Generalise: favourites at 5/1+ in big-field non-handicaps beat their price

Reasoning

Why this hypothesis exists: what the lab saw, the question it asked, the mechanism it proposed, what it predicted, and the result that would prove it wrong. Written before the test was run.

1. Observed
   • H199-favourite-at-a-big-price-in-a-big-handicap (H199: favourite at a big price in a big handicap) was GREEN in the earlier research: A/E 1.14 on 1086 bets (periods 1.16 / 1.08 / 1.23); now PROMISING. O-a934fc73
2. AskedDoes the market-consensus effect also hold for favourites at 5/1+ in big-field non-handicaps (maidens, novice races)? answered
3. MechanismWhen a large, open field has an unusually long-priced favourite, that horse still attracted the most money. That money is informed (yard and market confidence), and it is worth more than the price, which is set by comparison with typical 5/1 shots.
4. Why the market misses itA 5/1+ favourite looks unconvincing, and the price is anchored to how ordinary 5/1 shots perform, most of which are third to fifth choice.
5. Predictedpositive, A/E 1.05 to 1.12 on 400+ bets
6. Falsified ifA/E 1.00 or below on 300+ bets
7. Rule sketchis_handicap = 0 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 12
8. question id

Mechanism and rationale

When a large, open field has an unusually long-priced favourite, that horse still attracted the most money. That money is informed (yard and market confidence), and it is worth more than the price, which is set by comparison with typical 5/1 shots.

The rule

is_handicap = 0 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 12

Reads: is_handicap, is_favourite, sp_decimal, field_size

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

019b94d0892aa869be9a2f11c91f30dd8847185614a33cf4a7ae52c892a7b292

Frozen

not frozen

The definition can still change.

Confounders

• Debutants in maidens: a 5/1+ favourite can be a newcomer whose price reflects stable gossip, not form. This effect differs from open-field consensus and is tested by the is_maiden = 0 refinement.
• Irish maiden hurdles and bumpers with big yards (Mullins/Elliott), where several stable runners split the money and the favourite tag is close to arbitrary.
• Few qualifying races: big-field non-handicaps with a favourite at 5/1 or bigger are rare, so a handful of big-priced winners can carry the result.
• Joint or co-favourites: when the market is split this evenly, being 'favourite' carries little information.
• SP vs BSP: H199's sister edge existed only at BSP (+16.6% BSP, -9.6% SP), so A/E against SP-implied odds may not translate into profit at SP.
• Overlap with killed H234 (non-handicap favourites at 5/1+ (sp_decimal >= 5) in 12+ fields: A/E 0.99 on 208). This rule is a subset of H234's, so most of these bets have already been measured once.

Control

{
  "description": "The lab compares against the other runners in the same races. If the 5/1+ favourite is underpriced, the rest of the field should come in at or slightly below par (A/E about 0.95 to 1.0), so the overround is paid by the others and not by the favourite. The control rule tests the anchoring claim directly. It measures 'ordinary' 5/1 to 8/1 shots ranked 3rd to 5th in the same kind of race (big-field non-handicaps). If the favourite's edge is real, these runners should sit at or below par. If they also beat their price, the effect is a property of the price band or race type, not of the market's consensus on the favourite.",
  "rule": "is_handicap = 0 AND is_favourite = 0 AND sp_rank >= 3 AND sp_rank <= 5 AND sp_decimal >= 6 AND sp_decimal <= 9 AND field_size >= 12"
}

Discovery result

fail broad rule · 16 Oct, 15:46

Bets

6

0 winners

A/E

0.00

95%: 0.00 to 4.66

SP return

-100.0%

level stakes

Betfair SP return

–

6 bets with a BSP

Strike rate 0.0%. p = 1.0. Same-race control: the other runners in these races had A/E 1.69, so the selection did not beat its own races. Zone: discovery. Evaluator eval-1.0, data 6a4e98be18e83933

Falsification

No attacks yet. Only promising ideas are attacked.

Sealed holdout (vault)

Not queried. Only a frozen idea can spend its one query against the 2020-25 holdout.

Forward record

No forward record. Shadow tracking starts after the vault and counts only races run after the freeze.

Lineage

Agent runs

• designer v1ok · 33s · 16 Oct, 15:45

  Output (schema-checked)

  {
    "rule": "is_handicap = 0 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 12",
    "refinements": [
      {
        "label": "exclude maidens (debutant confounder)",
        "rule": "is_handicap = 0 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 12 AND is_maiden = 0"
      },
      {
        "label": "huge fields only (14+, matching H199's field threshold)",
        "rule": "is_handicap = 0 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 14"
      }
    ],
    "control": {
      "description": "The lab compares against the other runners in the same races. If the 5/1+ favourite is underpriced, the rest of the field should come in at or slightly below par (A/E about 0.95 to 1.0), so the overround is paid by the others and not by the favourite. The control rule tests the anchoring claim directly. It measures 'ordinary' 5/1 to 8/1 shots ranked 3rd to 5th in the same kind of race (big-field non-handicaps). If the favourite's edge is real, these runners should sit at or below par. If they also beat their price, the effect is a property of the price band or race type, not of the market's consensus on the favourite.",
      "rule": "is_handicap = 0 AND is_favourite = 0 AND sp_rank >= 3 AND sp_rank <= 5 AND sp_decimal >= 6 AND sp_decimal <= 9 AND field_size >= 12"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Debutants in maidens: a 5/1+ favourite can be a newcomer whose price reflects stable gossip, not form. This effect differs from open-field consensus and is tested by the is_maiden = 0 refinement.",
      "Irish maiden hurdles and bumpers with big yards (Mullins/Elliott), where several stable runners split the money and the favourite tag is close to arbitrary.",
      "Few qualifying races: big-field non-handicaps with a favourite at 5/1 or bigger are rare, so a handful of big-priced winners can carry the result.",
      "Joint or co-favourites: when the market is split this evenly, being 'favourite' carries little information.",
      "SP vs BSP: H199's sister edge existed only at BSP (+16.6% BSP, -9.6% SP), so A/E against SP-implied odds may not translate into profit at SP.",
      "Overlap with killed H234 (non-handicap favourites at 5/1+ (sp_decimal >= 5) in 12+ fields: A/E 0.99 on 208). This rule is a subset of H234's, so most of these bets have already been measured once."
    ],
    "mechanism": "In a large, open field, a horse that is still favourite at 5/1 or bigger drew the most money from bettors who had many alternatives. That consensus reflects informed confidence (stable, connections, market) and is worth more than its price implies. The price is set as though the horse were an ordinary 5/1 shot, most of which are third to fifth in the betting.",
    "market_miss": "A favourite at 5/1 or bigger looks weak, so bettors and layers anchor its price to how typical 5/1 shots perform, and most of those are third to fifth choice. They do not credit the extra information in being top of the market in a big, open field. H199/H235 suggest this happens in handicaps. This test asks whether it also holds outside them.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.05 to 1.12 on 150-400 bets (likely fewer than the hypothesis's hoped-for 400, given H234's 208 bets at 5/1+)",
      "falsified_if": "A/E 1.00 or below on 150+ bets, or the control (3rd-5th choices at 5/1 to 8/1 in the same race type) beating its price by as much as the favourites do"
    },
    "notes": "Warning: this rule is almost a strict subset of H234 (sp_decimal >= 5 instead of >= 6), which was KILLED at A/E 0.99 on 208. The prior is weak, and the qualifying sample will probably be under 208 and possibly below the 150-bet floor. The spec keeps the investigator's rule unchanged so that it is pre-registered honestly. A pass here would mean the effect is confined to the 6.0+ band, and that pattern looks like a price-band slice, so falsification should treat it sceptically. If the refinements show anything, check whether a few long-priced winners carried the result. Judge profitability at BSP, in line with the H199 finding."
  }

Timeline

1. Job 1906 finished

   16 Oct, 15:46 · engine:compute-1

2. H-0bc150c2 Discovery → KILLED fails: 6 bets, needs 150; A/E 0.0, needs 1.05; p 1.0, needs 0.05 or lower; periods [], each needs 1.0; A/E 0.0 against same-race control 1.687, needs 0.03 above (not a niche pass: only 6 bets; A/E 0.0 below 1.05; p 1.0 above 0.5; does not beat the same-race control 1.687)

   16 Oct, 15:46 · engine:evaluator

3. discovery result for H-0bc150c2 (broad rule): fail

   16 Oct, 15:46 · engine:evaluator

4. H-0bc150c2 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 15:46 · engine:orchestrator

5. compute-1 took job 1906 (evaluate, attempt 1)

   16 Oct, 15:46 · compute-1

6. Job 1904 finished

   16 Oct, 15:46 · engine:ai-1

7. Queued evaluate on the compute lane: discovery run of the broad rule

   16 Oct, 15:46 · engine:scheduler

8. H-0bc150c2 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 15:46 · agent:designer@v1

9. ai-1 took job 1904 (design, attempt 1)

   16 Oct, 15:45 · ai-1

10. Queued design on the ai lane: write the pre-spec

    16 Oct, 15:45 · engine:scheduler

11. New idea H-0bc150c2: Generalise: favourites at 5/1+ in big-field non-handicaps beat their price (generalise)

    16 Oct, 15:45 · agent:investigator@v1

Every change is an event with an actor: an agent and its version, an engine, a human, or the importer. Current stage: Killed.