Pricing

Appendix A

Dichotomous Logistic Regression

Throughout this book we referred extensively to the logistic regression methodology as a means to either calibrate the logit price-response functions of chapter 1 and chapter 2 or estimate the logit bid-response probability functions introduced in chapter 4. While the two contexts exhibit many similarities in terms of their final outcome (e.g., both functions are inverse S-shaped and approach zero at some high prices), from a methodological standpoint, they require two distinct sets of statistical tools. First, the logit price-response functions are computed using nonlinear regression models that attempt to minimize the sum of squared errors between the observed demand and the demand expected to materialize ...