Market dynamics chapters explained why arbitrage exists: YES-plus-NO parity, cross-market gaps, sibling trees, structural bundles. Platform chapters showed where you can trade. This chapter is how—capital on both sides, rule matching, partial fills, and when a gap is real versus a headline illusion.
Arbitrage is not free money. It is manufacturing a locked or bounded payoff when joint mispricing exceeds all-in friction.
Arbitrage chapters in market dynamics taught the types of gaps; platform chapters taught where capital sits. Execution is the unglamorous middle: wallets funded, rules matched, sizes clipped to depth, and a partial-fill story written before you start.
What you must verify first
Before any leg, confirm identical payoff definitions—same resolution source, same cutoff timezone, same event boundary. A Fed cut on one venue and a “Fed action” market on another may differ by one meeting. Currency and withdrawal paths matter: capital trapped on one side converts arb into a directional loan.
Gross edge must beat fees, slippage, gas, tax uncertainty, and a haircut for dispute risk. Size must fit the shallower book; walking both ladders is mandatory. If you cannot complete the second leg, the first leg is a bet. Legal access on both sides is gate one from venue selection—without it, the spreadsheet is fantasy.
Cross-platform binary sketch
Suppose matched rules on “Fed cuts 25bp at March meeting.” Kalshi YES ask 52¢, Polymarket NO ask 44¢. You pay 96¢ for $1 if rules truly match. Four cents gross sounds tasty until Kalshi fees, chain costs, a slippage buffer, and partial-fill risk leave half a cent—skippable for retail unless size is large and automation is fast.
Professionals often want three cents net on thousands of dollars because one stuck leg is gamma risk, not a bad day. Retail should default to skip when net edge after stress-testing fees is below a pre-set penny threshold.
Single-venue parity lag
On one book, YES ask 48¢ and NO ask 49¢ sum to 97¢. Buying both yields $30 gross per thousand pairs before fees—if both fill. Send limits together; if one fills, immediately work the other. Gaps on marquee contracts often live under a second when bots are awake; your edge may be infrastructure, not courage.
Unhedged leg risk is the story: you thought you were collecting four cents of bond-like yield; you are now short a binary view into the headline.
Partial fills and recovery
If leg A fills and leg B rejects or moves, recompute edge. Complete B only if edge remains; otherwise unwind A at a loss smaller than naked exposure. Withdrawal halts and dispute announcements freeze arb programs—mark open risk honestly.
Cross-contract violations on one venue—nomination rich versus president cheap on the same path—are usually tree trades, not two-app races, unless both legs list identically elsewhere. Structural index-versus-basket arbs demand spreadsheet proof of weights and expiries; default retail should paper-trade until joint payoffs are boring.
Capital, float, and tiers
Split bankroll across rails you actually use—USD regulated versus on-chain stablecoins—with a float buffer for partials. Fifty-fifty is a starting point; adjust for where depth lives. Manual traders live on alerts and spreadsheets for three-cent-plus edges; semi-automation pairs legs; pros co-locate APIs.
Kalshi-plus-Polymarket is the common serious pair when rules match; capped venues block scale. Treat arb books as low-variance, operationally risky positive EV—closer to bond substitutes than directional heroics.
Risk register in prose
Rule divergence is the silent killer—highlight text diffs before size. Invalid or void outcomes deserve a five-to-ten percent edge haircut. Gas spikes belong in the fee row. Account freezes mean stop the program, not hope. Reputationally, arb is not manipulation, but spammy false mappings pollute the ecosystem—be precise.
EV framing
Arb trades have low outcome variance if legs lock. Skill is operations and reading, not storytelling. Kelly on directional books does not copy cleanly: reserve bankroll for stuck legs and withdrawal timing. Journal both fills and basis at entry so monthly review can show whether arb sleeve actually added cents after frictions.
Retail versus pro expectations
If you are manual, assume marquee parity gaps are gone unless you have an alert on a slower listing. Your niche may be obscure cross-contract violations on one venue, or a new market before bots map it—not the Fed meeting everyone watches. Pros win on latency; retail wins on reading and patience when rules are new.
Tax and accounting note
Two legs on two apps can mean two taxable events and two fee schedules. Net edge is after both, not after the exciting spreadsheet cell. Keep a simple ledger: date, legs, gross cents, fees, gas, outcome.
When to walk away
If rule text differs by one word on timing, walk. If withdraw is delayed on either side, walk. If net edge is below your pre-set floor, walk. Arb programs that “almost work” are how directional blow-ups start.
Simulator habit on hybrids
Before any cross-venue clip, run the same mental simulator you use on pool markets: if exit would slip five cents, your locked edge was never three cents. Arb without exit depth is a one-way door.
Human error checklist
Map event IDs in a spreadsheet: Kalshi ticker, Polymarket slug, resolution PDF link, last verified date. Most arb blowups are wrong market, not slow fingers. Spend sixty seconds on mapping before you spend sixty dollars on legs.
Key ideas
Match payoffs, not headlines. Size to worse depth; recover partials fast. Treat arb as operations-heavy, low-variance EV—not free money.
What comes next
When you already have exposure and want to trade variance for certainty, hedging replaces some arb packages.
Alerting without automation
Retail can win with alerts on rule-matched pairs and a pre-computed minimum net edge. When the alert fires, run the checklist—do not click from memory. Most errors are mapping mistakes, not slow clicks.
Misconceptions
“Arb is free money” ignores partial fills and rule drift. “Same headline means same contract” kills accounts. “I will arb with one wallet funded” guarantees one-leg risk. “Bots took it so there is no edge” may mean your threshold is wrong, not that economics vanished.
Operational excellence—capital float, mapping, recovery—is the retail moat.
When net edge is thin, smaller clips succeed more often than hero size that partial-fills and leaves gamma.
Next: Hedging: Locking in Profits and Reducing Risk