Window effect with Markov-switching GARCH model in cryptocurrency market

The non-stationarity of cryptocurrency is mainly attributed to structural breaks. Many studies use the rolling windows to deal with structural breaks. However the selection of windows is an open question without a systematic answer. This study investigates the window effect on in-sample coefficient estima-tion and out-of-sample forecasting. The results provide evidence on the stability of coefficient estimation under various window selections. However, in forecast, some specific window size shows much better accuracy of left-tail predictions in stable patterns. It provides a possibility to get better out-of-sample forecast by choosing a window from the historical data. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This study investigates the impact of window selection on in-sample coefficient estimation and out-of-sample forecasting, finding that different window selections result in varying levels of stability in coefficient estimation and some specific window sizes show better accuracy in left-tail predictions. It suggests the possibility of achieving better out-of-sample forecasts by choosing a window from historical data.