05 — The Arbitrage Playbook

Box, fly, carry — the geometry of "free" money.

Three structures define spread arbitrage. Most are arbitraged away in liquid markets — which is exactly why knowing the no-arb price tells you where real mispricing (and real risk) lives. Every figure here is cited in /research/01–04.

STRUCTURE · 01

The Box Spread

A 4-leg options box (bull call + bear put, same strikes/expiry) locks a known payoff = strike width. It's a synthetic zero-coupon loan: SPX European boxes trade $900M+ notional/day as a cleared repo market, financing at SOFR + 20–31 bps. CME has settled options off box rates since Dec 2022.

Fair value = (K₂ − K₁) · e^(−rT)
// pure arb gone — MMs close it in ms
// American options ⇒ early assignment

⚠ WSB 2019: $57k loss on a $5k account — used American equity options, got assigned. Boxes are riskless only on European (SPX/ES). [01]

STRUCTURE · 02

The Butterfly (Fly)

+1 near / −2 middle / +1 far. Pure bet on curvature — the 2nd derivative of the curve. It profits when the belly bends vs the wings, with near-zero level/slope risk. Capital edge is huge: SPAN margins a CL fly at ~$200–800 vs ~$6,500 for an outright — 90%+ less.

Fly = S(t₁,t₂) − S(t₂,t₃)
= P₁ − 2·P₂ + P₃
// curve encodes physical info ⇒ not
// fully arbed away (unlike option boxes)
STRUCTURE · 03

Cash-and-Carry

The hard no-arb bound from cost-of-carry. If contango exceeds full carry, buy physical front, store it, sell the back — a locked, financed profit. The 2009 & 2020 floating-storage trades ran this at supertanker scale.

F(T) = S · e^((r + u − y)·T)
// u = storage, y = convenience yield
// Apr 2020: Cushing full → CL −$37.63

The bound only holds while storage exists. When it runs out, spreads blow past any carry model. [01/02]

Why retail box-arb is dead

The "alligator spread" problem: a box crosses four bid-ask spreads plus commissions. At $0.05/leg × 4 = $0.20 slippage, you've eaten any theoretical edge before you start. Market makers close real mispricings in milliseconds. The financing use (borrow/lend at the box rate) is real; the arbitrage profit is not.

Why the fly still pays

Unlike an option box — whose payoff is mechanically fixed — a futures curve encodes physical information that evolves continuously: OPEC cycles, refinery turnarounds, inventory shocks, seasonal demand. That creates genuine, recurring dislocations in curvature that a pure-arb desk can't fully erase. The fly is the cleanest, cheapest way to express them.

From the desk — FinTwit pulse

What practitioners are actually saying.

Harvested live from X (handles cited). The calendar spread isn't a textbook abstraction to these traders — it's their physical-balance gauge. Full pass in /research/04_twitter_x.md.

@akylles · @JKempEnergy · @Rory_Johnston

The M1–M2 spread is the truth serum

The front time-spread is the canonical read on real supply tightness vs geopolitical froth. Normal WTI M1–M2 backwardation sits ~$0.40–0.60/bbl; the Mar-2026 Hormuz scare peaked near ~$6/bbl — a 10+σ extreme.

@JavierBlas · @TradeLinesEF

Headlines flip the regime in hours

Time-spreads fell 30–50% on Hormuz-deal rumours within 24–48h while flat price stayed backwardated. "Steep backwardation = disruption premium, not structural." Any mean-reversion calendar trade must respect headline-driven intraday flips.

@PermianPirate · physical trader, Geneva

Backwardation self-corrects

In backwardation, physical holders dump the front month to capture carry — which itself compresses the spread. A built-in mean-reversion force. Flat price tanking on rumours while the arb widens = the physical market isn't buying the paper move.

@navnoorquant · @sev_kach

USO is the retail trap

USO bleeds 50%+ annually in contango via negative roll yield. Even a correct directional view fails when routed through an ETF in contango. Direct futures calendar spreads are the clean vehicle to harvest the same roll — not pay it.

Deep-dive sources — box financing & SPAN margin, the 2020 storage break, practitioner takes (Reddit + FinTwit/X) — in /research/01–04_*.md. FinTwit handles cited as posted; claims are practitioner opinion, not verified fact.