Real but Priced: Why Most Betting Angles Earn Nothing

A year of testing the classics against the market

Here is the trap almost every improving punter falls into. You read that front-runners win disproportionately. You check it — true. You start backing early speed, and slowly, mysteriously, you lose money anyway.

Nothing was wrong with the fact. What was wrong was the test. Beating chance and beating the market are different achievements, and only the second one pays.

The residual lens

Any factor can be tested two ways. Against chance: do these horses win more often than random? Against the market: do they win more often than their odds imply? The second test is the only one connected to money, and it works by adding up the win probability the public assigned to a group of horses and comparing it with how often they actually won. Zero means the crowd had them exactly right.

The idea has a distinguished history — it is the lens the great Hong Kong betting syndicates worked through — and its sharpest illustration is the champion jockey whose mounts, backed blindly, lose: not because he is bad, but because everyone knows he is good, and the price says so with interest. The question is never "is this factor real?" It is "what part of it has the crowd missed?"

What a year of testing found

We ran our entire registry of measured effects — hundreds of them, built on roughly 2.3 million runs — through that lens. The results were humbling in a specific, instructive way.

Real and priced, worth nothing as bets: every draw and gate bias we measured, including Hong Kong's famous inside draws; elite trainers, jockeys and sires; horses running career-bests; class droppers as a blanket rule; two-year-old patterns; surface switches; hot stables. All genuinely predictive. All bet down to the take, and then some — top-quintile entity quality returns about minus twenty per cent flat, which is worse than random, because the crowd overpays for obvious class.

Real behaviour with no edge attached: trainers demonstrably target weak race days, and demonstrably peak horses for the big ones — the peaking is visible in the data at overwhelming significance. Turn either into a bet and you get nothing, or worse: the peaking signal reverses, because the market over-respects reputation on the days reputation is most visible.

The rule of thumb that fell out of it all: if an angle can be described as one factor, or two factors crossed, the crowd already has it. What survived our sweep was narrow, local and multi-conditional — the kind of thing nobody computes in their head.

Why this is good news

A closed angle has real value: it stops costing you money. Half of what we learned this year is a list of things we no longer pay for.

And the market's efficiency has a precise limit. It prices factors well and combinations imperfectly — a market can have the draw right, the trainer right and the form right, and still misprice the horse where all three intersect, because almost nobody prices twenty things jointly. That is why the surviving edges in our data are combinations, and why model-built form products exist at all. It is also why the sharpest single number we hold is the per-market efficiency benchmark: some markets (Japan, Australia) are brutally well priced, others (Scandinavia) leave room, and knowing which is which decides where an edge is even possible.

How to use it

Every time you meet a betting angle — in an article, a tipping thread, your own head — ask the second question. Not "is it true?" but "who doesn't know this?" If the answer is nobody, the price has eaten it, however real it is. Save your capital for the places where the crowd's arithmetic runs out: combinations, thin markets, and information that takes actual work to compute.

That philosophy is baked into how we present form: the ratings and the model do the twenty-factor arithmetic, and the value framework is described here.