset pricing models and provides an important idea and solution for high-dimensional financial data processing.The second contribution is that this study models the asset pricing of a complex stock market as a whole based on the perspective of the momentum spillover impact of corporate affiliations.Based on the deep learning framework,we innovatively propose a multi-relationship fusion-based graph neural network algorithm to achieve dynamic capture and essential portrayal of corporate correlations in the stock market.With the help of this deep learning framework,we are able to effectively explore the nature of stock market volatility,promote the research process of asset pricing based on the perspective of listed companies themselves,and achieve a breakthrough in research on capturing dynamic spillover effects for asset pricing based on firm correlations.The third contribution is that this study evaluates and demonstrates the effectiveness of deep learning in capturing risk spillovers and expected returns of equity assets as a whole.Based on the real volatility of the stock market,it is the first attempt to use deep learning to capture equity market risk spillovers and attempts to assess the rationality,effectiveness,and robustness of deep learning modeling under a unified evaluation framework with traditional asset pricing,and to measure the interpretability of its economic significance,the accuracy of its predictive power,and the profitability of its practical operation.This study responds to the needs of the times and provides a complete set of intelligent analytical solutions for empirical asset pricing,which can capture the complex and subtle effects of market factors on the stock market,making it possible to accurately quantify asset risk premiums.Moreover,by gaining insight into the evolution of local risks,the goal of preventing systemic risks in capital markets can ultimately be achieved.It provides theoretical references and decision aids for practical operations in the fintech industry,such as market timing,portfolio selection,and risk management.In general,this study proposes an asset pricing model based on multi-factor representation learning and firm momentum spillovers,which can reasonably capture the essential factors affecting the expected returns of equity assets for the real situation of stock market operation.With the advanced deep learning intelligent computational methods,the model portrays the high dimensional,and interactive multi-factor data characteristics of equity assets,as well as the momentum spillover effects based on firm-dependent relationships,which can provide insight into the complex and dynamically changing market operation,reveal the intrinsic mechanism of asset risk premium,and thus better explain the excess returns of China’s equity assets,and ensure the healthy and sustainable development of the stock market.It can provide decision support and advice reference for policymakers,listed companies,and all market investors.