Exploring the Predictive Power of Trading Volume on Stock Price Movements
Authors
Trisha, Sadia Maliha
Issue Date
2025.17.12
Degree
Master of Business Administration
Publisher
Dublin Business School
Rights holder
Rights
Open Access
Abstract
This study examines whether trading volume significantly contributes to the predictive accuracy of stock price forecasting models. Trading volume is often viewed as a proxy for liquidity and sentiment, yet its independent predictive value remains unclear. Using daily S&P 500 index data from 2010 to 2025, three experimental datasets were constructed: a Volume-Only dataset, a Price-Only dataset, and a combined dataset with Principal Component Analysis (PCA) applied. Four machine learning algorithms—Random Forest, Support Vector Regression (SVR), XGBoost, and LightGBM—were tested under a leakage-free chronological split.
Results show that trading volume alone had no predictive power, with models yielding negative R² values. Price-only models performed strongly, with SVR achieving R² = 0.8770, confirming the persistence of stock prices as the dominant predictor. Adding volume did not improve accuracy, and in some cases reduced performance. PCA confirmed that volume represented an independent but weak component.
The results indicate that trading volume adds very little supplementary information, but historical prices remain the primary factors of forecasting accuracy.