Bayesian dynamic panel models for tourism research

Abstract

This paper introduces several innovative dynamic panel data models that allow variations in slope coefficients both across time and cross-sectional units. We replace time variation with a dynamic (autoregressive) component, and introduce several variations of the so-called Mundlak device in which random intercepts are linear function of the average values of the regressors. We develop all our models in a Bayesian framework, and test their performance using an interesting application on the impact of advertising on firm sales. We provide technical details of all these models and present tools to compare their performance in a Bayesian framework. Moreover, model averaging and posterior model pools are presented to gain more insight into the relationship between advertising and sales.

Diagnostic testing in Bayesian analysis

“…Step 3: Compute conditional posteriors of model parameters distribution, the Bayesian theorem transforms the prior data beliefs into posterior (or after data) beliefs. Several recent papers have discussed the specific advantages of the Bayesian approach, including its flexibility in handling more complicated models and its non-reliance on asymptotic approximation , 2019c. As mentioned, the Bayesian approach performs better in small samples and provides inferences based on the data at hand, and not on some imaginary samples that we did not really observe.…”

“…Within the context of panel regression, the Bayesian approach does not depend on instrumental variables (as is the case with generalized method of moments [GMM]) and can address the incidental parameter problem. It also performs better when T is small and is more flexible in some advanced modeling contexts such as random coefficient dynamic panel models (Assaf and Tsionas, 2019c).…”

“…There are also strong implications in terms of the inferences that can be reported from the fitted model. Accordingly, the hospitality and tourism literature has recently placed a strong emphasis on the importance of a model's specification (Assaf and Tsionas, 2019c). However, none of the existing studies have discussed the issue of diagnostic checking within a Bayesian framework.…”

More general panel data models for hospitality and tourism research

“…All data were collected from the COMPUSTAT database. The same application was used in Assaf and Tsionas (2019a).…”

“…While Mundlak-based models with heterogeneity have been proposed by Assaf and Tsionas (2019a), the authors did not fully consider autocorrelation and heteroskedasticity. In the proposed models we have conditional (on x it ) heteroskedasticity as the error depends on x it .…”