Arbitrage · Can the spread actually be locked in?
Test synchronized execution, settlement constraints, and capital usage; distinguish arbitrage from risk-taking trades.
Observation
The same asset quotes differently across venues, or a redeemable basket deviates from a theoretical constraint. The screen spread has not yet deducted trade direction, executable depth, fees, financing, transfers, or settlement risks. Statistical arbitrage relies more on relationship repair and must be distinguished from lockable cash-flow constraints.
Mechanism
Participants able to execute and settle both sides compress spreads. Segmented capital, inventory limits, and venue risks let spreads persist. Without reliable conversion or delivery commitments, similar prices alone do not constitute an arbitrage mechanism.
Hypothesis
Prefund finite inventory at two venues. Attempt a trade only when the buy-side ask and sell-side bid leave a positive spread after all costs at a common executable quantity. Never assume instantaneous inter-venue transfers. Price inventory replenishment and profit realization too.
Data
Use synchronized two-sided books, quantities, fees, balances, withdrawal/transfer status, and settlement constraints. Reject signals if quote-arrival gaps are long enough to invalidate one side. Balances constrain execution; they are not a footnote added after a backtest.
Signal
Size to the minimum of both sides' depth, inventory, and risk limits. Produce two-leg plans and single-leg failure handling: cancellation, remedial fills, or reducing unhedged exposure within authorized risk. Without cross-system atomic execution, submitting plans does not mean both legs filled.
Backtest
Retain each leg's orders, fills, cancellations, rejections, and balance changes. Compare finite with infinite inventory; differences reveal capital and rebalancing constraints. Without a validated latency distribution, do not assert “actually lockable” returns. An executability upper bound is a valid first step if labeled as such.
Falsification
Increase delays, reduce depth, freeze withdrawals at a venue, or reject one leg. Returns depending on unlimited inventory, instantaneous transfers, or midquote fills invalidate the original conclusion. Later price convergence cannot erase interim one-leg losses or settlement failure.
Exercise and project acceptance
Build a finite-inventory, two-venue event table: quote arrivals, both leg submissions, fill confirmations, balances, and recovery actions. Deliberately reject one leg and apply the original rules. Submit net-spread calculations, time constraints, capital usage, and sources of failure losses. “Riskless” cannot replace these fields.
Connect execution evidence to Advanced E and capital governance to Advanced F.