Evolved small recurrent networks beat transformers in stock prediction and trading
Forecast Accuracy Is Not Trading Profit: Evolving Small Recurrent Networks for Stock Return Prediction
Neural and Evolutionary ComputingComputational Engineering, Finance, and ScienceMachine Learning
Summary
Forecasting models usually try to predict stock returns as accurately as possible. The authors show that smaller, evolved recurrent networks not only predict better but also lead to more profitable trading decisions compared to big transformer models. These small networks run efficiently on simple hardware and make better use of longer-term patterns in stock prices. This challenges the idea that bigger, more complex models always perform better in finance.
What this means in practice
- •For quantitative trading teams: Use evolved small recurrent networks to improve trading decision profitability and reduce hardware needs compared to transformers.
- •For embedded system developers: Deploy ultra-efficient small neural networks for real-time time series predictions on low-power devices like Raspberry Pi.
Authors
Jonathan Chang, Zimeng Lyu
Abstract
Time series forecasting models are typically compared on pointwise error, which scores a prediction in isolation from the decision it is produced for, and a lower forecast error does not imply a better decision downstream. A parallel debate asks whether modern transformer architectures forecast better than recurrent and other lightweight models. We compare linear, fixed recurrent, transformer, and mixing based architectures against recurrent networks evolved by neuroevolutionary architecture search, evaluating each on forecast accuracy and on the net return of a daily long/short strategy. All models are fit on a pooled panel, one network trained across the whole universe. Across four mid-cap portfolios and three trading years, the evolved networks rank first on both forecast accuracy and net trading performance, while the second most accurate model loses money once positions are formed and costs are charged. The advantage tracks a horizon match, since rank IC for the evolved networks rises from a one-day to a ten-day scoring horizon while every model above 300 parameters declines. They are also the cheapest end to end: a CPU-only search of 16 minutes yields 66-weight networks that predict in 10.8~$μ$s on a Raspberry Pi Zero, against transformer baselines of up to 817,153 parameters that require GPU training.