Panel Markov-switching models of economic phenomena: Three applications

A popular approach to structural change of a model's specification is to allow each entity to take on the characteristics of one of two types. An entity's transition between the two types is governed by a probability of remaining in the current type or switching to the other type. This dissertation extends these Markov-switching models to panel data where heterogeneity can be cross-sectional and temporal. Three applications are developed.; The first essay studies four-digit industry returns-to-scale. Each industry can be characterized in a given year as “constant” or “increasing” returns-to-scale. A panel Markov-switching model endogeneously identifies a statistically significant increasing returns-to-scale behavior for virtually all of the industries studied in at least one year. The presence of increasing returns to scale has important implications for macroeconomic models.; The second essay extends the univariate panel Markov-switching mixture to a bivariate system. The behavior of 25 of the largest financial stocks is characterized by a latent flow of information driving both a stock's return volatility and volume. The Markov-switching panel vector autoregression (M-PVAR) identifies one of the regimes as a “high volume” and “high volatility” regime and the second regime as a “low volume” and “low volatility” regime. The M-PVAR provides a dramatic improvement in fit over OLS linear models and hence supports an information flow characterization of stock behavior.; The third essay explores two stylized facts of commodity returns—the co-movement of seemingly unrelated commodities and conditional heteroscedasticity of the residuals. The first characteristic is virtually non-existent in daily and monthly commodity data in the 1990's. The specification of the conditional variance is found to be better modeled as a Markov-switching mixture process rather than a GARCH specification.