For as long as sports have existed, sport gambling has been around the entire time. There are indeed records of citizens betting on the outcome of races during the very first Olympic games that were held in Greece in 776 BC (University of Pennsylvania History Department website). A few centuries later, Romans coming from all parts of society would gather in arenas to wager on the outcome of gladiator fights. Of course, times have changed, and so has the way gambling works. Nowadays, online sport gambling companies use algorithms to ensure that they can maximize their own profit, at the expense of the casual fan.
But is it possible to develop a strategy to ensure the casual gambler wins money? The clear and definite answer is no. If there was a way to assuredly make money using sports gambling, companies would fix it. Remember, their goal also is to make money.
However, this does not stop us from developing a strategy to at least gamble smartly and somehow improve chances! To do so, a data set, provided on Kaggle (see link at the end) was used. After cleaning the data and filtering for what was needed (i.e. Premier League matches), it is a total of 228 Premier League matches, spanning between December 2016 and May 2018 that could be used for analysis. For each match was provided the odds of each possible outcome (home win, tie and away win) as well as the result of the match.
Five basic strategies were first used. The hypothesis is that 1$ would have been gambled on every match, for a total of 228$ spent. Here are the returns that would have been observed, for each strategy.
|Betting Strategy||End Total ($)||Returns|
|Always bet on the home team||213.27$||-14.73$|
|Always bet on the away team||223.05$||-4.95$|
|Always bet on a tie||208.15$||-19.85$|
|Always take the lowest odd||241.67$||13.67$|
|Always take the highest odd||201.70$||-26.3$|
Surprisingly enough, always taking the safest bet by taking the lowest odd (representing the likeliest outcome) would have led to a total reward of 13.67$ (about 6%). What is not surprising is that all other strategies have led to losses ranging between 2% and 11%.
Like in finance, diversification likely is what leads to the best results. Diversification is the idea of managing risk by investing in various ways, to minimize the importance of a single factor on a portfolio (Investopedia.com). In our case, if we always take the lowest odd, all it takes is a few more upsets than usual to see the rewards go down.
In order to identify the best approach, I had to create various ranges using the home team winning odds. What this means is that for a given odd, a specific action would be taken. Here are the optimal ranges that were identified using the Excel Solver with the goal of maximizing the total returns while forcing the solver to choose only one betting option per match.
|Home team winning odds||What to bet?||Observations||Returns|
|From 0 to 1.19||Away team win||18||44$|
|From 1.2 to 2.95||Home team win||155||162.35$|
|2.96 to 15||Away team win||55||62.45$|
Concerning decimal odds, a 1.2 means that the event has 1/1.2 chances of happening (83%). Logically, if the odds for the home team are really high, it means that the away team is expected to win. We see that using those ranges would have given us a return of almost 18%. The solver combines risk and safety, as it suggests to bet against the home team when odds are too low (3 upsets in those 18 games were more than enough to make up for the 15 bets that were not successful, since the reward is so much bigger), but otherwise stays on the safe side. It is also interesting that in this strategy, the gambler would never bet on a tie.
Again, gambling on Premier League games should be done for fun and is not suppose to lead to rewards (betting companies want to make money too, you know!). It should also be noted that the strategy developed here might work for this small data set, but not at all times. Nevertheless, if you are to gamble on Premier League matches, having a strategy and diversifying will give you the best chances of breaking even!
Below is the original data set link
Below are the codes used.