Essays on risk and econometrics

This thesis contains the outcome of two projects: risk and property prices, and computational aspects of model averaging. In the first project, we investigate the effect of objective and subjective earthquake risk embedded in Japanese property prices. We employ a multivariate error components regression model to exploit a rich dataset containing transaction prices and various characteristics of residential properties. Long-run earthquake risk data is provided by the Japan Seismic Hazard Information Station and short-run earthquake probabilities are simulated from a temporal epidemic-type aftershock sequence (ETAS) model. While the long-run and short-run earthquake probabilities are seen as objective measures of earthquake risk, we elicit a subjective measure of risk from the data by means of a parametric family of probability weighting functions. The estimated shape of the probability weighting function provides insight on how people's perception of small and large earthquake probabilities is reflected in property prices.
In the second project we study the properties of the weighted-average least squares (WALS) estimator. The idea of model averaging emerges from the insight that model selection and estimation should not be seen as two separate steps, but rather as one integrated procedure. Model averaging estimators do not select one best-fitting candidate model but estimate a whole range of candidate models and assigns weights to each of the candidate estimates. We explore the computational properties of WALS and develop statistical packages that enable the computation of WALS estimates.