Expected Value (EV)

The average modelled return per unit allocated to a selection, given the estimated probability and the market price. Positive EV means the selection is profitable on average in the model.

Expected value (EV) is the average return per unit allocated, computed from the estimated win probability p and the decimal market price O:

EV = p × O − 1

EV is in units of allocation. EV = +0.30 means a 1-unit selection returns 1.30 on average over many repetitions (0.30 profit). EV = −0.05 means it loses 0.05 per unit on average.

Why EV alone isn't enough:

  • EV doesn't tell you about variance. A +5% EV selection at price 50.0 is a roulette wheel; the same +5% at price 1.5 is steady. Use Kelly to size the allocation to the price profile.
  • EV is only as good as your p. Overestimate p by 5pp and a "+30% EV" selection might really be +10% or negative.

The TrapStats EV pipeline:

  1. Model produces raw prob_win.
  2. Calibrator maps it to a calibrated probability.
  3. Shrinkage blends with market: pw_eff = (1−λ)·prob_win + λ·(1/price), λ=0.30.
  4. ev = pw_eff × price − 1.
  5. Gates: skip if ev < 0.25 (too marginal) or ev > 0.50 (implausibly high → model-overconfidence flag).
  6. Allocate from EV-tier table (1%/2%/3% of the test portfolio).

The "Why this pick" panel on each pick shows all five steps explicitly.

← All terms