Best odds measure for gauging win probability

[superhans]

Quick question... In a horse race market with high volume it's generally accepted that the probability of each horse winning is well approximated by 1 / (its decimal odds). But which odds should you use for this?

1. The SP.

2. The real-time best back or lay price - or an average of the two.

3. The weighted average across the whole market, calculated as (SUM(odds * money matched)) / (total money matched).

I've been using #1 but SP it seems can sometimes lie way outside the bid-ask zone between the best back and lay prices, making me wonder if it is really a good measure.

One of Gary's sample sheets using the COM interface has a neat algorithm for calculating #3, but I've found that can also lie outside the bid-ask spread. So I'm wondering if #2, just a straight average of the top back and lay prices might be the best solution after all.

Any thoughts much appreciated!

[jokerjoe]

I'd use the mid-point of the back-lay prices, normalised so that all probabilities sum to 100%.

[superhans]

Thanks - yeah, that's what I think I'm leaning towards myself now. I find it odd though that the weighted mean approach doesn't necessarily give something in that Back-Lay range - I was expecting that to somehow be the optimal measure of the market consensus...

[negapo]

It depends on the purpose.
The Back/Lay spread middle point and the SP can be seen as present expectations and the weighted average as pass expectations. If you want to take the present expectation i would go for the last price matched (if recent). That is a real market price as opposed to the middle point between the spread (in liquid markets this is not so much of a problem as spreads are very tight).
Normalizing the midd-point odds has some assumptions that can be OK for some purposes. It depends on your specific case.
Hope it helps
Regards

[superhans]

Thanks, negapo - that helps!