Every budget, valuation range, cash flow projection and year-end estimate is a confidence interval, usually an unstated one. This article is about the finding that those intervals are systematically too narrow, and that this particular error resists nearly everything.
Key Takeaway
Moore and Healy reconcile "the three distinct ways in which the research literature has defined overconfidence: (1) overestimation of one's actual performance, (2) overplacement of one's performance relative to others, and (3) excessive precision in one's beliefs." And: "Overprecision appears to be more persistent than either of the other two types of overconfidence."[1] In a decade-long survey, "only 36% of future stock market realisations fall within their 80% confidence intervals", with an average interval width of 9.4 percent against a historical range the authors put at about 40[2].
Our Grades For These Claims
Applying the scheme from the first article in this series.
Grade A for the three-way distinction. A theoretical paper in a leading journal, its abstract obtained verbatim from the authors' own institutional repository.
Grade B that overprecision is the most persistent of the three, which is the paper's own summary sentence and which we did not see the supporting evidence for.
Grade B for the CFO figures, which reach us through a column written by the study's own authors rather than the paper.
Grade C for our own reconciliation of those figures, because our check produces a lower number than the study reports and we could not resolve the gap.
Our position: the distinction is solid, the practical finding is striking, and one piece of its arithmetic does not tie out for us, which we report rather than smooth over.
A Note On Method
Everything here is verified to August 2026.
We obtained the 2008 paper's abstract verbatim from the lead institution's repository[1]. We did not obtain the paper.
The CFO figures come from a research column written by the study's authors describing their own work[2], and one further figure from the working paper version[3]. We did not obtain the published paper.
The consequences of overprecision are reported from a preprint's literature review[4], which is a summary rather than a source, and we flag it at each use.
All arithmetic is ours. The volatility figures in this article are implied by us from two reported numbers and are stated by nobody.
This article discusses forecasting research. It is not investment advice, and the market figures in it are used only to check a published claim.
Three Things, One Word
The distinction, and it is the paper's main contribution.
Moore and Healy published The Trouble with Overconfidence in Psychological Review, 115(2), 502–517, in 2008[1].
They describe the paper as "a reconciliation of the three distinct ways in which the research literature has defined overconfidence: (1) overestimation of one's actual performance, (2) overplacement of one's performance relative to others, and (3) excessive precision in one's beliefs."[1]
Put plainly, and this is ours: overestimation is thinking you did better than you did. Overplacement is thinking you did better than others. Overprecision is being too sure of whatever you believe.
Two observations, ours.
These are not three flavours of one thing. They are three different measurements, and the paper exists because the literature had been using one word for all of them.
And the structure is exactly the twenty-eighth article's, which found three distinct effects filed under framing. Advice built on one of them transfers badly to the others, and that is the recurring cost of a shared label.
They Run In Opposite Directions
The finding that proves the distinction is not pedantry.
"Experimental evidence shows that reversals of the first two (apparent underconfidence), when they occur, tend to be on different types of tasks. On difficult tasks, people overestimate their actual performances but also believe that they are worse than others; on easy tasks, people underestimate their actual performances but believe they are better than others."[1]
Three observations, ours.
On a hard task the two move in opposite directions at once: too generous about your own score, too pessimistic about your ranking.
On an easy task both reverse. You underrate your score and overrate your standing.
Which means a single question about whether people are overconfident has no answer without specifying which kind and on what difficulty. That is a considerably more useful result than a headline effect.
The One That Persists
The sentence this article is built on.
"Overprecision appears to be more persistent than either of the other two types of overconfidence, but its presence reduces the magnitude of both overestimation and overplacement."[1]
We did not obtain the evidence for this and report the authors' own summary.
Three observations, ours.
More persistent is the practical point. Two of the three reverse depending on task difficulty. The third apparently does not.
The second clause is subtler and worth sitting with: overprecision's presence reduces the magnitude of the other two. Being very sure of your beliefs makes you less likely to misjudge your score or your rank, because certainty about the world crowds out error about yourself.
And it explains why overprecision is the least discussed of the three in business writing. The other two are about self-image and make for better stories. This one is about the width of a number.
Ten Years Of CFOs
The study that puts this squarely in an accountant's territory.
Ben-David, Graham and Harvey published Managerial Miscalibration in the Quarterly Journal of Economics, 128(4)[5], with a working paper version circulated earlier[3].
Writing about their own work, the authors report: "CFOs are severely miscalibrated, which means that the actual market return rarely falls within the CFO's 80% confidence bounds, only 36% of future stock market realisations fall within their 80% confidence intervals."[2]
And: "The average confidence interval provided by CFOs was only 9.4%, much narrower than the actual 40% range based on historical stock market volatility."[2]
Three observations, ours.
These are chief financial officers, being asked a question inside their professional competence, in a survey run repeatedly over years.
An 80 percent interval landing 36 percent of the time means the stated confidence was wrong by more than a factor of two, in the direction of certainty.
And the definition the authors work with is worth keeping: "Miscalibration is defined as excessive confidence about having accurate information."[2] That is overprecision under another name.
We Checked The Arithmetic
Because two published numbers should be consistent with each other. Our own calculation, using a normal approximation; the volatility figures below are implied by us and stated by nobody.
An 80 percent interval spans roughly 2.56 standard deviations. So a 9.4 percent width implies the forecaster is behaving as though annual volatility were about 3.7 percent.
The 40 percent range the authors describe as historical implies about 15.6 percent.
One observation. On these implied figures the CFOs were forecasting as if the world were about 4.3 times calmer than it was. That is a cleaner way to state the finding than a coverage percentage, because it names what was actually mis-estimated, which is variance rather than direction.
A Discrepancy We Could Not Resolve
Reported because our check does not tie out, and hiding that would be worse than the discrepancy.
If true volatility is the 15.6 percent implied above, an interval of plus or minus 4.7 percent should contain the outcome about 24 percent of the time.
The study reports 36 percent.
Our figure is lower than the observed one, meaning the intervals performed better than their average width alone would predict.
Three possible reasons, ours and unresolved.
The intervals were not all the same width, and an average of widths does not combine with an average of coverage the way our calculation assumes.
Realised volatility over the sample period may have differed from the long-run figure the 40 percent range is based on.
And returns are not normal. If the middle of the distribution is fatter than a normal curve, a narrow interval catches more than our approximation predicts.
One observation. We did not obtain the paper and cannot settle this. The direction of the finding is unaffected either way; a 36 percent hit rate on an 80 percent interval remains a large miss.
The Honest Interval
What a correctly stated range would look like, and why nobody says it. Ours, on the implied volatility above.
At the implied 15.6 percent, a genuine interval would need to be about 21 points wide for 50 percent confidence, 40 points for 80 percent, 51 points for 90, and 61 points for 95.
Three observations.
An honest 80 percent interval on this quantity is forty points wide, which is a range so large that stating it feels like admitting you do not know.
Which is the actual barrier, and it is social rather than cognitive. The correct answer is embarrassing to say out loud in a meeting, to a lender, or to a client.
And it generalises past markets. Any forecast of a genuinely uncertain quantity has an honest interval wider than the one people are comfortable presenting, which makes the width of a stated range partly a social artifact rather than an estimate.
It Showed Up In The Policies
The consequence, which is what makes this more than a curiosity about surveys.
The authors report: "The results suggest not only that CFOs are miscalibrated, but also that firms with miscalibrated executives appear to be more aggressive in their corporate policies. In other words, the overconfidence of the executive shows up in the company's policies."[2]
From the working paper: "A one standard deviation shift in long-term overprecision is associated with a shift of 0.6 of a percentage point (t = 2.7) in corporate investment (relative to a mean of 8.7 percentage points and a median of 5.4 percentage points)."[3]
Three observations, ours.
That is an association, and the authors' own phrasing is appear to be. We report it as a correlation with a reported t-statistic, not as a demonstrated cause.
The magnitude deserves attention rather than the significance. A shift of 0.6 points against a mean of 8.7 is roughly a fourteenth of the average investment rate, which is real but not dramatic.
And the mechanism is worth naming: the working paper cites an argument that a miscalibrated manager chooses a higher level of debt because she overestimates her firm's ability to meet its liabilities[3]. Underestimating variance is precisely how a survivable leverage decision becomes an unsurvivable one.
Why Feedback Does Not Fix It
The observation that makes this difficult to remedy.
A preprint's literature review states: "Frequent feedback should help people calibrate their confidence; being routinely wrong should reduce confidence in the next forecast. However, the robustness and durability of overprecision suggests this corrective may be incomplete."[4]
This reaches us through a preprint's summary of the literature, not from a primary source.
Three observations, ours.
The expectation is entirely reasonable: someone whose ranges keep missing should widen them.
That it apparently does not happen connects to the twelfth article in this series, which found that feedback interventions made performance worse in over a third of cases. Being told you were wrong is not the same as learning how wrong you tend to be.
And it suggests the remedy is not experience but record-keeping. A forecaster who cannot recall their previous ranges cannot notice a pattern in how often those ranges missed.
What Overprecision Does
The catalogue, from the same preprint's literature review and therefore reported at one remove.
It states that overprecision contributes to "managers issuing too much debt when they underestimate the volatility of their firm's future", to "issuing excessively precise and inaccurate earnings forecasts", and leads people to "do too little to protect themselves against low-probability risks"[4].
We obtained none of the underlying studies and report this list as a literature review's characterisation.
Two observations, ours.
Note that all three are variance errors, not direction errors. Nobody in that list was wrong about which way things were going; they were wrong about how far.
And doing too little about low-probability risks is the one a small business feels hardest. Insurance, redundancy, a cash buffer and a second supplier are all purchases whose value depends entirely on the tails you believe in.
Why This Explains The Advice Article
A connection the sources make explicitly, and it closes a loop in this series.
The same review states that excessive faith in one's beliefs "can also lead people to discount others' views", and that overprecision "blinds people to the need to consider other perspectives"[4].
Two observations, ours.
The thirty-seventh article reported that people favour their own opinion over advice even when the advice is good, and could not say why. This is a candidate mechanism: if you are too certain of your own estimate, correctly weighting someone else's requires you to be less certain than you are.
We flag that connecting these two literatures is ours, not something either source proposes, and that the review sentence is a summary rather than a finding we have read.
Does It Survive An Organisation
The obvious hope, and a paper that asks it in its title.
A reference list identifies Meikle, N. L., Tenney, E. R., and Moore, D. A. (2016), Overconfidence at work: Does overconfidence survive the checks and balances of organizational life?, Research in Organizational Behavior, 36, 121–134[6].
We did not obtain it and report the title only.
Two observations, ours.
The question is the right one, because the hope is always that review, approval and challenge will catch what an individual misses.
And we cannot tell you the answer, which we would rather say than infer from a title. We note only that the previous article in this series found groups deciding in sequence discarding most of their information, so the assumption that organisational structure corrects individual error is not one this series has found much support for.
A Fifteenth Variant
The running sourcing note.
One academic reference list renders the 2008 paper's title as The trouble with overconfidece[6], dropping a letter from the word the paper is about.
One observation, ours. That brings this series' count of bibliographic variants to fifteen. It is trivial and it is in a peer-reviewed venue's reference list, which is the point: the citation layer of the literature is noisier than the literature.
What To Do
Say which kind you mean. Overconfidence is three measurements, and two of them reverse direction depending on task difficulty.
Widen your ranges, then widen them again. CFOs asked for an 80 percent interval on market returns landed inside it 36 percent of the time, with an average width the authors put at 9.4 percent against a historical 40.
State the interval, not the point. A single number is a confidence interval of width zero, and it is the least honest form a forecast can take.
Expect the honest range to be embarrassing. On our own implied figures a genuine 80 percent interval here is about forty points wide, and the barrier to saying that is social rather than cognitive.
Keep a record of your own past ranges. Feedback alone appears not to correct this, and you cannot notice a pattern in your misses without the previous ranges written down.
Watch the variance decisions specifically. Debt levels, insurance, buffers and single-supplier arrangements are all priced off the tails you believe in.
Treat a very confident colleague as a forecasting risk, not a forecasting asset. Narrower ranges are not better ranges, and the literature associates certainty with discounting other people's views.
Do not assume review will catch it. A paper asks in its title whether overconfidence survives organisational checks and balances, and we could not obtain its answer.
The Limits Of This Analysis
Several caveats matter. This article discusses forecasting research and is not investment advice; market figures appear only to check a published claim. Everything is verified to August 2026. We did not obtain any paper discussed in full. The 2008 abstract was obtained verbatim from the lead institution's repository, and none of its supporting evidence. The CFO figures reach us through a column written by the study's own authors plus one figure from a working paper version, not from the published paper. The consequences of overprecision are reported from a preprint's literature review, which is a summary rather than a source, and none of the underlying studies were obtained. All arithmetic is ours, uses a normal approximation, and the volatility figures are implied by us from two reported numbers and stated by nobody. Our reconciliation of those two numbers produces 24 percent coverage against the 36 percent the study reports; we offer three possible explanations, resolve none of them, and did not obtain the paper that would settle it. The association between miscalibration and corporate policy is reported as a correlation with a reported t-statistic, in the authors' own hedged phrasing, not as demonstrated cause. The connection drawn to the article on advice discounting is ours and is proposed by neither source.
Frequently Asked Questions
What are the three types of overconfidence?
Why does the distinction matter?
What did the CFO study find?
Did their arithmetic check out?
Why doesn't experience fix it?
What should I actually change?
References
- Institutional repository record for Moore, Don A., & Healy, Paul J., The Trouble with Overconfidence, Psychological Review, 115(2), 502–517, 2008, reproducing the abstract, on the paper presenting a reconciliation of the three distinct ways in which the research literature has defined overconfidence, being overestimation of one's actual performance, overplacement of one's performance relative to others, and excessive precision in one's beliefs; on experimental evidence showing that reversals of the first two, appearing as underconfidence, tend to occur on different types of tasks, with people on difficult tasks overestimating their actual performances while believing they are worse than others, and on easy tasks underestimating their performances while believing they are better than others; on the paper offering a theory to explain these inconsistencies; and on overprecision appearing to be more persistent than either of the other two types of overconfidence, while its presence reduces the magnitude of both overestimation and overplacement. Note: the lead author's institutional repository, reproducing the abstract. We obtained the abstract only and none of the paper's supporting evidence. kilthub.cmu.edu
- Research policy column by the authors of the managerial miscalibration study describing their own work, on miscalibration being defined as excessive confidence about having accurate information, citing Alpert and Raiffa (1982), Lichtenstein and colleagues (1982), Ronis and Yates (1987), Kyle and Wang (1997) and Moore and Healy (2008); on CFOs being severely miscalibrated, with the actual market return rarely falling within their 80 percent confidence bounds and only 36 percent of future stock market realisations falling within those intervals; on the average confidence interval provided by CFOs being only 9.4 percent, much narrower than the actual 40 percent range based on historical stock market volatility; and on the results suggesting not only that CFOs are miscalibrated but that firms with miscalibrated executives appear to be more aggressive in their corporate policies, with the overconfidence of the executive showing up in the company's policies. Note: a column written by the study's own authors, not the peer-reviewed paper. Our principal source for the CFO figures. cepr.org
- Working paper version of Managerial Miscalibration by Ben-David, Itzhak, Graham, John R., and Harvey, Campbell R., National Bureau of Economic Research working paper series, on a one standard deviation shift in long-term overprecision being associated with a shift of 0.6 of a percentage point, with a t-statistic of 2.7, in corporate investment, relative to a mean of 8.7 percentage points and a median of 5.4 percentage points; on that evidence resonating with Kahneman and Lovallo (1993), who theorize that executives are often optimistic and overconfident about their managerial skills and their ability to mitigate risk; on Heaton (2002) showing that optimistic managers invest more because they perceive negative net present value projects to be good projects; and on Hackbarth (2008) arguing that a miscalibrated manager chooses a higher level of debt because she overestimates her firm's ability to meet its liabilities. Note: a working paper version, not the published article. We obtained these passages and not the paper's methods or full results. nber.org
- Preprint literature review on overprecision and surprise, on overprecision being overconfidence in the accuracy of one's beliefs; on that excessive certainty appearing when people think they know how their friends will behave, when doctors are too certain of a favored diagnosis, and when managers issue excessively precise and inaccurate earnings forecasts; on frequent feedback being expected to help people calibrate their confidence while the robustness and durability of overprecision suggests this corrective may be incomplete; on overprecision contributing to managers issuing too much debt when they underestimate the volatility of their firm's future and to excessively precise and inaccurate earnings forecasts; on it leading people to do too little to protect themselves against low-probability risks; and on excessive faith in one's beliefs leading people to discount others' views and blinding them to the need to consider other perspectives. Note: a preprint's literature review, which is a summary rather than a primary source. We obtained none of the underlying studies it characterises. biorxiv.org
- Academic reference list confirming Ben-David, I., Graham, J. R., and Harvey, C. R., Managerial miscalibration, The Quarterly Journal of Economics, volume 128, issue 4; and identifying Moore, D. A., and Schatz, D. (2017), The three faces of overconfidence, Social and Personality Psychology Compass, 11(8), 1–12, together with Radzevick, J. R., and Moore, D. A. (2011), Competing to be certain (but wrong): Market dynamics and excessive confidence in judgment, Management Science, 57, 93–106. Note: citations only, used to confirm the journal, volume and issue of the CFO study. We obtained none of these papers. arxiv.org
- Academic paper reference list identifying Meikle, N. L., Tenney, E. R., and Moore, D. A. (2016), Overconfidence at work: Does overconfidence survive the checks and balances of organizational life?, Research in Organizational Behavior, 36, 121–134; Moore, D. A., and Schatz, D. (2017), The three faces of overconfidence, Social and Personality Psychology Compass, 11(8), e12331; and Malmendier, U., and Taylor, T. (2015), On the verges of overconfidence, Journal of Economic Perspectives, 29(4), 3–8. Note: citations only; we did not obtain any of these works and report the 2016 title without its answer. A separate academic reference list renders the 2008 paper's title as "The trouble with overconfidece", dropping a letter. arxiv.org
This article discusses forecasting research and is not investment advice. No paper discussed was obtained in full. The CFO figures reach this article through a column written by the study's own authors rather than the peer-reviewed paper. All arithmetic is the authors' own; the volatility figures are implied by them from two reported numbers and are stated by nobody, and their reconciliation of those numbers produces a lower coverage figure than the study reports, which is disclosed and unresolved.