A winning sequence can tempt a trader to change position size or skip checks. If those changes exceed the plan, a few later losses may erase earlier gains. This article examines overconfidence as one possible explanation while keeping observable rule deviations separate from a psychological diagnosis.

What overconfidence really is in trading

Moore and Healy distinguish three meanings: overestimating personal performance, placing oneself too high relative to others, and being excessively certain about the precision of a belief. Ask which claim needs checking. A confident mood, one profitable trade or a larger lot does not alone identify any of these patterns.

Barber and Odean studied common-stock investments of 66,465 households at a US discount broker using 1991–1996 data. More active groups had lower net returns in that sample, with trading costs important. The authors interpreted the findings as consistent with overconfidence models; this was not an experiment establishing that every extra trade harms every investor, or a measurement of present-day retail Forex performance.

Check the claim behind increased confidence

Separate an observable decision from its possible explanation. Skipping a required order check or exceeding a documented risk allowance can be recorded directly. Calling it overconfidence requires additional evidence about the belief behind it; the same action might involve a calculation error, a changed plan or another cause. Write the reason before placing the order rather than infer it solely from a later loss.

A profit can occur after an off-plan entry, and a loss after a correctly executed rule. That is why process and outcome need separate records. There is no established threshold of three rewards after which everyone becomes overconfident. Following a rule also does not prove that the strategy has positive expectancy.

A narrow forecast needs a coverage check

“I expect close to100” is more testable if a range is recorded beforehand. The same centre need not mean the same error tolerance.

Scroll sideways to see the full diagram

Two predefined ranges, four outcomesOn one axis,A99–101 contains only100. B97–103 contains98,100,102,not104.WIDTH AND COVERAGE ARE DIFFERENTLater values:98 ·100 ·102 ·104A:recorded range99–10198100102104B:recorded range97–10398100102104A:width2;contains1 of4.B:width6;contains3 of4.A wider range more easily covers outcomes.

Fictitious four-value series without monetary units. Ranges are recorded before outcomes and unchanged. No probabilities are assigned:these are neither confidence intervals nor a personality test. Moore and Healy — The Trouble With Overconfidence distinguishes excessive precision from overestimation and comparison with others.

Four fictitious values are checked against both predefined ranges. Endpoints are included.

Keep boundaries and outside outcomes in the journal. Higher coverage alone does not establish a better forecast:widening can reduce usefulness. Four cases cannot establish lasting calibration or a particular person's overconfidence.

Review changes in risk and procedure

Compare planned risk rather than lots alone. In a hypothetical linear instrument, one lot changes P/L by USD 10 per pip. A USD 100 price-risk allowance with a 20-pip stop gives 0.5 lot before costs; a 40-pip stop gives 0.25 lot. Keeping 0.5 lot in the second case doubles planned price risk to USD 200. These assumed specifications must be checked for a real instrument, and an ordinary stop does not guarantee that loss cap.

A change in size can follow a predefined rule using current balance or equity, or a changed stop distance or pip value. Record the rule and total exposure, including open positions and related risks. Neither professional status nor a winning week establishes how someone will behave next. A short sequence does not justify ignoring limits or identify a universal sample size that proves an edge.

A hypothetical payoff series: wins can still sum to a loss

The following invented series shows why wins need amounts as well as counts. Assume eight outcomes of +USD 100 and six of −USD 200, each before additional costs. No actual trader, week or psychological cause is inferred, and the listed amounts are outcomes rather than a recommended risk allowance.

An explicit hypothetical payoff series
Wins8 × USD 100 = +USD 800.
Losses6 × USD 200 = −USD 1,200.
Total14 outcomes give −USD 400 before additional costs, despite 8/14 ≈ 57.1% wins.

The total is USD 800 − USD 1,200 = −USD 400. A majority of wins therefore need not give a profit. The table does not reveal whether different position sizes, stop distances or another factor caused the losses. Overconfidence and attempts to recover a loss can overlap, but neither has one established speed or guaranteed path to account failure.

A good decision versus a good outcome

Check the precision of a recorded forecast too. In an invented ten-event, two-outcome exercise, six events occur. Forecast A always declares 90% and B always 60%, both assumed before outcomes. Define a comparison score as the mean squared error (p − y)², with y = 1 when the event occurs and 0 otherwise. A gives [6 × 0.1² + 4 × 0.9²]/10 = 0.33; B gives [6 × 0.4² + 4 × 0.6²]/10 = 0.24. Lower is better under this stated score.

The forecasts have the same predicted direction, but A’s greater confidence incurs more error on the four non-events. Do not retrofit B as an earlier forecast after seeing results. Ten cases neither identify a personality trait nor establish a future probability of 60%; they show what a hit/miss record leaves out.

In the journal, preserve the entry criteria, size calculation, changes to stops, fills and costs, then assess compliance separately from net result. An unchanged standard lot is not automatically correct when inputs change. A few weeks of notes do not automatically separate luck from skill; review the strategy and the execution as distinct questions.

What to do tonight

Turn the review into observable checks. The following steps are examples of process design, not clinically validated treatment or a guarantee of profitability.

  • Recalculate the allowance. Record the capital basis, stop distance, pip value, costs and existing exposure. A streak does not replace the calculation; no single safe percentage applies to every account.
  • Compare process and outcome. Check the required entry criteria and documented reasons for any change. Retain an off-plan profit as an off-plan decision rather than excuse it retrospectively.
  • Use a defined pause condition. If checks are being skipped or risk raised impulsively, stop opening new positions while arranging management of existing ones. Five wins do not establish a universal one-day reset.
  • Record the opposite scenario. Write what would contradict the idea and how orders would be handled. This is an analytical exercise, not a requirement to predict failure.
  • Track calibration with uncertainty. Sixty wins in one hundred past trades are a sample frequency. Record estimates prospectively and retain misses; that frequency does not establish the next trade’s exact probability.

Under an illustrative model of independent trades each with a constant sixty-percent win probability, five specified consecutive wins have probability 0.6^5 = 0.07776, about one in 12.86 such blocks. Overlapping runs and changing probabilities behave differently. The sequence alone does not prove skill or justify increasing risk.

To examine related questions, recency bias discusses weighting recent observations; the trader’s ego considers how decisions are explained. Survivorship bias addresses missing unsuccessful cases, and a trading journal keeps reasons alongside outcomes. Risk management develops separate exposure and loss calculations.