Impact of COVID-19 on Stock Indices Volatility: Long-Memory Persistence, Structural Breaks, or Both?

Abstract

The onset of the COVID-19 pandemic has increased volatility in financial markets, motivating researchers to investigate its impact. Some use the GARCH family of models to focus on long-memory persistence, while others use Markov chain models to better identify structural breaks and regimes. However, no study has addressed the occurrence of these two phenomena in a unified framework. Since both are important features of the data, to ignore one or the other could lead to poorly specified models. The outcome would be incorrect risk measurement, with implications for risk management, Value at risk, portfolio decisions, forecasting, and option pricing. This paper aims to fill this gap in the literature. We assemble an international dataset for 16 stock market indices in three continents over the period from August 1, 2019 to February 18, 2022, totalling 669 business days. Using R, we estimate 80 GARCH family models, 16 pure Markov-Switching models, and 900 combined GARCH/ Markov-Switching models using daily stock market log-returns. We allow for two volatility regimes (low and high). We also measure and incorporate News Impact Curves, which show how past shocks affect contemporaneous volatility. Our main finding, across estimated models, is that COVID-19 affected both long-memory persistence and volatility regimes in most markets. To describe the specific impact in each market, we report News Impact Curves. Lastly, the first wave of COVID-19 had a much greater impact on volatility than did subsequent waves linked to the emergence of new variants.

Portuguese Agrifood Sector Resilience: An Analysis Using Structural Breaks Applied to International Trade

“…After this division, the sample parameters are estimated, and the equality of the coefficients between the two samples is tested using the F statistic [30,41]. According to [43], this test has a weak capacity when the structure point is not known, and it is not robust when heteroscedasticity and autocorrelation are not guaranteed, existing according to the author's probability of rejecting the null hypothesis of stability due to autocorrelation issues in the errors or distinct variances. When you have time series, it is interesting to make a contrast between the null hypotheses of temporal homogeneity of the model versus the hypothesis of having produced a change.…”

Developing Exp-FIGARCH Hybrid Models for Time Series Modelling

“…There are some financial indices that are known to be nonlinear, volatile and long memory (Rahman and Jibrin (2018), Benrhmach et al (2020), Ojeda et al, (2021), de Oliveira et al, (2022), Jibrin et al, (2022), and Jiang et al, (2023)). The ExpAR-ARCH, ExpAR-GARCH and other hybrid models introduced in the time series literature lack the strength to handle these three identified time series characteristics at the same time.…”