This opinion by Raghu Rajan places the blame on skill biased technological change and increased inequality leading to a policy to increase homeownership among low and moderate income as a way to close this gap as the root cause of the crisis. This is a merely a deft way of blaming Fannie and Freddie and the CRA (Community Reinvestment Act).
The CRA and Freddie and Fannie has its passionate (?) defenders, among them:
1. Menzie Chinn,
2. Paul Krugman,
3. Mark Thoma
BTW, Jim Hamilton thinks that GSEs did play a role, albeit not the principal role.
I am thinking of playing a game:
These are as many factors that I could think of that can be the cause of the financial crisis (in no particular order, and I'll update the list as I think of more):
1. Greed (aka rational utility maximizing investment bankers and investors)
2. Lack of regulation and oversight
3. Complexity of securities
4. Media spreading rumors
5. Eliot Spitzer - his removal of Greenberg from AIG coincided with AIG insuring CDS
6. Dr Doom aka Nouriel Roubini, Meredith Whitney, or any or all bearish financial analysts, bloggers and commentators
7. Monetary policy - interest rates were too long for too long and then started rising too quickly triggering defaults
8. Herd behavior among bankers and investors
9. Over-optimisim among bankers and investors
10. Fannie Mae/Freddie Mac and policy to increase homeownership aka Community Reinvestment Act
11. Interconnectedness of financial system
12. Rating agencies
13. Short sellers
14. Securitization
15. Mortgage fraud
16. Inequality caused by skill biased technical change (SBTC)
17. Globalization/free trade possibly causing SBTC
18. And just for the heck of it, global warming.
19. Animal spirits.
20. Financial models & over-reliance on models
21. Moore's law and increase in computing power
22. Mark to market
23. Too big to fail/Moral hazard
24. Pay for performance and the economists who designed lavish pay packages of investment bankers.
25. Leverage
26. Proprietary trading
The name of the game is "causality jenga". We have these factors or building blocks of the crisis on one side and then there is the crisis on the other and for this purpose we can just define crisis as the events of the fall of 2007 beginning with the failure of Bear Stearns. If we start pulling out these building blocks (factors) we would ask ourselves, would the event still happen?
And as we do this, I think analysts need to ask themselves the chain of events that led to the crisis. In other words, a mechanism. Rajan has laid one out above which I think is incomplete. In the parlance of some other social sciences, we need to think clearly about what factors are 'primal' and what others are 'mediators'. If the chain can be broken somewhere could the crisis have been averted?
Or perhaps we can start by identifying necessary and sufficinet conditions. Or perhaps this is asking too much of economists.
Thursday, July 15, 2010
Econometric Models
This post really annoyed me (emphasis mine):
The ARRA, the fiscal stimulus act passed last year, gave the Council of Economic Advisers an impossible job: measuring how many jobs the act created. Here is the CEA's latest attempt. As far as I can tell, there are two kinds of evidence here.
First, there are model simulations. That is, the CEA took a conventional Keynesian-style macroeconomic model and used those set of equations to estimate the effect the stimulus should have had. Essentially, the model offers an estimate of the policy's effect, conditional on the model being a correct description of the world. But notice that this exercise is not really a measurement based on what actually occurred. Rather, the exercise is premised on the belief that the model is true, so no matter how bad the economy got, the inference is that it would have been even worse without the stimulus. Why? Because that is what the model says. The validity of the model itself is never questioned. (Moreover, the fact that other organizations simulating similar models come to similar conclusions is no evidence about the validity of the model's simulations. It only tells you the CEA staff did not commit egregious programming errors when running their computer simulations.)
This criticism is vacuous in the following sense:
1. It is valid not just for this model, but any econometric model.
2. It is valid also for any DSGE models that assume exogenous 'deep' parameters.
3. It is essentially valid for any econometric model that uses instrumental variables or 'fancy' causal type analysis. (Can any one say valid instruments?)
4. Any model is only as good as its assumptions.
The following is just an off-the-cuff remark and I'll probably get into trouble for this: No one really is convinced by any econometric or statistical evidence especially if it is a one-off study. We are only convinced if we already believe in the first place and in this case, the author has already discounted the effects of ARRA (before any evidence was even presented) and therefore dismisses this evidence as unconvincing.
Even if I were agnostic about the effects of the stimulus, this one study would really do nothing to convince me that it is effective. Different models, using different assumptions (robustness is what some may call it) really is the key to trying to persuade. After all, models (econometric or otherwise) really are rhetoric in disguise.
The ARRA, the fiscal stimulus act passed last year, gave the Council of Economic Advisers an impossible job: measuring how many jobs the act created. Here is the CEA's latest attempt. As far as I can tell, there are two kinds of evidence here.
First, there are model simulations. That is, the CEA took a conventional Keynesian-style macroeconomic model and used those set of equations to estimate the effect the stimulus should have had. Essentially, the model offers an estimate of the policy's effect, conditional on the model being a correct description of the world. But notice that this exercise is not really a measurement based on what actually occurred. Rather, the exercise is premised on the belief that the model is true, so no matter how bad the economy got, the inference is that it would have been even worse without the stimulus. Why? Because that is what the model says. The validity of the model itself is never questioned. (Moreover, the fact that other organizations simulating similar models come to similar conclusions is no evidence about the validity of the model's simulations. It only tells you the CEA staff did not commit egregious programming errors when running their computer simulations.)
This criticism is vacuous in the following sense:
1. It is valid not just for this model, but any econometric model.
2. It is valid also for any DSGE models that assume exogenous 'deep' parameters.
3. It is essentially valid for any econometric model that uses instrumental variables or 'fancy' causal type analysis. (Can any one say valid instruments?)
4. Any model is only as good as its assumptions.
The following is just an off-the-cuff remark and I'll probably get into trouble for this: No one really is convinced by any econometric or statistical evidence especially if it is a one-off study. We are only convinced if we already believe in the first place and in this case, the author has already discounted the effects of ARRA (before any evidence was even presented) and therefore dismisses this evidence as unconvincing.
Even if I were agnostic about the effects of the stimulus, this one study would really do nothing to convince me that it is effective. Different models, using different assumptions (robustness is what some may call it) really is the key to trying to persuade. After all, models (econometric or otherwise) really are rhetoric in disguise.
Wednesday, July 14, 2010
The mathematization of medicine
The wisdom of this in the field of economics has been debated and is perhaps also debatable. But what about medicine? I am not in a position to judge how useful nor am I in a position to say how widespread and popular it is but I came across the following
From Cardiovascular Mathematics by Luca Formaggia, Alfio Quarteroni, and Alessandro Veneziani (Eds.). The following is from Chapter 2
From Cardiovascular Mathematics by Luca Formaggia, Alfio Quarteroni, and Alessandro Veneziani (Eds.). The following is from Chapter 2
Some gems from Robert Hall
This interview has been blogged about elsewhere but I wanted to save the bits that I found interesting:
[on automatic recession dating]:
Actually, long ago, in the 1980s, we sponsored a project that informally, unofficially put out a recession probability index that Jim Stock and Mark Watson prepared. It didn’t work very well in the 1991 recession, so they stopped doing it after that.
And it didn’t work for fairly typical reasons. That was the first recession that wasn’t accompanied by a decline in productivity, so it looked somewhat different. So their historical relationships weren’t as stable as they hoped.
That’s one of the main reasons why automatic rules haven’t worked. People have done research on the machine approach for years. In fact, when I was a graduate student and took a computer science course, my project was to write software that would automate this. So it’s not a new idea. But it’s never worked very well.
Region: It would have missed the 1981 recession if we’d used the two negative GDP quarters rule.
Hall: You mean 1980.
Region: Right, 1980.
Hall: 1981 was no problem. The 1980 recession was just one quarter. And people have said that the 1980 recession was actually just sort of a prelude to the ’81 recession. We say no, but it’s been said.
Region: It seems it’s more of an art than a science then. Hall: It’s a classification problem that the world seems to want an answer to, but it has a shifting structure, and dealing with the shifting structure is the issue. We try very hard to achieve historical continuity.
We don’t doubt for a second—and I don’t think anyone else does either—that we know when there’s a recession. In all the data we look at, certainly in the period when we’ve had reliable data, which is since World War II, there’s never been an episode that’s somewhere halfway between a recession and a nonrecession. Every recession has been clear. And they all see unemployment shoot up and typically see GDP decline.
We do face issues though. With the most recent revisions of GDP, the 2001 recession essentially doesn’t exist. It was a flattening, but as emphasized on our Web site, there are issues of depth, duration and dispersion, but there was neither depth nor duration in what happened in ’01. By the alternative measure of total output, real gross national income, the 2001 recession is quite apparent.To me, it’s not an issue because that’s just looking at GDP. If we look at employment, as I did in a 2007 Brookings paper on the “Modern Recession,”—by “modern recession” I mean one in which productivity rises…
[on the state of macroeconomics]
Region: The past few years seem to have brought about a crisis of confidence in the economics profession, with critics suggesting that macroeconomics has failed in some fundamental way. It’s a topic addressed by [Minneapolis Fed President] Narayana Kocherlakota in our Annual Report this year. Do you agree that the macro profession failed the nation during the financial crisis?
Hall: I don’t. There are two parts to the issue. First, did macroeconomists fail to understand that a highly levered financial system based in large part on real-estate debt was vulnerable to a decline in real-estate prices? No way. Many of us pointed out the danger of thinly capitalized banks. We had enthusiastically backed the idea of prompt corrective action in bank regulation, so that banks would be recapitalized well before they became dangerously close to collapse. We watched in frustration as the regulators failed to take that action, even though they had promised they would.
Second, did macroeconomists fail to understand that financial collapse would result in deep recession? Not at all. A complete analysis of that exact issue appears in an extremely well-known and respected chapter in the Handbook of Macroeconomics in 1999, written by Ben Bernanke, Mark Gertler and Simon Gilchrist. Depletion of the capital of financial institutions raises financial frictions to levels that distinctly impede economic activity. In particular, credit-dependent spending on plant, equipment, inventories, housing and consumer durables collapses. That chapter is an excellent guide to the depth of the current recession.
I would have liked him to have answered the question: How should the government response? Although he does talk a little about fiscal policy in the beginning of the interview, I did not think that he adequately responded to how the fiscal stimulus should have been structured (he says in the interview that too little of it went into increasing aggregate demand directly) and whether monetary policy would have been prefered i.e. he did not answer Kocherlakota's claim that economists were not able to provide a play book to respond to the crisis.
[on automatic recession dating]:
Actually, long ago, in the 1980s, we sponsored a project that informally, unofficially put out a recession probability index that Jim Stock and Mark Watson prepared. It didn’t work very well in the 1991 recession, so they stopped doing it after that.
And it didn’t work for fairly typical reasons. That was the first recession that wasn’t accompanied by a decline in productivity, so it looked somewhat different. So their historical relationships weren’t as stable as they hoped.
That’s one of the main reasons why automatic rules haven’t worked. People have done research on the machine approach for years. In fact, when I was a graduate student and took a computer science course, my project was to write software that would automate this. So it’s not a new idea. But it’s never worked very well.
Region: It would have missed the 1981 recession if we’d used the two negative GDP quarters rule.
Hall: You mean 1980.
Region: Right, 1980.
Hall: 1981 was no problem. The 1980 recession was just one quarter. And people have said that the 1980 recession was actually just sort of a prelude to the ’81 recession. We say no, but it’s been said.
Region: It seems it’s more of an art than a science then. Hall: It’s a classification problem that the world seems to want an answer to, but it has a shifting structure, and dealing with the shifting structure is the issue. We try very hard to achieve historical continuity.
We don’t doubt for a second—and I don’t think anyone else does either—that we know when there’s a recession. In all the data we look at, certainly in the period when we’ve had reliable data, which is since World War II, there’s never been an episode that’s somewhere halfway between a recession and a nonrecession. Every recession has been clear. And they all see unemployment shoot up and typically see GDP decline.
We do face issues though. With the most recent revisions of GDP, the 2001 recession essentially doesn’t exist. It was a flattening, but as emphasized on our Web site, there are issues of depth, duration and dispersion, but there was neither depth nor duration in what happened in ’01. By the alternative measure of total output, real gross national income, the 2001 recession is quite apparent.To me, it’s not an issue because that’s just looking at GDP. If we look at employment, as I did in a 2007 Brookings paper on the “Modern Recession,”—by “modern recession” I mean one in which productivity rises…
[on the state of macroeconomics]
Region: The past few years seem to have brought about a crisis of confidence in the economics profession, with critics suggesting that macroeconomics has failed in some fundamental way. It’s a topic addressed by [Minneapolis Fed President] Narayana Kocherlakota in our Annual Report this year. Do you agree that the macro profession failed the nation during the financial crisis?
Hall: I don’t. There are two parts to the issue. First, did macroeconomists fail to understand that a highly levered financial system based in large part on real-estate debt was vulnerable to a decline in real-estate prices? No way. Many of us pointed out the danger of thinly capitalized banks. We had enthusiastically backed the idea of prompt corrective action in bank regulation, so that banks would be recapitalized well before they became dangerously close to collapse. We watched in frustration as the regulators failed to take that action, even though they had promised they would.
Second, did macroeconomists fail to understand that financial collapse would result in deep recession? Not at all. A complete analysis of that exact issue appears in an extremely well-known and respected chapter in the Handbook of Macroeconomics in 1999, written by Ben Bernanke, Mark Gertler and Simon Gilchrist. Depletion of the capital of financial institutions raises financial frictions to levels that distinctly impede economic activity. In particular, credit-dependent spending on plant, equipment, inventories, housing and consumer durables collapses. That chapter is an excellent guide to the depth of the current recession.
I would have liked him to have answered the question: How should the government response? Although he does talk a little about fiscal policy in the beginning of the interview, I did not think that he adequately responded to how the fiscal stimulus should have been structured (he says in the interview that too little of it went into increasing aggregate demand directly) and whether monetary policy would have been prefered i.e. he did not answer Kocherlakota's claim that economists were not able to provide a play book to respond to the crisis.
When is aggregate demand high
This question puzzled me: If aggregate demand is so low, why are profits so high?
If I remember what's left of my economic knowledge correctly, aggregate demand is simply C+I+G (in a closed economy) and this in turn equals Y or GDP.
So is Tyler asking why are profits so high when output is so low? This doesn't really make sense and I decided that it has to be the output gap that is high (that makes aggregate demand low).
From FRED the picture for corporate profits:
And from FRBSF a picture of the output gap (I wish these folks would make their data available for download):
The questions I would have are (and bearing in mind that output gap measures can be vastly different, e.g. see here for example):
1. Is the cyclical component of profits pro- or counter-cyclical to the output gap?
2. Same question vis-a-vis unemployment and profits.
If I remember what's left of my economic knowledge correctly, aggregate demand is simply C+I+G (in a closed economy) and this in turn equals Y or GDP.
So is Tyler asking why are profits so high when output is so low? This doesn't really make sense and I decided that it has to be the output gap that is high (that makes aggregate demand low).
From FRED the picture for corporate profits:
And from FRBSF a picture of the output gap (I wish these folks would make their data available for download):
The questions I would have are (and bearing in mind that output gap measures can be vastly different, e.g. see here for example):
1. Is the cyclical component of profits pro- or counter-cyclical to the output gap?
2. Same question vis-a-vis unemployment and profits.
Sticky expectations or just cheap
In an old post I thought that perhaps I had sticky expectations or that I was just cheap. But Nick Rowe explains that it is possible that I am not alone - that via comments by Richard Serlin, that consumers expect prices to be stable.
One of the arguments against sticky prices is that there are no microfoundations for this. Yet economists make all kinds of unrealistic assumptions all the time - sometimes with very little foundation - that agents are rational and utility maximizing or that preferences are Cobb Douglas or when it suits them, preferences are non-separable. Sometimes technology is Cobb-Douglas and sometimes it is CES again depending on which paper you are reading. These assumptions are all over the place and there is really no one set of assumptions that are made for all papers.
The argument for sticky prices posted above is that consumers do not like prices that change frequently. One of the hard things about economic modeling especially in dynamic models is the real world equivalent of a "period". In a model with 50 periods and prices that change every period can be considered "too frequent". The real world equivalent is usually a "quarter" or perhaps even a "month" - so are price changes every quarter really too frequent? Shoe prices that fluctuate every month I would consider to be too frequent, but if the price changes every quarter I can possibly deal with that.
One month ago, the price of a case of 18-pack Horizon Organic milk was $13.49 at our Giant grocery store nearby. Two weeks ago it was $13.99 and a few days ago it was $14.49. Is this too frequent? Again, it depends on the good as some papers have shown. (See yesterday's post and the links therein for references.) It also depends on whether these price changes will stick - i.e. is it worth my time to drive 10 or 20 minutes somewhere else to get it cheaper (or to find out that it is the same price!) Morever, where firms have the leeway to substitute for cheaper inputs or smaller portions (e.g. restaurants) they will do so.
To summarize, prices are sticky and models that explore price stickiness are a part of the development of economic knowledge. How important is this stickiness I am undecided. I am also unconviced that there is sufficient work done on the frequency of price changes and the possibility of substitutions - and I think it is because the data is not easy to come by. It is also not clear how consumers respond to price changes of various goods. In other words, a research agenda and hopefully enough grants to sustain a whole generation of economists.
One of the arguments against sticky prices is that there are no microfoundations for this. Yet economists make all kinds of unrealistic assumptions all the time - sometimes with very little foundation - that agents are rational and utility maximizing or that preferences are Cobb Douglas or when it suits them, preferences are non-separable. Sometimes technology is Cobb-Douglas and sometimes it is CES again depending on which paper you are reading. These assumptions are all over the place and there is really no one set of assumptions that are made for all papers.
The argument for sticky prices posted above is that consumers do not like prices that change frequently. One of the hard things about economic modeling especially in dynamic models is the real world equivalent of a "period". In a model with 50 periods and prices that change every period can be considered "too frequent". The real world equivalent is usually a "quarter" or perhaps even a "month" - so are price changes every quarter really too frequent? Shoe prices that fluctuate every month I would consider to be too frequent, but if the price changes every quarter I can possibly deal with that.
One month ago, the price of a case of 18-pack Horizon Organic milk was $13.49 at our Giant grocery store nearby. Two weeks ago it was $13.99 and a few days ago it was $14.49. Is this too frequent? Again, it depends on the good as some papers have shown. (See yesterday's post and the links therein for references.) It also depends on whether these price changes will stick - i.e. is it worth my time to drive 10 or 20 minutes somewhere else to get it cheaper (or to find out that it is the same price!) Morever, where firms have the leeway to substitute for cheaper inputs or smaller portions (e.g. restaurants) they will do so.
To summarize, prices are sticky and models that explore price stickiness are a part of the development of economic knowledge. How important is this stickiness I am undecided. I am also unconviced that there is sufficient work done on the frequency of price changes and the possibility of substitutions - and I think it is because the data is not easy to come by. It is also not clear how consumers respond to price changes of various goods. In other words, a research agenda and hopefully enough grants to sustain a whole generation of economists.
Tuesday, July 13, 2010
If an old model is wrong, does a new model make it right?
I enjoyed this article:
... We almost never directly observe what is going on beneath the surface of 70 percent of the planet, and yet US fishing rules and regulations demand that scientists predict how many fish are in a given sea.
So scientists and fishermen and everyone else rely on computer models that mimic what is known about fish. Into the models goes information like size, age, growth rate, how many fish will die of natural mortality (predation, disease, moving away from the area) and how many are taken in the fishery.
Lobsters lack body parts like ear bones that help to reveal the age of other species; as a result, modelers use lobster size instead. But even size can be misleading, because lobster growth rates also vary with temperature: the warmer the water, the faster they grow. And since most lobsters move around throughout the year and over the course of their lives, their growth rate does not stay the same.
Variable growth rates can have surprising effects. According to Dr. Yong Chen, a fisheries scientist at the School of Marine Sciences at the University of Maine, the lobsters that grow into the size that can be legally harvested during any given year can include individuals born over a seven-year time span. In other words, the lobster on your plate could be four years old and the one on your friend's plate could be eleven years old, even if they are both one-pound lobsters that were caught in the same trap.
These facts complicate computer models. Throughout the 1990s and early 2000s, the National Marine Fisheries Service model consistently underestimated the number of lobsters in the sea, and therefore overestimated the percentage being caught in the fishery each year, leading federal scientists to believe that overfishing was occurring in the Maine lobster industry. Yet year after year, catches went up and research surveys recorded higher and higher numbers of lobsters. Clearly the model wasn't working.
Fast forward to 2008: The Atlantic States Marine Fisheries Commission and the National Marine Fisheries Service officially adopted a completely new model that estimated the lobster population as "not overfished."
This transformation occurred largely thanks to the talent and tenacity of Dr. Yong Chen.
According to Chen, there are four main areas where his model improved on the prior version. "We included the inshore trawl data from Maine and other state surveys, in addition to federal survey data; we had better catch data to work with than before; we had more realistic biology built into our virtual lobsters; and we used a statistical approach that incorporates margins of error in our inputs (this approach uses Bayesian statistics)," he said.
But how do we know that the new model is right versus less wrong?
... We almost never directly observe what is going on beneath the surface of 70 percent of the planet, and yet US fishing rules and regulations demand that scientists predict how many fish are in a given sea.
So scientists and fishermen and everyone else rely on computer models that mimic what is known about fish. Into the models goes information like size, age, growth rate, how many fish will die of natural mortality (predation, disease, moving away from the area) and how many are taken in the fishery.
Lobsters lack body parts like ear bones that help to reveal the age of other species; as a result, modelers use lobster size instead. But even size can be misleading, because lobster growth rates also vary with temperature: the warmer the water, the faster they grow. And since most lobsters move around throughout the year and over the course of their lives, their growth rate does not stay the same.
Variable growth rates can have surprising effects. According to Dr. Yong Chen, a fisheries scientist at the School of Marine Sciences at the University of Maine, the lobsters that grow into the size that can be legally harvested during any given year can include individuals born over a seven-year time span. In other words, the lobster on your plate could be four years old and the one on your friend's plate could be eleven years old, even if they are both one-pound lobsters that were caught in the same trap.
These facts complicate computer models. Throughout the 1990s and early 2000s, the National Marine Fisheries Service model consistently underestimated the number of lobsters in the sea, and therefore overestimated the percentage being caught in the fishery each year, leading federal scientists to believe that overfishing was occurring in the Maine lobster industry. Yet year after year, catches went up and research surveys recorded higher and higher numbers of lobsters. Clearly the model wasn't working.
Fast forward to 2008: The Atlantic States Marine Fisheries Commission and the National Marine Fisheries Service officially adopted a completely new model that estimated the lobster population as "not overfished."
This transformation occurred largely thanks to the talent and tenacity of Dr. Yong Chen.
According to Chen, there are four main areas where his model improved on the prior version. "We included the inshore trawl data from Maine and other state surveys, in addition to federal survey data; we had better catch data to work with than before; we had more realistic biology built into our virtual lobsters; and we used a statistical approach that incorporates margins of error in our inputs (this approach uses Bayesian statistics)," he said.
But how do we know that the new model is right versus less wrong?
Subscribe to:
Posts (Atom)