Robustness checks

Strategy robustness testing for crypto

A strategy robustness test checks whether a backtest result is real or the product of luck and overfitting, by reshuffling the trades, nudging the settings and testing on dates the strategy never saw. Keel is a strategy builder and backtester for Hyperliquid perpetuals, and robustness checks inside its backtests are coming soon. Today every backtest includes fees, slippage and funding, and you can re-run it with other settings and dates to check by hand.

Sign up free What you can do today
By Keel Research Team
Updated October 15, 2026
Coming soon

Robustness checks

Robustness checks inside Keel backtests are coming to Keel. They are not in the product yet.

In Keel today

  • Backtest over the dates you choose, with fees, slippage and funding
  • Change a setting and re-run; every version and backtest is kept
  • Paper trade on live Hyperliquid prices with no money at risk
What it is

Three ways a backtest can fool you

Luck in the order of trades. A backtest is one path through history. Reshuffle the same trades thousands of times, which is what a Monte Carlo test does, and you see the range of drawdowns the strategy could have had. The bad end of that range is a better guide to risk than the single drawdown in the report.

Settings that only work in one spot. Move each setting a step either way and run again. A real edge fades slowly as you move. An overfit one falls apart when a lookback changes from 14 to 15.

Results that only exist in-sample. Test on dates the strategy never saw while you were building it. Measures like the probability of backtest overfitting and the deflated Sharpe ratio put a number on how much of a result to believe after you have tried many versions.

Why it matters in crypto

Small samples, fat tails

Crypto returns have fat tails: a handful of days can make or break a year. A strategy with a few dozen trades over a couple of years can show a great Sharpe ratio by chance, and one crash can be most of its drawdown.

Many coins are young, and a strategy that trades a basket keeps meeting new listings and coins that fade away. That makes each backtest a smaller sample than it looks.

So these checks matter more in crypto, before any money goes behind a result.

A real example

How much the result moves with the settings

The library tests every strategy across a set of settings. A tight spread is a good sign. One tall bar among weak ones is a warning.

Funding smoothing / Rebalance buffer / Pinned universeSharpe after costs
  1. 20 / 0.2 / top60_launch1.62
  2. 20 / 0.2 / top100_launch1.61
  3. 10 / 0.2 / top60_launch1.58
  4. 20 / 0.15 / top60_launch1.57
  5. 20 / 0.15 / top100_launch1.53
  6. 20 / 0.1 / top60_launch1.52
  7. 10 / 0.15 / top60_launch1.50
  8. 10 / 0.2 / top100_launch1.48
  9. 20 / 0.1 / top100_launch1.45
  10. 5 / 0.2 / top60_launch1.44
  11. 10 / 0.1 / top60_launch1.43
  12. 10 / 0.15 / top100_launch (default)1.39
  13. 5 / 0.15 / top60_launch1.36
  14. 5 / 0.1 / top60_launch1.32
  15. 10 / 0.1 / top100_launch1.31
  16. 5 / 0.2 / top100_launch1.28
  17. 5 / 0.15 / top100_launch1.24
  18. 5 / 0.1 / top100_launch1.18

Each bar is a full backtest of one setting. Funding Carry (Cross-Sectional, Hyperliquid), Hyperliquid, Aug 2024 to Sep 2026, after trading fees, slippage and funding. Past results don't predict future ones.

See Funding Carry (Cross-Sectional, Hyperliquid) in the library
In Keel today

Check a strategy by hand today

The pieces of a robustness test that you can run in Keel now.

Re-run with nearby settings

Change one setting a step at a time and run the backtest again. Every version and every backtest is kept, so you can line them up.

Test on dates you held back

Set the start and end date of any backtest. Build on one stretch of history, then check the finished strategy on a stretch you kept aside.

Real costs in every run

Every backtest charges trading fees and slippage and applies Hyperliquid's hourly funding to open positions, so a setting that trades a lot pays for it.

How the backtest works

Free calculators for your own results

Keel's free tools reshuffle a list of trades or returns in a Monte Carlo test, and estimate overfitting and the deflated Sharpe ratio from numbers you paste in.

Open the Monte Carlo tool
Coming soon

Robustness checks

Robustness checks inside Keel backtests are coming to Keel. They are not in the product yet. Sign up free and we'll email you when it ships. Your account works today: build a strategy, backtest it and paper trade it while you wait.

Sign up free
FAQ

Common questions

What is a strategy robustness test?

A set of checks that ask whether a backtest result would survive small changes: a different order of trades, slightly different settings, and dates the strategy never saw. A strategy that passes them is more likely to hold up live.

How do I know if my backtest is overfit?

Warning signs are a great result from few trades, a result that collapses when one setting moves a step, and a curve that only works on the dates you built it on. Test nearby settings, test unseen dates, and count how many versions you tried before this one.

What does a Monte Carlo test show?

It reshuffles or resamples a strategy's trades many times to show the range of outcomes the same edge could have produced, including deeper drawdowns than the one in the backtest.

Does Keel run robustness checks on backtests?

Not yet. Robustness checks inside Keel backtests are coming soon, and if you sign up we'll email you when they ship. Today you can re-run a strategy with nearby settings and on dates you held back, and use the free calculators on your results.

What is the probability of backtest overfitting?

An estimate of how likely it is that the best of many tried versions does worse than the median version on data it was not chosen on. A high value means the selection process, rather than the strategy, produced the good backtest.

How realistic are Keel's backtests?

Every backtest charges trading fees and slippage on each trade and applies Hyperliquid's real hourly funding to every open position. It is still a simulation: fills are modelled, and past results don't predict future ones. How Keel works explains how the backtest and the live run use the same compiled strategy.

Start with a backtest

Sign up free and we'll email you when it ships. Your account works today: build a strategy, backtest it and paper trade it while you wait.

Sign up free