A crypto backtest return distribution shows how individual modeled outcomes are spread, not just their average. To review one, keep every comparable net closed outcome, sort a copy from worst to best, and inspect the center, middle range, extremes, and concentration. Preserve the original order too. A sorted view can expose outliers, but it cannot show when losses occurred or predict what a live market will do next.
What a crypto backtest return distribution answers
A headline return compresses a historical path into one figure. A distribution asks different questions:
- Where do most closed outcomes sit?
- How far apart are the middle observations?
- How large are the most adverse and favorable outcomes?
- Does the average depend on one or two extremes?
Those questions help describe the tested sample. They do not establish the probability of a future outcome. Before reading the shape, confirm that every observation uses the same definition.
Choose one comparable outcome unit
A distribution can be built from net quote-currency PnL, percentage return on committed capital, or a risk-normalized unit. These are not interchangeable. A 10-unit result from a large position should not be compared casually with a 10-unit result from a much smaller one.
Write the unit beside the distribution and keep these boundaries fixed:
- closed outcomes only, with open PnL reported separately;
- the same fee, spread, slippage, funding, and borrowing treatment;
- the same capital basis and position-sizing convention;
- the same market, side, date range, and strategy revision, or clearly labeled groups;
- one rule for partial exits and multiple fills.
If the unit or cost treatment changes halfway through the ledger, split the sample or restate it on a common basis before comparing the observations.
An equal-average example with different tails
Consider two invented eight-outcome ledgers, already sorted from worst to best. Every value is a normalized test unit after modeled costs:
| Ledger | Sorted net outcomes | Mean | Median | Win rate |
|---|---|---|---|---|
| A | -2, -1, -1, -1, 2, 3, 4, 4 | +1.0 | +0.5 | 4 of 8 |
| B | -10, -1, -1, -1, 2, 3, 4, 12 | +1.0 | +0.5 | 4 of 8 |
Both ledgers have the same mean, median, and count of positive outcomes. Ledger B still contains a much larger adverse extreme and depends more heavily on one favorable extreme. The shared summaries do not make the historical samples equivalent.
The figures illustrate arithmetic only. They are not Gimmer results, exchange data, cryptocurrency market data, a target distribution, or evidence about future performance.
Read the center, spread, and tails separately
Center: compare the mean and median
The mean adds every observation and divides by the count. The median marks the middle of a sorted sample, using the average of the two middle values when the count is even. A large gap between them can be a prompt to inspect skew and extremes, not a verdict by itself.
NIST’s Engineering Statistics Handbook explains mean, median, and alternative measures of location, including why some location measures are less affected by tail observations. That general statistical reference does not define a trading threshold or validate a backtest.
Spread: inspect the middle range
Quartiles divide a sorted sample into sections. The distance from the lower quartile to the upper quartile is the interquartile range. It can summarize the middle half without letting the single largest gain or loss set the entire scale.
Software can use different percentile conventions, especially for a small sample. Record the method rather than comparing two quartile values whose calculation rules are unknown. Pair the distribution with the crypto backtest volatility checklist when the return interval and dispersion calculation also matter.
Tails: locate concentration and extremes
Mark the worst and best observations, then recalculate the aggregate result without the largest few outcomes as a sensitivity check. This does not justify deleting inconvenient trades. It shows whether a conclusion depends heavily on a narrow part of the tested sample.
Small samples make tail estimates especially fragile. Use the backtest sample-size framework to examine decision count, time coverage, market coverage, and concentration before interpreting a sparse distribution.
Do not sort away the original sequence
Sorting makes the distribution readable, but it removes time order. Two ledgers with the same sorted values can reach different drawdowns when the outcomes occur in a different sequence. Keep an untouched chronological ledger beside the sorted copy.
The Monte Carlo sequence-risk guide explains how separate resampling analysis can challenge the order of observed closed outcomes. It also keeps an important limit visible: resampling rearranges or repeats the available observations; it does not create new market evidence.
Separate distribution shape from the payoff total
A distribution view does not replace the closed-trade ledger or its aggregate checks. Rebuild gross winning and losing totals with the profit-factor ledger checklist, then inspect whether a few outcomes dominate either side.
Group the outcomes by market, direction, position-size band, and predefined market condition when the sample permits. A combined distribution can hide that one subgroup supplied most favorable observations while another carried most losses. Define those groups before hunting for a flattering slice.
Where Gimmer fits in the review
Gimmer’s public Running Backtest guide describes a historical workflow with report measures, open and closed positions, and an Execution ledger that can be checked against the report. It also says to treat the result as evidence about the tested strategy revision and history, not as automatic approval.
This article does not claim that Gimmer exposes a dedicated return-distribution chart, export, or percentile calculation. Use the visible closed-position and execution evidence available for the exact run, record the outcome unit, and keep open exposure outside a closed-outcome shortcut.
Crypto backtest return distribution checklist
- Freeze the run. Record the strategy revision, markets, dates, interval, sizing, and cost settings.
- Choose the unit. State whether each observation is net PnL, return on a declared capital base, or a risk-normalized result.
- Set the boundary. Keep closed outcomes in the distribution and report open PnL separately.
- Reconcile the ledger. Check entries, exits, amounts, costs, and timestamps against the report.
- Preserve two views. Keep the chronological ledger and make a separate sorted copy.
- Read center and spread. Record the mean, median, quartile method, middle range, and sample count.
- Inspect the tails. Mark extremes and test how much the aggregate depends on the largest few observations.
- Challenge the conclusion. Repeat the same contract on a locked out-of-sample period before treating the pattern as durable.
Frequently asked questions
Is average return enough to evaluate a crypto backtest?
No. The average does not show the sample size, median, spread, tails, sequence, drawdown path, open exposure, or dependence on a few observations. Read it as one historical summary inside a wider review.
Should open positions be included in a closed-outcome distribution?
Not without an explicit and comparable valuation rule. A cleaner default is to keep open PnL and exposure visible beside the closed-outcome distribution rather than treating an unrealized position as if it were already closed.
Does a stable historical distribution predict future returns?
No. It describes the selected historical data, strategy revision, execution assumptions, and outcome unit. Later market conditions, liquidity, costs, fills, outages, and behavior can differ.
Make the distribution reproducible
A crypto backtest return distribution is most useful when another reviewer can rebuild it from the same closed outcomes and get the same center, spread, and tails. Keep the unit, costs, open-position boundary, percentile method, sample coverage, and original order attached to the chart or table.
For the next review, open Gimmer’s Running Backtest guide, freeze one completed historical run, and create a sorted copy of its comparable net closed outcomes. Read that distribution beside the original sequence and out-of-sample evidence, not as a forecast.