firlandsfarm wrote: ↑Fri Dec 31, 2021 3:52 am
napshnap wrote: ↑Thu Dec 30, 2021 6:28 pm
Alright, I maybe misread it, it sounded like you can't calc (convert) odds from ratings, weights or other metrics at all.
I agree that you can't calc "true odds" (cause "true odds" is a controversial entity).
No I don't think you misread that bit. there are many stabs at converting ratings to odds I just don't see them able to stand up to mathematical theory. By calculating I mean put a set of ratings into a spreadsheet and have it calculate the respective odds with no interaction with the user.
I thought I had cracked it with what I will refer to as a calculated Monte Carlo algorithm based on creating random numbers related to each runner's rating and calculating the probability each runner's 'random number' was the highest. Obviously the runner with the highest rating would be most likely that it's random number would be the greatest (the winner) thus introducing big is best. But at the same time the random number of the runner with the smallest rating can sometimes be the largest albeit rarely but acknowledging that every donkey has it's day! The formula used calculates the outcome for an infinite number of Monte Carlo iterations (I found performing actual Monte Carlo simulations too time consuming when using a large number of iterations for greater accuracy) ...
LargestRandomNumberFormula.png
(yeah, I spent more time researching how to calculate the formula than I did finding it! I had three specimen calculations I could test my interpretation against so I was satisfied I had converted it into Excel correctly.)
I then compared the algorithm calculated probabilities/odds with market history race by race (it took my laptop 3.5 days to run the algorithm across my racing database) and found a clear bias towards the higher rated runners i.e. the calculated odds were consistently below market odds so I concluded the results were not accurate. It could of course be the ratings that are at error and not the algorithm. I'm not using my own ratings so can only back test against ratings in the market (Proform, Adrian Massey etc.) but the problem there is accessing a good history to test against.
I'm thinking of introducing what I am calling a 'tilt' to sets of ratings. By that I mean reduce the higher ratings and increase the lower rating just to see if it brings the results more in line with the market prices. I'm not trying to back fit the algorithm but when 90% of results on favourites trend in one direction it's clear it's not correct!
I'm testing on favourites because of the neutrality of price over the long term.
I appreciate I have probably gone off-topic with this comment but I would like to see alternative ways people create odds from ratings.
Good post.
I always look to simplify things where possible.
My view on producing a rating for each runner is based on my own experience(s) reading books/forums, testing my own data...etc This doesn't mean I've solved it, because horse racing is an imperfect sport where there is always an incomplete picture.
My approach is to take a set of attributes that I think affect/influence an outcome of a race, weight them and produce a rating. In some cases, combine certain attributes that strengthen the overall rating.
The one thing I found was that the top-rated was often the favourite and well backed. So this told me that the market makers seemed to be taking into account similar information into their algo. I also quickly learned, that sometimes, the bottom rated would win, but the price was a value one comparing the market price with my own tissue price. I also learned that if a favourite was poorly rated, they ran poorly too.
Taking all that into account, there are many ways I can utilise one set of ratings. This includes In-Play if the race is going the way I visualised it.
I also tend to avoid tweaking things as horses aren't machines and will have off days, be unlucky, or the jockey might make a mistake.
My early attempts were shite, but I learned from them. I also believe there are websites/resources that present the information better and would save a lot of time, but I enjoy the process.