Samyak Jayanth

A 15-Year Empirical Comparison of Deep Learning, Tree-Based, and Regression Models for Stock Market Forecasting

Started Aug. 13, 2024

Abstract or project description

This study explores the effectiveness of various machine learning and statistical models - Linear Regression, Lasso Regression, Random Forest, XGBoost, and Long Short-Term Memory (LSTM) neural networks - for forecasting the next-day closing price of Tesla ($TSLA) stock. Using historical daily data from 2010 to 2025, the models incorporate a comprehensive set of engineered technical indicators, including moving averages, momentum oscillators, volatility measures, and price levels, to capture both short-term and long-term market dynamics. Model performance is evaluated against robust baselines such as the previous close, previous open, and midpoint of the prior session, with accuracy assessed using mean squared error (MSE) and directional accuracy metrics. The results reveal the challenges of achieving a significant predictive edge over simple heuristics in financial time series and highlight the relative strengths and weaknesses of advanced machine learning models versus traditional statistical approaches.