The Utilization of Machine Learning for The Prediction of CSX Index and Tax Revenues in Cambodia

ABSTRACT: The machine learning model LSTM was applied and compared with the linear regression machine learning model for the CSX index of the Cambodia Securities Exchange. The time span covered in this study was from February 1, 2019, to October 12, 2023, totaling 1146 days. Out of these, 917 days were classified as Train data, accounting for 80% of the total sample size, while 229 days were classified as Test data. Additionally, the same techniques were used to forecast the tax revenues of the Royal Government of Cambodia, which were
collected by the General Department of Taxation. Despite using monthly data, the tax revenues were analyzed over a longer period of 25 years, starting from January 1998 to August 2023. The Train data consisted of 247 months, equivalent to 80% of the total sample size, while the Test data accounted for 61 months, approximately 20% of the total sample size. The machine learning model, LSTM, demonstrated superior performance over the linear regression machine learning model in predicting the daily movement of the CSX Index of the Cambodia
Securities Exchange, as evidenced by the root mean square error. The Test data revealed that the estimated root mean square error of the linear regression machine learning was 64.78, while the LSTM machine learning produced a lower error of 33.08. However, when applied to tax revenues data, the linear regression machine learning outperformed the LSTM machine learning, with estimated root mean square errors of 219.31 and 1601.87, respectively.
KEYWORDS: CSX Index, Tax Revenues, Machine Learning, Linear Regression, LSTM