What do you think a projection is?
All of the statistics i have been talking about are descriptive, not predictive. A projection will be based on past data, then using probability a number is projected based off of the same data that is used. You could use any number of sample sizes as the basis of your projection. For example career stats, however it is fair to question how useful data from four or five seasons ago is for projecting a players production for the next season, when more recent data is available, and more likely to be closer to what the player will do in the following season than the data from 5 years ago will be.
There are advantages of using a larger sample size. I feel that 3 seasons of data is enough of a sample in most cases to find the players current mean or average performance in the next season, and therefore subsequently in each season after that.
If I use 3 seasons as my data set or career stats. Both would be the basis for making projections, just a different sample size. There are several years where I have done both.
One of the things that happens when you are looking over player careers and projecting their career stats, is you will notice that the players rookie seasons are dissimilar to seasons that follow it. So that data from the rookie season is often off somewhat. So I will drop the rookie season of a players career stats from the sample size for this reason. Nice thing about a 3 year window is that it does this automatically once a 4th season of data for each player becomes available.
You ask why wouldn't I project out for Cooper to age 32 or 9 seasons. I did explain this in some detail already, in regards to the change and turnover for each team over the course of 3 seasons, it often causes the ancillary factors around the player such as coaching and supporting personnel to change from what was happening for a player that long ago (4 or more years) is not the situation they are in any more. This is also related to the idea of trying to find a players mean, or average performing season, which I think the 3 year sample size does a good job of doing that, finding the players current mean or average, which does change over time, as the player changes, and the situation around the player changes as well.
I do not have confidence in a projection for Cooper 4 years or longer from now still being valid based on the sample size currently available to me. My best guess or projection for what Cooper will do in 2017 is based on his two seasons already played in the league. A forward looking projection would be based off of what he has done, perhaps with a modification to his targets going up to the 150 range which would make him more competitive with other WR 1s like Green on a PPG basis, than he has been. But that is just a guess as far as one thing that I pretty much expect to happen.
There is also the possibility that Cooper continues to perform at a similar level in 2017 that he already has in 2015 and 2016. If he does put up similar numbers again for a third season, what would you project him for in 2018? I would project him to score the same amount of points again, as I have 3 seasons of data telling me who Cooper is in his current situation. If Cooper performs better in 2017 than he has before, then that improvement goes into the sample and his outlook or projection moving forward improves. If Cooper has a worse season than before, then that gets calculated into his value and overall it declines.
i acknowledge that two seasons likely isn't enough for me to know what Coopers average season looks like yet. It also has issues because I would like to not include rookie data in my sample for him, but that will not be possible until 2018.
So why don't I project Cooper out for nine seasons? For all of the same reasons why I don't use data from 9 seasons ago to project for the current NFL. Too much has changed for me to have confidence in its relevance or validity for the players value now.
If you did want to project Cooper out for 9 seasons, it would likely be based on whatever method you used to determine Coopers average NFL season. Maybe you have some bonus and discount in there based on player productivity curves, such as what Couch does. However the average season would still be the same thing carried forward every season of the players career that you want to count as him performing the same as what his current average is every year. For nine seasons, instead of three.
For the purpose of this discussion lets say Cooper is worth 34 VBD (which is what he has averaged over his two seasons) every season.
If I use that value for 3 seasons, that would be 102 VBD points.
If I use that value for 9 seasons, that would be 306 VBD points.
What happens though if Cooper has a injury or for some other reason his career just never reaches the upside that I do think is there for him though? You will have paid 3 times as much for Cooper because of those extra six years than I would have paid for him using my shorter time frame.
That is what I am trying to avoid. Overpaying for potential. Potential does not help me win now, points do.
One thing that does help if you are going to project over a long time frame such as 9 years is to recognize the
time value of money, or in this case, fantasy points. I was discussing this with Adam Hardstad a year or so ago and I forget the exact details, but basically each year after the first year had a percentage discounted for uncertainty, using finance principles of present discounted value, so for example 10% less for the 2018 season, 20% less for the 2019 season, 30% less for the 2020 season. The level of uncertainty increases for each season out so the rate of discount should be increasing with time, however Adam had a way to prevent this discount from ever becoming 100% at which point the value of the player would be nothing. However even if the discount rate stayed the same each season, by the time you got to season nine it would be a 90% discount of Coopers 2017 projected value, or next to nothing, just from time present discount calculation.
That would still be better than just valuing Cooper as being worth 34 points every season for the next nine seasons though.
Just working this out with the 10% hypothetical discount each year from year one. would be
34
30.6
27.2
23.8
20.4
17
13.6
10.2
6.8
Total 183 points instead of 102.
I am not saying one way is better than another, that is a matter of preference and what works for you. The player values are going to be very similar when applied equally to all players using any of these methods, because they will all be based fundamentally on what the players performance has been in the sample size used.
If you have ever noticed, projecting next season, no matter how well it is done, will contain a lot of errors and bad guesses, or projections. Each year you extend that out compounds that error. So bad inaccurate projections will not hurt quite as much when you are only counting 3 seasons, as they will if you were counting nine. If you have a bad projection, you will likely fix it at some point in the following season based on new information gained from the previously completed season. So what did you really gain by projecting out all those extra seasons?