Arbitrage: The Textbook Version Is Gone
Arbitrage is buying and selling the same asset simultaneously in two places at different prices, with matched size, so the position carries no market exposure. Textbooks call that a position without risk. In practice, execution, settlement and counterparty exposure all remain, and machines close the gap first.
How it works
Arbitrage has a strict definition that most uses of the word ignore. It is buying and selling the same asset at the same moment, in two places at different prices, in matched size — so the position carries no market exposure.
The matching is the whole idea. Long in one venue and short the same quantity in another leaves a difference, not a direction. Textbooks call that a position without risk; execution, settlement and counterparty exposure say otherwise.
The clean version is effectively unavailable to a person. The differences are small, brief and taken by machines that pay less to trade than you do: high frequency trading desks and market makers.
The kinds worth naming
Three mechanisms account for most of the real thing. Cross-venue is the plainest: the same asset quoted lower on one exchange than another, bought and sold at once.
Cash-and-carry sets spot against a future. You buy the asset, sell the contract, and collect the difference as it runs to expiry. Conversions rebuild an instrument from related ones and trade the copy against the original.
What retail is sold under the name is none of those. It is a spread position: two related things traded in the hope the gap closes, which can widen first. Statistical arbitrage and pairs trading are the honest names.
Small edges invite leverage, which is where it ends. A strategy earning fractions of a per cent must be scaled to matter, and scale turns a rare adverse move into a terminal one.
Measured here, exposure does not scale cleanly. Resetting exposure each bar before costs, the base returned 3.61%; two times exposure made 6.61% against a naive 7.22%, three times 8.93% against 10.82%. Drawdown went from 3.76% to 7.45% and 11.08%.
Execution risk is the one that bites. One leg fills, the other does not, and what you thought was matched is suddenly directional.
In practice
Crypto is where most people meet the pitch. Two venues quote the same coin differently, and the fees are where the difference goes.
The thinner leg decides the size you can do. Liquidity is rarely equal across venues, so volume on the worse side sets the trade and the order book sets your fill.
Convergence keeps no schedule. A difference that closes in a week and one that closes in six months pay the same amount; only the second ties up capital for months.
A hedge that is not open is not a hedge. When one venue trades while the other is shut, an opening gap lands on a single live leg.
There is no obvious stop loss on a spread. The legs move against each other by design, so the level that says you are wrong is a judgement.
One round trip on this site’s history costs 2% of a median bar’s range. A matched trade pays it on both legs, crossing the bid-ask spread four times.
Testing a claim before you believe it
Price both legs in full before calling anything an arbitrage. Count the trading fee on the buy, the trading fee on the sell, the withdrawal fee, the network or transfer fee, and any currency conversion on either side. A pitch quotes the price difference and leaves the rest for you to find.
Then time the transfer. Ask how long the asset or the cash takes to move between venues, and what the price can do while it is in transit. If the difference can close before you arrive, the trade was never simultaneous, which means it was never arbitrage.
Then ask what happens if only one leg fills. That question has a specific answer — a directional position, in a size chosen for a hedged trade, at a price you did not pick — and that answer decides whether the idea is worth doing.
What arbitrage is not
- Not a bet that two prices will converge. That is a spread, and it can widen first.
- Not matched if the legs fill at different times. A delay makes it directional.
- Not arbitrage once an asset must be moved. A transfer is not simultaneous.
- Not something a subscription hands you. A retail screen shows only priced differences.
When it fails
In a quiet market the spread simply does not move. A trading range can hold a difference open for weeks, and waiting costs every other use of the money.
On the smallest bars the cost is most of the move. The round trip here exceeds a tenth of a bar’s range on 15 of the 576 bars, and the smallest bar spans 0.022 against a median of 0.493.
The difference often exists for a reason. A withdrawal freeze, a trading restriction or a venue nobody trusts explains most prices that look wrong.
One venue is holding your money. Transfer limits, maintenance windows and failed networks are counterparty risk, and they arrive when the difference is widest.
Leverage sets the deadline. A levered spread can be closed by a margin call on the losing leg long before convergence arrives.
And the relationship can change. Convergence versions rest on a historical relationship holding; when the reason behind it goes, the spread has no obligation to close.
The original data
Of the 31,760 videos in research/search-study-corpus.jsonl, 896 teach scalping and three teach
arbitrage. Scalping runs a median of 19,949 views across 353 channels; arbitrage manages 2,143
across three channels, and the statistical version has one video, one channel, thirteen views.
The gap is not difficulty — arbitrage cannot be sold as an achievable retail activity, so nobody makes
the video. Counts: research/broker-coverage.json.
Then the cost, which is the whole argument. One round trip on this site’s shared 576-bar history is
0.0098 price units: 2% of a median bar’s range and 45% of the smallest, per site/measure_series.py
into research/series-measurements.json. An arbitrage edge is by definition smaller than the differences
inside a normal bar, so on quiet bars the cost is most of the opportunity: the business belongs to
whoever pays the least to trade. Price both legs, every fee and the transfer time before calling
anything an arbitrage; if the edge survives, ask who can do it faster.
Related
Statistical arbitrage is the honest name for the convergence version, and sets out what it does and does not hedge. High frequency trading takes the real differences: seven videos in the corpus cover it at a median of 121,256 views. And the bid-ask spread page carries the cost figure used here.
Every arbitrage anybody has brought me turned out to be a spread they expected to close. The word gets used because it sounds like certainty, and the thing underneath is a directional bet with extra steps. What I ask now is whether the position is matched in both venues at the same moment, and what happens if only one side fills. Nobody selling it has ever had a good answer to the second half.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.