Sunday, 5 May 2019

Let's bet against the best player in the world

It seems as if over the past few months, I've been betting against Michael van Gerwen with increasing frequency - if I look back through my entire spreadsheet of bets (which, at least if you look at the sidebar on the left in desktop view, I've only actually backed van Gerwen nine times - three of them were in the worlds (vs Lewis, Hopp and Anderson), another two were against Anderson (in the Grand Slam and Players Championship semi finals), then you need to go back to the 2017 Matchplay (against Bunting) before I bet him again.

Meanwhile, I've bet against him 29 times - nine times in the Euro Tour alone (all of them lost except the Keegan Brown win), then before that in the Grand Slam (yes, I was on Clayton in the group stage), another four times in the Euro Tour (getting a win through Ian White in the one that Hopp won), then it gets to pre-Dart Connect ages where I had very little data to work with, or no data and was basically winging it, so probably not overly relevant going forward.

Something jumps out straight away - why are we betting van Gerwen in mostly longer games, but laying him in shorter games? How much of a difference does game length make? Fortunately my data model can demonstrate this - first, let's take the last twelve months of data and see how things vary with van Gerwen against Cross:


This is what gets spit out of the master computer - you'll see that if they were to play in the Euro Tour right now, the model thinks van Gerwen would win two out of three, but if they were to play in the final of the Matchplay now, van Gerwen would win over 10% more often. Now let's try against someone a bit further down the pecking order, to see what the shift is like there:


I've picked a random in form player for this year, and included the date slider that I have to show that I'm using just this year - it swings a bit more, but it's not by that much more - 14% as opposed to 11%. What about if we pick someone who isn't in form:


It's the same kind of story really. Where the difference in match format doesn't alter things as much is where the players are extremely close to begin with - this year, MvG is at 51.37% in a best of 11 against Gerwyn Price, but 52.49% in a Matchplay final. Similarly, if the opponent has next to no chance to even win a best of 11, there's not much winning chances to be reduced in the first place (Michael Barnard's figures were, shall we say, interesting). Could it be the case that bookmakers aren't correctly adjusting for the increased variance that there is in a short course format? Maybe, but given that outside of the worlds there aren't actually any games that get to a very long format for more than the quarter finals onwards (Grand Prix as well I guess, but that's double in and an entirely different beast), maybe it's the other way around and they don't shorten the van Gerwen price enough in a long event. I did look back at some of the worlds posts, and did see that van Gerwen was 1/25 in the game against Joyce, which is probably one of the few games that we'd see van Gerwen play in a long format against someone who isn't also pretty decent, what'd he be priced up as if he played Joyce in the next Euro Tour they're in?

Let's look at that Monk figure. If we were to say that is actually 25% to make it a round number, that would equate to a fair price of 3/1. But when was the last time that anyone saw a van Gerwen opponent priced up that short on the Euro Tour? This year I've bet on the following in the Euro Tour:

Mervyn King (13/2)
Jermaine Wattimena (6/1)
Keegan Brown (17/2)
Krzysztof Ratajski (17/2)
Dimitri van den Bergh (5/1)
Raymond van Barneveld (11/2)
Nathan Aspinall (7/2)
Steve Lennon (7/1)
Stephen Bunting (6/1)

Only Aspinall out of those is even close to a 3/1 price, and I think you'd consider all of these players to be at the very least on par with Monk right now.

Let's take a bunch of top 50's - the top 50 in the Order of Merit, the (active) top 50 in the FRH rankings, the top 50 in PPT this season, and the top 50 on the FDI (who's also provided win chances against van Gerwen for his top 100 players), and see what our two projection systems would give them against van Gerwen, and then convert to odds. We'll use decimal (if you don't get decimal odds, google a converter) because it's easier in Excel obv:


Where a ranking is in red, it's because they didn't make the (active, where applicable, hence Ross Smith not being red in the FRH column because Taylor's still drakking around in the 30's) top 50 non-MvG ranking for that particular section. Lendel shipped me the percentages for the top 100 in his ranking, but Lewis, Norris and van de Pas were all outside (at 127, 181 and 185 respectively as far as I can see) - so I've assigned them what the number 100 was projected to be, but if you're thinking of betting on any of those three at all, you're nuts, so it may just be as easy to chuck them out of the data whatsoever. There's not going to be a huge difference between 6.4% and the actual figure anyway.

You'll see that of the players that I bet on, the market price was in the region of 6/1, with some slight variations - that's 7.00 in decimal (the quick conversion is to add one to the first figure then divide it by the second). Peter Wright's 9/2 (so 5.5) for his game this evening (which means the bet on Bunting lost obv). Splitting the difference between our two methods, it's only the very bottom few players where you'd consider not betting on them if offered 6/1 - on my rankings, it's basically nobody.

You'll note that I'm getting much higher winning chances than Lendel is for the non-MvG player - significantly so in some cases. There's a few possible reasons for this - firstly, as far as I can tell, the Elo system is just using wins and losses, maybe also adjusting for size of victory - so it's not picking up that van Gerwen's been pretty ordinary in an awful lot of matches, but he's still been getting the cake in more or less everything he's entered, so his Elo rating will be staying high in relation to the competition.

Secondly, I'm just using data from this season - where not only has van Gerwen not been playing that well, so his overall figures will be down, he's also possibly been running bad - his consistency score is much better than the database average, because his losing average is that high, over 95 - a point and a half higher than anyone in the database, and three points higher than anyone not called Price, Cross or Hughes (with a 100 leg minimum in 2019, only 18 players are even over 90). This'll have the knock-on effect of possibly underrating MvG in games against almost anyone, although Gurney, Payne, Suljovic, Searle, Monk, Dobey, Huybrechts, Clemens, Klaasen (really), Adie and Labanauskas all have better consistency ratings than van Gerwen. van Gerwen's Elo at the start of the season will have been pretty high on account of just having binked the worlds, it is not as if everyone is starting on 1500 from the start of 2019.

Thirdly, I assume Lendel incorporates the Premier League and the Masters into his dataset, whereas I don't include any unranked events. van Gerwen winning the Masters and doing decently enough in the Premier League isn't going to have harmed his rating significantly.

It is interesting to see where our projections don't differ by much - Anderson is I guess because of a small sample size where he's not been great, so his numbers are lower than they should be. That clump of players between Aspinall and Lewis - Beaton excepted - all of them have perfectly fine consistency scores, which may be reflected in Lendel's ratings putting them closer to mine than they otherwise would be. Whitlock being fairly close in odds? I'm not sure how quickly Elo can fall, you can only lose one game in a tournament, so if you're able to get a scrappy win before going out, put together a run out of nowhere, or when losing you lose to decent players, maybe your Elo doesn't adjust quickly enough for the level of actual play you're putting in. On the flip side, look at Steve West - he's not actually been playing bad darts, his points per turn, overall, has been around where his overall ranking is, he's just been a bit unlucky with some close results and draws, that'll end up dropping his Elo a bit.

So do we just bet against MvG until further notice? If you're just using my figures, then go for it. Lendel's figures end up giving a bit of a closer line to the actual market, but compared to prices you actually see, they're still in the range of it being anywhere from a slightly decent bet to at least not a significant mistake (he'd have Lennon, Bunting and Wright at 15/2, 5/1 and 4/1, compared to the 7/1, 6/1 and 9/2 that's generally available). I'm not seeing anything to suggest stopping betting against MvG if our opponent is somewhat competent, at the very worst, if we assume that the bookmakers know what they're doing and are actually pricing things up correctly, if we assume that they want to limit potential liabilities by putting van Gerwen shorter than he should be on name value alone, then as long as we have enough books to choose from that there's not huge vig, then taking the best price we have on the other guy can't really be a mistake.

Finally, no, that 1.88 isn't a typo in the Cross row. On this year's data Cross projects to win the game - just.

No comments:

Post a comment