Increase Alpha: Performance and Risk of an AI-Driven Trading Framework

TL;DR

This work addresses the challenge of extracting stable, low-overhead alpha from market inefficiencies using a domain-focused AI framework. It combines feed-forward and recurrent networks with expert-curated features to produce daily directional signals for 814 U.S. equities, avoiding large transformers to maintain speed and reliability. Through a comprehensive methodology—including multi-horizon signal generation, a cloud-based scenario analysis grid, rigorous accuracy testing with p-values and confidence intervals, and a six-quarter walk-forward portfolio—the study demonstrates superior risk-adjusted returns (Sharpe > 2) and low drawdowns relative to the S&P 500. The results persist across market regimes, including post-January 2025 turbulence, and the framework shows practical applicability with scalable cloud deployment, dynamic rebalancing, and explicit considerations of leverage and diversification, highlighting the potential of interpretable, computationally efficient AI in modern asset management.

Abstract

There are inefficiencies in financial markets, with unexploited patterns in price, volume, and cross-sectional relationships. While many approaches use large-scale transformers, we take a domain-focused path: feed-forward and recurrent networks with curated features to capture subtle regularities in noisy financial data. This smaller-footprint design is computationally lean and reliable under low signal-to-noise, crucial for daily production at scale. At Increase Alpha, we built a deep-learning framework that maps over 800 U.S. equities into daily directional signals with minimal computational overhead. The purpose of this paper is twofold. First, we outline the general overview of the predictive model without disclosing its core underlying concepts. Second, we evaluate its real-time performance through transparent, industry standard metrics. Forecast accuracy is benchmarked against both naive baselines and macro indicators. The performance outcomes are summarized via cumulative returns, annualized Sharpe ratio, and maximum drawdown. The best portfolio combination using our signals provides a low-risk, continuous stream of returns with a Sharpe ratio of more than 2.5, maximum drawdown of around 3%, and a near-zero correlation with the S&P 500 market benchmark. We also compare the model's performance through different market regimes, such as the recent volatile movements of the US equity market in the beginning of 2025. Our analysis showcases the robustness of the model and significantly stable performance during these volatile periods. Collectively, these findings show that market inefficiencies can be systematically harvested with modest computational overhead if the right variables are considered. This report will emphasize the potential of traditional deep learning frameworks for generating an AI-driven edge in the financial market.

Figures (12)

• Figure 1: Boxplot showing the distribution of Sharp Ratio, cumulative return/PnL (%), and Drawdown for different cases in the scenario analysis.
• Figure 2: Confidence interval plots for the period beginning from the first observation to the end of 2024. The stocks selected from the optimization process are ordered from lowest to highest accuracy on the horizontal axis. The solid green line shows the average accuracy for each ticker across the sample and the shaded areas shows the 95% confidence interval. The left plot shows long signals, while the right plot depicts short signals.
• Figure 3: Confidence interval plots for the period of the first two quarters of 2025. The stocks selected from the optimization process are ordered from lowest to highest accuracy on the horizontal axis. The solid green line shows the average accuracy for each ticker across the sample and the shaded areas shows the 95% confidence interval. The left plot shows long signals, while the right plot depicts short signals.
• Figure 4: Visualization of the strategy’s returns and risk, compared with the macro benchmark (S&P 500) and the stock's buy-and-hold.
• Figure 5: Scatter plot showing the long accuracy of the trading signals, before and after January 2025.