🔬 HHV/LLV ETH Trend-Catcher — Experiments

Explanation

baseline = the real algo (with stops) · after = the option structure (replaces it) · futures_hold = passive hold to the option's expiry, NOT the algo

long options — DTE 14d

long options — DTE 7d

Year by year — is it better every year?

Result: the baseline is the real algo (+241.5% 2024+, n=462, with its stops). The grey curve that looked negative before — futures_hold — is a passive hold of the future to the option's expiry (−81% at 7d): the algo exits in ~3h median, the passive hold sits through the whole 7/14d option life, so the gap between them is the algo's exit discipline.
Among the held-to-expiry options the convex, uncapped payoff wins: long_single is clearly best (14d +64% total, avg +0.66%/trade) despite a low win rate (~31%) — the one fat winner pays for the many theta losers. Butterflies cap exactly where the trend would pay → bad. Debit spreads sit in between. 14d > 7d (the move needs time to beat theta). But note: the active algo (+241%, trades every signal with stops) is a denser, different trade set (n=462) than the debounced option events (one position per DTE window) — same period, not the same trade count.
M2 vs Tardis: verified against the Deribit chain (tab 11) the M2 price ≈ Tardis mid at 7d/14d (ratio ≈ 1.00–1.03). So these long entries are priced about right — no large hidden bias either way. (Earlier we feared a ~2× overstatement; Tardis does not support that.)

Explanation

before = the algo's long future · after = the same call, exited a different way (hold / TP / at the algo signal)

Call exit rules — 14d

Call exit rules — 7d

Which strike? — ATM vs OTM (+1 / +2 / +3 on the $50 grid)

Strike result: ATM wins the total — 7d ATM +72.3% is the max; every step OTM gives up total (+68.7% / +58.9% / +48.4%). The leverage does rise OTM but only barely — ROI 1.11× → 1.18× (7d) while premium more than halves (3.14% → 1.35%): payoff shrinks almost in step with cost, no free lunch. MaxDD shrinks OTM (7d -21.5% → -12.0%) because less premium burns on losers.
Why: the algo exits after median ~7h, so the realised move is small — the ATM call has the highest delta (~0.5) and captures the most of a small move. The "more leverage at 1%" of OTM only pays on big moves, which the fast exit rarely sees. M2 caveat doubles down OTM: far-OTM is exactly where M2 is least reliable (tends cheap, tab 11) — so the OTM totals are more optimistic than the ATM ones.

Result — the exit rule decides everything, but calls-on-longs are the weak side of the book. A profit-target underperforms: it caps the few fat convex winners that carry the payoff, costing more than the theta it saves — every TP either loses or barely breaks even (7d TP 1.5× -1.1%, 14d TP 2× -74.7% at a -193.6% DD). Holding to expiry bleeds theta (14d +21.8% at maxDD -188.1%). Exiting the call when the algo exits the future is by far the best exit rule — but it does NOT beat the future: 7d +72.3% / 14d +46.8% vs the algo's +96.6%. It gives up return and only modestly trims the drawdown (maxDD -21.5% 7d / -20.3% 14d vs the future's -27.6%), so on return-per-drawdown it is roughly a wash on 7d (3.5 → 3.4) and worse on 14d (2.3). Win rate is only 16% (7d) / 25% (14d) — a long-only call on a 44%-win directional signal that exits in a median ~7h can't out-earn the linear future; it pays off on the rare fast move, not the median trade. Mechanic: short hold → little theta; winners keep intrinsic + residual time value; losers capped at the premium. The workhorse is the put-on-shorts side (tab 4), where the same structure roughly doubles return-per-drawdown.
Honest caveat (M2 + mid): the exit @ algo / TP marks reprice the call's residual extrinsic value on the M2 surface, at mid. Selling back a multi-day option costs bid/ask in reality, and M2 short-DTE is only ~1.0–1.2 accurate (tab 11) — both eat exactly that residual time value. So the direction (short exit ≫ hold-to-expiry / TP) and the ranking are robust, but the exit_sync level (+72.3%) is an upper bound — the 2026 Tardis-chain test (combined tab) validates the M2 level against the real Deribit mid. Longs/calls only here; shorts/puts are the mirror in tab 4.

Explanation

before = the algo's short future · after = the same put, exited a different way (hold / TP / at the algo signal)

Put exit rules — 14d

Put exit rules — 7d

Which strike? — ATM vs OTM (+1 / +2 / +3 below spot on the $50 grid)

Strike result (mirror of the call side): ATM wins the total — 7d ATM +110.9% is the max; every step OTM gives up total (+99.8% / +99.8% / +88.9%). The leverage rises OTM but only barely — ROI 1.13× → 1.25× (7d) while premium more than halves (3.19% → 1.40%): payoff shrinks almost in step with cost, no free lunch. MaxDD stays tiny across all strikes (7d -7.6% … -3.5%, ATM the shallowest) and win-rate rises OTM (7d 31% → 42%). Why: the algo exits after median ~7h → small realised move; the ATM put has the highest delta (~0.5) and captures most of a small drop. M2 caveat doubles down OTM: far-OTM is where M2 is least reliable (tends cheap, tab 11) — so the OTM totals are more optimistic than the ATM ones.

Result — this is the workhorse of the whole book. hold→expiry is catastrophic (14d -15.1% at maxDD -224.5%); TPs are mediocre; only exit @ algo works (7d +110.9% at maxDD -7.6%, 14d +107.3% at maxDD -6.5%). It does NOT beat the algo on return — the algo short future earns +144.9% vs the put's +110.9% (7d) / +107.3% (14d), giving up ~a quarter of the gain — but it collapses the drawdown: maxDD -19.3% → -7.6% (7d) / -6.5% (14d), a ~60% cut. On return-per-drawdown that lifts 7.5 → 15 (7d) / 16 (14d) — roughly 2× the algo. That risk-axis win, not a return win, is the whole case for the structure.
Why the put side beats the call side: the algo's short future is a stronger baseline than its long (+144.9% / ret-DD 7.5 vs the long's +96.6 / 3.5), shorts exit even faster, and the bought put's downside is capped — so the put book shows a tiny maxDD (-7.6%) where the call book sat above −20%. Win rate is still only 31% (7d) / 39% (14d) — it makes its money on the size of the moves it catches, not their frequency. Same mechanic: the short hold (~7h) pays almost no theta, winners keep the move, losers are capped at the premium.
Honest caveat (M2 + mid): the exit @ algo / TP marks reprice the put's residual extrinsic value on the M2 surface, at mid; a real resale pays bid/ask and M2 short-DTE is only ~1.0–1.2 accurate (tab 11) — both eat that residual time value. Direction (short exit ≫ hold/TP) and ranking are robust; the level (+110.9%) is an upper bound. This is the put/short mirror of the call/long tab 3.

Explanation

before = the real algo on ALL trades (futures, long+short) · after = call-on-longs + put-on-shorts (single + best debit spread W300), both exit @ algo, merged into one curve

Combined book vs the algo on ALL trades

Year by year — equity curve per calendar year (algo ALL vs combined, reset Jan 1)

Result — the option book trades return for a much smaller drawdown: lower CAGR, ~half the DD, higher return-per-drawdown. Every leg is anchored to the algo's actual futures fill (the breakout level it really traded), so the option and the future live on the same realised path — no entry-timing look-ahead. Merging calls-on-longs + puts-on-shorts into one book turns the algo's +241.5% at maxDD -33.6% (ret/DD 7.2) into +183.2% at maxDD -17.7% (7d, ret/DD 10.3) / +154.0% at -13.5% (14d, ret/DD 11.4). So it gives up ~a quarter of the return (+241.5% → +183.2%) but cuts the drawdown ~2× (-33.6% → -17.7%) and lifts return-per-drawdown from 7.2 to 10.3. The worst single trade drops from -3.5% to -2.7% (downside capped at the premium on every leg). Mechanic: the algo wins ~47% of trades with asymmetric P/L; swapping the future for an ATM option held to the algo's own exit caps each loser at a small theta bleed while winners keep their directional move — but a bought option also pays premium on every trade, which is why total return lands below the linear future. Win rate 24% (7d) / 33% (14d).
Year by year shows the same trade in every regime — the option book earns less in the trending years but always at a smaller DD ('24 algo +132.7%/DD -12.6% → 7d +105.0%/DD -9.5%; '25 algo +127.8%/DD -14.6% → 7d +84.7%/DD -9.1%). And 2026: the algo is -19.0% with a -33.6% intra-year drawdown; the option book also loses but far less — -6.5% 7d / -3.0% 14d at -17.7% / -13.5% DD — the capped downside shrinks the hole by ~half but does not turn it green.
Spreads in the picture (best width W300 = sell a wing $300 further OTM on each side): 7d +106.2% at maxDD -18.6% (ret/DD 5.7), 14d +45.3% at maxDD -16.3% (ret/DD 2.8). The spread is cheaper (the sold wing pays for part of the bought leg) but caps the upside and forces buying back residual extrinsic at exit, so it trails the single on both return (+183.2% → +106.2% at 7d) and return-per-drawdown (10.3 → 5.7; 14d 11.4 → 2.8). The single ATM combined leads; the spread's only edge is the lower absolute capital at risk per trade. Nothing removed — both widths are in the table above. Per-side spread detail (both widths, long & short) lives in tab 6.
Honest caveat (M2 + mid): both legs reprice residual extrinsic on the M2 surface, and both the M2 entry and the exit value are mid — a real resale pays bid/ask (median spread ~3.9%) and short-DTE M2 is only ~1.0–1.2 accurate (tab 11), so the level is optimistic. The 2026 ultimate test below re-prices this exact book on the real Tardis Deribit chain and the M2 price level holds (M2 vs Tardis within a couple of points). Direction and ranking are robust; the absolute totals are an upper bound. Per-side detail: call/long in tab 3, put/short in tab 4.

2026 ultimate test — Model 2 vs the real Tardis Deribit chain

Result — M2 matches the real chain, and the honest 2026 book loses less than the algo. On the identical matched book the two price sources land within a few points: 7d M2 -3.7% vs Tardis -9.2% (maxDD -13.7% vs -14.9%); 14d M2 -6.6% vs Tardis -16.1% (-14.3% vs -19.6%). Entry‑premium ratio μ 0.95 (7d) / 0.99 (14d) — M2 sits ~1–5% under the real mid, so live is a touch worse, never richer. Both the pricing check AND the strategy result now agree with the rest of the page: on this losing 2026 window the algo future is -21.3% at maxDD -33.6%, while the real Tardis option book is only -9.2% (7d) / -16.1% (14d) — it still loses, but far less than the algo, exactly the "cut the drawdown" story. The fake +60% / 85%-win was pure entry-timing look-ahead and is gone.
Honest caveats: Tardis is hourly only — entry & exit are single hourly mids (a real resale still pays bid/ask, so live is a touch worse). 7d/14d snap to the nearest listed expiry (realised DTE ~6.8d / 13.9d). And this is one ~4‑month regime — it calibrates the M2 level for 2026, it is not a full out‑of‑sample equity.

Explanation

Debit spread instead of the single call (same exit rule)

Debit-spread result: the exit rule lifts the spread too — held to expiry the spreads trail the single call and carry deep drawdowns (W150 14d hold -4.2% at maxDD -70.8%), but exit @ algo turns every width into a small-DD winner. Cost is lower as expected (7d W150 1.79% vs single 3.14%) — but total is lower and ROI is no better (1.07× vs 1.11×): on the buy-back you must repurchase the short call's residual time value, which eats exactly the extrinsic the single call keeps whole. The cost saving and the capped payoff roughly cancel. Where the spread does shine: drawdown. 7d W150 maxDD -9.6% (worst -0.7%) vs single -21.5%. On return-per-drawdown it flips by tenor: the single leads at 7d (3 vs W150 3) while the narrow W150 edges ahead at 14d (1 vs single 2). Smaller absolute profit, even smoother curve. M2 caveat doubles: two legs reprice residual extrinsic, two legs pay bid/ask on resale.

And butterflies? — structure risk/reward under the same exit rule

Butterfly result: the paper risk/reward does not materialise under an early exit — the fly has the lowest ROI of all structures (7d 1.01× / 1.06× vs debit 1.07× and single 1.11×). Mechanic: the peak (W − debit at the body) only forms near expiry; at ~7h with most of the life left the fly is still rounded, so we never collect what we paid for — whereas the single call keeps its residual extrinsic whole. What the fly does give: a very high win-rate (14d W150 47%) and a tiny drawdown (14d W150 maxDD -2.0%, worst -0.5%) — the cheapest, lowest-return corner. On Ret/DD it splits by tenor: the single call leads at 7d (3), the narrow debit spread at 14d (1) — the butterfly is last on both. For early-exit risk/reward the debit spread beats the butterfly; the fly only pays off if held near expiry, which the ~7h algo exit never does. M2 caveat triples here: three legs reprice residual extrinsic, three legs pay bid/ask on resale, and short-DTE M2 error compounds across the structure.

Explanation

baseline = the real algo (with stops) · after = the sold credit spread (defined risk, replaces the future) · futures_hold = passive hold to expiry, NOT the algo

3a · with the trend (bull→put-CS, bear→call-CS)

credit with-trend — DTE 7d

credit with-trend — DTE 14d

3b · fade / against the breakout (bull→call-CS, bear→put-CS)

credit fade — DTE 7d

credit fade — DTE 14d

Year by year

Result: the baseline is the real algo (+241.5% 2024+, n=462). The grey futures_hold is the passive hold to the option's expiry, not the algo — that is why it sits negative. Against the algo every credit-spread variant gives up the trend: with-trend selling loses across the board (high win rate 63–81%, but the losers run far — the breakout drives straight into the sold side). Fade is slightly positive at 7d (call_cs_off50_W150 +11%, win 75%) — confirming the post-breakout pullback; at 14d it flips negative (the trend continues). Neither approaches the active algo.
M2 vs Tardis: at 7d/14d M2 ≈ Tardis mid (ratio ≈ 1.00–1.03, tab 11), and a credit spread nets the bought leg against the sold leg so most of any residual bias cancels. The collected credit here is therefore close to realistic — the old "selling is ~2× too rich" worry is not supported by the chain. The read stands on its own: with-trend selling into a breakout is structurally bad; only the short fade has any (fragile) substance.

Explanation

before = long only over the overnight window · after = long + sold call (call on top)

Combined — future + sold call (after) vs future only (before)

Structure	n	Total %	Avg %/trade	Win %	Max DD %	Worst %	Best %
future only (before) ⟵ baseline	522	+94.85	+0.182	49	-30.08	-7.13	+9.09
atm	522	+206.92	+0.396	67	-20.43	-5.95	+3.38
atm+1	522	+187.55	+0.359	55	-25.34	-6.34	+5.37
atm+2	522	+180.42	+0.346	52	-26.63	-6.50	+8.63

Call overlay alone (premium collected − capped P/L)

Structure	n	Total %	Avg %/trade	Win %	Max DD %	Worst %	Best %
atm	522	+112.07	+0.215	82	-34.41	-7.14	+1.90
atm+1	522	+92.71	+0.178	92	-14.27	-5.10	+1.42
atm+2	522	+85.58	+0.164	97	-5.68	-2.86	+0.99

Year by year — is it better every year?

combined — total % per year

Year	future only	atm	atm+1	atm+2
2021	+30.54	+127.31	+111.32	+100.66
2022	+6.30	+43.01	+26.13	+16.55
2023	+42.85	+28.10	+40.02	+43.14
2024	+29.38	+27.90	+30.00	+33.99
2025	+5.47	-6.38	-1.47	+5.06
2026	-19.70	-13.00	-18.45	-18.97
ALL	+94.84	+206.92	+187.55	+180.42
Δ vs future only (variant − base, per year)
Δ 2021	+0.00	+96.76	+80.78	+70.12
Δ 2022	+0.00	+36.71	+19.84	+10.26
Δ 2023	+0.00	-14.75	-2.83	+0.29
Δ 2024	+0.00	-1.48	+0.62	+4.61
Δ 2025	+0.00	-11.85	-6.94	-0.41
Δ 2026	+0.00	+6.69	+1.24	+0.72

Result (call on longs): the sold call lifts the long's overnight return (combined_atm +206.9% vs long-only +94.9% over 522 nights) and cuts the maxDD (−20.4% vs −30.1%). Further OTM = less premium, less upside given up: the overlay alone gets smoother (ATM+2 win 97%, overlay maxDD −5.7%) but adds less; ATM collects/caps the most — its overlay worst (−7.1%) are strong up-gap nights where the call runs fully ITM. Year-by-year: the edge is front-loaded in 2021–22; ATM goes negative vs long-only in 2023/24/25, only ATM+2 stays roughly neutral every year — the headline total is carried by the early bull years, it is not better every year.
M2 vs Tardis: overnight ATM credit ratio ≈ 1.0–1.2 (tab 11), recomputed PnL on Tardis mid ≈ M2 (call ATM M2 +6.93 vs Tardis +6.97); far-OTM calls if anything cheap. Level roughly real, ~4-month overlap.

before = short only over the overnight window · after = short + sold put (put on top)

Combined — future + sold put (after) vs future only (before)

Structure	n	Total %	Avg %/trade	Win %	Max DD %	Worst %	Best %
future only (before) ⟵ baseline	194	+56.31	+0.290	48	-22.01	-9.47	+14.34
atm	194	+55.83	+0.288	61	-18.99	-8.29	+4.54
atm-1	194	+60.42	+0.311	55	-20.87	-8.49	+6.60
atm-2	194	+61.22	+0.316	51	-19.99	-8.89	+7.31

Put overlay alone (premium collected − capped P/L)

Structure	n	Total %	Avg %/trade	Win %	Max DD %	Worst %	Best %
atm	194	-0.48	-0.002	78	-43.74	-13.55	+2.05
atm-1	194	+4.11	+0.021	90	-35.06	-12.26	+1.27
atm-2	194	+4.91	+0.025	94	-22.59	-10.56	+1.19

Year by year — is it better every year?

combined — total % per year

Year	future only	atm	atm-1	atm-2
2021	-6.57	+16.07	+10.19	+5.63
2022	+41.09	+43.70	+55.17	+51.36
2023	-4.10	+3.94	-1.62	-2.80
2024	+27.09	+4.75	+9.35	+15.16
2025	+0.46	-11.50	-11.91	-7.32
2026	-1.65	-1.13	-0.76	-0.82
ALL	+56.31	+55.83	+60.42	+61.22
Δ vs future only (variant − base, per year)
Δ 2021	+0.00	+22.64	+16.76	+12.21
Δ 2022	+0.00	+2.61	+14.08	+10.28
Δ 2023	+0.00	+8.04	+2.48	+1.30
Δ 2024	+0.00	-22.34	-17.74	-11.93
Δ 2025	+0.00	-11.96	-12.37	-7.78
Δ 2026	+0.00	+0.52	+0.89	+0.83

Result (put on shorts): the covered put adds almost nothing. The put overlay alone barely earns (overlay ATM −0.5%, ATM−1 +4.1%, ATM−2 +4.9% over 194 nights), so combined ≈ short-only (combined_atm +55.8% vs short-only +56.3%; ATM−1 +60.4%, ATM−2 +61.2%). The reason is structural: the short's overnight edge is the sharp down-move, and the sold put gives exactly that back (overlay worst −13.6% at ATM on a hard down-night). DD is trimmed only mildly (≈ −19 to −21% vs −22%). Win-rate of the overlay is high (78–94%) but the few losing nights are large. Mirror of the call side, but the short window is smaller (194 vs 522 nights) and the payoff far less convex.
M2 vs Tardis: the short put is the one bucket where M2 was a touch generous (overnight put ATM M2 −0.89 vs Tardis −1.62, tab 11) — i.e. the real put overlay would be marginally worse, reinforcing "adds little".

before = the future (long or short) over the window · after = future + sold option (call on longs, put on shorts)

Combined — future + sold option (after) vs future only (before)

Structure	n	Total %	Avg %/trade	Win %	Max DD %	Worst %	Best %
future only (before) ⟵ baseline	716	+151.16	+0.211	49	-50.12	-9.47	+14.34
atm	716	+262.75	+0.367	66	-35.23	-8.29	+4.54
+1	716	+247.97	+0.346	55	-40.84	-8.49	+6.60
+2	716	+241.64	+0.337	52	-43.99	-8.89	+8.63

Option overlay alone (premium collected − capped P/L)

Structure	n	Total %	Avg %/trade	Win %	Max DD %	Worst %	Best %
atm	716	+111.59	+0.156	81	-67.42	-13.55	+2.05
+1	716	+96.81	+0.135	91	-47.90	-12.26	+1.42
+2	716	+90.48	+0.126	96	-25.58	-10.56	+1.19

Year by year — is it better every year?

combined — total % per year

Year	future only	atm	+1	+2
2021	+23.97	+143.38	+121.51	+106.29
2022	+47.38	+86.70	+81.30	+67.92
2023	+38.75	+32.04	+38.41	+40.34
2024	+56.47	+32.65	+39.35	+49.14
2025	+5.93	-17.88	-13.38	-2.25
2026	-21.35	-14.14	-19.21	-19.79
ALL	+151.16	+262.75	+247.97	+241.64
Δ vs future only (variant − base, per year)
Δ 2021	+0.00	+119.41	+97.54	+82.32
Δ 2022	+0.00	+39.32	+33.91	+20.54
Δ 2023	+0.00	-6.71	-0.35	+1.59
Δ 2024	+0.00	-23.82	-17.12	-7.33
Δ 2025	+0.00	-23.82	-19.31	-8.19
Δ 2026	+0.00	+7.21	+2.14	+1.56

Result (all together): merging every long-night and short-night into one equity (522+194 = 716 nights), the sold-option overlay lifts the total (combined_atm +262.8% vs future-only +151.2%) and cuts the merged maxDD (−35.2% vs −50.1%). Almost all of the lift comes from the call side (longs); the put side (shorts) contributes little, as the PUT tab shows. The merged future-only DD (−50%) is deeper than either side alone because long- and short-nights stack into one curve. Further OTM trims the total but keeps more upside (combined +1 +248.0%, +2 +241.6%).
M2 vs Tardis: ratios ≈ 1.0–1.2 at the overnight tenor (tab 11); pattern across strikes is the firmer read on the ~4-month overlap.

Explanation

before = the real algo (futures, directional) · after = sell credit spread on the signal (no future)

Combined — every long→put-CS + every short→call-CS (vs the real algo)

PUT side — long signals → put credit spread (vs algo longs)

CALL side — short signals → call credit spread (vs algo shorts)

Result — selling credit spreads on the signals badly trails the algo. Every leg is anchored to the algo's actual futures fill (the breakout level it really traded), so there is no entry-timing look-ahead. The real algo makes +532% (win 47%) over the full 942-trade history; the best combined credit book reaches only +83% (3d +2, win 86%, maxDD -42.0%), with offsets/horizons spanning -16% … +83% — a fraction of the future. The two sides split sharply. The PUT side (sell put on longs) is the only earner — put 3d +2 +115% (win 90%, maxDD -9.4%), put 3d atm +88% (maxDD -24.3%) — smooth and high-win, but still well below the algo's longs (+243% / maxDD -32.9%). The CALL side (sell call on shorts) is the killer — negative almost everywhere: best is call 1d +2 +4%, worst call 3d atm -104% (maxDD -117.0%). In an up-drifting market the bearish call-CS keeps getting run over and hands back the algo's short edge (+290% / maxDD -19.3%). Win rates look high everywhere (65–96%) but the wins are capped (≤ width − credit) against full −width losers — the classic credit-spread payoff that a trending market punishes. M2 vs Tardis: spreads net the bought against the sold leg, so the residual M2 bias largely cancels; 7d/14d M2 ≈ Tardis mid (tab 11). The pattern over offsets/horizons is the read.

Explanation

before = the real algo on its longs (futures) · after = sell put below + buy call above (no future) · also vs long calls from exp 2

Risk reversal — vs the algo's longs AND vs the long calls (exp 2)

Entry premium — net credit (paid to you) or debit (you pay) per variant

Put leg alone (sell put below) — the consistent earner