Decomposing venture evaluation: Learning about belief formation

Abstract Research Summary Evaluating early‐stage ventures is difficult because observable signals of future quality are weak and noisy. How closely are evaluators' assessments related to venture quality, and do their subjective judgments add to the information already available about a venture? Using more than 1600 screening decisions from a venture incubator, we examine how judges use application information to evaluate ventures by rating them on a set of 14 criteria and providing an overall recommendation. Judges' beliefs as recorded in the grades of the criteria explain their recommendations well, but they are only weakly associated with venture quality. We document some systematic miscalibration in how criteria are weighted, though the miscalibrations largely even out when examining the correlation between recommendations and quality. Within the models we estimate, beliefs and recommendations predict venture quality no better than ventures' observable attributes. Experienced judges are more selective but identify high‐quality ventures only slightly better than novice ones. Managerial Summary Accelerators and incubators rely on expert judges to screen early‐stage ventures, but how much does this expertise add? Using more than 1600 screening decisions at a venture incubator, we assess how well judges identify high‐quality ventures and where their judgment falls short. We find that judges' recommendations are only modestly accurate. Judges form a recommendation from venture information by first rating a set of 14 criteria, but the beliefs formed are themselves only weakly tied to venture quality. Judges also misweight the evidence: several characteristics that matter most for quality carry little weight in their recommendation, while others sway decisions despite mattering little. Experience helps at the margin; seasoned judges are more selective and better at filtering out weak ventures, but they don't spot winners better than novices. A simple model using only the objective data that ventures report in their applications predicts quality as well as the judges do.