Implicit price estimation: Measuring the impact of nonfinancial factors in environmental management decision-making

This dissertation develops and applies a new method termed implicit price estimation to measure the implied values associated with nonfinancial factors in investment decisions. The application uses a proprietary dataset of environmental improvement projects considered by a major U.S. corporation during 1993-1994. The revealed price estimates show that decision makers placed a significant value on achieving environmental improvements that address publicly stated corporate environmental objectives, beyond the value that was readily measured and incorporated into traditional net present value (NPV) estimations. Prices so obtained may prove useful in guiding future decision-making, in assessment of biases and inconsistency in existing decision processes, and in communication to regulators and the public of the decision weights that the firm places on nonfinancial environmental factors.;The implicit price estimation method is an application of a modified probit model to financial and nonfinancial decision data that yields coefficients directly interpretable as prices in an internal capital market. Decision models to be estimated must include one financial measure along with one or more nonfinancial measures that are to be priced, with all these measures expressed in absolute terms. The standard scaling of probit coefficients is replaced by a scaling that allows for direct interpretation of coefficients. I use the projects' estimated NPV (a single absolute measure of financial performance) along with absolute physical measures of environmental impacts, as the independent, explanatory variables in a model explaining project accept/reject decisions. Estimated coefficients are directly interpretable as prices reflecting the per-unit portion of expected project value that is not captured in the NPV measure. I find that an "Arctanit" maximum likelihood estimation model, based on the Cauchy distribution and not previously employed in accounting research, serves better than similar probit and logit models.;The empirical analysis features likelihood ratio-based testing of coefficient estimates. I show by simulations that likelihood ratio tests are better calibrated and more powerful than Wald-based tests in detecting statistically significant relationships.;Also, I introduce a deterministic alternative termed Multivariate Pareto Analysis that may have useful application in quality control and other decision-making arenas.