Confidence, Overconfidence, and Narcissism

Module 1 showed that CEOs are selected for confidence and optimism. This module asks what happens when those traits exceed the evidence — and it separates six words that practitioners use interchangeably and researchers do not.

RESEARCH FINDING The central finding is about shape, not direction. In a study of 111 technology CEOs, higher measured narcissism was associated with bolder, more changeable strategies and more volatile results — but not with better or worse average performance (Chatterjee & Hambrick, 2007). Pooled across 37 studies, CEO narcissism showed small positive associations with innovation and essentially no reliable link to average firm performance (Cragun, Olsen & Wright, 2020). Overconfident CEOs, by revealed-preference proxies, acquire more and their deals are received worse (Malmendier & Tate, 2008), yet in innovative industries they produce more innovation per R&D dollar (Hirshleifer, Low & Teoh, 2012).

INTERPRETATION Read together: overconfidence and narcissism widen the distribution of outcomes. They do not move its center. And the width depends on how much latitude the CEO has — discretion is the amplifier (Li & Tang, 2010).

This module works on the Traits term and on its interaction with Organizational Context — specifically governance and discretion, which decide whether an overconfident CEO's variance is contained or unleashed. It closes with a practical rubric, productive conviction versus epistemic arrogance, for reading any confident statement, including your own.

Overconfidence buys you a wider distribution of outcomes, not a better one — and how wide depends on how much nobody can stop you.

The confident CEO who is right looks like a genius and the one who is wrong looks like a cautionary tale, and the research says they were drawn from the same distribution. The question a board should ask is not "is this CEO confident?" — they all are — but "what is the variance this CEO's confidence produces, and can the company survive the left tail?"

Narcissism: the variance finding

RESEARCH FINDING Chatterjee and Hambrick (2007) built an unobtrusive narcissism index for 111 CEOs of computer hardware and software companies (1992–2004) from four proxies: prominence of the CEO's photograph in the annual report, prominence in press releases, first-person-singular pronoun use in interviews, and cash and non-cash pay relative to the second-highest-paid executive. Higher-scoring CEOs showed greater strategic dynamism and grandiosity, made more and larger acquisitions, and produced more extreme and fluctuating firm performance. On average their firms performed neither better nor worse than others. The index is a proxy, the sample is one industry, and the authors note narcissists may select into dynamic firms.

RESEARCH FINDING Chatterjee and Hambrick (2011) introduced "capability cues" — contextual signals about one's current ability — and found that CEO risk taking generally rose after positive cues and fell after negative ones. Highly narcissistic CEOs were much less responsive to objective performance feedback (recent firm results) than other CEOs, but considerably more responsive to social praise (media coverage, awards). Tested on risky capital outlays (1992–2006) and acquisition premiums (2001–2008) in US public companies, with the same proxy index.

RESEARCH FINDING Cragun, Olsen and Wright (2020) meta-analyzed 37 studies and concluded findings were "mixed and potentially dependent upon methods." Reported meta-analytic correlations: firm financial performance ρ ≈ .06 (k = 19, N ≈ 8,307; significant but small, with ROA and Tobin's Q individually non-significant); innovation/growth composite ρ ≈ .09; R&D spending ρ ≈ .11; new-product innovation ρ ≈ .14; risk taking ρ ≈ .03 (not significant); leverage ρ ≈ .07 (not significant); M&A frequency/size ρ ≈ .01 (not significant); CEO duality ρ ≈ .09; firm size ρ ≈ .08. The bibliography flags the effect sizes as extracted from a posted full text rather than the publisher page. Small k, heavy reliance on the Chatterjee–Hambrick index, and heterogeneity across measures limit all of these.

RESEARCH FINDING Buyl, Boone and Wade (2019), in 92 US commercial-bank CEOs (2006–2014), found archival markers of narcissism predicted riskier pre-crisis policies, especially with option-heavy pay; strong board monitoring dampened this; and narcissist-led banks recovered more slowly after September 2008.

Overconfidence: the revealed-preference lineage

RESEARCH FINDING Malmendier and Tate (2005) classified CEOs as overconfident when they persistently failed to reduce personal exposure to company risk — holding deep-in-the-money options past the point a rational diversifier would exercise (the "Longholder" proxy). In a Forbes 500 panel (1980–1994), such CEOs' corporate investment was significantly more sensitive to internal cash flow, especially in equity-dependent firms — consistent with overestimating project returns and viewing external finance as overly costly. The proxy could also reflect inside information, risk tolerance or signaling; the authors test and argue against these.

RESEARCH FINDING Malmendier and Tate (2008), using both option-holding and press-portrayal proxies, found overconfident CEOs about 65% more likely to make an acquisition, strongest for diversifying deals and when internal financing was available, with the market reacting more negatively (announcement return about −90 basis points versus −12 basis points for other CEOs). Malmendier, Tate and Yan (2011) found Longholder CEOs 37–49% less likely to issue equity when raising capital, leaving firms with higher leverage.

RESEARCH FINDING Hirshleifer, Low and Teoh (2012), in a large panel (1993–2003) using the same proxies, found overconfident CEOs' firms invested more in R&D, obtained more patents and citations, and achieved more innovation per R&D dollar — with higher stock-return volatility and no evidence of lower sales, ROA or Tobin's Q. The innovation advantage appeared only in innovative industries.

RESEARCH FINDING Ho, Huang, Lin and Yen (2016), in US banks (1994–2009) with an option-holding proxy, found overconfident-CEO banks loosened lending standards and raised leverage more before crises (market leverage roughly 5 percentage points higher) and suffered larger default increases, bigger performance drops and higher failure rates (about 10% versus 4%) during them. Fahlenbrach and Stulz (2011) found bank CEOs with the most "skin in the game" fared worst in 2008 and lost on average about $31.5 million personally — consistent with misjudged risk rather than self-dealing.

Discretion as amplifier

RESEARCH FINDING Li and Tang (2010), surveying 2,790 CEOs of Chinese manufacturing firms, found survey-measured hubris positively related to firm risk taking, and markedly more so under greater managerial discretion — munificent but complex markets, low organizational inertia, greater intangible resources, CEO–board chair duality, and CEOs who were not politically appointed. Single-country, cross-sectional, survey-based.

The paradox and the superstar

RESEARCH FINDING Zhang, Ou, Tsui and Wang (2017), in two studies of Chinese CEOs (63 and 143), found humility and narcissism were not mutually exclusive, and that the combination of high humility and high narcissism was associated with the strongest innovation outcomes, mediated by "socialized charisma." Narcissism was self-reported (NPI-16); humility was other-rated; correlational; one country.

RESEARCH FINDING Malmendier and Tate (2009) found CEOs who won prestigious business-press awards subsequently underperformed relative to their own prior record and matched non-winners, received higher pay, spent more time on outside activities, and ran firms with more earnings management — strongest under weak governance. Hayward, Rindova and Pollock (2004) supply the theory: media attribute firm actions to the CEO's disposition, the CEO internalizes it, and persists with the actions that produced the celebrity.

Where the evidence is weak

Nearly everything above uses proxies: photographs, pronouns, pay gaps, option-holding, press words. Each carries a known contamination (pay gaps reflect governance and size; option-holding reflects information and liquidity). Meta-analytic effects on performance are tiny and measure-dependent (Cragun et al., 2020). Direct measures (Zhang et al., 2017; Li & Tang, 2010) are self-report in one country. The sharpest evidence — banks in 2008 — is one crisis. The variance-not-mean claim rests mainly on one single-industry study plus the pattern across the rest; it is presented here as INTERPRETATION with strong support, not as a settled law.

Six words that are not synonyms

FRAMEWORK Practitioners use these interchangeably. Keep them apart.

Confidence is a belief in your capacity to act effectively. It can be calibrated or not; calibrated confidence is simply accuracy about yourself. Every CEO has it (Module 1), so its presence tells you nothing.

Ambition is a motive: the desire for achievement, scale or standing. It is not a belief about anything. Ambitious people can be humble about their odds and still want the outcome badly.

Self-belief — self-efficacy — is confidence attached to a specific task: "I can run this integration." It is the most useful of the six because it is testable against a track record.

Overconfidence is a calibration error: believing your estimates are more accurate, your projects more valuable, or your odds better than the evidence supports. It is not a personality; it is a mistake, and it can be measured by comparing belief to outcome. The Malmendier–Tate lineage captures it through revealed preference: the CEO who will not diversify is betting on themselves beyond what a rational person would.

Narcissism is a personality dimension: a grandiose self-view, a need for admiration, sensitivity to social praise, and a tendency to see others as audience. It differs from overconfidence in that it is about identity and audience, not about probability estimates. Chatterjee and Hambrick (2011) captured the difference precisely: narcissistic CEOs discount objective results and amplify risk after praise. An overconfident non-narcissist misjudges the odds; a narcissist misjudges who the game is for.

Grandiosity is the behavioral expression: the bold, visible, attention-getting action — the transformational acquisition, the headquarters, the keynote. Chatterjee and Hambrick (2007) measured it as an outcome of narcissism. It is what the board sees.

INTERPRETATION The chain that matters for practice: ambition and confidence are the fuel and are universal among CEOs; self-belief is the part you can audit; overconfidence is the part that costs money; narcissism is the part that makes the overconfidence resistant to correction; grandiosity is the symptom.

Variance, not mean

The most important idea in this module is a picture. Take a population of companies led by low-narcissism CEOs and plot their five-year returns: a bell curve with a certain center and width. Now take the high-narcissism CEOs: RESEARCH FINDING the same center, wider curve (Chatterjee & Hambrick, 2007). More big wins. More disasters. Same average.

INTERPRETATION This explains almost everything confusing about the practitioner conversation. The bold CEO who won is real and is on the right tail. The bold CEO who destroyed the company is real and is on the left. Both are cited as proof of a theory; neither is evidence about the center. The meta-analytic correlation with performance of ρ ≈ .06 (Cragun et al., 2020) is what a wide, centered distribution looks like when you average it.

The overconfidence lineage tells the same story with different proxies. Hirshleifer, Low and Teoh (2012) found overconfident CEOs produced more innovation and more volatility; Malmendier and Tate (2008) found more acquisitions and worse market reaction; Lee, Kim and Bae (2020) found founder-led firms — measurably more overconfident on proxies (Lee, Hwang & Chen, 2017) — produced more patents at both quality extremes, breakthroughs and duds. Variance is the consistent signal.

What this means for a board is a change of question. Not "is this CEO's confidence good?" but "what width of outcomes does this confidence produce here, and can we survive the left tail?" For a venture investor with a portfolio, wide variance is the business model. For a bank with depositors, it is the failure mode — which is why Ho et al. (2016) found 10% versus 4% failure rates.

Discretion is the amplifier

FRAMEWORK Personality × Power. A trait produces outcomes only in proportion to the latitude the CEO has to act on it (Hambrick, 2007). RESEARCH FINDING Li and Tang (2010) showed the hubris–risk link was markedly stronger when the CEO also chaired the board, faced less inertia, or had more intangible resources to redeploy. Buyl et al. (2019) found board monitoring dampened narcissism's effect on bank risk. Malmendier and Tate (2009) found the superstar effect strongest under weak governance. Malmendier and Tate (2008) found overconfident acquisitions most likely when internal cash was available — cash is discretion.

INTERPRETATION So the variance a given CEO produces is not fixed. It is the product of the trait and the constraints. The same narcissist runs a contained company under a strong chair and a strong CFO with covenants, and an unconstrained one with duality, a compliant board and a cash pile. Module 11 develops this fully. The practical point here: if you are worried about a CEO's overconfidence, the cheapest intervention is usually on the discretion side — governance, capital discipline, decision rights — not on the personality side.

When overconfidence is adaptive

RESEARCH FINDING In innovative industries, overconfident CEOs are associated with more R&D, more patents and more innovation per R&D dollar, without lower sales or ROA (Hirshleifer et al., 2012). Founders — more overconfident on proxies — sustain more exploratory patenting than the professionals who replace them, at similar spend, in a sudden-death quasi-experiment (Lee, Kim & Bae, 2020). Roughly 80% of US CEOs are "very optimistic" (Graham, Harvey & Puri, 2013), which suggests the trait is not merely tolerated but selected for.

INTERPRETATION The adaptive conditions share a structure: the downside is bounded and the upside is unbounded. Innovation portfolios, early-stage ventures and exploratory bets are convex — a string of failures costs a little, one success pays for all. Overconfidence there is a feature, because the calibrated person under-invests in convex bets. And Zhang et al. (2017) suggest the combination of narcissistic drive with other-rated humility produced the strongest innovation — conviction plus a correction mechanism.

When it is destructive

RESEARCH FINDING The destructive conditions have the opposite structure: bounded upside, unbounded downside. Acquisitions — especially diversifying ones with internal cash (Malmendier & Tate, 2008). Leverage — Longholders lean on debt (Malmendier, Tate & Yan, 2011), and overconfident bank CEOs raised leverage and loosened lending before crises (Ho et al., 2016). Ignoring feedback — narcissistic CEOs discount results and amplify risk after praise (Chatterjee & Hambrick, 2011), and award winners underperform afterward (Malmendier & Tate, 2009).

INTERPRETATION The common thread is that overconfidence removes the correction mechanism exactly where the correction is most valuable. In a convex bet, being wrong is cheap and correction is optional. In a concave one — a levered acquisition, a bank's loan book — being wrong is catastrophic and correction is everything.

Productive conviction versus epistemic arrogance

FRAMEWORK The test is a contrast between two statements:

"We can probably accomplish this. Here is the evidence, the assumptions, and what would make us change course."

versus

"We will accomplish this. People who disagree simply don't understand the vision."

Both are confident. The first keeps confidence connected to evidence and names its own kill criteria; the second treats disagreement as a deficiency in the listener. The first is the Two-Sentence CEO — "We're going to do this" with "I was wrong. Change the plan." held in reserve. The second has amputated the second sentence.

Run it on real statements: board decks, all-hands transcripts, your own emails. Look for three markers. Probability language — is the claim hedged with an honest estimate or asserted as fact? Named assumptions — can the speaker say what has to be true? Falsifiability — has the speaker said what would change their mind? A statement with all three is productive conviction whatever its boldness. A statement with none is epistemic arrogance whatever its politeness.

This is a fictional composite.

Pellucid Systems is a listed software company: $180 million in annual recurring revenue, 900 employees, dual-class shares giving the founder-CEO, Victor Hale, 58% of the votes with 14% of the equity. Hale is, by any observer's account, high on every marker this module discusses: his photograph fills a page of the annual report, his earnings-call answers are dense with "I," and he is paid 4.1 times his CFO.

In 2019 Hale bet the company on rebuilding the core platform from scratch, against the advice of his CTO and two board members. The rebuild took twenty-two months, cost $60 million, and lost the company two of its ten largest customers during the transition. It also produced a product two years ahead of the market. By 2022 ARR had doubled and Pellucid's share price had tripled. Hale was profiled twice as a visionary. The board members who had opposed the rebuild rotated off.

In 2023 Hale announced the acquisition of a competitor for $310 million — 1.7 times Pellucid's cash and a premium of 46% — financed with a term loan and convertible notes. The board approved it in a single meeting. The CFO's memo noted that the target's churn was rising and that the integration would require the platform team that was still stabilizing the rebuild; the memo was circulated but not discussed. Hale's response to an analyst who questioned the price was that the analyst "had never built anything."

The acquisition lost 30% of the target's revenue in eighteen months. Pellucid breached a covenant in 2025, renegotiated at a cost of $9 million, cut 15% of staff, and traded back to its 2021 price.

INTERPRETATION Here is the point. Hale's board treated the 2019 rebuild as evidence about Hale's judgment. It was evidence about his variance. The rebuild and the acquisition were draws from the same distribution: bold, self-authored, resistant to dissent, with a wide range of possible outcomes. One landed on the right tail and one on the left, and the board's error was to read the first as a reason to remove the constraints before the second.

Notice the amplifiers. Dual-class control, a board that rotated out dissent, cash on hand, a CEO who responded to praise more than to the CFO's churn data (Chatterjee & Hambrick, 2011) — every discretion-enhancing condition Li and Tang (2010) describe was present. And notice the statement to the analyst: no probability, no assumptions, no kill criteria, and disagreement attributed to the listener's deficiency. It was epistemic arrogance in one sentence, published, and nobody on the board ran the test.

Put four archetypes into Hale's seat in 2023, facing the same target at the same price after the same triumphant rebuild.

The Visionary CEO — Hale himself — sees the acquisition as the rebuild's second act and the premium as the price of speed. Gain: if the target's churn was a fixable product problem and Pellucid's platform fixes it, the combined company dominates. Cost: the outcome distribution is wide, the financing makes the left tail fatal, and the Visionary's own success has removed the people who would have said so.

The Operator CEO reads the CFO's memo first, sees a platform team that is not ready, and passes or delays. Gain: no covenant breach, no layoffs, a rebuild that finishes stabilizing. Cost: the competitor may be bought by someone else, and an Operator who passes on every wide-variance bet in a software market eventually runs a slow company in a fast industry. Overconfidence is adaptive in convex settings, and an Operator may under-invest in them.

The Public Company CEO without dual-class control faces a board that can actually say no. The same proposal goes through a special committee, an independent fairness opinion and a financing structure the CFO can live with. Gain: a contained version of the bet — maybe a lower price, maybe an earn-out on churn. Cost: slower, and a board that might block a deal the Visionary would have gotten right. This is what discretion-limiting governance does: it narrows the distribution in both directions.

The Technology CEO with a humility practice — the Zhang et al. (2017) combination — wants the deal as much as Hale does and still writes down what has to be true, asks the CTO to argue the case against, and sets the walk-away price before the negotiation rather than during it. Gain: the conviction survives with a correction mechanism attached; if the churn data is disqualifying, the CEO finds out before signing. Cost: the process is slower and visibly less heroic, and a CEO who publicly names kill criteria is easier to hold to them.

INTERPRETATION The Visionary and the Operator sit at the two ends of the aggression–caution dial; each has a context that rewards them. The Public Company CEO shows what governance does to variance. The fourth archetype is the module's answer: the same conviction, with the second sentence kept in reach.

Overconfidence and narcissism climb four rungs of the Overuse Ladder at once.

FRAMEWORK Confidence → arrogance: the CEO's certainty is treated as evidence, and dissent as a failure of the dissenter. Optimism → delusion: forecasts detach from track record; the CFO's memo is circulated but not discussed. Vision → fantasy: the story about the last success becomes the plan for the next one, regardless of whether the conditions match. Risk tolerance → recklessness: the bounded-downside bet that worked is repeated with unbounded downside — internal cash becomes leverage, exploration becomes acquisition.

The mechanism is the one Hayward, Rindova and Pollock (2004) theorized and Malmendier and Tate (2009) measured: success is attributed to the CEO's disposition, the CEO believes it, and persists with the actions that produced the attribution after those actions stop paying. Chatterjee and Hambrick (2011) add the diagnostic: the narcissistic CEO's risk appetite tracks praise rather than results. Once the correction mechanism is gone, variance does the rest.

Early warning signs a CEO or board could notice:

• The CEO's public profile is growing faster than the company's results, and the CEO enjoys it (Malmendier & Tate, 2009).
• Risk appetite rises after an award or a profile, not after a good quarter (Chatterjee & Hambrick, 2011).
• Acquisitions are proposed when cash is available rather than when targets are (Malmendier & Tate, 2008).
• Dissenting board members or executives have rotated out and been replaced by admirers.
• The CEO's statements fail the productive-conviction test: no probabilities, no named assumptions, no kill criteria, disagreement attributed to the listener.
• Leverage has risen and the CEO treats equity as "too cheap to issue" (Malmendier, Tate & Yan, 2011).
• The CEO holds deep-in-the-money options unexercised well past vesting — the Longholder pattern — while urging the board to bet the balance sheet.

The correction is rarely to make the CEO less confident, which the population does not permit. It is to restore the constraints — governance, capital discipline, a CFO with standing, pre-registered kill criteria — that convert unbounded variance into bounded variance, and to insist on the second sentence.

1. Take your last three significant decisions and write, for each, the probability you assigned at the time and the outcome. Were you calibrated, or were you drawing from a wider distribution than you thought?
2. When did you last change a decision because of results, and when because of praise or criticism from outside? Which moved you more?
3. Which of the six constructs — confidence, ambition, self-belief, overconfidence, narcissism, grandiosity — would your CFO say describes you? Would you agree, and how would you know?
4. Find a statement you have made to your board or company in the last year. Run the productive-conviction test: probability, assumptions, kill criteria. Which markers are missing?
5. List the constraints on your discretion — board, covenants, ownership, a strong number two. Which have you loosened in the last three years, and after which success?
6. Is your current biggest bet convex (bounded downside, unbounded upside) or concave? If concave, what is the correction mechanism, and who owns it besides you?
7. What would have to be true for your current strategy to be wrong? Have you said that sentence to anyone?

• Confidence, ambition, self-belief, overconfidence, narcissism and grandiosity are different constructs: the first three are universal among CEOs and uninformative; overconfidence is a calibration error that costs money; narcissism makes it resistant to correction; grandiosity is the visible symptom.
• Narcissism and overconfidence widen the distribution of outcomes without moving its center: bolder strategies, more volatile results, no average advantage (Chatterjee & Hambrick, 2007; Cragun et al., 2020).
• Discretion is the amplifier: the hubris–risk link strengthens with duality, low inertia and available cash, and weakens under board monitoring and strong governance (Li & Tang, 2010; Buyl et al., 2019; Malmendier & Tate, 2008).
• Overconfidence is adaptive in convex settings — innovation, exploration, early-stage bets — and destructive in concave ones — levered acquisitions, bank leverage, and any situation where feedback is being ignored (Hirshleifer et al., 2012; Malmendier & Tate, 2008; Ho et al., 2016; Chatterjee & Hambrick, 2011).
• Test every confident statement, including your own, for probability language, named assumptions and kill criteria. "We can probably accomplish this. Here is the evidence, the assumptions, and what would make us change course." is productive conviction; "We will accomplish this. People who disagree simply don't understand the vision." is epistemic arrogance.