Hey Tim, Ranch West had an ELO system based on Prime Power. I don't know how well it worked.

What is an ELO system??

That's a new one to me.:smile:

I have been working on an ELO based system for horse racing and wondering if anyone has done something similar and what was your experience with the system. I get a decent return but not fully profitable based on betting only one horse per race. What are others seeing? — Tim Romero

What a great first post!

Welcome.

I think the biggest challenge you face is that racing is far different than (say) chess because (almost) every chess player in (almost) every game is TRULY TRYING.

There needs to be a mechanism for lightening (or removing completely) races where a given horse was unsuited for some/any reason - including being off form.

This makes it difficult to maintain.

What is an ELO system?? — ponyplayer

A relative rating system - think Chess Federation ratings.
(i.e. who you beat or lose to matters)

It is similar to multi-player ranking system that you see in video games. Basically all horses have a "rank" and the expectation is that a higher ranked horse will beat a lower ranked horse. Based on the result of a race, the rankings are adjusted such that a if a lower ranked horse beats a higher ranked horse then more points are awarded the winner and vice versa.

What is an ELO system??

That's a new one to me.:smile: — ponyplayer

What is an ELO system?? — ponyplayer

It is similar to multi-player ranking system that you see in video games. Basically all horses have a "rank" and the expectation is that a higher ranked horse will beat a lower ranked horse. Based on the result of a race, the rankings are adjusted such that a if a lower ranked horse beats a higher ranked horse then more points are awarded the winner and vice versa.

There needs to be a mechanism for lightening (or removing completely) races where a given horse was unsuited for some/any reason - including being off form. — Dave Schwartz

Yes, systematically, that implementation is very difficult, I am working on a compromise where failure to finish ITM would result in a less server loss in ranking points. Obviously horses finishing near the bottom are not really "trying" anymore so the penalty for the bottom half needs to be minimal. One of the main issues is that points earned by a horse are not really "equal". Adding class to the point system looks to be my next step in development.

In our system, we have something called "The Ask."

Just imagine that only the G and B races should really count.

Just imagine that only the G and B races should really count. — Dave Schwartz

Wouldn't that severely limit your past history data to a very small sample size?

Wouldn't that severely limit your past history data to a very small sample size? — Tim Romero

Yes.
Which would you prefer... LESS and RIGHT or MORE but WRONG?

LOL - A well-known quandary for us data guys.

Is anyone around who can share about the 90% Confidence Interval stuff?

Been away for a while, but I'm back handicapping again and some talk today at PA about Ranchwest reminded me off the good stuff here.
I made a system baserd on BRIS PP only, from an idea I got years ago from a tape from the last Handicapping Expo.

I used the top 3 PP horses. I had, through research, established a minimum odds for each rank of PP and if the 1 rank PP horse was higher than my munimum, it was the bet. If not, I go to #2, and lastly, #3.

No bets beyond #3, unless, rarely, #3 and #4 were exactly tied.

If #1 is 6-5 and my min is 9-5, no play. Keeps me off all low prices, but with so little time to "handicap" a race. I can play a lot more races. I wait to decide until the last minute to bet.

I'll dig up the old chart tonight and post the minimumsd for each rank. It is not a really large sample, all the races were actually bet by me at the time. But a database could do a large sample easily.

Tom

Here is the chart I used-

This data was collected from a spreadsheet using Dabve's What's winning at your track video, with the 7-4-1 weighting of top 3 finishers. In my study, PP #1 won at 55% because I scanned each race and did not use races that had red flags, like two or more unknowns.

The results of all the races were - 491 races,# wins, w%, avg payout odds based on avg pay

[code
]PP1 160 33% 5.26 8-5
PP2 111 23% 8.51 3-1
PP3 63 13% 11.68 9-2
PP4 60 12% 14.01 6-1
PP 5 46 9% 18.09 8-1
PP>5 51 10%

[/code]