24/7 algo trader | alpha discovery crypto/stock/algo trading futures/options equity curve FIRST. collab: m3quantlab@gmail.com
- Tweets 27
- Following 29
- Followers 5
- Likes 0
Same stock list. Three totally different books.
Equal (1/N): fair, small-cap tilt — but you sell winners / buy losers every rebalance.
Cap (∝ MV): cheap, huge capacity — megacaps dilute the factor.
1
1
Factor (∝ g(S)): highest ceiling — fat tails concentrate unless you winsorize.
You’re not picking a formula. You’re picking which cost you can live with.
#Quant #PortfolioConstruction #M3QuantLab
1
ARIMA only lets d be an integer. Some series need a fraction.
ARFIMA: fractional d, hyperbolic ACF, long memory.
Short horizon: AR ≈ ARFIMA. Longer horizon: ARFIMA pulls ahead.
Cost: d is noisy in samples. No long memory? Skip it.
#Statistics #Quant #TimeSeries #M3QuantLab
6
Correlated factors lie about whose IC is real.
Value ↔ size. Momentum ↔ vol. Without orthogonalization you cannot tell signal from hitchhiking.
Gram–Schmidt:
1) G1 = F1
2) G2 = F2 − b·G1 (regress, keep residual)
3) G3 = F3 − b1·G1 − b2·G2
1
8
…same as QR → take Q
Cost: the cleaned factor is a residual, not the named style. Refresh when regimes shift.
#TradingFactors #Quant #FactorInvesting #M3QuantLab
10
Raw factors lie when size and sector sit inside them.
P/E sorts → bank books, not value books.
Vol sorts → small-cap books, not low-vol books.
Neutralize size (regress on ln mkt cap) + industry (dummies or within-industry z). Refresh the coeffs. Then trust the IC.
8
Factor extremes can fake a working IC. One name at 500 vs a 0–20 cloud can sell you a backtest that won’t survive live.
Delete · winsorize · transform — pick before you trust the IC.
Full note on Patreon (paid).
11
Capital source ≠ wallet size.
Own: flexible but capacity-capped.
Client: redemptions hit when the book is already worst.
Institutional: hard DD/liquidity boxes (I once saw 0.2% daily NAV rewrite a mandate).
Full note on Patreon (paid).
11
Situational Awareness: ~439% H1 after fees. Concentrated AI hardware, high leverage, transparent 13F book.
Good thesis != good strategy. Concentration + leverage-liquidity mismatch ends games before the thesis does.
Full note on Patreon (paid).
9
BTC crypto-native derivatives OI mix (Glassnode · CME out):
Options share: 26.5% (2021-12-31)→ 50.9% (Aug 23)→ 53.1% (Sep 17)
Perps 46.7% / dated futures 2.4% (as-of Aug 23)
Dated % of futures book: 36.5%→ 4.9% (2021-01-01 → Aug 23)
Structure — not a price call. Own series.
15
Payward (Kraken parent): US onchain perps via Hyperliquid HIP-3 (reg. pending).
NinjaTrader Clearing (FCM) · Bitnomial DCM/DCO + HIP-3 · Hyperliquid book.
Allowlisted by BOTH — not open HL for all US users.
No launch date / assets / leverage / fees.
CoinDesk · BW (Sep 16)
12
Senate CLARITY (H.R. 3633) cloture on motion to proceed FAILED.
Vote 234 · Sep 15 ~2:19 PM ET
49–50–1 (needed 60)
BTC (our yfinance hourly):
13:00 ET close ≈ $76.9k
Vote-hour bar low ≈ $74.9k → close ≈ $75.9k
Source: senate.gov roll call 234
8
Momentum isn't just up a lot. Continuity asks grind vs gap.
Same-direction day share + normalized run length → path-quality score that sizes SPY. Jump tapes chop; continuous tapes carry.
Direction without path quality is half a signal.
5
BTC 25Δ skew (Deribit/Glassnode) — options, not spot Fib.
1w flipped call-prem Aug 19; 1m Aug 21 — first 1m >0 after a year of put-prem.
Into Sep 16 FOMC (Sep 14): 1w +0.50 · 1m −0.53 · 3m/6m put-rich.
Live book: FOMC week slightly call-rich; Dec put-rich.
Own rebuild. Not FA.
6
Fed week · BTC into Sep 16 FOMC.
Fib (our OHLC): Jun low $58,076 → Aug high $81,347
23.6% $75,855 · 38.2% $72,457 · 50% $69,711 · 61.8% $66,965
Spot Sep 14: $77,665 — still above 23.6%.
FedWatch high-80s / Polymarket high-70s for +25 bp.
Own series. Not FA.
4
US spot BTC ETF net flows (Farside):
Sep 8: −$46.6M
Sep 9: −$120.2M
Sep 10: −$282.7M (ARKB led)
3-day: −$449.5M
Sep 11: +$6.0M
Sep MTD still ~+$247M
Not a regime flip — redemption streak stalled after CPI. Demand delayed, not abandoned.
4
US spot BTC ETF net flows (Farside):
Sep 3: +$730.8m
Sep 4: +$174.6m (−76%)
Sep 4: only IBIT + FBTC still printed. Everyone else $0.
Not sentiment flipped — the scheduled bid went quiet. Phantom liquidity if you trade ETF flow like a wall.
9
Alt-side leverage flipped above BTC.
Coinalyze (~8:01 PM ET 2026-09-06):
Perp OI $59.5B · BTC $24.0B (40.3%) · Non-BTC $35.5B (59.7%)
ETH $13.8B · HYPE $2.9B · SOL $2.8B · ZEC $2.0B · XRP $1.3B · DOGE $648M
Leverage off BTC into thinner books ≠ signal. Watch funding + BTC.D.
75
Extreme funding = crowding meter, not a crystal ball.
Deribit BTC-PERP (2026-09-06): ~0% ann.; 30d MA ~4.0%
1y P10 −0.9% / P90 +8.8%
After crowded longs (n=164): +0.62% 1d / +1.51% 7d vs +0.09%/+0.60% baseline — not a clean fade.
OI: Deribit ~$1.7B; OKX swap ~$2.1B (−18%/99d).
6
