How to Trade Arbitrage: Spreads, Costs, and Execution Risk

What Is an Arbitrage Strategy, and How Does It Differ from a Directional Trade?

Arbitrage is built on one idea: the same economic risk is priced differently in two places, and that difference can be captured. The U.S. Securities and Exchange Commission describes the mechanism in its bulletin on exchange-traded funds: authorized participants buy where the price is lower and sell where it is higher, and their actions themselves bring the price back toward fair value. There is an important caveat: in that market, this is done by a small group of large professional participants, not just anyone.

“Arbitrage is the practice of profiting from differences in prices in two or more markets.”

• U.S. Securities and Exchange Commission, investor bulletin on exchange-traded funds, https://www.sec.gov/investor/alerts/etfs.pdf

The difference from a conventional trade is fundamental. In a directional position, the trader predicts where the price will move. In an arbitrage combination, the market's direction hardly matters: the two legs are opened in opposite directions, so a rise in the asset adds exactly as much to one leg as it takes away from the other. The profit lies in the spread between them, while the entire risk shifts from forecasting to execution.

Arbitrage must therefore be evaluated differently from a conventional setup. There is little point debating whether the asset will rise; what matters is whether you can execute at both prices. The overall testing discipline remains the same as for any system: the strategy rules and the method used to test them must be written down before the first trade.

Which Arbitrage Strategies Are Used on Exchanges?

Arbitrage strategies can be classified by what exactly is diverging in price:

1. Cross-exchange arbitrage. The same asset is priced differently on two venues. You buy on the cheaper venue and sell on the more expensive one; the difference is the profit before costs.
2. Spot-futures combination (cash-and-carry). A futures contract trades above the underlying asset. You buy the asset, sell the contract, and wait for the prices to converge by the settlement date.
3. Reverse combination (reverse cash-and-carry). This is the mirror image: the contract is cheaper than the asset, so the asset is sold and the contract is bought.
4. Triangular arbitrage. A discrepancy among three currency or trading pairs on one venue, where the product of the exchange rates does not reconcile.
5. Statistical arbitrage. The difference is not fixed by a contract but inferred from the historical behavior of two related instruments.

The first four classes rely on a relationship enforced by market mechanics. The fifth relies on statistics and therefore behaves differently; it is covered in a separate section below.

How Cross-Exchange Arbitrage Works

The structure looks simple until real money is involved. A price difference between two exchanges lasts for seconds, while the funds must already be available on both venues: transferring the asset during the trade is not feasible because network confirmations take minutes or hours. Capital must therefore be split between the venues in advance, and the return must be calculated on the entire deposit tied up, not on just one leg.

Diagram 1. Sequence of steps in a cross-exchange arbitrage combination. Source: exchange deposit and withdrawal mechanics.

Speed matters more than analysis here. NYU Stern lectures state the first two requirements for success in pure arbitrage directly: access to real-time prices and instantaneous execution. The practical figures look just as unforgiving. The author of an analysis on Habr ran his own data collector and recorded 15 256 arbitrage opportunities: he estimated that 90.8 percent were short-lived price spikes that were virtually impossible to act on manually, while only four passed a manual review for suitability in real trading. This was someone else's experiment, not our measurement, but it gives an honest sense of the scale.

A manual trader is physically unable to submit two orders within the same second on different venues, so these combinations are almost always automated. The relevant questions then concern testing the algorithm itself, not reading a chart.

How a Spot-Futures Cash-and-Carry Strategy Works

A futures contract and its underlying asset are linked by time. When settlement is still far away, the contract may cost more than the asset because its price incorporates interest, storage, and participant expectations. As the settlement date approaches, fewer reasons for the difference remain and the prices converge: by that point, holding the contract until delivery and holding the asset itself become equivalent.

The difference between the cash price and the price of the nearest contract is called the basis, while the risk of its unexpectedly widening or narrowing is called basis risk. Both terms appear in the glossary of the U.S. Commodity Futures Trading Commission. It is important to understand that convergence does not necessarily produce an exact zero: the profit is not the entire basis, but only the portion that remains after deductions.

In practice, the trader buys the asset, simultaneously sells the contract, and holds both legs. The projected result of the combination is known in advance and equals the entry basis before deductions. Everything associated with holding the position must still be deducted: interest on borrowed funds, margin required for the short leg, variation margin, and fees. A separate guide explains how margin, collateral requirements, and futures contract expiration work.

Diagram 2. The life of the basis in a spot-futures combination. Sources: definitions of basis and basis risk in the U.S. Commodity Futures Trading Commission glossary and the derivation of fair futures prices in NYU Stern lectures.

Perpetual contracts have no settlement date, so their mechanics are different: instead of convergence toward delivery, they use a periodic funding rate exchanged between buyers and sellers. Expiration rules cannot be applied to these instruments; they require a different set of calculations.

How Statistical Arbitrage Differs from a Fixed Price Difference

In the first four classes, the combination is supported by a market rule: a contract must converge with the underlying asset, and three currency pairs must close the triangle. Statistical arbitrage has no such obligation. There is only the observation that two instruments historically moved together and a bet that they will converge again after diverging.

In its description of educational material on arbitrage strategies, the Moscow Exchange explicitly separates these concepts: futures-underlying arbitrage is treated separately from statistical arbitrage and trading the volatility of the basis. The venue classifies the first combination among the least risky. That is its assessment, not ours, but the distinction itself is important: these strategies cannot be measured by the same standard.

Diagram 3. Strength of the relationship between legs in different types of arbitrage. Sources: the classification of arbitrage types in Moscow Exchange materials and the definition of basis in the CFTC glossary.

The practical conclusion is simple. A pair may diverge more than it ever has historically and never return: a company's business may change, or the structure of demand for an asset may shift. Statistical arbitrage therefore requires a stop-loss order and a position limit, just like any directional trade, whereas in a classical combination the settlement date itself serves as the convergence deadline.

How to Calculate the Executable Spread Using Bid and Ask Prices

The main beginner mistake fits into a single line. A trader sees that an asset is priced at 99.90 on exchange A and 100.50 on exchange B, then calculates a difference of 0.60 percent. But 99.90 and 100.50 are last trade prices, while an order can only be executed against a matching order in the book. A purchase executes at the ask and a sale at the bid, and both prices are worse than the figure displayed in the terminal header. The actual entry prices will be closer together, and the calculated 0.60 percent will shrink before the first fee is even charged.

The formula for the executable spread is:

gross_spread = (bid_sell_leg - ask_buy_leg) / ask_buy_leg * 100%
net_spread   = gross_spread - fees - slippage - holding_costs - transfer_costs

Before entering, follow these steps:

1. Open the order book on both venues and capture prices at the same moment.
2. Use the ask where you are buying and the bid where you are selling. Do not use the last trade price.
3. Check that the intended size fits within the order book on both sides without reaching deeper price levels.
4. Calculate the gross spread using the formula above.
5. Deduct all costs and review the result. A negative net spread settles the question.

This calculation has an academic form that is useful because it states its assumptions honestly. In NYU Stern lectures, the fair value of a futures contract is derived from a simple idea: buying a contract and borrowing money to buy the asset are two routes to the same result, so they should cost the same. But caveats follow: the model assumes that money can be borrowed and invested at the same rate and that buying and short-selling involve no costs. Once the borrowing and lending rates are separated and costs are added, fair value stops being a single point and becomes a range. The practical meaning is direct: while the spread remains within that range, there is no trade. Profit begins only beyond its boundaries.

The order book exists in the exchange terminal, not on the chart. Technical analysis platforms usually lack access to the order book and third-party data, so the executable spread must be measured where the orders actually reside, and the trade must be executed there as well.

Which Fees and Costs Can Consume the Price Difference?

The price difference on the screen is the return before deductions. Everything else is deducted along the way, often twice: an arbitrage combination has two legs, and almost every cost is paid on each of them.

Diagram 4. Order of deductions from the spread. Sources: trading-venue fee schedules and contract specifications.

The execution fee model also matters. Under the common maker-taker model, the venue pays a rebate to the participant who posts a quote and adds liquidity, while charging the participant who removes that liquidity. Exchange fee documents list these rates separately, and the figures differ. Arbitrage is time-sensitive by definition and therefore tends to remove liquidity, so it pays the more expensive side of the fee schedule on both legs at once.

Why the Two Legs Must Be Executed in Coordination

This is the most underestimated risk in arbitrage. While both legs are open, market direction does not matter to you. As soon as the second order fails to execute, that neutrality disappears: you are left with an unhedged long or short position that you never intended to hold and for which you calculated neither a stop-loss order nor a position size.

This can happen routinely. The order on the second venue may be only partially filled because the order book lacks sufficient size. A transfer network may delay confirmation. A venue may slow order acceptance during a burst of activity. By then, the price difference that motivated the trade has usually disappeared, and the other side must be completed at a worse price.

Partial execution is not unusual. A U.S. Securities and Exchange Commission bulletin on order types examines exactly this situation using a simple example: in a fast market, half the size is executed at the expected price and the other half at a higher price. For a standalone purchase, this is a problem; for an arbitrage combination, it breaks the entire structure.

This leads to three rules:

1. Set the order size according to the least liquid side of the combination, not according to the desired profit.
2. Use limit orders where possible. Regulators describe the tradeoff directly: a market order provides execution certainty but leaves the price uncertain, while a limit order protects the price but may not execute at all.
3. Write down a contingency plan for a one-sided trade in advance: close it immediately at a defined loss or complete the second leg according to a specific rule.

A separate guide uses protective orders to explain exactly how a limit order differs from a market order and what execution risk means; the mechanics are exactly the same here.

How to Check Liquidity, Size, and Partial-Execution Risk

The check takes a minute and follows these steps:

1. Review the total order-book size across the price range your order will reach, separately for each venue.
2. Compare that size with the size of the leg. The greater the share of visible liquidity consumed by the order, the further it moves the price away from your calculation. In practice, problems begin long before an order consumes the entire book.
3. Check the instrument's average daily turnover on both venues, not only on the one where the spread looks more attractive.
4. Remember that venues use different data sources and that the same ticker on the spot and derivatives markets shows different volumes.
5. Include a buffer: the calculated spread should withstand execution worsening by at least a few hundredths of a percent.

Even on a major venue, some assets can have thin liquidity, and the price may reverse before it reaches a more distant order. A separate guide explains how to read depth and liquidity levels on a chart, but the final decision must still be based on the order book of the venue where you execute.

Educational Calculation: When a Positive Spread Turns Negative

All figures below are synthetic. The table illustrates the calculation mechanics. It does not present an expected return: every venue has its own fees, funding rates, and order-book depth, and those are the figures that must be used. Note that the prices in the table have already been taken from the order book, so the difference here is smaller than the one visible in the last trade prices in the section above.

Diagram 5. Step-by-step breakdown of the educational calculation. Source: synthetic example; figures are hypothetical.

Now let us make the example worse. Suppose the second leg is only partially executed: 60 percent of the size is filled, while the remainder stays in the order book. The remaining 40 percent of the first leg becomes a directional position and must be closed at the current price. If the market moves 0.10 percent against you during those seconds, the loss on the unhedged portion exceeds the entire profit on the hedged portion.

This leads to a practical rule: calculate the result using the worst execution you are prepared to accept. If the idea survives only when both legs receive perfect fills, that is already the answer, and there is no reason to test it in a live account.

Risks and Limitations

• An analysis tool does not guarantee a profit. A signal on the chart is a reason to check the order book, not an instruction to enter a trade.
• Markets are volatile, and trading involves the risk of losing capital. If the second leg remains open, the combination becomes an ordinary directional position.
• Past results do not guarantee future results. A divergence that historically closed may never close again.
• Costs can consume narrow spreads entirely. Fees, funding rates, and network charges change, so old calculations quickly become outdated.
• Capital is tied up on two venues at once, which means venue risk and withdrawal risk must also be included in the cost of the idea.
• Position size must be calculated in advance. The guide to risk per trade explains how much money may reasonably be lost on a single attempt.

Disclaimer. This is not personalized investment advice. Markets are volatile, and trading involves the risk of losing capital. Past results do not guarantee future results.