Boosting Financial Econometrics via ML | Research In School

The research was presented by Aditya Punatar, a 10th grader, and Himansu Takur, who is pursuing a degree in Electronics and Telecommunication Engineering. Both students collaborated on this project during the summer of 2024.

Their work focuses on the use of the Upper Half Plane Model and hyperbolic geometry to improve the forecasting of stock prices. Traditional forecasting models like LSTM and GARCH often struggle with capturing market volatility and sharp transitions. The proposed hyperbolic model addresses these limitations by analyzing stock price trajectories through turning angles and hyperbolic distances, offering a more nuanced understanding of market dynamics.

The research was built upon established models like ARMA (AutoRegressive Moving Average) and LSTM (Long Short-Term Memory) networks, with the key innovation being the integration of the hyperbolic model. This new approach enhanced the models' ability to handle nonlinear relationships in stock data, which traditional models had difficulty capturing. Through this research, they demonstrated how combining machine learning with advanced geometric methods could significantly improve forecasting accuracy.

The students tested their models using historical stock price data from Yahoo Finance, and the hyperbolic model outperformed both ARMA and LSTM in predicting market behavior, with a notable decrease in the root mean square error for key stock prices like Apple and Google.