"Using an old-fashioned terminology, the eurozone has an original sin, which is that it is not an optimal currency area. At the same time, if you ask me, “Should I marry my friend X?” I may tell you, “No, I don’t think you are compatible, you are going to end up divorced.” But that’s a very different question from, `Should I get a divorce now that we are married and have a mortgage, three kids in school, two cars, and a dog?'
"Like it or not, we got married to the Germans, and the Germans got married to the Spaniards. We need to make this work, because breaking up now would be way too costly. What we need is a reform of the euro. In terms of incentives, you need to tell countries that they will not face economic crises alone, that there is going to be money from the European Union that will help the Netherlands going through a rough patch in the same way that federal taxes and transfers will help if California suffers a bad period. That would imply, for instance, moving toward a bigger European Union budget and creating some European bond system. There is a lot of discussion among European economists about how to design such a thing. But there also need to be constraints. For this to be sustainable, fiscal discipline and cleaning up the house really needs to be done. There has to be a great bargain between those who point out the need for making financial and economic crises easier to go through and those who emphasize that, in the long run, rules are very important. That’s the big question mark: Is the political process within Europe going to be able to deliver that solution?"
The State of Macro
"In the mid-1990s, we learned as a profession how to build models that are dynamic, that take the randomness of the economy seriously, and that incorporate price and wage stickiness. That class of models started being called DSGE, which is the terribly unsexy Dynamic Stochastic General Equilibrium acronym. I think these models really clarify a lot of aspects of, for instance, how monetary policy interacts with aggregate activity, and we learn a lot from them.
"The second big leap, which we have had over the last 10 years, is a big revival in models with heterogeneity. In the standard basic model that we teach first-year graduate students, there is one household. But, of course, we know this is not a description of reality; we have people who are older versus younger, college-educated versus not college-educated, unemployed versus employed, high-income versus low-income. Both solving these models and taking them to the data was such a large task that, until around 10 years ago, not that many people wanted to use them. This led to criticisms of representative agent models with only one type of agent, but we didn’t have that many alternatives. But over the last 10 years there has been a tremendous jump in our computational capabilities. This iPhone on my desk is computationally more powerful than the best supercomputer on the planet in 1982. That means we can do a lot of things that even 10 years ago we couldn’t. ...
"The problem is that a lot of this exciting, backbreaking research has not transpired outside of the relatively small group of people working on the frontier. ... If you take the best 20 macroeconomists of my generation, of course they don’t agree on everything, but the things they talk about are very different from the type of things you will see on Twitter or the blogosphere. The conversation sometimes looks like two very different worlds. Sometimes I see criticisms about the state of macro saying, `Macroeconomists should do X,' and I’m thinking, `Well, we have been doing X for 15 years.' ...
"Many of the people who are currently very critical of macro are in another generation, and some of them may not be fully aware of where the frontier of research is right now. They also have plenty of free time, so it’s much easier for them to write 20 pages of some type of exposé, if they want to use that word, on the state of macro. This raises a more general issue of whether academia in general and the economics profession in particular have the right incentives to transmit some of these learnings from the frontier to the general public."
The Particle Filter Story
"I once made a joke at a conference that the particle filter pays for my mortgage. Now a lot of people ask, `How is your mortgage going?' and I say, `Nearly done.'
"Let me give you an example of what the particle filter does. In early 2018 we entered a time of high volatility in the stock market. The problem with volatility is that it is not directly observed: I can go to the back pages of the Financial Times and find a value in the table for a stock’s price, but there is no number to express its volatility. What you need is a statistical model that will let you learn about volatility from things you can actually observe, in this case, the variations of the stock market from one day to the next. This is called filtering — learning about things that you haven’t seen from things you can see.
"The original filters were developed for the space program. The idea is you are the guy in Houston with a joystick, and you see the satellite but can’t get its exact position because you are measuring with radar and there is noise. What you are trying to figure out is how much to push the joystick to the left or right given what the radar is telling you.
"For the longest time the most important filter was the Kalman filter. It requires two assumptions: that the world is linear, and that noise comes from a normal distribution, or is `well behaved.' Those assumptions prevent it from handling many, many questions in macroeconomics. The best example is volatility because it can only be positive: You can have a lot of volatility or very little, but you cannot have negative volatility.
"So when I was a graduate student, I was very interested in coming up with methods that could extend filtering to these types of environments. I spent a lot of hours browsing through math journals, and I heard about this new generation of methods called sequential Monte Carlo, which is a complex name for something quite simple: A classic question in a basic probability class is if you throw two die, what is the probability that the sum of the two is five. You have to calculate the probability that the first is a one and the second is a four, and so on, and when you do that homework you always make a mistake because you forget one combination. Alternatively, you could throw the dice one million times. Of course, in real life you can’t do that, but computers can do it for you.
"In the 1990s, some people came up with the idea of applying Monte Carlos recursively to filtering problems. I learned about these new methods, and I thought gee, this can be done in economics as well. So I came back to my office and got my dear friend and co-author Juan Rubio and I explained to him, `This can work,' and he said, `Yeah.' I said, `Well, let’s write a paper.' So we wrote the paper, my most-cited paper probably, and it still pays for my mortgage."