Skip to content
Kestrel

Performance, in full

Everything below is derived from 2,745 recorded trades across 10 instruments between 2021-09-08 and 2026-09-03. Nothing on this page is illustrative and nothing is rounded in our favour.

Average result per trade
0.0611 R
R = one unit of risk
Win rate
37.7%
losses outnumber wins by design
t-statistic
2.27
see the caveat below
Sharpe ratio (annual)
1.02
this is what sets consistency

Outcome distribution by holding period

Each row is 20,000 simulated runs at 1.00% risk per trade, net of the 5% performance fee. A single run is one possible future; the spread between the bad and good columns is the honest answer to “what will I make”.

Held forBad case (p5)Typical (median)Good case (p95)Ends in profitDrawdown (median / p95)
1 month-12.2%+1.9%+18.8%58.5%8% / 16%
3 months-17.8%+6.6%+38.0%66.1%14% / 26%
6 months-20.9%+14.4%+66.4%72.8%18% / 33%
1 year-22.0%+30.9%+121.0%80.5%24% / 41%
2 years-17.6%+72.3%+263.0%88.5%30% / 49%
3 years-8.6%+126.0%+467.6%93.1%34% / 54%
5 years+22.0%+290.4%+1193.3%97.3%39% / 59%

Result by calendar year

The losing years are in this table because leaving them out would make every other number on this site meaningless.

YearTradesWin rateBefore feeAfter fee
20219332%-1.4%-1.4%
202238133%-15.2%-15.2%
202367238%+41.1%+39.0%
202462838%+27.2%+25.9%
202559839%+50.4%+47.9%
202637342%+80.8%+76.7%

How these numbers were produced

What the simulation does

Each trade in the record is an R-multiple — the result expressed in units of the risk taken. A run draws 550 of them at random, with replacement, and compounds them at 1.00% risk per trade. Twenty thousand runs give the distribution shown above.

What costs are included

Real broker spread measured per instrument, five points of slippage per side, and overnight financing charged per night held rather than prorated. Costs are not optional and are already inside every figure on this page.

Why the t-statistic overstates confidence

The instruments in this book were chosen from a much wider search. A wider search finds a better-looking best result even when nothing is there, so 2.27 is inflated relative to a figure produced by testing one pre-chosen strategy once. Treat it as suggestive.

What this is not

It is not a live client track record. It is simulation over recorded historical trades. Live trading is recent and the live sample is still far too small to confirm or contradict any of this.

Why the drawdown matters more than the return

The median run gives back 24% from a peak at some point in a year, and the worst five percent give back 41%. Most people who lose money with a working system lose it by stopping during that stretch.

Independent bootstrap, and its limit

Runs are drawn independently, which does not model the tendency of bad stretches to cluster together. Real drawdowns can therefore run somewhat deeper than the table shows. We would rather say that than quietly let the model flatter itself.

Want to see what a specific amount does over a specific number of years? Run it through the calculator. It shows the bad case beside the good one, because only seeing one of them is how people end up surprised.