Jivlain
Hammelmann Appreciation Society
AFL Fantasy Club Board Champion
AFL Fantasy Club Board Champion
AFL Fantasy Club Board Champion
Brisbane Lions - Charlie Cameron Player Sponsor 2020
Brisbane Lions - Charlie Cameron Player Sponsor 2019
Brisbane Lions - Hugh McCluggage Player Sponsor 2018
Brisbane Lions Board Tipping Champion
Brisbane Lions - Mitch Robinson Player Sponsor 2017
- Sep 19, 2015
- 8,339
- 14,746
- AFL Club
- Brisbane Lions
- Other Teams
- Swansea City
- Thread starter
- Moderator
- #26
This is an awesome discussion gents; well done. I love Fantasy and I'm learning data science via R, so this is amazing on both fronts. Jivlain, have you compared the correlation of expected prices to actual from the Footywire website? They have some price predictions (predicted changes at n+1, n+2, n+3); I'm looking at some of their data on a weekly basis, but I haven't bothered to check whether their predictions are accurate.
Loving your work! Thanks!
I have all the actual prices already - I have a scraper for that - and calculate the factors with a system of simultaneous equations using last week’s price and the last k rounds as coefficients on one side and the current actual prices on the other. The resulting correlation of estimated price to actual price is really close - R squared of 0.999 or so.
There is some mysterious factor of last year’s performance in the prices according to the rules - and there clearly are residuals that should be explainable - but so far neither my eyeballs nor the simultaneous equations have found any correlation between the residuals and last year’s performance.
Given the already tight correlation it must be pretty small however.
Either way, seeing as it is apparently tied to last year’s performance - I figure it doesn’t change much. Plus I’ve noticed that there is a correlation between last week’s residuals and this one’s. So my “predicted prices” are actually not the output of my formula.
Instead, I estimate the current price again based on my formula. I then estimate the next week’s price using my formula and predicted score. And then I add the difference between them to the current price and that is my predicted price - I may not know the mysterious factor but that should control for it pretty neatly.
So I reckon the next big step forward would be to make my score estimation better - obviously it’ll remain limited and could never have guessed what Telly was about to do