The extension from independence to dependence between jump frequency and jump size in Markov-modulated jump diffusion models

• An extension of the Markov-modulated jump diffusion model.
• An empirical investigation of the relationship between jump frequency and jump size.
• Using EM and SEM algorithms to estimate the model parameters.
• Revealing a phenomenon of dependence between jump frequency and jump size.

We set out in this study to investigate the relationship between jump frequency and jump size for the 30 component stocks of the Dow Jones Industrial Average (DJIA) index, extending the Markov-modulated jump diffusion model from independence to dependence between jump frequency and jump size. We propose an estimation method for the parameters of the Markov-modulated jump diffusion model based upon dependence between jump frequency and size, with our results indicating that when abnormal events occur, the Markov-modulated jump diffusion models with both state-independent jump sizes (MJMI) and state-dependent jump sizes (MJMD) outperform the pure jump diffusion (JD) model in terms of capturing the risks of jump frequency and jump size. Based upon Akaike Information Criterion (AIC) and Schwarz Bayesian Criterion (SBC), our results further indicate that for 23 of the component stocks, the MJMD model may be better suited, as compared to the MJMI model. Finally, our empirical observations reveal that the behavior of jump risks in the stock markets, including jump frequency and jump size, is not independent, since these phenomena are found to coincide during both financial crisis periods and stock market crashes, with the largest jump size risks, during certain periods, being accompanied by either systematic or idiosyncratic risks.