The Hot Stove Effect in Venture Capital: How Adaptive Sampling Generates a Negativity Bias

https://doi.org/10.65161/rec1Wp0YFUBB9kF2T

Venture capital (VC) decision-making operates under bounded rationality and sparse, noisy feedback. This paper examines the Hot Stove Effect, a negativity bias that arises when the intensity of future sampling depends on the outcome of early sampling. Using a stochastic simulation of a two-period investment problem, we analyse how an adaptive sampling policy, under which a favourable initial signal raises and an unfavourable one lowers the number of subsequent draws, distorts the beliefs of learners ranging from naive averages to Bayesian updaters. We derive a closed-form expression for the expected belief and show that it is strictly negative whenever the commitment following a positive signal exceeds that following a negative one. The bias is attenuated, but not eliminated, by Bayesian updating with a correctly centred prior: a rational learner remains negatively biased in the mean because the sample size is endogenous to the sign of the early signal. Independently of the updating rule, a majority of learners (about 58.5% in the baseline) end below the true value, because the underestimation share is governed by the sign of the accumulated evidence and not by the learning rule. At low volatility the resulting belief distribution is negatively skewed, with its mode near the true value and a heavy pessimistic tail; for the Bayesian learner the asymmetry reverses as volatility rises, while the underestimation share does not move at all. The findings indicate that the systematic undervaluation of neutral opportunities in VC follows structurally from payoff-contingent sampling and persists even under rational inference.