Causation and the Financial Crisis

Robert Waldmann

Matthew Yglesias discusses the financial crisis and the idea of causation. He shows that he was a philosophy major. I’m delighted. The question is whether the answer to the question “What caused the financial crisis?” is of the form “A, because if A had not been true, then there wouldn’t have been a financial crisis.” Yglesias notes potential problems with that approach. I critique after the jump.

OK jumpers, here I admit that, while I am delighted to have philosophy applied to finance, I think that among other things Yglesias made up some pseudo problems. He is commenting on an excerpt from a post by Barry Ritholtz

Understand that this is a theoretical discussion based on counter-factuals — what is likely to have occurred if various elements leading up to the crisis were different. We are trying to discern the differences between primary and secondary factors, separating the causes from the exacerbators.

Whenever someone asserts as a cause an event or force relative to a particular outcome, you should always ask: “Is this a “BUT FOR cause of that outcome?” In terms of a specific result or outcome, “But for” this factor, how would the outcome have changed? Would the result have been the same or different?

I will comment on the passage quoted by Yglesias (note I have not read the post).
That little passage is vulnerable to a serious practical critique. It relies a lot on the word “likely.” Now one thing we have learned from the crisis is that it is a bad mistake to look at the most likely outcome and assume it is certain. In practice, the sort of result we can get from trying to understand the causes of the crisis is that it probably wouldn’t have occurred without those causes. That’s not good enough for financial regulation.

More to the point, we can figure out that a crisis in 2007-8-9 would have been unlikely but for the following factors, but we can’t conclude that without those factors another crisis is unlikely to ever occur. Obviously different crises can have different causes. If the same factors exacerbate crises with different causes, it might be better to address the exacerbators and not the causes of the current crisis.

This is not the criticism Yglesias makes. I’m going to try to be relatively brief for now and focus on two issues. First he notes that but for causation and legal liability are different. He has examples in which people are, and should be, punished for damage they didn’t cause in the sense that it wouldn’t have happened but for their actions.

Then he writes “One is that when something very large-scale and bad happens, we probably have a few different interests. One is in punishing perpetrators and one is in preventing recurrence.” Huh ?!?! Matthew Yglesias and I are inclined to be consequentialists (in my case at least it means I was inclined to be a utilitarian — was exposed to convincing arguments against utilitarianism and watered down my beliefs to something vague enough to be unrefuted).

I at least absolutely believe that the only possible justification for a punishment is to prevent a recurrence. I definitely do not think that we have an legitimate interest in punishing perpetrators except to the extent that it prevents an occurrence. I had thought that Yglesias agreed with me on this. He certainly has repeatedly described himself as mostly consequentialist.

This implies that I oppose vengeance for the purpose of vengeance. That, in turn, doesn’t imply much in practice. I think one effective way to prevent a recurrence of the crisis would be to tar and feather 10% of bankers chosen at random as collective punishment. I do not advocate this course of action, because tarring and feathering is torture and because there are costs of having terrified bankers which are even worse than the occasional financial crisis due to excessive risk taking.

Then Yglesias comes up with an abstract example.

The second is that when something complicated happens, it’s plausible that you’ll see confusing webs of preemption that defy simple counterfactual analysis. Imagine that A, B, C, D, E, F, and G all happened. And suppose that to produce a crisis you need A, plus either B or C, plus any two of E, F, and G. If you try to do a “but-for” analysis on all seven factors, you’re going to reach the conclusion that “A caused the crisis” while B, C, D, E, F, and G all may have contributed in some way but didn’t actually cause it. But it’s not clear that that’s the right thing to say. And it’s definitely not clear that the correct policy response is to focus on A. Maybe A had lots of really beneficial effects, while B and C are both useless.

Here he is probably making a correct analysis of how attempted “but for” causal analysis will mislead us. However, the shorthand of referring to factors with letters is key. This is one case in which language influences causal analysis (in a way in which id doesn’t influence logical inference). Let’s say there is another language in which the same word is used with two English meanings B and C. In Italian the same word “fare” means “to make” or “to do” and the same word “forza” means “strength” and “force.” On the other hand there are two translations of “to be” “essere” and “stare” and of “to love”If they do the but for analysis they will conclude that both A and “B or C” are causes.

This means that we can make a mistake if we take the division of entities into categories to be either a known truth or a harmless convenience. Here Yglesias doesn’t say that understanding what caused something is unimportant. He notes that it is hard. I tend to share his belief that analysis of causation by English speaking lawyers and judges is invalid, because they take the English language as know to be right and true (unlike all other languages) or, more exactly, because they assume that the decision to reason in English is harmless, innocent and doesn’t matter.

Philosophers know that arguments which seem valid are proven to rest on unexamined conventions. They do this by inventing new words. I agree with Yglesias that one problem with legal efforts to find causation is that they rely to much on plain English and are too suspicious of neologisms and jargon.

I am not joking.

OK that was the punchline but we are after the jump so I go on.

In the simple case of A, B, C, D and E, where each can be true or false it is possible to construct the whole set of possibilities (there are 32) and say which imply a crisis. This means that the same long boring report is written whether we think that “A or B” is an elementary category and saying “A but not B” a pointless quibble or if we think that “A” and “B” are elementary categories and “A or B” is derived from them.

In the real world, it is absolutely impossible to get anywhere without risking having been lead there by the irrelevant history of the evolution of language. The innocent language in which every possible state of the world has it’s own name has an infinite vocabulary and no one can learn it. IIUC philosophers have long since given up on finding a language such that using that language can’t lead us astray.

Just to go on and on some of Yglesias’s other examples are interestingly irrelevant.

After the Ritholtz passage he continues

I’m broadly sympathetic to that account, but it’s worth emphasizing that there are a lot of complications. Consider this example from L.A. Paul and Ned Hall, “Causation and Preemption”:

Suzy and Billy, two friends, both throw rocks at a bottle. Suzy is quicker, and consequently it is her rock, and not Billy’s, that breaks the bottle. But Billy, though not as fast, is just as accurate: Had Suzy not thrown, or had her rock somehow been interrupted mid-flight, Billy’s rock would have broken the bottle moments later.

So you can’t say that the “but for Suzy’s toss, the bottle wouldn’t have broken.” But I think a normal person would want to say that Suzy broke the bottle. There’s a lot of work done on this and other philosophical topics. But I would note that in a politics/policy context it often just depends on what we care about.

Yes and, in this case, what we care about is preventing a recurrence so, for our purposes, Billy’s and Suzy’s actions are the same. They are like B and C in the abstract example.

In plain English, “Sally broke the bottle” is true and “Bill broke the bottle”. The example shows how we can’t completely describe the events usefully with two sentences of only 4 words each. This is not a surprise.

OK this is the example which caused me to write this post.

Imagine I’m in the elephant house at the zoo on a crowded day. A couple of elephants get loose and start stampeding through the crowd. I have a gun on me, and spot some dude I don’t like and decide to take advantage of the chaos by shooting him in the end. Eventually, it turns out that 80 percent of the people who were in the elephant house that day wind up dead as a result of the stampede. From a “but-for” perspective, it’s not clear that I actually caused the dude’s death. But from a legal perspective, it’s clear that I’m a murderer. The point of the statute is precisely to punish people who shoot other people in the head. But from a policy perspective, the bigger issue here is arguably elephant stampedes rather than guns at the zoo. At a minimum, elephant stampedes are killing more people.

Again huh ?!?!? I would say that the purpose of the murder statute is to prevent murder and the purpose of preventing murder is to keep people alive. To mean, punishing murderers is a means to an end not an end in itself. So I think the character named “I” should be jailed. We jail for 2 reasons and both have to do with intention and not “but for” causation. We want to deter murderers. Punishing someone for killing someone who would have died anyway but he didn’t know it works just as well. Deterrence depends on what people think will happen to them and people generally don’t kill other people if they know the other person is about to get stomped by an elephant (some bloodthirsty person might and there is no point in deterring him but no harm either so punishing people for killing people who are going to die anyway is just as useful as a deterrent (also we are all going to die sooner or later)).

Another purpose of jail is to incapacitate. We lock people up because the crime they have committed convinces us that they are likely to commit more crimes. In theory if we could determine who would commit crimes in the future consequentalism tells us we should lock them up before they have committed a crime. The rule that people are to be locked up only after they have committed a crime is needed, because the alternative is to grant some person or group arbitrary power to lock others up if they say they think the others will commit a crime.

Our forecast of future murder depends again on intent.

Now I will argue that I am really really fanatically consequentialist and have been for a long long time. When I was in 4th grade, I said that the prison sentence for attempted murder should be longer than the prison sentence for murder. My logic was based on incapacitation (I’m pretty sure I didn’t believe in deterrence at all then). The prediction of future murders depends on the intent I argued (as I argued above). Plus if the attempted murder is unsuccessful we know that there is someone that the defendant wants to kill. If the murder had been successful, maybe the murderer would be satisfied.

This is very consequentialist thinking. It totally freaked out my classmates and teacher.

Years later, I understood what was wrong with my reasoning. I was assuming that we knew for sure that an act was an attempted murder. In the real world, severe punishment for attempted murder would lead to locking up people whose aim was assault not murder. Indeed the word lead me astray. There are many degrees of intent from trying and seeing if it works to absolute determination to make it work. The difference between a murder and an attempted murder may be partly chance, but it can also be partly based on determination.

My proposal was like locking up people who will commit crimes but haven’t yet. Under absurd assumptions about what we know, it makes consequentialist sense, but in the real world, we can’t know that much or trust people who claim to know that much.