Tuesday, 27 October 2015

How To Successfully Bet On NHL And Which Stats To Use

Analytically, hockey has many similarities to soccer. Both are invasive, fluid team sports, where games are won by the number of goals scored and conceded and there are a large number of readily identifiable and recorded precursors to scoring, namely shots.
Therefore, the overlap in how teams and players are statistically evaluated, both descriptively and predictively, in the two sports has much common ground.
Methods of analysis for NHL betting
The Poisson distribution used extensively to model past outcomes and predict future matchups in soccer also adequately describes the likely number of goals scored and allowed in regulation time in hockey. The sudden death nature of overtime scoring in hockey is better described as a binomial distribution.
Inevitably, Bill James’ Pythagorean Expectation has also been adapted for hockey in an attempt to identify those teams who may be experiencing unsustainable levels of luck. But, you need to account for the awarding of a point for losses in overtime or during a shootout.
However, the most accessible tool to evaluate the NHL involves counting shots.
Even the wins or losses from an 82 game sample size for a single NHL team in the regular season will contain elements of random variation that may mislead when attempting to assess a side’s true abilities in a relatively low scoring sport such as hockey.
Goals are more numerous than wins, and shots, both on and off target, even more so. The more extensive shots at goal data are seen as better predictors to future performance in the NHL than goals or simple wins, losses or points.
Collecting "Corsi" and "Fenwick" data
“Corsi” named after the originator of the stat describes shots that are directed towards the net and they include goals, saves, misses as well as blocked shots. This may be quoted as a single number, either for or against for a single side, as a differential of total shots or as a ratio of the total shots taken in a game or over a series of matches.
A similarly named stat, “Fenwick” removes blocked shots from the calculation and more recently attempts have been made to improve the predictive ability of shot data by weighting each individual component differently.
These core hockey stats can be further subdivided by collecting shot numbers when the score line is close and teams are playing in a more normal tactical setup.
Score effects, also referred to as “game state” in football, can skew shooting data, as trailing sides may adopt more risky approaches in search of a goal, while the team which leads may devote more of its resources to defence to maximize their chances of winning the game.
Shot domination, particularly in hockey is often taken as a proxy for possession of the puck and a side which dominates possession in hockey would appear to be more likely to be successful in the long term.
Therefore, by demonstrating an ability to consistently out shoot the opposition is often taken as evidence that a side excels in the process that will bring success, even if short term bouts of good or bad fortune may prevent this ability from revealing itself in a team’s current win and loss record.
The pitfalls of PDO
Unsustainable levels of shot conversion rates or save percentages are powerful short-term factors, which may temporarily elevate a side’s record above or reduce it below their core abilities.
These are combined to create the final useful hockey stat, “PDO”, or SPSV% to the NHL. Typically, a team’s PDO, which is the sum of their save percentage and their scoring percentage, will cluster around 100, although the stat is sometimes scaled to 1 or 1,000.
PDO values that are either significantly in excess or below 100 are often signs that a side has experienced unsustainably high levels of good or bad luck.
However, the tendency over time is for such extreme luck-driven outcomes to become less extreme in the future, and PDO values that may enhance or confound the rewards from a side’s Corsi or Fenwick data should be expected to gravitate towards 100.
It has become increasingly recognised that a team’s win record, even over a season is the product not only of the talent of the players, but also natural variation the exists in any sample of data and small sample sizes are frequently imprecise measures of true talent.
Just as a coin may produce “heads” in excess of the expected 50% through natural variation, a side may convert and save shots, and ultimately, win hockey games by a combination of luck and skill, with the former unlikely to persist at the extremes.
A team’s 5v5 Fenwick ratio from one season shows a much higher correlation to their wins in the following season compared to their PDO under the same playing conditions from the previous campaign.
Therefore, when projecting a team’s performance, either using last year’s data or a recent run of results, shot data should perhaps carry more sustainable predictive weight as measured by either Fenwick or Corsi ratios, rather than a particularly high or low PDO value, which will have a tendency to regress towards the league mean of 100.