Two-Leg Cross-Venue Arbitrage vs. Put-Call Parity Arbitrage — Execution, Risk, and Which One Actually Pays

Executive Summary

Crypto options admit two canonical "riskless" trades. Direct two-leg arbitrage buys a contract on the venue quoting it cheap and sells the identical contract on the venue quoting it rich; the position is market-neutral by construction and converges at expiry. Put-call parity (PCP) arbitrage trades a call, a put and a delta-one instrument (spot or perp) against each other on a single venue when the relationship C − P = S − K·e^(−rT) is violated, locking the deviation.

We put both through the same lens — execution workflow, full risk inventory, capital efficiency, and measured opportunity flow — using OptionsLab's own data: 30 days of live two-leg signals (46,917 raw signals, 711 distinct contracts) and a same-day parity scan across 922 matched call/put pairs on ClickOptions, Deribit, Binance and Bybit.

The headline result: in today's crypto market structure, direct two-leg arbitrage dominates on both return on capital and manageability. Parity across major venues is already tight — near-the-money PCP deviations have a negative median after crossing the spread (−0.3 to −9.9 bps of spot before fees), and the apparent exceptions are dominated by measurement artifacts rather than money. PCP scanning still earns a place in the stack — but as a monitoring and quote-quality layer, not as a first-line trading strategy.

Strategy 1 — Direct Two-Leg Cross-Venue Arbitrage

The trade

The same European option — same underlying, strike, expiry — trades on multiple venues. When venue A's ask sits below venue B's bid by more than round-trip costs, buy on A, sell on B. Both legs are the same instrument, so the combined position has zero market exposure: whatever the underlying does, the payoffs cancel and the locked-in gap is realized in full by expiry (or earlier, if the gap inverts and the position can be unwound at a profit).

Execution workflow

1. Detection. Continuous cross-venue quote comparison flags a fee-adjusted positive gap (this is the signal feed the Option Screener tracks).
2. Validation. Re-quote both legs at top of book; check executable size on each side; net out taker fees and, on the sell venue, the margin requirement.
3. Dual-leg execution. Both legs fire as close to simultaneously as possible. The residual risk of the strategy lives almost entirely in this step (see risks).
4. Margin management. The bought leg is fully paid; the sold leg carries maintenance margin that grows if the option goes in-the-money. The position is riskless at expiry but not margin-free along the way.
5. Convergence. At expiry both legs cash-settle against the same underlying print and the gap is banked.

What the flow actually looks like (30 days of live data)

Two numbers in that table look contradictory and are not — this distinction matters more than anything else in this paper. The gap is large as a fraction of the option premium (~17%), but the capital deployed is dominated by the sell-leg collateral, not the premium. Measured against total capital, one cycle nets roughly +0.3–0.4% — which is exactly what the live ArbHunter account has realized: +21.8% total, ~67% CAGR, −0.11% max drawdown, 8.79 Sharpe since inception. The equity curve is a staircase because every step is a converged arbitrage, not a directional bet.

Timing is bimodal: nearly a fifth of opportunities die within a minute (fast MM re-quoting), while two-thirds of flagged contracts stay intermittently actionable in episodes spread over hours — the same contract drifts in and out of arbitrage as quotes move. Automation captures the first class; humans can only capture the second.

Risk inventory

• Legging risk (the big one). If leg two fails after leg one fills, you hold a naked directional position. Mitigations: pre-trade size checks, IOC orders, automatic unwind rules.
• Margin/liquidation risk on the short leg. The arb is riskless at expiry, but a large adverse move inflates maintenance margin mid-flight. Under-collateralized accounts can be force-closed at the worst moment — turning a riskless trade into a realized loss.
• Settlement-index mismatch. The two venues settle against their own indices. They track each other closely, but a settlement-print difference of a few bps is a real (small) noise term.
• Venue/counterparty risk, ×2. Capital sits on two exchanges for the life of the trade.
• Fee drift. The gap is computed net of fees at signal time; fee-tier changes or unexpected settlement fees eat into thin gaps.

Strategy 2 — Put-Call Parity Arbitrage

The trade

For European options, C − P = S − K·e^(−rT) must hold; with short-dated crypto options, r ≈ 0 and the relation is effectively C − P = S − K (using a perp or dated future as the delta-one leg replaces S with the future and absorbs the rate term). When the market violates it:

• Conversion (market says calls rich / puts cheap): buy spot, buy the put, sell the call → guaranteed K at expiry. Profit = K − (S + P_ask − C_bid) − costs.
• Reversal (calls cheap / puts rich): short spot (or perp), sell the put, buy the call. Profit = (S + P_bid − C_ask) − K − costs.

Three legs, one venue. Same "riskless at expiry" logic as the two-leg trade — but the inefficiency being harvested is internal mispricing between an exchange's own call, put and delta-one markets, rather than a gap between exchanges.

What we measured (live scan, 13 July 2026)

Across 922 matched call/put pairs with two-sided quotes on four venues, restricted to near-the-money strikes (±5% of spot), evaluating the best executable side (conversion or reversal) gross of fees:

The medians are the story: after crossing the bid-ask spread on both option legs, the typical parity "deviation" is negative — i.e., there is nothing to take. The p90 tail (roughly 40–150 bps gross) looks tempting until you inspect it: it concentrates in thin quotes and stale books, and a realistic cost stack (two option taker fees, one spot/perp fee, plus slippage — order of 10–20 bps of spot) consumes most of what survives inspection.

The false-positive problem — an instructive artifact

Our scan initially showed one venue with 87% of pairs in apparent violation and a constant ~350 bps offset across every ATM strike. A constant offset across strikes is the signature of a reference-index mismatch — the venue prices options against its own index/forward while the scan compared against another exchange's spot — not of 350 bps of free money sitting in size. Deep in the tails, the artifacts get worse: wide, stale, or one-sided quotes on far strikes produce spectacular fake edges (we saw four-digit dollar "violations" at 68–72k strikes that no order could actually collect).

This is the operational core of PCP arbitrage: the screen is only as good as its inputs. Every candidate must be validated against the venue's own index, its own forward, executable size on all three books, and quote freshness. The two-leg strategy has no analogous problem — both legs are the same instrument, so there is no model, no index choice, and no parity relation to get wrong.

Risk inventory

• Three-leg legging risk — strictly worse than two legs; the failure surface grows combinatorially.
• Funding-rate risk. Using a perp as the delta-one leg (standard on crypto venues, since shorting spot is awkward) exposes the position to funding payments for its whole life. A reversal held two weeks through positive funding can bleed more than the locked edge.
• Capital intensity. The spot/perp leg posts full delta-one notional (or its margin), so the locked edge — measured in bps of spot — translates into basis points, not percent, of deployed capital per cycle. The two-leg trade's ~0.3–0.4% per cycle on capital is an order of magnitude better.
• Margin on two short legs (short call in a conversion; short put + short perp in a reversal) with the same liquidation dynamics as above.
• Settlement/pin risk. At expiry the option settles against the settlement index while the spot/perp leg is closed at market — small basis risk exactly at the moment the profit is supposed to be banked.
• Model/reference risk. As demonstrated above — the dominant practical risk is trading an artifact.

Head-to-Head

Verdict. On current crypto market structure, the direct two-leg trade wins on every axis that matters: it has measurably positive flow, an order of magnitude better return on capital, fewer legs, no funding exposure, and — critically — a screen that cannot be fooled by reference-index choices because both legs are the same contract. PCP violations at takeable size are essentially arbitraged out on the major venues; what a naive parity scanner surfaces is mostly measurement error. Where PCP does pay is for market makers who are already warehousing inventory on one venue and can lean on parity as a relative-value pricing constraint — a different business than taker-side arbitrage.

Is PCP Worth Adding to OptionsLab?

As an execution strategy: not as first line. The measured edge after spreads is negative at the median, the capital efficiency is structurally poor, and the funding leg imports a risk the two-leg trade simply doesn't have.

As an analytics layer: yes, and cheaply. The same matched four-venue chain that powers our Cost Analysis section already contains every input a parity monitor needs. A "Parity Watch" module would: (1) compute conversion/reversal edges per venue against that venue's own reference, (2) enforce size floors and quote-freshness filters to kill the artifact classes documented above, and (3) flag genuine, sized violations as a quote-quality signal — a venue drifting off parity is either mispricing (opportunity) or publishing a bad index (warning), and both are worth knowing about. It also strengthens the two-leg engine: a cross-venue gap that is also a same-venue parity violation is more likely to be a stale quote than a real gap.

Method Notes

Two-leg statistics: 30 days of live signal-channel data to 13 July 2026 (46,917 signals, 711 distinct contracts), windows measured as first-seen → last-active per contract in the screener database. Parity scan: OptionsLab benchmark snapshot of 13 July 2026, 922 call/put pairs with two-sided quotes across ClickOptions, Deribit, Binance, Bybit; near-ATM = strikes within ±5% of Binance spot; r treated as 0 for the short-dated tenors scanned; edges reported gross of fees and before size/freshness filtering, which is precisely why we treat the tails as suspect. ArbHunter performance figures are the live account's time-weighted series.

This article is market-structure research, not investment advice. Arbitrage strategies carry execution, margin, and counterparty risk; "riskless at expiry" is not riskless along the way.