Bayesian-adjusted estimates in project selection - a comparison study of Bayesian and non-Bayesian decision makers with empirical evidence from the pharmaceutical industry

Companies select projects to invest in based on uncertain estimates of their performance. Theories and empirical evidence suggest that if the uncertain estimates are taken at face value, the true performance of the selected projects tends to be lower than estimated, causing the decision makers (DMs) to experience post-decision disappointment. Taking prior information into account through Bayesian adjustment can result in more realistic estimates of the project performances and thus higher expected performance among the selected projects. However, Bayesian adjustment makes it less likely to predict extreme outcomes and, consequently, may lead to missing out on big wins. This thesis studies the differences in the investment strategies of Bayesian and non-Bayesian DMs and outcomes of these strategies. This is done by employing a combined approach of both qualitative and qualitative research methodology. The quantitative approach of this thesis is in the form of a mathematical model that is used both to derive analytic results and for Monte Carlo simulation. The qualitative approach of this thesis is utilized to test theoretical findings empirically. The key results reveal that when fewer than 50% of project proposals would truly perform well (e.g., have truly positive NPV), a Bayesian DM invests in too few and a non-Bayesian DM to too many projects. Moreover, the average ex post performance of the projects funded by a Bayesian DM is higher than that of a non-Bayesian DM. However, a non-Bayesian DM will have a higher proportion of funded projects that result in big wins, but also a higher proportion of projects that result in losses. If, on the other hand, more than 50% of project proposals would truly perform well, the roles of a Bayesian and non-Bayesian DM are reversed. The less accurate the performance estimates, the more pronounced the differences between the investment outcomes of a Bayesian and a non-Bayesian DM. These analytic results are testified empirically in the R&D portfolio selection decisions in the pharmaceutical industry. Accordingly, the decision-making environment of the pharmaceutical industry displays characteristics of an environment with high estimate errors that amplify the differences in outcomes of a Bayesian DM’s versus a non-Bayesian DM’s investment decisions. As the DMs in this industry show quintessential characteristics of non-Bayesian DMs and the observed empirical outcomes perfectly coincide with the theoretical outcomes for non-Bayesian investment decisions, our theoretical findings are well-reflected empirically. From a theoretical perspective, this thesis contributes novel analytic results on the differences between the investment strategies adopted by Bayesian and non-Bayesian DMs and validates these results with empirical evidence. From a practitioner’s point of view, this thesis gives insights into how estimation uncertainties affect investment decisions and outcomes. Understanding such effects can help managers make better-informed decisions