Free courseLesson 05 of 10

Backtest it honestly

Run a test you can believe: real costs, real fills, the same risk on every trade. Then explore the variations.

What a backtest is, and how a test cheats
The numbers on your report
Exploring thousands of variations

What a backtest is

A backtest runs your rules over past prices and shows what would have happened. It's a simulation. It can't tell you what comes next.

Costs go in, always

The toolkit adds commission and slippage to every trade by default. Never switch them off to see a nicer number.

Fills you could really have had

A test can quietly assume you got a price that never traded, or that your target was hit in the same minute you got in. The toolkit checks every fill against one-minute data and flags the ones that look wrong.

And when one candle touches both your stop and your target, the stop counts first, unless one-minute data shows the target came first.

The same risk on every trade

Risk the same dollar amount on each trade. Then results are comparable, and one oversized trade can't make a weak rule look strong.

Read it year by year

A rule that made everything in one year and nothing in the others is not a good rule. Ask for results per year, every time.

The numbers on your report

Profit factor. Everything the winners made, divided by everything the losers lost. Above 1 means the rule made money after costs.

Drawdown. The deepest drop from a high point to the next low.

Result over drawdown. What the rule made, compared with the worst stretch you'd have sat through.

Number of trades. A handful of trades tells you nothing. The more trades, the more the result means.

Explore the variations

Every idea has a few settings that can be varied, such as the size of the target. Your assistant can now run hundreds or thousands of variations of your rule and rank them.

What it ranks on. Result over drawdown, with a minimum number of trades. Never the share of winners.

Plateaus, not peaks. A setting only counts if its neighbors hold up too. One great result surrounded by bad ones is luck.

Every variant is counted. The more you tried, the more a pretty result deserves doubt. The count goes on every report.

Do it with your AI assistant
Run my strategy over the last three years with the kit's standard costs and fixed dollar risk per trade. Then check the test for cheating: fills at prices that never traded, exits counted in the same minute as the entry, and trades without costs. Report results per year and tell me plainly if anything looks too good to be true. Then build section 6 of the blueprint if it isn't built yet and explore the settings of my strategy: rank by result over drawdown with a minimum number of trades, show me the top results together with their neighbors, and tell me how many variants you tried.
You're done when
  • Your report includes costs
  • The fill check ran, and you know what it flagged
  • You've seen the result for each year separately
  • You've explored the variations and know how many were tried