Skip to content
Menu
Finabase Learn

Investor psychology Lesson 14 of 15

Articles · Behavioral finance

Process over outcome: judging investment decisions under luck

Why a good result does not prove a good decision, what the outcome bias experiments found, and how many years it takes to tell an investing edge from luck.

On this page
  1. What the experiments found
  2. One decision, many outcomes
  3. How long does it take to tell skill from luck?
  4. What judging the process means
  5. What helps
  6. How Finabase helps
  7. Frequently asked questions
  8. Summary
  9. Check your understanding
  10. Exercise

Outcome bias is judging a decision by how it turned out instead of by what was known and done when it was made. A risky bet that paid off gets called smart. A careful decision that lost money gets called a mistake. Poker players call it "resulting".

In most of life the shortcut works well enough, because results and decision quality usually agree. Investing is the exception. Over any period an investor can actually observe, results are mostly noise, and the same decision made on a different day would have produced a very different number. This lesson shows the size of that noise on real data, what the experiments on outcome bias found, how long it takes to separate skill from luck, and what it means in practice to judge a process.

What the experiments found

Jonathan Baron and John Hershey ran five studies in which people evaluated decisions made by others. In one set of cases a doctor decides whether to operate on a patient, knowing the odds; the operation either succeeds or the patient dies. In another, someone chooses between gambles with stated probabilities. The readers were given everything the decision maker knew, including the odds, so the outcome told them nothing new about whether the choice was sound.

It changed their ratings anyway. When the result was good, readers rated the thinking behind the decision as better, rated the decision maker as more competent, and were more willing to let that person decide for them in future. Many of the same readers said, when asked, that the outcome should not matter. They knew the rule and could not apply it.

Rebecca Ratner and Kenneth Herbst showed how this changes behavior with money. Participants chose between two stockbrokers, one with a higher past success rate than the other (54% against 43%), and sensibly chose the better one. When that broker's next recommendation lost money, many were unwilling to use the same broker again, though one loss said almost nothing about a 54% success rate. The feeling about the result overrode the information about the odds.

One decision, many outcomes

The clearest way to see the gap between decision and outcome is to hold the decision fixed and vary only the luck. The decision here is as plain as investing gets: buy the SPDR S&P 500 ETF and hold it for 12 months. The figure shows the result for every starting month from March 1993 to November 2025, with dividends.

One decision, many outcomes: SPY over 12 months, by starting month 0 20 40 60 −50% Number of starting months 393 starting months from March 1993; total return. 18% ended with a loss; the range runs from −43% to +56%.
SPDR S&P 500 ETF (SPY), total return over 12 months for each starting month from March 1993 to November 2025 (393 overlapping periods).
Twelve-month result Value
Median +14.4%
One start in ten was below −11.2%
One start in ten was above +30.2%
Worst (started March 2008) −43.4%
Best (started April 2020) +56.2%
Share of starts with a loss 17.8%

Two investors make the identical decision, one in March 2008 and one in April 2020. A year later the first has lost 43% and the second has gained 56%. Nothing distinguishes their reasoning, their research or their discipline. If you grade them by the result, you will conclude that one is a fool and the other a genius, and you will have learned nothing true about either.

Now reverse it. Suppose an investor puts an entire retirement account into a single small company on a tip, and it triples. The outcome is excellent. The decision exposed everything to one company's fate without any basis for the odds, and repeating it will eventually end badly. Outcome bias is most dangerous in this direction, because a lucky win teaches the investor to do it again, with more money.

How long does it take to tell skill from luck?

If results are noisy, more results should help. They do, but slowly, and the arithmetic is worth seeing once.

Suppose a strategy truly beats its benchmark by an average of 2% a year (its edge), and its yearly result relative to the benchmark varies with a standard deviation of 6% (its noise; see tracking error and information ratio). Statisticians ask whether an observed average is distinguishable from zero using a t-statistic, and conventionally want it to be about 2 or more.

How many years of results it takes to tell an edge from luck 0 1 2 3 4 0 5 10 15 20 25 30 35 40 Usual bar for “probably not luck” (t = 2) Edge 2% a year, noise 6% (ratio 0.33) Edge 3%, noise 6% (ratio 0.5) Edge 6%, noise 6% (ratio 1.0) 36 years 16 years 4 years Years of track record t-statistic of the average yearly edge (edge ÷ noise × √years)
How the statistical evidence for an edge grows with the square root of time, for three ratios of yearly edge to yearly noise. Illustrative: assumes independent years and a constant edge.

A 2% yearly edge would be a remarkable achievement, and it would take a working lifetime of results to be reasonably sure it was not chance. A three-year or five-year record, which is what fund advertisements and brokerage statements show, cannot settle the question.

The same arithmetic runs in reverse for luck. Take 1,000 managers with no skill at all, each with a 50% chance of beating the market in any year. After five years about 31 of them will have beaten it five years in a row. After ten years, about one will have a perfect ten-year record. With thousands of funds and millions of individual traders, a spotless record somewhere is guaranteed.

Two large studies applied this logic to real funds. Eugene Fama and Kenneth French compared the actual spread of US stock fund results with the spread that luck alone would produce and concluded that few funds have enough skill to cover their costs. Laurent Barras, Olivier Scaillet and Russ Wermers estimated that 75.4% of funds had no skill in either direction after costs, 24.0% were unskilled and only 0.6% were skilled. Why most active funds underperform covers the fund evidence in detail.

What judging the process means

A process is the part of a decision that was under your control when you made it. For an investment decision, that comes down to a short list.

Process question Controllable? Outcome question
Did I have a reason I could write down? Yes Did the price go up?
Did I look for evidence against it? Yes Did the news go my way?
Was the position sized so that being wrong was survivable? Yes How much did I make or lose?
Did I decide the exit before entering, and follow it? Yes Did I sell at the top?
Did I know roughly how often this kind of decision works? Yes Did it work this time?

This gives four combinations, and the two off-diagonal ones are where the learning is.

Good outcome Bad outcome
Good process Deserved. Repeat. Bad luck. Repeat, and check the size.
Bad process Dumb luck. The most dangerous square. Deserved. The cheapest lesson available.

A worked example

An illustrative trader, Ana, has a tested rule with a 55% win rate, wins and losses of equal size, and risks 1% of a $50,000 account ($500) on each trade. Its expectancy is positive: 0.55 × $500 − 0.45 × $500 = $50 per trade (see edge and expectancy).

In one month she makes 20 trades by the rule and wins only 8. She loses 12 × $500 − 8 × $500 = $2,000, or 4% of the account. Is the rule broken?

With a true win rate of 55%, winning 8 or fewer of 20 happens about 13% of the time, roughly one month in eight. The outcome is unpleasant and entirely consistent with the process working. If Ana abandons the rule now, she is grading by outcome. If she keeps it without thought through 200 trades at a 40% win rate, she is ignoring evidence: that result would be far outside what luck allows. Judging the process does not mean ignoring results. It means asking how many results there are and how surprising they are.

The harder case is the opposite month. Ana breaks her rule, doubles her size on a hunch and wins $3,000. Nothing in the account will tell her this was the worst decision of the year. Only a record of what the rule said can.

What helps

No study shows that an investor who "focuses on process" earns more. The supportable claims are about what can and cannot be learned from results.

  • Write the reason and the odds first. A decision can be graded on its process only if the process was recorded before the outcome. See investment decision journal.
  • Grade the decision and the outcome separately. Two marks per decision, one for each. Over time the off-diagonal squares show where luck has been hiding a problem or a strength.
  • Count before concluding. Twenty trades or three years is a small sample. Ask how likely the result would be if nothing had changed.
  • Be most suspicious of big wins from broken rules. They are the outcomes that teach the wrong lesson most persuasively.
  • For funds and advisers, weigh costs and method over recent rank. Costs are known in advance. Recent rank is mostly the first table in this lesson.

How Finabase helps

The backtest page reports trades, wins, losses and win rate for a set of signals over a chosen period, which puts the sample size next to the result. The transaction history in paper trading lists every filled order with its date and price, a record of what was actually done that does not change with the outcome. The SPY chart on the MAX range shows the spread of one-year results that the histogram above summarizes.

Frequently asked questions

What is outcome bias?

Outcome bias is rating a decision as good or bad according to its result, even when the result depended largely on chance and adds no information about the quality of the choice.

What does "process over outcome" mean in investing?

It means judging decisions by what was controllable when they were made (the reasoning, the evidence, the position size, the exit plan and whether the plan was followed) and treating any single result as one noisy sample.

How many years of performance prove skill?

More than most records contain. An edge of 2% a year against 6% of yearly noise needs about 36 years to pass the usual statistical test. Shorter records can be suggestive, but three or five good years are well within what luck produces.

If a decision made money, was it not a good decision?

Not necessarily. A concentrated bet on a tip can pay off and still be a decision that would ruin the investor if repeated. The test is whether you would want to make the same decision, with the same information, a hundred times.

Summary

Outcome bias grades decisions by results. Baron and Hershey showed that people do this even when they have all the information the decision maker had and agree that the result should not count. In investing the gap between decision and result is wide: one fixed decision, a year in SPY, has returned anywhere from −43% to +56% depending on the starting month, and a real 2% edge takes about 36 years to distinguish from luck. Among funds, studies find very few with skill beyond their costs. Judging the process means recording the reasons, odds and size before the result, grading decision and outcome separately, and asking how surprising a run of results really is before changing course.

Check your understanding

5 questions on this lesson. Each answer comes with the reason, and nothing is scored.

  1. Two investors each bought an S&P 500 fund and held it for a year. One started in March 2008 and lost 43%; the other started in April 2020 and gained 56%. What can you conclude about their decisions?

    1. The second investor made the better decision
    2. They made the same decision; the difference is the period, which neither controlled
    3. The first investor should not have invested
    Show the answer

    B. They made the same decision; the difference is the period, which neither controlled Holding the decision fixed and varying only the starting month produces the whole range of results. Grading by result would call one a fool and one a genius for the same reasoning.

  2. A strategy has a true edge of 2% a year and yearly noise of 6%. About how many years of results does it take to reach a t-statistic of 2?

    1. 3 years
    2. 9 years
    3. 36 years
    Show the answer

    C. 36 years Years needed = (2 × noise ÷ edge)² = (2 × 6 ÷ 2)² = 36. Evidence grows with the square root of time, so short records cannot separate a modest edge from luck.

  3. A trader breaks her own rule, doubles her position on a hunch and makes $3,000. In the process and outcome grid, where does this trade sit?

    1. Good process, good outcome
    2. Bad process, good outcome: the square most likely to teach the wrong lesson
    3. Good process, bad outcome
    Show the answer

    B. Bad process, good outcome: the square most likely to teach the wrong lesson The account shows only the gain. Without a record of what the rule said, the lesson learned is to break the rule again, with more money.

  4. A rule with a true 55% win rate wins only 8 of 20 trades in a month. That happens by chance about 13% of the time. What is the reasonable response?

    1. Abandon the rule: it has stopped working
    2. Keep following it and keep counting: the result is unpleasant but well within what luck produces in 20 trades
    3. Double the position size to recover the loss
    Show the answer

    B. Keep following it and keep counting: the result is unpleasant but well within what luck produces in 20 trades One month in eight looks like this with nothing wrong. The same win rate over 200 trades would be strong evidence against the rule. Sample size decides how much a result says.

  5. Of 1,000 managers with no skill, each with a 50% chance of beating the market in a year, about how many will have beaten it five years in a row?

    1. None
    2. About 31
    3. About 250
    Show the answer

    B. About 31 1,000 ÷ 2⁵ ≈ 31. With enough participants, impressive records appear by chance, which is why a five-year streak is weak evidence of skill.

Exercise

One task to do in Finabase. Nothing is graded or saved.

Open Backtesting (signed out, it shows a sample; signed in, open one of your own). Read Trades, Wins, Losses and Win rate. Then look at the list of trades: find the longest run of losing trades in a row. Before you look, guess how long it will be. Would you have kept following the rule through that run if it had been your money?

Open Backtesting
What you should find

A rule that wins about half the time will usually show a run of four or more losses somewhere in a few dozen trades. Such a run is an outcome and says little about the rule. What says more is the total number of trades: with fewer than about 30, neither the win rate nor the losing run is strong evidence either way.

Sources and further reading

  1. Baron and Hershey — Outcome bias in decision evaluation, Journal of Personality and Social Psychology 54(4), 1988
  2. Ratner and Herbst — When good decisions have bad outcomes: the impact of affect on switching behavior, Organizational Behavior and Human Decision Processes 96(1), 2005
  3. Fama and French — Luck versus skill in the cross-section of mutual fund returns, Journal of Finance 65(5), 2010
  4. Barras, Scaillet and Wermers — False discoveries in mutual fund performance: measuring luck in estimated alphas, Journal of Finance 65(1), 2010
  5. Finabase chart data — SPDR S&P 500 ETF (SPY) monthly closes and dividends, February 1993 to October 2026