The seven cognitive biases that show up in real trade data

Last reviewed 19 May 2026.

If you have read Thinking, Fast and Slow you already know the vocabulary. You know that loss aversion means losses hurt about twice as much as equivalent gains feel good, and you can define the disposition effect in your sleep. What you probably do not know is whether you, specifically, do any of these things with real money — because knowing the concept and seeing it in your own account are two entirely different acts. Self-report is worthless here. People who can recite the entire heuristics-and-biases literature still hold their losers.

This post is about the evidence, not the theory. Each of the seven biases below maps one-to-one onto a tag StackViewer applies to your trades, and each one leaves a footprint you can compute by hand from a broker export. For every bias you get four things: a one-line definition, the exact trace it leaves in trade data, a spreadsheet test you can run this afternoon, and — this matters — the honest counterpoint, the case where the same pattern is a deliberate, rational strategy rather than a bug in your wiring.

How a bias leaves a footprint in data

A cognitive bias is invisible. Its consequences are not. Every bias, if it is real and repeated, changes at least one of four measurable things: which trades you take, when you take them relative to price and news, how big you size them, and how long you hold them. You can reconstruct all four from a plain export — most brokers give you a CSV with symbol, side, quantity, price, and timestamp. That is enough to build every test in this article; you do not need anything fancier than a pivot table.

The unit that does most of the work is the round trip: a buy paired with the sell that closed it. From a round trip you get the realised profit or loss, the holding period in days, and the entry and exit prices. Group your round trips into winners (closed for a gain) and losers (closed for a loss), and most behavioural patterns fall out of two summary numbers per group: the count and the average holding period. Here is the manual calculation for the single most-cited pattern, the disposition effect, so you can see the shape of the method before the product is mentioned at all.

Worked example — the Proportion of Gains Realised (PGR) vs Proportion of Losses Realised (PLR). On any given day, look at every position that is currently showing a paper gain and count how many you sold that day; divide sold-winners by total winners available to sell. That is PGR. Do the same for losers to get PLR. Suppose over a quarter you had 200 winner-days and sold on 60 of them (PGR = 0.30), and 150 loser-days but only sold on 15 (PLR = 0.10). Your disposition ratio is PGR ÷ PLR = 3.0 — you realise gains three times as readily as losses. Odean's 1998 study found a population ratio around 1.5; anything materially above 1.0 says you cut winners faster than losers. The maths is arithmetic. The interpretation is where judgement comes in.

The seven biases, one by one

Below, each bias is treated the same way: definition, data trace, self-test, counterpoint. The self-tests all assume you have exported round trips into a spreadsheet. None of them require the product — that is deliberate. If you would rather not build the pivot tables, StackViewer applies these exact seven tags automatically, but the maths belongs to you either way.

1. Disposition effect. Definition: the tendency to sell assets that have risen and hold assets that have fallen. Trace: average holding period on losers longer than on winners, and a PGR/PLR ratio above 1. Self-test: two AVERAGEIF formulas — one for the mean holding days of gain round trips, one for losses — and the PGR/PLR calculation above. Counterpoint: if you run a trend or trailing-stop system, cutting losers quickly and letting a few winners run is the entire point; a low, deliberate stop that fires on losers will look identical to disposition and be exactly right. The tell is whether the exits were rule-based or reactive.

2. Overconfidence. Definition: overestimating the accuracy of your own judgement, usually after success. Trace: position size (as a percentage of portfolio) trending upward after a run of wins, and trade frequency rising. Self-test: plot position size against a rolling count of your last five outcomes; if size climbs when the recent record is green, you are sizing on mood, not edge. Counterpoint: scaling exposure as an edge is statistically validated is a legitimate technique — the difference is fixed risk-per-trade and a written rule, versus "I'm hot right now." Overconfidence also famously drives over-trading; more trades is not more skill.

3. Loss aversion. Definition: weighting a potential loss roughly twice as heavily as an equivalent gain, which distorts risk-taking near red positions. Trace: averaging down into losers, widening or removing stops, and taking larger risk specifically to avoid booking a loss. Self-test: flag every add to an existing position and check whether the add happened above or below your average cost; a strong skew toward adding below cost is the footprint. Counterpoint: pre-planned tranche buying with a hard total-risk cap is a valid accumulation strategy and will look like averaging down. Loss aversion is the version with no cap and a moving stop.

4. Anchoring. Definition: fixing on a reference number — usually your purchase price or a past 52-week high — and judging every later decision relative to it. Trace: sell and add decisions clustering suspiciously close to your entry price ("I'll sell when I'm back to breakeven") or to a prior peak. Self-test: compute exit price as a percentage of entry price for every round trip and build a histogram; a spike right around 100% (breakeven) is anchoring you can see. Counterpoint: sometimes the entry price coincides with a genuine technical support or supply level that other participants also watch — in that case the anchor is a real feature of the market, not just of your head.

5. Recency bias. Definition: over-weighting the most recent data and under-weighting the base rate. Trace: your buys rotate toward whatever sector or style has led over the last few weeks, and away from recent laggards, regardless of longer-term evidence. Self-test: tag each buy with the trailing four-week return of its sector at purchase, then check whether your purchases skew heavily toward high recent-return sectors. A strong tilt is recency. Counterpoint: if you run a documented momentum or trend-following mandate, buying recent strength is the strategy — recency is only a bias when it overrides a plan you claim to follow.

6. Confirmation bias. Definition: seeking and acting on information that supports an existing thesis while discounting what contradicts it. Trace: every trade in a given name points the same direction, you never trim on disconfirming news, and there is no record of you revisiting the thesis. Self-test: for your five largest positions, list every trade and note whether any acted against the prevailing thesis; an unbroken one-way record is the pattern. Counterpoint: a genuinely high-conviction, concentrated strategy also produces one-way trading — the honest distinction is whether the conviction was sized deliberately and stress-tested, which is exactly what a conviction tag is for. See our piece on tagging conviction on every trade for how to separate the two.

7. Herding / FOMO. Definition: buying because others are buying and you fear missing the move, rather than on independent analysis. Trace: entries cluster after large multi-day runs on elevated volume, often near local highs. Self-test: tag each buy with the asset's trailing five-day return and relative volume at entry; a concentration of buys after big up-runs on heavy volume is the herding footprint. Counterpoint: a deliberate breakout trader also buys strength on volume — with a defined invalidation level and position size. FOMO is the same entry with no plan for being wrong.

A one-hour spreadsheet audit of your own log

You can run a first pass on all seven in about an hour. Export at least twelve months of round trips — a longer window matters, because a handful of trades will not distinguish a tendency from noise. Build one row per closed round trip with entry date, exit date, entry price, exit price, quantity, and realised P&L. From that single table: derive holding-days (exit minus entry), a win/loss flag (P&L sign), exit-as-percent-of-entry, and position size as a share of portfolio value at entry. Those four derived columns feed every test above.

Then compute the seven summary numbers — disposition ratio, size-after-wins slope, add-below-cost share, breakeven-exit spike, recent-return tilt, one-way-trade share, and post-run buy share — and, crucially, write next to each one whether the pattern was rule-based or reactive. That single annotation is what separates a bias from a strategy, and it is the reason self-report alone fails: the number tells you the behaviour happened; only the reason tells you whether it was a mistake. This is also where our companion post on what your trade data actually reveals goes deeper on turning raw exports into signal, and the winners you sold too early piece drills into the disposition case specifically.

One warning on interpretation. Every test above is correlational, and every counterpoint exists because the same footprint can be produced by a disciplined strategy or by an unexamined habit. The data cannot tell you which. It can only surface the trades worth a second look. The value is not a verdict — it is a shortlist of decisions to re-examine while you still remember why you made them.

What this doesn't tell you

These footprints are behavioural, not clinical — a high disposition ratio is evidence of a tendency, never a diagnosis, and a rational rule-based strategy can generate every single pattern in this article. The tests also need history: with fewer than a few dozen round trips you are reading noise. StackViewer is built for months-to-years investors, not day traders, and it is not a real-time tool — brokers sync three times a day and values refresh roughly every 30 minutes via read-only OAuth (SnapTrade, 31 brokers), so intraday bias-hunting is out of scope. Custom or unlisted assets your broker's API does not return are excluded, and coverage gaps exist for some EU brokers, so your audit is only ever as complete as what the API exposes. Finally, none of this predicts returns or recommends any action — it analyses decisions you have already made.

Frequently asked

Can you actually measure cognitive biases from trade data?

You can measure the behavioural footprints biases leave, not the mental state itself. Realised-gain versus realised-loss ratios, average holding time on winners against losers, position sizing after a run of wins, and how often you re-enter a stock you just sold all leave numbers in an exported trade history. Those numbers are evidence of a tendency; they are not proof of intent, and a rational strategy can produce the same pattern. Treat them as prompts to review a decision, not verdicts.

Which cognitive bias costs investors the most money?

For most long-term investors the disposition effect — selling winners too early and holding losers too long — is the most expensive and the easiest to see in data, because it is the sum of hundreds of small realised-gain-versus-realised-loss decisions. But cost depends entirely on your strategy: a disciplined trailing-stop trader may show a disposition pattern that is intentional and correct. Measure your own realised gain/loss ratio and holding times before assuming any single bias is your biggest leak.

This article is for information only. It is not financial, investment or tax advice, and nothing here is a recommendation to buy or sell any security. StackViewer analyses decisions you have already made; it does not predict returns. Past performance does not indicate future results.