I want to warn you about a deadly mistake that bettors make as my lesson for today. My last three educational emails about regression to the mean, separating signals vs noises, and when to “double up” were very well-received, so I’ll continue the series today with another powerful lesson. If you missed my previous three lessons, you can read about them below:
- Click here to read my sports betting educational lesson #1: How to Avoid Back-Fitting Data When Handicapping Sports.
- Click here to read my sports betting educational lesson #2: How to Use “Regression to the Mean” to Win When Betting on Sports.
- Click here to read my sports betting educational lesson #3: When Is It Smart to “Double up” on Bets?
Today is my 4th sports betting lesson for you: “The Coincidence Trap in Sports Betting”
Here’s a question I get asked often: “What’s your secret to creating so many winning betting systems?”
One approach I use is to always identify weaknesses and find all potential flaws in a betting system before approving it for use. In this email, I’ll walk you through a deadly mistake that people often make when developing betting systems.
Many people fail at developing winning systems because they automatically attribute correlation to causation. While two events can happen together, it does not automatically mean that one event is the cause of the other. In fact, the correlation might just be due to what is known as a coincidence.
When researching past historical data, you might find a trend that seems to show correlation. That, however, does not mean that you can expect the trend to automatically repeat. That’s because when you’re looking at any data set where there are many variables involved, you’re almost guaranteed can find correlations just out of coincidences alone.
So just because the numbers align in some special way based on past historical data, it doesn’t mean they are likely to do so again in the future unless you have a good understanding of how the relationship works. The problem is that most of the time, it comes down to a coincidence.
Here’s an example: Let’s say that you’re looking at past historical data of sports games to see how teams in the NBA are likely to respond after getting blown out by double digits. Here’s your hypothesis: An NBA team will likely come back and cover their next game if they lose by at least 20 points on their last game. Let’s go test out this hypothesis by going back and checking on past data.
So, you go back into the past records to see what happens when teams get blown out by at least 20 points. Let’s say you checked for the last 10 years of data. Regrettably, you found that the chances of an NBA team that lost by 20+ points to come back and cover on the next game is basically 50-50. So there’s no trend.
But wait! Lo and behold…You found that when a team loses by exactly 24 points, they were able to come back and cover the spread on the next game by 60% of the time over the last 10 years!
What a gold mine, right? Have you discovered something huge? Just wait for an NBA team to lose by exactly 24 points, and then bet on them to cover on their next game. Based on historical data, you can expect them to cover 60% of the time!
Here’s the problem with that strategy: You’re cherry-picking data to force a positive trend. Unless you can come up with a rational explanation as to why 24 is the magic number of points an NBA team must lose by (and not 23, 25, 32, etc.), then there is no reason to believe that this trend would hold up. Usually you can tell when a trend is bad if there is an arbitrary cutoff without any rational explanation.
In this case, ask yourself: Why 24 points? Why would it not work when a team lost by 23, or 27, or 29? Why did it cover 60% of the time on the next game only when an NBA team lost by exactly 24?
If you cannot come up with a rational explanation to the question above, then chances are: You’re simply cherry-picking data to force a positive correlation. As a result, you’re simply fooling yourself with a false positive That’s a deadly mistake that will lead you to the poor house.
The reason why NBA teams tend to cover the next game when they just lost by exactly 24 points based on the last 10 years of data might very well can just be explained by the fact that it’s simply a coincidence.
Coincidences are bound to always happen when you’re looking at vast amount of data. If your original hypothesis is that NBA teams who get blown out by at least 20 points are likely to come back and cover the next game, then that’s a lot of data points to cover. A team could come back and cover very well when they lost by exactly 20, or 21, or 22, or 23, etc., all the way up to 50+
There are 30 numbers between 20 and 50. Out of these 30 numbers, some of them are bound to show a positive correlation with winning on the next game simply based on coincidence alone. So even if you discovered that 24 was the magic number, that doesn’t mean that it’s a good trend. It just simply mean you found a coincidence, and coincidences happen literally all the time when you’re looking at vast amount of data. Coincidences don’t make for good positive correlations that will hold up to future events.
This problem gets worse: Since coincidences happen all the time when looking at vast data, you could run into many shady touts in the sports betting industry that claim they’ve got a betting system that has won at some incredible numbers over the years. They could actually even be honest when saying that.
But here’s the problem: Most observed correlations are due to coincidences, and not real trends. If anyone digs hard enough, they will eventually find something that wins at incredible numbers. Maybe it’s MLB teams losing by 13+ coming back the next game. Maybe it’s NBA teams that score more than 116 points for at least 5 games straight that will likely go Over the total on the next game. Maybe it’s NFL teams that lost exactly 5 consecutive games to likely cover the next one. The list goes on. The lesson to learn here is that when looking at large amounts of data, you WILL run into many correlations. But chances are, those correlations happened simply because of coincidences.
Here’s the key: When you do find a positive correlation, you must ask: “Why?” Then, you must be able to give a rational explanation. If you can come up with a reasonable, rational, and probable explanation, then you might be onto something. Otherwise, you are setting yourself up for a false positive because you might be fooling yourself with what could just simply be a coincidence.
Even when you are able to come up with a rational explanation, you still need to test for more data before jumping to a conclusion. This is typically done by throwing out all the results of your test so far, and then testing your hypothesis again on other unrelated years to see if the correlation still holds up.
For example: If you found that +3 underdogs covered very well in the NBA in the last 5 years, then the proper way to test whether that might be a coincidence or a real trend is first to throw out all of those data, then go back and test at least 5 more years before that, to see if the results still hold up. You do NOT include the initial 5 years of test in the results, because those were the years that you used to form your hypothesis, and those results would end up skewing the data.
That’s sneak-peek preview of the rigorous method I go through to develop all of my betting systems. What you’ve read so far is simply just the very start. It doesn’t even scratch the surface. The process to develop a winning betting system is far more involved and meticulous. This, however, is a good start for you to peek inside and see what it takes to create a true winning sports betting system.