Saturday, 20 November 2010

2010 season review: prediction

This review is about how good practice sessions are for predicting qualifying, and how good practice sessions and qualifying are for forecasting race winners. After this, I’ll write a final review summarising betting success or lack thereof during 2010.

I allocated 1 point for every correct prediction. So, if P2 had Vettel the fastest and he subsequently got pole that’s 1 point. If he then won the race then (in the numbers for the winner) that’s also 1 point, and a further 1 point from qualifying.

In addition, for pole position, I allocated 0.1 points for a prediction of a chap that came within a tenth of pole. So, entirely at random, if Alonso came within 0.002s of Vettel for pole, and P2 had Alonso as the fastest, that’s 0.1. I included this because I think a chap that close to pole is both capable of getting it and close enough to be very layable.

So, which practice session was best for forecasting pole? Here are the stats:
P1 = 4 = 21%
P2 = 2.4 = 13%
P3 = 7.2 = 38%

This is good stuff, because I’ve said throughout that P3 (with the low fuel qualifying simulation run) is best for predicting the pole-sitter, and that P2 is probably the least useful as it includes heavy fuel running.

What about the race winner, though? Again, I looked at the practice sessions, and qualifying, and here’s what emerged:
P1 = 5 = 26%
P2 = 4 = 21%
P3 = 5 = 26%
Q = 8 = 42%

Unsurprisingly, pole is the best predictor, but still less than 50% correct. Interestingly, all three practice sessions are about the same in terms of predictive power, getting the answer right about a quarter of the time.

We need to consider the Vettel Effect when looking at these numbers. He got pole more than half the time this year, and failed to get a win from pole pretty often. So, his lack of reliability (mechanical as well as personal) will have a dramatic impact on the results.

So, what happens if we remove Vettel from the equation by cutting out races where he got pole? He got 10 poles, so stripping those races reduces the number to 9 (just under half). Here are the qualifying stats:
P1 = 0 = 0%
P2 = 0.1 = 1%
P3 = 3.1 = 34%

Race stats:
P1 = 2 = 22%
P2 = 1 = 11%
P3 = 1 = 11%
Q = 5 = 55%

With the exception of qualifying predicting the race result, all other sessions see reduced predictive power. However, the grid becomes a better than evens forecaster, getting it right 5 times out of 9.

Will this hold true next year? KERS, especially if some engines have superior systems, may give an almost automatic number of passes off the grid for some cars. If the Pirellis do (intentionally) degrade more than this year’s tyres then this will increase the impact of strategic decisions (deciding when to pit) and tyre management. Neither of these will affect qualifying, so I fully expect P3 to be best at predicting that again next year, but both would affect race results, making things a bit more mixed up at the sharp end.

However, we also need to look at the predictive results for each individual race. There are some oddities (no-one saw Hulkenberg’s stellar pole in Brazil coming, for example), but a few patterns too. Here are the individual race predictions for pole and the race win (NB numbers are totals of all practice sessions/all practice sessions and qualifying):
Bahrain = 0/1
Australia = 0.1/0
Malaysia = 2/0
China = 0/1
Spain = 0/1
Monaco = 0/1
Turkey = 0/1
Canada = 1/2
Europe = 1/2
UK = 2/1
Germany = 1.1/1
Hungary = 2/1
Belgium = 1/0
Italy = 0/1
Singapore = 0.2/1
Japan = 2/3
Korea = 0.1/0
Brazil = 0/2
Abu Dhabi = 2.1/3

There are clearly quite a few races with little foresight offered by practice/qualifying, but a few do stand out as being predictable. Abu Dhabi and Japan especially, with high qualifying predictability in the UK and Hungary. Whether this is repeated next year remains to be seen, but is something I’ll try to remember to keep my eye on.

Although P3 is the best qualifying predictor and qualifying is the best race predictor neither got it right even half the time, and this ought to be considered for next season.

The next, and last, review will be about how the tips I offered (particularly in the latter half of the season) turned out.

Morris Dancer


Omnium said...


Some interesting thoughts. I'd be slightly concerned that your data samples here are too small to glean very much meaning anyway. Not trying to put you off, but I'd be careful about trying to read too much into it. With that cautionary note it's perhaps interesting that your numbers suggest P2 is the worst pole predictor given that it takes place at a similar time to qualification.

One thing that perhaps stood out for me this year was that the lesser teams pretty much had all their good results towards the beginning of the season. There's good reason for that obviously, but maybe looking at the pre-halfway and post-halfway point balances for the teams might be instructive. I'll perhaps try to look at this if I have some time.

I hope that the next posts contain details of great victories! My season was very good on all my long term bets - even though the Rosbergs of this world didn't deliver me the monster results I'd hoped for. However whether by bad luck or angering the gods I pretty much lost on all of the race day bets I made. Vettel in particular seemed to always trip me up, whether I was with him or against him. Mind you Webber did me the odd good turn.

Some early 2011 prices around now too. Good hunting!

jamei said...

This is interesting stuff, and often overlooked by other betfair users ;-) I'm reminded of the guys who made megabucks by taking the time to work out that the odds of hole-in-one in a major golf tournament were around 1 in 3, when everyone else (including the bookies) assumed it was a lot less likely than that. Not such big discrepancies to be found here of course, but any solid data is useful.

I think there are enough data samples and your conclusions for practice and qualifying are sound, though I'd be wary of making any decisions on the basis of last season's results at individual circuits.

With KERS and adjustable rear wings, the race result can surely only get less predictable, but this won't affect qualifying which is by it's nature more predictable anyway.

I'm going to do a bit of number-crunching myself on the three qualifying sessions - I assumed the fastest drivers wouldn't try very hard especially in the first session when they're in little danger of going out. However I noticed during the season they do seem to push, so I'm wondering if poll or the top 10 can be predicted by the order in the first two sessions.

As to the 2011 championship, I'll initially be trying to get good odds on Rosberg and Schumacher - their car surely can't be any worse than last year's, especially given they've been developing it longer than anyone else.

Morris Dancer said...

On Schumacher: I was tempted by the 20/1 Ladbrokes had for him to take the title, but this fell to 14/1 after he got 2nd in the 2nd Pirelli test [ironically being relegated to 7th after he was judged to have cut the chicane].

I do agree that it's better to take a general view of things (ie P3's the best of the practice sessions but isn't perfect, rather than using the percentages I included as if they were precise and bound to be the same next year).

I find it a bit hard to assess who is better, today, of the Mercedes drivers. I ignore the first half of the season because Schumacher clearly wasn't up to speed. In the latter half, he had some better results, but it's worth considering Rosberg had back luck (being taken out by Webber in Korea, for example) so that may well flatter Schumacher. Anyway, it's only 14-15 weeks until Bahrain, and we'll find out then.

Omnium said...

The tyre tests have been interesting in that Schumacher seems to have preferred the new Pirellis - perhaps Massa too. Maybe their somewhat poor results have been sue to the tyres really not suiting their styles. Another factor to throw into the mix!