Four essays on forecasting evaluation and econometric estimation.

摘要

This dissertation investigates methods for evaluating forecasting models and econometric estimation for the willingness to pay. The first chapter introduces a new concept for the forecasting evaluation, time distance. In modeling series with leading or lagging indicators, it is desirable to begin comparing models in terms of time distance. This chapter formalizes the concept of time distance in terms of various metrics, and investigates the behaviors of these metrics. It is shown that under some circumstances, time distance metrics indeed perform better in forecasting than standard measures (such as MSPE), and that some time-distance metrics outperform others.;Chapter 2 applies time distance to models of inflation. The commonly used metric of forecast performance, mean squared forecasting error, measures how vertically close the forecast is to the series to be predicted; time distance instead measures the leading and lagging properties, often leading to significantly different conclusions about relative performance. The effect of data snooping is examined with White's (1997) reality check bootstrapping technique. This paper specifically addresses the problem of using measures of core inflation. The core rate is identified using statistical dominance and risk-reduction dominance. Time distance selects commodity price as a leading indicator, in contrast with the results of using mean squared forecast error.;Chapter 3 demonstrates that models which may appear to have different properties may in fact produce residuals that differ only in subtle ways. By analyzing the relationships between model residuals the problems in distinguishing between models can perhaps be discovered, as illustrated by the econometric examples considered. Regressing residuals gives the long-memory residual, which is the difference between two models, but this difference is very subtle and deeply hidden, which explains why the traditional standard technique does not find this difference.;The last chapter studies the difficulty with estimating statistics related to the maximum willingness to pay (WTP) distribution that the information avaliable to the analyst in carrying out this task, under even the best of circumstances, inherently informationally deficient. In this chapter, we look at ways in which (a) economic theory can help to provide restrictions on the nature of the WTP distribution, (b) how fairly non-controversial assumptions about the characteristics of the WTP distribution can dramatically bound the range of key statistics of the WTP distribution, and (c) how both general prior information from a wide range of previous studies on a wide range of different goods and specific prior information related to the good can in conjunction with the observed data greatly reduce confidence intervals concerning the main statistics of interest of the WTP distribution. We augment the available data with these forms of outside information through a sequence of models starting with a non-parametric approach and concluding with a Bayesian hierarchical model.