The purpose of this project is to develop a model that predicts stock market crashes for an asset management firm in South Korea with over 8 billion USD under management. We utilize high-frequency signals known for their ability to forecast extreme negative returns in the stock market based on academic research. By leveraging 12 diverse predictors and conducting replication and out-of-sample tests on more than 30 recent studies, we have found that creating a composite signal through ensemble learning significantly enhances predictability compared to standalone signals. Collaborating with our client's in-house IT experts, we have successfully implemented the system, allowing internal portfolio managers to access real-time crash likelihood information.

The common issue with widely-used mean-variance portfolio optimization is its reliance on historical information. Theoretical implementation of Sharpe Ratio maximization necessitates the estimation of expected returns for individual stocks in the portfolio. However, estimating expected returns from historical returns can lead to over-allocation on stocks that experienced significant historical growth. This bias can overstate the expected return of the optimized portfolio. To tackle this problem, we adopt multi-factor models developed in existing studies to estimate expected returns. Through in-sample and out-of-sample tests of optimized portfolios based on factor returns, we consistently outperform various portfolios that rely solely on a wide range of historical returns. This approach mitigates the biases associated with historical return-based estimation, providing more accurate and robust portfolio allocations.

Thanks to advancements in data collection and processing, research on market predictability and cross-sectional return abnormality has grown significantly. We have designed a portfolio optimization algorithm that incorporates these alpha signals and estimates the second moment from forward-looking  option market information. With a convex optimization with net-zero systematic exposure constraints, we create a long-short market-neutral strategy to capitalize on abnormal returns. The platform's versatility allows easy integration of new alpha signals, making it ideal for advisory services in a discretionary fund. We collaborate with registered investment advisors and asset managers to achieve superior risk-adjusted returns, while effectively monetizing insights from abnormal returns.