Engine-computed reference · 6×5 grid · 30 cells
Edge Variance Kelly Reference Grid
The half-Kelly bet fraction after Bayesian shrinkage for every combination of estimated edge and edge-estimate uncertainty across this grid. Each cell is a live run of the Position Sizing under Edge Variance engine; no value on this page was entered by hand.
When the edge is itself an estimate, the naive Kelly fraction overbets: the more uncertain the estimate, the more the optimal size shrinks toward zero. This grid shows the variance-adjusted Bayesian fractional Kelly across estimated edge (rows) and the standard deviation of that estimate (columns), all under half-Kelly scaling. The fraction runs up to 1.199 at edge 0.12 with the tightest uncertainty (0.005); every one of the 30 cells sizes below its own naive deterministic fraction (30/30), because uncertainty always penalizes size. The CVaR column quantifies the 5% tail loss of the chosen fraction. For the underlying idea, see edge-variance sizing for LLM signals. Education only — not investment advice.
Bayesian fractional Kelly by edge mean × estimate uncertainty
Rows = estimated edge mean (expected per-bet return). Columns = edge
estimation uncertainty (standard deviation of the edge estimate). Each
value is the engine's
fractionalBayesian output.
| edge \ stddev | 0.005 | 0.010 | 0.020 | 0.040 | 0.080 |
|---|---|---|---|---|---|
| 0.01 | 0.100 | 0.100 | 0.099 | 0.097 | 0.089 |
| 0.02 | 0.200 | 0.200 | 0.198 | 0.194 | 0.177 |
| 0.03 | 0.300 | 0.299 | 0.298 | 0.291 | 0.266 |
| 0.05 | 0.500 | 0.499 | 0.496 | 0.484 | 0.443 |
| 0.08 | 0.800 | 0.798 | 0.794 | 0.775 | 0.709 |
| 0.12 | 1.199 | 1.198 | 1.190 | 1.163 | 1.064 |
Headline metric: Bayesian fractional Kelly. The CSV download below also carries the deterministic and conservative fractions and the 5% CVaR per cell.
Provenance
- Engine
- Position Sizing under Edge Variance
(
position-sizing-edge-variance) — computed live from/engines/position-sizing-edge-variance.js - Grid
- edge mean ∈ {0.01, 0.02, 0.03, 0.05, 0.08, 0.12} × stddev ∈ {0.005, 0.010, 0.020, 0.040, 0.080} = 30 cells
- Fixed inputs
- outcome_variance=0.05, kelly_fraction=0.5
- Computed
- 2026-05-23, recomputed in CI on every build
The engine is deterministic: the Bayesian shrinkage and CVaR are closed-form, so the same input always returns the same output. The full method — the prior, the posterior shrinkage of the edge, and the tail-risk calculation — is documented at the Position Sizing under Edge Variance methodology page. For how conviction maps onto a scaled Kelly fraction, see conviction-scaled Kelly.
Reference grids are planning aids, not financial, tax, or investment advice. No cell is a recommendation to bet at that fraction.