In 2012, a risk analyst named Nassim Taleb wrote about the “Green Lumber Fallacy” in his book “Antifragile”. Taleb explained how one of the most successful commodity traders of all time earned his wealth buying and selling green lumber – even though he did not know what green lumber was. The premise of the message is that too much data in markets can be detrimental. A better approach for many traders is to rely on small amounts of data and be deaf to the noise.
I have decided to put prose to practice for the next nine months and will only be using three simple betting models to the bet on MLS matches – a league I have not bet in before – and see if there is truth to the fallacy. I will use a combination of Total Shots Ratio, Expected Goals and Poisson Distribution to generate odds for each match and identify wagers accordingly.
Total Shots Ratio is the simplest of the three methods. It compares the shot attempts for against the shot attempts allowed to generate a percentage. This value gets compared to the rate of points earned per match to spot over and under performing sides.
Poisson Distribution, in short, is a way of generating the probability of a given number of events (goals) occurring during a set time frame (match).
Expected Goals is a little more complicated. It requires mapping each shot attempt of each match using distance and angle intercepts. The probability of scoring gets attached to each shot attempt. The sum of the quality of shots is used to predict performance.
Here are the projected probabilities I generated for each scoreline of each match this weekend:
As is the case with any model, when there is limited data the variance can be SIGNIFICANT.
This weekend the three models priced four of the nine matches (HOU v ATL, ORL v DC, DAL v RSL, KC v NYC) to a point where there was no advantage identified in the home, draw or away markets.
The three standouts were Toronto (vs Columbus), Philadelphia (vs New England) and Portland (vs Los Angeles). The three models had Toronto and Philadelphia and Los Angeles all installed as heavy favourites – prices too good to be true. Nevertheless, Toronto at 1.70 Philadelphia at 1.99 and Los Angeles at 3.18 are worth a wager.
Toronto kept all of the main pieces from their record-setting team last season and should threaten another all-time high in 2018. Columbus has to deal with losing 60% of their offence and 31 expected goals from last season in Kamara and Meram.
In all the sims I generated, Toronto was never higher than 1.60 – closer to the correct price than what the model suggests at 1.37 (too much variance).
New England returns Cody Cropper in goal after letting in 50 goals last season (13 more than expected). The market never caught up to adjust for the horrific road play of the Revolution last season (1-3-13, 6 pts, -31GD) and still have not for the season debut this weekend.
The highest sim price I generated for Philadelphia was 1.77 – again, closer to the correct price than what the model suggest at 1.34 (way too much variance).
It is always difficult to peg an away win in MLS (1.87 home goals vs 1.03 away goals avg. per match), and while the price has a ton of noise in it, the market is quite clearly overreacting with Los Angeles. The club came off an all-time poor season a year ago from an expected goal, total shot ratio and points per game standard.
The Galaxy have added a ton of talent, but until it comes together, the pricing needs to reflect the market standard. This healthy Timbers side should be much closer to 2.00 but certainly not as low as the 1.61 price the model suggest (way, way too much variance).
With the true prices all falling somewhere between the extreme model price and the market price, there is reason to back all three.
Adam’s Recommended MLS Week 1 Bets:
- Toronto FC 1.70
- Philadelphia Union 1.99
- Portland Timbers 3.18