Predicting Extreme Daily Regime Shifts in Financial Time Series Exchange/Johannesburg Stock Exchange—All Share Index

During the past decades, seasonal autoregressive integrated moving average (SARIMA) had become one of a prevalent linear models in time series and forecasting. Empirical research advocated that forecasting with non-linear can be an encouraging alternative to traditional models. Linear are often compared mixed conclusions terms superiority performance. Therefore, aim this study is build early warning system (EWS) model for extreme daily losses financial stock markets. A logistic tree (LMT) used collaboration average-Markov-Switching exponential generalised conditional heteroscedasticity-generalised value distribution (SARIMA-MS-EGARCH-GEVD) estimates. five-day exchange/Johannesburg exchange-all share index (FTSE/JSE-ALSI) period 4 January 2010 31 July 2020. The set into two-stage framework. Firstly, SARIMA fitted returns order obtain independently identically distributed (i.i.d) residuals fit MS(k)-EGARCH(p,q)-GEVD i.i.d residuals; while, second stage, we set-up EWS model. results estimated MS(2)-EGARCH(1,1) -GEVD revealed highly volatile giving expected duration approximately 36 months days regime 58 2 two. We further found any degree above 25% implies there will no losses. Using seven statistical loss functions, SARIMA(2,1,0)×(2,1,0)240−MS(2)−EGARCH(1,1)−GEVD proved most appropriate predicting regimes as it was ranked at 71%. Finally, exhibit reasonably overall performance 98%, sensitivity 79.89% specificity 98.40% respectively. indicated success classification rate 89% prediction 95%. This promising technique EWS. findings also confirmed 63% 51% both training sample validation correctly classified. useful decision makers sector future use planning. Furthermore, base researchers conducting studies on emerging markets, have been contributed. These important risk managers investors.