 Okay. Well, thank you very much to the organizers for having me here today. I'm delighted to have the opportunity to present this paper, which is joint with Yuri Gorodnichenko and a student of his, Mauricio Willate, who will be on the market this year. This paper is going to be motivated to a large extent by the following figure. So this shows the trajectory of U.S. real GDP, since the Great Recession. So the black line here is real GDP and the red line was the projected level of potential GDP from the Congressional Budget Office prior to the start of the Great Recession. So what you can see is that relative to this anticipated level of GDP, the U.S. economy has fallen far behind and has not recovered back to this trend. Now, what's happened since then is kind of interesting in terms of these estimates of the path of potential GDP is that gradually over time the Congressional Budget Office has revised downward and downward and downward and downward its estimates of the potential level of U.S. GDP. So that at this point you would say, well, the output gap in the U.S. is zero. We have recovered. We're back to the potential level of GDP. But what you can see from this figure is that this view that we've recovered that the output gap is zero is basically coming entirely from the fact that we've revised downward our estimates of potential GDP, not because GDP has recovered back to anything like its previous trends. So there's several views that are associated with this figure, different ways to interpret it. One is what we're going to call somewhat loosely the Fed view, which is the CBO is largely correct in having revised downward its estimates of potential GDP. And so we should take more or less literally this view that the output gap is zero. If the output gap is zero, that means it's time to bring monetary policy back to a more normal level, undo the kind of quantitative easing policies that we've done and start raising interest rates. This is precisely what the Federal Reserve has been doing. Another view is what we'll call the hysteresis view. It's to say, well, what happened here is that we had very large demand shocks, which brought down the potential level of GDP over time. There's a variety of mechanisms that we can think about, about why demand shocks can ultimately reduce the potential level of GDP. But if you take this view of the world and you could say, well, this has the level of potential depends on the kind of demand side policies that we pursue. And so if this is the world we live in, then in principle, expansionary policies at this point could reverse these declines in the potential level of GDP. In this case, you could justify having much more expansionary monetary policies than according to the Fed view to try and bring potential GDP back up. A third interpretation, and this is the one we're going to push for in this paper, is what we're going to call the mis-measurement view. And this is going to be that this perception that potential GDP has declined in the way that the CBO describes is likely to be too pessimistic. The potential probably has not declined anywhere near as much as what is estimated here, which suggests that there would remain a very large output gap, and therefore there remains a lot of room for further expansionary monetary and fiscal policies. So this is the view we're going to advocate in this paper. And it's going to have two parts to getting there. So the first part is going to look at historical estimates, real-time estimates of potential GDP, like those that I just showed you from the CBO, and ask, well, what do we know about these things? We're going to behave in the way that we would like, high-quality real-time estimates of potential GDP to follow. And so what we're going to do is we're going to take a wide range of estimates of potential GDP from a variety of different sources for the U.S. but also for other countries, and we're going to study their behavior. Now the way that we're going to assess the quality of these estimates is to look at how they respond to different kinds of identified shocks. So the basic idea is we're going to say, look, potential output should be responding to supply side shocks and should be largely independent of demand side shocks. So this is what we're going to do. We're going to take some supply shocks. We're going to take some demand shocks. And we're going to ask, well, when you have a supply shock, do we see real-time estimates of potential GDP respond immediately and quickly in the right direction when you have demand side shocks, do we see these real-time estimates of GDP respond at all? And what we're going to find is that these estimates of potential GDP respond gradually to all of these shocks. So it appears as if the organizations that are estimating potential GDP, it looks as if they're incapable of differentiating between these two kinds of shocks in real-time and they're adjusting their estimates of potential GDP in the direction of actual output, kind of regardless of what the source of the change in GDP actually turns out to be. In fact, we're going to be able to replicate the actual dynamics of real-time estimates of potential GDP very easily by creating our own pseudo-estimates of potential GDP through an HP filter. And an HP filter is basically just a moving average. So what it's going to do is whenever you have persistent changes in GDP, of course an HP filter of GDP is going to gradually move in the direction of the change in GDP regardless of what the source is. So we're going to show this can reproduce very closely these estimates of actual estimates of potential GDP. So the interpretation that we draw from this is that when you see large downward revisions in estimates of potential GDP, like the ones from the CBO since the Great Recession, this doesn't really tell you anything about whether the changes in GDP are actually coming from permanent supply side changes. All it's really telling you is, well, the change in GDP has been persistent and so we're always going to see downward revisions in these estimates of potential after persistent changes in GDP regardless of what the source actually is. So then in the second part of the paper, we want to say, well, can we do any better? Can we create estimates of potential GDP that actually differentiate between the supply and demand shocks? And here we're going to take a number of approaches. Our benchmark is going to be to use Blanchard and Qua decomposition in real time in the Blanchard and Qua approach. It's kind of a very simple representation of output dynamics that decomposes changes in output between supply and demand. We're going to use this approach to construct real time estimates of potential GDP, which we're going to define as changes in GDP that are coming only from the identified supply side shocks. And we're going to show that these estimates of potential GDP in real time respond in the right way to the kinds of identified supply and demand shocks that we used in the first part of the paper. So they correctly respond quickly to supply side shocks. They don't respond to demand side shocks. So this suggests that in real time it's possible to differentiate between supply side and demand side shocks when we're trying to look at what's happening to potential GDP. And this approach is going to suggest much smaller revisions to potential GDP since the Great Recession. We're also going to consider a variety of other approaches to estimating potential GDP. One approach following Galli is going to be to identify technology shocks. Another approach is going to be to exploit information from consumption to figure out what's happening to potential GDP. A third approach is going to be to look at nominal variables like inflation, a Phillips curve approach to infer what's happening to the output gap. All these approaches are going to give us the same qualitative answer, which is it looks like the potential GDP has declined much less than what the CBO is estimating. It appears as if there remains a significant output gap in the U.S. So in practice there's a variety of approaches that are used to estimate potential output by different organizations. One approach is the production function approach. So it's kind of from the ground up. You try and estimate what would be the level of capital and labor that the economy would be using if it was more or less at its natural level. What is the level of productivity? You plug these into a production function and that gives you an estimate of potential output. A second approach, a class of approach is that people use are more statistical methods. Here you observe, say, GDP and you ask can we differentiate between permanent changes in GDP, transitory changes in GDP so you can do this through things like an HP filter, unobserved component approaches, or multivariate versions of this. This is a very common approach to estimating potential across a wide range of agencies. A third approach is more structural, also going to be more recent. So this is you write down a DSG model. You estimate the structure of the model, the shocks that are driving the economy, and then from this you infer afterwards what is the potential level of GDP that's consistent with the dynamics that you've estimated. In this paper we're going to focus on estimates that come from the first two approaches. This is because the more structural DSG approach has only been used much more recently and so we don't really have time series of these real-time estimates coming from DSG models that we could study, plus no one would provide us with their real-time estimates from these DSG approaches in the central banking community. So we don't know what these would have said. And then what we're going to suggest is to combine the statistical approach with some economic restrictions to try and estimate potential GDP in a different way. So we're going to look at estimates, as I mentioned, from different set of agencies. So for the U.S. we're going to focus on estimates that come from the Congressional Budget Office as well as from the Federal Reserve. And then we're also going to look at estimates of potential GDP for other countries and so these are going to come from the International Monetary Fund, the OECD, and then we're also going to look at some private sector forecast. So from the private sector we don't observe estimates of potential GDP per se but we do have from consensus economics estimates of the long-run growth rate of output and so we're going to say, well, you know, if you think business cycles are relatively transitory events then the long-run growth rates of output are going to be telling you something about the potential levels of GDP. Okay, so we're going to have a wide range of these real-time estimates of potential output and so what we want to ask is how do these estimates of potential GDP respond to different kinds of shocks? Okay, so we're going to assess this through a very simple econometric framework. So specifically what we're going to do is we're going to take revisions in the estimates of potential output. So that's the delta log y star at any moment in time and we're going to regress it on identified structural shocks. That's the epsilons. Okay, we're basically just going to trace out the impulse responses of these real-time estimates to identify structural shocks and for comparison I'll show you what's happening to actual output after these same shocks. Okay, in terms of the shocks we're going to use three different kinds of supply and three different kinds of demand shocks and here our terminology in terms of what is a supply and what is a demand shock is basically just going to follow from whether the shock appears to have permanent or transitory effects on GDP. Okay, so in terms of the supply shocks we're going to focus primarily on TFP shocks. We're also going to look at tax shocks and oil price shocks and then in terms of demand shocks we're going to have monetary policy shocks and then two types of fiscal shocks. One is the military spending shocks from Valerie Raimi and another is government spending shocks identified as in Arbaq and Gurudnichenko. So we have these shocks for the U.S. When we go to the cross-country studies we will only have the TFP shocks, the oil price shocks, the monetary shocks and one measure of fiscal shocks. But in the interest of time I'm only going to show you responses to the TFP shocks and the monetary shocks here. Okay, so we will run these regressions tracing out the impulse responses of real-time estimates of potential GDP to these different kinds of shocks using these different estimates of potential output. So it's going to turn out that all of the estimates of potential output display the same qualitative properties in response to supply and demand shocks. So I'm just going to show you a couple kind of representative ones. So this is going to be the TFP shocks and the monetary policy shocks for the real-time estimates coming from the International Monetary Fund. Okay, but again this is more or less what we observe for all the different real-time estimates. Okay, so here's the first result. So this is the effect of a productivity shock on GDP and on real-time estimates of GDP. Okay, so the black line in the shaded area, this is the response of actual output to one of these TFP shocks. So you can see output jumps up and stays at this permanently higher level. The question is, well what happens to real-time estimates of potential GDP after one of these shocks? That's the blue lines here. What you can see is you observe a very gradual increase in these real-time estimates. Okay, so the agencies that are constructing these estimates of potential GDP are not recognizing in real-time that TFP has jumped up and at the level of potential GDP has jumped up as well. Okay, so instead you get this very gradual response in these real-time estimates. And this is what happens after demand shock. So here we have a monetary policy shock, right, which we know has transitory effects on GDP. Here the black line is the average response across the range of 20 or so countries that we have from the IMF. So you get a decline in GDP which reverses over time. So this is the transitory demand shock. What happens to the real-time estimates of potential GDP? Well that's the blue. So what you can see this is gradually declining in the direction of the actual change in GDP. Okay, so these agencies, these organizations that are estimating potential GDP, are finding or implicitly saying that when you have one of these transitory demand shocks, it's affecting potential GDP. So in the paper we have a wide range of robustness checks which I'm not going to run through here other than to say you can use different approaches to identifying the impulse responses. You can use different ways of identifying the shocks, different samples, etc. One thing that's kind of interesting is if you look at the CBO, you have estimates of potential GDP at different time horizons. And so you can ask, well how does the entire term structure of the estimates of potential GDP change after one of these shocks? And this is kind of interesting because one story you could tell for the monetary policy shocks is you could say, well if I write down a New Keynesian model with capital and things like that, then the natural level of GDP can change even in response to a monetary shock as say the level of capital goes down over time, then for a while you can have a decline in the natural level. But because the shock is transitory, ultimately the natural level goes back and so you wouldn't see any long-run changes in the potential level of GDP. And so you can ask, well, for those agencies that provide long-run forecasts of potential GDP, even if in the short run they're responding, it should be the case that their long-run estimates are not changing. In fact, if you look at the response of the long-run estimates, they track the response of the short-run estimates. So the entire term structure of the estimates of potential are moving together. So it doesn't really matter what the horizon is that you use in terms of doing these estimates, you'll find the same thing. So these transitory shocks are leading to downward revisions even in the long-run estimates of potential GDP from these agencies. So why is this happening? So one interpretation is if you're using statistical approaches like an HP filter to estimate potential GDP, you should expect to see precisely this. So why do we think these agencies are doing this? So here is the revisions in the estimates of potential GDP from the green books. And this is what you would get if you were trying to estimate potential GDP in real time using an HP filter in the U.S. You see that you can very closely reproduce these estimates. And this is true for the green books, this is true for all these other agencies. The real-time estimates that they produce look a whole lot like the kind of moving average that comes from, for example, the HP filter. And this is true in the impulse responses as well. So what we can do is we can create estimates of potential GDP ourselves in real time using very simple statistical methodologies like the HP filter and ask, well, do they look like, even conditional on the shocks, do they look like what these agencies are producing? And the answer is yes. So here the red impulse responses that we have are the impulse responses that we would get if we were estimating potential GDP using an HP filter. You can see they line up almost perfectly with the actual responses of the estimates of potential GDP from these agencies. Mechanically, it's pretty clear why this is going to happen. If you're using an HP filter, your estimate of the trend is going to capture these slow-moving components of changes in GDP. So it doesn't really matter what the source of that change is. If you have a demand shock which pushes output down for a while, your HP filter is going to move in that direction. So to the extent that these agencies are relying in part on statistical representations of potential GDP, it's not surprising that we find these kinds of responses. But the interpretation that you should draw from this is, therefore, when we see downward revisions in estimates of potential GDP, like we've observed since the Great Recession from agencies like the Congressional Budget Office or the IMF, this doesn't really tell you that they've identified shocks that are driving potential GDP. It's just saying, well, we've had a persistent decline in actual GDP. And so the methodologies that we use to estimate potential GDP are kind of mechanically decline, regardless of what the actual source of these changes may be. So the question is, what would other approaches imply? Or can we create estimates of potential GDP that are going to differentiate between supply and demand shocks in real time? And so what we're going to do is we're going to try a variety of approaches. These are off-the-shelf approaches, essentially because we couldn't come up with anything more clever than what was already out there. And so we'll show you what these different approaches do. So let me start with the first one. The first one we use is the Blanchard and Qua approach. So Blanchard and Qua says we're going to decompose changes in output into supply and demand shocks. And we identify these based on the assumption that only supply side shocks have long-run effects on GDP in a var. And so what we can do is we can apply this in real time. So take real-time series of GDP and say for each period we're going to apply Blanchard and Qua. And we're going to create a measure, an estimate, of potential output in real time from the Blanchard and Qua methodology, where the definition of potential output that we're going to use is that component of GDP coming only from supply side shocks. So we create these new estimates of potential output in real time. And then we ask, well, how do these respond to those same supply and demand side shocks that we were using before? And what we find there is that the estimates from the Blanchard and Qua methodology can differentiate in real time between the supply and the demand shocks. So they respond quickly to all of the supply shocks and they don't respond at all to the demand shocks. So they can do what the real-time estimates of potential output from these other agencies could not do. So then if we ask, well, what do these estimates imply for what's happened to potential output in the U.S. since the Great Recession, we get something that looks like this. So after a few years, the Blanchard and Qua approach says, yes, there is a decline in potential output since the Great Recession, but it's nowhere near as large as what the CBO estimates. And it implies that the output gap has not declined by anywhere near as much as what the CBO and other agencies would have us believe. Okay, so this is one methodology. There's a bunch of different approaches that we can think about to try and estimate potential that use some economic theory to say something about potential GDP. So one was the Blanchard and Qua approach. So I'm going to show you what these different approaches imply relative to both the initial CBO estimate as well as the more recent CBO estimates of potential output. So this is the Blanchard and Qua approach that I've just told you about. So another approach is to say, well, let's take a narrower view even of what is potential GDP. And let's say what should be moving potential GDP is changes in technology. So we can identify changes in technology using the structural VAR approach proposed by Galli in real time again, and then create real-time estimates of potential GDP as the changes in GDP that are coming exclusively from changes in technology, the identified shocks in technology, that's going to give you an estimate of potential which looks like this red line, which is even higher than the Blanchard and Qua, which is probably not surprising given that we're taking an even more restricted view of what can move potential GDP here. It has to be specifically technology shocks. Another approach is one that comes from John Cochran, and this is kind of a, I think this is a very clever approach that he suggested. So this is meant to identify permanent changes in GDP, and his suggestion was if you want to separate between transitory and permanent changes in GDP, you should look at consumption, because by the PIH consumption should respond only to permanent changes in GDP. So if you don't see consumption changing when income is changing, that should tell you that the change in income is a transitory one. When you see consumption changing along with changes in GDP, that should tell you something about the fact that the change in GDP is expected to be permanent. So if you use that to estimate potential GDP, then you're going to get, for the U.S., this kind of yellowish line, which again says it looks like potential GDP has gone down relative to the pre-crisis estimates of the trend, but again it hasn't gone down anywhere near as much as the CBO would have us think. Okay, a final interpretation, a final approach that we use is to rely on inflation dynamics. So the most natural approach that we think to say something about gaps is to relate them to changes in nominal variables. So this is what we do here. We say, well, let's use a Phillips curve to back out the level of potential GDP since the Great Recession. Now if you want to use a Phillips curve approach, it's a little delicate as to how you're going to do it, right? Because inflation dynamics are kind of tricky, recent inflation dynamics are kind of tricky to explain within a Phillips curve framework. So one interpretation of recent inflation dynamics is that the Phillips curve has become a whole lot flatter in recent periods, and so this is how you can explain the fact that inflation does not appear to have moved very much with output. But if you think that the Phillips curve has become very flat, then it tells you that inflation dynamics are very uninformative about what's happening to the gap. So Stock and Watson showed long ago that as the Phillips curve gets flatter and flatter, your uncertainty about the implied level of potential GDP or the natural rate of unemployment kind of blows up. So if you think the Phillips curve is really flat, then you should think that inflation is very uninformative about what's happening to potential GDP. But another interpretation is that, well, maybe the Phillips curve is fine, right? And there are ways of writing down the Phillips curve that are consistent with more recent inflation dynamics. And one way that you can do that is by using an expectations augmented Phillips curve, conditioning on real-time estimates of inflation expectations. So if you use that kind of Phillips curve, then you don't observe changing slopes over time, you don't have a missing disinflation puzzle. So you observe kind of a very well-behaved Phillips curve over time. It's not nearly as flat as the other approach would tell you. And so when you use that approach, what you find is because inflation remains below real-time estimates of future inflation, the net tells you that there remains a significant output gap. How big is the gap? Well, according to our estimate of the slope of the Phillips curve, it gives you this green line here, which suggests, again, that the gap is in the same neighborhood as what was suggested by the Blanchard and Croix approach. So all these approaches are more or less giving us the same answer, which is, yes, there has been some decline in potential GDP since the Great Recession, but it's not as big, so there remains a pretty significant output gap. So this is more or less what we did in this paper. Again, there's kind of two parts. So one is asking, are these real-time estimates of potential GDP doing what we think they should be doing? Are they responding in the right way to supply and demand shocks? And here the answer seems to be no, which means these estimates are really not telling us very much in real time about the sources of changes in GDP and therefore whether we should expect those to be reversed over time. And when we create other estimates of potential GDP, they paint a very different picture of the output gap in the U.S. today. So I'll stop there. Thank you very much.