How Sharp Are Vegas Handicappers In College Basketball?

How Sharp Are Vegas Handicappers In College Basketball?

Genesis

I’ve been an avid sports fan all my life, particularly football and basketball. One of the highlights of the year for me is the NCAA men’s basketball tournament, a.k.a. “March Madness.” As a former basketball player I have always appreciated the higher quality of play that occurs in the third and fourth rounds of the tournament. From a pure fan perspective, it is my single most favorite sporting weekend of the year. Prior to developing my own proprietary, statistical basketball metrics I regularly struggled to handicap the games in the third and fourth rounds of March Madness. The “David & Goliath” matchups that occur in the first and second rounds of the tournament aren’t present; the weaker teams have been weeded out of the tournament.

During my college years I was first exposed to sports gambling. While I found it exciting, I found the experience to be frustrating overall. I was just a middling-to-bad sports handicapper at that time of my life. Years later I was invited to go to Las Vegas for the opening weekend of the basketball tournament (rounds 1 and 2). I decided to apply some of my professional skills as a stock and bond analyst/trader to see if I could meaningfully tilt the odds in my favor during our annual trips to “The Strip.”

I developed a series of four metrics that measured quality of defense, pace of play, efficiency, and consistency that looked promising in testing. At the sportsbook my real-time results over a three-year span the win-rate with my metrics was approximately 76%. Typically my wagers were less than $200 per game and limited to the strongest consensus bets highlighted by my basketball metrics. I was pleased that I could cover the costs of my trip and take a little home with me for my efforts. To be truthful, my overall trust-level wasn’t high enough to throw significant amounts of capital behind betting my metrics. I simply hadn’t performed a rigorous testing of my metrics to confirm to myself that it was no more than a successful lark.

That has since changed, however. I have performed rigorous statistical analysis on my own metrics in an effort to see if any improvements could be made. In doing so, I was curious how much opportunity is routinely available in the Vegas lines. In short, I was curious to know exactly how far off are Las Vegas handicappers when to it comes the NCAA basketball tourney.

Getting accurate historical information on college basketball on both performance statistical lines isn’t the easiest task to accomplish but I managed to secure data for March Madness for 2013-2014 for free. The results were illuminating.

At this point I’d like you to guess along with me. My preconceived idea was that tournament lines would be pretty close to the actual outcome. I suspected that the average differential between the Vegas line and the actual outcome would be 4-6 points with a standard deviation of 2-3 points. Further, I suspected that this differential would narrow in each subsequent round of the tournament and that the standard deviation of error would also tighten as the tournament went along. What would your guess be?

For your info, the NCAA basketball championship tournament has four play-in games to complete the field of 64, then 63 games to determine a championship over the course of a three week span ending sometime near April 1st. Within my two year data sample I have 134 games but the results were startling similar from year to year.

In the table above I’ve broken down the average standard differential of the Vegas line to the actual outcome. Given the limited sample size there is some volatility in the per round results. However, there is some nascent evidence that the error decreases as teams get eliminated but the standard deviation of outcomes stays stubbornly high. Over the course of 134 games, however, the data are remarkably consistent in my two-year sample; expect the actual outcome to differ by about seven points from the Vegas line with a 6.0 standard deviation. As a result we can conclude with a 67% confidence level that the actual outcome will be 1-13 different than the Vegas line. That is a GIGANTIC amount of opportunity for a handicapper to play with, much more than I would have guessed before embarking on this study.

If this sample is representative of college basketball handicapping in Vegas, then it seemed reasonable to me that the teams with the lowest performance variance should beat the line with a high degree of regularity. Hence I designed a series of metrics to measure for low performance variance in basketball stats. I won’t be more specific on those metrics, I intend to keep them for myself only.

And The Results Are In…

The tables below are the output tables generated from those metrics for both the 2013 and 2014 March Madness tourneys. The Vegas line is in bright red and my picks are shaded either by olive green, bright yellow or dark gold cells. Olive is a lower confidence pick, call it a 1 star pick. Bright yellow is a higher confidence pick, call it a two star pick. Dark gold is the highest level confidence pick, call it a three star pick.

The combination of metrics I used in real-time while in Vegas had a 76% success rate. After returning from Vegas I realized that my measures of performance variance were significantly more accurate and revised my output table to reflect as much. The results of back-testing the 2013 and 2014 tournaments was staggering: 2013 results were 58-7-2 versus the line for the entire tournament while 2014 posted a 56-8-3 success rate versus the line for the entire tournament. For the two-year span that’s a 114-15-5 record versus the line.

In 2013 the metrics picked 38 favorites, 27 underdogs and two pick ‘em teams. In 2014 the metrics picked 36 favorites, 30 underdogs and 1 pick ‘em games. During the 2013 tourney the 2 star and 3 star picks went 17-3-1. In 2014 the 2 star ad 3 star picks went 14-1-1. The winning percentage for the 2 star and 3 star picks was slightly higher than that of the 1 star picks but the real advantage occurred as the winners covered the line at a much higher than average margin, typically the average margin plus one standard deviation. For the two year span, that means that 2 star and 3 star picks covered the line in excess of 13 points on average. The “comfortable” margin of cover suggests that 2 star and 3 star picks would be well-suited for some sort of parlay strategy that adds leverage against the casino.

Obviously these results sound much too good to be true. I am a natural skeptic myself so if I were reading this I would take it with a grain of salt. After all, I am only working with a two-year sample of data. In response I would offer that the less refined version of my metrics yielded a 76% success at the Wynn Sportsbook in 2014. I won’t be able to confirm or deny my new basketball metrics’ success rate until next March, however.

Now The Gridiron Experiment Blog

Hopefully I’ve demonstrated what potential opportunity there is to exploit the Vegas line in college basketball if you’re measuring the right metrics in your handicapping. I’ve done some preliminary testing of college football metrics that looks promising but the work is incomplete. Hence the creation of this blog. This blog will be aimed at handicapping college football games and potentially NFL games. This is an ongoing real-time experiment. I am as curious as all of you to see if I can successfully handicap football games something north of 62% success rate clip.

More to follow soon.

Jack

NCAA Tourney Round By Round Results 2013-2014

Round 1 Results

Round 2 Results

Round 3 Results

Round 4 Results

Round 5 & 6 Results