Beat the Geek: Actuarial tips on AFL and NRL this season

Read­ing time: 3 mins

Tay­lor Fry is cel­e­brat­ing the 10th year of its ‘Beat the Geek’ AFL and NRL tip­ping com­pe­ti­tion which pits tip­sters skills against the expert mod­el­ling pow­er of an actu­ar­i­al and ana­lyt­ics firm. Here, Gra­ham Tay­lor lets us in on com­mon tip­ster habits, what works and of course, who will win the 2017 pre­mier­ships!

How would you describe ‘Beat the Geek’ competition?

A Our “Beat the Geek” com­pe­ti­tion was con­ceived in 2008 by a for­mer col­league, James Sul­li­van, to engage and enter­tain our clients.  We ini­tial­ly devel­oped three mod­els (two GLMs and a Boost­ed deci­sion tree mod­el) to tip the win­ners of AFL games and invit­ed our clients to com­pete against these mod­els, with prizes for the best tip­sters to encour­age take-up.  We added a NRL com­pe­ti­tion in 2010 and this year is the 10th anniver­sary of our comps.

What have you learned about how people tip, what works and why?

There is a wide spec­trum of tip­sters (we are all just part of a dis­tri­b­u­tion).  Three com­mon approach­es include:

  • Slow and steady – wins the race – these peo­ple gen­er­al­ly tip favourites and often do well.
  • All or noth­ing – these tip­sters will take a few chances each round. They often win the week­ly prize (most win­ners picked cor­rect­ly in a round) although only very rarely win our comps – there’s typ­i­cal­ly a strong inverse cor­re­la­tion between week­ly round win­ners and comp win­ners.
  • Hit and hope – employ more of a ran­dom approach to tip­ping. At least they make every­one else look good.

Q Can you let us in on how the models work?

As already men­tioned we built two Gen­er­alised lin­ear mod­els (GLMs) and a Boost­ed deci­sion tree mod­el.  Due to the lack of data (and some lazi­ness in col­lect­ing addi­tion­al data) the GLMs are fair­ly sim­ple.  The tips can be viewed as ‘odds-plus’ mod­els, since bet­ting odds are by far the most use­ful pre­dic­tor –some of our tip­sters such as Michael Price (AMP) had already worked that out them­selves. The odds have tend­ed to car­ry small bias­es over time, so the addi­tion­al fac­tors we mod­el (even a sim­ple home ground fac­tor) can improve the sig­nal slight­ly. An inter­est­ing aspect of fit­ting the GLMs is that lim­it­ed amount of data requires sig­nif­i­cant care in the set­up choice of fac­tors. This is quite dif­fer­ent from some of our client mod­els where big datasets allow every­thing to be test­ed togeth­er. The boost­ed deci­sion tree mod­el builds a ‘black box’ mod­el, which is hard­er to inter­pret although it does pro­duce a rank of vari­able sig­nif­i­cance.

Q How have the models fared over the years?

A  I would give a score­card of 7 to 8 out of 10.  The mod­els have always fin­ished in the top half of our client com­pe­ti­tion, they have won the comps twice (one year fin­ish­ing 83rd out of 108,000 on the wider  web­site). The NRL mod­els have tend­ed to fin­ish high­er on the leader­board than the AFL mod­els.  Over the years, the more man­u­al GLM mod­els have done mar­gin­al­ly bet­ter than the boost­ed deci­sion tree mod­el (hope­ful­ly my man­ag­er is read­ing this – there’s still some hope we don’t get replaced by robots).

Q Can money be made from modelling sports data?

A  Yes (David Walsh seems to have done OK).  A few years ago a deriv­a­tives trad­er and pro­fes­sion­al gam­bler approached us about sup­ply­ing him with cer­tain infor­ma­tion from the mod­els.  We had noticed that the mod­els were par­tic­u­lar good at pre­dict­ing a cer­tain aspect of a match and, on the lim­it­ed amount of data, reduced the book­ies’ typ­i­cal mar­gin from around 5% in their favour (that is, over the long run you get 95% back on a bet) to mod­er­ate­ly in our favour (slight­ly over 100%).  We did con­sid­er the busi­ness case – the size of the bets required, the rel­a­tive­ly small num­ber of AFL and NRL match­es and the non zero prob­a­bil­i­ty of ruin meant that we decid­ed against pur­su­ing it.  One relat­ed find­ing is that fixed odds bet­ting agen­cies do not always get it right.  A num­ber of years ago book­ies typ­i­cal­ly offered the same odds for a half-time/full-time draw in AFL and NRL irre­spec­tive of the head to head odds – this is clear­ly wrong, since well-matched teams are much more like­ly to tie. They have since amend­ed this, main­ly by reduc­ing odds on games expect­ed to be close.

 Hav­ing said all that almost every man and his dog has tried to beat the odds and the vast major­i­ty have failed.

Who should we back this AFL/NRL season (premiere predictions?) and why?

GWS in the AFL and Mel­bourne Storm in the NRL, although bor­ing­ly these are the same as the bet­ting mar­ket favourites.  If you are tak­ing part in a tip­ping comp this year then my rec­om­mend­ed strat­e­gy would be to tip with the book­ies’ favourites except for 4 or 5 times over the sea­son – if you get those cor­rect you will have a good chance of win­ning the comp!

Head­er Image by Scott Nolan.

CPD Actuaries Institute Members can claim two CPD points for every hour of reading articles on Actuaries Digital.

About the author

Graham Taylor

Graham has been at Taylor Fry since it formed in 1999. He uses statistical modelling to understand a diverse range of data, from New Zealand’s welfare system, to listed company financials that drive shareholder return, to predicting the outcomes of AFL and NRL matches!

Comment on the article (Be kind)


No Comments

Also this month