Stock Price Prediction Using MIDAS Regression in the Colombo Stock Exchange, Sri Lanka

Abstract

The Colombo Stock Exchange (CSE) experiences frequent fluctuations driven by political and economic factors, which undermine investor confidence and complicate investment decisions. The objective of this research is to evaluate the forecasting accuracy of the Mixed Data Sampling (MIDAS) model in predicting the All Share Price Index (ASPI) within the Sri Lankan context. This study employs the Mixed Data Sampling (MIDAS) model to forecast stock prices using the All Share Price Index (ASPI) and the Standing Lending Facility Rate (SLFR) as key variables. Monthly ASPI and quarterly SLFR data from January 2018 to December 2024 were utilized. The MIDAS regression is estimated using polynomial lag structures to capture the delayed impact of interest rate changes on stock market performance. Unit root test, including the ADF & PP results, confirmed that all variables are integrated of order one (I(1)). The MIDAS model was then used to forecast the All Share Price Index (ASPI) for the period January to December 2025, yielding an average predicted value of 4.65%. The forecasting accuracy metrics showed a low MAE of 0.1390%, RMSE of 0.2459%, and MAPE of 2.3042%. According to Lewis’s (1982) criteria, a MAPE below 10% indicates high forecasting accuracy. This study highlights the effectiveness of the MIDAS model in integrating mixed-frequency data and its practical value for emerging markets like Sri Lanka. The findings offer valuable insights for investors, financial analysts, and policymakers aiming to enhance decision-making through advanced forecasting techniques. Therefore, they should integrate mixed-frequency forecasting models like MIDAS into monetary and financial decision-making to enhance predictive accuracy, support data-driven policy formulation, improve investor confidence, and promote sustainable economic growth.