If you have ever Googled "how to make money trading", you have seen the same thing over and over: confident charts going up, screenshots of "an algorithm with 200% annual returns", promises of "a system that never loses". They all sound the same way — more confident than honest work with the market should sound.
And that, in fact, is the main signal you should learn to react to. Confidence in trading almost never comes from quality — it comes from the fact that the person did not run the necessary checks. Real work looks different: more skepticism, fewer round numbers, and almost always a modest result that you arrive at through several layers of self-deception.
I spent a few months building such a system on crypto — not to impress anyone with it, but to learn how to tell a real result from four things that look a lot like one. What follows is what a trading strategy is, which numbers are actually worth understanding, and how exactly those numbers lie. No formulas, plain language.
What a systematic strategy is
A systematic strategy is a rule that replaces intuition.
Instead of "I have a feeling Bitcoin is going up", a concrete condition that a computer can check. For example: "if the price right now is above its average over the last six months, and it has been rising over the last couple of months — hold it. Otherwise — sit in cash". No news, no gut feel. Every day we check the condition — true or false.
The advantage of a rule over intuition is not that it is smarter. It is that a rule can be tested on history: run it on data from past years and see what would have happened if you had traded by this rule from 2018 to today. That test is called a backtest. And the backtest is the main source of self-deception in trading, because it almost always produces a prettier result than what would have actually happened.
Sharpe ratio: the one number worth understanding
Out of all the trading jargon, there is one metric a beginner should know — the Sharpe ratio. From here on I will just call it "Sharpe".
Sharpe is not "how much you earned". It is how much profit you got per unit of stress — per percent of swings on the way there. A strategy that returned +50% over five years and grew more or less steadily has a high Sharpe. A strategy that also returned +50% but lost half its capital twice along the way has a low Sharpe — same profit, but it rattled you a lot harder.
A rough scale to anchor on:
- Sharpe ~0.5 — weak, barely distinguishable from luck.
- Sharpe ~1 — respectable; trustworthy if the strategy has survived honest checks.
- Sharpe ~2 — excellent, rare for a private researcher.
- Sharpe above 3 — almost always either a bug in the code or overfitting to the past. This is a red flag, not a reason to cheer. The best funds in the world have spent decades around Sharpe 1.5–2.5. If a home-made strategy from the internet shows Sharpe 5, somebody has fooled themselves or you.
If you take one thing from this section, take this: in trading, big pretty numbers are a reason for suspicion, not delight.
Three ways a backtest lies
On to the main part. A backtest looks like a truth machine: a program ran through historical data and honestly counted what you would have earned. In practice there are three classic ways a backtest starts lying — and all three are everywhere on the open internet.
1. Peeking into the future
The trickiest of the bugs, because you cannot see it from the chart.
Imagine the rule: "buy at the start of each day if that day closes up". Sounds great — we only enter green days. But at the moment of buying in the morning, we do not yet know how the day will close in the evening. A backtest that accidentally uses the closing price of the day to make the decision "in the morning of the same day" produces fantastic results — and they consist entirely of time travel.
That example is cartoonish; in real code the bug hides more subtly. Somewhere a function pulled a price "one day forward", and the strategy used it without knowing. The equity curve looks beautiful. Without a separate check, you cannot tell real talent from an accidentally built time machine.
2. Forgotten fees
When you buy crypto on an exchange, the exchange takes a fee — usually 0.1–0.4% of the amount. Sounds tiny. On one trade — really tiny. But an active strategy can do 50, 100, 200 trades a year.
And here a regular backtest often lies by default: it calculates the result as if trades were free. On paper the strategy earns 50% a year; in reality after fees — it loses 10%. "Best fund of the generation" becomes nothing thanks to one line of code that no one added.
When you see a pretty chart of a strategy — the first and cheapest question is: "are fees in this?". In half the cases the answer is "no", and the whole result falls apart on this single question.
3. Searching through variants until something looks good
This is the most subtle place — and the one where even honest researchers usually fall in.
The idea is simple: you try not one strategy, but 500 variants of one idea — different parameters, different thresholds, different combinations. Out of 500 variants, one is guaranteed to give a pretty Sharpe by pure chance. That is just how statistics work: if you flip 500 coins 50 times each, one of them will show 35 heads out of 50, and that does not mean the coin is magical — it means you searched long enough through ordinary ones.
Then a person, unaware of this trap, shows the world only the best variant. "Sharpe 2.5! Backtest is clean, fees are included!" Everything is honest — except that the result is built on pure luck during the search.
There is a rigorous statistical correction for this — it is called DSR (deflated Sharpe ratio). You do not need to know how it is calculated. You only need to know one thing: if someone shows you a strategy and does not mention how many variants they tried before settling on "this one" — ask them. The answer almost always wipes out the beauty.
The fourth impostor — a different kind
Suppose you have avoided all three traps. The backtest is honest: no peeking, fees included, no search. The strategy returned +80% over five years.
That leaves a question almost nobody asks first, but everyone should: is this actually better than doing nothing and just holding Bitcoin?
The answer is often "no, it is not". And then your complicated strategy, with multiple conditions and months of code, loses to the "buy and forget" button. The strategy is not broken. It is simply unnecessary.
This is a different class of question. The first three traps are defects that make a bad strategy look pretty on paper. This question stays valid even when the strategy is perfectly clean. The backtest honestly answers "how much"; the question of "is it worth it" is one the backtest does not answer at all. That question is asked by the researcher — or it is not asked, and then years of work go into something worse than simply doing nothing.
What came out of mine
Back to my crypto strategy — it is here not as an achievement, but as an illustration of how the checks above work in practice.
The first, naive backtest showed Sharpe 1.38 over eight and a half years of history. By the scale above — an excellent result, better than most professional funds. The picture was convincing.
Then I started peeling off layers. First I checked whether the result depended on the quirks of one specific exchange: I ran the same rule on prices from a different exchange. The signal matched almost perfectly — a good sign — but it turned out that independent verification was only possible for the last four and a half years; before that, the second exchange's data simply did not exist. On the window where honest verification is possible, the Sharpe came out to 0.72.
Then I applied the DSR correction — "is this maybe the result of a search?". The correction pulls the estimate back down to the level that would be realistic if I had searched blindly among 500 variants for the best configuration. After that correction, what remains is roughly zero.
And here is the key thing. Each of these numbers is not a drop, it is a successful check. The backtest showed 1.38 not because the strategy is that good — but because a naive backtest does not know how to subtract self-deception. Each successive layer of checking subtracted one kind of illusion, and what remained is the honest estimate.
This is what disciplined work with a strategy looks like: you start with a big pretty number, and it melts as you check it. The people who sell you a confident "yes" — they did not peel these layers off. When someone promises you a steady +30% a year, they are not doing it because they found something the rest of us missed. They are doing it because they did not get to the questions on which confidence falls apart.
What to take with you
You do not need to remember the names of the checks. You only need to learn to ask three questions of any strategy chart — your own or someone else's:
- Is the strategy peeking into the future? Is it using data for its decision that, in reality, was not available at that moment?
- Are fees included? Every trade costs money — is that money subtracted from the result?
- Is this the result of a search, and is it better than simply buying and holding? How many variants were tried before settling on this one, and does the final strategy beat plain hold?
If any of these questions has no clear answer — you do not have a result, you have a promise.
What is next
The strategy used in the examples is currently running in paper-trading mode — no real money, it just records every day what it would have done. In a few months I will write a follow-up: what real-time work showed, where the honest estimates matched practice and where they diverged. No promises about the outcome — the point of the exercise is precisely that I do not know how it will end.