Strategy tester & historical replay

Rules are replayed against stored daily closes — the same history the scores use. Results describe the past only. A rule that worked across one five-year window can fail completely in the next one.

Entry rules — buy when all checked conditions are true

Hold each position for30 days

Historical result

2021-08-042023-10-12
Trades
16
Win rate
56.3%
Average return
4.29%
Median return
0.54%
Best trade
53.5%
Worst trade
-26.5%
Compounded
52.7%
Buy & hold
-25.7%

The rule was in the market 480 days (60% of the window). Compare the compounded figure against buy & hold before concluding anything — fewer days exposed is worth something on its own. Fees, slippage and taxes are not modelled.

EntryExitReturn
2023-08-252023-09-24+0.77%
2023-07-252023-08-24-10.48%
2023-06-112023-07-11+18.08%
2023-05-112023-06-10-4.19%
2023-02-122023-03-14+13.64%
2022-12-272023-01-26+37.80%
2022-11-092022-12-09+7.79%
2022-09-242022-10-24+2.15%
2022-08-192022-09-18-6.80%
2022-06-132022-07-13-9.98%
2022-05-112022-06-10+0.31%
2022-04-102022-05-10-26.48%
2022-02-202022-03-22+10.39%
2022-01-052022-02-04-4.19%
2021-11-222021-12-22-13.63%
2021-09-202021-10-20+53.50%

Signal power by regime — did the score predict anything?

2021-08-142023-10-12 · 790 scored days

Every past day is re-scored using only the data available on that day, then paired with what actually happened next. Days are split into a bull regime (above a rising 200-day average) and a bear regime. A signal is only useful if high scores were genuinely followed by better returns — and if that holds in both regimes, not just the one where everything went up.

Bull regime (above a rising 200-day average)

284 days
Rank correlation (IC)
-0.04
High minus low
Regime baseline
-3%
Score bandDaysAvg 30dUp rate
Low (0-40)0
Mid (40-60)71-2.75%28%
High (60-100)213-3.06%32%

inconclusiveToo few days in the outer score bands to compare.

Bear regime (below or under a falling 200-day average)

506 days
Rank correlation (IC)
-0.02
High minus low
+2.4pp
Regime baseline
+0.4%
Score bandDaysAvg 30dUp rate
Low (0-40)44-2.22%59%
Mid (40-60)316+0.80%53%
High (60-100)116+0.15%43%

inconclusiveHigh scores were followed by 2.4pp higher 30-day returns than low scores — but the gap is small enough to be noise.

Caveats worth keeping in front of you: forward windows overlap day to day, so the effective sample is far smaller than the day count suggests; one asset over five years is a handful of real market cycles, not a large sample; and only the price-derived part of the score is tested here, because social, derivatives and on-chain feeds only have history from the day scanning started.

Historical replay

On 2023-10-12, bitcoin closed at $26,750.51 and the tape read Neutral.

RSI(14)
46.8
50-day avg
$26,622.93
200-day avg
$28,025.79
Drawdown from high
-60.4%
30d change (before)
3.5%
Volatility (annualised)
25.2%
What happened next: 30d
still ahead
What happened next: 90d
still ahead
  • Up 3.5% over 30 days
  • 60% below the running high
  • Realised volatility 27%