Research Ideas & Findings

Working thesis

Crypto has no fundamental anchor — no cash flows, dividends, or fair value. Price is flow- and reflexivity-driven, so momentum/trend persists and trends run far (no valuation ceiling or floor to stop them). Mean-reversion only exists (a) in the volatility dimension (IV/RV revert) and (b) as fleeting microstructure liquidity bounces — never as price-level reversion to a fundamental. Every finding below is consistent with this lens.

✓ VALIDATED works, with evidence · ◐ PARTIAL works but fragile/conditional · ✗ DEAD tested, no edge · ○ OPEN next experiment

Validated edges 5

✓ VALIDATEDMomentum

Trend-following (price) — the robust price edge

Go WITH the move. The direct corollary of 'no anchor → trends persist'.

Evidence: ETH band-breakout / HHV-LLV: +0.30–0.42% per trade EVERY year, ungated, with stop.

Takeaway: Design for it: cut losers fast, let winners run (trailing, not fixed TP). Low win-rate, large avgWin/avgLoss. The opposite profile of a fade.

✓ VALIDATEDEnvelope (Nik)

Nik envelope bot + ATR chop-filter (ETH 1h)

Fade the 7.94/5.43% EMA envelope, exit at the FAR band, only when ATR is below its rolling median.

Evidence: Combined +53R (2024+), ret/DD 1.43, positive every year from 2022. Without the ATR filter: −42R. Reproduced faithfully (fixed a day-reset bug).

Takeaway: The ATR chop-filter IS the strategy. Far-band exit = positive skew: 25% win-rate carried by huge ~16% winners, small k·ATR losers → survives stops.

✓ VALIDATEDVol premium

Vol-Risk-Premium / option selling

The mean-reversion that actually works in crypto lives in the VOL dimension, not price.

Evidence: IV > RV systematically; vol is mean-reverting (hard stylized fact). = the existing options book.

Takeaway: Sell premium when IV elevated. Anchor-independent (risk-transfer premium, not a valuation bet).

✓ VALIDATEDFilters (cross-algo)

ATR-low & ADX-weak gates (robust over ALL 5 algos)

Two filter findings that hold across HHV, Donchian, both SuperTrends, and SOL MA-cross.

Evidence: ATR_low > base > ATR_high everywhere. ADX_weak > base > ADX_strong everywhere.

Takeaway: Enter trend algos in the CALM / non-extended state. ADX-strong = trend already extended = breakout fails — this REVERSES the common 'strong ADX = good' intuition.

✓ VALIDATEDFilters (Nik)

Combined trend+vol filter study — DONE

Nik's MACD+ATR / ADX+ATR / MA+ATR combos, direction-aligned, full long/short + by-year, vs the existing matrix. → see the Trend+Vol Combos page.

Evidence: Robust cross-algo: ATR-low>base>ATR-high; ADX-weak>base>ADX-strong. MACD-alignment big on breakouts (HHV 5.7→10.9, Donch 3.1→6.6), not on MA-cross algos.

Takeaway: Direction-alignment helps direction-AGNOSTIC algos (channel breakouts), is redundant on algos that already embed direction (MA-cross). Long/short split exposes dead sides (SOL longs ~0).

Partial / fragile 4

◐ PARTIAL / FRAGILEMR (BTC)

BTC ER-low envelope fade — the one crypto price-MR

Fade a 4% band on BTC 15m ONLY in ER-low (chop). Exit at the MA (revert to middle).

Evidence: +16R, ret/DD 2.89, year-consistent. But ~19 trades/yr, and it's an island at exactly 15m·4% — every neighbour (other TF / depth) is negative.

Takeaway: Real but sparse & fragile. A complement sleeve, not a standalone earner.

◐ PARTIAL / FRAGILEFilters (Nik)

MACD direction-alignment — algo-specific

Take a trade only if its direction matches the MACD histogram sign.

Evidence: Big win on CHANNEL-breakout algos (HHV 5.7→10.9, Donchian 3.1→6.6); neutral/negative on MA-cross algos (SuperTrend, SOL).

Takeaway: MA-cross algos already embed a direction signal → MACD-alignment is redundant there. Direction-agnostic breakouts benefit. Not a universal add-on.

◐ PARTIAL / FRAGILEScalp (BTC)

Short-MA micro-scalp (SMA5, no stop)

Fade a fast 4% flush off a 3–5 bar mean — the microstructure overshoot bounce.

Evidence: Ungated-positive: +26R / ret-DD 3.52 at 5bps. But +19R/1.78 at 10bps and only +3R at 20bps round-trip.

Takeaway: Real microstructure edge, but it lives entirely in the 10–20bps execution-cost band → a fill-quality / latency game, not a strategy edge for us.

◐ PARTIAL / FRAGILEFilters

MA-direction gate (MA_up) — regime-biased

Trade only when the macro MA is rising.

Evidence: Improves every algo (HHV ret/DD 5.7→11.7) — but 2024+ was up-biased.

Takeaway: This is a macro long-bias bet, not a true filter. Suspect regime-dependence; needs a down-regime to trust.

Dead ends (with reasons) 4

✗ DEAD ENDMR

ETH / SOL / HYPE envelope fade (price-level)

Fade dips below an MA envelope, any timeframe / parameter.

Evidence: No robust edge anywhere; ETH ≈ 0 (2021-only), SOL/HYPE strongly negative. Ungated 3% fade = −271R.

Takeaway: No anchor → no restoring force. These markets are momentum-dominated intraday. Don't revisit.

✗ DEAD ENDStat-arb

BTC/ETH spread mean-reversion

Fade the deviation of the BTC↔ETH relationship (the spread has a 'relative' anchor).

Evidence: Non-stationary: ADF p=0.76, Hurst>0.5, half-life ~500d, VR≈1. β=1 ratio fade −345% / Sharpe −3.09. Hedge-β version is a 2021-22 artifact, negative 2025-26.

Takeaway: The two mega-caps ARE the market factors → worst possible stat-arb pair. Crypto stat-arb (if any) lives in sector pairs / market-residuals, not BTC/ETH.

✗ DEAD ENDFilters

ER as a filter for far-band-exit strategies

Apply our ER-low (chop) gate to Nik's envelope bot.

Evidence: ER-low gate: −59R vs +53R with Nik's ATR filter. ER+ATR combined: −12R.

Takeaway: The filter must match the exit. ER-low fits revert-to-MIDDLE fades (our BTC bot); it kills ride-to-FAR-BAND exits because the clean run to the far band never completes in chop.

✗ DEAD ENDOptions timing

Open a calendar INTO an active downtrend

Calendar = short gamma + long vega + positive theta; wants price near the strike.

Evidence: Feb-2026 crash: wide/OTM calendars −30% to −65% on debit during the down-move; +60% to +108% in the post-crash stabilization.

Takeaway: A calendar is a bet that the trend STOPS, not that it continues. Start it after vol rolls over, not mid-crash. Selling calls (with-trend) is the opposite, safer bet in a downtrend.

Open — next experiments 4

○ OPEN — NEXTMomentum

Cross-sectional momentum across the coin universe

Rank many coins, long the strongest / short the weakest. The untested BREADTH lever.

Evidence: Per-asset momentum edge already proven on ETH; breadth → smoother equity than any single-asset sleeve.

Takeaway: The real answer to 'not enough trades / no smooth equity'. Highest-priority open idea — structurally rests on the no-anchor thesis.

○ OPEN — NEXTCarry

Funding-rate carry (market-neutral)

Harvest structurally-positive perp funding (long spot / short perp, delta-hedged).

Evidence: Retail is structurally long-biased → funding skews positive. A flow premium, not a valuation bet.

Takeaway: Smooth, low-DD, anchor-independent — the most 'Sven-shaped' idea. Tail risk: basis blow-ups / exchange risk; different infra.

○ OPEN — NEXTFilters

Vol-CONTRACTING gate (vol-slope) vs vol-LEVEL

Gate on volatility falling (SMA5 < SMA20 of ATR), not just on its level.

Evidence: Conceptually orthogonal to ATR-level; vol-slope axis already built in the engine.

Takeaway: Possibly additive to the ATR-low finding — test whether 'vol contracting' beats 'vol low' as the entry gate.

○ OPEN — NEXTOverlay

S3+1 call-side as a covered-call overlay

Sell weekly 8% OTM calls + light 1x daily hedge, on BTC already held.

Evidence: 2026: +1.24% net / 4mo on full notional, 15/16 weeks green, worst week −0.00%. Hedge eats 62% of premium.

Takeaway: Tiny as a standalone (~4%/yr unlevered) but additive as an overlay on held BTC, and directionally aligned with a downtrend. The 1x (light) hedge is the only net-positive-on-Exec variant.

Living document — edit the IDEAS list in build_ideas_page.py to add or update. Each entry carries the evidence and the honest verdict, negatives included. Numbers are net of fees; short windows (2024+ engine / 16-week Tardis 2026) carry sample-size caveats.