In response to questions we received over the weekend about why we parlayed Alistair Overeem and Rashad Evans in two separate 3-team parlays – one each with Francisco Rivera and Jacob Volkmann – we are writing this article to explain how there can be value in placing parlay bets and why other times they should be avoided.
Statfox.com gives us a good overview of how parlays can be good as well as how they can be bad. Some of the information they provide is a great basis for this article.
Common Belief That Parlays Cost More
Assuming fair odds are given by the bookie (i.e. individual odds are multiplied together to find resulting parlay odds), the Statfox article shows us that a bettor actually pays slightly less vigorish. Vigorish is defined as what is charged by the bookie. Obviously, paying less to the bookie is a positive of parlays.
It is important though, to make sure you are actually getting fair odds on your bets. Betonline.ag offers reduced juice on big parlays. This means that people are cashing in big on their large parlays since they are receiving better than fair odds on their bets.
You Make Less Money Betting Parlays vs Individual Bets
Wrong again, it is shown in the table on page 4 of the Statfox article that a bettor actually makes more money, assuming they expect to win at least 55% of the time. As the probability of winning increases (we expect to have a long term winning percentage well over 55% with our model, see bottom of article for estimations) so does the expected value.
Why Don’t You Just Parlay Everything?
As shown by Statfox, and assuming we are getting fair odds, (which we are at BetOnline.ag and Bet365) our expected value would be much higher if we simply parlay every possible fight we can. So why don’t we do this? Assuming our average correctness is 70%, (slightly lower than we expect) the probability of winning a 2 play parlay becomes just 49%. A 3 team parlay is 34%. While total profit in the (very) long run would be greater, our bankroll may not be able to withstand repeated losses if they all occur before we have a win. Wins would be big, but less frequent, and we prefer to avoid this volatility.
The strategy we use for parlaying is as follows. When a fighter is heavily favoured by the oddsmakers, we try and find another fighter or two on the same card who are also heavily favoured. This way, we will not lose as much should something unexpected happen. (Evans and Overeem both losing would be a great example!) In the case of a major upset, we are only losing 1 unit for the whole parlay while our profit is greater overall if everything goes as planned. Our data and model accuracy indicate a favourable outcome happening more often than not when we parlay this way using our models picks.
So Parlays Are More Profitable IF Your Accuracy Is High: What Do You Project Your Accuracy To Be?
We test our model by randomly selecting a single observation from our data (in our case, a fight from the past), then allow the model to calculate the expected winner based on a comprehensive list of statistics and variables. Since this fight has already happened we can immediately see if the model was right. By repeating this thousands of times, we effectively simulate random sampling. It is important to note that this method assumes past trends in MMA will continue and identifying them will aid in predicting future outcomes. Additionally, these numbers are not our ACTUAL historical prediction accuracies, but they do provide some great insight into the strengths of our model. An event like UFC 156 hurts our profits, but seeing variability and unexpected upsets improves our model because it now contains the stats that led to that particular upset.
Here are the results of using 10,000 random observations:
Overall model accuracy historically approximately 80%.
Heavyweight: 89% and accounts for 11% of total observations
Light-Heavyweight: 92% and accounts for 10% of total observations
Middleweight: 78% and accounts for 15% of total observations
Welterweight: 77% and accounts for 19% of total observations
Lightweight: 80% and accounts for 21% of total observations
Featherweight: 76% and accounts for 15% of total observations
Bantamweight: 70% and accounts for 7% of total observations
Flyweight: 100% and accounts for 3% of total observations
What Do These Numbers Tell Us?
Firstly, the flyweight percentage is on a very small number of observations so for now it can basically be disregarded. However the most important thing these numbers tell us is that Light Heavyweight fights are relatively predictable, while Bantamweight fights are the most unpredictable. The way we will use this in our future bets is that we will be very unlikely to bet more than 1 unit on a bantamweight or featherweight fight unless the fighter is very heavily favoured. In the Light-Heavyweight division, we will likely bet 2 units on the fighter our model indicates will win. The other divisions will be regarded on a case by case basis, taking into account our probabilities, the odds, and other variables. The predictions the model makes for Featherweight will be scrutinized more strongly than the predictions for Heavyweight for example. Most importantly, the overall accuracy of 80% makes us believe that we are capable of great success in the near future!