Overconfidence is the tendency of individuals to overweight their own private information relative to public information. In a sample of 2,994 entrepreneurs, 81% believed their chances of success were at least 70%. In reality, about 75% of new businesses fail within five years. In On the Evolution of Overconfidence and Entrepreneurs, Bernardo and Welch (2001) model informational cascades where overconfident entrepreneurs broadcast their private signals. In a group of 500 with signal precision p = 0.51, one entrepreneur's group benefit is roughly 114 times the individual cost.
What the Study Found
In a no-entrepreneur group with signal precision p = 0.51, the probability of an incorrect cascade approaches 48% even in large groups. The limiting probability of an incorrect cascade equals q²/(p²+q²) when the entrepreneur proportion is zero, and falls to qᵏ/(pᵏ+qᵏ) for any positive proportion. In a 500-person group with p = 0.51, the twentieth entrepreneur's marginal cost is 0.0035 against a 0.11 benefit to the other 499 members. In a contest with N = 100, p = 0.6, and k = 4, entrepreneurs rise from 7.5% to about 8.2% next generation. At p = 0.51, k = 12, and N = 100, entrepreneurs survive at only 1–2%, far below the group optimum of 0.425.
Methodology
The paper develops a theoretical model rather than an empirical dataset, extending the informational-cascade framework with overconfident agents. It compares normal, fully rational individuals against entrepreneurs who underweight public information, and compares groups containing entrepreneurs against groups without them. Because no closed-form solution exists for the cascade probabilities, the authors compute comparative statics numerically using a state-time recursion. A group-selection displacement model then traces how the proportion of entrepreneurs evolves across generations.
Key Statistics
| Metric | Finding | Context |
|---|---|---|
| Bad-cascade probability without entrepreneurs | Approaches 48% | Signal precision p = 0.51, large groups |
| Group benefit vs. individual cost | ~114× larger | N = 500, p = 0.51, first entrepreneur with k = 4 |
| Marginal cost vs. benefit of 20th entrepreneur | 0.0035 cost vs. 0.11 benefit | N = 500, p = 0.51, k = 4 |
| Surviving entrepreneur frequency | 1–2% vs. group optimum 0.425 | N = 100, p = 0.51, k = 12 |
| Social welfare function | E[V(λ)] = λ·E[V_OC(λ)] + (1−λ)·E[V_R(λ)] | λ = entrepreneur proportion; maximized to find λ* |
| Limiting incorrect-cascade probability | qᵏ/(pᵏ+qᵏ) for λ > 0 | q = 1−p, k = entrepreneur critical state |
Why This Matters
The model offers a discipline for when behavioral assumptions like overconfidence belong in economic theory. It frames the bias as a positive externality. Individuals who follow their own signals pay a private cost but reveal information that rescues their group from herd-driven errors. Behavioral finance models that treat overconfidence as a primitive gain a survival mechanism that explains why the trait persists. The framework predicts overconfidence should be most useful in large groups, with low-precision information, and at moderate rather than extreme levels.