Next: a forward test runs daily until January. Does the live code run without look-ahead? Do 80% intervals hold? Only then does it go near our risk card. Not financial advice. Research log. #Bitcoin #crypto #volatility

Fine print: Edge vs GARCH is clear before the spot ETFs, a tie after. At 50–100 days the 250-day average ties MSM. And an old finding of mine, "calm begets calm", turned out to be plain volatility clustering.

At 1 day NATR ties MSM. It sees intraday highs and lows. MSM sees one return a day. ATR tells you how wild the last two days were. Not how wild the next month will be.

Round 3: NATR. ATR(14) on 4h candles over price, the "volatility" most dashboards show. Given its own scale, refitted the same way. 20 days ahead: loses to MSM in 9 of 9 coins (t 4.6). GARCH and the plain average beat it too.

What the edge is: tracking slow changes in the volatility level. Strip the level out and the edge goes. It does that even in a simulated world where MSM is the true model. So "level or dynamics" was a false choice.

Round 2: every model refitted every 125 days. MSM still beats GARCH at every horizon, 1 to 100 days. t 2.7 at 20 days, ahead in 8 of 9 coins. The stale level had inflated the edge, not created it.

I read it as "MSM only won because it forgets the old level more slowly". Published "no multi-scale edge". Too quick.

Round 1, parameters fixed after training: MSM beat GARCH (t 2.8). But the plain 250-day average beat everyone. Training-years volatility was 2–4x the test years'. Crypto has calmed down a lot since 2021.

The race, pre-registered: MSM (multifractal) vs GARCH(1,1) vs HAR vs a plain 250-day average. 9 Binance coins, forecasting volatility 1 to 100 days ahead. Main horizon 20 days, |t| ≥ 2 to call a winner.

Nine days ago I said Bitcoin isn't multifractal. Then I raced a multifractal volatility model against GARCH on 9 coins. It won. Just not for the reason the papers give

BTC's extreme hourly moves got rarer after Jan 2024 ETFs. Ran the control: 7 no-ETF coins thinned the same way. Alts moved FIRST — Q3 2023, after BlackRock filed. BTC followed in Q4. Not the ETFs. The market turned. reddit.com/r/algotradingcryp…

Next (November): not timing the market doesn't mean it can't pick coins. Do BUY coins beat the market over the same 24h? Rules are written down. I'll post it whichever way it goes. Not financial advice. Research log. #crypto #Bitcoin #quant #algotrading

What I take from it: • the system describes the last two days well; it isn't a forecast • a fast reaction isn't skill if the input contains the event • run the known-answer test first. Mine failing taught me more than a pass

So I locked a stronger check (AR ±0.4, SNR 6.3) before touching real data. Real data, 2022–2026, 145k coin-dates: +1.2 bp, SNR 0.23. Hurst across coins: median 0.515, 80% between 0.44 and 0.59. They barely differ. Turned it off.

Test 3: a bias-corrected Hurst exponent. Do coins with H>0.5 keep trending more over 3 days? First the known-answer check, and it failed twice. Even knowing each coin's true type, an AR ±0.2 trend difference only reaches SNR 3.0. The design was at its limit.

Test 2: four BTC shocks since August (>3 sd in 4h, all up). Share of the signal change made in the first turn after the shock: 66%, 100%, 95%, 100%. Then nothing new. The indicators contain the shock bar, so they jump. Fast, but for free.

A trap: if the market has momentum, a signal that copies the past "predicts" the future for free. So hold the past 48h fixed and ask what's left about the next 24h. SNR −0.49. Top vs bottom fifth of BUY share: next-day market differs by 0.001%.

Test 1: BUY share vs the market, from 48h before the signal to 48h after. Past 48h: correlation 0.17–0.28 at every lag, outside the noise band. Next 48h: about zero. It remembers. It doesn't foresee.

My signal system rates ~470 Binance coins every 4 hours. It looks smart on a dashboard. Does it know anything the price hasn't already shown? Three pre-registered tests this week. Same answer three times

Live card, updated every 10 minutes: bitdataview.com/ Full write-up of the two rejections behind it is on Reddit reddit.com/r/algotradingcryp… #Bitcoin #crypto #orderflow #quant