This article discusses how to design deep-learning models for empirical cross-sectional asset pricing by combining them with financial priors, focusing on prediction-target design, market-environment modeling, and network-architecture choices.
As China’s quantitative investment industry and market environment have matured, traditional pricing factors derived from price and volume data have been mined extensively. Hundreds or even thousands of public research reports have already been published, and some hedge funds even claim to hold tens of thousands of factors. Yet these star-like factors are only different quantitative manifestations of a few underlying forces, such as the illiquidity premium and the idiosyncratic volatility anomaly. They cannot diversify risk when the market enters unusual regimes. In extreme episodes, such as February 2024, most price-volume factors fail or reverse together, causing even larger swings for investors. The performance of domestic quant funds this year has illustrated exactly that. From this angle, these alpha factors are really just special forms of beta.
To obtain genuinely independent excess returns, quantitative researchers have to do the core job of value discovery and develop new factors. There are three main routes:
Among these three sources of new factors, this article focuses on building alternative factors from new data plus new methods in order to capture mispricing caused by investors’ limited attention. Concretely, I split the discussion into similarity momentum and text analysis.
Quantitative trading relies on backtesting to iterate on investment strategies and improve out-of-sample portfolio performance. This article uses formulas together with code to introduce a selected set of return-risk metrics for portfolios, and also discusses how to evaluate return-prediction models and risk-prediction models.
Quantitative investment strategies often rely on mathematical programming to determine portfolio weights. One reason is to balance objectives such as return and risk more scientifically; another is to use optimization as the layer where subjective and objective constraints can be introduced in a unified way. This article focuses on short-horizon, long-only equity alpha strategies built on multi-factor theory, and reviews practical portfolio optimization methods. It also summarizes mainstream approaches for forecasting the key inputs required by optimization: expected return, risk, and transaction cost.
Traditional finance assumes that market participants are perfectly rational, which is obviously at odds with reality, so it cannot explain the many forms of mispricing that appear in practice. Behavioral finance treats investors as boundedly rational and combines traditional financial theory with psychology to study which irrational behaviors create mispricing and how those biases connect to asset returns. Factor investing treats human behavioral bias as a source of excess return, so it is worth reviewing the main results of behavioral finance as a whole and linking them to the anomalies seen in markets.
Most introductions to static-blog building focus only on choosing a blogging system and hosting the generated files, which leaves beginners confused and unsure how to put the pieces together. This article walks through the full technology stack and the basic setup used for a static blog from writing, to publishing, to maintenance. Used as an index, it gives a full-picture view of static-blog creation.
SHAP (SHapley Additive exPlanations) is a game-theoretic, model-agnostic approach to machine learning interpretability. It can quantify each feature’s contribution to a single prediction while also aggregating local explanations into a global view of the model. SHAP has strong theoretical guarantees and, thanks to substantial engineering optimization, is also practical in real-world workflows. This article introduces the core theory behind SHAP and shows several example visualizations.
The role of a risk model in an alpha strategy is to predict the covariance matrix of expected returns for the underlying assets, so that portfolio variance can be minimized for a given expected return and the strategy’s Sharpe ratio can be improved. A risk model is no less important than a return model, yet compared with return modeling there is surprisingly little discussion of how to build one. One reason is that most investors simply buy commercial solutions such as Barra; another is that risk models must be judged at the portfolio level and cannot be evaluated through a simple benchmark. This article gives a systematic introduction to the definition and implementation of risk models, including shrinkage estimators, expert factor models, and data-driven statistical models.
In this section, we revisit Quant 4.0 from a systems perspective and explore how all of its components can be integrated into a single architecture. The figure below shows a suggested Quant 4.0 system framework, including an offline research system for quantitative research and an online system for live quantitative trading.
The unexamined life is not worth living.
This article first explains why I chose to pursue an alpha strategy, then reviews the first phase of strategy R&D in the second half of 2023, outlines the tasks of phase two, and finally looks ahead to phase three.