A precise number. An uncertain assumption.
A model might calculate an exact probability, but that result is only as reliable as its assumptions. Learn how to read conditional probabilities without turning them into promises.
Updated 14 Nov 2026 · Original educational material · Hypothetical examples, not live market data
What does a 60% probability say?
Suppose a hypothetical trading rule wins with a probability of 60% on each trade. That does not mean six of the next ten trades must win. It means each trial is assigned a 0.60 win chance under the chosen model.
A particular sequence could contain three wins, seven wins or even none. The model says some counts are more likely than others, but it does not identify the winners. Entering 60% into a calculator does not make it a measured future probability for an actual strategy.
For the mathematics below, assume each trade is an independent and identically distributed Bernoulli trial: a win or a loss, with fixed p = 0.60. Real market outcomes often violate these simplifying assumptions.
Ten trades produce a range of possible outcomes.
Under our fixed independent-trial assumptions, the number of winners among ten trades has a binomial distribution. The chance of exactly six winning trades is approximately 25.08%. Six wins is the most likely single count here, yet appears in only about one quarter of ten-trade sequences.
This is a probability model, not real trading history, a measured win rate or an estimate of how Bitcoin will perform.
Swipe sideways to see 0 through 10 winning trades →
Six winners is not guaranteed.
This answer can be mathematically exact given its assumptions, yet a poor real-world forecast if the true success probability differs or outcomes are dependent. It also says nothing about profitability: six small wins may not cover four large losses.
Three kinds of uncertainty.
Outcome uncertainty: future sequences can vary even if the probability model is perfectly specified. The bar chart illustrates that variation.
Parameter uncertainty: we may not know the true win rate, payoff distribution or loss variability. Six wins among ten observed trades gives an estimated historical fraction of 60%, but such a small sample does not establish a stable future 60% chance. Reliable statistical intervals need a defensible sampling procedure.
Model uncertainty: the structure itself can be wrong. Trades may share exposures, success probabilities can shift between regimes, and a two-outcome model may omit unusually large losses or execution failures. Precise calculations cannot correct such omissions.
Independence is not automatically realistic.
Market conditions can persist. Several trades may fail during the same liquidity shock; their outcomes, payoff sizes or probabilities may be connected. A model of identical independent trades may describe losing streaks and tail outcomes poorly in such conditions.
Risk of Ruin computes a conditional first-passage probability under its stated assumptions. Monte Carlo produces simulated paths from a user-entered distribution, not a validation of that distribution. Scenario & Stress Lab shows sensitivity to chosen parameters, not empirically measured confidence bands.
Probability, confidence and certainty are different.
A probability describes an event under a defined model. A frequentist confidence interval is generated by a procedure designed to cover a fixed unknown parameter at a stated long-run frequency under its assumptions. It is not generally a direct posterior probability distribution for the parameter after a single sample.
A p-value is not the probability that a tested hypothesis is true. It quantifies how extreme data are relative to a specified null hypothesis and procedure. Repeated searching and selection can distort its interpretation, which is one reason the backtesting guide discusses multiple testing.
Claims of “95% confidence” or “exact probability” should identify the target event, method, data source, time horizon and assumptions. Without that context, a precise-looking number can mislead.
How our existing models should be read.
| Tool | Conditional result | Not verified |
|---|---|---|
| Risk of Ruin | Chance of ever touching a specified starting-capital floor within a finite trade horizon | Actual future win rate or exchange execution |
| Monte Carlo | Distribution of simulated paths from chosen assumptions and random seed | Future market prices or the presence of a real edge |
| Scenario & Stress Lab | Changes in modeled results when assumed parameters vary | Observed market-data confidence limits |
Transparent mathematics remains useful. The key is to separate calculations conditional on assumptions from empirical evidence that those assumptions hold.
Three common mistakes.
“60% means six wins out of ten.”
Individual samples vary, even when the underlying chance is stable.
“Exact output means accurate forecast.”
Correct arithmetic cannot fix an incorrect probability input or missing tail event.
“Past win rate equals future probability.”
Small samples, multiple testing, dependent trades and regime changes make estimates uncertain.
Frequently asked questions.
Does a 60% win rate guarantee profitability?
No. Payoff sizes, commissions, slippage, position risk and unusual losses all matter.
Why is the most likely outcome only about 25%?
Probability is spread across eleven possible win counts. Exactly six is the tallest single bar but not the only result.
Can ten past trades establish the true probability?
They provide an observed fraction, but a very uncertain estimate of an unknown future win chance. The sample may also be unrepresentative.
What if losing trades tend to cluster?
Then the independent Bernoulli model may be inadequate. A suitable dependence and regime model would require additional evidence.
Does Monte Carlo predict the next Bitcoin candle?
No. Market History Lab currently samples hypothetical trade outcomes from input assumptions; it does not forecast live asset prices.