Is My Trading Strategy Just Luck?

You have a track record, and it made money. Maybe you're looking at a backtest, maybe you're looking at the last few weeks of your live account. Say it returned +40%, or it won 12 of your last 20 trades. It feels like you found something.

But is what you found skill or luck?

The good news is that this question is answerable with common metrics you probably already have at hand. The trick is reading them in the right order, and as a whole, with the right mindset. This post will walk you through it. By the end, you'll have the tools to evaluate not just the strategy that brought you here, but every strategy that gets sold to you online.

(For reasons that might become obvious, trading influencers, especially the ones that sell you hyped-up golden-egg strategies on X or YouTube, don't often have the incentive to teach their audience how to tell skill from luck.)

One reassurance before we start: there is no bad news here. Finding out a strategy is dumb luck today spares you from betting real money on it tomorrow. You might be briefly disappointed. But trust me on this: the disappointment of losing money is much sharper than the disappointment of not making it.

Step 1: Establish the Mindset

TLDR: You want your strategies to have skill instead of luck. Skill is repeatable. Assume your strategy is chance, then prove luck can't explain the results.

Let's start by defining skill, ie. the characteristic you want your strategy to have when it isn't luck: a real, repeatable advantage in its rules that makes your average trade profitable, and profitable for a reason that keeps working.

(Quants have another name for this: an edge. Skill and edge are the same thing, used synonymously here. An edge is not "the strategy that made the most money." A strategy with no edge can get rich by luck, and a strategy with a real edge can be modest.)

By default, we should assume that a strategy is effectively a coin flip, and any profit is noise. That's called a null hypothesis: the basic assumption that your data must now disprove.

You might ask: why is the basic assumption "your strategy is chance" and not "your strategy is bad"? Well, "bad" is itself a claim that needs proving. The neutral baseline is the one that assumes nothing: outcomes that are exactly what randomness alone would produce, neither skill nor anti-skill.

Once the null hypothesis is established, our goal becomes not to prove your strategy is good, but to show that pure luck fails to explain your results. When chance can't account for what you're seeing, the only explanation left is that something real is going on. It's proof by elimination.

Step 2: Did the strategy make money?

TLDR: A profit can be genuine skill, or it can be a handful of lucky trades. Even a small one fools you. To separate those two scenarios, you need a different tool than "look at how much it made."

It's the number everyone looks at first. What's your PnL?

(Sanity-check: PnL stands for profit and loss, and you're looking for a positive number.)

Here is a real result from my own backtest fleet. An EMA 50/200 crossover strategy on ETH, 4-hour candles, returned +15.12% over the test window (see the run). Modest but positive.

It also means almost nothing.

PnL tells you what happened once, on one slice of history. It says nothing about whether this feat is repeatable. And notice the kind of number it is. Not some eye-popping return you'd reflexively distrust, just a believable +15%. That's the more dangerous kind, because it's the one you'd actually act on. This particular result came from only 36 trades, and as the next sections show, that's nowhere near enough to tell a real edge from a lucky streak.

If you've got a strategy running on a live account or on a paper-trading account, and you want to see PnL's fickle nature in action, you've got a few options:

1. slice your trading history into time-bound chunks and pull the PnL from each
2. let it run long enough to watch its performance as the market varies, then do #1
3. invest in running a backtest over different time periods

If you're running a backtest, run your strategy over a few different market conditions: a rising stretch, then a falling one, then something choppy. Almost certainly, some runs will come back positive and others negative.

Yes, ideally you're looking for strategies that return a positive PnL over most time windows, or at least over the market regimes that fit your strategy. For example, the EMA Cross tends to work better in trending markets. But don't let PnL alone determine whether a strategy is worth sticking to.

Step 3: What percentage of trades were successful?

It's the second number everyone looks at. What's your strategy's win rate?

(Sanity-check: Win rate is the % of your total trades that resulted in profit, and you're looking for a number over 50%.)

A strategy's win rate suffers from the same fickle nature as PnL and can be observed by slicing your trading history into the same time-bound chunks or runs. You'll surely find that your win rate slides up and down, sometimes drastically, depending on your time window.

Win rate suffers from an extra issue, too: luck can fake it too easily.

For example: flip a coin 20 times. You expect 10 heads, but 12 wouldn't surprise you. Getting 12 or more heads out of 20 happens about 25% of the time with a perfectly fair coin. So one in four no-skill strategies "wins 60% of its trades" over a 20-trade record, on nothing but chance.

A 60% win rate over 20 trades is therefore not evidence of skill. The real question isn't "did I win more than half my trades?", it's "did I win by enough, over enough trades, that chance becomes an unreasonable explanation?"

A brief detour: the problem with sample size

PnL and win-rate suffer from the same underlying issue: sample bias.

Consider our 20-trade example.

These trades are a sample: a finite slice of every trade the strategy could ever take. The number you actually care about is the strategy's "true" win rate, ie. the value it would settle on if it ran forever, across every market condition it will ever meet. Unfortunately, that's an infinite dataset no one can ever collect. You never observe the true value directly. You only ever see the sample, and the sample is a noisy, incomplete estimate of the infinite dataset.

(Bias sneaks in too: test on one asset, in one rising market, and your sample is not just small, it's tilted.)

A confidence interval measures the width of the gap between the estimated value and the "true" value. It's a range that says: given the limited sample you observed, here is where the "true" value plausibly lives. That's what the "+/-" sign means in writing, or what people mean when they say "give or take a few".

When we talk about strategies, fewer trades mean a wider interval on all the common metrics, because a small possibly-biased sample can't pin down the true value.

Take the EMA Cross strategy from Step 2. Its win rate came in at 33% (12 wins in 36 trades), give or take about 15 points. That "give or take" is the confidence interval, and here it runs from a weak 20% up to a coin-flip 50%. Thirty-six trades can't pin it down any tighter, so this sample can't tell you where in that range the true win rate actually sits.

Step 4: How consistently did the strategy make money?

TLDR: Win rate ignores the size of your wins and losses. The Sharpe ratio doesn't, and the lower bound of its confidence interval is the closest thing to a real verdict you'll get. If even that low end stays above zero, you're probably looking at skill.

So the win rate counts how many trades won, but ignores how big your wins and losses were. That gap is where a lot of strategies live or die.

We can easily imagine a strategy that only wins about one trade in three. It still makes money, because on the trades it wins, it wins big and on the trades it loses, it loses small. A strategy can just as easily win 70% of the time and slowly bleed out, if those occasional losses are large enough. So counting wins only gets you so far. The better question is how consistently the strategy made money.

The number that answers that is the Sharpe ratio.

(Sanity-check: the Sharpe ratio is your average return per trade divided by how much those returns bounce around. Higher is steadier. Roughly: under 0.5 is poor, 0.5 to 1 is okay, 1 to 2 is good, and above 2 is excellent, though anything that good deserves a suspicious second look.)

Because it divides by that bounce, the Sharpe ratio rewards a strategy for making money the boring way: a little at a time, reliably, instead of one lucky moonshot followed by a string of losses.

Now we run the exact same coin-flip test we ran on win rate, except on the Sharpe ratio. Put a confidence interval around it and look at the lower bound. If even the pessimistic end of that range sits above zero, the edge survived the haircut, and you can reasonably call it real. If the interval dips below zero, your answer is that you don't have proof yet. This is the number PerpForge's verdict actually turns on, and the one I'd trust over PnL and win rate combined.

Let me put two real results from my fleet next to each other:

On SOL, 4-hour candles, an EMA 21/50 strategy returned +12.75% over 115 trades (see the run). Its Sharpe interval sits entirely above zero, from about 0.36 to 1.26. Real skill.

Now back to the ETH strategy from Step 2: +15.12%, over just 36 trades (see the run). A slightly bigger return. It's Sharpe interval runs from slightly below zero to comfortably positive. It might be skill, or it might be luck. Thirty-six trades isn't enough for us to tell.

Two returns in the same ballpark, with opposite verdicts. The thing that separated them was the number of trades behind each one: 115 is enough to rule out luck here, and 36 is not. So when two strategies make about the same amount, trust the one with the most consistent track record.

A brief detour: the problem with comparing multiple strategies

This one is for anyone backtesting or paper-trading a whole batch of strategies at once.

Picture 1,000 monkeys, each flipping a coin to decide every trade. Pure chance guarantees a spread: most monkeys land near 50%, but a few, purely by luck, land way above it. Line all 1,000 up, rank them by win rate, and the monkey on top looks like a genius. But it's just the lucky tail of a random crowd, and a crowd that size always has a lucky tail.

Now swap the monkeys for variations of a strategy. You test a few lookback periods, a couple of symbols, a handful of timeframes, and without really noticing, you've quietly run hundreds of experiments. The best-looking one is partly just "best by luck," no matter how clever the underlying idea is. The more variants you try, the better your top performer looks, with no extra skill anywhere in the system. Sorting a leaderboard by raw performance is, quite literally, sorting for luck.

(This is one face of what people call overfitting, and it's why a "best performers" list with no significance check beside each row is close to useless.)

Across the whole family of EMA-crossover variants I tested, only 8 of the 52 I judged cleared the significance bar, fewer than one in six, and several of the biggest headline returns were sitting right in the pile that didn't. I go through all of them, receipts included, in a companion post.

Step 5: Skill is not the only thing to consider

TLDR: Passing the coin-flip test means the edge is probably real, but not all you need to think about when commiting to a strategy.

Say a strategy clears every bar above: real edge, plenty of trades, a Sharpe interval safely over zero. There's still one more question, and it's a personal one. Could you actually sit through it?

A strategy with a genuine edge can still put you through a drawdown (a long, deep stretch where your account is underwater) that lasts months and swallows a third of your balance before it recovers. On a spreadsheet, riding that out looks easy. At 2am, watching a real balance bleed, most people tap out at the worst possible moment, right before the recovery. If you can't sit through the bad stretch, the edge never gets the chance to pay you back.

So once a strategy passes the statistical test, look at its worst drawdown, both how deep it got and how long it lasted, and ask yourself whether you'd have held on.

Putting it together

So before you trust any track record, yours or a stranger's, walk down the same short ladder every time:

1. It made money, but how much of that is one lucky slice of history?
2. It won often enough, over enough trades, to rule out a coin flip?
3. It made money consistently, with a Sharpe interval whose lower bound stays above zero?
4. And if it passed all that, could you stomach its worst drawdown?

If a result can't clear step 3, you don't really have a strategy yet, you have a sample that's too small to mean anything. Your next move is to go gather more trades. For live or paper trading, that means setting up the strategy and letting it run. For backtests, that means either widening the test window or moving to a different asset where the strategy might generate more trades.

One last bit of honesty, since this whole post is about not fooling yourself. Passing the coin-flip test tells you a pattern is unlikely to be chance. It does not promise you'll make money. Real trading has costs, and there are things even a good simulator doesn't model yet (funding payments between longs and shorts, for one). A backtest is in-sample by nature, judged on the same history it was drawn from, so treat a passing grade as a strong hypothesis about the future, never a guarantee.

If you want to see this test running on real numbers, the PerpForge leaderboard is public and free to browse, no signup. Every backtest on it carries its significance verdict right next to its returns, busts and coin-flips included. Looking is free. Testing your own idea is the fun part.

PerpForge is an educational simulator. No real money is traded, nothing here is financial advice, and it's meant for adults 18 and over.