Forecasting the KSE-100 index during Novel Coronavirus (COVID-19) Using Time Series Models

Abstract

Forecasting stock market behavior is challenging due to its nonlinear, complex, and volatile nature, making it a key area of interest for financial experts and data scientists. Accurate predictions provide valuable guidance for investors in decision-making. The COVID-19 pandemic significantly impacted global economies, causing substantial declines in major stock markets. This study applies various time series forecasting models to predict the KSE-100 Index closing prices over short- and medium-term horizons (7, 15, and 30 days) and identifies the most accurate model based on out-of-sample performance. Using daily KSE-100 data from January 2 to December 31, 2020, the study also examines correlations with COVID-19 epidemiological variables. Empirical results indicate a negative correlation between daily COVID-19 confirmed cases and KSE-100 closing prices (-0.179, 95% CI: -0.333 to -0.053) and between daily COVID-19 deaths and KSE-100 closing prices (-0.188, 95% CI: -0.325 to -0.044). For the 30-day forecast, the drift method outperformed others in MASE, MAPE, MAE, and RMSE, while the simple exponential method was most accurate for 15-day forecasts, and the Holt method performed best for 7-day forecasts. These findings highlight the impact of the pandemic on market behavior and provide insights for investors and policymakers.

Keywords:

KSE-100, COVID-19, ARIMA, PSX, Error trend seasonal, Neural network autoregressive