My Backtest Made 11% A Month. It Was Just A Bet On Bitcoin.
10 October 2026
Trading Strategy Diary
What this is and is not: a historical backtest on free Binance data (spot daily 2024-01-11 → 2026-10-09; perpetual futures daily 2021-01 → 2026-10, with real funding prints). No positions, no real or paper money, no P&L. Nothing was built or deployed. It failed a pre-registered test. Nothing will trade it.
The question I asked
"Did we include the likes of Zcash and SUI?" No. At $23.5B and $4.6B, ZEC and SUI are 90–470x over the $50M micro-cap ceiling, so the size filter dropped them first. The micro-cap result (probe 8: no long-only edge in the 160) said nothing about large caps, so I ran the same method there.
The hindsight trap, with numbers
I tested two universes because "today's top 20" is badly biased.
- FIXED = today's top 20 by market cap (stablecoins, gold tokens, wrapped assets excluded). These are top 20 because they went up. Hindsight is built in.
- PIT (point-in-time) = each day, the top 20 by trailing 30-day volume, using only data known that day. 103 different names passed through it. This is the honest one.
| Buy & hold (2024-01-11 → 2026-10-09) | |
|---|---|
| FIXED top 20 (biased) | +50.1% |
| PIT top 20 (honest) | −76.9% |
That is about 127 points of hindsight. ZEC is the biggest survivorship example: +5,222% since Jan 2024, and it is #7 today because of that run. SUI is in today's top 20, but −15% since Jan 2024 and −80% from its peak. What the honest basket held while it was hot: WLD −93%, AVAX −84%, LINK −73%, TRUMP −70%.
32 tests, one survivor
I ran 32 cross-sectional signal tests across both universes. FIXED: nothing passes. PIT: one result — 30-day turnover long/short: long the most-traded quarter, short the least-traded, 30-day hold: +10.0% net per month, t=3.52 (Binance spot prices; the short side assumed via perps; funding costs not modelled at this stage).
One hit in 32 tests. That is the multiple-testing trap. It is also what luck looks like. I wrote the pass rules before I let this lead run a second test.
The rules, written down before the test
- 2021–23 out-of-sample: positive, with t ≥ 2.
- At least 4 of 6 calendar years positive.
Both must hold. This is the centre of the entry — the rules were written before the test ran, not after.
The result
Probe 12: Binance USDT perps, 653 contracts including 132 delisted, real funding prints for all 217 contracts that ever made the point-in-time top 20. Long top 5 / short bottom 5 by 7-day turnover, 30-day hold, 0.1% per side (fees + slippage combined).
| Window | Mean / month | t |
|---|---|---|
| 2024–26, funding off (replication) | +11.7% | 2.93 |
| 2024–26, funding on | +10.8% | 2.69 |
| 2021–23, out-of-sample | −5.1% | −1.33 |
| Full 2021–26 | +2.2% | 0.81 |
By calendar year (2021 → 2026): −13.3%, −4.9%, −6.8%, +12.3%, +8.9%, +12.9%.
Verdict: NOT CONFIRMED. Rule 1 failed outright (−5.1%, t = −1.33). Rule 2 failed 3/6. The headline "+11% a month" (formally +11.7%/month, 2024–26, funding off, replication) sounds like a trading-guru screenshot. Because the rules were written first, the test killed it with zero dollars at stake.
Funding wasn't the killer
Longs paid 0.16% per hold and shorts earned 0.17% per hold — funding nets to roughly zero over the whole window.
What the strategy actually was: long BTC and ETH 100% of the time (plus SOL and XRP), and short LINK, AVAX, FIL, LTC, AAVE. It made +6.0%/month in down months and lost 2.6%/month in up months. That is a Bitcoin-dominance bet that happened to fit the 2024–26 regime — a bet wearing a statistics costume.
Even the good years were fragile
Shifting the start day moved the 2024–26 result anywhere from +2.5% to +14.8% per month. Same window, same rules, different entry date. A real edge doesn't depend on which day of the month you start.
What's left standing
The trend filter (hold only above the 50-day MA) reliably cuts drawdowns — BTC max drawdown −53.0% → −31.4% — but does not reliably add return. It's a risk-control tool, not an edge. It gets one line and no more.
Two no-gos in crypto: micro-cap (probe 8) and large-cap (probes 11/12), both with the arithmetic shown. The process held: the rules were written before the test, the test ran, and it said no. That is what "the process you need to go through to make money" actually looks like — not a green curve, not a course pitch.
The next probe moves to US micro-cap equities, where the data is free, structured, and audited. SEC filings first. The local LLM writes the research notes, but every number in them has to trace back to a filing, and a script rejects any that don't.