Does anything beat buy & hold on Pyth?
Every setup we tested on Pyth — ranked out-of-sample, corrected for multiple testing, and forward-tracked in public from the day this page published. The honest answer is the headline.
Beat buy-and-hold in both windows — but can't be told apart from selection luck.
Beat buy-and-hold in both the full window and out-of-sample but its OOS Sharpe 0.55 did not clear the 3.31 selection hurdle (best-of-N luck cannot be ruled out). Buy-and-hold benchmark: -43.8% CAGR over 3.8 years (-76.7% CAGR in the out-of-sample window).
Educational research from historical backtests — not investment advice. Past performance does not predict future results.
Pyth (PYTH): Beat Buy-and-Hold in Both Windows, Couldn't Beat the Selection Hurdle
Crypto sets a brutal baseline: Pyth compounds at -43.8% annualized just for holding, and it charges drawdowns near -60.2% for the privilege. Against that backdrop we ran 411 indicator configurations on PYTH. The best of them — Connors RSI on the daily timeframe — beat buy-and-hold in both the training and holdout windows, with a profitable out-of-sample trade profile across 55 trades. That is genuinely uncommon here. But 'best of 411' is exactly the phrase that should make you slow down, which is what the second paragraph is for.
When you pick the winner from 411 attempts, the result is partly signal and partly luck of the draw. Our hurdle corrects for that: with this many tries and 1.1 years of holdout data, an out-of-sample Sharpe needs to clear 3.31 before we trust it. This one landed at 0.55 — profitable, but statistically indistinguishable from the luckiest of hundreds of tries. Add that crypto structure keeps changing — new venues, new participants, shifting volatility regimes — and a pattern from the past has no obligation to repeat. Treat this as a research lead, not an edge.
Every figure above is computed from our own backtests — nothing is estimated or invented. Hypothetical results; not investment advice.
The setups that ranked, and what they are
Fully mechanical — no discretionary steps, no hidden filters. We name the indicator and publish its record; we keep the settings.
Connors RSI
What it is: An oscillator rule: it reads a bounded momentum gauge and takes exposure when it turns from an extreme. Long/flat only — no leverage, no shorting. The parameters we tested are not published.
How it was tested: signals evaluated at daily-bar close, the position changes on the NEXT bar, 0.08% cost per side. Same engine, same cost model, same out-of-sample split as every other number on this site.
Out-of-sample (last ~30% of the window, never used to pick this setup): Sharpe 0.55 · alpha +93.5% · 21 trades over 1.1 yrs.
Connors RSI-2
What it is: A mean-reversion rule: it fades stretched moves and exits as price reverts toward its average. Long/flat only — no leverage, no shorting. The parameters we tested are not published.
How it was tested: signals evaluated at daily-bar close, the position changes on the NEXT bar, 0.08% cost per side. Same engine, same cost model, same out-of-sample split as every other number on this site.
Out-of-sample (last ~30% of the window, never used to pick this setup): Sharpe 0.32 · alpha +78.5% · 19 trades over 1.1 yrs.
Fisher Center-of-Gravity
What it is: An oscillator rule: it reads a bounded momentum gauge and takes exposure when it turns from an extreme. Long/flat only — no leverage, no shorting. The parameters we tested are not published.
How it was tested: signals evaluated at daily-bar close, the position changes on the NEXT bar, 0.08% cost per side. Same engine, same cost model, same out-of-sample split as every other number on this site.
Out-of-sample (last ~30% of the window, never used to pick this setup): Sharpe 0.06 · alpha +70.4% · 28 trades over 1.1 yrs.
Since publication — including if it loses
Buy-and-hold is ahead so far — we publish that too. That is the point. Marked to market nightly from real prices, rules frozen at publication, as of 2026-09-11. Currently LONG.
We tested 411 setups (indicator × parameters × timeframe) on Pyth. Only setups with ≥30 trades qualify (202 did). Setups are ranked by out-of-sample Sharpe — the last ~30% of history, which standard-parameter rules never saw during selection. Because picking the best of 411 tries mines even the holdout, the VALIDATED verdict additionally requires the top setup’s OOS Sharpe to clear a selection hurdle of 3.31 (√(2 ln N)/√T) AND positive alpha in both windows. Of the eligible setups, 99.0% had positive out-of-sample alpha (median OOS Sharpe -1.32) — the table below is truncated, but this summary covers all of them. How we test: methodology. The scoring contract is published in full; the parameters of each individual setup are not.
Top 20 of 202 eligible setups
Ranked by out-of-sample Sharpe. Full + out-of-sample columns, costs included. Hypothetical.
| # | Setup | TF | Total ret | Sharpe | Max DD | Win | Trades | α vs B&H | OOS Sharpe | OOS α | OOS trades |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Connors RSI | Daily | +217.4% | 0.71 | -60.2% | 58.2% | 55 | +79.7% | 0.55 | +93.5% | 21 |
| 2 | Connors RSI-2 | Daily | +156.3% | 0.64 | -55.6% | 57.7% | 52 | +72.2% | 0.32 | +78.5% | 19 |
| 3 | Fisher Center-of-Gravity | Daily | +79.2% | 0.51 | -72.2% | 48.9% | 90 | +60.6% | 0.06 | +70.4% | 28 |
| 4 | Fractal Adaptive MA | Daily | -36.7% | 0.03 | -70.8% | 32.6% | 95 | +32.4% | -0.0 | +68.2% | 26 |
| 5 | FRAMA 100 Trend | Daily | -64.5% | -0.31 | -80.7% | 34.2% | 79 | +19.8% | -0.03 | +68.0% | 21 |
| 6 | FRAMA 30 Trend | Daily | -59.3% | -0.16 | -81.4% | 36.9% | 84 | +22.6% | -0.06 | +65.9% | 21 |
| 7 | Lorentzian Classification | Daily | -58.5% | -0.17 | -84.0% | 47.6% | 126 | +23.0% | -0.07 | +59.2% | 51 |
| 8 | Stochastic RSI | Daily | -30.3% | 0.18 | -70.7% | 46.7% | 30 | +34.7% | -0.17 | +59.8% | 11 |
| 9 | Median MA | Daily | -27.2% | 0.04 | -74.4% | 41.9% | 43 | +35.8% | -0.23 | +66.4% | 10 |
| 10 | Elder Impulse | Daily | -56.1% | -0.28 | -74.9% | 38.9% | 90 | +24.2% | -0.23 | +62.7% | 25 |
| 11 | Chande Forecast Osc. | Daily | -73.2% | -0.27 | -87.5% | 42.0% | 100 | +14.3% | -0.29 | +50.9% | 25 |
| 12 | Stochastic Fast (5,3) | Daily | -82.4% | -0.37 | -92.7% | 36.2% | 127 | +6.9% | -0.32 | +45.5% | 36 |
| 13 | Jurik MA (approx.) | Daily | -77.1% | -0.46 | -86.8% | 36.9% | 84 | +11.4% | -0.33 | +58.1% | 18 |
| 14 | Williams %R (7) | Daily | -82.4% | -0.5 | -91.5% | 39.1% | 92 | +6.9% | -0.37 | +53.6% | 24 |
| 15 | MBFX Timing | Daily | -82.4% | -0.5 | -91.5% | 39.1% | 92 | +6.9% | -0.37 | +53.6% | 24 |
| 16 | Bollinger 10 (x1.5) Break | Daily | -63.1% | -0.97 | -70.5% | 45.0% | 60 | +20.6% | -0.38 | +68.6% | 13 |
| 17 | FRAMA 200 Trend | Daily | -83.2% | -0.74 | -87.3% | 34.8% | 69 | +6.1% | -0.43 | +47.7% | 25 |
| 18 | Arnaud Legoux MA | Daily | -82.0% | -0.49 | -92.1% | 31.5% | 89 | +7.2% | -0.5 | +50.4% | 24 |
| 19 | Schaff Trend Cycle | Daily | +109.8% | 0.54 | -40.6% | 48.4% | 31 | +65.6% | -0.51 | +57.3% | 11 |
| 20 | Detrended Price Osc. | Daily | +51.4% | 0.48 | -67.8% | 52.9% | 102 | +55.5% | -0.51 | +39.2% | 28 |
Hypothetical backtests with 0.08%/side costs. Not investment advice — see the full disclaimer.
These are historical backtests of mechanical rules. They are educational research, not investment advice, not a recommendation, and not tailored to you. Educational information only — not investment advice. Hypothetical backtested results; past performance does not guarantee future results. Trading involves risk of loss.