Stock price prediction using ARIMA model: Evidence from Colombo Stock Exchange

Abstract

Purpose: The core objective of the research is to investigate the forecasting capability of the Autoregressive Integrated Moving Average (ARIMA) model for predicting short-term stock prices in the Colombo Stock Exchange (CSE) of Sri Lanka. Design/Methodology/Approach: The data gathered on a daily basis through the CSE Price Index, ranging between July 1, 2014, and June 30, 2024, was analyzed by using the Box-Jenkins approach. The selection of the optimum models was based on the minimum of Akaike Information Criterion and Schwarz Bayesian Criterion. The Autocorrelation Function, Augmented Dickey-Fuller Test, and error test measures, such as Mean Absolute Percentage Error, were considered for validation and for assessing the goodness of fit of the forecasting results. Findings: From the Autoregressive Integrated Moving Average (ARIMA) model analysis, the ARIMA (2,1,1) model was the best, with an MAPE of 3.9%, indicating strong forecasting performance. For the Autoregressive and the Moving Regression tests, both were highly significant at the 1% level, supporting the idea that past price variations contain useful information for predicting prices. Results suggest that partial weak-form inefficiency exists in the Sri Lankan Stock Market. Research limitations/ Future research directions: The study uses a univariate linear approach and does not account for exogenous variables, nonlinearity, or structural breaks. The findings of this approach would be more relevant to short-term linear predictability. The approach would not account for non-linear phenomena that could be prevalent in an emerging economy. Originality: The current research is among the first 10-year empirical validations of the ARIMA model's predictive accuracy in the Sri Lankan market, as the study’s results provide theoretical and practical insights into predictive modelling and market efficiency