"The waiting is killing me."

Excellent post! Really interesting stuff!

There are dozens of well-supported posts on this site, with insights that pre-date the “official” rebuild, calling for pieces that have yet to be added and whose absence is evident every game, cough RH 3L PK C with FO%>50 cough.

But everyone who posts or comments here IS human. And humans tend to behave in ways that are predictable, as the links remind us. So I would accept the idea that even well-reasoned arguments come with human baggage and biases. No sense pretending that this is the pre-eminent bastion of objectivity on the planet; that just reinforces how human we really are.

I was sort of expecting that somebody would accuse me of the gamblers fallacy. That is the idea that if a coin has come up heads ten times in a row then it most be more likely to come up a tail on the 11th flip than it was on the first flip. There used to actually be a betting system for coin flip games known as a martingale where you just kept doubling your bet until you won. Much modern probability started from attempts to show it invariably ended in bankruptcy.

Over time mathematicians actually invented/discovered a series of functions with great usefulness in stochiastic calculus which are known as martingales. Simply put, there are two kinds of probalistic spaces in the world; ones with independent variables and ones with interdependent variables. There are many great tools for analysing spaces where the variables are independent, you’ve all probably heard of random walks which is one of the most popular approaches. There aren’t so many if the space is interdependent, as any team competitive sport is, the best known approach is to use a Dobb Martingale and an Azume inequality. Again simplifying, in a probability space like hockey occupies while we still can’t predict when the reversion will occur we can identify stopping times which may, or may not, be predictive and may or may not reveal the process (hockey outcomes) to be a true martingale, a semi-martingale, a super martingale, a sub martingale, or a local martingale. Once we know which it is we can more cleverly choose our mathematical tools of choice for further analysis.

Hockey (if all we are concerned with is the intransitive nature of team interactions) is a local martingale and thus such things as hockey winning or losing streaks are almost impossible to call, often persisting longer than the math would predict. However, when viewed in terms of team performance over longer times it begins to conform (in the sense that the best fit algorithm is one) to a super martingale. Believe it or not there are raging debates within mathematics like any other field, say hockey blogs. I come from a camp that would argue that if hockey is a super martingale then we should be able to find a Stopping Time that is predictive based on past history. We should be able to call the reversion while not being able to predict its magnitude (mostly reversions, even in interdependent spaces, are only to the average – though there are obvious exceptions).

The first thing that becomes apparent as you do that sort of analysis is that Michael’s paper on luck has hit the nail on the head. Skill makes the manifold for returning to the mean sticky. This is probably self-evident but good players make good teams even in a sport with a lot of luck. Similarly, bad management can keep a team from rising up. The reversion is sticky both ways.

Complexity theory says that minor pertubations can make a complex system wickedly unstable, often very quickly. Management no longer has an impact (this is one of the hardest parts of transmitting the insights of complexity, chaos, and game theory to corporate management – they need to accept that conventional management has situations where it is nigh on useless – or as many of you here would say about the Oilers – management is always useless). Thus, there is a point where the instability of the system predicts the outcome not active decisions by any part of the system. This is Malcolm Gladwell’s famous tipping poiint theory in action. In fact in many well known examples this is the moment power laws come into play. See the Social Network or read The Accidental Billionaire.

I am merely suggesting that in trading Penner the Oilers management tipped their system into the kind of instability in which their own decisions are now utterly irrelevant. Faginni and others would suggest that they should try to make the instability worse in an attempt to make it better. Thus my “trade anything that isn’t tied down” which I see as a great way to create utter chaos a state that seems to be pretty essential to avoiding the fate of Toronto, Atlanta, Florida, etc.

Derek – Loss aversion applies just as well whether it is the movement of a beloved player or the loss of your dominant meme. Hereabouts that meme is get a shut down D, a 3C who can take faceoffs, a real goaltender, and veterans to support the rookies because they need to be sheltered. In a couple of years, with a little luck we will be good again. A long rebuild pretty much ensures all those things are off the table. So we react emotionally to the loss of the opportunity to see that we were right and assume we were right and that this trade must then be complete crap. We then extend the bias (supporting it with frequent stunning examples of what Tvesky and Kahneman call the “law of small numbers”) to the rebuild. It must also be crap. The trade of course also challenged the status quo, it changed the game space from being about how to round out the team to how to rebuild it and much of our accumulated knowledge is now worthless and frankly I am not taking it well and suspect you aren’t either. Think of all the brilllant work on whether or not face offs matter, now the questions are more like what is the optimal mix of draft picks, what PPI should we be aiming for, etc. It ain’t going to be pretty and new voices and ways of thinking may well emerge.

It eased Tambellini’s fevered brain. It resolved his cognitive dissonance. He no longer had to worry about his best player leaving for nothing. Plus he got those draft choices. Each is exactly like a lottery pick. They each bring a chance at the grand prize; a great player. The combination was enough to overcome his natural commitment to the status quo.

i’m all for using cognitive research to better understand hockey behavior, but it’s indefensibly reductive to take general principles that effect the decision-making habits of the population at large and claim that they explain why one person made one particular choice. especially when that choice had the kind of research and analysis tools available that nhl teams can bring to bear on trades and signings. you can’t know what, if any, of this information tambellini used or didn’t use, but you can’t deny that it was far more than that presented to allais’ test subjects concerning their hypothetical vacations.

An interesting article, I think perhaps what you don’t deal with is that while Steve Tambelini may have analysed this as “known” versus “unknown” whereas trading Penner for assets increased his certainty, most fans have interpreted the situation in a diametrically opposed fashion.

The difficulty most of us have with this trade is that Penner represented a “known” quantity, and a fairly valuable one. It was a sure thing that we had him for the rest of the season, and next season. While Steve Tambelini may have felt “uncertain” about whether he could be resigned, the attempt to resign Penner could be made and at worst he could be traded next year at the trade deadline to prevent the team from seeing him walk away without nothing.

Instead we are faced with a return that if calculated generously has approximately a sixty percent chance of producing a regular contributing NHL player, and the chances of that player being equally valuable are substantively less than that.

Hence we have a scenario to use your earlier example of 100% chance of a 1 week vacation at a four and half star Chez Penner resort immediately, with the possibility of winning additional time at said resort immediately or at some other unknown resort at some time in the future for an unknown period of time.

This is contrasted with a scenario where we have several possibilities of approximately 20-30 percent of winning an undefined stay at a resort of undefined quality.

Now in economics its said to be rational to attempt to maximize one’s utility. Here we have a sure thing, and if that is not extended into a longer stay we recieve a chance of some other value in an ill defined time in the future. We known we are getting something immediately, and the quantity may improve over time.

The alternative that Tambellini chose involves a substantial chance of the “magic beans” scenario where the oilers scratch their lottery tickets and get a big “Sorry better luck next time sucker”, and even presuming a payoff may only be of an equal or lesser value than they already had, with a slight possibility of being more valuable. Overall strong likelihood of less total utility than the first scenario.

The fans have looked at this and said “we have chosen the scenario less likely to produce the most value, our GM is irrational”. I don’t think this speaks to risk aversion among the fan base so much as the deficiency of any analysis being performed by our management.

Scott,

As I said in my post, fans prefer to judge by the here and now rather than the ultimate outcome. On the other hand in any real world results do matter. Your idea that the first and second Pronger trades aren’t related makes analysis easier but not more accurate. Consider if I bought Gold when it was at 400 dollars but managed to that gold at 30 cents on the dollar. I paid $120 and ounce. Today I sold it at $1200 an ounce with the market at $1400. I made a great deal on the first transaction, got hosed on the second, but overall came way out ahead. Thats the Pronger trade and I am not conflating them, merely pointing out that for those of you who keep preaching asset management while you can fairly say the second transaction was bad you can’t ignore the aseet management value of the two transactions combined.

In my gold trade example the point is the total return not only exceeds the investment but out performs the market. I wasn’t suggesting that people who belong to the instant evaluation school of rating trades, (and remember in the original post I did say that is the vast majority of evaluators), were in anyway shape or form changing their sentiment on the second trade. In fact, I was implying that there is a movement amongst people who use the longer “results” driven view of a trade to the position that an argument can now be made that the trades taken as a whole were good asset management. So on that we agree. I’d also agree with you that the first trade in the Pronger series was a long way beyond good and the second was really bad. The net effect however was good asset management.

The point I was trying to make are, that there are a small fraction of people who will take the “wait and see” approach on evaluating any trade and that there position isn’t wrong it is just different from the rest of us. They have turned out to have a point when looking at the building of organizational value through the Pronger trade at this stopping point, at this moment in time. That has to make you wonder about the validity of an argument that goes we screwed up the Pronger trade thus we must have screwed this up.

I think there is great validity in looking at each and every transaction an organization makes on the merits of that specific transaction. I just think we keep rushing to judgement and making those evaluations before the evidence is in. For the organization you would hope that over time they will learn from each transaction and improve their overall decision making and improve their transactional skills. To get there they have to wait for actual results. They can’t afford to base their organizational analysis on guesses or they and we are all doomed.

The thing about the response to the Penner trade that made me so curious is that there was an instant assumption that we had to have lost the Penner trade. None of the arguments made in defence of that position really held up but all were presented with great rhetoric and tremendous emotion. It couldn’t have been more disconnected from the approach that needs to be taken by management when doing the same evaluation. It made me curious. You must surely wonder from time to time why the Oilers management never comes close to doing any of the things we think are smart. I get that you keep assuming it is because they are stupid but for me the Penner trade is the moment when I started asking myself how sure am I that they are stupid? If they aren’t, what else could be going on here?

I do get that there is a reasonable body of evidence to begin evaluating the performance of the Oilers since the lockout and it doesn’t look pretty. However, it lacks context and definitions. If we took the roster that the Oilers had in the last year pre-lockout and matched it up against the one they have today it should be much worse if they have had terrible asset management. Shouldn’t it? Seems a reasonable test in any case. However, how are we going to compare the assets and their value? Are we going to bring in to the mix issues of salary and salary cap, profits, ticket sales, merchandizing. As fans none of that probably matters much but I am sure it matters a lot to the organization. You are also spot on when you suggest there had to be a degree of interdependence between all those transactions. We just don’t know how to describe and quantify it, we may not even know how to qualify it. Never mind that we have no clue how much of it is luck and how much of it is skill. How much does management matter?

What I am trying to say is that while I agree with you that evidence on its surface appears to support your view that Oilers have terrible asset management skills we are a long ways from proving it is true and certainly shouldn’t be using it in evaluating their current behavior. Our assumed previous bad management is being widely used to prove the Penner trade was bad. However, we have no rigorous proof that the Oilers are that bad, nor can we assume the argument applies to Tambellini with his even smaller sample size. Don’t we need a separate analysis for him? Wouldn’t we then need context, how does he compare to the field of all other GMS, also to GMS with similar years on the job, etc.?

Its not simple at all, neither trying to understand how we rate trades nor how we should. The more I think about it, the more I read all the responses and counter responses to my post the the more certain I am becoming that irrationality and imperfect knowledge haunt both management’s trading behavior and how the trades are evaluated by fans and the media.

I get I can be difficult to argue with and I really appreciate time you and Derek have put into this.

Your relegation argument is actually a very good one and honestly I think we’d all be happier as fans if there was some relegation system. As you say there isn’t. Thus, I would counter by saying what was the potential future value of the pre-lockout roster. I think you’d agree we can get close on that based on current value. The trickier problem is to calculate future value of this roster which we have to do to avoid retroactive bias. Those would be the values you would need to compare. I have no idea how to do that.

Your argument for the Penner trade is sound. There are things you could add. That is things like possible improvements in future outcomes of current players from extra minutes of playing time. What I continue to be puzzled by is that in the light of all that you then appear to me anyway to be attacking the trade with what you admit is a flawed argument, albeit a compelling one.

Hi Scott,

Well I’d start off by asking if you have even tired applying any sort of statistical test of validity to the assumption we keep reading about 27% of picks in that range becoming actual NHLers? The sample size may be adequate and I could be full of crap. In which case game, set, and match.

I’d also think you could improve the argument against the trade by coming back to my point that Dustin Penner is a lot more than simply an actual NHL player. I tried using a metric of games played (okay I know it is flawed – look at our great in the locker room D-man) but you have to start somewhere. It is still a guess to project Penner but we can at least bracket the likely games he will play in his career. If you simply do that something like 6% of all picks are as good or better.

I then tried PPI and that really gives Dustin an edge – now only 3.5% seem to me to be as good or better. You can’t know that the actual numbers have predictive power but you can make the argument that Penner is better than average and that there are even fewer better than average players in the draft than average players and thus the chance of replacing Penner this way is smaller than it would be for a more average player. That then would allow you to argue Penner has a uniqueness of X and only n of all players in the history of the NHL have been higher than X. The entire population of players may be a large enough sample that it would pass statistical significance.

The other thing you might be able to do is go the other way. What would it cost us to replace Penner’s skill set in the open market. For that I’d probably look for a player with comparable PPI and another metric I like is shots per minute, I’d probably add in corsi (sorry Don). That should be a fairly simple calculation since most would be quickly eliminated. The argument by analogy is as cursed by the law of small numbers as the 27% rule but has the beauty of including more variables. If as you increase the number of variables the outcome of the argument remains the same then you are probably on the right track.

You might be able to get there even quicker. You could just take the players above and below him in goals at his position (or goals at any position) over the 5 years and see how expensive they are both in terms of salary (we do it in arbitrage work all the time) and in terms of the picks it would cost to get yourself one. Or go for a value metric like I mentioned above (PPI + shots per minutes + corsi) an keep moving up and down the lists of scorers who have done roughly as well as Penner. Find strongly comparable players and you can make a true value argument. Fail to find them and Penners very uniqueness begins to make the argument for you.

Those are some preliminary thoughts.

In general I think something is mroe compelling if we can understand it intuitively or emotionally, much math has an elegance and is intellectually compelling but doesn’t correspond with our common sense view of things and is not appealing emotionally.

Certainly each draft pick most have a different real world value and should descend from Teubert (viewed as a pick – since he was picked 13) then LAs first, then the conditional. In my average world you should be able to say the average value of a draft player should be the value of all drafted players (however you choose to define it) minus the number of picks in that years draft. Then you should be able to give each pick a draft value as a proportion of the chance of drafting an average player. If they scattered at random throughout the draft as they would be in my fictional world the value of each one is equal to the number of picks left in the draft divided by the number of picks in the total draft.

Lets keep the math simple and say that there are 300 draft picks, and the value of players is scattered randomly amongst them.

Then Teubert = 287/300*VAPD = .956 VAPD
Then 1st round = maybe say 278/300*VAPD = .927 VAPD
Then conditional pick = maybe say 270/300*VAPD = .90 VAPD

If VAPD must ultimately coorespond to VORP in an average NHL
Then 1.6 = .956 .5 + .927.5 + .90*.5 + X
1.6 = .478 + .4635 + .45 + X
X = .2085

Sorry, missed a step in the first post

VP = 1.6 VORP = Vp1/VORP + Vp2/VORP + Vp3/VORP + X
thus 1.6 = Vp1 +Vp2 + Vp3 +X
1.6 = 3VORP + X
1.6 = 1.5 + X
X = .1