Portal de Periódicos da CAPES

Revolutionizing Active Investing With Machine Learning

2024; RELX Group (Netherlands); Linguagem: Inglês

10.2139/ssrn.4675439

1556-5068

Mukul Pal, Regina Fenesi, Oliviu-David Cigan, Alexandru-George Berciu, Radu-Ciprian Tiric, Florina Pal, Dan Todor, Eva H. Dulf,

Complex Systems and Time Series Analysis

This paper introduces a novel machine learning method aimed at enhancing the capabilities of active asset managers in navigating the complexities of selection systems. These complexities encompass idiosyncratic factors, subpar prediction methods, and competition from entrenched systems like Market Capitalization (MCAP) weighted benchmarks, which have historically challenged the industry. Active managers often fail to outperform the market, leading to a decline in the active asset management sector's growth, contrasted with the rise of passive management, especially post the introduction of Index funds in 1976. The authors apply the [3N] non-linear systems approach, starting by redefining MCAP benchmarks as closed systems. They introduce "Discrete Decile Steps," a method that categorizes stock information into dynamic states that rise or fall, indicating performance trends. This approach is augmented with a Random Forest Regressor to predict these states, forming a prediction system independent of idiosyncratic elements, aiding managers in making more effective selections. The paper also presents a thought experiment using elevators as a metaphor for stock movements, illustrating the importance of predicting the direction of these 'elevators' to enhance selection and reduce underperformance risk.