6 minute read · Ultimo Research Desk · Reviewed 5 Sept 2026
Arbitrage and Latency Strategies: What They Are, What They Require and Where They Fail
Arbitrage seeks to capture a price difference for economically equivalent exposures by buying one side and selling another, while latency strategies focus on differences created by the timing of information or quotes. They depend most on the price difference being simultaneously executable and larger than every fee, delay, hedge, settlement and operational cost.
What it is
Pure arbitrage is a relative-price activity. Examples include buying an asset on one venue while selling the same asset on another, trading a security against a related future or forward, or exploiting a conversion relationship between instruments. Statistical arbitrage is broader: it uses a modelled relationship that can diverge and therefore carries estimation and market risk. Latency strategies are a subset of timing-sensitive activity in which a faster participant reacts to a price or information change before a slower quote or venue has updated.
These activities exist because markets are fragmented, information travels at finite speed, order books have queues and equivalent exposures can be priced differently for a short period. Institutional desks may use exchange memberships, co-located servers, direct market access, proprietary market data, automated order management, financing lines and controls for settlement and inventory. The infrastructure is part of the method, not an optional enhancement.
A retail CFD account usually provides a quoted contract rather than simultaneous access to the underlying exchange order books. The quote, timing, hedging and contract terms are different from a two-venue exchange arbitrage. Latency arbitrage against a broker’s own price feed is treated by most brokers, including under standard client agreements, as abusive trading, and our own trading rules follow that practice. It can lead to account closure. That is a factual description of the contract and market structure, not a statement that every relative-value analysis is prohibited.
What it requires
Arbitrage requires two or more genuinely accessible prices, the ability to transact both legs, and a method for keeping the exposures matched. Direct market access, co-located infrastructure and high-quality market data are common requirements for latency-sensitive institutional activity. The system must measure the age of each quote and reject stale or non-executable prices.
Capital is needed for both legs, collateral, settlement timing and inventory while one leg remains unfilled. A spread that appears riskless before fees can become adverse when the second order misses. The instruments must have compatible contract sizes, currencies, corporate actions, trading hours and settlement rules. Legal, compliance and market-access restrictions are part of the operating environment.
Technology must be tested for clock synchronisation, order duplication, feed loss, network failure and partial fills. Human screen time is less important for automated latency strategies than continuous system monitoring and the ability to stop the process safely. A statistical arbitrage model requires additional research time because its relationship can break even when the two prices have moved together historically.
How it is implemented
An implementation defines the relationship, the executable price on each side, the minimum difference required after costs and the order sequence. A pure arbitrage process attempts to buy the cheaper exposure and sell the dearer exposure as close to simultaneously as possible. It then closes or settles both legs under a defined rule. A latency process measures the time between a reference price change and the target quote, then submits an order subject to queue position and fill uncertainty.
Position size is limited by the smaller available quantity, the collateral requirement and the maximum tolerated unmatched exposure. The process records expected and actual prices for each leg, including rejected, cancelled and partial orders. A price difference that cannot be traded at both sides is an observation, not an arbitrage opportunity. The same applies when the contract terms create a material basis difference.
Worked example
Assume one unit is quoted at 99.90 on venue A and at 100.10 on venue B. If the cheaper buy and dearer sell are both filled, the gross price difference is 100.10 − 99.90 = 0.20. Assume fees of 0.03 on each leg, total slippage of 0.04 and a hedge or inventory cost of 0.02. Total cost is 0.03 + 0.03 + 0.04 + 0.02 = 0.12.
If both prices remain executable, the residual is 0.20 − 0.12 = 0.08 price units.
If the second price moves before the sell fills and the realised difference is only 0.05, the residual is 0.05 − 0.12 = −0.07 price units. The numbers are illustrative. They show why simultaneous execution and the cost of unmatched exposure determine the result; seeing two prices on a screen is not enough.
Costs
Fees and dealing costs are paid on both legs. Bid and ask differences, exchange or venue charges, data fees, financing, borrow, settlement and currency conversion can all reduce the observed gap. Latency-sensitive systems also require hardware, network connectivity, co-location and software maintenance. These are operating costs rather than optional enhancements.
Slippage and queue position determine whether the displayed difference can be captured. A partial fill can leave a directional position. A failed hedge can turn a relative-value operation into an outright market exposure. Historical tests that use simultaneous mid-prices and ignore message timing do not reproduce the actual cost structure.
Where it fails
Arbitrage fails when one price is stale, indicative or not executable; when the second leg cannot be filled; or when the instruments are not economically equivalent. It can also fail after a market halt, feed interruption, corporate action, contract change or settlement delay. Statistical relationships can diverge for fundamental reasons.
Latency strategies face competition from faster systems, changing exchange rules and uncertain queue position. A retail account can also face contractual restrictions when activity depends on exploiting a provider’s quote delay. Behavioural errors include treating a visible difference as riskless, increasing size after an earlier favourable outcome, or leaving one leg open when the matching leg has failed.
Who it suits and who it does not
Institutional arbitrage and latency strategies suit desks with direct access, specialised infrastructure, technical staff, capital for both legs and formal risk and compliance controls. A statistical relative-value approach requires research capacity and the ability to withstand a relationship breaking.
They do not suit a retail CFD account that has access only to a provider’s quote and not to the relevant underlying venues. They are also unsuitable for anyone who cannot monitor unmatched exposure, settlement and operational failure. The distinction is structural: the required prices and execution rights are not the same.
What the evidence says
Budish, Cramton and Shim (2015) analysed the high-frequency trading arms race and argued that continuous-time market design creates value for very small speed advantages. The paper is directly relevant to why latency strategies exist, but it is a market-design analysis, not evidence that a retail account can reproduce institutional conditions.
Hendershott, Jones and Menkveld (2011) studied algorithmic trading and liquidity using exchange data. Their findings concern the relationship between algorithmic activity and market quality, not the outcome of a specific arbitrage operation. The academic literature documents speed, liquidity and market-structure effects; it does not turn a delayed retail quote into a reliable arbitrage venue.
Sources
- Budish, Cramton and Shim (2015), “The High-Frequency Trading Arms Race: Frequent Batch Auctions as a Market Design Response”, DOI.
- Hendershott, Jones and Menkveld (2011), “Does Algorithmic Trading Improve Liquidity?”, DOI.
- Menkveld (2013), “High Frequency Trading and the New-Market Makers”, DOI.
- John J. Murphy, Technical Analysis of the Financial Markets (1999), Google Books.
Educational only: Arbitrage and latency strategies depend on simultaneous access to executable prices and can fail through leg, technology, settlement or contract risk.
Written by the Ultimo Research Desk and checked against our own contract specifications and client agreement before publication; reviewed again when those change. Educational only — nothing here is a recommendation to trade. Spotted an error? Tell us.
