📈 Quantitative Trading Strategy using OLS, Ridge, and Lasso

Overview

We compare:

Ordinary Least Squares (OLS) Ridge Regression Lasso Regression

The goal is to evaluate predictive performance and assess whether linear models can generate profitable trading signals.

Data

Source: Yahoo Finance (via yfinance) Asset: AAPL Period: 2010-2025

Features Engineered

Lagged returns (1, 2, 3 days) Moving averages (5, 10 days) RSI (Relative Strength Index) MACD (Moving Average Convergence Divergence) Rolling volatility (10-day) Volume-based features

Target

Next-day log return

Models Used

Linear Regression (OLS) Ridge Regression (L2 regularization) Lasso Regression (L1 regularization)

Evaluation Metrics

RMSE R² score Directional Accuracy Sharpe Ratio Statistical significance

Strategy Go long if predicted return > 0 Go short if predicted return < 0

Results

Compared model performance across OLS, Ridge, and Lasso Lasso used for implicit feature selection Evaluated cumulative returns and Sharpe ratio

Key Insights

Regularization helps reduce noise in financial data Lasso identifies most relevant predictors Linear models have limited predictive power but can provide directional signals

Future Improvements Add transaction costs Use multiple assets (pairs trading / cross-sectional)

Tech Stack involved Python Pandas, NumPy Scikit-learn Matplotlib SciPy