BaselinePolicies
Helpers & Utilities · AiDotNet.Finance.Trading.Evaluation
No-skill baseline policies to rank a learned agent against. If a trained portfolio manager cannot beat these on the untouched holdout, it has learned nothing worth trading (see the honest-eval discipline).
Methods
| Method | Summary |
|---|---|
Func<Vector<T>, Vector<T>> EqualWeight(int tradableCount) |
Equal-weight long book: allocate 1/N to each of the tradableCount assets every step (the environment holds the mix, so this is effectively rebalanced buy-and-hold — the classic no-skill control). |
Func<Vector<T>, Vector<T>> Flat(int tradableCount) |
Flat (all-cash) policy: never take a position. |
Func<Vector<T>, Vector<T>> Momentum(int windowSize, int totalColumns, int tradableCount) |
Cross-sectional MOMENTUM baseline: go equal-weight long the tradable assets whose price rose over the observation window (recent winners), flat the rest. |
Func<Vector<T>, Vector<T>> RiskParity(int windowSize, int totalColumns, int tradableCount) |
RISK-PARITY (inverse-volatility) baseline: allocate long weights proportional to 1/σ of each tradable asset’s recent returns, so every name contributes roughly equal risk (calmer assets get more capital). |