Price question.

[Padawanbewbie]

Is there a calculation for estimating the implied probability of a price going from we'll say 12s to 2s in the latter stages of a race?
I understand that if a horse is trading at 12 then the implied probability is 100/12 = 8.33%
But how do I calculate it getting to 2s or any price bar 1.01.
Thanks!

[sionascaig]

You would need the probability distribution of the prices, a mean & a variance..

[ShaunWhite]

If you assume markets are efficient (which they are) then there's the same probability of a move as the potential gain from it. Eg you'll double your money 50% of the time and you'll triple it 1/3 of the time. There's no 2 prices, random or otherwise where that isn't the case. That's also evident via logic aside from the maths. If you could just pick two prices and that move was more likely than the risk associated with it then this game would be a piece of cake.

These things are true when you look at the data at scale, so to profit you have to identify the specific situations where the entry, the exit, or the move has a positive expectancy.

[Padawanbewbie]

Thanks very much for the informed replies.
Much appreciated.

[ShaunWhite]

To save you some time, if P1 is efficient and P2 is efficient then, p1-p2 must also be efficient.

So, if you're backing first then the likelihood of a price shortening is 100/(p1/p2)

eg
Dobbing : enter at 3, exit at 1.5 .... 100 / (3/1.5) = 50%
Take a tick : enter at 3, exit at 2.98 = 99.3%
Let it run : enter at 3, exit at 1 (ie don't exit) = 33.3%

Making money on the 3 -> 1.5 means you need a strike rate better than 50%, and you need 1 tick too often to be viable. 3.0 winners obvioulsly need to win > 33.3% of the time but it's nice to see why from a different angle. If p2 is 1 then it simplifies to 100/p.

With the right data you can query all the possible values of p1 and p2 for every selection that's run for years to try and find anomalies, ie frequency > 100/(p1/p2). They exist, but any whole set of data always has a degree of distribution which is why backfitting is a hazard, and when tested more rigourously with in & out of sample data etc they're not reliably predictable by the pure numbers themselves. That's where an individuals edge comes in, either visually such as race reading or by combining other metrics such as TPD. Picking two numbers for a take-profit or stop-loss across the card sadly isn't going to work.

[ShaunWhite]

Here you go

Screenshot_26.jpg

[Korattt]

very interesting thread, thanks OP for this