Risk Management

Risk management in Matchpoly isn't a feature layered on top of the trading logic — it's the architecture. Every position opening, every order, and every strategy activation passes through a set of hard limits before execution. Those limits have veto power over everything else in the system.

// POSITION SIZING

Matchpoly sizes positions using the Kelly Criterion — a formula that calculates the mathematically optimal fraction of capital to allocate based on the estimated edge and the odds. Running full Kelly produces the fastest theoretical capital growth but also enormous swings; in practice, it creates drawdowns that are psychologically and operationally difficult to hold through. Matchpoly runs at approximately one-quarter of full Kelly, which significantly smooths variance while preserving most of the long-run growth advantage.

Regardless of what the Kelly calculation suggests, no single market position can exceed 5% of total capital. No single sport can represent more than 25–30% of deployed capital on any given day. And at least 30% of total capital stays undeployed at all times as a reserve — available for high-confidence opportunities and as a buffer against simultaneous losing positions.

Slippage is modeled before every trade. If filling the desired size would require walking up the order book far enough that the average fill price materially erodes the edge, the order is reduced or rejected. Executing at bad prices is worse than not trading at all.

// DRAWDOWN CONTROLS

Drawdown is tracked at three timescales simultaneously. Each timescale has a warning threshold and a halt threshold.

Daily: a 3% loss triggers a warning — all new position sizes are cut in half and the worst-performing strategy in that sport pauses. A 5% daily loss triggers a full halt: no new positions, all open orders cancelled, and no resumption until the cause is understood.

Weekly: the warning triggers at 8% down for the week, halt at 12%. Monthly: warning at 15%, halt at 20%. These thresholds exist because losing streaks happen in any probabilistic system. A -4% week inside a good month is noise. A -12% week is a signal that something in the market environment or the model has changed, and trading aggressively through it compounds the damage.

When a halt triggers, the system does not attempt to trade its way back. Existing positions are held through to resolution unless there's a specific reason to exit. No new positions open until a post-mortem review identifies what drove the drawdown and confirms it's been addressed.

// CORRELATED EXPOSURE

Markets tied to the same game aren't independent. A moneyline position on Team A, a spread position on Team A, and a prop bet on Team A's quarterback all move together when the game goes badly. Matchpoly tracks total correlated exposure per game — the sum of all dollar exposure across every market linked to the same event. When that total crosses a cap, no new positions open in that game regardless of how attractive any individual market looks.

This prevents the system from inadvertently building a large concentrated bet disguised as multiple small positions. A single bad game outcome should never be able to materially damage the overall portfolio.

// THE KILL SWITCH

A global kill switch operates at the infrastructure level — separate from the application code, so a software bug that disables other components can't disable the kill switch. It cancels all open orders and freezes new position opening. It does not force-close existing positions, because closing positions under emergency conditions means paying market prices and incurring gas costs at the worst possible time. Existing positions hold until natural resolution.

The kill switch triggers automatically on a monthly halt threshold breach, on API connection loss longer than 60 seconds during live trading, on three consecutive failed transactions, or on detection of any unexpected open position. It can also be triggered manually. When it fires, alerts go out immediately and a full snapshot of system state is logged for review.