Criscent Birungi

About me

I am currently conducting research in stochastic control and reinforcement learning for lifecycle retirement and annuitization. I am also a PhD Mitacs Research Fellow at Desjardins Global Asset Management (DGAM). In this role, I contribute to designing a global stock recommendation system using an AI model that integrates financial data and market trends to classify securities based on their expected one-month return.

Outputs

Recommending the right securities at the right time: A global system based on optimized decision-tree algorithms for dynamic factor solutions

This paper studies a machine-learning stock-ranking framework for cross-sectional equity selection using point-in-time global data. We evaluate four ensemble regressors: LightGBM, XGBoost, Random Forest, and AdaBoost-R, trained to predict next-month returns from standardized fundamental and technical characteristics. Forecasts are translated into long-only, short-only, and long-short portfolios through a transparent quintile-sorting rule. Across the tested specifications, the models generate economically meaningful cross-sectional signals, with technical and liquidity-related variables contributing substantially to short-horizon predictive power. Differences across models reflect trade-offs between flexibility, stability, and sensitivity to fast-moving market information. Turnover analysis shows that the long-short strategy operates at trading levels consistent with short-horizon cross-sectional approaches, with annual one-way turnover stabilizing near 50% and declining after 2020. A fundamental-only specification confirms that analyst expectations and valuation variables remain informative even in the absence of technical factors. All results are reported gross of transaction costs and do not incorporate shorting or financing costs. Overall, the results indicate that, under the tested assumptions and design choices, machine‑learning models can extract useful cross‑sectional signals when applied within a transparent and implementation‑aware empirical framework.