I walk-forward tested bull, range, and bear. All three lost money

The short version

A regular backtest trains and tests on the same stretch of history, so it's always a little bit cheating. Walk-forward testing is the honest version: re-fit on a rolling window, trade the next unseen chunk, slide forward, repeat. Every segment was genuinely out-of-sample at the moment it traded. I ran this across 11 pairs, split the results into bull, range, and bear regimes, and expected at least one regime to come out ahead. None did.

The regime split

16,262 walk-forward trades total, split by the market regime each trade fell into:

Read the win rate column again. 85.1% in bull. 82.1% in range. 80.8% in bear. Every single regime won more than four trades out of five. And every single regime still has a profit factor under 1.0 -- meaning gross losses outweighed gross wins -- and a negative dollar total. Combined: -$1,419.25 across 16,262 trades.

Per-pair, it's worse than the averages suggest

Eleven pairs went into the walk-forward run: BTC, ETH, SOL, BNB, XRP, ADA, MEGAUSDT, NAORISUSDT, EWJUSDT, GOOGL, and MSFT (the last three are tokenized stock perps, not crypto). Only two of the eleven finished net positive out-of-sample:

EWJUSDT's +$4.15 came from 25 trades -- too small a sample to call a working edge, just a coin flip that landed right. NAORISUSDT's +$41.17 across 1,798 trades is the only result here with enough volume to take seriously, and I already wrote about why I don't fully trust it either: its own quarter-by-quarter profit factor dropped below 1.0 in the most recent quarter I measured.

MEGAUSDT is the one that stings. It's the pair I've used in half a dozen posts on this blog as the textbook case of a DCA-propped, high-win-rate illusion -- 95%+ win rate on the in-sample deep test, real profit, but a max adverse excursion north of -14% hiding under the hood. Here, on genuinely unseen walk-forward data, it still wins 85.6% of trades. And it still loses money. The in-sample version looked better specifically because it had already seen the data it was "tested" on.

Why winning 4 out of 5 trades wasn't enough

Profit factor is gross profit divided by gross loss. A profit factor of 0.63, like the bear regime posted, means for every $1.00 in winning trades, the same bucket of trades lost $1.59. Win rate never shows up in that ratio at all. You can stack an 85% win rate on top of a 0.86 profit factor without any contradiction -- it just means the 15% of losing trades were, on average, large enough to swallow most of what the 85% of winners made.

That's the mechanical reason a regime with an 80%+ win rate can still bleed money every time: the size of the loss matters as much as how often it happens, and nothing about "I won most of my trades" tells you anything about size.

What I'm actually doing with this

This run doesn't kill the strategy family outright -- NAORISUSDT's 1,798-trade result is thin evidence of something, not nothing. But it does kill any claim that this approach has a regime where it reliably works. Bull didn't save it. Range didn't save it. Bear, predictably, didn't save it either. When I see a backtest claiming a clean edge in exactly one regime from now on, the first thing I check is whether someone ran the other two regimes and quietly left them out.

Related calculators

If a win-rate-heavy pitch is why you're here, these tools separate the number that sounds good from the number that matters:

• Risk of ruin calculator — see how win rate and risk-per-trade combine into real survival odds
• DCA survival calculator — test how deep a "winning" averaging strategy can still get hurt
• Drawdown calculator — the number that erases a high win rate fastest
• Position size calculator — size losses so one bad regime doesn't wipe five good ones
• RektCalc Learn hub — liquidation, funding rates, margin modes and more, explained plainly