Monte Carlo is a broad label for repeated randomized experiments, not a promise that each simulation creates a fresh market path. Here, the product example repeatedly redraws a completed backtest’s trade outcomes. Trade resampling asks how that chosen record behaves under a specified rearrangement or redraw. It can expose sequence sensitivity in one backtest. It cannot create new market history, prove the trades are representative or forecast a live return. Start by stating what is sampled and what question the output is meant to answer.

A permutation keeps the trades; a bootstrap redraws them

A permutation shuffles a fixed set of outcomes without replacement. Every observed trade appears once in each reshuffled path. With fixed dollar outcomes and additive equity, the final total stays the same, but the order can change the interim peak-to-trough drawdown.

A bootstrap samples with replacement. An outcome can appear several times in one path while another observed outcome is absent. The final total can therefore change between draws, even though every draw comes from the same original record. Neither operation is a new strategy run: it does not regenerate prices, recalculate position size from a changed equity balance or model a fresh execution path.

A 12-trade example

Assume fictional closed-trade net results of six gains of $100 and six losses of $80, starting from $1,000. Treat the dollar outcomes as already net of the costs included in this example. Both sequences contain the same 12 outcomes, so both finish at $1,120, or +12%.

If the six losses come first, equity falls to $520 before recovering. The drawdown from the initial $1,000 peak is 48%. If losses and gains alternate starting with a loss, equity first falls 8% to $920; the later peak-to-trough declines are smaller. Both sequences still end at the same value. This comparison isolates ordering; it does not say either ordering is more likely or that the underlying strategy has a 48% future drawdown probability.

same trade outcomes can produce different equity drawdowns when the order changesTwo fictional 12-trade equity paths use the same six plus-100-dollar gains and six minus-80-dollar losses with 1,000 dollars of starting equity. Both finish at 1,120 dollars, but losses clustered first produce a 48 percent drawdown, compared with an 8 percent drawdown when outcomes alternate starting with a loss. Fixed-dollar additive illustration, not a forecast.SAME NET OUTCOMES · DIFFERENT INTERIM DRAWDOWN$1,000$750$500StartTrade 12Closed trade · equity dollarsBOTH PATHSEnd $1,120 · +12%Losses firstMax drawdown 48.0%AlternatingMax drawdown 8.00%Same outcomes · different pathFictional: 6 × +$100, 6 × −$80 · fixed dollar additions · no costs beyond those in each net result
Two fictional 12-trade equity paths use the same six plus-100-dollar gains and six minus-80-dollar losses with 1,000 dollars of starting equity. Both finish at 1,120 dollars, but losses clustered first produce a 48 percent drawdown, compared with an 8 percent drawdown when outcomes alternate starting with a loss. Fixed-dollar additive illustration, not a forecast. Illustrative teaching graphic · not live market data

With replacement, a particular 12-draw path may contain, for example, eight losses and four gains, or the reverse. Those paths change the final net sum. A percentile across many such draws summarizes the selected resampling rule applied to this sample; it is not a confidence interval for tomorrow’s market.

What SuperQuant’s diagnostic actually does

The product’s “Bootstrap trade-order risk” diagnostic takes the current run’s closed-trade net dollar outcomes and samples them with replacement. It adds each sampled dollar amount to starting capital. The default run uses 1,000 resamples, the implementation bounds an explicit count from 100 to 5,000, and fewer than 10 closed trades produce an “insufficient” result. A seed derived from the strategy and data fingerprints makes the same frozen run reproducible.

The diagnostic reports descriptive percentiles for final return and maximum drawdown, along with the share of simulated paths that end below starting capital and the share whose modeled equity reaches zero. In the interface, a label such as “loss in 18%” means 18% of this finite resampling run ended negative. It is not an estimate that a live strategy loses with 18% probability.

Each sampled trade remains a fixed dollar result. The calculation does not preserve chronological clusters or duration, reapply current market costs, recalculate size as equity changes, create intrabar gaps, or model outages, liquidation rules and actual venue fills. Costs already present in the source trade’s net value stay embedded in that observed result; missing or understated source costs are not repaired by resampling. The bundled product histories are synthetic and cannot establish market performance.

SuperQuant bootstraps a frozen run's net trade outcomesThe product diagnostic requires at least 10 closed trades, samples observed net dollar results with replacement, adds each sampled result to starting capital, and summarizes the simulated return and drawdown distributions. A fingerprint-based seed reproduces the same frozen-run result. It does not model price paths, serial dependence or new execution costs.WHAT THE CURRENT PRODUCT DIAGNOSTIC DOES1 · INPUTfrozen run's closednet dollar outcomes2 · GATEat least 10 tradeselse insufficient3 · REDRAWwith replacement1,000 default · 100–5,0004 · ADDfixed net dollarsto starting capital5 · REPORTreturn · drawdownsample frequenciesNot replayed: chronology · changing size · new cost assumptions · market executionSame strategy + data fingerprint → deterministic seed · descriptive sensitivity onlyA simulated share ending negative is not a probability of future loss
The product diagnostic requires at least 10 closed trades, samples observed net dollar results with replacement, adds each sampled result to starting capital, and summarizes the simulated return and drawdown distributions. A fingerprint-based seed reproduces the same frozen-run result. It does not model price paths, serial dependence or new execution costs. Illustrative teaching graphic · not live market data

Choose a method that matches the question

Use a shuffle to illustrate how the same recorded outcomes can create different equity paths. Use a with-replacement bootstrap when you specifically want to examine sensitivity to which observed outcomes are drawn into a same-length sample. If the research question depends on neighboring observations, drawing individual trades independently discards that sequence structure.

The stationary bootstrap introduced by Politis and Romano resamples consecutive time-series segments with exponentially distributed lengths; the arch package documents this method in its time-series bootstrap reference. Keeping observations together within a segment can retain some short-run dependence that independent single-trade draws discard. It is a different procedure from SuperQuant’s trade-order diagnostic, and its stationarity and weak-dependence assumptions must be examined for the data and statistic being studied.

Here is an index-only illustration. Start with nine ordered observations numbered 1 through 9. One possible length-nine sample with three consecutive blocks of length three is 2, 3, 4 | 7, 8, 9 | 1, 2, 3. The original order remains inside each selected block; the next block can restart elsewhere, and an observation can be repeated or omitted. The stationary bootstrap randomizes block length rather than fixing every block at three. In the arch 8.0.0 documentation, the StationaryBootstrap input is the average block size. Three observations here is only a teaching choice, not an estimated or recommended setting.

stationary bootstrap keeps selected observations in consecutive blocksIndex-only teaching illustration: one possible length-nine resample uses original observations 2–4, 7–9 and 1–3 in that order. It is not market data; block length three is only an example, not a recommended setting.PRESERVE LOCAL ORDER INSIDE SELECTED BLOCKSOriginal ordered observations123456789One sampled path234789123Block 1 · indices 2–4Block 2 · indices 7–9Block 3 · indices 1–3Illustrative blocks only · stationary bootstrap randomizes block length; average size must be justifiedOriginal order retained within blocks · not market history, not a recommended setting
Index-only teaching illustration: one possible length-nine resample uses original observations 2–4, 7–9 and 1–3 in that order. It is not market data; block length three is only an example, not a recommended setting. Illustrative teaching graphic · not live market data

The unit matters. A block of three trades means three consecutive trades, not three hours; trade durations and idle intervals can vary. If the question is about dependence across hourly or daily returns, use a time-indexed series at that frequency and align the statistic with it. Inspect several defensible average block sizes and report the choice. A block bootstrap can keep some local runs together, but it cannot restore a missing market regime, fix selection bias or establish that a nonstationary sample represents the future.

Report the input period, trade count, net-result definition, position-sizing rule, sampling method, replacement rule, seed, iteration count and all requested percentiles. Show the full distribution and the original run beside it. If repeated inspection of resampled outputs changes the strategy, record that research choice and reserve later data for a fresh evaluation. Continue with backtest metrics, overfitting checks and the trading-journal guide. This is a diagnostic explainer, not a risk forecast or investment advice.