Streaks Are the Design

The arithmetic of a run

A losing streak is not evidence about a method; it is a prediction a win rate makes. If a system wins 45% of the time, a loss occurs 55% of the time, and the probability of six losses in a row is that loss rate raised to the sixth power — 2.77%. The figure sounds small and is anything but: over five hundred trades the expected number of such runs is around fourteen, and the probability of meeting at least one inside a hundred trades is above ninety per cent. A streak is not a malfunction at this win rate; it is what a hundred trades look like. What varies across traders with the same edge is what the run costs, and the arithmetic is a single expression: the capital left after a run is one minus the risk fraction, raised to the length of the run. At 1% risk six losses leave about 94% of capital, which is a bad fortnight. At 5% they leave 73.5%, and the recovery requires a 36% gain, which is a project rather than a fortnight. At 10% they leave barely half, which is the point at which the decision is no longer about trading but about whether to continue. Nothing about the edge changed across those three rows. This is why the calculation belongs in the plan rather than in the reaction. The question the plan will be asked in the middle of a run is whether six losses can be tolerated, and the honest answer at that moment is distorted by everything the run has just done to the trader. Computed in advance, while the run is hypothetical, it becomes a design question with a clear answer: the size is chosen so that the streak the win rate predicts is survivable, and then the streak is recognised as a prediction rather than a failure. That recognition is the whole psychological content of the lesson — the trader who has done the arithmetic recognises the prediction, and the trader who has not experiences it as proof. A 45% win rate, one losing run — Chance of six losses in a row: 0.55⁶ ≈ 2.77% · Expected such runs in 500 trades: About 13.8 ← · Chance of at least one run in 100 trades: Above 90% ← · Capital left after six losses at 1% risk: 94.15% · At 5% risk: 73.51% — and 36% needed to recover · At 10% risk: 53.14% The same table can be read as a sizing rule: pick the risk fraction at which the streak your win rate predicts is an inconvenience rather than a decision about whether to continue, and then the streak costs nothing but time.

Why the streak feels like evidence

The reason a losing run is so persuasive is that it arrives with an explanation attached, and the explanation is usually about the trader. Attention after a loss is narrower, the last few trades are more available than the previous hundred, and the run has a shape — it feels like a deterioration rather than a sequence. None of that is information about the method. The method has a win rate, the win rate predicts runs of this length at this frequency, and the forecast was available before the first loss of the run. There is a useful asymmetry to hold on to instead. A streak that is within the predicted frequency is evidence that the system is behaving as designed, and the correct response is to keep taking the trades. A run that is materially outside the prediction — more losses than the recorded win rate can plausibly produce, or losses taken outside the rules — is genuinely evidence, and it points at an execution problem rather than at a market change. Distinguishing the two requires the record from the earlier lessons: a journal that tags planned and impulse trades is what tells a trader whether a run was the plan or a deviation. The second-order defence is to decide in advance what a run will trigger. A rule that halves the size after a stated drawdown, or that pauses after a run longer than the prediction allows, converts the moment of maximum emotional pressure into the execution of a decision made calmly. Without such a rule, every run ends with a judgement call made by the person the run just happened to, which is the weakest available arrangement for a decision about risk. • The run is predicted by the win rate; the frequency is checkable in advance. • Within prediction, keep taking the trades — the streak is the schedule. • Outside prediction, look at execution before looking at the market. • Decide in advance what a run triggers: a size cut or a pause, with a stated trigger. The most expensive version of this is a size increase after a run, on the reasoning that the system is “due”. A run says nothing about the next trade, and raising the fraction exactly when the account is at its smallest compounds the drawdown the streak produced.

Is it a slump, or is it noise?

A losing run can mean three very different things, and they call for opposite responses. It can be the expected tail of an unchanged process, in which case nothing has changed and the correct action is to keep executing. It can be a regime change — the strategy’s edge has been competed away, or the market it was built for has stopped behaving that way — in which case continuing to execute is the expensive choice. Or it can be a deterioration in the operator: worse sleep, a bigger position, less attention. The run itself cannot distinguish them, which is why the emotional response of “something is wrong” is a hypothesis rather than a conclusion. The way to separate them is to test the *inputs* rather than the outcome. Has the win rate on matched trades moved beyond what the sample could explain by chance? Has the average win and average loss changed? Are the trades still coming from the setup the system describes, or has the definition quietly drifted toward whatever is available? A run of losses that arrives with the same average win, the same average loss and the same setup frequency is noise by construction: the process has not changed, and stopping is a reaction to randomness. A run that arrives with the losing trades getting larger, or the setups becoming vaguer, is information. This is also the point where the two classic fallacies meet, and they look like opposites. The **gambler’s fallacy** says a loss makes a win more likely — that the process owes you something. The **hot-hand belief** says a win makes the next one more likely, that form is real. Both are statements about a sequence containing memory; for an independent process both are wrong, and the arithmetic in this lesson is exactly the arithmetic that shows why a six-loss run is unremarkable at a forty-five percent win rate. The correct stance in the middle of a run is that the sequence itself is not evidence, while the inputs might be — and the only way to tell them apart is a rule written before the run started. What the same six-loss run can mean — Same win rate, same payoffs, same setups: Noise — the process is unchanged and the run is the tail · Same setups, but average loss rising: Information — sizing or exits have drifted, and the drawdown is deeper than designed · Fewer setups, vaguer criteria: Regime change — the edge has degraded or the market has moved ← Write the rule before the run: the conditions under which you would stop trading a strategy, and the number of trades you would need to see before deciding. A rule written mid-drawdown is a decision made by the person the rule was supposed to protect you from.

Two opposite mistakes from the same streak

A run of losses produces two contradictory errors, and the same person can commit both inside a week. One is the belief that the process has broken and must be changed; the other is the belief that a win is now due. Both readings come from the same source: treating a random sequence as though it kept score. The first is the hot-hand reading in its pessimistic form. Six losses in a row must mean something is wrong — and sometimes it does, because a market regime has shifted, a data feed has changed, or a rule is quietly not being followed. But the diagnosis has to come from the mechanism, not from the run. The check is short and specific: has anything about the inputs changed? Is the strategy trading the setups it was specified to trade? Have the fills changed? A run of losses is not evidence on its own, because runs of this length are ordinary and would have looked identical in the backtest that justified the method. The second is the gambler’s fallacy: six losses make the next one more likely to win, as though the process owed something. In a genuinely independent sequence it does not, and the belief is most destructive where the expectancy is negative, because it converts a losing method into a method that escalates size exactly when it is least able to absorb the outcome. The two errors are mirrors, and which one appears tells you more about the holder’s recent state than about the market. Underneath both sits the **clustering illusion** — the well-established tendency to see streaks in sequences that are genuinely random, and to read them as structure. The classic demonstration asks subjects to judge which of two sequences was produced randomly, and they consistently choose the one with more alternation and fewer runs, because real randomness contains longer runs than intuition allows for. This is why the arithmetic matters more here than in most places: six losses at a workable win rate is not a rare event, and computing its frequency once changes what a streak means the next time one arrives. The observation stops being a signal and becomes a prediction that came true. There is exactly one situation in which the run is genuine evidence, and it is not about the run. If a mechanism has changed — a rule violated, a feed altered, a market’s microstructure shifted, an edge competed away — then the losses are a symptom of a real change, and the right response comes from identifying it. The distinction to hold is between asking what the streak says about the process and asking what has changed about the process. Only the second question has an answer, and only the second one justifies action. • A losing run may mean a broken mechanism or an ordinary sequence; the mechanism decides. • The gambler’s fallacy escalates size exactly where the method cannot absorb it. • People read randomness as clustered, because they underestimate how long real runs get. • Pre-compute the run distribution once, and the next streak becomes a prediction rather than a signal. Log the mechanism checks, not the streak. A record of “inputs unchanged, rules followed, setup as specified” is what makes it possible to sit through a run that the arithmetic says should arrive roughly once every fifty trades.

Sources

• Runs, streaks and the arithmetic of losing sequencesStandard probability; risk-of-ruin literature
• Drawdown recovery and the cost of a percentage lossFixed-fractional sizing research
• Streak expectations and psychological responseSteenbarger, “The Daily Trading Coach”