Same trades. Different paths.

Three questions, not three interchangeable forecasts

Permutation

Shuffle all N observed trades without replacement. Each outcome appears exactly once. Under fixed-size additive accounting, every path ends at the same equity. The order can still change the drawdown.

IID bootstrap

Draw N trades independently with replacement. The empirical one-trade distribution is held fixed, but counts and terminal P/L vary. Serial dependence is discarded.

Circular block bootstrap

Draw fixed-length contiguous blocks with replacement, with uniformly sampled start positions and wrap-around. Concatenate, then truncate to N trades. Local dependence is retained within blocks, not across their joins. This is not the stationary bootstrap with random block lengths.

Every method uses the same trade count, capital and seed. The same seed means repeatability, not paired identical scenarios across different algorithms. Inspect the exact fixtures, all block lengths and known-answer tests.

Definitions and practical interpretation

At closed-trade checkpoint t, Eₜ = E₀ + Σ P/Lᵢ. The running peak includes starting capital. Absolute drawdown is peakₜ − Eₜ. Percentage drawdown is 100 × (peakₜ − Eₜ) / peakₜ. Maximum absolute and percentage drawdowns can occur at different checkpoints.

The threshold output is the fraction of simulated paths with maximum absolute drawdown greater than or equal to your USD threshold. Quantiles use the empirical nearest-rank rule. P95 describes this resampling distribution, not a 95% confidence bound on future loss. A zero exceedance count does not establish zero real-world risk.

The displayed paths are actual simulations chosen at P05, P50 and P95 of maximum USD drawdown. They are not pointwise percentile curves or confidence bands. The original ledger is shown separately. The horizon is exactly N completed trades, not a calendar period.

What this tool cannot establish

• No compounding, position resizing, deposits, withdrawals, margin calls or liquidation. Equity may fall below zero and drawdown may exceed 100%.
• Only closed-trade P/L is observed. Open-position losses and intraday drawdown are not modeled.
• Historical dependence may not persist. No simulated outcome can introduce an unobserved type of loss.
• No fill model, strategy replay, firm-specific pass probability or statistical certification of an edge.
• Net input retains its costs. Gross input remains gross. Use the futures ledger cost audit before resampling when cost treatment needs reconciliation.
• Single-trade, zero and constant ledgers are valid arithmetic controls, but have little or no resampling uncertainty. A longer record does not automatically remove selection bias.

The run is limited to 2 million sampled trades across selected methods and 10,000 simulations per method. Computation runs in a dedicated worker and can be canceled. Input and results remain in memory until you leave or reload. A downloaded summary contains derived figures you choose to save, but never the individual trade rows.

Continue the research

• How shuffling and block sampling change the question
• Why losing streaks and account barriers differ
• Parameter stability and robustness beyond one risk simulation
• Chronological validation and reused holdouts