Hacker News Reader

Comment by DigitalJack | Hacker News Reader

I'm not discounting her complaints, but the following is not a confusion of cause and effect:

"Silver confuses cause and effect. We didn’t have a financial crisis because of a bad model or a few bad models. We had bad models because of a corrupt and criminally fraudulent financial system."

A= Financial Crisis B= Bad Models C= Fraudulent System.

Nate said "A<-B" Author says "A<-B<-C"

That is not a mix up of cause and effect.

Author's main complaint seems to be that Nate assumes bad models are an accident, and Author claims they were intentional.

Again, not a mixup of cause and effect. At worst it's a naive interpretation of the correct cause.

View on HN

There's some equivocation here I think:

"Bad Model" in Silver's view is really "bad modelling" -- a problem with the statistical technique (frequentist vs. bayesian) used to model the data.

"Bad Model" in O'Neil's view is really "bad data" -- intentionally skewed numbers as a consequent of conscious or sub-conscious eliding or redactory effects on the data populating the model, for purposes other than accuracy: financial gain, for instance.

O'Neil's point is that even a frequentist model would have been effective had the data not been massaged. On this view O'Neil's view is correct: the cause was not "bad modelling" but "bad data." The "bad data" in the model was an "effect" of corruption, not the root "cause" of the financial meltdown.

"O'Neil's point is that even a frequentist model would have been effective had the data not been massaged."

This post really has nothing to do with techniques or "bad data" or a bayesian vs. frequentist debate.

Her point is that bad modeling was entirely incidental and inconsequential as a cause of the financial meltdown. To repeat an example I used previously, if I smashed your head in with a hammer, Silver would essentially be saying "a hammer caused davesims death."

The statement may be technically true, but it's a shallow analysis of cause and effect. It's unsatisfying because it gives too much significance to an incidental link in the complete chain of events.

> This post really has nothing to do with techniques or "bad data" or a bayesian vs. frequentist debate.

I disagree, but maybe I'm missing something. This early quote, regarding 'bayesian vs. frequentist' seems to sum up O'Neil's view of Silver fairly well to me:

"What is not reasonable, however, is for Silver to claim to understand how the financial crisis was a result of a few inaccurate models, and how medical research need only switch from being frequentist to being Bayesian to become more accurate."

And later on regarding 'bad data,' a point she reiterates several times:

"In other words, it’s not that there are bad statistical approaches which lead to vastly over-reported statistically significant results and published papers (which could just as easily happen if the researchers were employing Bayesian techniques, by the way). It’s that there’s massive incentive to claim statistically significant findings, and not much push-back when that’s done erroneously, so the field never self-examines and improves their methodology. The bad models are a consequence of misaligned incentives."

I do think in the above quote O'Neil is equivocating between the idea of a 'bad model' and the skewed data or overreaching significance applied to the model.

Her fundamental point, which she could have done a better job communicating I think, is that it wasn't the models themselves that were bad, bayesian or otherwise, but the data in the models and at times the inordinate significance applied to those models that was the real cause of the meltdown.

These are particulars to the examples she provided. The unifying idea is that people often choose models or data based on self interest, even at the expense of accuracy. When discussing the cause of an event, it's not enough to say "a bad model is to blame" or "bad data is to blame." At this point, the model or data are instruments serving the self interest of the modeller, and they are incidental in the cause and effect relationship.

That's a good summary of the point I think O'Neil was trying to make contra Silver.

>On this view O'Neil's view is correct: the cause was not "bad modelling" but "bad data." The "bad data" in the model was an "effect" of corruption, not the root "cause" of the financial meltdown.

So it goes that corruption caused bad data, which caused a bad analysis by the model(she claims that both the data and model are bad), which caused the financial meltdown.

That does not invalidate Silver's point. It merely points out that Silver's analysis may be inadequate.

Silver's point can be technically valid, but ultimately shallow and irrelevant. That seems to be what she's arguing.

What I understand is that she means that corruption lead to the financial crisis. The bad models/data were just a smoke screen, or an instrument to get rich quicker.

She's actually saying that

A<-C AND B<-C

implying that B wasn't necessary for A as long as C.

Even so, still not a mix up of cause and effect. Wrong cause in this case.

I agree.

Also, I'll address the accusation that Nate is confusing cause and effect.

Nate's assertion is A<-B If he had confused cause and effect than that corrected form would be B<-A, or, that the financial crisis caused bad models.

So the Author's mistake is that (A<-C AND B<-C) != (B<-A), or, that her attack on Nate is inconsistent with what she believes is the correct logic.

The author isn't saying that A<-B is a confusion of cause and effect. She is saying that B is itself an effect of C, not a cause. (A better way to say it would be "B is not a root cause".)

I don't think the title is her thesis, possibly just link-bait for her usual readership -

This seems to be her thesis:

To be crystal clear: my big complaint about Silver is naivete, and to a lesser extent, authority-worship.

linkbait from a website called mathbabe.org? Surely you don't mean that?

Downvoted.

Having a domain name that is short and rememberable != linkbait.

Surely none of her audience expect to see pics of babes on mathbabe.com

The bad models were a consequence of a fraudulent system. So by definition, the models weren't the cause of the financial crisis. Therefore the bad models were an effect, in the same way the financial crisis was an effect.

If this is true, then the author's claim (mix-up of cause and effect) must be correct.

But it's not.

She states:

>Rather, the entire industry crucially depended on the false models. Indeed they changed the data to conform with the models, which is to say it was an intentional combination of using flawed models and using irrelevant historical data (see points 64-69 here for more).

So C(corruption) directly leads to B(bad model). She does not argue that A(everything falling apart)<-C anywhere in the piece. Silver is claiming that A<-B. He is not commenting about C. So while Silver's claim may not identify the absolute root cause, the author does not actually prove that his cause and effect analysis is flawed. If anything, she has shown that it may be incomplete.

Silver isn't just claming that A<-B. He is claiming that fixing B will fix A, and that fixing B is straightforward: just use good models.

mathbabe is saying that fixing B is not straightforward, because B<-C, and fixing C is very, very difficult. Bad models exist because people have incentives to make them bad.

She's also saying, I think, that fixing B by itself won't fix A, because B is not the root cause; C is. Fixing B without fixing C just means that C will manifest itself somewhere else, and A will still happen. In other words, there are many causal routes from C to A, and fixing B by itself only blocks one of them. So A<-C is true regardless of the state of B.

But it's not [true].

What, specifically, isn't true?

So C(corruption) directly leads to B(bad model). She does not argue that A(everything falling apart)<-C anywhere in the piece.

The point of the piece is that the financial meltdown stemmed from corruption, not from models. If this wasn't her point, then why else would she have written this piece?

She wrote this piece because she thinks he's only citing the proximate cause and not the ultimate cause. At least that's how it reads to me.

She doesn't appear to be disputing the idea that following bad models caused the crisis — instead, she just seems to be saying Silver missed the fact that the models were deliberately bad rather than simply a poor application of math.

A few quotes from the author herself:

"the entire industry crucially depended on the false models."

"Silver gives four examples what he considers to be failed models at the end of his first chapter, all related to economics and finance. But each example is actually a success (for the insiders) if you look at a slightly larger picture and understand the incentives inside the system."

She's not disputing that the models were false or even that acting in accordance with the models is what caused the crisis. She's disputing that this was accidental.

What, specifically, isn't true?

An argument doesn't have to contain a false statement to have no baring on the discussion.

And more generally it's quite possible for a thing to have multiple causes, and in this universe this is generally true for everything we witness. So if A causes C that doesn't mean that B can't also cause C "by definition". I'm not aware of any serious philosphical framework that says that events must have only a single cause, though many philosophers from Aristotle onwards[1] but great stock in the "final cause" of things, but that would be either "God" or "The Big Bang" depending on your religion or lack thereof and nobody was arguing for those.

And as an aside, the words "by definition" in an argument that isn't on it's surface about definitions is generally a bit of a red flag.[2]

[1]http://en.wikipedia.org/wiki/Teleology [2]http://lesswrong.com/lw/nz/arguing_by_definition/

She does not prove that claim, though. She argues that corruption causes bad models. For Silver's cause and effect analysis to be incorrect, she needed to prove that the inaccurate models had absolutely nothing to do with the financial meltdown.

The models can be part of the meltdown without really being a cause.

If I choose to shoot you with a gun, and have a variety of appropriate guns, it is not the fault of the specific gun chosen that it was used to kill you. I had motive, opportunity, and alternate means available.

"For Silver's cause and effect analysis to be incorrect, she needed to prove that the inaccurate models had absolutely nothing to do with the financial meltdown."

If I smashed your head in with a hammer, would people claim the hammer was the cause of your death? Is "hammer caused death" the end of the story, or even particularly important to the related series of causes and effects?

She's arguing that it's not, that Silver's cause and effect analysis is irrelevant because his purported cause is purely incidental. Intentionally inaccurate models were simply an instrument designed to further the self-interest of individuals.

(And further, she doesn't need to provide a "proof" to make this claim. But she does offer a compelling argument.)

A is the "financial crisis", not "everything falling apart".

Everything falling apart would be hyper-inflation in addition to capital control and gold confiscation : )

That could be 'D'.

Latter or sooner we'll learn if D <- A ; )

The easiest way to understand this is that even if the models were accurate, the financial crisis would still have occurred. So blaming the models would be confusing cause and effect regardless of whether you think the models were wrong intentionally or unintentionally.

That said, I didn't read the book so I have nothing to say about whether or not Nate Silver actually does this.

The models certainly contributed. (Note: I have read this far in the book.)

The modeling error in question was independence; that is, if you have five mortgages, each with a 5% change of default, then these can be packaged up as an AAA security as follows: you only lose your money if all five default. A bit riskier package is that you lose if 4/5 default. And so on, each with different returns.

If they are independent, the p(default) = 1/20^5. If they are dependent, it is 1/20. Now multiple mortgage pool size by a 100 or 1,000 or 10,000 (?) and see how far off the estimated risk is. :)

Now combine this with a 30-to-1 leverage when buying these "AAA" securities.

(This was quite a good problem to work through with my daughter to see what that little "independence" assumption means. :)

His main point here was that modeling failures are typically due to out-of-sample conditions; when the housing bubble broke, the markets were might more tightly coupled across the country than the modelers assumed. While they could have seen this kind of dependence if they looked to Japan, there was no such precedence in the US in recent history.

This is a beginner mistake which I have a hard time believing the entire mortgage market failed to spot.

My personal belief is that they knew of the flaw of the models but did not care since their personal incentives were more profitable if the model was not fixed.

As Michael Lewis found in the Big Short. His Liar's Poker (from the 80's) explains the mentality. As he says in the former, he thought the latter would be the be-all end-all.

But the kids in Business School would constantly write him and ask him for new ways to rig the game.

In order for this to be true, she needed to make the argument that it is. She's not saying anything to that effect at all.

If her point wasn't "corruption caused the financial meltdown," then why did she write this piece?

The simplest explanation is that she was, in fact, arguing that corruption caused the financial crisis.

You're imposing a cause/effect dichotomy that doesn't actually exist. Most causes are effects of something else. When we talk about mixing up cause and effect, we mean that A is said to have caused B when in actuality A was caused by B. Showing that A was caused by C does not mean that we've mixed up cause and effect, because nobody said A caused C.

"The bad models were a consequence of a fraudulent system. So by definition, the models weren't the cause of the financial crisis. Therefore the bad models were an effect, in the same way the financial crisis was an effect."

This is faulty reasoning. A pool ball may go into a pocket because it was hit by a cue ball. In this case the cue ball hitting the other ball is the cause of the other ball falling into the pocket. Yes, you can look back into the chain of causes to find one further back, e.g., the fact that a person used a pool cue to hit the cue ball into the other ball. This doesn't mean that the first cause you found was not a cause, and certainly doesn't mean that "by definition".

If a corrupt and fraudulent system could have brought about the financial crisis by some means other than causing bad models (which I assume is true), then you can say that the bad models were not a necessary part of the causal chain. But you can't say that they weren't part of the causal chain at all. As a matter of fact (at least according to assumptions everyone is making in this topic's thread) they were.

I'm not sure you can get to the financial crisis without faulty models. Risk models, especially on packaged securities, are crucial to setting ratings as a proxy for risk. Those ratings are key to determining pricing, maximum leverage, and which portfolios can buy them.

If the models had been more accurate - for example, if they had more properly reflected the way that the value of a bundle of mortgages will suddenly lose value as housing prices decline - banks would not have been able to pile up such huge risks and high leverage.

You can only justify things like 30X leverage if you believe the securities in question have essentially no risk - because at that leverage, a 3% loss in value means you're wiped out.

I'm not sure you can get to the financial crisis without faulty models.

Sure you can. The root cause of the crisis was people gaming the system. There are many ways to game the system; faulty models are only one of them.

Okay, but all routes to that gaming - at least at the institutional level - ran through being able to do it without it impacting credit ratings. Credit ratings and security pricing are very dependent on statistical models for risk, in particular correlation and downside outcomes.

all routes to that gaming

All routes to that particular method of gaming. There are other methods.

Exactly. And she finds this mistake so dire she claims it is "malicious" without backing that up at all. Perhaps that was just hyperbole.

I don't think the author is proposing an "A<-B<-C" system. She is stating that the models are really blameless in the situation when the information provided to the models are wrong. The cause and correlation issue arises there.

Silver assumes that the systems fail because the models are bad. O'Neil is instead claiming those are just correlations and not a cause and effect relationship. Basically the models are bad and the systems failed because the people providing the data were corrupt. Using your example: "A<-C" and "B<-C"

It's known as proximate and ultimate causation. [0] An Aristotelian analysis would say the fraudulent system or its corruption was the final cause, the models were the formal cause.

[0] http://en.wikipedia.org/wiki/Proximate_and_ultimate_causatio...

It's actually both. Bad models and bad incentives, from black swans and the models that ignored them to corrupt practices and liars.

Reply on news.ycombinator.com