Inventory & Rebalancing

Atomic on-chain arbitrage has a beautiful cheat code: the flash loan. You borrow a million dollars, do the trade, and repay it inside a single transaction — if any step fails, the whole thing reverts and you never needed the capital at all. Cross-chain arb does not get that toy. There is no flash loan that reaches across chains, because a transaction on Chain A cannot atomically wrap a transaction on Chain B. The money has to be yours, and it has to already be sitting on both chains before the opportunity shows up.

That single fact — capital must be pre-positioned, finite, and idle — is the whole economic story of this lesson. It turns a strategy that looks like free money per trade into a capital-intensive business with a return-on-capital problem.

Why you must pre-position on both sides

Picture a currency-exchange kiosk at an airport. To serve a tourist who wants euros for dollars, the kiosk must keep both dollars and euros in the till. It cannot conjure euros the instant a tourist walks up — if the euro drawer is empty, the deal simply doesn’t happen, no matter how good the rate is. Cross-chain arb is exactly this kiosk, run across two blockchains.

Definition. Inventory is the working capital you hold on each chain so you can fill a leg the moment a price gap opens. For our canonical setup you hold the buy-side currency (USDC) and the sell-side asset (ETH) on each chain, so you’re ready to trade in either direction:

Worked example. You pre-position $2,000,000 of working capital on Chain A and $2,000,000 on Chain B. Total capital committed:

That’s $4,000,000 parked across two chains before you’ve made a single dollar of profit, just so both kiosk drawers are stocked.

When it matters

Pre-positioning matters the instant you size a cross-chain strategy. If you budget like an atomic arber — “I only need gas, the flash loan does the rest” — you’ll model a capacity that doesn’t exist. The real question isn’t “how big is the gap?” but “how much inventory am I willing to lock on each chain, idle, to chase it?”

Same-direction flow drains one side

Now suppose the mispricing is persistent: ETH is reliably cheaper on A than on B, so every trade is “buy on A, deliver on B.” Back at the kiosk: every customer hands you dollars and wants euros. Your euro drawer empties, your dollar drawer overflows, and eventually you’re out of euros and have to close the window — even though customers are still lining up.

Drag the slider below until one side hits zero. This is the whole pitfall in one picture: push net flow toward “sell on B” and watch the B drawer drain while A overflows. When B hits zero you’ve hit capacity for this cycle — the gap can stay wide open and you still can’t trade.

Definition. Same-direction (one-way) flow is a sequence of arbs that all consume the same chain’s inventory. Each trade subtracts your trade size from the depleting side:

where $s$ is the per-trade size and $n$ the number of trades. Inventory hits zero when $n = \text{Inventory}_B^{(0)} / s$.

Worked example. You start with $2,000,000 on B and trade $200,000 of ETH per arb:

You get about 10 trades before B is exhausted and the window slams shut — regardless of how many more juicy gaps appear in block 11.

When it matters

It matters the moment your flow stops being two-directional. A market that oscillates (cheap on A, then cheap on B, then back) self-rebalances for free. A market with a structural one-way lean — one chain that’s persistently the buy side — drains you steadily and forces paid rebalancing. Know which regime you’re in before you commit inventory.

Capacity is capped by inventory, not opportunity

Here’s the line that should be tattooed on every cross-chain desk: throughput per cycle is set by your inventory on the depleting side, divided by your trade size — full stop. The size, frequency, and lusciousness of the gap don’t enter the formula. You already met this shape in the prerequisite course as the AMM-depth ceiling: pool depth capped how much you could trade before slippage ate the edge. Same idea, new bottleneck — now it’s your own capital that’s the depth.

Definition / formula. Cycle capacity is the number of arbs fundable before you must rebalance:

Bigger inventory buys more trades per cycle — but it also locks more capital. The dial that turns up capacity is capital, and capital is never free.

Worked example. Doubling depleting-side inventory from $2,000,000 to $4,000,000 (at $s$ = $200,000) doubles trades per cycle from 10 to 20 — but it also doubles total capital locked to $8,000,000. You bought throughput with capital, one-for-one.

When it matters

Capacity-as-inventory matters whenever you’re forecasting volume or pitching size. “We can capture this all day” is only true up to inventory ÷ trade size trades — after that you’re idle until a (costly, slow) rebalance. Forecast in cycles, not in opportunities.

Rebalancing cost & time

So B is empty and A is overflowing. To reopen the window you must rebalance: rebridge inventory from the full side back to the empty side. That costs the bridge fee, and — because this is itself a cross-chain transfer — it eats the settlement/finality wait too. While the capital is in transit it is unusable: not trading, not earning, just floating between chains like luggage on a delayed flight.

Definition / formula. The rebalancing fee is the bridge rate times the amount moved; amortized per trade it’s that fee spread across the trades the rebalance re-enables:

where $r$ is the bridge rate (5 bps here), $M$ the amount rebridged, $g$ the gross spread captured per trade, and $n$ the trades the rebalance unlocks.

Worked example. You rebridge $1,000,000 back to B at a 0.05% bridge fee:

plus roughly 4 minutes of downtime while that $1,000,000 settles (no trading on those funds meanwhile). That $1,000,000 refills 5 trades of $200,000. Amortized, the rebalance costs:

On a $200,000 trade, $100 is 5 bps of drag — small if you rebalance rarely, but a recurring tax that grows the more often you have to do it. Rebalance after every 2 trades instead of every 5 and the per-trade tax climbs sharply.

When it matters

Rebalancing cost matters most exactly when the opportunity looks best: a strong, persistent one-way gap. The stronger the lean, the faster you drain, the more often you rebalance, and the heavier the amortized tax. Net it in before scaling, not after.

The capital drag (return on capital)

Now the honest part — the part that turns a trader’s grin into an accountant’s frown. You’ve locked $4,000,000 across two chains. That capital is idle: it isn’t earning staking rewards or lending yield; it’s sitting in drawers waiting for tourists. The right scorecard for cross-chain arb is therefore not profit per trade. It’s return on capital (ROC): net profit per period divided by the capital you tied up to earn it.

Definition / formula. Return on capital over a period:

And the opportunity cost is the yield that same capital would earn elsewhere (staking/lending) — the benchmark your arb must beat.

Worked example. One cycle: ~10 trades, each $200,000 at ~0.7% net edge (our clean post-fee, post-drift assumption from the prior lessons):

So $14,000 per cycle on $4,000,000 locked — a per-cycle ROC of $14{,}000 / 4{,}000{,}000 = 0.35\%$. Now annualize and benchmark. Suppose a cycle (drain + rebalance + redeploy) takes about a day; very roughly 250 trading cycles a year gives:

That looks fantastic — but it assumes a fresh, full cycle every day with no idle gaps and rebalancing already netted out. Thin the gap, add downtime, rebalance more often, and that number falls fast. The discipline is: compare whatever you actually clear against simply lending the $4,000,000 at, say, 5% ($200,000/yr risk-light). If your honest, rebalancing-and-downtime-adjusted arb nets less than $200,000 a year, you’d have been richer — and calmer — just lending the capital.

When it matters

ROC matters at the only moment that counts: deciding whether to run the strategy at all, and at what size. A backer doesn’t care that each trade was green; they care whether their $4,000,000 beat the boring alternative. Lead with ROC, net of rebalancing and downtime — and be honest about the cycles you’ll actually complete, not the ones you could in a frictionless dream.

Recap