Do Quantor's trading bots beat the market?

No. Across 264 out-of-sample walk-forward tests — 6 strategies, 11 major pairs, 2 timeframes, 2 cost levels — on real Binance data and after fees and slippage, zero of them showed a credible return edge that beats simply holding the asset (buy-and-hold). Quantor does not promise returns. What the same tests showed was lower drawdown: 108 of the 264 lost less in bad stretches. That is risk reduction, not extra profit.

What is a walk-forward test?

A walk-forward test repeatedly trains or calibrates a strategy on one slice of history and then evaluates it on the next, unseen slice — out-of-sample. Rolling that window forward across the whole dataset measures how the strategy would have performed on data it never got to fit, which is a far more honest gauge of real edge than a single in-sample backtest that can be curve-fit to look great.

Methodology · 6 min read

We ran 264 walk-forward tests on our own bots. None beat just holding.

Q: Can I reproduce Quantor's walk-forward results?

Yes. The test uses free public Binance kline data (no API key, no account) and runs as a single command in Quantor's open methodology. It fetches real history, runs each strategy out-of-sample after a realistic round-trip cost, and prints a verdict table with the alpha versus buy-and-hold for every combination. The 0-of-264 result is not a claim you have to trust — it is a number you can regenerate.

Quantor · Solo-founded · 2026-06-24 · ← All posts

We tested our own crypto strategies 264 different ways. After fees, not one of them reliably beat simply holding the coins. We're publishing that anyway — because the number that makes us look bad is exactly the number a trading tool should be willing to show you.

What we tested

A walk-forward analysis is the honest version of a backtest: train on one slice of history, evaluate on the next, unseen slice, then roll the window forward across the whole dataset. It measures how a strategy does on data it never got to fit — which is where curve-fit backtests fall apart.

We ran the full grid:

6 strategies (EMA cross, breakout, time-series momentum, our adaptive regime strategy, mean-reversion, MA-timing)
× 11 major pairs (BTC, ETH, SOL, BNB, XRP, ADA, AVAX, LINK, DOGE, DOT, TON)
× 2 timeframes (1h, 15m)
× 2 cost levels (~0.30% and ~0.60% round-trip)

That's 264 out-of-sample evaluations, on real public Binance data, after fees and slippage, each one benchmarked against buy-and-hold over the same window. The benchmark matters: a huge long-only return in a bull market is beta, not edge. Real edge means beating the benchmark — alpha = strategy return − buy-and-hold return.

The result: zero

Of the 264 combinations, the number that showed a credible return edge which also beats holding, after costs, was 0. Not "a small edge." Not "in some markets." Zero.

Most platforms in this industry run on cherry-picked backtests and "+340% APY" banners. If a trading tool can't show you the test that makes it look bad — ask why.

What the same 264 tests did show

The strategies weren't useless — they were just honest about what they do. In the published run below, 225 of the 264 reduced drawdown versus holding: they sat out some of the worst stretches and lost less when the market fell. They don't make you more — they can lose you less, on rules you can audit.

That's the whole honest shape of a rules-based, risk-gated bot: not extra profit, but a smaller hole and a smoother ride — and, crucially, a number you don't have to take on faith.

You can reproduce it yourself

This isn't a screenshot we're asking you to trust. The test runs on free public Binance data — no API key, no account — and prints a verdict table with the alpha-vs-buy-and-hold for every single combination. It fetches real history, runs each strategy out-of-sample after a realistic round-trip cost, and tallies how many show a credible edge. Run it on a falling week or a rising one; the shape holds. The 0-of-264 is regenerable, not rhetorical.

One command per cost level — this is exactly how the dataset below was produced:

quantor wf \
  --symbols BTC/USDT,ETH/USDT,SOL/USDT,BNB/USDT,XRP/USDT,ADA/USDT,AVAX/USDT,LINK/USDT,DOGE/USDT,DOT/USDT,TON/USDT \
  --tf 1h,15m --bars 1500 --folds 5 --feeBps 10 --slipBps 5  --emit run_low.csv
quantor wf  <same symbols/tf>  --feeBps 20 --slipBps 10 --emit run_high.csv
# 132 rows each → 264 combinations

Download the raw per-run data: walk-forward-264-2026-06-25.csv — all 264 rows: strategy, pair, timeframe, cost level, trades, win-rate, net return, profit factor, max drawdown, folds profitable, and the alpha vs buy-and-hold for each. Open it in any spreadsheet and check our arithmetic.

One honest detail you'll find in that file, because we won't bury it: on this run, 151 of the 264 beat buy-and-hold on the point estimate (column beats_buyhold). That looks like a win — it isn't. Zero of them did it consistently across the out-of-sample folds (column credible_edge), which is the only bar that separates a real edge from one lucky window. Point-estimate wins are noise; surviving the folds is signal — and nothing here survives. Numbers shift slightly each run as the rolling window moves; the shape — 0 credible edge, most combos cutting drawdown — does not.

So what are we actually selling?

Not a return — we don't promise one, and our own testing is why. A process: consistent rules, hard stop-losses, a regime filter that stands down in dangerous markets, withdraw-disabled keys, and a signed, verifiable record. You can run the whole thing in paper mode first — real strategy, zero money. Risk reduction stated plainly beats a profit number we couldn't honestly stand behind.

The honest pitch in one line: we don't beat the market on returns — we aim to lose less getting there, and we let you check every claim. See the methodology on /performance, why lower drawdown can matter more than higher returns, and whether crypto trading bots are a scam. Crypto trading carries substantial risk of loss — Quantor does not guarantee returns.