Persistence in Variance, Structural Change, and the GARCH Model

Abstract

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.. American Statistical Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Business &Economic Statistics. This article examines the persistence of the variance, as measured by the generalized autoregressive conditional heteroskedasticity (GARCH) model, in stock-return data. In particular, we investigate the extent to which persistence in variance may be overstated because of the existence of, and failure to take account of, deterministic structural shifts in the model. Both an analysis of daily stock-return data and a Monte Carlo simulation experiment confirm the hypothesis that GARCH measures of persistence in variance are sensitive to this type of model misspecification.

Level changes in volatility models

“…This effect is reported in empirical and simulation studies like Diebold (1986), Lamoureux and Lastrapes (1990), Hendry and Neale (1991), Mikosch and Starica (2004), Rapach and Strauss (2008), Rapach et al (2008), among others. An analytical argument is found in Hillebrand (2005).…”

The Study of Causal Relationship between Stock Market Indices and Macroeconomic Variables in Cote d’Ivoire: Evidence from Error-Correction Models and Granger Causality Test

“…As a result, this hypothesis implies that the distributions of underlying parameters are nonstationary over time. (Clark, 1973), (Epps & Epps, 1976), (Tauchen & Pitts, 1983), (Harris, 1986), (Lamoureux & Lastrapes, 1990), and (Richardson & Smith, 1994) have presented evidence supporting the mixture of distribution hypothesis from their studies of the stock return volatility-volume relationship. These findings have suggested that stock price data be generated by a conditional stochastic process with a changing variance parameter which can be proxied by volume.…”

Non-Parametric Regression Methods

“…In the IGARCH class of models a shock to the conditional variance is persistent in the sense that it remains important for future forecasts of all horizons. Lamoureux and Lastrapes (1990) argued that large persistence may actually represent misspecification of the variance and result from structural changes in the unconditional variance of the process. They have shown that occasional discrete shift in the mean level of volatility causes substantial upward bias in estimates of the volatility persistence.…”