Establishment and Optimization of Multi-factor Model

Pan Xu2020 · DOI

Abstract

Quantitative investment offers significant advantages over traditional methods through rapid data acquisition, rigorous processing, and objective transaction logic. While still in its infancy in China, the research on quantitative models tailored to the Chinese stock market presents substantial development potential. This paper reviews the evolution of asset pricing, tracing the journey from the single-factor Capital Asset Pricing Model (CAPM) to the Fama-French five-factor model, and explores the integration of machine learning to navigate the complexities of modern financial landscapes.

Introduction

The fundamental objective of asset pricing is to explain the variance in stock returns based on risk. For decades, the Capital Asset Pricing Model (CAPM) served as the industry standard, positing that a stock's return is proportional to its market risk. However, as markets evolved, "financial anomalies" emerged that the CAPM could not explain.

This inadequacy is particularly pronounced in emerging markets like the Chinese A-share market. Established in 1990, the A-share market operates under different mechanisms than mature Western markets, meaning Western models often require significant optimization or "factor correction" to be effective. The challenge for modern researchers is to bridge the gap between traditional linear models and the non-linear complexities of the current market.

The Evolution of Quantitative Investment

The development of quantitative investment can be categorized into three distinct historical stages:

1. The Foundation (1950s–1960s)

The field began with Harry Markowitz’s mean-variance theory (1952), which laid the groundwork for modern portfolio theory. This was followed by William Sharpe’s CAPM (1966), which introduced the critical distinction between systemic and non-systematic risk.

2. The Multi-Factor Era (1970s–2010s)

As the limitations of CAPM became apparent, researchers moved toward the Arbitrage Pricing Theory (APT) and eventually the seminal Fama-French three-factor (1993) and five-factor (2015) models. These models expanded the scope of risk by adding dimensions such as company size (SMB), value (HML), profitability (RMW), and investment patterns (CMA).

3. The Intelligence Era (Present)

Modern finance is currently witnessing the integration of non-linear algorithms. To manage the "factor explosion"—where researchers now track hundreds of potential indicators—the industry is transitioning from linear regressions to machine learning techniques. Methods such as Support Vector Machines (SVM), Random Forests, and Genetic Algorithms (GA-SVR) are increasingly used to identify complex patterns in high-dimensional data.

Challenges in the Chinese A-Share Market

A critical finding in recent literature is the "Asian Exception." While the Fama-French five-factor model is robust in the US, Eastern Europe, and Latin America, it often performs inconsistently in Asian markets. In China, empirical studies have shown that the simpler three-factor model occasionally outperforms the five-factor model because factors like "investment pattern" (CMA) and "profitability" (RMW) show inconsistent results across different study intervals.

Furthermore, the Chinese market is characterized by "factor overload." While researchers have identified over 300 potential factors, many are redundant. One cited study found that out of 94 common factors, only 12 provided truly independent information. This redundancy necessitates more sophisticated selection methods than traditional regression can provide.

Machine Learning and Model Optimization

To address these challenges, Chinese scholars have focused on domestic optimization. By tailoring factors to local market characteristics, researchers have successfully built expanded models (e.g., an 8-factor model) that achieve stable excess returns compared to the CSI 300 index.

Machine learning has proven particularly effective in this environment. Random Forest algorithms, in particular, frequently outperform SVMs and traditional linear models. Their ability to handle non-linear relationships and high-dimensional data without overfitting makes them ideal for stock selection. Quantitative strategies combining technical indicators (such as MACD and RSI) with Random Forest have consistently outperformed the SSE 500 Index in weekly stock selection tests.

Conclusion and Future Directions

The Chinese market is still maturing and does not yet fully adhere to the strict assumptions of traditional theoretical pricing models. Future research must focus on:

  • Factor Correction: Developing specific factors that account for the unique regulatory and behavioral environment of the A-share market, such as addressing the lack of a significant momentum effect.
  • Deep Learning: Moving beyond standard machine learning to utilize deep learning for mining complex, hidden information in large-scale financial datasets.
  • Real-time Adaptation: As new data collection methods emerge, the multi-factor model must evolve into a living system that constantly discovers and discards factors to adapt to rapid market shifts.

Summary updated

Abstract

Compared with the traditional investment method, quantitative investment has the advantages of fast data acquisition, rigorous processing method and free transaction logic. Quantitative investment in China is still in its infancy, so the research on quantitative model and quantitative investment strategy suitable for Chinese stock market has great development space and investment value. On the basis of capital asset pricing theory, quantitative trading develops from single factor to multi-factor model. Multi-factor stock selection strategy has become one of the most widely used strategies in the field of quantitative investment in China due to its advantages such as flexibility and strong ability to adapt to the market. With the development of big data and machine learning algorithms, scholars continue to combine artificial intelligence with multi-factor stock selection to build portfolios to obtain excess returns. Based on the capital asset pricing model and Fama-French five-factor model, this paper explores the future development direction of the multi-factor model. With the development of new theories, new methods, new data collection and new problems, new impact factors and new algorithms will be constantly discovered.
BibTeX
@misc{xu2020establishment,
  author = {Pan Xu},
  title = {{Establishment and Optimization of Multi-factor Model}},
  year = {2020},
  doi = {10.25236/icbdem.2020.014},
}
Figure