06 · Arbitrage in Practice

Arbitrage is the trading world's closest thing to a "money printer": no directional bets — only the certainty of spread convergence. In reality, though, arbitrage is full of traps: spreads that refuse to converge, liquidity drying up, fees devouring profit, black swans blowing up. This article lays out the core logic of arbitrage and seven mainstream plays — principle, operational steps, risk points, who they suit — ending with the 5 most common ways arbitrageurs die.

The Core of Arbitrage

What arbitrage earns is essentially the certainty of spread convergence, independent of price direction. One sentence:

Spread = the price difference of the same asset across markets/contracts/instruments
Arbitrage = buy the undervalued side + sell the overvalued side, close when the spread converges

Remember the first principle of arbitrage: spreads exist because markets are temporarily irrational; the arbitrageur's job is to wait for rationality to return. But if the market never returns to reason, you are the one being harvested. So the real skill isn't "finding spreads" — it's "assessing whether the spread will converge, and when".

1. Cash-and-Carry (Futures-Spot) Arbitrage

Principle

Futures and spot prices have a theoretical relationship: futures price = spot price + cost of carry (funding interest + storage + freight + insurance). When the futures-spot gap (basis = spot price − futures price) deviates from fair range, an opportunity appears:

• Contango (futures > spot): basis negative → theoretically "buy spot + sell futures", hold to delivery and capture convergence;
• Backwardation (futures < spot): basis positive → reverse: "sell spot (or short via securities lending) + buy futures", again waiting for convergence.

Numeric example (gold):

Gold spot 600 CNY/g, December futures 615 CNY/g, one-month carry ≈ 5 CNY/g
→ fair futures price should be 605 CNY/g; actual premium 10 CNY/g (abnormally rich)
→ Arb: buy 10 kg spot (6M CNY), sell 10 lots December futures
→ At delivery the gap reverts to carry level; locked profit ≈ (10 − 5) × 10,000 g = 50,000 CNY

"Long backwardation / short contango" logic: without touching physical delivery, when basis is positive (backwardation), going long futures equals "buying future goods cheaper than spot today"; profit arrives as the market normalizes (backwardation narrows). Shorting futures in contango works symmetrically.

Steps

1. Compute theoretical basis (cost-of-carry model);
2. Compare with actual basis; act only when deviation exceeds costs (fees + funding + market impact);
3. Open both legs: buy the cheap side, sell the rich side, matched by notional;
4. Hold until basis reverts and close, or hold to delivery (requires spot channels/warehousing);
5. Monitor basis throughout; set a "basis won't revert" stop-loss line.

Risk points

• Basis may not revert: in extreme markets (squeezes, roll squeezes), basis can keep widening;
• Heavy capital usage: buying spot requires full payment; returns diluted by cost of capital;
• Delivery risk: individuals cannot take physical delivery — must swap or roll;
• Squeeze: when the spot side is cornered, shorts are forced to buy back at ruinous prices.

Suited to

Institutions, industrial players, investors with large capital (millions+).

2. Calendar (Inter-Temporal) Arbitrage

Principle

Contracts of different months on the same instrument (e.g. rebar January vs May) differ mainly by time value (funding interest, storage, seasonal supply-demand). When the far-near spread deviates from its normal range, do long near month + short far month (or reverse) and bet on convergence. Directional exposure is hedged; only the "calendar spread" risk dimension remains.

Positive/negative spreads:

Numeric example:

Rebar Jan contract 4,000 CNY/t, May 4,200 CNY/t; normal spread ~150 CNY/t
→ actual spread 200 CNY/t (too wide, near month undervalued)
→ Positive spread: buy 10 lots Jan + sell 10 lots May
→ Spread reverts to 150: profit (200−150) × 10 t × 10 lots = 5,000 CNY

Steps

1. Establish the historical normal range of near-far spreads (mean ± 2 standard deviations);
2. Open when the spread breaches the range with an explainable logic (inventory, maintenance, seasonality);
3. Use liquid benchmark month pairs; avoid illiquid near months;
4. Close or roll before the delivery month approaches (last 10 trading days);
5. Take-profit: spread back at range midpoint; stop-loss: spread extends to historical extremes (e.g. 3 standard deviations).

Risk points

• Structural widening: long-term fundamental shifts (new capacity online) move the whole "normal range";
• Roll slippage: spread and slippage during main-contract switches;
• Squeeze: near-month squeeze forces the positive-spread short into massive losses;
• Margin usage: two-sided positions mean high margin and heavily compressed leverage.

Suited to

Intermediate investors with 100k-1M capital who tolerate low returns. Small swings, clear logic — a good starting point for learning arbitrage, but beware turning "positive/negative spreads" into disguised directional betting.

3. Cross-Market Arbitrage

Principle

The same commodity trading on different markets should converge in price (after freight, tariffs, FX). When domestic-international gaps turn abnormal, buy the cheap market + sell the expensive one and wait for convergence. Classic pairs:

FX is the most critical variable in domestic-international arbitrage: SHFE and London gold are denominated in different currencies; spreads must be converted into a common currency. With USD/CNY at 7.2 and London gold at $2,000/oz:

Theoretical SHFE price = 2000 × 7.2 ÷ 31.1035 ≈ 463 CNY/g (1 oz ≈ 31.1 g)
If actual SHFE gold is 480 CNY/g → domestic premium 17 CNY/g (abnormally high)
→ Arb: buy London gold + sell SHFE gold, wait for premium to normalize

Steps

1. Normalize units and currency (CNY/g vs USD/oz; barrels vs tonnes);
2. Compute theoretical spread including FX, freight, tariffs, premia/discounts;
3. Open both legs when actual deviation exceeds operating cost;
4. During volatile FX periods hedge the exchange rate simultaneously (forwards/options) or cut size;
5. Close when the spread converges — never bet direction.

Risk points

• FX risk: sharp RMB moves swallow the entire spread profit (one of the most common deaths);
• Time mismatch: different trading hours onshore/offshore; overnight gaps;
• Liquidity differences: night sessions vs offshore hours misaligned — closing one leg may find no counterparty;
• Cross-border capital: deposit/withdrawal restrictions, FX controls, conversion costs;
• Policy risk: trade or quota changes permanently shift the spread midpoint.

Suited to

Investors with overseas accounts, FX/tariff fluency, and meaningful capital. Domestic-international spreads are where professional institutions camp permanently — retail should be extra careful: the "onshore-offshore inversion" you read about in the news is usually already the endgame after institutions finished positioning.

4. Cross-Commodity Arbitrage

Principle

Upstream-downstream commodity prices are strongly correlated: processing margins (crushing, smelting, rolling) define the fair range between them. When margin compresses to extremes or balloons absurdly, long "the undervalued leg", short "the overvalued leg", betting on margin normalization.

Classic logics:

Numeric example (crush spread):

Soybeans 4,000 CNY/t, meal yield 78%, oil yield 19%
Meal 3,800 CNY/t, soybean oil 8,000 CNY/t
→ output value = 3800×0.78 + 8000×0.19 = 2964 + 1520 = 4,484 CNY/t
→ crush margin = 4484 − 4000 − 100 fee ≈ 384 CNY/t (historical mean ~150)
→ Margin abnormally high → buy beans + sell meal/oil, wait for margin to fall back

Steps

1. Understand the industry chain (who processes whom, yield ratios, cost structure);
2. Compute processing margin; plot its historical range (mean ± std dev);
3. When margin exits its range with explainable fundamentals, position against it (short high margin, long low margin);
4. Match legs by yield ratio (e.g. 1 lot beans ≈ 0.78 lots meal + 0.19 lots oil, adjusted for contract multipliers);
5. Set margin-reversion targets and extreme-value stops.

Risk points

• Permanently changed ratios: breeding cycles, capacity exits shift the spread midpoint;
• Contract specification mismatch: quality standards and delivery rules break "1:1 hedging";
• One-sided market risk: in macro crashes/rallies the two legs desynchronize — a "pseudo-hedge";
• Margin calls: while margin keeps extending, both legs float against you, forcing top-ups.

Suited to

Investors who know the industry chain and can research (capital 100k-1M). Cross-commodity arb demands continuous fundamental tracking — "research-driven arbitrage", a higher bar than cash-carry or calendar arb, but the most interesting logic of all.

5. ETF Arbitrage

Principle

An ETF has two prices: the secondary-market trading price (real-time matching) and NAV (real-time weighted basket value). When they diverge (discount/premium), the creation/redemption mechanism locks the spread:

• Premium (market price > NAV): buy the stock basket → create ETF units → sell on-exchange → pocket premium;
• Discount (market price < NAV): buy ETF on-exchange → redeem into the basket → sell shares → pocket discount.

Numeric example:

ETF NAV 2.000 CNY, market price 2.010 (premium 0.5%)
→ create 100k units: basket costs 200k, sell on-exchange for 201k
→ gross 1,000 CNY; net ~500-800 CNY after commissions, creation/redemption fees, slippage

Steps

1. Monitor premium/discount rate (premium = market price ÷ NAV − 1);
2. Trigger only beyond the cost line (~0.2%-0.5%, depending on liquidity);
3. Direct creation/redemption requires thresholds (typically 500k-1M units or equivalent value); small accounts can only trade the premium's fluctuation on-exchange — that's speculation, not arbitrage.

Risk points

• Creation latency: seconds-to-minutes delay between creation and selling; NAV moves eat the spread;
• Thresholds and fees: creation/redemption fees, constituent slippage, T+0/T+1 constraints (some ETFs allow same-day redemption);
• Liquidity trap: small ETFs show fat premiums you cannot actually sell into;

Suited to

Institutions/professionals with programmatic capability or large capital.

6. Crypto Market Arbitrage

Principle

Crypto trades 24/7, exchanges are numerous, leverage tools abundant — the densest arbitrage terrain globally:

• Spot-perpetual funding-rate arbitrage: perpetuals never settle; "funding rate" balances longs and shorts. When funding is positive (longs pay shorts), long spot + short perpetual collects funding each cycle (usually every 8 hours);
• Cross-exchange spreads: the same token prices differently on Binance, OKX, Coinbase etc. (especially during big moves); buy low, sell high;
• Cash-carry/calendar: same as the futures chapter, but shorter tenor and smaller size.

Numeric example (funding arb):

BTC spot 60,000 USDT; perpetual funding annualized +25% (common in bull markets)
→ deploy 100,000 USDT: long spot + short 1 BTC perpetual (hedged)
→ collect ~25% × 100,000 = 25,000 USDT funding annually
→ after fees, spread losses, slippage: realistic annualized ~10%-18%

Steps

1. Pick exchanges and instruments; confirm funding rates, position limits, withdrawal fees;
2. Open equal-sized legs: buy spot + short perpetual (size = spot quantity × price);
3. Each settlement cycle (8 hours), verify funding receipts;
4. Exit when funding turns negative (or fees erode the arb return);
5. Cross-exchange spread play: monitor → fire both orders fast → close on receipt.

Risk points

• Extreme speed requirements: cross-exchange gaps close within tens of seconds; manual execution almost always loses — API automation required;
• Platform risk: exchanges pulling plugs, halting withdrawals, or vanishing (black-swan death is especially common in crypto);
• Funding-rate volatility: bull-to-bear flips turn funding negative instantly and the arb loses in reverse;
• Imperfect hedge: underlying/caliber differences between spot and perpetual funding calculations;
• Capital movement costs: fiat ramps, on-chain fees, slippage eat small-spread profits.

Suited to

Intermediate investors with programming ability who accept platform risk. Funding arb is one of few "quasi-arbitrages" open to individuals, but it is essentially a beta strategy earning volatility compensation: collecting rent in bulls, paying rent in bears — not passive income.

7. Statistical Arbitrage (Concept)

Principle

Statistical arbitrage (Stat Arb) skips fundamental logic: use historical data to model the relationship between two assets' prices; when the relationship deviates, long the cheap one and short the rich one, betting on reversion. Core concepts:

• Mean reversion: many pairs' price differences (spreads, ratios) oscillate around a long-run mean; the farther the deviation, the higher the reversion probability;
• Cointegration: two price series each random-walk, but their linear combination is stationary — i.e. "this spread won't run away". After testing cointegration, fading deviations beyond 2 standard deviations is textbook stat arb.

Numeric example (pairs trading):

Kweichow Moutai vs Wuliangye: historical spread mean 50 CNY, std dev 8 CNY
→ current spread 70 (> mean+2σ = 66)
→ short Moutai + long Wuliangye (equal notional), bet on reversion
→ close at spread 50, earning ~20 CNY/share × position

Steps

1. Find highly correlated asset pairs (same sector, chain, or index constituents);
2. Run cointegration tests on historical data; compute spread mean and standard deviation;
3. Enter at ±2σ deviations, exit at mean reversion, stop-loss beyond ±3σ;
4. Match positions by "equal notional / equal beta" to control portfolio volatility;
5. Roll tests continuously: if the relationship structurally breaks, retire the pair.

Risk points

• Relationship failure: statistical patterns may be historical coincidence; after structural change the spread never reverts (stat arb's biggest killer);
• Backtest overfitting: parameters reverse-engineered from history fail live;
• One-sided markets: in systemic crashes the shorted leg squeezes — both legs lose together;
• Capital requirements: 10-20 diversified pairs are needed for smoothing — hard for individual capital to sustain.

Suited to

Quant-literate investors with 1M+ capital. Stat arb is quant funds' home turf; for individuals it's mostly "learning the concept" — understanding cointegration and mean reversion helps judge any spread's convergence odds.

5 Common Ways Arbitrage Traders Die

What these 5 share: arbitrageurs die from treating low risk as no risk. True arbitrage discipline is defining "what if it doesn't converge" before discussing "how much we make".