 Peter asked me to write a paper for this conference. I thought this would be a very nice opportunity to look at two sets of facts which have been confronted to in the last seven years and go a bit deeper and see where this took me. So the two facts which are going to be the topic of this presentation, the first is that clearly, and this was shown by a graph by Mario earlier, output has not returned to the old trend after the crisis. And the question is, what's going on? The second is that if you go back seven years and you asked yourself, suppose that unemployment is going to increase as it actually did, what would you have expected inflation to do? And I think most of us would have expected disinflation maybe at a substantial rate, especially in the light of the Great Depression experience. And while there has been a decrease in inflation and some deflation in some places, this relatively small decline is known as the missing disinflation. And again, I wanted to look at the facts more closely. So having done this, I decided to enroll Orenio Ciroti, who is somewhere in the audience, to do the hard work. Then I talked to Larry and Larry thought this was an interesting set of ideas and he joined the team, so this is a three-authored paper. So what I'm going to do is revisit the first fact in the light of hysteresis hypothesis and revisit the second fact and looking at the relation between unemployment and inflation. And having done this, I will draw some implications from which we'll see about the next paper will actually go deeper in that direction. Let me emphasize the last line of this. This is very much an exploration in the sense that I think we have found interesting facts. We have, I think, intriguing hypotheses, but there is no question that there is a need for more work both at the empirical level and at the theoretical level. So let me just plunge. So the motivation for the first half of the paper is a graph that, again, very similar to one that Mario showed, which shows the evolution of output real GDP in the US and in the euro since 2000. And as you can see, if you look at the US, it looks like there was a level shift around the time of the crisis and continuation on something which looks a bit like the trend before the crisis, but surely lower than the old trend. For Europe, things look even worse in the sense that not only do you have this decrease in level, but it looks as if the growth rate thereafter is also lower. And so the question is, again, what happened? So what we decided to do was using the data set that we have at the fund, look at basically all the recessions we could look at. So we decided to look at 23 countries. These are advanced economies, 50 years, starting typically 1960 something. Look at all the recessions and then try to see what happens after recessions. Is what we just saw unusual, or is it a more general feature? So to do this, we decided to use a non-parametric method, something very intuitive, very visual. So the first thing you need to do is to define the recessions. And basically what you do, you use a formal algorithm, but the idea is you basically look at peaks and troughs and you call them recessions. Then you eliminate a few on other grounds. But that gives us about 120 recessions. For each recession, then you have to decide what was the pre-recession trend. And clearly you don't want to compute the trend using the data all the way to the beginning of a recession because there might have been a boom, in which case you would be over estimating the pre-recession trend. So what we do is we compute the trend using the data up to two years before the recession on the assumption that the last two years may be contaminated by a boom. And we use an additional condition which is based on credit booms. And if there is a credit boom in the pre-recession period, we even more careful and basically remove a few more years in order to eliminate that possibility. So then we have a trend and then the question is, where do we, so we position the trend two years before the recession, typically. And then we compute the trend using either four years before that or 10 years before that. There are arguments for doing it both ways. We agnostic, you're going to see the results. And this way we can basically draw a line which is anchored two years before a recession with a pre-recession trend. And then we look at where output is relative to that. That's a very simple, conceptually very simple way of doing things. We do it for log real GDP and log real GDP per working age population to remove some of the underlying trend in population. So let me show you two graphs of how this looks when you do it. The first one is going to be for the US. The second one is going to be for Portugal. So if you look at the US, there's about six or seven recessions, you can see the trend which are the red lines with the one standard deviation interval around them. Look at the last one, the current one and you can clearly see how output is not returning to the trend. It looks like it's lower. Very much what you saw in the earlier picture. Otherwise, if you look at the other recessions, it's less obvious. Some of them have some of the same. The one at the end of the 70s also has a trend which clearly is above output thereafter. So it is not the current case. It's not the only case, but it's one case. The one, the graph which I find most interesting is this one and I've taken Portugal, not because we are in Portugal, but because it's symptomatic of many European countries. If you look, most of the time, the pre-recession trend is above output or output is below the pre-recession trend in nearly all the episodes. And one way of describing it is it looks as if every recession is followed by lower growth, right, the lower trend after the recession. It looks as a series of discontinues, decreases in underlying growth and each time, or nearly each time, there is a recession that I think already suggests an interpretation which I'll come to. So these are, we do this for other countries. Sorry. This is the tabulation of what we get. Let me not get into all the numbers. Look at the line which says log real GDP. So the first line, basically we define, we classify the episodes in three groups. The ones where there doesn't seem to be a sustained gap, output more or less returns to trend. Again, not exactly, but statistically, insignificantly so. And then we divide the rest, the episodes between those which have a sustained gap and those which have an increasing gap. So sustained gap will be a level shift down. An increasing gap means that the post-recession trend is lower in effect than the pre-recession trend. And I think the results are quite striking, which is the proportion of recessions which are followed by an output gap either sustained or increasing is about two thirds of all recessions, right? And of those, two thirds of those are followed by an increasing gap. So what we've seen this time is not exceptional. It may be special, but it is not exceptional. So based on this, we decided to look at the type of recessions, recessions associated with disinflations, with financial crisis, with all prices and so on. And so having anticipated the next slide, I go back and I say, okay, so what do you do with the fact that I just showed you? And this is a crucial slide in terms of interpretation of the results. I think there are three ways of interpreting these results. The first one is indeed hysteresis, which is recessions have long run effects. They do something. We have stories for how they might lead to an increase in unemployment which will create an output gap. That's hysteresis, but you've seen in the data there's something stronger, something that our discussant Larry Ball is called super hysteresis, which is a lower growth, not just a lower level. I think it's a bit harder to tell stories, but you can tell stories about recession leading to changes in institutions which change the allocation, which change productivity growth, but we don't have a very good story for it, but at least it's possible that this would be the case. So the first explanation says, well, be very careful, it's hysteresis or even something stronger, super hysteresis. The second explanation says, well, correlation is not causality. You have third causes, oil prices, financial crisis. They have an acute effect at the beginning, but the effect is also there in the long run. An increase in oil prices has an effect on potential output. So in this case, there's no causality from the recession to growth, just due to something else. Then there's a third explanation, which is reverse causality, which is causality from lower growth to the recession. And the idea here has been explored by a number of people, which is that for reasons which I'll take as exogenous, you basically have movements in underlying growth over time, so some decades have higher underlying growth and some decades have less. What happens is people are in a world in which they think underlying growth is X, and then underlying growth slows down or they perceive underlying growth to be slowing down in the future. They learn something. What happens as they do this is that they realize the future is not as bright and they cut spending. Firms cut investment, households cut consumption, and you get a recession in response to the anticipation of bad news. So in this case, the causality runs from information about the future or a change in what's happening to the economy now underlying and leading to recession. So these are three different explanations, and now I go back to my previous set of comments, which is how do you distinguish between both basically by looking at recessions, which are caused by different factors? These different theories have different implications. Given the time, I may skip some. So what we find here, I'm not going to go into the details, is we find indeed that recessions which are caused by clear supply shocks, so oil prices or financial crisis are much more likely to be followed by an output gap thereafter. So there's clearly something there about supply shocks which plays a role. Although this does not distinguish between the common cause of the hysteresis explanation. One interesting conclusion is we look at recessions due to what we think are associated with what we think are mainly demand shocks. So we look in particular at the recessions which were engineered through intentional disinflations and more or less every country has had one doing the sample. And what we find here is that even in that case, there's a fairly large proportion of recessions which are followed by an output gap. If you look at the circles here for with intentional disinflation, 36% of recessions associated with the intentional disinflations have an output gap thereafter. You might have expected zero, it's not zero. So there seems to be something there. Very few of these recessions have an increasing output gap of a time. So let me conclude the first part. What do we conclude? And you can see that there are many potential ways of interpreting the evidence. My own sense is given the number of recessions which are followed by lower output growth, not just lower output level. And given that I don't think we have a very good story for super hysteresis, I tend to think of reverse causality as being an important factor which is that we have these underlying movements in potential growth. We realize things are not great, spending comes to a halt, there is a recession. This coincides with a number of observations. Bob Gordon has noticed that before the recession starts, typically productivity slows down. He has called this the end of expansion productivity decline. This would fit the story fairly well. There's a number of elements which suggest that the causality may run this way. At the same time, it could well be that this is at work but hysteresis is also at work and the conclusions from the demand induced recessions suggest that maybe hysteresis is at work in which case the causality also runs from recessions to later output output growth. This has extremely different implications for my trip policy, these two explanations of the same fact and I'll return to it later. So let me now move to the second part of the paper and I realize that I'm not managing my time as well as I should. So let me try to give you the highlights here. So the second is the missing disinflation. Why is it that we had a decline in inflation but we never got true deflation or true negative large deflation? And this is a graph which showed what happened, the decrease in inflation but not the dramatic one. So what we did here was to extend something that had been done in the World Economic Outlook two years ago, which is to estimate a Phillips curve which is given here which is a very conventional Phillips curve. It has inflation as a function of the unemployment gap, the difference between the actual unemployment and the natural rate. A function of expected inflation taken as long term expectations from surveys, lagged inflation, a term of pi M reflecting of a relative price of imported goods to capture imported inflation in a near term. What's special about it is that we allowed all the coefficients to move through time to try to capture evolutions and see how they had changed. So we allow the slope coefficient, the theta which is an important coefficient here to move. We allow the unemployment rate, natural unemployment rate to move as well. We allow the coefficients on long term expectations and past inflation to move. Technically what we do is use a Kalman filter, a non-linear Kalman filter to do this, but the intuition is very clear. Each of these coefficients we allow to follow random work with finite variance. Again, we more or less the same data set as before. We estimate it over 50 years, 20 countries, country by country and let me show you the results. So this is what we find for the two important coefficients. The first one is what is the weight on long term expectations and so you have this on the left. The black line is the median across countries and the blue band is the interquartile range and what you see is that the coefficient decreased in the 70s to about 0.5 and now increased to 0.7, 0.8. So indicating I think something we know which is that there is better anchoring of expectations. There is more of an effect on actual inflation. The one that we were focusing on is the slope of the Phillips curve and as you can see here, there has been a dramatic decline in the slope of the Phillips curve of the time. The timing is interesting. For the most part this took place from the mid 70s to the early 90s. Division more or less by three for the coefficient. Since then it has remained about constant and indeed during the crisis it has not declined further. Now what we do in the next few slides is actually look country by country so let me just show you United States and Germany. This is the estimated natural rate in red with a standard deviation band and then these are the corresponding coefficients for the anchoring coefficient and the slope of the Phillips curve. Look at the bottom two. What you can see is that for the US the slope of the Phillips curve has more or less steadily decreased, stabilizing around the early 90s and for a country like Germany it has decreased even more. It's not very different from zero. Now there was a fairly long interaction with my discussant who I think who argued that basically these coefficients were still significant. Here you can see that the standard deviation which is zero which would be very bad news saying that there is no reliable relation between the output gap or the unemployment gap and inflation, he has convinced me that maybe that conclusion is too strong. I think what's safe to say is the coefficient is very small relative to what it was and in some countries it is probably close to insignificant but I would not push it probably as much as in the current version of the paper. Let me pass on this. We do various robustness tests which you can look at at your leisure and then talk about the might repositions implications of each of our two tentative findings. Now for the first one the question is whether you believe it's whether it's causality or it's hysteresis or super hysteresis and this has dramatically different implications for the type of might repositions that you should pursue. If you think it's hysteresis then basically you want to be extremely aggressive in maintaining output because if you don't it might well be forever and that's going to be an issue so this leads to a much more aggressive might repositions that would be without hysteresis. If you believe that it's reverse causality then the warning is what you think maybe a Keynesian recession is due in fact to a slow down in growth so that the output gap may actually be smaller than you think and Mario had exactly that figure earlier in his presentation basically could be that underlying growth has decreased and you have to take this into account so it actually makes you much more careful about the size of the output gap that you would otherwise be so one tells you be more aggressive and the other says no be more careful so long as we don't have a good solution a good interpretation or a reliable interpretation then this tension is going to be there. What about the might repositions implications of the second one? So it's clear that if the effect of the unemployment gap or the output gap on inflation is small and uncertain then in order to achieve an inflation target you're going to have to move the output gap a whole lot this is true if it's small it's even more true if it's uncertain and that raises the issue of whether a single mandate is really the appropriate thing to do to the extent that you now have a mandate to keep inflation on target but you have an instrument which doesn't work very well or forces you to move the output gap a whole lot then the issue of a trade off clearly comes up so it seems to me that if we're really in a world in which this coefficient is small and somewhat uncertain then this really pleads for a dual mandate more explicitly in the sense that it's easier to focus on the output gap as well as inflation as just inflation. Let me end with what I see as a puzzle and a challenge if our results are robust which is that there's an interesting tension between the fact that there's clearly increasing credibility of central banks and we saw this in the coefficient which was estimated in which people believe very much that inflation will go back to the target of a central bank while at the same time the ability of a central bank to achieve that target if we believe that the coefficient of a Phillips curve has become small and uncertain is actually smaller so at the same time as people believe the central bank is more is going to deliver the ability of a central bank itself to deliver this inflation target this inflation rate equal to the target may have decreased and this might be bad news for the future. Let me stop here Peter. Thank you. Okay, I'm delighted to be here. Time is short so I'll get right to the point. There are two parts to this paper. The first part on hysteresis I think is very interesting and informative. The second part less interesting and informative and being a good discussant and having it be easier to be negative than positive I'll spend most of my time talking about the second part. Okay, so relatively briefly on the first part about hysteresis overall I think this is very well done and very sensible and balanced and I don't really have much to add. Let me just highlight a couple of points. First of all I think I essentially agree with Olivier. There's the odd result that if one goes out and looks for evidence of hysteresis due recessions have long-term effects after the recessions one tends to succeed beyond one's wildest dreams. There seem to be these very large effects on output or even effects on output growth, super hysteresis. The effects that jump out are so large that they're hard to believe and that makes us worry about reverse causality of what's really going on and I basically agree with Olivier that there are some puzzles here that need more research. Let me just throw out one tiny little idea about what in the world could this super hysteresis be if it really exists. There's interesting work by John Haltenwanger and others suggesting that recessions dampen entrepreneurship. Fewer new firms are started, perhaps fewer new technologies are introduced. Who knows, that might be one channel through which a recession can impede the growth process. I'll just throw that out. Finally in this part of the paper let me just emphasize the policy implication. That if hysteresis effects exist then there's a clear implication to put a bluntly that a single mandate for price stability is not a good idea. My understanding of the logic for the single mandate is that the only thing the central bank can control in the long run is the price level or inflation or nominal variables and therefore it makes sense to focus policy on those variables. If there's hysteresis that means that monetary policy can have long term effects on output and employment and focusing only on price stability might lead to success for price stability but very poor outcomes potentially for output and employment which would not be optimal. Oops. Okay, moving on to the second part of the paper and again I'll be fairly blunt. I think the authors here are not assistant professors whose spirits will be crushed by my criticism so I can be fairly blunt. So and actually I should say Olivier's remarks I think were somewhat more modest and moderate than the paper itself so I'm replying if it seems like I'm a little over the top I'm replying to the paper as opposed to what Olivier said. The paper suggests this idea that maybe the unemployment inflation relationship is broken down so such a degree that central banks are gonna have trouble controlling inflation and we should worry about whether central banks can really control inflation anymore and that I think is a little bit far-fetched. Let me first try to cast doubt on that idea by appealing to your intuition, asking a question and I realize actually we have the technology here we could have set up voting but I'm not prepared to do that so I'll just ask you to answer these questions introspectively. Suppose that President Draghi came back on stage and announced that the ECB was raising the target for interest rates by 300 basis points not for any good reason just because the executive board is in a bad mood. How would that affect the economy? For me that would tend to raise my forecasts of unemployment tend to reduce my forecasts of inflation or correspondingly what if President Draghi announced there was a 300 basis point reduction in the interest rate target. I think that would probably lead to lower unemployment everything else equal higher inflation. If you gave the same answer to those questions that I did then in your heart you believe the ECB faces some unemployment inflation trade off and can move along that trade off and control inflation. Now of course we all know my second thought experiment is not possible it would not be possible for obvious reasons to reduce the interest rate target by 300 basis points. So there is we can of course debate about quantitative easing there is potentially a reason to worry about in the current environment how well can the ECB control inflation but that's because of the zero bound on interest rates that's a completely separate issue from whether the Phillips curve is broken down which is what this paper emphasizes. Okay so going a little bit beyond appealing to your intuition let me give the results of a little bit of empirical work that I've done jointly with my co-author Sandeep Mazumdar. So I'll give some empirical results and then compare them to the paper that I'm discussing. So let's estimate a very simple Phillips curve inflation depends on expected inflation and economic activity inflation is core inflation specifically the weighted median of price changes I'll come back if I have time to why I think that's a good measure of core inflation. Expected inflation is long-term inflation expectations from surveys of forecasters economic activity is either unemployment or output in either case detrended with the HP filter and I'm gonna estimate this simple specification with quarterly data for the US and for the Euro area. So there will be four versions of the analysis for the two economies each with output and unemployment and the results simply are that it seems to me there's a very clear trade-off between activity and inflation and no particular evidence of instability in the relationship. Okay now this I fear is difficult to read so let me just summarize what it says. If the Phillips curve is unemployment on it there are similar results for the US and the Euro area the coefficient on unemployment in the Phillips curve is in the neighborhood of minus 0.5 or minus one half. At least for the United States back in the 1970s, 1980s this coefficient was probably closer to one so we've gone perhaps from one to a half so I certainly agree that the coefficient has fallen compared to the 1970s or early 80s but it still is substantial and the T statistics on the coefficient are somewhere above five so they're very significant statistically. If you use output rather than unemployment in the Phillips curve the coefficients are about half as large which is consistent with Oaken's law. Now this, well I don't know how easy or hard this is to see but these are scatter plots and I'll tell you what they are so the graphs on the left show the unexpected part of inflation graphed against the output gap so for the US and for Europe there's a positive relationship between inflation and the output gap positive relationship. If we use unemployment there's a negative relationship between, I'm not sure they said this right, there's a negative relationship between inflation and unemployment, negative relationship between inflation and unemployment so that looks like a Phillips curve to me. The red dots in the graph are the observations since 2007 and it doesn't seem as though the red dots are on a different line than the black dots so no obvious evidence of a breakdown in the relationship or a change since the onset of the Great Recession. Okay, now let me get back to the paper that I'm supposed to be discussing, compare my results to the Blanchard et al results and again what they emphasize in the paper is that at the end of the sample in 2014 the coefficients on unemployment are no longer statistically significant so to look a little more closely at the idea of statistical significance I'm gonna compare 95% confidence intervals for the unemployment coefficient in 2014. In my simple specification there's a single fixed unemployment coefficient in the Blanchard specification that varies over time. I'm looking at estimates for 2014 so for the specification that I just looked at in the last couple of graphs these were the 95% confidence intervals of the unemployment coefficient. Again clearly bounded away from zero a very significant relationship. Now if we look at the Blanchard specification they don't look at the Euro area as a whole they go country by country so I'm looking at the three largest countries in the Euro area plus the country that we happen to be in at the moment and overall my reading is that what happens when you move from the simple specification to the more complex one is that you widen the confidence intervals so except for the case of Germany the confidence intervals expand in both directions in the Blanchard specification they expand downward so they hit zero and the coefficient is constrained not to be less than zero. The upper bound also rises so I think my summary of the Blanchard et al results is by themselves they don't say very much about what the Phillips curve coefficient is. It could be zero, it could be over one or even over two for Portugal so the estimates are just not very precise. The simple specification that we looked at before gives much more precise estimates of the Phillips curve slope. Okay so what accounts for the greater precision of the simple specification? I think the obvious answer is that it's much less flexible or more parsimonious. My specification has one parameter which is fixed their specification has four parameters each of which is allowed to vary in a flexible way over time. I think generally in econometrics if you allow lots of flexibility and lots of time varying parameters it makes it more difficult to estimate parameters so I think what we learn from the exercise in this paper is that if you add enough flexibility in your specification it's possible to turn a significant coefficient into an insignificant coefficient but frankly I don't think we learn all that much from that. Furthermore I don't really think there's a very good motivation for the complex specification in the paper. Again if you look back at the scatter plots I presented the simple linear non-time varying Phillips curve fits the data pretty well so I think there's a strong case for just stopping there and not much motivation for introducing all the complexities of the Blanchard specification. To be a little more constructive to try to reconcile the different results we see here one thing that the authors might do is to try imposing the restriction that the unemployment coefficient has been constant since 2000 or perhaps since 1993 I can lecture that they would not reject that restriction and then also with that restriction that the unemployment coefficient will be significant consistent with what I found. Another possibility would be to allow the unemployment coefficient to vary over time but not in an unrestricted way to assume that it depends on the trend inflation rate and there's a lot of previous work suggesting that's a reasonable specification and I can lecture in that case that they would find that the unemployment coefficient does vary as trend inflation changes but that it's always significantly different from zero. So there's a technical point about measuring core inflation and why the weighted median is a good measure which I will skip taking seriously the time limits and just get to the policy conclusions. So two things that I take away from this paper which actually in the end are pretty similar to Olivier's points of view. Well at least in part. First of all the paper suggests again that we should worry about perhaps central banks cannot control inflation so well or have to move the output gap huge amounts to control inflation. Again I think that's a little bit far-fetched since inflation targeting started in 1992 central banks have done a good job of controlling inflation and there's no reason to think that that ability is deteriorated. In my final six seconds the quotation I have here I think is the quintessential statement of the belief in a single mandate that the job of the ECB is to keep inflation less than but close to 2% and if that happens policy is a success. Again I have a longer version of this but I think the basic point is obvious that if there are hysteresis effects a central bank might be very successful at controlling inflation but unsuccessful otherwise so we need to explicitly have a full employment mandate. Thank you.