Just wondering whether there has ever been an analysis/piece of work done on the quality of Tennis Abstracts predictions vs actual outcomes. . . .??
They did their own a couple of years ago and it was well above any prediction that pure rankings would make in terms of accuracy. But I cant remember what the level of accuracy was. I think it is largely in line with bookies odds, from memory, which is probably what you would expect?
Just wondering whether there has ever been an analysis/piece of work done on the quality of Tennis Abstracts predictions vs actual outcomes. . . .??
They did their own a couple of years ago and it was well above any prediction that pure rankings would make in terms of accuracy. But I cant remember what the level of accuracy was. I think it is largely in line with bookies odds, from memory, which is probably what you would expect?
I think I was being a little bit tongue in cheek! Mainly referring to ones we tend to see on here!!
Just wondering whether there has ever been an analysis/piece of work done on the quality of Tennis Abstracts predictions vs actual outcomes. . . .??
They did their own a couple of years ago and it was well above any prediction that pure rankings would make in terms of accuracy. But I cant remember what the level of accuracy was. I think it is largely in line with bookies odds, from memory, which is probably what you would expect?
I'd actually have expected the bookies' odds to be better predictors, given they can take everything into account though offsetting that a bit is the weight of money effecting them.
I personally don't really see the Tennis Abstract percentages as strictly predictors but percentages based on the algorithm and its inherent limitations in not allowing for everything. But fair enough if others call them as such.
Anyway, well above pure rankings and largely in line with bookies' odds I would see as pretty impressive as a 'predictor'.
I'd like to know about any system that beats bookies' odds
Just wondering whether there has ever been an analysis/piece of work done on the quality of Tennis Abstracts predictions vs actual outcomes. . . .??
They did their own a couple of years ago and it was well above any prediction that pure rankings would make in terms of accuracy. But I cant remember what the level of accuracy was. I think it is largely in line with bookies odds, from memory, which is probably what you would expect?
I'd actually have expected the bookies' odds to be better predictors, given they can take everything into account though offsetting that a bit is the weight of money effecting them.
I personally don't really see the Tennis Abstract percentages as strictly predictors but percentages based on the algorithm and its inherent limitations in not allowing for everything. But fair enough if others call them as such.
Anyway, well above pure rankings and largely in line with bookies' odds I would see as pretty impressive as a 'predictor'.
I'd like to know about any system that beats bookies' odds
I'm not sure if predictions and forecasts are the same thing, but they call them forecasts. And they are based purely on ELO and surface weighted ELO at that.
they run several thousand simulations in the background of a match up and the outcome of those simulations is the forecast percentage. Not sure what those simulations do or how they work, are they called Monte Carlo simulations?
Definitely a massive surprise to see Toby and Jake lose. I admit I got a little carried away myself and started looking at where an all-GB final would take them in the rankings. This has definitely been a bit of a week to forget - onto week 40!
That's more what I thought from a 2025 post that the match forecasts are seemingly based on Jeff's ELO ratings, as you first indicated with its at times limitations for using for forecasts that he refers to.
Your previous post and link before this one refers to something he calls jranks with then 'Monte Carlo simulations'. The post seems to eminate from 2012 ( ah, I belatedly see that it's actually got 2012 in the link title ) and when I look for 'jranks' I see rankings from 2013. So I suspect these rankings and that post and explanation are very out of date and have been superceded
This link above would appear to give the current situation. Interesting all his analysis as to such as where he feels using ELO ratings is more and less relliable.
-- Edited by indiana on Friday 2nd of October 2026 02:48:02 PM
That's more what I thought from a 2025 post that the match forecasts are seemingly based on Jeff's ELO ratings, as you first indicated with its at times limitations for using for forecasts that he refers to.
Your previous post and link before this one refers to something he calls jranks with then 'Monte Carlo simulations'. The post seems to eminate from 2012 ( ah, I belatedly see that it's actually got 2012 in the link title ) and when I look for 'jranks' I see rankings from 2013. So I suspect these rankings and that post and explanation are very out of date and have been superceded
This link above would appear to give the current situation. Interesting all his analysis as to such as where he feels using ELO ratings is more and less relliable.
-- Edited by indiana on Friday 2nd of October 2026 02:44:41 PM
i think you're correct, ELO got developed by him after JRanks, so that is probably out of date and the link above is more current. I am not sure if he still runs Monte Carlo Simulations for forecasting though, I thought he did, but the latest article doesn't mention that...