RESEARCH · DERIVATIVES REFERENCE | Python / Julia | Interactive reference | Research corpus | Citation
A block of derivative strategies, bundled into one book.
Block is a member of the Blaque Baux family. The core repo is the engine and blueprint — a governed, systematic platform (Julia) with a venue-agnostic execution controller and a Layer-3 live-money safety gate. Block points that engine in its own direction and inherits the governance wholesale.
Not investment advice. Educational/research software. Nothing here is validated. See LICENSE.
git clone --recursive https://github.com/blaquebaux/block.git
julia --project=engine -e 'using Pkg; Pkg.instantiate()' # one-time engine setupA meta-sleeve spanning the four derivative blocks — equity, FX, rates, commodities — combined into one book. The organizing question: those blocks are more interlocked than people credit (a stronger currency hurts the home country's exporters; higher rates raise the cost of doing business and hit equity), so a book that spans all four must map those linkages — to know whether it is genuinely diversified or secretly one macro bet, and whether the interlocks are alpha or only risk.
Full detail in research/README.md. The scorecard:
| # | Question | Verdict |
|---|---|---|
| 1 | Are the 4 blocks one interlocked factor? | |
| 2 | Currency → home equity (H1)? | ✅ real — strong yen vs local Japan −0.47; but wrapper cancels it (+0.01) and it's export/EM-only |
| 2 | Rates → equity (H2)? | ✅ real but regime-dependent — stock-bond corr −0.36 (2016–21) → +0.11 (2022+) |
| 3 | Is the interlock tradeable? | ❌ not lead-lag alpha (priced in); ✅ cross-asset trend +0.43 and 4-block hold +1.19 at 4.1/5 bets |
The synthesis: the intuition is right — the blocks genuinely interlock — but the research reframes it. The links are real yet weak (the four blocks still carry ~4.6 of 8 independent factors, far more than the ~19% within one equity block — they are more diversified than interlocked). H1 holds only in local currency terms and is cancelled inside the USD wrappers you'd actually trade; H2 holds only in an inflation regime and flips sign in a growth-scare one. And the interlocks are priced instantly (no lead-lag edge). What earns its keep is owning the four blocks as a diversified book (+1.19 Sharpe, 4.1/5 bets) and trend-following them — the linkages are for hedging and sizing, not stat-arb.
Research: three bodies of work, all on the record. (1) The original cross-asset linkage study — a
diversified basket + risk framework, not stat-arb (scorecard above). (2) The seven-block premium map
(equity-vol · rates · credit · commodities · FX · municipal · funding, below) and its graduation into a
governed cross-asset keeper book (live/crossasset_allocator.py +
live/crossasset_live.jl) — gold + gated duration + equity-vol-gated FX carry routed
through the Layer-3 safety gate, dry-run PASSES. (3) The "modify to death" variation program — ten
catalog structures each varied on real data or a market-standard model, synthesized in the whitepaper
"Where the Fee Lives". The 120-catalog itself remains a reference layer;
nothing here is validated to the spine's live-money bar.
Block's second and larger half is the platform's derivatives reference layer: the definitive institutional taxonomy (equity, FX, rates, credit, commodities, municipals, funding), made computable and held to the same verify-before-claim discipline as the rest of Blaque Baux.
catalog/— 120 strategy building blocks. A machine-readable taxonomy (derivatives_catalog.json,schema.json): parties, cash-flow legs, parameters, payoff, variants, and the documented combination/hedging plays for each. Every entry is a building block, not standalone alpha — the value is in composition. Status ladderreference → spec → implemented → validated.pricing/— reference pricers, one per major class, keyed to the catalogidand each proven against its textbook identity (all pass): IRS (par swap = 0 value), FX forward (covered interest parity), Black-Scholes option (put-call parity), CDS (credit triangle).analytics/— the cross-cutting layer: index calculation + the divisor (rebalance continuity), price-movement correlation → basket variance / diversification ratio / implied correlation, beta & sensitivity, dividends & corporate-action adjustments, and the index-vs-basket basis. 12 identities, all verified.docs/derivatives_framework.mdties it together; two plays are on the record — a collar guardrail (SPY Sharpe 1.01→1.39, tail capped, at the cost of upside) and the correlation-regime finding (diversification collapses in stress, exactly when it's needed).
The through-line: every readout is a guardrail, not a signal — what a bespoke book actually owns and where it's exposed. None of these blocks is a money-maker alone; the value is the comfort and control they lend.
Block's next phase: take each catalogued derivative and vary it to death — combine the carry of one with the convexity of another, the derivative-of-a-derivative, ATM vs OTM, hedged vs naked — and suss out, with the family's honest method (P&L, skew, the fat-tail toolkit, benchmark vs the nulls), which added component is paying the buyer versus paying the desk. Finance firms earn a fee on every component bolted onto a bespoke instrument; the research question is which components genuinely add return, cut risk, or are pure fee-in-a-suit.
The synthesis paper —
docs/derivatives_honesty.html("Where the Fee Lives"). All ten variations distilled into one law: a modification never conjures return — it relocates risk onto an axis the buyer isn't watching (the tail, an unobservable mark, or the funding leg), and the honest exceptions prove the rule. The reproducible detail follows below.
Variation #1 — straddles (research/block_straddle_variations.py):
a Black-Scholes P&L simulator on the real SPY path (implied vol swept, since VIX isn't on the feed), across
long/short straddle and OTM strangle. Findings: (1) no edge without the vol-risk-premium — at fair pricing
every variant is ~±0.2 Sharpe noise; the entire return is implied>realized. (2) Long vol is a cost (the
bleed profile), worst when you buy a fat post-spike premium that mean-reverts. (3) The short strangle "wins"
in-sample (+0.92 Sharpe, 72% win-rate) — and that's the trap: its catastrophic tail simply isn't in
2016–2026 monthly bars, but brace's SVXY lived it (−95%, Feb-2018). The
straddle is a pure VRP instrument; the variants reshape the premium/tail tradeoff, not the bet — and the
best-looking one is the most dangerous.
Variation #2 — the delta-hedged short straddle (research/block_straddle_deltahedged.py):
sell the ATM straddle and delta-hedge daily, stripping the directional luck to isolate the pure vol premium
(realized vs implied variance). Finding: the first modification that genuinely improves the book —
removing the directional noise earns the same premium at a higher Sharpe (+0.21→+0.48 at a 10% VRP; vol
9%→7%), which is exactly why vol desks hedge. But it doesn't remove the short-gamma tail (the worst month
is still a big loss on a vol spike), and at fair pricing (mult=1.0) it's still ~0 Sharpe — no edge without the
premium. Honest caveat: the lift is gross of daily-rehedging cost (cheap on SPY, ruinous on an illiquid
underlying). So the added component (hedging) pays the buyer — cleaner harvest, same bet (brace, refined).
Next: regime-gate the hedged short (harvest in calm only) and calendar/diagonal (own the far tail cheaply) — the modifications that attack the tail.
Variation #3 — calendar & diagonal: attacking the tail (research/block_calendar_diagonal.py):
own a far-dated option to cap the short-vol catastrophe. Two structures vs the naked short. Calendar
(short 1M + long 3M straddle, same strike) turns out to be net long vol — the far leg cancels the near's
move and adds vega; it's a bleed/long-vol variant financed by near theta, not a safer harvest. Diagonal
(short 1M ATM straddle + long 3M ±10% OTM wings) is the structurally correct tail cap — but on this data it
made things worse (worst month −7%→−9%, Calmar +0.11→−0.02): the ±10% wings almost never triggered because
2016–2026 monthly moves rarely reach 10%, so they were pure premium drag. The honest trap runs both ways:
you can't judge a tail structure on a sample without the tail — the wings look like waste for the same reason
the naked short looks safe (the SVXY-style −95% catastrophe isn't in monthly bars). Verdict: unprovable on
monthly data — the diagonal is the right structure (bounded loss) but its value only shows in a real crisis
this sample lacks. The methodological lesson is Block's core: tail structures need tail data (daily/intraday
or a crisis window) — brace's real SVXY −95% outweighs any monthly backtest.
Variation #3b — the diagonal on DAILY marks through 2018 & 2020 (research/block_diagonal_daily.py):
re-run with daily valuation so the crisis paths are visible, and it resolves #3 both ways. The cap is real
and large: in COVID the naked short's −21.6% intra-drawdown becomes the diagonal's +6.9% gain; in
Feb-2018 Volmageddon −1.8% becomes +3.0% — the far wings pay exactly when needed, turning an open-ended
tail into a bounded one. But the insurance is expensive: full-sample, the diagonal's daily-mark drawdown
(−43%) is worse than the naked short's (−29%) — paying wing carry every calm day grinds a deeper long-run
bleed than the crises it prevents. So it's not a free lunch: you trade an acute, potentially terminal
blow-up (brace/SVXY) for a chronic, survivable bleed. The "right" short-vol structure is a ruin-aversion
choice, not a Sharpe one — and only tail data could reveal it. Next: variance/vol swaps (the cleaner VRP
instrument), then the rates & credit blocks.
Variation #4 — variance vs vol swaps: the pure VRP, and its convexity (research/block_variance_swaps.py):
the straddle family kept circling the vol premium through strikes and paths; the variance swap is the pure
instrument (pays realized − strike variance, no strike/delta/path), and the vol swap is its linear cousin.
Comparing them prices the premium cleanly and exposes the convex tail. Two honest findings, both of which
corrected a prior: (1) "fair" pricing is a loser. On a trailing-realized strike, both swaps lose heavily
even at mult=1.0 (vol −0.66 Sharpe, var −0.92) — trailing realized systematically under-forecasts the
fat-tailed forward realized, so you're short a violently negative-skewed forecast error. This is the whole
reason the market strike (VIX) sits above realized — the VRP is the fee for that un-forecastable spike; you
only turn a profit once you add a real premium buffer (the vol swap needs a 19% cushion to reach +0.47).
(2) Convexity is terminal. variance = vol², so a k× spike costs a var swap ~k² vs the vol swap's ~k: worst
month −1779% vs −333%, skew −4.53 vs −2.24, COVID −892% vs −267%, ruin-months 26 vs 9 — the exact
convexity that detonated short-variance books in 2008. Verdict: the vol swap dominates the var swap
(same premium, linear survivable tail vs quadratic terminal one), but neither is free — both need a real VRP
cushion to profit and both can still be wiped by one spike (maxDD −100%). brace at heart, in cleanest form:
the strike matters more than the structure, and the tail is the whole story. Next: the rates block
(swaps/swaptions) and the credit block (CDS/index) — genuinely new economics beyond the vol premium.
Variation #5 — swaptions: the rate-vol premium and its two tails (research/block_rates_swaption.py):
the rates block of the catalog, modeled as options on duration (payer swaption ≈ put on TLT, receiver ≈ call).
Findings: (1) a real rate VRP — selling the rate straddle pays and scales with the premium (Sharpe +0.17
fair → +0.89 at a 19% VRP). (2) But rates have two tails, not equity's one, and the violent one is
the side I didn't expect: the worst single-month tail is the receiver / rates-down side (short receiver
worst month −16%, skew −2.66 — the Mar-2020 flight-to-quality bond spike), while the payer / rates-up tail
(2022) is a slow grind (−17% cumulative, worst month only −8%). (3) A twist vs the equity arc: rates
contain their worst year (2022) in-sample, so the short-straddle Sharpe is more trustworthy than equity
short-vol's — but negative skew both ways keeps it insurance-writing, and the Bermudan early-exercise feature is
fee-in-a-suit for a seller. Verdict: the rate-vol premium is real and double-tailed; sell it sized for
both tails, never as clean carry.
Variation #6 — CDS index tranches: the correlation trade, and 2008 (research/block_credit_tranches.py):
the credit block's defining machine (tranche / first-to-default / Nth-to-default / synthetic CDO), modeled with
the market-standard one-factor Gaussian copula. Findings: (1) tranching creates nothing — the invariance
check is exact (Σ tranche loss = portfolio expected loss); it only redistributes the same loss by seniority,
so the structuring fee buys slicing, not edge — the purest fee-in-a-suit. (2) The senior "AAA safety" is a
correlation bet: at ρ=0.1 the super-senior tranche loses 0.0% (bulletproof on paper), but as ρ→0.9 the
senior [7–15%] jumps ≈5× and the super-senior to 1.8–3.9% — in 2008 ρ→1 and the "impossible" AAA losses
simply happened. (3) The trade is the equity/senior split: equity is short correlation (corr-delta
−55%), senior is long correlation risk — selling a senior tranche is writing systemic insurance (the
steamroller at portfolio level). Verdict: the catalog's most infamous instrument makes a senior claim look
safe by hiding a correlation tail inside a flattering metric — the metric flatters, the tail decides, in its
most consequential form. 2008 was the bill for selling systemic-correlation insurance and stamping it AAA.
Variation #7 — CMS: the convexity fee, and where the margin hides (research/block_rates_cms.py):
the CMS family of the catalog (constant-maturity swap, CMS spread option, range accrual). A CMS pays a long-tenor
swap rate on a short schedule, forcing a convexity adjustment (the fair CMS rate sits above the forward, by
Jensen). Findings: (1) the adjustment is real, not fee-in-a-suit — its sign is model-independent (Jensen),
and a 10y CMS genuinely carries ~25bp, a 30y ~60bp+. (2) But the margin hides in the vol mark: the same 10y
CMS is worth 9bp at σ=15% and 49bp at σ=35% (the 30y swings ~97bp) — purely from the vol assumption the
buyer can't observe. The fee isn't a charge, it's a mark. (3) The CMS spread option (10y−2y steepener)
tells the same story on correlation — its value falls 51% as the rate-rate ρ goes 0→0.9. The unifying
law (with #6): exotic value lives in an unobservable parameter (σ for CMS, ρ for the spread option and the
tranche), so the "fair price" flatters and the assumption decides — the metric-flatters/tail-decides law
applied to pricing. Verdict: CMS convexity is honestly-earned value, but it's the catalog's clearest case
of a fee that hides in an assumption rather than a line — the "innovation" is the desk's information edge on an
unobservable, not new economics. Rule: when value depends on σ or ρ, ask for the σ or ρ, not just the price.
Variation #8 — commodity structures: crack spread · Asian option · three-way collar (research/block_commodity_structures.py):
three commodity-block structures chosen to span the honest range — the program isn't only about traps.
(1) Crack spread (UGA gasoline − USO crude, the refiner margin) is a real, diversifying exposure: near-zero
equity correlation (+0.05) and — rare — positive skew (+1.25; the tail is on the upside, +95% in 2020 as crude
collapsed but products held). Not standalone alpha (Sharpe +0.34), but a genuinely uncorrelated positive-skew
sleeve. (2) The Asian (average-rate) option is the modification that pays the buyer: averaging cuts the
effective vol (terminal 21% → average 12%, ≈/√3), so it's 44% cheaper — and the discount is fair, a better,
cheaper hedge for anyone with continuous/averaged exposure. Not every modification is desk margin. (3) The
three-way collar is the trap: "widen the zero-cost band" by re-selling a deep put, and below that strike you're
long the crash again — worst 3-mo −61% vs the plain collar's −10% (skew −2.16 vs −0.28); in 2020 oil it blew
through exactly where you thought you were protected. The sharpened rule (with #6, #7): a modification pays
the buyer when it genuinely reduces risk (Asian averaging) and the desk when it hides risk (the
three-way's re-sold tail, the tranche's correlation, the CMS's vol mark) — read the payoff to the tail, not the
brochure; and some structures are honest exposure, neither trick nor trap (the crack spread).
Variation #9 — FX exotics: the target-forward & the barrier put (research/block_fx_exotics.py):
the FX block is the catalog's densest field of tail-hiding "corporate hedges." On GBP (which carries the 2016,
2020 and 2022 crashes): (1) The target-forward (TARF) is asymmetry weaponized — its best outcome is capped
at the +5% target (vs a plain forward strip's +141% symmetric upside), its worst is −303% (2× the strip's
−151%, from downside leverage), skew −1.56, and any small win knocks it out (only ~1% of years reach the
cap). You cannot win big, you can lose without limit — 2015-CHF and a graveyard of EM corporates in one payoff.
(2) The down-and-out put sells you a 23% discount by removing the tail: in the windows that actually crashed
through the barrier the vanilla put paid 6.9% while the down-and-out paid 0.0% — crash insurance that cancels
itself in a crash. Verdict: FX exotics sold to hedgers are the sharpest form of the program's law — marketed
as risk reduction while they add catastrophic risk (the TARF inverts the tail into leverage; the barrier put
deletes the cover). The discount/enhancement is always the tail, priced out. The honest FX counterpart is the
average-rate forward (averaging genuinely cuts vol — a fair, cheaper hedge). Read the payoff to the tail,
not the brochure — this closes the option-family sweep across vol, rates, credit, commodity, and FX.
Variation #10 — linear/structural: funding, basis, convention (research/block_linear_structural.py):
the catalog's linear entries (IRS, basis/OIS/zero-coupon swaps, cross-currency swap, NDF, asset swap, total
return swap, quanto, funding legs) carry no option tail, so the risk moves to other axes. (1) The TRS is the
TARF's linear twin: synthetic leverage for a ~0.4%/yr spread, but the tail is the funding-pull / margin
wipeout — on real SPY, 3× is wiped by COVID (−34%), 5× by 2020 and 2022, 8× by 2018 too; Archegos (2021, ~5–8×
via TRS) is the proof. The visible fee is trivial; the funding leg is the risk. (2) The cross-currency basis
is a "free arbitrage" that isn't — a fee for balance-sheet scarcity that gaps in stress (JPY ≈ −140bp in
Mar-2020) — and, honestly, not recoverable from an equity/ETF feed (flagged, not faked). (3) Quanto/convention
value hides in an unobservable correlation, like the CMS σ and tranche ρ. The grand unifying law (all 10
variations): a modification never conjures return — it relocates risk onto an axis the buyer isn't watching:
options hide it in the tail, exotic pricing in an unobservable mark, linear/structural in the funding
leg or basis. The honest exceptions prove the rule — the Asian, the average-rate forward, the crack spread, and
the weather derivative (≈zero market beta by construction) genuinely serve the user, fairly priced. The one
rule the whole program earns: read the payoff to the tail, the mark, and the funding — never the brochure.
The "modify to death" program now spans all six catalog blocks — vol (#1–4), rates (#5, #7), credit (#6), commodity (#8), FX (#9), and linear/structural (#10) — a complete honest map of where a derivative's fee actually lives.
Rates #1 — does bearing duration pay? (research/block_rates_termpremium.py):
off the vol premium entirely, into new economics. Duration buckets (SHY/IEI/IEF/TLT) as total-return excess
over cash (BIL), plus a 100d-trend gate. The honest scorecard: (1) the term premium was negative this
decade — every bucket lost to cash (TLT −2.0%/yr) and the tail scaled straight with duration (TLT maxDD
−52%, 2022 −32%); buy-and-hold duration did not pay 2016–2026. (2) But bonds are the crisis mirror of the
vol/equity block — and that's their value: positive skew (opposite equity's), a flight-to-quality rally in
COVID-2020 (+17.7% TLT) exactly as the short-vol book detonated, and a different catastrophe (2022 inflation,
not a growth scare). You don't hold duration to earn — you hold it to be long the other side of equity's crash.
(3) Trend-timing bounds the tail but doesn't make edge: the 100d gate cut 2022 from −32% to −11% and halved
maxDD while keeping the 2020 upside and positive skew, but standalone Sharpe stayed ~0 (IEF +0.13, TLT −0.18) —
a risk overlay, not alpha. Discipline note: the naive same-bar signal faked Sharpe +1.17; the one-bar-lag
correction cut it to +0.13 — caught, both reported, honest one kept. Verdict: duration is a near-null on
carry but a real positive-skew, equity-crisis-mirror diversifier — the honest reason a balanced book (family:
balanced / bonds) holds it.
Next: carry/roll-down across the curve, then the credit block (CDS/index) — the default-risk premium.
Rates #2 — carry / roll-down: the curve slope as the signal (research/block_rates_carry.py):
an upward-sloping curve pays you to hold duration (coupon + roll-down); an inverted one doesn't — so the right
signal is the level of the slope, not price trend. Slope reconstructed price-only from ETF distribution
yields (IEF yield − BIL cash yield). Findings: (1) carry is a genuinely distinct signal — it correlates
just −0.09 with the trend gate, agreeing only 43% of days (trend reads fast price momentum; carry reads the
slow curve state). (2) Combining them is the best rates result yet: trend-AND-carry on IEF lifts Sharpe
−0.14 → +0.30, maxDD −28% → −9%, skew +0.17 → +0.64, 2022 −16% → −4%, on just 30% of days (TLT the
same shape) — trend dodges the acute 2022 price crash, carry avoids the 2023 negative-carry inversion, each
covering the other's blind spot. (3) Honest caveats: the distribution-yield proxy lags (carry-alone rode
the full 2022 crash before registering the inversion in 2023), and even combined the excess return is thin
(~+1%/yr) on one big regime — the win is risk (bounded tail, positive skew), not carry harvested.
Verdict: carry is a real, distinct signal and trend+carry is the honest best way to hold duration —
but it's a risk overlay on a crisis-mirror diversifier, not a harvested premium: the curve doesn't pay you to
hold duration this decade, it tells you when duration is least dangerous. Next: the credit block (CDS/index)
— the default-risk premium, the first genuinely earned premium to test since the VRP.
Credit #1 — real premium, or equity beta in a bond wrapper? (research/block_credit_premium.py):
strip the rate risk with a duration-matched Treasury to isolate the pure spread (IG = LQD−IEF, HY = HYG−IEI),
then test the residual with the family toolkit (Jensen's alpha vs SPY + crisis correlation). Findings: (1) a
gross spread premium exists — duration-stripped, IG earns +1.6%/yr and HY +3.0%/yr, so credit does pay for
default risk. (2) But it's equity beta in a wrapper, and underpaid: beta to SPY +0.22 (IG) / +0.39 (HY),
and Jensen's alpha is negative (−1.4% / −2.3%) — after the equity risk you're taking, the premium doesn't
vanish, it goes negative (equity-like downside, sub-equity upside this decade). (3) The crisis tell is
definitive: on equity's worst 5% days the spread crashes with stocks (corr +0.71 IG, +0.86 HY) and
the skew is negative (IG −1.77) — zero diversification exactly when you need it. The finding is the
contrast: credit is the mirror-opposite of duration — duration is negative-carry / positive-skew /
crisis-mirror; credit is positive-carry / negative-skew / crisis-correlated. "Fixed income" is two opposite
factors bolted together, and only the duration half actually diversifies an equity book. Verdict: the
default-risk premium is not an independent earned premium here — it's underpaid equity beta wearing a bond's
illiquidity; for a cross-asset book, HY credit is not diversification (hold equity directly, or duration for a
real hedge). Next: gate credit on a risk-on regime (does timing rescue it?), then FX / commodities carry.
FX #1 — the carry premium: the classic earned return, and its crash tail (research/block_fx_carry.py):
FX carry is the most-cited independent premium in macro — borrow the low-yielders, lend the high-yielders.
Tested two ways: the packaged DBV ETF (2016–23) and a self-constructed G10 basket ranking six CurrencyShares
ETFs by their own trailing distribution yield (long top-2 / short bottom-2, dollar-neutral). Findings: (1)
construction validated — the yield ranking is textbook (AUD/GBP/CAD high, CHF/JPY the funders). (2) The
premium is real but thin — +2.4%/yr constructed (Sharpe +0.32), +1.4% DBV. (3) The steamroller is real —
negative skew on both (−0.54 / −0.57) and it partly unwinds in risk-off (corr +0.57 with equity on its
worst days). (4) But it's the least costume of the earned premia: vs credit, carry's equity beta is lower
(+0.22 vs +0.39), crisis-correlation milder (+0.57 vs +0.86), and Jensen's alpha is ~0 (−0.6% / −0.1%) rather
than deeply negative — carry keeps a genuine independence credit lacked. Verdict: of VRP / credit / carry,
FX carry is the closest thing to a real standalone premium — but it's thin, crash-prone, and its alpha is ~0,
not positive: you're paid roughly fairly for bearing an equity-correlated tail. The nickels are real; so is
the steamroller. Next: regime-gate carry to dodge the unwinds, then commodities carry (backwardation / roll).
FX #2 — regime-gated carry: can timing dodge the steamroller? (research/block_fx_carry_timed.py):
carry's crashes are vol spikes, so gating exposure off in risk-off should dodge the unwind. The real test is
whether a gate fixes the skew and cuts the crisis-correlation, not just trims the mean. Four lag-safe gates
tested; the result is clean. The equity-vol gate wins decisively — crisis-corr +0.57 → −0.01, maxDD
−21% → −12%, skew −0.54 → −0.36 (the steamroller removed), while Sharpe holds (+0.32 → +0.31) and Jensen
alpha flips positive (−0.6% → +0.6%) — the first genuine positive independent alpha in the whole
earned-premium arc, and it came from dodging the tail, not a fatter premium. The wrong gates fail, honestly
reported: own-vol over-trims (Sharpe → +0.15), the dollar gate worsens skew (−1.00) and keeps crisis-corr
+0.39, momentum whipsaws (+0.09). The lesson — match the gate to the failure mode: carry fails in equity-vol
spikes, so an equity-vol filter is the right signal and price/dollar trend are not — "match the signal to the
sleeve," now proven at the gate level. Caveat: the winning alpha is small (+0.6%) on one decade and
un-costed for turnover — a real improvement, not a validated keeper. Verdict: timing turns FX carry from a
fairly-priced, crash-prone premium into a small positive-alpha, tail-managed one — the arc's first true (if
thin) edge, and exactly on the program's thesis, the edge is risk management, not the premium. Next and last
asset block: commodities carry (backwardation / roll-yield) — completing Block's four-block cross-asset map.
Commodities #1 — roll yield / backwardation (research/block_commodities_carry.py):
the last asset block. Commodity carry is roll yield — read from front-vs-deferred ETF pairs (USO/USL oil,
UNG/UNL gas). Findings: (1) the roll tax is the whole game — WTI front bled +4.0%/yr to the 12-mo deferred,
and natgas front lost −12.1%/yr (an +11.2%/yr roll tax) almost purely to contango; the curve state is the
return. (2) Carry-timing dodges the catastrophes — holding the front only when backwardated turned oil-2020
from −68% to +11% and fixed skew (−0.82 → −0.13); the one direct harvest is shorting persistent contango
(natgas L/S +7.5%), but oil L/S was −1.8% when its curve flipped to backwardation — real but regime-dependent,
with brutal idiosyncratic tails (maxDD −60% to −95%, oil went negative in 2020). (3) Independence is partial,
and gold is the prize — broad commodities carry moderate equity beta (+0.32) but hedge inflation (2022 +18%);
gold is the standout: Sharpe +0.70, near-zero equity beta (+0.08), low crisis-corr — the most
independent, most diversifying single asset in the whole study.
Municipal #1 — the muni spread (research/block_muni_premium.py): the muni
spread (MUB−IEF, HYD−MUB) tested like credit #1. Munis are the "good credit" — about half the equity beta
of corporates (IG-muni β +0.11 vs +0.22; HY-muni +0.19 vs +0.39), because they default on politics/rates, not
earnings. But the pre-tax spread is thin (+0.9% / +0.7%) with negative Jensen alpha (−0.6% / −1.8%) and a
March-2020 liquidity tail (skew −1.15 / −1.52, crisis-corr +0.61). The real premium is the tax code: MUB's
3.5% yield is a 5.8% taxable-equivalent (top bracket) vs IEF's 4.2% — a +1.6%/yr after-tax pickup the pre-tax
return can't show. Verdict: for a taxable investor munis dominate corporate credit (lower beta + tax pickup),
but the edge is the exemption, not a harvestable spread — neither is a standalone premium.
Funding #1 — the front-end money-market premium (research/block_funding_premium.py):
isolated with floating-rate pairs (FLOT−USFR, MINT−BIL) that strip duration entirely. A real, independent,
steady carry (+0.5…0.75%/yr at near-zero equity beta) — but pure short-liquidity: the most extreme negative
skew in the whole study (−9.3 to −23.6), where the worst day (−6.9%) erases ~a decade of carry and it gaps in a
freeze (Mar-2020, crisis-corr +0.62). The Sharpe is an illusion — FLOT−USFR's +0.15 becomes +0.61 once you
exclude March-2020 alone; one month is the risk. Verdict: fairly-priced liquidity-crisis insurance-selling
— thin pennies for a rare violent gap; it belongs in a book as sized insurance-selling, never as "high-Sharpe
carry." The flattering calm-Sharpe is the exact trap the whole arc keeps exposing: the metric flatters, the tail decides.
All seven catalog categories now tested with one honest toolkit (equity/vol · rates · credit · commodities · FX · municipals · funding). Standalone "premia" mostly dissolve: VRP is a crash tail dressed as income; credit is underpaid equity beta in a wrapper; munis are better credit but the edge is tax, not spread; FX & commodity carry are priced roughly fairly with violent tails; funding is insurance-selling whose Sharpe is an illusion. The durable edges are not premia — they are (1) diversification (duration's crisis-mirror positive skew, and gold's near-zero beta actually hedge an equity book) and (2) risk-management / timing (the equity-vol gate that gave FX carry the arc's only positive alpha; curve-timing that dodges the commodity roll-tax). And one signature recurs in every block — the tail is the whole story, and headline metrics flatter it. Block's founding thesis holds: the value of a cross-asset book is in combination and risk control, not in any single harvested premium — the edge is the portfolio, not the trade.
The payoff: assemble only the validated keepers into one book and test whether the combination clears the
bar no single block did. Validation (research/crossasset_keeper_book.py):
equal-weight three sleeves — gold, gated duration (IEF, trend AND curve-carry), equity-vol-gated FX
carry — that are near-uncorrelated (cross-corr +0.17 / −0.05 / −0.22). The book's Sharpe +0.83 exceeds every
standalone sleeve (gold +0.72), with low equity correlation (+0.15), a −10% max drawdown, and positive crisis
years (2020 +12.7%, 2022 ~flat). As a cash-funded overlay it lifts an equity core from Sharpe +0.75 → +0.93
(1×) and dominates 60/40 (+0.70). Honest caveats: the book is gold-dominated (duration & FX carry earn their
place as low/negative-correlation ballast, not return), it's a Sharpe/return enhancer not a drawdown hedge,
and it leans on one decade. But the thesis is validated — the edge is the portfolio — so it graduates to the
governed allocator: live/crossasset_allocator.py emits today's target book
(as of the last settled close), and live/crossasset_live.jl routes it through the
engine's Layer-3 safety gate (preflight, idempotency, reconciliation, HWM, kill switch) — no LLM in the order
path. Dry-run PASSES the gate (5 names, gross 1.0x). The 120-catalog variation program that this groundwork
enables is now complete — ten structures across all six blocks, synthesized in "Where the Fee Lives".
Blaque Baux is a quantitative research initiative and a subsidiary of Carter Warrens. BlaqueBaux.com is the home for the work; the code lives here on GitHub — open to study, test, and build bespoke strategies on top of.
Anyone can point an AI at a market. The edge is understanding what the data actually says — and turning it into something you can act on. We test relentlessly and put most of it on the record as rejected, with the reason; what survives is built, governed, and validated before it is ever called real. That combination — honest research, reproducible evidence, and execution you can trust — is why Carter Warrens leads on strategy and implementation, not merely uses the tools everyone now has.
This repo is one sleeve of the Blaque Baux family — a single governed engine steered in many directions. The core repo is the base/blueprint and holds the full family roster.
engine/ the Blaque Baux platform (git submodule -> blaquebaux/base)
catalog/ 120-block derivatives taxonomy (JSON + schema + index) — the backbone
pricing/ reference pricers (IRS / FX forward / Black-Scholes / CDS) + identity self-test
analytics/ index / correlation / beta / corporate-action layer + self-test
docs/ derivatives_framework.md — the seven considerations, computed and proven
research/ cross-asset linkage sketches + the collar & correlation-regime plays
live/ governed live drivers (once a sleeve graduates to paper A/B)
MIT. (c) 2026 Carter Warrens.