The problem with this model is that you can't factor in things like the Raptors being a really bad playoff team. You can't just look at previous years either, because team members change, coaches change. Raptors with more or less the same team have consistently under performed in the playoffs.
Well I don't know how shoes' model works, but I would like to say a few things
Most importantly, it's a model. All models are going to have problems. Even for modeling simple things in life it's difficult to capture a model that fully describes a process. Models will have their strengths and weaknesses, and that's ok. A lot of times it would be too costly or unrealistic to try to capture every single dynamic in play, and there will always be lurking variables out there that most don't even know is affecting a process.
That goes beyond sports. My current job has me modeling investments. One assumption that a lot of models in the industry make is that investors are risk-neutral (they neither sick nor avoid risky assets).
Logically, that assumption sucks. People on average are risk averse(they avoid risk). If people were risk-neutral CDs at banks would be extinct, and the insurance industry would be doomed.
But it turns out that assumption doesn't really affect the modeling of a lot of different assets that are already purchased, only ones that have a provision to be 'returned' under certain circumstances.
Would the model be better if we could capture the risk aversion of investors? Of course. But each investor is different. And capturing just how risk averse each investor is is extremely costly, time consuming, and I would argue misleading because it depends on the input of the investors and you could get a survey bias. That simplification saves a lot of time and money, and it turns out doesn't come up as an issue as often as one would initially think.
Ok back to the basketball world. I don't know how Shoes model works, but he absolutely
can factor in the Raptors recent playoff history. But there are immediately a few problems:
- How much does past performance correlate into future performance?
- How do you quantify being 'really bad', and have 'really bad' teams in the past gone on to defy that label to win a championship. Moreover, does that even matter since these are different teams, different human beings entirely we are talking about.
- Like you said, different players, coaches also factor in. How does Demare Carroll affect the Raptors 'choke' factor? A hell of a lot less than LeBron James would. But how much less?
- How much of it has to do with youth? Do other teams with similar aged key contributors face such fates? And what age does this 'wear off' if it does?
Any time you add additional parameters to the model it becomes more complex. There are actually theorems out there about measuring if additional parameters are worth adding to a model. A trade off between complexity and accuracy.
Using the past seasons could also very easily lead to overfitting the data and getting a model that heavily favors those that have won in the past, and heavily punishes those who have lost early. Such a model would punish the Warriors last season because before that they were first round exits. Is such a model good, and the Warriors 'beat the odds?' I don't know, I don't model these things
But look how much complexity would be added, and questions that would arise from something as simple as a playoff adjustment factor.
Models are models. Simplifications of real life processes. Understanding their strengths and weaknesses is vital, and models can still perform extremely well without capturing every statistical relationship out there.