It’s fun (and bit discomfiting) to look at the Oilers’ lineup and realize just how young it is. Only 6 players on their roster are 30 years of age or older, their top line has an average age of 20.7, and their 2nd line has an average age of 23.7. Obviously, this team will only go as far as its Dream Teens will take it, but what kind of an impact will the veterans have? In this post I’m going to try to predict what the 10+ year NHL veteran skaters are going to do this year for the Oilers.

In previous posts, I’ve used a lot of regression and linear optimization techniques to try to figure out how a player will perform. The main reason for that is with so few data points (prior season for younger players), it’s impossible to glean much of a trend intrinsic in the data series itself, so you need to find a population of comparable players to figure how how they did in similar situations as our subjects. With 10+ year veterans, we have at least 10 prior seasons of data to draw upon, so we can concentrate solely on that player to see what kind of insight we can draw from their progression as an NHL player alone.

To accomplish this, I’ll use a forecasting technique known as Exponential Smoothing. There are three main versions of this technique, and I’ll be using a version known as double exponential smoothing. I’ll spare you the math details, but basically this is a technique which uses smoothing factors to figure out what weight to give a) prior data points in an exponentially-decreasing fashion and b) any trend in previous data points. The technique produces an estimated actual for each preceding time period, and then compares this to the actual figure. I’ll again use Solver in Excel to minimize the sum of squared errors between the estimated actual and the actual observation. This will come up with the best two smoothing factors that fit past data. Then you can simply propagate the formulae to represent a future data point, which in this case is the upcoming NHL season.

I performed this analysis separately for the points per game levels for each of Ryan Smyth, Shawn Horcoff, Eric Belanger, Darcy Hordichuk, and Nick Schultz.

The two smoothing constants are seen as a and b in the above tables. The higher ‘a’ is, the more value it puts in more recent observations — in Nick Schultz’ formula, ‘a’ is so high each forecast is essentially just the last observation’s value. The lower ‘b’ is, the less trend amplification it imparts — in the cases of Horcoff, Hordichuk, and Schultz, the forumla decides that no useable trend can be discerned at all.

Using these point per game levels, we can then forecast what they would score over 82 and 48 game equivalent seasons:

So, do I believe these results? Out of all of them, I’d say Horcoff’s is likely a touch high, considering he’s not starting the season on any power play unit. However, if Hartikainen doesn’t produce results with the kids, Horcoff would be the direct substitute on the Wizard’s power play unit. Belanger will also be in tough to get this many points while leading what I’d like to call the ‘Participation Ribbon’ line, or the Oilers’ 4th line. I also don’t think Krueger will have the same compunction as Renney in forcing Belanger off a power play unit. Nick Schultz may very well score more than a point every 10 games if he stays partnered with Justin Schultz — we’ll call this the Martin Marincin effect, as the lanky Slovak rookie was the beneficiary of quite a few bonus points while skating with J. Schultz in Oklahoma City this year. I have the highest confidence in the forecasts for both Ryan Smyth and Hordichuk. Smyth’s total is a decline from previous years, but still reasonably high considering he’ll get more icetime than a usual 3rd liner gets at even strength and he’ll be on the 2nd power play unit for the entire season. Hordichuk… well, let’s just say he can swing a sack of door knobs.

## One Comment

Yay for the focus on the vets. Thank you! I agree with your assement that the results seem a touch high, but suspecting that it is just reflecting the methodology limitations. And in other news, this one we are totally going to chat about this week. I’m totally intrigued by the math and am wondering about potential other applications. Now I am off to look up “compunction” in my online dictionary. 😉