What a backtest can prove, and what it cannot

A backtest is the first tool in this track that can flatter you. Your watchlist could be stale and your screener could be incomplete, but neither ever told you that you were right. A backtest does, in confident percentages, and most of this course is about not being fooled by your own program.

The one honest claim

A backtest supports exactly one sentence:

Under these assumptions, on this data, a rule stated in advance would have produced this sequence of results.

Every word in that sentence is load-bearing. Under these assumptions — costs, slippage, fills, all of which you chose. On this data — one history, one vendor, one adjustment method. Stated in advance — or you are describing the past, not testing a rule. Would have — nobody traded it.

That claim is genuinely useful. It kills bad ideas cheaply, which is most of what research is. What it cannot do is tell you what happens next.

Three things it cannot prove, ever

That the rule works. One history is one sample. A rule that survived it might be sound or might have been lucky, and a single backtest cannot separate those. Unit 4 is about narrowing the gap; nothing closes it.

That you would have followed it. The equity curve assumes a person who took every trade without hesitating, including the one after a fourth consecutive loss. You do not know that person.

That the future resembles the sample. Rates, market structure, participants and fees all change. A rule fitted to one regime is a description of that regime.

The line this course does not cross

A backtest is research. It is not advice, not a recommendation, and not a claim that anybody should do anything with money.

This matters more here than anywhere else in the track, because a chart with a rising line and a Sharpe ratio beside it looks like a recommendation whether or not you meant it as one. If you publish results — a repo, a blog post, a screenshot — say plainly that it is a historical simulation and that you are not a licensed adviser. That is not a disclaimer bolted on at the end; it is an accurate description of the artefact.

The data terms are the same as in courses 1 and 2. EODHD's plans are, in their own words, "intended for personal use only as commercial use requires a more thorough approach to licensing and data use", and a backtest run for yourself is personal use. Selling signals derived from it is a different conversation, and the licence page in this lesson's sources is where it starts.

What you will actually have

A backtester in the same repository as your watchlist and screener: candles loaded through your existing proxy, a rule you can state in one sentence, costs visible on the first curve you ever draw, four bias tests you wrote yourself, and an exported report you would be comfortable handing to someone who wanted to pick holes in it.

The finance behind it

Why historical averages flatter almost everyone who quotes them, from the finance side: Why do "average returns" often look too good?

Try it now

Write the one honest sentence for a rule you are curious about, filling in every blank: which assumptions, which data, stated when. If you cannot fill in "stated in advance" — because you picked the rule after looking at a chart — you have learned something before writing a line of code.