The same asset trades at visibly different prices on different platforms, all day, in public. It looks like free money and it is one of the most reliable ways for a retail trader to lose some, because every element that makes the gap visible is also an element that makes it unavailable.
A price on a comparison site is the last trade or the mid quote on that venue, and neither is a price you can transact at. The last trade already happened, possibly for a size much smaller than yours; the mid is the midpoint between two prices you cannot get, because you buy at the offer and sell at the bid.
Correcting for that alone removes a substantial share of any observed difference. You would buy on the cheaper venue at its offer, not its mid, and sell on the more expensive one at its bid, not its mid. The real gap is between those two, and it is narrower than the displayed one by the sum of the two spreads before anything else has been counted.
Then there is size. The prices quoted are for whatever is at the top of each book, and an arbitrage worth the effort requires size, which walks both books in the unfavourable direction. The gap that exists at minimum size frequently does not exist at any size worth trading, and that is not a detail: it is why the gap persists at all.
A visible price difference that nobody has taken is usually not an opportunity that nobody noticed. It is one that does not survive being taken.
Assume the gap survives the spreads. It now has to survive the trading fee on the buy and the trading fee on the sell, which is two fees, both at the taker rate because arbitrage requires immediacy on both legs. It has to survive your own market impact on both books, in opposite directions, both against you.
It has to survive the transfer, which costs a network fee on the way and frequently a withdrawal fee set by the venue above that network cost. And it has to survive any conversion, if the two venues quote against different units, which adds a spread twice more.
Adding those honestly against a real fee schedule removes most observed gaps entirely. The exercise takes ten minutes and it is worth doing once with actual numbers rather than assumed ones, because the conclusion is stable: on liquid pairs between established venues, the difference that survives all four is close to nothing on the overwhelming majority of occasions.
The naive version of the trade is to buy on one venue, move the asset to the other, and sell. That sequence contains a period during which you hold the asset, in transit, unhedged, while the price does whatever it does. That period is minutes on a fast chain and considerably longer on others, and it is longer still because the receiving venue requires confirmations before crediting.
During that window the gap you were capturing can close, reverse, or become a loss much larger than it was ever going to be a gain. This is the single most common way the trade goes wrong, and it is not bad luck: gaps close because other people are also closing them, and the closing frequently happens during exactly the minutes your capital is in flight.
It gets worse in the situation that produces the largest gaps. Big price differences appear during violent moves, which is when networks are congested, confirmations are slow, and venues delay withdrawals. The conditions that create the opportunity are the same conditions that prevent the transfer, and that correlation is not accidental.
Anybody doing this seriously does not transfer. They hold balances on both venues in advance, so that when a gap appears they sell on one and buy on the other simultaneously, with no asset in transit at all. The trade becomes instant and the transfer problem disappears.
What replaces it is capital. You now hold the asset on one venue and the settlement currency on the other, permanently, in sufficient size to act. That capital is exposed to both venues as counterparties, it earns nothing while it waits, and the inventory drifts as you trade, so it has to be rebalanced periodically, which is the transfer you were avoiding arriving on a schedule instead of in the moment.
That is the actual economics of the strategy: you are being paid to hold inventory in two places and provide immediacy to whoever wants it. It is a real business and it is a market-making business rather than an arbitrage, with a return that has to compensate for the capital, the counterparty exposure, and the operational work of keeping it balanced.
Serious arbitrage does not move assets. It holds them everywhere in advance, which converts a trading problem into a capital and counterparty problem.
The intuition that a wider gap is a better opportunity is exactly backwards, and understanding why removes most of the temptation. A gap persists because something is preventing it from being closed, and the something is usually the thing that would also prevent you.
The most common causes are informative. Withdrawals are suspended on the cheaper venue, so nobody can move the asset out and the price there is a claim rather than a price. Deposits are halted on the expensive one. The venue is inaccessible from most jurisdictions, so the pool of participants who could close the gap is small. Or there is a solvency concern, and the discount is the market pricing the risk that the asset cannot be retrieved at all.
Each of those turns the gap into information rather than an opportunity, and the information is usually bad. A large, persistent discount on one venue is one of the more reliable early indicators that something is wrong there, and traders who treat it as free money are buying into exactly the situation the discount is warning about.
Triangular arbitrage exploits inconsistency between three pairs on the same platform: if the rates between three assets do not multiply out consistently, a cycle through all three returns more than it started with. It removes the transfer problem entirely, since nothing moves between venues, and it removes the counterparty question because everything happens in one account.
It also removes almost all of the profit, because the same absence of obstacles means everybody can do it, and the mispricings are consumed by automated systems in milliseconds. What remains for a human is the appearance of an opportunity on a screen that refreshes more slowly than the market corrects itself.
The honest description is that this is a fully solved problem on any venue worth trading on. Persistent triangular inconsistencies are a sign of a venue whose matching is slow or whose pairs are thin, which are properties you would not want for other reasons.
Participants who have capital pre-positioned on many venues, direct connections rather than public interfaces, fee tiers that a retail account cannot reach, and automated execution measured in milliseconds. They are not capturing large gaps; they are capturing very small ones, very often, at a scale where a fraction of a percent repeated thousands of times is a business.
That description also explains why the visible gaps look large to a retail participant and are unavailable. What those firms have taken is everything down to their own cost floor, and what remains above that floor is what nobody with lower costs could reach, which is a different statement from nobody having noticed.
This is worth stating without discouragement, because the same reasoning identifies where a smaller participant does have an edge, and it is not speed. It is in instruments and venues too small for the automated participants to bother with, where the gaps are real and the reason nobody has taken them is that the size available does not justify anybody's infrastructure. That is a genuine niche and it is a very different activity from watching a comparison page.
Three things, none of which involve running the trade. First, the price differences on comparison sites are a poor guide to anything, and using them to decide where to trade is using a number that includes none of your costs. Second, a persistent large discount on one venue is a warning rather than an offer, and it is worth checking why before it is worth acting on.
Third, and most usefully, the same arithmetic applies to a decision every trader actually makes: which venue to execute on. Comparing the price you would pay after spread, fee and impact, on the venues you already have accounts with, is the useful version of this analysis. It captures a small, reliable saving on every trade rather than a large, unreliable one on a trade you cannot execute.
The general lesson is one that transfers well beyond this strategy. An opportunity that is visible to everybody, on a public page, with no barrier to acting on it, has already been acted on by somebody with lower costs than you. The question worth asking about any apparent gap is not whether it exists but what is preventing it from closing, and the answer is usually the same thing that would prevent you.
Because each venue has its own book, its own participants and its own liquidity conditions, and closing a difference requires moving capital, which costs money and takes time. A gap smaller than the cost of moving between them is the normal state rather than an opportunity.
Four things before you start: the two spreads, since you buy at the offer and sell at the bid rather than at the mids being compared; two taker fees; your own impact on both books in opposite directions; and the transfer cost. Against a real fee schedule, most observed gaps disappear entirely.
Because during the transfer you hold an unhedged position while the price moves, and the gap frequently closes in exactly those minutes, since other people are closing it too. It is worse during the volatile periods that create the largest gaps, when networks are congested and venues delay withdrawals.
They pre-position balances on both venues so that both legs execute simultaneously with nothing in transit. That converts a trading problem into a capital problem: the inventory earns nothing while it waits, is exposed to both venues as counterparties, and has to be rebalanced periodically.
Because something must be preventing it from closing, and that something is usually withdrawals suspended, deposits halted, restricted access, or a solvency concern. A large persistent discount on one venue is among the more reliable early warnings that something is wrong there.
It removes the transfer and counterparty problems and removes almost all the profit with them, because the same absence of obstacles means automated systems consume the inconsistencies in milliseconds. Persistent triangular gaps indicate a venue with slow matching or thin pairs.
In instruments and venues too small for automated participants to justify their infrastructure, where the gaps are real and nobody has bothered. That is a genuine niche and it is a different activity from watching a comparison page, requiring the same operational work at a smaller scale.
Apply the same arithmetic to the decision you actually make: which of your existing accounts to execute on. Comparing the price after spread, fee and impact on each captures a small reliable saving on every trade, which is worth more than a large unreliable one you cannot execute.