This chapter introduces two innovative AI techniques for asset management: Corrective AI (CAI) and Conditional Parameter Optimization (CPO). Unlike fully autonomous AI systems, which face significant challenges in financial environments, these techniques assist or optimize human decision-making processes. CAI corrects human or algorithmic errors, while CPO dynamically adjusts resource allocations based on changing conditions. Both techniques have been successfully deployed in financial markets, showcasing AI's potential in practical, high-stakes applications.
CAI focuses on enhancing existing human or algorithmic decisions rather than making them from scratch. Inspired by meta-labeling techniques discussed by Dr. Marcos López de Prado, CAI has shown remarkable results. For example, it successfully guided a crisis alpha strategy at QTS Capital Management during the COVID-19 pandemic, accurately predicting market crises and providing a significant return boost. CAI’s effectiveness in finance, one of the most data-intensive and competitive fields, suggests its broad applicability across various industries, from oil exploration to semiconductor manufacturing.
CPO addresses the limitations of traditional optimization methods by incorporating machine learning. Traditional techniques often fail to adapt to external, time-varying conditions, leading to suboptimal outcomes. CPO uses ML to train models that optimize parameters based on historical data, making it suitable for scenarios like adjusting stop-loss levels in trading strategies or optimizing inventory management in retail. This approach allows for continuous and flexible parameter adjustments, enhancing performance in volatile or uncertain environments.
The chapter highlights several case studies demonstrating the power of CAI and CPO. In one example, an oil exploration firm used CAI to improve predictions about well productivity. Instead of replacing their trusted formula, CAI augmented it, leveraging additional variables and uncovering complex relationships. Similarly, a semiconductor manufacturer applied CAI to optimize its manufacturing processes, resulting in real-time adjustments and improved efficiency. These examples illustrate the versatility and effectiveness of AI-driven enhancements across industries.
Feature engineering is crucial for AI success. The chapter describes creating cross-sectional and time-series features for financial applications. Cross-sectional features, such as earnings yield and dividend yield, are normalized to remove survivorship biases. Predictnow.ai’s database includes over 6,000 active and delisted companies, ensuring comprehensive, point-in-time accuracy. Time-series features, like Fama-French factors, are used for market-wide analyses. Transforming cross-sectional factors into hedge portfolios further enhances their predictive utility, enabling better risk management and strategy optimization.
Several unique features developed by Predictnow.ai are introduced. The NOPE (Net Options Pricing Effect) feature captures market maker delta-hedging activity, providing insights into price movements. The Canary indicator signals market danger based on dual momentum strategies, while the Carry feature calculates roll yields across various asset classes, aiding in predictive modeling. The Order Flow feature, based on high-frequency tick data, measures buyer-seller imbalances, offering a robust signal for price prediction. These features showcase advanced techniques that go beyond conventional indicators.
CAI's impact on a daily seasonal Forex strategy is demonstrated using the EURUSD pair. The strategy exploits the “invoice effect,” where currencies depreciate during local working hours and appreciate during U.S. hours. Initially, the strategy achieved a Sharpe ratio of 0.88. However, after applying CAI, which adjusted bet sizes based on predicted probabilities of profit, the Sharpe ratio improved to 1.29, with reduced drawdowns. This case study exemplifies how CAI can significantly enhance trading strategies by refining existing models.
CPO's methodology is explored through an example involving a Bollinger Band-based mean reversion strategy on GLD (gold ETF). The strategy has three adjustable parameters: hedge ratio, entry threshold, and lookback window. Traditional optimization methods fixed these parameters based on historical data. In contrast, CPO uses ML to predict daily returns, dynamically adjusting parameters to maximize performance. The result was a significant improvement in Sharpe ratio and return stability, demonstrating CPO’s ability to adapt to market changes effectively.
The chapter contrasts conventional parameter optimization methods, such as using fixed or rolling historical data, with CPO's adaptive approach. Traditional methods often fail to respond to rapid regime changes, while CPO leverages a vast set of market features to optimize parameters daily. By incorporating technical indicators as input features, CPO makes informed, data-driven adjustments, ensuring the strategy remains robust and responsive to evolving market conditions.
The greatest potential of CPO lies in portfolio optimization. Unlike traditional methods, such as Equal Weights, Risk Parity, and Markowitz’s Mean-Variance optimization, CPO adapts capital allocations based on market regimes. A case study on a tech-heavy portfolio showed that CPO outperformed conventional methods, especially during market downturns, by intelligently allocating to cash or defensive positions. The flexibility to optimize for various objectives, such as minimizing expected shortfall, makes CPO a versatile and powerful tool for portfolio management.
The chapter presents comparative results, demonstrating CPO’s superiority over traditional methods in multiple scenarios, including ETFs, crypto portfolios, and FX trading strategies. In all cases, CPO delivered higher Sharpe ratios and better risk-adjusted returns, particularly in volatile markets. The adaptive nature of CPO enables portfolios to thrive across different economic conditions, providing a compelling argument for incorporating AI-driven optimization into asset management practices.
AI's role in asset management is not about replacing human judgment but enhancing it. Techniques like CAI and CPO represent a pragmatic approach to integrating AI, delivering measurable improvements in decision-making and resource allocation. By leveraging AI for corrective and adaptive purposes, asset managers can achieve better performance, manage risks more effectively, and gain a competitive edge in the ever-evolving financial landscape.
