Weekly Quantitative Finance Paper Digest (Jan 24‑Jan 30, 2026)

MarketGAN: Multivariate financial time-series data augmentation using generative adversarial networks

This paper introduces MarketGAN, a factor‑based generative framework for generating high‑dimensional asset returns under severe data scarcity. It embeds an explicit asset‑pricing factor structure as an economic inductive bias and generates returns as a single joint vector, preserving cross‑sectional dependence, tail co‑movement, and temporal dynamics. MarketGAN employs generative adversarial learning with a temporal convolutional network (TCN) backbone that models stochastic, time‑varying factor loadings and volatilities, capturing long‑range temporal dependencies.

Using daily returns of a large US stock set, the authors find MarketGAN outperforms traditional factor‑model‑based bootstrapping methods in matching empirical stylized facts of asset returns, including heavy‑tailed marginal distributions, volatility clustering, leverage effect, and especially high‑dimensional cross‑sectional correlation structure and tail co‑movement among assets. In portfolio applications, covariance estimates derived from MarketGAN‑generated samples outperform those from other methods when factor information is at most weak, demonstrating tangible economic value.

Alpha Discovery via Grammar‑Guided Learning and Search

The paper addresses the core problem of automatically discovering formulaic Alpha factors in quantitative finance. Existing methods often ignore grammatical and semantic constraints, relying on exhaustive search in an unstructured, unbounded space. The authors propose AlphaCFG, a grammar‑based framework that defines and discovers Alpha factors that are syntactically valid, financially interpretable, and computationally efficient.

AlphaCFG uses an Alpha‑specific context‑free grammar to define a tree‑structured, size‑controllable search space, formulates the Alpha discovery problem as a tree‑structured language Markov decision process, and solves it with grammar‑aware Monte‑Carlo tree search guided by value and policy networks that are also grammar‑aware. Experiments on Chinese and US stock datasets show AlphaCFG surpasses state‑of‑the‑art baselines in both search efficiency and trading profitability. Beyond trading strategies, AlphaCFG serves as a general framework for symbolic factor discovery and refinement in quantitative finance, including asset pricing and portfolio construction.

Generating Alpha: A Hybrid AI‑Driven Trading System Integrating Technical Analysis, Machine Learning and Financial Sentiment for Regime‑Adaptive Equity Strategies

The paper proposes a hybrid AI‑driven trading strategy that combines (1) trend‑following and directional momentum capture via EMA and MACD, (2) mean‑reversion price normalization detection using RSI and Bollinger Bands, (3) market sentiment analysis through FinBERT, (4) machine‑learning signal generation with XGBoost, and (5) regime‑adaptive exposure adjustment based on volatility and return environment.

Over a 24‑month period the system achieves a final portfolio value of $235,492.83, yielding a 135.49% return over the initial investment. The hybrid model outperforms major benchmark indices such as the S&P 500 and Nasdaq 100 during the same period, demonstrating strong flexibility, lower downside risk, higher profits, and validating the use of multimodal AI in algorithmic trading.