I Simulated 2,486 Leveraged Entries on Real Candles. At 50x, Being Right Didn't Save You.

Every leverage argument I've ever had ends in vibes. Someone says 50x is fine with a tight stop, someone else says never above 3x, and neither of them has a number. So I stopped arguing and pulled the candles.

The setup is deliberately dumb, because dumb is testable. I took 120 days of daily candles for 11 USDT perpetuals straight off the Bybit public API — 6 April to 3 August 2026. Then I opened a position at every single daily close and held it for seven days. If any low inside that week reached my liquidation price, the trade counted as dead. Then I did the whole thing again short. That's 113 entries per pair per side, 2,486 positions at each leverage setting, and no cherry-picking of entries whatsoever.

Liquidation distance assumes a 0.5% maintenance margin, so 20x liquidates about 4.5% below entry, not 5%. Funding is not charged. Both choices make the results friendlier than reality.

The headline table

Median across the 11 pairs, share of entries liquidated within one week:

Look at the bottom two rows side by side. That's the finding I didn't expect to be so clean.

Direction stops mattering somewhere around 25x

This window was a bear drift. Every one of the 11 pairs finished lower than it started — BNB was the mildest at −3.2%, AVAX the ugliest at −29.4%, BTC in between at −9.2%. So shorts had the wind at their back for four straight months. They were on the right side of the whole move.

At 5x that advantage shows up exactly where you'd expect: zero median liquidations short versus 3.5% long. At 10x it's already gone — 18.6% versus 19.5%, a rounding error apart. By 50x the correct direction was liquidated 78.8% of the time and the wrong direction 84.1%. Being right for four months bought a five-point improvement on a number that was already a catastrophe.

That's the part traders get backwards. High leverage doesn't amplify your edge, it deletes your exposure to your edge. Your thesis needs weeks to play out; your liquidation distance is smaller than a normal Tuesday.

How big is a normal Tuesday?

The median pair in this sample had at least one day where price dropped 11.25% from open to low. Not over the window — inside one candle. ADA's worst was 14.13%, AVAX's 14.66%. Median 120-day peak-to-trough drawdown across the group was 39%.

Now put that next to the second column of the table. At 20x you're betting that 4.5% of adverse movement never happens in a week, on assets that routinely do triple that in a day. At 50x you're defending 1.5%. The spread on some of these pairs is a meaningful fraction of that.

Per-coin, and why "just trade majors" only half works

At 10x, coin choice is worth real money: BNB 5.3% versus WIF 38.1%. That's a seven-fold difference in how often you get carried out, from nothing but the ticker. At 50x the same two pairs read 77.0% and 92.0%. The gap compresses because at that distance you're no longer measuring the asset, you're measuring whether any wick happened at all. And a wick always happens.

So "trade majors, not memecoins" is genuinely good advice — at sane leverage. As a defence for a 50x position it's a rounding error.

What I'd been getting wrong

I used to think about leverage as a dial on position size. Wrong frame. It's a dial on how long you're allowed to be wrong. At 5x, a week of ordinary chop can't touch you and your thesis gets to breathe. At 20x the median entry in this sample didn't survive seven days. Not the bad entries. The median one.

The other thing this killed for me: the idea that a tight stop makes high leverage safe. It doesn't, it just changes who closes the trade. If your stop is inside normal daily range, you get stopped constantly and pay fees for the privilege. If it's outside normal daily range, at 50x it's behind your liquidation and does nothing at all. There's no leverage setting where the stop rescues the math.

Run your own pair through max leverage for a drawdown and pick the number from the drawdown you actually expect, not from the exchange's slider. Then check where that puts your liquidation price and whether a single bad candle reaches it.

My own answer after running this: nothing above 10x on anything held longer than a session, and 5x on anything with SUI's or WIF's volatility profile. That's not caution talking. That's the table.

Fair warnings about this data

One 120-day window in a downtrend is one sample, not a law. A four-month rally would flip the long and short columns and leave the 50x rows roughly where they are, which is the point. Daily candles also understate the damage — intraday wicks that reverse within the same day are invisible here, so real liquidation rates on the high-leverage rows are higher than what I measured. So is the effect of funding, which I left out entirely.

None of that rescues the numbers. It all pushes the same way.

The four numbers I check before setting leverage

1. My pair's worst single day in the last few months. If it's bigger than my liquidation distance, the leverage is wrong. Median here was 11.25% — that alone rules out everything above 10x.
2. Liquidation distance after maintenance margin, not the naive 1/leverage. At 50x it's 1.5%, not 2%. Liquidation calculator.
3. The drawdown I'd need to survive to be right. I put that into max leverage for drawdown and take whatever number it gives me, even when I don't like it.
4. How many of these I can lose in a row. A 59% liquidation rate isn't a bad week, it's ruin on a schedule. Risk of ruin prices that.

→ Liquidation price · → Max leverage for drawdown · → Volatility drag · → Learn the math