Predicting stock market returns in turbulence through the adaptive neuro-fuzzy inference system

Purpose Accurate stock market forecasting impacts the decision-making of firms, investors and macroeconomic policies. This study analyzes the performance of the neuro-fuzzy technique for estimating the stock market in turbulent and normal states for firms listed on the Pakistan Stock Exchange (PSX). This study aims to evaluate the accuracy of the forecasting of neuro-fuzzy. It further considers the ex ante uncertainty and information asymmetry theories. Design/methodology/approach To determine one-day-ahead forecasting, we consider the dataset of 15 nonfinancial firms for two volatile market periods (2017 market fluctuations and COVID-19) and one stable market period (2018–2019). The methodology employs two adaptive neuro-fuzzy inference systems (ANFIS): the controller and the stock market process. Results suggest an accurate prediction of stock market trends using the proposed system in all three market periods. We evaluated the results based on hit rate percentage, root mean square error (RMSE) and graphical representation. Findings Results from the controller model were used to drive the stock market process model. In contrast, results from the process model indicate that stock market volatility, changes in stock prices and previous-day stock returns can effectively predict following-day stock returns using ANFIS. Practical implications The findings would be helpful for investors in yielding higher returns during stable and turbulent market periods. Furthermore, it would aid managers, policymakers and analysts in decision-making and target-setting. Originality/value This study uses the ANFIS to predict the stock market return in turbulent phases.