Modeling and Forecasting Stock Market Volatility of CPEC Founding Countries: Using Nonlinear Time Series and Machine Learning Models

Abstract

The highly sensitive, nonlinear, and unpredictable stock marketbehaviours are always challenging for researchers. Stock markets ofPakistan and China, i.e., KSE-100 and SSE-100, respectively, are thetwo most attractive stock markets after the official announcementof CPEC. Thus, the daily closing price of KSE-100 and SSE-100 Stockreturns are used to evaluate the volatility forecast performance ofthe machine learning technique, GARCH family and the nonlinearregime-switching models. The findings of this study revealed that thestandard GARCH model is the best-fitted model based on Akaike’sInformation Criteria (AIC) and Bayesian Information Criteria (BIC).Furthermore, the forecast performance of the machine learning LSTMmodel outperforms other models based on RMSE for SSE-100. Incontrast, the forecast performance of CGARCH for SSE-100 and theMarkov-regime-switchingmodelforKSE-100outperformsothermodelsbased on MAE, MAPE, and SMAPE evaluation criteria. It is also revealedthat the predictive power of the machine learning model is very closeto CGARCH and MRS model; therefore, the LSTM model can be used asan alternative to GARCH and regime-switching models for stock marketvolatility. These findings will help national and international investors,policy-makers, geographical economists, and industrialists to use thebestforecastmodeltomakebetterpoliciesandgaintremendousprofit.