Forecasting daily stock market trading volume using Machine Learning

Abstract: Today, brokers within the stock market brokerage industry are having difficulties with accurately forecasting the trading volume that is conducted by their customers. This is especially a problem during periods of exceptionally high or low trading volumes. Solving this problem would lead to both monetary savings in terms of server costs and operational planning issues. This thesis uses three Machine Learning models (Random Forest Regressor, Linear Regression, and Support Vector Regression) to predict daily trading volume. In Machine Learning, features are variables that act as explanatory variables for the dependent variable, in this case, the daily trading volume. The primary focus of this study is to evaluate and analyze which types of feature categories are the most important. Therefore, this study uses a variety of features divided into five different categories (Temporal, Historical, Market, External, and Customer). The results from the models trained using each individual feature category are compared against each other. Secondly, this study also focuses on analyzing the performance of all feature categories together. A Naive model of a 20-day rolling average is used as a benchmark to evaluate the results. The findings of this study indicate that Machine Learning models perform better than the proposed Naive approach when predicting daily stock market trading volume. However, the difference is of a small nature. Further, the Historical feature category is the category that performs best and can therefore be argued to be the most important category when predicting daily trading volume. However, the results of this study are not of statistical significance. The findings of this study can be relevant to the research field and can be used in future studies to further investigate the feature importance in stock market trading volume prediction.