The trouble with the tout industry
Examining the five biggest problems with selling picks
Originally Published: October 29, 2014
<cite class="source"> By Jeff Ma | ESPN.com</cite>
<cite>AP Photo/Las Vegas ***************/John Locher</cite>Bettors are always looking for an advantage, and touts can seem like an easy path to one.
"Those who can, do: those who can't, teach." -- George Bernard Shaw, "Man and Superman"
The sports betting equivalent of the quote above is "Those who can, bet: those who can't, tout." Touts, or people who sell picks for a living, have been around as long as the betting industry itself and have helped give the industry some of its (deserved) perceived seedy reputation.
There are many inherent flaws with selling picks, which I'll detail below, but first and foremost, it's important to understand that the best handicappers I know don't sell their picks. As a consumer, you must ask yourself: If someone is so good at picking games, why are they selling that info?
The usual refrain is that touts can make more money selling the info than they can actually get down on games themselves due to liquidity issues. But this highlights one of the big moral hazards with selling picks: If you are any good in a limited liquidity sport (like NCAA basketball), you will move lines when you release picks, and only a few of your customers will be able to benefit from your info.
In fact, the minute you decide to sell your picks, you naturally must be concerned with selling and marketing said picks. And how you decide to market those picks poses a big ethical challenge -- a challenge where most touts fall short.
Below I'll look at the biggest red flags I see surrounding the tout industry and provide a personal guideline I follow when giving picks.
Lesson 1: The context of stats is paramount
When touts sell picks, they feel compelled to create a narrative around those picks. Often that narrative plays well if stats and numbers are attached. "The Detroit Lions are 10-1 against the number in their last 11 Monday night games," a tout may claim as a supporting stat for a play on the Lions.
But the problem with these narratives is they likely mean nothing at all.
Not only is 11 games not a substantial sample size, but many different players and coaches participated in those Monday night appearances for Detroit over the course of many different seasons. This statistic has no predictive relevance.
The more meaningful stats are those that aren't in the mainstream and represent a contrarian opinion from the public. For example, teams in the NFL can become overrated early in the season when they have a high turnover differential. Turnover differential is difficult to predict in the future, so teams that perform well largely due to turnover differential are likely to be overrated going forward if they can't sustain that turnover rate.
One example this season is the 4-3 Cleveland Browns (ranked fourth in the NFL with a plus-7 turnover differential, but a defense that's 30th in rushing yards allowed, 20th in pass yards allowed and 20th in opponents' yards per play).
Similarly, the 6-1 Arizona Cardinals are ranked second in the league in turnover differential, yet rank in the bottom third offensively and defensively in yards per play. If they can't maintain that turnover differential, they are likely to have some struggles in the second half of the season.
Against-the-spread trends are especially troubling, as they sound great when quoted, but the reality is they likely have very little predictive value. A team that has covered five games in a row is no more likely to cover in its sixth game than any other team. In fact, you could make a case that a team that has covered five games in a row is likely to become overrated by the linesmakers or the public. Again, though, without actual predictive stats, these line trends mean nothing.
Make sure whenever you hear a stat being used to illustrate a point or sell a product, you look beyond the numbers and try to understand the significance, if any, of the statement itself.
Lesson 2: Be wary of selection bias
Many touts will quote their record publicly, but it won't be their complete record -- just a subset that paints their performance in the best light. (i.e.: "I'm 10-1 in my last 11 Monday night games.") Some will just quote their recent performance when it suits them best ("I'm 65 percent over the last four NFL weeks"). This is a common mistake called selection bias, where there is an inherent flaw in the sample that is being used to draw conclusions. Anyone can make a case with numbers if they get to choose their sample and sample size.
In order to really assess someone's performance, you need to look at a statistically significant amount of time -- and that is certainly more than 11 games or four weeks over the course of one NFL season.
Lesson 3: Sports aren't necessarily stable over time
I'm always wary of those that say they have a "mathematical model," because it's often an excuse to not fully disclose their methods (more on that later).
But let's assume for the moment that a tout has created a good model and has back-tested that model successfully. Without knowing more about his/her method, there's no way to know if that model is stable over time. Let's take a model that tries to predict NFL scores and uses six years of historical data to make its predictions, for example. A lot has changed in the last few years in the NFL. An increased emphasis on penalties in the defensive secondary, along with better QB play, has led to higher scoring. In the early 2000s, the average score for a team hovered around 20 points; now the average team score is closer to 23 points.
Similarly, the penetration of advanced analytics into on-court strategies in the NBA has fundamentally changed that sport. Teams are taking more 3-pointers than ever: In the 2009-10 season, teams averaged 18.1 3-point attempts per game; this past year, that number rose to 21.5. So a model built on historical data is likely to run into some issues unless it has significant in-season adjustments.
That is a challenge for all model builders: How much data do you look at and how much do you weigh recent data? In all cases, these are complicated questions, and without some explanation of a tout's model, you don't know how he/she is handling these challenges.
Examining the five biggest problems with selling picks
Originally Published: October 29, 2014
<cite class="source"> By Jeff Ma | ESPN.com</cite>
"Those who can, do: those who can't, teach." -- George Bernard Shaw, "Man and Superman"
The sports betting equivalent of the quote above is "Those who can, bet: those who can't, tout." Touts, or people who sell picks for a living, have been around as long as the betting industry itself and have helped give the industry some of its (deserved) perceived seedy reputation.
There are many inherent flaws with selling picks, which I'll detail below, but first and foremost, it's important to understand that the best handicappers I know don't sell their picks. As a consumer, you must ask yourself: If someone is so good at picking games, why are they selling that info?
The usual refrain is that touts can make more money selling the info than they can actually get down on games themselves due to liquidity issues. But this highlights one of the big moral hazards with selling picks: If you are any good in a limited liquidity sport (like NCAA basketball), you will move lines when you release picks, and only a few of your customers will be able to benefit from your info.
In fact, the minute you decide to sell your picks, you naturally must be concerned with selling and marketing said picks. And how you decide to market those picks poses a big ethical challenge -- a challenge where most touts fall short.
Below I'll look at the biggest red flags I see surrounding the tout industry and provide a personal guideline I follow when giving picks.
When touts sell picks, they feel compelled to create a narrative around those picks. Often that narrative plays well if stats and numbers are attached. "The Detroit Lions are 10-1 against the number in their last 11 Monday night games," a tout may claim as a supporting stat for a play on the Lions.
But the problem with these narratives is they likely mean nothing at all.
Not only is 11 games not a substantial sample size, but many different players and coaches participated in those Monday night appearances for Detroit over the course of many different seasons. This statistic has no predictive relevance.
The more meaningful stats are those that aren't in the mainstream and represent a contrarian opinion from the public. For example, teams in the NFL can become overrated early in the season when they have a high turnover differential. Turnover differential is difficult to predict in the future, so teams that perform well largely due to turnover differential are likely to be overrated going forward if they can't sustain that turnover rate.
One example this season is the 4-3 Cleveland Browns (ranked fourth in the NFL with a plus-7 turnover differential, but a defense that's 30th in rushing yards allowed, 20th in pass yards allowed and 20th in opponents' yards per play).
Similarly, the 6-1 Arizona Cardinals are ranked second in the league in turnover differential, yet rank in the bottom third offensively and defensively in yards per play. If they can't maintain that turnover differential, they are likely to have some struggles in the second half of the season.
Against-the-spread trends are especially troubling, as they sound great when quoted, but the reality is they likely have very little predictive value. A team that has covered five games in a row is no more likely to cover in its sixth game than any other team. In fact, you could make a case that a team that has covered five games in a row is likely to become overrated by the linesmakers or the public. Again, though, without actual predictive stats, these line trends mean nothing.
Make sure whenever you hear a stat being used to illustrate a point or sell a product, you look beyond the numbers and try to understand the significance, if any, of the statement itself.
Many touts will quote their record publicly, but it won't be their complete record -- just a subset that paints their performance in the best light. (i.e.: "I'm 10-1 in my last 11 Monday night games.") Some will just quote their recent performance when it suits them best ("I'm 65 percent over the last four NFL weeks"). This is a common mistake called selection bias, where there is an inherent flaw in the sample that is being used to draw conclusions. Anyone can make a case with numbers if they get to choose their sample and sample size.
In order to really assess someone's performance, you need to look at a statistically significant amount of time -- and that is certainly more than 11 games or four weeks over the course of one NFL season.
I'm always wary of those that say they have a "mathematical model," because it's often an excuse to not fully disclose their methods (more on that later).
But let's assume for the moment that a tout has created a good model and has back-tested that model successfully. Without knowing more about his/her method, there's no way to know if that model is stable over time. Let's take a model that tries to predict NFL scores and uses six years of historical data to make its predictions, for example. A lot has changed in the last few years in the NFL. An increased emphasis on penalties in the defensive secondary, along with better QB play, has led to higher scoring. In the early 2000s, the average score for a team hovered around 20 points; now the average team score is closer to 23 points.
Similarly, the penetration of advanced analytics into on-court strategies in the NBA has fundamentally changed that sport. Teams are taking more 3-pointers than ever: In the 2009-10 season, teams averaged 18.1 3-point attempts per game; this past year, that number rose to 21.5. So a model built on historical data is likely to run into some issues unless it has significant in-season adjustments.
That is a challenge for all model builders: How much data do you look at and how much do you weigh recent data? In all cases, these are complicated questions, and without some explanation of a tout's model, you don't know how he/she is handling these challenges.
