ACCUSCORE SIMULATION 101

AccuScore’s been around for a few years now and for the most part, we’ve been hush-hush on revealing details of how our simulations work.

If You’re So Good Why Don’t You Move to Vegas and Bet for a Living?

In around 1,000 meetings with representatives from the media, other companies, investors, and the competition, I’ve been asked this question probably 999 times.  The one time I wasn’t asked this question was by my wife who hates gambling.

I have two thoughts / reactions to people that ask this question:

1. I bet you think you’re clever asking that question?
2. You’re not clever, it’s a stupid f**king question.   If I moved to Vegas and had $200,000 I’d probably end up with $250 to $300K at the end of the year.  So why would I move my anti-gambling wife to Vegas to make less than I make helping to run AccuScore?

If you’ve taken the time to read other AccuScore articles on ROI, accuracy, and performance, you’ll see that we don’t claim to be a crystal ball that makes a poor man rich.  We are a service that allows over 70% of our paying subscribers to report that for every $1 they spend on AccuScore, they make between $1.25 to $2.50 in profit.  The 30% that do not see a profit are inevitably those customers that are in a constant search for the crystal ball and immediately cancel the subscription the second they don’t double their money.

Why We Started Simulating In the First Place

AccuScore’s Simulations were created to fill a glaring void in sports statistical analysis.  My contention since I was a kid is that sports authorities either misuse statistics or misinterpret them.  For example, I remember being 8 years old and wondering why the Colts (before that damn Irsay family bolted for Indy) wouldn’t just run the ball twice on 2nd and 8 when the RB averaged 4.5 yards per carry.  It took me until I was 30 to actually take the time to look at yards per carry and realize that if you consider a run of 4, 5, or 6 yards as “average” run for most RBs, that runs of this length only occur around 25% of the time.  If a run of 4, 5 or 6 yards is so rare, why is it relied on as an indicator of rushing quality?

The reason is the sports statisticians and the public aren’t willing to look at stats like these the right way.  They don’t take the time to realize that rushing yards is not a statistic that is normally distributed and therefore reporting on average is totally inappropriate.

In recent times, there has been a major push to boil down individual player performance into a single measure of efficiency or even a single value like Fantasy Points.   I don’t think these efficiency stats are bogus, I just don’t think they should be used to make any real judgments on player quality.  For example, Dwayne Wade may be one of the most efficient players in the NBA, but the the New Orleans Hornets would not be better if they traded David West and a healthy Peja Stojakovic for Wade.  With Chris Paul, the Hornets are better off with great perimeter shooters like Stojakovic and West to nail jumpers off of Chris Paul penetration.  Efficiency ratings may help settle some debates or start intriguing debates, but I do not think they really help determine a player’s value to his team or to another team.
We started AccuScore to take all the stats that are available (box scores, play-by-play data, situational stats, etc.) and get the most value out of them as possible. 

AccuScore’s goal is to maximize predictive value from historical statistics.

Because we’re 100% reliant and quantified statistics, we only forecast based on tangible, quantifiable variables.  We do not directly incorporate motivation or other subjective variables.  We are the first to admit that these intangibles do impact game outcomes and player performance.  This is why we are not a crystal ball.  There are too many intangibles that can interfere with what our tangible stats predict to have a forecasting system that is crystal ball accurate.

If stats make up 60% of the game and the rest comes down to intangible variables, it would explain why AccScore’s overall gambling accuracy is in the 53-54% range. 

Our customers who are professional handicappers often pride themselves on properly factoring for intangibles so these customers tend to use AccuScore as the objective statistical advisor and when our advice matches their other systems’ predictions they have the 4 star lock.  Other customers are thrilled with generating consistent profits and therefore go strictly with AccuScore’s picks. 

Our only unsatisfied customers are those that cannot accept losses and would prefer to be the first ones investing into a Ponzi scheme, or they are gamblers that want to be given 4 star locks, do not want to do any homework on bets, and still want to make big profits.

How AccuScore's Simulation Works

Our Intellectual Property lawyers highly advise against publishing the rest of this article because they worry people will try to copy us.  To alleviate their concerns I’m only going to run through one sport because that will give you a sense of how all our sports work.  Second, for every detail that is included below there are 10 to 20 more that aren’t revealed without which, the simulations would totally suck, so I’m not too concerned about IP theft.

ACCUSCORE’S NFL SIMULATION METHODOLOGY

AccuScore’s NFL simulation program involves three primary steps.

STEP 1: CREATE MATHEMATICAL VERSIONS OF PLAYERS AND COACHES BASED ON PAST PLAY-BY-PLAY PERFORMANCE UNDER COMPARABLE CONDITIONS

There are a number of environmental conditions (ex. home vs away, dome vs outdoors, grass vs turf, rain, snow, wind, etc.) and match-up statistics (ex. #1 against the pass defense, #10 rushing offense, etc.) that correlate to both individual player performance and team play calling.  AccuScore’s NFL simulation stores all past game data (play-by-play and game box score data) along with team match-up statistics and environmental conditions. 

The environmental and match-up conditions for an upcoming game are determined by using a proprietary method of weighting past data with over 30 different match-up and environmental conditions, which then allow AccuScore to create a series of mathematical formulas that describe the following:

1. Play Calling.  The types of plays a team will call in different conditions and game situations.  For example, AccuScore’s multivariate analysis may determine that Team X may pass 55% of the time on 1st down, with the ball on their own 30 yard line, with neither team being up by more than 3 points, in the first half, indoors on the road.  It may determine that same team will pass 37% of the time when on 3rd down with 4 to go, on the opponent’s 30 yard line up by more than 9 points.  AccuScore’s simulator will store over 20 different play calling and team performance statistics describing their performance under different conditions and situations.

2. Individual player performance.  Each player is described using over 70 different simulation parameters that cover all possible player statistics (offensive, defensive, special teams).  For example, Peyton Manning’s player parameters would include things like taking 100% of a team’s snaps when playing, fumbling the snap 0.5% of the time, getting sacked 5.7% of the time, targeting Reggie Wayne 28.5% of the time, handing off to Joseph Addai 56.3% of the time, etc.  These are just a handful of statistics used to describe what each player does on the field.  Individual defensive players are described by how well opposing offensive players have performed against their team when they are active as well as their individual contributions to the defense. 

After step 1 is complete, AccuScore has mathematical versions of active players and play calling strategy that are based on past performance but precisely tailored to match known upcoming game conditions.

STEP2: SIMULATE A SINGLE GAME ONE PLAY AT A TIME

Much like a video game simulation, AccuScore’s NFL simulation plays each game one play at a time using the mathematical versions of players and coaches generated after step 1.  Because AccuScore is not a video game, we can simulate an entire game in fractions of a second while following all the rules and procedures involved with an NFL football game. 

While it may take 45 seconds for an offensive coordinator to call in a play, the team to huddle up, snap the ball, complete a pass for 8 yards and have the WR run 6 yards after the catch before getting tackled in bounds, in AccuScore simulations it takes milliseconds.  However, even if it is taking milliseconds the play called, the progression the QB makes before targeting a receiver, the potential for an incompletion / completion, the subsequent yards after the catch and tackle by a linebacker or safety are all 100% based on the mathematical formulas derived in Step 1.  These are 100% realistic approximations of what these players and coaches have done under the current conditions and game situation.

The outcome of a play is based on the simulation parameter but there is a random component. For example, say LaDainian Tomlinson runs up the middle and has a 9 percent chance of rushing into the second level of the defense (based on past performance of him and the offensive line), but the opposing defense and middle linebacker has a 5 percent chance of allowing a RB to into the secondary.  In the simulation the computer will use the offensive and defensive simulation statistics to create an overall probability of Tomlinson rushing for say over 8 yards (getting past middle linebacker). Say this value is 6 percent.  It would then randomly generate a value between 0 and 1.  If that value is say 0.03 it would then simulate a Tomlinson run that goes for over 8 yards.  The precise yardage is then determined by a series of additional offense vs. defense simulation parameters (calculated in Step 1).

STEP 3: REPEAT SIMULATION 10,000 TIMES

The Tomlinson example above describes a situation where he runs for over 8 yards.  If the random number generated were higher, say 0.78, it would have resulted in Tomlinson limited to a run of 8 or less.  How much less would again be determined by a series of calculations based on the offensive and defensive player statistical formulas generated in Step 1.

Even though each play-by-play game simulation relies on the same simulation formulas, by introducing the element of random chance each actual simulation can be slightly or extremely different than the previously run simulation.  For example, Peyton Manning may complete 65 percent of his passes over 10,000 simulations, but in 10% he may complete just 40% or less, in another 10% of simulations he may complete over 80% of his passes.

Each player’s performance and the plays being called by the computer versions of the coaches will vary (+/-) around the simulation parameters calculated in Step 1.  By repeating each simulation 10,000 times AccuScore can determine how the individual variation ultimately translates into key outcomes of the game including, but not limited to:  1) Each Team’s Chances of Winning; 2) Individual Player Statistics based on Simulation Averages and/or Simulation Ranges; 3) the Correlation of individual player statistic(s) to the ultimate outcome of the game (ex. How does a QB’s interceptions correlate to his team’s chance of winning?).

POWER OF SIMULATIONS

AccuScore is running a Monte Carlo simulation which is defined as a “class of computational algorithms that rely on repeated random sampling to compute their results."  Monte Carlo simulations are useful in studying complex mathematical systems with a large number of variables / moving parts.  While physicists, hurricane trackers and economists may be the people one thinks of when discussing Monte Carlo simulation applications, AccuScore’s founders feel that the NFL and other sports can not only benefit from Monte Carlo play-by-play simulation, they can be entertained and informed by them as well.