Is Your Equity Curve Skill or Luck? The Reshuffle Test

No. A smooth equity curve is proof of one favorable ordering of your trades — nothing more. Skill lives in the set of trades your strategy took: their expectancy, their consistency, whether the edge survives out of sample. The smoothness of the curve, by contrast, depends heavily on the order those trades happened to arrive in, and order is close to accidental. Two strategies with identical trades can show a serene staircase or a stomach-churning sawtooth purely from sequencing.

The reshuffle test separates the two. It keeps every trade exactly as your backtest produced it — same wins, same losses, same sizes — and only varies the order, thousands of times. Because the multiset of returns never changes, the final equity is identical in every shuffle. What changes is the path: the drawdowns, the longest losing run, the depth of the worst trough. If your realized curve sat at a lucky percentile of that distribution, then part of what looked like skill was actually a kind ordering.

This is a specific, narrow use of [Monte Carlo simulation](/learn/monte-carlo-simulation-backtests): shuffle-without-replacement, which isolates path-dependence and nothing else. It tells you how much of your curve's calm was structural versus how much was luck of the draw. It says nothing about whether the underlying edge is real or profitable — that is a separate question addressed by out-of-sample testing and significance corrections. Educational only; trading involves substantial risk of loss.

Because drawdown is path-dependent and final return is not. Final equity is the product of all your per-trade multipliers, and multiplication commutes — the order does not change the product. Drawdown, though, is the deepest trough relative to the running peak, and that depends entirely on when the losers land. Cluster your losing trades together and you carve a deep hole. Space them out among winners and the curve barely dips. Same trades, same endpoint, completely different worst case.

A concrete illustration: imagine ten trades, seven winners and three losers. If the three losers happen to fall consecutively early — before the winners have built a cushion — the account visits a much lower low than if a loser, two winners, another loser, more winners, and the last loser are interleaved. Nothing about the strategy changed. Only the sequence did, and the sequence is not something you control or can count on repeating.

import numpy as np

def max_drawdown(returns):
    eq = np.cumprod(1.0 + np.asarray(returns, float))
    return float((eq / np.maximum.accumulate(eq) - 1.0).min())

def reshuffle_drawdowns(returns, n=20000):
    r = np.asarray(returns, float)
    return np.array([max_drawdown(r[np.random.permutation(len(r))])
                     for _ in range(n)])

Each shuffled path is built forward from the trades themselves — no future value touches any earlier step, so there is no [look-ahead bias](/learn/look-ahead-bias) introduced by the test. The spread of reshuffle_drawdowns output is the honest picture of how much sequencing luck was baked into the single curve you were admiring.

That is the number the reshuffle test exists to produce. Once you have the distribution of maximum drawdowns across thousands of shuffles, you locate your actual realized drawdown within it. If your historical ordering produced a drawdown milder than 80% or 90% of the shuffles, your real curve was near the lucky end — most alternative orderings of the very same trades would have hurt more. The calm you saw was substantially a gift of sequence.

If, instead, your realized drawdown sits near the median of the shuffle distribution, then the curve you saw is roughly typical for these trades, and its smoothness reflects the trade set rather than fortunate timing. That is a more trustworthy result — not because the edge is validated, but because you are not fooling yourself about the path. And if your realized drawdown sits at the ugly end, you actually had unlucky sequencing and the trade set is calmer than it looked.

| Where your realized drawdown sits | Reading | | --- | --- | | Milder than ~85% of shuffles | Lucky ordering — the smooth curve was partly luck | | Near the median | Typical path; smoothness reflects the trades, not timing | | Deeper than most shuffles | You had unlucky sequencing this run |

The practical consequence is the same regardless: plan around the distribution, not the single path. If a lucky ordering flattered you, live trading will likely re-draw from the full distribution — including the deeper draws you never saw. This is a statement about path risk, not about expected profit.

It does not prove your strategy has an edge. Reshuffling holds the trade set constant, so if that set is the product of overfitting, every shuffle inherits the same illusion. A curve-fit strategy can pass the reshuffle test — sit at a perfectly ordinary percentile — while having no real predictive power at all. The test speaks only to path-dependence, not to whether the trades themselves mean anything. Validate the edge separately with the [Overfitting Check](/backtest/overfitting) and the concepts in [overfitting and curve-fitting explained](/learn/overfitting-curve-fitting-explained).

It also assumes trades are exchangeable — that shuffling their order is a fair thing to do. If your strategy has genuine serial structure (positions that pyramid, losses that cluster in volatile regimes, trades whose outcomes correlate), plain shuffling destroys that structure and can paint an optimistically smooth distribution. In that case the shuffle understates real path risk, and you need [walk-forward analysis](/learn/walk-forward-analysis) to see how the strategy behaves across changing regimes rather than across arbitrary reorderings.

And because it keeps your exact trades, the reshuffle test says nothing about sampling uncertainty — whether you would even get these trades again. For that you want a bootstrap (resampling with replacement) and an honest count of how many trades you have; see [how many trades is enough](/learn/how-many-trades-is-enough). The reshuffle test is one narrow, useful lens: it isolates sequencing luck. It is not a verdict on profitability, and this article is educational only, not financial advice.

Read the reshuffle result as a humility check on the curve, then act on the distribution rather than the single path. If your realized ordering was lucky, do not size or plan against the smooth line you saw — size against a pessimistic percentile of the shuffled drawdowns, because that is closer to what future sequencing can hand you. A lucky backtest ordering is not a promise the market renews.

Sequence the work sensibly. First establish that the trade set is real: enough independent observations, confirmation on data the strategy never trained on, no leakage. Our notes on [in-sample versus out-of-sample](/learn/in-sample-vs-out-of-sample) and [why most backtests fail](/learn/why-most-backtests-fail) cover that groundwork. Only once the edge survives those checks does the reshuffle test add value — at which point it tells you how much of your comfortable-looking curve to actually trust.

You can run the reshuffle on your own results with the free [Backtest Health Check](/backtest): it shuffles your trade list, plots the drawdown distribution, and marks where your realized ordering landed. Pair it with the significance literature — Bailey and Lopez de Prado's [Deflated Sharpe Ratio](/learn/deflated-sharpe-ratio), and White's Reality Check and Hansen's SPA test for comparing many strategies without fooling yourself. The reshuffle test answers one question honestly: was the calm skill or sequence? It will never tell you the strategy is profitable. Trading carries a substantial risk of loss.