Verliesreeksen en ruïnerisico: wat de kansen echt zeggen

This article explains losing-streak probabilities and risk limits without claiming a win rate for TIC or for any strategy. The worked case assumes 100 independent trades, each with a hypothetical 40% probability of winning and 60% probability of losing. That is a teaching model, not a forecast.

How likely is a losing streak?

For one specific block of eight trades, independence makes the probability of eight losses:

But a 100-trade sequence contains many overlapping places where a run can begin. You cannot multiply 1.68% by 93 because those windows overlap and are not independent. Exact dynamic programming tracks the current run length after every trade and adds the probability mass that reaches eight losses.

Under the stated assumptions, the probability of at least one run of eight losses somewhere within 100 trades is approximately 48.97%. The exact result says something important: a run that looks extraordinary in one chosen eight-trade block can be fairly common when there are many opportunities for it to appear.

This is an at least once probability, not the chance that the next eight trades will all lose and not the expected number of eight-loss runs. Those are different questions with different answers. If you change the sequence from 100 trades to 20, or change the assumed loss probability, the result changes. A trader should therefore state the horizon and assumptions beside any quoted streak statistic; a percentage without them can mislead even when the arithmetic is correct.

The OpenStax geometric-distribution chapter explains the Bernoulli assumptions behind repeated independent trials. Markets often violate those assumptions because volatility, execution and strategy behaviour can cluster.

Does a 48.97% probability mean the strategy wins 40%?

No. We selected 40% only to demonstrate the mathematics. It is not TIC's win rate, not an estimate for your strategy and not evidence that future trades are independent.

A win rate also says nothing about payoff size. A system winning 40% with average wins twice average losses can have positive expectancy before costs; a system winning 70% can still lose if occasional losses are huge. Read sample-size limits in track records before treating a short history as stable.

How does risk per trade change drawdown?

Assume eight consecutive full-risk losses and no compounding within each row for a simple first comparison:

The compounded column uses (1 − risk)⁸. At 5% risk, the account has lost about one-third, not merely endured an uncomfortable week. Recovering from a 33.66% drawdown requires about 50.74% on the remaining capital. This asymmetry is why drawdown control matters; our drawdown guide shows the recovery arithmetic.

The table assumes that every loss is exactly the planned percentage of the balance immediately before that trade and that no deposits, withdrawals or gains interrupt the sequence. Real fills can breach a planned stop during a gap, while partial exits can reduce a loss. Fixed currency risk and fixed percentage risk also compound differently. Use the table to see sensitivity to position size, then stress-test your own rules with costs and execution uncertainty instead of treating its rows as promised outcomes.

For example, define ruin as falling below a pre-set balance at which the strategy must stop. Estimating the chance of crossing that boundary requires a model for both wins and losses, including their sizes, not just how often each occurs. If losses are larger during volatile periods, the constant-loss model understates that risk. If several positions respond to the same news, assuming independence also understates the chance of clustered damage. These are reasons to test adverse scenarios, not reasons to claim that a single precise ruin percentage applies to everyone.

What risk limits are more useful than a streak prediction?

Set limits from what you can survive, not from the average streak you hope to see.

1. Define failure before trading. Choose a maximum strategy drawdown or evidence threshold that triggers review, not impulsive doubling.
2. Cap risk per trade. Use a fraction small enough that a plausible streak leaves room to continue and investigate.
3. Cap correlated exposure. EURUSD, GBPUSD and gold can all express a similar dollar view; three positions are not always three independent risks.
4. Separate process from outcome. A valid loss can follow the plan; an invalid win can violate it.
5. Include costs and gaps. Planned loss is not guaranteed loss, especially around news and illiquid periods.

A useful stress test asks: “If the next eight trades lose at the planned stop, what percentage remains, and would I still follow the process?” If the answer is no, the position is too large regardless of how attractive the recent backtest looks.

When does a streak become evidence of a problem?

A streak becomes more informative when it arrives with structural evidence: fills are materially worse than the test, the market regime is outside the design, rules changed, signals are correlated, or the observed distribution has drifted beyond a pre-defined tolerance. Do not move that threshold after losses arrive.

Compare backtests with live execution, as explained in backtest versus live results, and evaluate payoff and downside-sensitive measures rather than win rate alone using Profit Factor, Sortino and Calmar.

Sources and calculation note

• OpenStax, Geometric Distribution
• Exact 48.97% figure: dynamic programming over 100 Bernoulli trials, state = current consecutive-loss count 0–7; probability reaching state 8 is absorbed. Assumptions: independent trials, constant loss probability 0.60, at least one run of eight.