StockGPT: Autoregressive Return Modeling for Stock Portfolios

Summary

StockGPT treats daily stock returns as sequences of numerical tokens and trains an autoregressive model to predict future returns. Its attention mechanism is intended to learn temporal patterns directly from return histories, avoiding the need to specify conventional price-based signals by hand. The paper evaluates predictions using long–short portfolios rebalanced daily or monthly.

The reported evaluation uses a held-out period and finds that portfolios based on the model span momentum and both short- and long-horizon reversal effects. The authors also report statistically significant alphas relative to leading stock-market factors. These results suggest that the model may capture return patterns beyond the selected benchmarks. The supplied description does not give portfolio construction details, risk-adjusted performance figures, transaction costs, or tests beyond the stated sample, so it does not establish how the approach would perform in other periods or after implementation costs.

Key ideas

• StockGPT models stock returns as token sequences in an autoregressive prediction task.
• Its attention mechanism learns patterns from return histories without manually designed price signals.
• Long–short portfolios formed from its predictions are tested at daily and monthly rebalancing frequencies.
• The paper reports exposure to momentum and reversal patterns and significant factor-adjusted alphas.

Tags

# StockGPT: A GenAI Model for Stock Prediction and Trading


# StockGPT: A GenAI Model for Stock Prediction and Trading









This paper introduces StockGPT, an autoregressive ``number'' model trained and tested on 70 million daily U.S.\ stock returns over nearly 100 years. Treating each return series as a sequence of tokens, StockGPT automatically learns the hidden patterns predictive of future returns via its attention mechanism. On a held-out test sample from 2001 to 2023, daily and monthly rebalanced long-short portfolios formed from StockGPT predictions yield strong performance. The StockGPT-based portfolios span momentum and long-/short-term reversals, eliminating the need for manually crafted price-based strategies, and yield highly significant alphas against leading stock market factors, suggesting a novel AI pricing effect. This highlights the immense promise of generative AI in surpassing human in making complex financial investment decisions.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.