It lived once.
The black line is the sequence we happened to observe. Its worst fall was —.
A SMALL EXPERIMENT ABOUT A BIG NUMBER
Would you
have quit?
Before we talk about Monte Carlo, probabilities, or robustness, I want you to make one decision with your own imaginary money.
MAKE THE DECISION↓
Below is the historical trade list behind a strategy published last year. It spans roughly sixteen years and 185 closed trades. Costs are included. The ending looks reassuring.
But you never get to trade an ending. You have to survive the middle.
BEFORE YOU KNOW
HISTORICAL TRADE LIST · COSTS INCLUDED
Draw your line in the sand.
Imagine watching this strategy in real time. How far below a previous high could it fall before you switched it off?
$3k$20k
YOUR RULE—
Keep it in mind. We are about to give this exact strategy other lives.
THE FIRST TRICK
The equity curve is not one object.
It is a deck of 185 trade results, read from left to right. Shuffle the deck and the final profit cannot move by a cent. But everything you must endure on the way can change.
THE LIFE THAT HAPPENED
There is a deck under the line.
Blue cards made money. Red cards lost money. Nothing mystical: the curve is their running sum.
Same destination. Different stomachache.
Every pale line contains the same winners and losers exactly once. Their endings agree. Their drawdowns do not.
Now the path becomes a population.
Each path that turns red breached your drawdown rule—a version of this strategy you said you could not keep trading.
Your number changes what “tradable” means.
Lock in a number above to personalize this result.
OBSERVED DRAWDOWN—
The observed path was unusually gentle under this particular shuffle model.
TYPICAL SHUFFLED DRAWDOWN—
DIRECTOR’S CUT
Let’s watch one difficult life unfold.
A cloud of lines is evidence. A single path is experience. This one contains the same 185 trades as the published curve—only the order changes.
READY
Press play. Watch what happens after the early confidence.
TRADE 0 / 185
?
PAUSE HERE
What did we actually learn?
Not that the market will rearrange these trades like cards. Not that every ordering was equally plausible. We learned something narrower—and useful:
Conditional on treating the trades as exchangeable, the path we observed was unusually easy to live with.
That condition matters. Markets have regimes. Losses can cluster. Signals can overlap. A shuffle destroys time.
THREE MACHINES, THREE QUESTIONS
“Run Monte Carlo” is like saying “use a map.”
Which map? For what journey? Each machine preserves a different part of history and quietly destroys another.
NOW FOR THE TRAP
Meet a strategy that Monte Carlo could not rescue.
It was selected as the best of twenty filter variants. In its chosen window, 83 trades produced a handsome curve. Trade shuffling looked ordinary. IID resampling usually ended in profit.
SPECIMEN B · SELECTED WINNER
SHUFFLE: NO FLAGBOOTSTRAP: MOSTLY PROFITABLE
Thirty-one unseen trades are waiting behind this line.
THE NEXT 31 OUT-OF-SAMPLE TRADES—
The previous curve did not merely have a rougher journey. The selected relationship did not survive this next unseen period.
THE FALSE COMFORT
The bootstrap called it bad luck.
Asked to imagine the next 31 trades by drawing repeatedly from the winning sample, the model placed the actual loss near the bottom of its distribution.
THE MODEL SAID—
of its imagined outcomes were this bad or worse.
The calculation was not broken. The question was too trusting. Every resample began by assuming the selected trades were genuine representatives of the future.
TEST THE INGREDIENT, NOT THE SOUVENIR
What if the signal never knew anything?
Scramble the signal across dates, rerun the strategy on the same market, and build a world in which that ingredient has lost its timing relationship with future price.
PRETENDING WE TRIED ONE IDEAp = 0.047
Only about five scrambled runs in one hundred beat the selected result. That looks persuasive.
Once the simulated null is allowed to conduct the same twenty-variant search, impressive winners appear routinely from scrambled ingredients.
The uncertainty did not begin with trade one. It began when the research contest chose which equity curve you were allowed to see.
A POCKET GUIDE
Start with the uncertainty—not the button.
Choose the question you are actually trying to answer.
SELECT A QUESTION
The method should follow the uncertainty.
A Monte Carlo result is a conditional sentence. The generator supplies the condition.
BACK TO YOUR NUMBER
You chose —.
Now you know what that rule would have meant across the strategy’s shuffled lives.
THE SENTENCE WORTH KEEPING
Monte Carlo does not show you the future. It shows you the consequences of the world you told it to imagine.
The hard part is not running more simulations.
It is earning the right to trust the ingredients.
Data, definitions, and receipts
Specimen A
A published historical trade list with 185 closed trades. We subtract $70 round-turn costs. Trade-order paths, IID bootstrap paths, and circular ten-trade block paths use 2,000 deterministic simulations.
Specimen B
An 83-trade strategy selected from twenty filter variants, followed by 31 subsequent out-of-sample trades. “Out-of-sample” does not mean live capital.
Drawdown
Peak-to-trough decline in cumulative one-contract P&L, including zero starting equity as an eligible peak. No compounding or position sizing is applied.
Permutation tests
Signal values are shuffled across dates and the strategy is rerun. The full-search null repeats the selection procedure across the tested variants. Results are conditional on the exact scrambling and search definitions.
What this does not prove
No simulation establishes future profitability, live execution quality, causal validity, or immunity to regime change. The article is an explanation of research questions, not investment advice.
Rebuild
The self-contained browser bundle is generated by build_data.py from the archived trade lists and permutation-test distributions.