ML for alpha is easy to overfit and hard to productionize. This list links to primary sources that stress validation, leakage, and deployment constraints. Read the foundations first, then large-scale empirical work, then nonlinear models, formula alpha, and how to evaluate what you mine.
Foundations and methodology
Start here before tuning hyperparameters. Finance-specific cross-validation and multiple-testing bias are where most ML alpha projects die.
Advances in Financial Machine Learning
Labeling, sample weights, purged k-fold CV, and backtest overfitting. The standard practitioner reference.
Multiple Testing and the Cross-Section of Expected Returns
Multiple testing in factor research. Adjusted t-stats and why published significance thresholds are too low.
The 7 Reasons Most Machine Learning Funds Fail
Short, blunt checklist on data leakage, false discovery, and production gaps in ML-driven funds. SlideShare mirror
Large-scale return prediction
Canonical empirical work comparing ML methods on equity cross-sections. Useful for understanding what actually works at scale versus in toy backtests.
Empirical Asset Pricing via Machine Learning
Neural nets, trees, and elastic net on 30,000+ stocks. Strong baseline for ML in cross-sectional equity prediction. Journal page
Deep Neural Networks, Gradient-Boosted Trees, Random Forests: Statistical Arbitrage on the S&P 500
Early systematic horse race of ML models on US equities. Useful for understanding model turnover and decay. Journal page
Taming the Factor Zoo: A Test of New Factors
Model selection framework for new signals. Relevant when your ML features overlap with known factors. Journal page
Deep learning and nonlinear signals
Nonlinear factor models, compression, and learning portfolio weights from latent structure — not just point forecasts on raw features.
Deep Learning in Asset Pricing
Autoencoder + factor model hybrid. Shows where deep learning adds structure rather than black-box fitting. Journal page
An Information Bottleneck Asset Pricing Model
Information-bottleneck constraints compress noisy inputs while preserving pricing-relevant signal in deep factor models. Addresses overfitting in low signal-to-noise equity data.
Scaling Conditional Autoencoders via Uncertainty-Aware Factor Selection
Scales CAE latent dimensions to K=50, then prunes factors by forecast uncertainty using Chronos, quantile XGBoost, and bootstrap ensembles.
Deep Reinforcement Learning in Factor Investment
CAFPO: conditional autoencoder factors feed a DRL agent for low-frequency long-short weights. Handles changing investable universes via factor compression instead of raw stock panels.
Formula alpha and LLM evaluation
Industry-style formula libraries, grammar-guided search, and scoring beyond a single backtest Sharpe — including when LLMs should act as researchers, not stochastic trading agents.
101 Formulaic Alphas
Industry-side alpha construction at scale. Good reality check on signal correlation and turnover.
Alpha Discovery via Grammar-Guided Learning and Search
AlphaCFG uses a context-free grammar and grammar-aware MCTS to search for syntactically valid, interpretable formulaic alphas. Tested on CSI 300 and S&P 500.
AlphaEval: A Comprehensive and Efficient Evaluation Framework for Formula Alpha Mining
Parallel, backtest-free evaluation across predictive power, temporal stability, perturbation robustness, financial logic, and diversity. Useful sanity check before you commit capital to a mined alpha library. ACM proceedings
AlphaForgeBench: Benchmarking End-to-End Trading Strategy Design with Large Language Models
Shows LLMs deployed as trading agents can be wildly non-reproducible even with deterministic decoding. Benchmarks LLMs as quant researchers that output executable alpha factors, decoupling strategy reasoning from stochastic order execution. ACM proceedings
FactorBench: A Portfolio-Aware Benchmark for Automated Factor Mining
~5,000 factors from nine mining methods (GP, RL, diffusion, LLM agents) on five index universes under one execution contract. In-sample IC often fails after style neutralization; LLM pools overlap heavily with Alpha101; dollar-neutral net Sharpe beats long-only headlines. No paradigm wins on all dimensions. Code
Before you ship
Walk-forward validation, purged k-fold, and explicit transaction cost models are non-optional. If Sharpe only exists in one in-sample window, fix the pipeline before adding features.
Also follow ongoing preprints at arXiv q-fin and journal pages via DOI.