Hey it’s the world cup season - Tapping into a Machine Learning based paper
Once again as a Singaporean citizen, there’re no other choices but to place my bets in Singapore pools. Betting in SG is not my preferred way to grow my wealth because of the crazy ~15% spread in Sg pools. In international sites, the spread is usually close to 3-5%. Oh well, that’s illegal.
But because this is the world cup season, I decide to make some small punts! And of course with some analytical slant.
I came across this academic paper (https://arxiv.org/pdf/1806.03208.pdf) written by Darmouth professors. They use a suite of features from individual players’ abilities, to teams’ rankings and even Countries’ GDP! Looking at all the data fields used, the data collection process must have been a nightmare. The authors then use Random Forest with simulations (think of Dr Strange forays’ into 14 milion possibilities) to predict the results. The probabilites for each team reaches each round is listed in 1 of the pages.
As the probabilities listed are accurate only for round of 16 (non-conditionally), I shall only place my bets for round of 16 should I find any undervalued opportunities. In addition, as one of the group A’s game has been played, I’ll ignore the analysis from Group A since the ‘true probability’ has already shifted.
Also, if my analysis shows that 2 of the teams in the same group are undervalued, I’ll only pick 1 of them because of the possible non-independence issues.
Results
- According to the table at the end of the web page, only 5 teams are undervalued now! And they are Saudi, Morocco, Iceland, Switzerland and Sweden.
- Since Saudi has been trashed, I shall ignore them.
- And since these countries are in different groups, the round-of-16 probabilities used are strictly independent
Disclosure: - I bought into Morocoo, Icleand, Switzerland and Sweden with bet sizes of 200, 350, 350, 350. The allocation is based on ‘True - SG odds difference’ & ‘True - SG probabilities difference’. Fingers crossed and wish me luck!
Update: - Switzerland and Sweden won but the rest lost. So, I won 97 (got back 1347) while risking 1250. A return of 8% in a space of 1 month. May not seem much - but that’s the return in 1 month! Ok, that’s not a lot. - Earning a profit doesn’t mean I’m right. I believe I had a sound framework; that’s why I went with my decision. I think that’s the most important.
#Analyzing UEFA champons league odds
dat = data.frame(team = c("uru","rus","egy","sau",
"spa","por","mor","ira",
"fra","den","per","aus",
"arg","cro","nig","ice",
"bra","swi","ser","coa",
"ger","mex","swe","kor",
"bel","eng","tun","pan",
"col","pol","sen","jap"),
sg = c(1.15,1.16,2.1,11,
1.05,1.12,3.5,6,
1.03,1.53,2.3,4.2,
1.02,1.45,3.15,3.2,
1.02,1.85,1.9,4,
1.07,1.80,2,3.25,
1.03,1.05,4,8,
1.25,1.5,2,3),
ml_prop = c(86.6,50.4,45.5,17.5,
88.4,67.5,30.3,13.8,
85.5,59.0,39.2,16.2,
81.6,65.9,15.8,36.6,
83.5,58.9,36.2,21.4,
86.5,41.5,54,17.9,
86.3,79.8,22.8,11.1,
79.2,60.6,39.7,20.5))
dat$ml = 1/(dat$ml_prop/100)
dat$underval = ifelse(dat$ml < dat$sg,1,0)
library(dplyr)##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, uniondat = arrange(dat,desc(dat$underval))
dat$sg_prob = 100*1/dat$sg
print(dat)## team sg ml_prop ml underval sg_prob
## 1 sau 11.00 17.5 5.714286 1 9.090909
## 2 mor 3.50 30.3 3.300330 1 28.571429
## 3 ice 3.20 36.6 2.732240 1 31.250000
## 4 swi 1.85 58.9 1.697793 1 54.054054
## 5 swe 2.00 54.0 1.851852 1 50.000000
## 6 uru 1.15 86.6 1.154734 0 86.956522
## 7 rus 1.16 50.4 1.984127 0 86.206897
## 8 egy 2.10 45.5 2.197802 0 47.619048
## 9 spa 1.05 88.4 1.131222 0 95.238095
## 10 por 1.12 67.5 1.481481 0 89.285714
## 11 ira 6.00 13.8 7.246377 0 16.666667
## 12 fra 1.03 85.5 1.169591 0 97.087379
## 13 den 1.53 59.0 1.694915 0 65.359477
## 14 per 2.30 39.2 2.551020 0 43.478261
## 15 aus 4.20 16.2 6.172840 0 23.809524
## 16 arg 1.02 81.6 1.225490 0 98.039216
## 17 cro 1.45 65.9 1.517451 0 68.965517
## 18 nig 3.15 15.8 6.329114 0 31.746032
## 19 bra 1.02 83.5 1.197605 0 98.039216
## 20 ser 1.90 36.2 2.762431 0 52.631579
## 21 coa 4.00 21.4 4.672897 0 25.000000
## 22 ger 1.07 86.5 1.156069 0 93.457944
## 23 mex 1.80 41.5 2.409639 0 55.555556
## 24 kor 3.25 17.9 5.586592 0 30.769231
## 25 bel 1.03 86.3 1.158749 0 97.087379
## 26 eng 1.05 79.8 1.253133 0 95.238095
## 27 tun 4.00 22.8 4.385965 0 25.000000
## 28 pan 8.00 11.1 9.009009 0 12.500000
## 29 col 1.25 79.2 1.262626 0 80.000000
## 30 pol 1.50 60.6 1.650165 0 66.666667
## 31 sen 2.00 39.7 2.518892 0 50.000000
## 32 jap 3.00 20.5 4.878049 0 33.333333