Smooth Transition Regression Models in Finance
This thesis covers the further development of smooth transition regression models and their applications in finance. Smooth transition regression models are used to model nonlinearity of regime-switching type in empirical applications. Usually, smooth transition regression models are used with univariate transition functions, that is, a single transition variable is supposed to govern the transition between regimes. With multivariate transition functions, the transition between regimes can be driven by more than one variable, each weighted by its estimated impact on the transition process. The first paper of the thesis transfers the use of multivariate transition functions to the panel smooth transition regression model and uses oil price forecast as an empirical application. Another commonly found restriction is that smooth transition regression models are applied with a single homogeneous transition function. If a model with more than one regime-switching variable is used, the transition between regimes can be either estimated with a single transition function for all regime-switching variables (homogeneous model) or with different transition functions (heterogeneous model) for each regime-switching variable. The second paper of the thesis proposes the use of panel smooth transition regression models with heterogeneous and multivariate transition functions. Credit default swap pricing is used as an empirical application. The previously introduced generalization of the smooth transition regression framework with heterogeneous transition functions increases the flexibility of the model. Using the heterogeneous model instead of the homogeneous one is the cautionary choice, because estimating the homogeneous model when a heterogeneous specification is appropriate generally leads to biased parameters. Therefore, the third paper of the thesis proposes a test for the decision between homogeneous and heterogeneous transition functions. Empirical size and power of the test are examined by a Monte-Carlo simulation.