Seeing how that data for those years has yet to be recorded, he cannot possibly compare the forecasted values to the actual values to test the accuracy.
Of course he can. His data comes from a model, so he can produce data for any 'years' he wants to.
It's entirely relevant to the accuracy of the model. You cannot use fictional data to try to understand the causes and variation in a real world variable.
The article doesn't address understanding causes and variation of a real world variable, it address the ability of a model to track the underlying model it was built from. Whether the underlying model was reality or just another model, if anything, ought to make the task easier.
I called out his "calibration" as bullshit.
Your work, as you've described it, is no different than what the article claims does not work.
Care to tell the class what you do for a living and your educational background?
Masters in Computer Science--not that my credentials are relevant to the strength or weakness of my argument.
We've already established that I have reason (both educationally and career-wise) to know WTF I'm talking about
First, your education and career may indicate that you ought to know what you're talking about, but the arguments you put forth show that you don't.
Second, this is perhaps the crux of matter addressed by the article. You, and I assume many other economists, think you know what you're doing, but you don't. You're practicing cargo cult science--going through the motions of statistical analysis without understanding what you're doing or why, and consequently producing garbage models that don't predict anything other than the 10 years of data you tested against.
He had two models. The first model produced hypothetical historical data (analogously, the data from 1950 to 2010). He then created a second model, built on part of the 'historical data' (1950-1999) and tested on the remainder (2000-2010). He then used the first model to produce another segment of data (2011-2020), and found that the second model did not predict this 'new' data at all.
You and he are doing the same thing; the fact that your 'historical' data comes from reality and his comes from a model is irrelevant so far as the second model is concerned. The phenomenon described by the article is known as 'over fitting the training data' in Machine Learning circles, and is widely known in other fields, according to the comments. Perhaps if you pursued a degree in something a little less soft than 'Applied Economics' you might understand the Math you're blindly applying.
If you've done a good job collecting a large enough data set and including the necessary variables, you'll have some pretty damn good predictions for the first part of your time series.
Did you even read the article? Allow me to quote the relevant portion:
The problem, of course, is that while these different versions of the model might all match the historical data, they would in general generate different predictions going forward
Regardless of your spurious claims about 'not calibrating models', the point remains that in any complex system multiple different models can be found that fit the small slice of time you use for accuracy testing, without fitting the unseen future data.
The great unwashed masses seem to be blissfully unaware that there is no such thing as 'chemical free chicken' or what ever marketing spiel the ad men decide to pitch to you.
This may come as a surprise to you, but not all words have a single canonical definition, nor are the multiple definitions they do have necessarily clear cut. Allow me to demonstrate, using the Google-Dictionary definition of chemical:
A compound or substance that has been purified or prepared, esp. artificially
As you can see, the term chemical implies, at least according to this one source, artificiality, without requiring it. Therefore, the obvious implication of saying that organic foods don't contain chemicals is that they don't contain artificial, unnatural, or extra chemicals (the credibility of such a claim is another matter, of course). This is the kind of mistake a computer would make; ignoring context in favour of some obscure interpretation that leads to a ridiculous conclusion. Or a troll, now that I think about it
The key difference is that for Banks, the missing $9 was loaned out to other people, and there is therefore an asset (the outstanding loan) to match the debt (the money missing from the account). For Full Tilt Poker, the missing money was used to pay the board of directors, so it's gone.
From the article:
The funds in [the player's] accounts were supposed to remain untouched and available for withdrawal at any time, according to the complaint. But that wasn't the case.
Instead, Full Tilt Poker didn’t have enough in those accounts to repay players, and, moreover, used whatever funds it could collect to pay board members and other company executives more than $440 million since April 2007.
So, being a coward is illegal in Britain? Sounds like a slippery slope to me.
This is a laughable straw man argument.
I sure did poke holes in the prosecution, for all you know.
As per other posts in this thread, Duffy did confess. So I know very well that you did not poke holes in the prosecution.
Yes, the defense was clearly not successful. As I said, IF THEY WERE COMPETENT.
This is a textbook True Scotsman fallacy.
The system was down for backups from 5am to 10am last Saturday.
