Example robustness report: trend-atr (ma=40, atr_mult=3.5)

About this sample. Strategy: trend following (long above a 40-day EMA, exit on a 3.5× ATR trailing stop), picked as the best of 25 combinations. The price data is a synthetic daily price series (a seeded random walk with bull, bear and choppy regimes, Jan 2018 to Dec 2025, 2,922 daily bars). It is not a real coin such as BTC. The parameters were tuned on the first 70% of the data (to Aug 6, 2023), then the report checks how the strategy held up on the untouched last 30%. Costs: 0.10% fee and 0.05% slippage per fill, long only. It's an example made with the Backtest Toolkit, not a forecast; with Toolkit Pro you can run the same kind of checks (out-of-sample split, Monte Carlo, parameter heatmaps) on your own data. ← All example reports (Backtest Toolkit)

start	2018-01-01
end	2025-12-31
total_return	+1150.7%
cagr	+37.1%
sharpe	1.00
sortino	1.52
max_drawdown	-48.5%
volatility	+39.0%

Summary

Mixed: treat with caution

Robustness score out of 100. Input: trend-atr_equity.csv (equity, 2922 rows, 2018-01-01 to 2025-12-31).

Check	Result	What it means
Out-of-sample degradation	warn	Sharpe 1.18 before 2023-08-07 vs 0.54 after (46% retained). Edges that survive usually keep at least half.
Year-by-year consistency	pass	5 of 8 calendar years were profitable.
Deflated Sharpe ratio	pass	Probability the true Sharpe is above zero: 100% before, 100% after deflating for 25 configuration(s) tried, skew 0.29 and kurtosis 10.1. Below 95% means the result is not distinguishable from the best of many lucky tries.
Monte Carlo	warn	2,000 resampled histories (bars). P(loss) 2%. Backtest max drawdown -48%, bad-case (p5) -73%. Plan for the bad case, not the backtest.
Parameter sensitivity	warn	25 parameter combos. 28% sit within 25% of the best sharpe (the 'plateau'); 88% are positive. A lone bright cell is a red flag; a broad plateau is reassuring.

Out-of-sample degradation

Sharpe 1.18 before 2023-08-07 vs 0.54 after (46% retained). Edges that survive usually keep at least half.

Year-by-year consistency

5 of 8 calendar years were profitable.

Deflated Sharpe ratio

Probability the true Sharpe is above zero: 100% before, 100% after deflating for 25 configuration(s) tried, skew 0.29 and kurtosis 10.1. Below 95% means the result is not distinguishable from the best of many lucky tries.

sr_period	0.0526
skew	0.2861
kurtosis	10.1359
n_obs	2921
n_trials	25
sr_benchmark_period	0.0015
psr	0.9978
dsr	0.9972

Monte Carlo

2,000 resampled histories (bars). P(loss) 2%. Backtest max drawdown -48%, bad-case (p5) -73%. Plan for the bad case, not the backtest.

	total_return	cagr	max_drawdown
p5	53.7%	5.5%	-72.5%
p25	425.0%	23.0%	-59.2%
p50	1112.0%	36.6%	-50.6%
p75	2794.1%	52.3%	-43.8%
p95	10245.5%	78.6%	-36.0%

Parameter sensitivity

25 parameter combos. 28% sit within 25% of the best sharpe (the 'plateau'); 88% are positive. A lone bright cell is a red flag; a broad plateau is reassuring.

Method notes

OOS split: last 30% unless a split date is given. Monte Carlo: bootstrap of closed-trade returns (or stationary block bootstrap of per-bar returns for equity input). Deflated Sharpe: Bailey & Lopez de Prado (2014), benchmark = expected max Sharpe of N null trials. Parameter sensitivity requires the optimization grid CSV. Engine: Crypto Backtest Toolkit Pro.

Backtests are hypothetical, use historical or synthetic data, and do not guarantee future results. Not financial advice.