The curse of overconfidence 

Why do managers and political leaders persist with their pet projects when faced with negative evidence? Margaret Thatcher’s riposte, “the lady is not for turning” made clear that record unemployment of over two million was not going to make her change course. Mao intensified the Great Leap Forward, despite reports of widespread starvation in rural China, with over 30 million people dying in the consequent famine. When Texas Instruments abandoned its foray into the home computer market, its share price rose 22 per cent.

To the ancient Greeks, the answer was hubris. In Aeschylus’ The Persians and Sophocles’ Antigone, the protagonist kings are consumed with ambition and arrogance. They ignore advice and omens, and persist in their chosen paths, precipitating their demise. Modern-day historians and political scientists invoke hubris to explain Napoleon’s doomed march on Moscow, Hitler’s over-ambitious expansion, Thatcher’s failure to reverse the poll tax, and Mao’s persistence with the Great Leap Forward. Economists invoke hubris to explain both the tendency of CEOs to persist with bad decisions, and the empirical finding that takeovers often reduce the acquiring firm’s value.

In a recent paper, we begin by considering a more rational explanation for a leader’s stubbornness in persisting with inefficient policies, based on the leader’s concern for her reputation. We then show that a leader’s overconfidence magnifies inefficiencies. Strikingly, even the mere perception that she might be overconfident has similar effects. We conclude that a culture where leaders are expected to be overconfident can have adverse consequences, even when the leader in question is not, in fact, overconfident.

Reputational concerns in the absence of overconfidence

Consider a leader who has undertaken a new project or a policy reform. She partly internalises the social value of the new policy but is also concerned about the public perception of her ability to choose good projects – say, because she seeks re-election. We show that, for the sake of her reputation, such a leader might be inclined to persist with the reform even after being faced with negative evidence. Suppose that the evidence is discouraging but inconclusive – it suggests that the new policy is more likely to fail than to succeed, but that failure is not inevitable. Moreover, the evidence is not observed by the public who can, at best, infer it from the leader’s actions.

A key asymmetry is that pressing on with the reform will eventually reveal its true quality, both to the leader and to the public, resolving all uncertainty. In contrast, repealing the new policy prevents any further learning about its quality. Continuing a policy to its conclusion therefore amounts to the leader publicly verifying her private belief about its quality: if she assigns probability p to the reform being worthwhile, she expects a public vindication with probability p and a public humiliation with probability 1-p. Cancelling the reform only reveals that the leader deemed the policy bad enough to warrant termination, although the public cannot ascertain whether the leader’s private information was moderately discouraging or outright damning. Thus, cancelling a policy has a reputational cost when the news is moderately bad, biasing the leader to persist with the reform.

Furthermore, there are strategic complementarities between the leader’s continuation decision and the public’s inference. If the leader is more stubborn and willing to continue the reform at worse beliefs, the public’s inference after a repeal is more adverse, increasing the reputational penalty for stopping. Consequently, there can be multiple equilibria, and an equilibrium where the belief threshold for stopping is lower (i.e. where a repeal necessitates more adverse information) is more inefficient.

Overconfidence

We examine the implications of the leader’s overconfidence in this context. Suppose that the leader and the public have different prior beliefs regarding the merit of the policy reform, with the leader having a more optimistic prior. Since continuing the project results in more information revelation, this choice becomes even more appealing for an overconfident leader. Moreover, a second, more subtle effect separately induces excessive project continuation. Since the public knows that the leader is overconfident, its inference after a repeal is even more adverse: imagine how bad the policy reports must have been to make the overconfident leader change her mind! This further raises the reputational cost of a repeal.

Indeed, this reputational cost obtains even when the leader is not overconfident but is merely perceived to be so. The negative inference drawn after a repeal is then identical to that drawn when she is, in fact, overconfident. Thus, perceived overconfidence, in conjunction with the leader’s private information, induces a greater tendency towards stubbornness, exacerbating inefficiencies.

One may take this argument further. Suppose the leader is not overconfident and is also not perceived to be overconfident. However, the leader believes that she is perceived to be overconfident. More generally, there may be mutual knowledge that the leader is not overconfident up to many levels, but not common knowledge: at some level, there is a perception that the leader is (believed to be) overconfident.

We show that equilibria in a game where it is mutually known, up to many levels, that the leader is not overconfident are anchored to equilibria in the game where the leader is in fact overconfident. Indeed, it may be “almost common knowledge” that the leader is not overconfident, but, nevertheless, the equilibrium in the game with actual overconfidence plays a powerful role in increasing inefficiency. If the (presumed) overconfidence is large enough, then there is a unique limit equilibrium – the most inefficient equilibrium of the game without overconfidence.

The culture of overconfidence

These results show that a culture where business and political leaders are expected to be substantially overconfident may have pernicious effects, even for leaders who are not. Indeed, it can be mutual knowledge to a high degree that the leader in question has the right beliefs. Nonetheless, the lack of common knowledge entailed by the culture entails that the most inefficient equilibrium is selected.

Our theory highlights a novel and important channel though which cultural stereotypes might determine outcomes: via higher-order beliefs. In a culture where leaders are expected to be overconfident, outcomes under actual overconfidence play a powerful role even when a leader is not in fact overconfident. This can be contrasted with the existing view of the role of culture, namely that it coordinates expectations and thereby selects among equilibria, much as history does.

One possible normative implication of our results is that individuals who are not stereotypically expected to be overconfident might prove better leaders, as they are under less pressure to pursue unprofitable projects. Empirical and experimental work finds that women are often less likely to be overconfident, and that they are also perceived as not being overconfident. Our results suggest that this perception may enable such individuals to change course, since they suffer a smaller reputational penalty.