What Is Revenge Trading? What the Evidence on Trading After a Loss Actually Shows

Revenge trading is the habit of putting on more size, or more trades, immediately after a loss in order to win that loss back before the session ends. The phrase is trading-floor slang, but the behaviour it names has been measured on futures data: in a study of the Treasury bond pit at the Chicago Board of Trade, a local who was down by late morning ran more risk than his own average over the rest of the session about 16% more often than a local who was up.

That study is the spine of this post. It shows where the extra risk came from, what the losing traders had to pay to get it, and the one place the effect did not show up at all.

What is revenge trading?

Revenge trading means adding contracts, adding trades, or accepting an entry you would normally skip, straight after a losing trade, with the aim of recovering the loss inside the same session. It is a description of behaviour: the CFTC's glossary carries entries for day trader, local, scalper, pit and open trade equity, and no entry for revenge trading or for loss aversion.

The reason the day matters is structural. The CFTC's glossary defines a day trader as someone who takes positions and offsets them inside one session before the close. For a trader working that way, the loss and the attempt to recover it land in the same accounting period, and there is no overnight gap in which the loss becomes last week's problem. The loss is on the screen, and so is the time left to erase it.

That makes the trading day a natural unit for measuring the behaviour, which is exactly how the futures study below was built.

Is there evidence that futures traders take more risk after a loss?

Yes, from one detailed study of professional futures traders. Joshua Coval and Tyler Shumway, both then at the University of Michigan Business School, read the 1998 audit trail of the CBOT Treasury bond futures pit and cut each trading day in two at 11:00 a.m. Their working paper is dated 4 May 2001, and the figures here are that version's.

The headline result comes from a logit model evaluated at the sample means. After a profitable morning, the chance that a trader's afternoon risk came in above his own average was 26.9%; after a losing morning it was 31.3%. On the simpler measures, a losing morning raised the likelihood of an above-average afternoon trade count by 28.6%, and of above-average trade size by 9.5%. In regression form, a one-standard-deviation fall in morning profit was associated with between 11% and 23% more trades in the afternoon, at sizes 5% to 11% larger.

One honest qualification: in the continuous regressions the authors' composite "total dollar risk" measure moved in the same direction but reached significance in only one of the four specifications, so the trade-count and trade-size results are the stronger half of the evidence. Measured in binary form in the logit tables, that same composite came in significant in two of three.

Which traders did the CBOT futures study actually cover?

Floor market makers in one futures contract, in one year. The data was the exchange audit trail for all of 1998: 236 full trading days and more than five million transactions, obtained from the CFTC through a Freedom of Information Act filing. Of 1,082 traders in the records, Coval and Shumway kept 426 as locals: anyone with at least 1,500 own-account trades and at least 100 trading days in the year. A local, in CFTC terms, holds trading privileges at an exchange and deals for their own book, historically from the floor, and the CFTC credits that activity with supplying liquidity. That gave a panel of 82,595 trader-days.

The pit session ran from 7:20 a.m. to 2:00 p.m., so the 11:00 a.m. split fell a little after the midpoint of the trading day.

Two limits belong on any claim made from this paper. The effect was measured inside a single session, and the sample is professional market makers in one contract on one exchange in one year — not retail traders, not screen trading, not equity index futures.

Why does the size of the loss change how a trader responds?

Because the response was asymmetric around zero. Coval and Shumway sorted traders each day into twenty morning-profit groups and ran the regressions group by group, using the composite dollar-risk measure and plotting the averaged coefficients inside two-standard-error bands. The increase in afternoon risk got larger as morning losses got larger, and the relationship ran up to roughly the 30th percentile of the morning-profit ranking, which is where morning profit crossed zero. Above zero it flattened: traders who had made money all took similar, below-average afternoon risk, and the slight rise in the top few percentiles was not statistically significant.

A kink at zero is what the authors were looking for. They read the shape against Kahneman and Tversky's 1979 estimate of a utility function that bends at zero, with the convexity over losses roughly matching the concavity over gains.

The split-day design is what makes that shape informative. Two rival explanations — overconfidence read through self-attribution, and the house money effect — both predict that risk-taking rises as profits rise. Loss aversion predicts the opposite. Splitting the day at 11:00 a.m. put those predictions on opposite sides of the same test, and the data came down on the loss-aversion side.

Does revenge trading actually cost money?

In this sample it bought worse prices. The authors flagged a trade as price-setting when a local bought above the previous trade price or sold below it, and asked who was doing that. A trader with a losing morning was 10% to 20% more likely to be the price setter than an otherwise comparable trader with a morning gain, and a one-standard-deviation morning loss made him 2% to 5% more likely to place one than on an average afternoon. Those regressions held when the sample was narrowed to trades that expanded a position, so it is not simply losing traders unwinding. Against a baseline rate of losing money the authors put at 32.9%, losing traders accounted for 38% of all the afternoon price-setting trades these market makers placed.

What happened next is the part worth sitting with. In the ten minutes after one of those trades, the price reversed 16.4% more when the trader who moved it had lost money that morning. The authors' reading is that the rest of the pit treated those trades as noise and was happy to take the other side.

And yet the return numbers are mild. Mean afternoon profit, in the paper's standardised units, was 0.076 after a losing morning and 0.078 after a profitable one. The standard deviation around those means was 1.22 after a losing morning and 0.88 after a profitable one. For this group of professionals, the extra risk widened the spread of outcomes far more than it moved the average.

Does the effect carry over into the next trading day?

Not in this data. When Coval and Shumway asked whether one day's profit explained the next day's risk-taking, the relationship disappeared: no detectable effect in the logit specification, and none in the continuous specifications under panel, pooled OLS and Fama-MacBeth estimation alike, with or without outliers removed. Their conclusion for this set of traders is that the loss aversion was pronounced only at the daily horizon.

The volatility tests behave the same way, with a caveat about which measure worked. Coval and Shumway aggregated morning losses across the pit three ways, and the two obvious ones — the share of locals down at 11:00 a.m., and the average morning profit across traders — were not statistically significant at all. Only the third was, and it is the least obvious of them: each trader's own estimated loss-aversion coefficient, summed across the traders who finished the morning ahead. A one-standard-deviation fall in that measure went with 12.2% more afternoon volatility sampled second by second. Sampled over ten minutes, or across the whole afternoon, the result lost its statistical significance, and the whole-afternoon figure was 5.2%. The authors call the volatility results far from conclusive, because each trading day contributes a single observation, and say in their conclusion that most of their power to detect effects on prices sits at the microstructure frequency.

Is revenge trading the same as the disposition effect?

They are two different behaviours that both get triggered by a loss. Revenge trading is about the next position: size and frequency go up after a realised loss. The disposition effect, named by Shefrin and Statman in 1985, is about the current position: losers get held too long while winners get sold too soon.

Terrance Odean tested the second one in the Journal of Finance in October 1998, reading the 1987-to-1993 trading records of 10,000 accounts held at a large discount brokerage. Aggregated over the whole sample, the proportion of gains realised was 0.148 against a proportion of losses realised of 0.098. Averaged account by account the gap is wider: 0.57 against 0.36. Odean reports that rebalancing and the higher cost of trading low-priced stocks do not account for it, and that subsequent portfolio performance does not justify it.

The two can compound in the same session: an unrealised loss left open is still consuming room, and the next position is sized against whatever room is left.

What rule does revenge trading run into on a funded account?

On a funded account here — a simulated account with a real payout on the profits — it runs into a drawdown line that only ever moves up, and nothing in the rules ends the session early. None of the programs currently sold carries a daily loss limit, whether evaluation, funded account or instant funding, so no tripwire closes the platform after a bad morning.

The rule that governs the account is the end-of-day drawdown. The line sits a fixed distance below the highest end-of-day balance, rises only on a new highest closing balance, never falls, and is monitored live, so a touch at any point in the session liquidates the account. Because it is live, the unrealised half counts (the CFTC's term for that figure is open trade equity), and the rule reads the loss before the trade is closed.

There is also a cost on the recovery side. On programs that carry a consistency rule, winning the money back in one outsized session creates a profit-concentration problem of its own, so the revenge day can cost time on a payout even when it works.

What does doubling size after a loss do to the room left?

It halves it, measured in points, and it does so whatever the balance. Take the $50K account used in the worked example in that drawdown article: a $2,000 drawdown with the line at $48,000, and a ceiling of four minis under the contract limits. ES moves in 0.25-point ticks worth $12.50, per the instrument list, so one contract is $50 per index point and four are $200. The whole $2,000 of room is ten ES points at the ceiling.

Now lose $600 in the morning. There is $1,400 of room left. Held at two contracts, that is 14 ES points. Doubled to four, it is seven — at the exact moment the remaining room is already smaller than it was at the open. That is why position sizing on a funded account is recalculated from the current distance to the line rather than from the number on the account label.

FAQ

Is revenge trading the same as overtrading?

Overtrading is one of the forms revenge trading took in the data. In the CBOT study, a losing morning raised the likelihood of an above-average afternoon trade count by 28.6%, and a one-standard-deviation fall in morning profit was associated with between 11% and 23% more trades that afternoon. Trade size rose too, by 5% to 11%, so the extra risk came from both more trades and bigger ones.

Do professional traders revenge trade?

The traders in the Coval and Shumway study were professionals — pit locals with at least 1,500 own-account trades in 1998, whose living came from trading — and the effect showed up clearly among them. The authors treat a result measured on full-time proprietary traders as a lower bound on how much behavioural bias matters for market participants generally, on the grounds that these traders' livelihoods depended on trading well.

Does a daily loss limit stop revenge trading?

A daily loss limit is a rule at some firms that halts trading once a set amount is lost in one session. No program sold here has one, so there is no automatic stop to the afternoon after a losing morning. The binding rule is the end-of-day drawdown, and a touch on that line closes the account rather than pausing it.

Does trading after a loss hurt performance?

In the CBOT sample the effect on average return was small: standardised afternoon profit averaged 0.076 after a losing morning and 0.078 after a profitable one. What changed was dispersion, from a standard deviation of 0.88 to 1.22, and price quality — prices these traders moved reversed 16.4% more over the following ten minutes than prices moved by traders who had made money before the 11:00 a.m. split.