TLDR: AlphaX is an AI-driven value investing strategy for the Brazilian stock market that uses fundamental and market data. Backtesting shows it significantly outperforms major benchmarks (Ibovespa, Selic) and traditional technical strategies (RSI, MFI, Stochastic) in terms of returns, risk, and risk-adjusted returns, while also addressing common biases in AI models. The research highlights future potential in integrating qualitative analysis using Large Language Models.
Autonomous trading strategies have long been a focal point in artificial intelligence (AI) research, with various techniques like neural networks and reinforcement learning being explored to create agents capable of trading financial assets. While many of these strategies demonstrate strong performance in simulations using historical data (known as backtesting), their effectiveness often diminishes when deployed in real markets, particularly concerning risk-adjusted returns. A new study introduces AlphaX, an AI-based strategy inspired by the classical investment paradigm of Value Investing, designed to mitigate these issues by controlling for biases and reducing the risk of overfitting.
Understanding AlphaX: An AI-Driven Approach to Value Investing
AlphaX is an autonomous equity investment strategy specifically developed for the Brazilian Stock Market. Its primary objective is to achieve returns that surpass Brazil’s main benchmarks—Selic (Brazil’s benchmark interest rate) and Ibovespa (the main stock market index)—while maintaining a controlled risk profile. The strategy automates core concepts and practices of Value Investing, integrating both fundamental and market data through the application of advanced AI techniques. Its development prioritizes best practices to minimize model overfitting and enhance generalization, ensuring more reliable performance in live trading conditions.
The Essence of Value Investing
Value investing is a time-honored investment approach that relies on in-depth fundamental analysis. It involves seeking out stocks priced below their true intrinsic worth, based on a company’s financial health and long-term prospects. Unlike short-term trading strategies, value investing is typically associated with long-term positions, often held for several years. The core idea is simple: assess a company’s intrinsic value by analyzing its financial statements, earnings potential, asset base, debt levels, competitive position, and overall business model. Investors then purchase shares of companies trading significantly below this estimated value, aiming for a comfortable “margin of safety.” This approach often requires investors to go against prevailing market sentiment, buying when others are selling, based on the conviction that the market will eventually recognize the company’s true worth.
How AlphaX Operates
AlphaX gathers data from two key Brazilian financial sources: B3 (the Brazilian Stock Exchange) for market data (daily prices) and CVM (Comissão de Valores Mobiliários, Brazil’s equivalent of the SEC) for financial statements (revenue, expenses, assets, liabilities, etc.). The strategy targets companies with a minimum of five years of historical data and high liquidity, excluding highly indebted sectors like retail and financial institutions due to their unique risk characteristics.
The AlphaX algorithm computes four key indicators: Profitability, Solvency, Valuation, and Growth, normalizing them on a scale from 1 (worst) to 5 (best). These indicators, along with common financial multiples such as Price-to-Earnings and Price-to-Book, are used to perform a price regression. AlphaX employs an ensemble of regressor algorithms, including Random Forest and Regression to the Mean, for this purpose.
For potential investments, AlphaX selects companies whose indicators are above the median (with a lower threshold for Growth). These selected assets are then ranked by their expected return, calculated as the percentage difference between the projected price and the current market price. Capital is uniformly allocated among the top ‘X’ ranked assets, where ‘X’ is a configurable parameter. If fewer than ‘X’ assets are selected, capital is allocated only among those chosen. In scenarios where no suitable assets are identified, the capital is allocated to a Selic bond, considered a risk-free asset in the Brazilian market.
Risk mitigation within AlphaX is further enhanced by a “Triple Barrier” framework. This method defines three conditions for closing an investment position: a “Take Profit” barrier (if the asset reaches its projected price), a “Stop Loss” barrier (if the asset declines by a predefined percentage, e.g., 10%), and a “Vertical Barrier” or maximum investment horizon (positions are closed at the end of the operational quarter or upon the release of new financial statements, allowing for re-evaluation).
Impressive Performance in Simulations
The AlphaX strategy underwent rigorous testing in simulated scenarios using real B3 quote data and CVM balance sheet data, covering 18 quarters from February 2021 to May 2025. The results demonstrate AlphaX’s superior performance compared to both the Ibovespa and Selic benchmarks in terms of Total Return and Compound Annual Growth Rate (CAGR). For instance, AlphaX achieved a Total Return of 97.9%, significantly outperforming Ibovespa’s 16.6% and Selic’s 55.1%. The strategy also surpassed a normalized version of Ibovespa (NIbov), which controls for survivorship bias, with AlphaX showing a 97.9% Total Return against NIbov’s 72.6%.
Furthermore, AlphaX was benchmarked against three widely recognized technical strategies: Relative Strength Index (RSI), Stochastic Oscillator, and Money Flow Index (MFI). AlphaX consistently demonstrated superior performance across all key metrics, including Total Return, CAGR, Max Drawdown (indicating lower risk), and risk-adjusted returns (Annualized Sharpe Ratio and Annualized Sortino). AlphaX’s Annualized Sharpe Ratio was 0.98, notably higher than the best technical strategy’s 0.43.
To provide a more robust evaluation, the study also utilized the Probabilistic Sharpe Ratio (PSR), which adjusts for non-normal return distributions and limited sample sizes. AlphaX showed a 97.4% confidence level of having a Sharpe Ratio above zero, a performance unmatched by any of the technical strategies analyzed.
Also Read:
- AlphaEval: A New Standard for Evaluating Financial Predictive Signals
- STONK: A New AI Framework for Smarter Stock Movement Prediction
The Future: Integrating Qualitative Analysis with AI
While AlphaX successfully automates many quantitative aspects of value investing, the paper acknowledges that traditional value investing also involves crucial qualitative analysis. This includes interpreting textual information from financial reports, assessing a company’s competitive landscape, understanding regulatory changes, and analyzing market psychology and behavioral biases. The authors highlight that current AI models, including AlphaX, do not yet fully encompass these subjective, non-numerical aspects.
However, the research points to the transformative potential of Large Language Models (LLMs) and Transformer architectures in addressing this gap. These advanced AI models show promise in extracting meaningful insights from unstructured textual data, identifying shifts in regulatory environments, and even interpreting market sentiment. Integrating such capabilities into autonomous agents represents a significant future challenge and a promising direction for developing a more comprehensive AI-based Value Investing framework.
In conclusion, AlphaX presents a compelling AI-driven strategy for value investing in the Brazilian stock market, demonstrating robust performance in simulations. The researchers plan to make the code for their backtesting and strategy implementation open-source in the future. For more detailed information, you can refer to the full research paper: AlphaX: An AI-Based Value Investing Strategy for the Brazilian Stock Market.
