 So good afternoon. We are gathered here for a first session on the future of macro and macro finance theory and models. Let me just say, as a matter of introduction, not as a matter of introducing the speakers, which I will not introduce, because they are very well known and respected, but as a matter of introducing the topics, that it is very proper that we start a discussion on the future of central banking with a discussion on the future of macro and a discussion on theory and models. And it is even more proper that we do it here when we're all gathered to contribute to our vice president, Vitor. It is very proper for central bankers to care about the future of macro and the future of modeling. We care a lot about models, theoretical models, and empirical models to quote, or maybe to misquote, max labor. As central bankers, we both have an ethics of responsibility and an ethics of conviction. The ethics of responsibility is about delivering on our mandate, which is bringing inflation close to 2%. I'm sure of it, or you will agree with that starting point. And the ethics of conviction, the Gizinox Ethic is about doing policy based on first principles informed by facts and framed by sound economic thinking, which is about having a sound macro framework and which is about having good models. And we're doing it seriously. And Vitor has been leading the effort here in the ECB to upgrade our models, revamp our macro thinking in the light of facts, as the president was saying, in the light of the lessons we've learned from the great financial crisis in particular, introducing financial frictions in our models, introducing home to mouth agents, and all kinds of imperfections that we've seen were important in the crisis. And it's not enough, and what we're going to discuss now is how to make it even better. And I always like to start a discussion on models, theoretical models, or empirical models by quoting General Eisenhower, who famously said that planning plans are useless, but planning is of the essence. So being on the receiving side of that discussion, I very much believe that models are useless, but modeling is of the essence. And I hope I will be maybe proven wrong on the first account and confirmed on the second account. So let me open the floor to John Mulbauer for the presentation of the first paper. Yes. And there should be slides. There are slides. See them here. Well, I greatly appreciate this opportunity to be here. And Mario Draghi said beautifully what I think about Vitor, just beautifully. Vitor has been very patient with me, has given me lots of time. We've had some wonderful conversations, not just about economics, but also music. And it's a real pleasure to be able to contribute to this occasion. Right. There is a paper. I'm going to talk about parts of the paper, just briefly the introduction. I'll skip the DSG New Keynesian debate most of it, but just touch on it. And then I'll talk about the positive aspect of what I want to talk about, which is financial stability and how evidence-based work that's a bit more open-minded can help improve our understanding of financial stability and integrate better with stress testing. And then, again, talk about the other big issue that central banks are concerned with, especially at the moment, which is the future of inflation and what drives inflation, is the ruff-lips curve and so on, and talk about some very recent work on modeling US inflation. So let me pay homage to be in with two intellectual giants who have influenced me since the 1970s. One is Joe Stiglitz. The other one is David Hendry. Joe's work on information economics had a profound effect on my thinking really throughout my professional life. He writes in his Nobel lecture that the attempts made to construct a new macro based on traditional microeconomics with its assumptions of well functioning markets was doomed to failure. And he particularly highlights the issue of people who lose income and aren't able to maintain their consumption because they don't have access to capital markets, at least in our good terms. And a second thing he says in his lecture is that information economics has alerted us to the fact that history matters. The path dependence is really important, something that Roger Farmer and I very much agree on. And dynamics are better described by evolutionary processes than by equilibrium. Now David Hendry is my other big influence and just give you two quotes from him. The false belief that database model selection is a subterfuge of scoundrels rather than the key to understanding the complexities of macroeconomies. That's been a mantra really most of my life. And then on economic forecasting where he's done really the leading work of anyone, understanding structural breaks, what you do when you're faced with a structural break, what you do afterwards is really, really important. So my brief comment about the New Canes in DSGE is well, it would have been nice. It wasn't new based on ideas made redundant by the asymmetric information revolution of Stiglitz and his colleagues. It wasn't Keynesian, it ignored coordination failures particularly between the real economy and finance and so useless for understanding financial stability. It wasn't dynamic enough because the leg structures implied by these models really don't fit the world at all. And it was hard to stochastic because stochastic is about statistical distributions. Radical uncertainty means in the time dimension the distribution is really quite different and cross-sectionary heterogeneity is missing. And hardly gender-equilibrium because it missed most of the system feedbacks that are important. And then the assumptions of rational expectations into separate optimization in the world that is subject to radical uncertainty and structural breaks has to be reformulated. So obviously repairs are necessary and many people agree and the ECB among others is leading in this work. Evidence to me is crucial. So there's lots of empirical evidence of course against the New Keynesian DSG model. The failure of aggregate consumption, the oral equation, something that Larry Christiana himself says is the most rejected equation in economics. The mountains of new micro-evidence on heterogeneity, credit constraints, buffer stock behavior, the influence of house prices on consumption in credit liberal economies, and of course the evidence against the New Keynesian Phillips curve both micro and macro. And then sadly in macro, evidence is seldom allowed to speak in the top journals. You have to go to the sub-journals, the specialist journals to allow macro evidence to show up. I think this is because of the pincer movement of the Lucas critique and the Sims critique, the incredible restrictions critique, which argue that we don't know what the theory is. Let's formulate a very, very general model and let the data speak that way. But of course, Bayesian BARs, which is what you have to, well, what the profession resorted to when faced with the curse of dimensionality are subject to compromise by strong priors and Bayesian restrictions. So let me turn to the positive side of my talk, the first positive side. So I want to talk about financial stability and think about it in a very general way and think about how empirical work can help appreciate the risks and understand how to integrate with stress testing. So I'd like to divide risks into three. So first of all, exogenous negative shocks, which for a small level of economy, really outside of the control of the policy makers. There could be a deterioration of the sense of trade. There could be a rising global interest rates. I mean, we haven't yet fully experienced the Trump shock, but who knows what that might do to global interest rates. There are all shocks, external credit supply shocks, physical risk, climate change, and a lot of things that small countries can't do anything about. A second thing they could do something about, which is the so-called fundamentals, things that look as part of the long-run solution, maybe fragile, I think of duration mismatch and credit supply, currency mismatch of debt, unsustainably weak financial regulation. These are all things that could have been avoided if the policy makers and the politicians had been on the wall. And then finally, the endogenous feedback loops, which can amplify risks, and the financial accelerator is what I'll talk about next. So let's think about positive and negative feedbacks. Let me start with the negative feedbacks, because obviously on the downside in a crisis, this is what really matters. So the buildup of debt, especially if the quality of lending has deteriorated ways on spending, a large expansion of the housing stock, which later weighs on prices. Well, this is something that varies across countries in the same way as the quality of lending is something that varies across countries and across time. A third negative feedback possibility is the increased saving for a down payment when house prices are income-wise. And it's really interesting that this negative feedback loop is quite high in Germany and France, but low in the UK and the US. And so that has implications for the stability of those economies. Now, quite a lot depends on the timing of negative feedbacks relative to the positive ones. So let me turn to the positive ones. First of all, extrapolated expectations of capital gains matter more when high leverage is possible. And that makes sense in first principles. It's also something my empirical work shows. Leverage amplifies gains. There's a leverage cycle. I expect the gains cause lenders to relax conditions because higher equity cushions make the loans look safer. Profitable end increase of the capital base. The third positive feedback loop occurs through residential investment, something we certainly saw in Ireland and Spain and the US. Booms that boost employment and income raising aggregate demand and feeding back. And then the fourth, the consumption channel. In countries with easy equity withdrawal, large collateral effects on consumption can occur, you know, think of the UK and the US. But time varying. In countries where equity withdrawal was not possible, that amplified mechanism is absent. So thinking about hedging net across countries and across time is really important. Of course, at the timing of the way these loops work. So here's a picture that my colleague, John Duca, designed showing how the household financial accelerator worked in the US. So we start off with the falling asset prices, the Norwegian housing crisis at the top, feeding into lower residential construction on the far left, weaker consumption second from the left. And then, of course, feeding back on the financial sector with bad loans rising, the ability of the financial sector to extend credit is very much reduced, spreads in the credit markets tighten, and this all feeds into lower GDP growth and feeds back onto lower asset prices, lower capital of financial firms and so on. I think most people recognize that story. So what are the implications for econometric models? Well, the FRB US model, which is still around, is not a DSG model and its example is being followed in many places where banks are now thinking about non-DSG models. It's good on expectations. I think it's very good on expectations, but I think it fails in other ways and there's some important lessons there. It imposes the net worth constraint on consumption. In other words, consumption depends on the aggregate of assets minus debt, lumping all assets together. So liquidity and credit shocks can affect consumption only through net worth, given income. So that means that debt necessarily has a trivial role relative to housing and stock market wealth. It missed the amplifying of feedback loops via the financial systems ability to extend credit. And in 2007 at Jackson Hole, Michigan gave a presentation showing the simulations of the Fed model, which showed that I think at 0.25% falling consumption relative to the baseline, so hardly a blip. It's got unstable parameters, the speed of adjustment in the consumption equation, which is the most important equation in the system, has almost halved in 10 years. And it's slow speed and inconsistent with focus and among central bankers about the real economy effects of monetary policy. So it's claimed in microfounded, but in my view it's not a structural equation in the more fundamental sense of the Cowles Commission. And the Fed model also lacks a decent residential investment equation. So how do we make progress? I think the encompassing principle is a really important principle for thinking about how to learn from data. So for applied work, we need to formulate models that encompass alternative theories. But allow the possibility that by imposing restrictions on this general model, a particular theory might be supported by the data. Something that David Henry and his co-authors proposed in the 1970s. Let me take the example of the Lifecycle Permanent Income Model and show how you can generalize it and learn from examining the implied restrictions, relaxing the implied restrictions. So here we have log of consumption is equal to a constant plus the log ratio of permanent income to current income plus the asset to income ratio plus log current income. That's the best approximation, the best log approximation of the story. Now you can generalize that. You can break net worth into different elements, liquid assets, debt, housing wealth, financial wealth. Most people think that cash is more spendable than pension wealth. And we know from theory that housing wealth is different from other wealth because housing is an assumption good as well. We can allow the coefficient on permanent income plus the current income different from one. Textbook says it's one. But if some consumers are myopic, it could be less. Moreover, the discount rate used to construct the discounted present value of future income should be far higher. That's what the microeconomics in the buffer stock theory says. Far higher than the conventional real interest rates that's used in textbook models. And then the intercept should be time varying. If down payment is required for mortgage, fall, saving for a down payment declines so that the consumption of income ratio can rise. And then because of shifting access to home equity, the coefficient on the house price effect or the housing collateral effect should be time varying. So we find for example, the new S and the UK was pretty much zero in the 1970s. And then with credit market liberalization, it rose very strongly. So what we need is a household equation system because of course, once you condition on these assets and debt and credit conditions, you need to estimate a system of equations so you can in terms of general equilibrium solve forward by working through the portfolio effects and then doginizing those. And in our work we extract credit conditions using a latent variable method. Credit conditions are latent variables from the same system. So in a ECB wooden paper on Germany and a new paper that's being published with Valerie Schodau, we estimate six equations for consumption, unsecured debt, mortgage debt, liquid assets, house prices and family income. And for France and Germany, we find very clear evidence of this dampening mechanism that I was talking about for the US. On the other hand, we find powerful evidence of this amplifying mechanism, time varying, which is physically strong in the upswing to the financial crisis and then very strong in the downswing. And just to show you that latent variables really do what they're supposed to do. The latent variable picks up everything that is not being explained by the rest of the model jointly in the set of equations. So in the house price equation, the mortgage stock equation, the consumption equation, it's a latent variable in all three equations. Now estimated from the French data, it turns out that it's incredibly close to the ratio, negative eight that is, to the ratio of non-performing loans relative to total bank loans. So the contraction of credit in the early 90s in France was very much connected with what was happening to the asset base of the banks and then the subsequent improvement and the new contraction later on. So, and of course that also then gives you the connection with the financial sector. You can connect what's happened in the banking sector with what's happened to loan conditions for the households and that's an important part of the whole story. Right, let me talk about the other topic that central banks are particularly concerned with which is inflation. Now in a paper that Janine R and I published in 2013, we forecast PCE inflation in the US. We looked it out of sample and performed it to different models and our key insights are that inflation is partly a process of relative price adjustment. Now that's not a new idea. Dennis Sargan actually, in his famous 1964 paper proposed exactly that and there's a great paper by David Hendry on the history of inflation in the UK with the same idea. What are the drivers? Well, unit labor costs, international prices and exchange rate and house prices are key elements of the modern solution. We find that including union density, a measure of labor market power, greatly improves the relevance of the unemployment rate. Now, because of the cost of dimensionality, such a problem in VARs, introducing a poor trade-off between the number of variables you can control for and the number of lags. We used a technique called posseminous longer lags, PLL, to give a better trade-off. The key intuition behind this is that the impulse response function becomes fuzzier as lags rise. In other words, the precise timing of shocks, a lag of 11 months or 13 months is not only precise, but there is an effect. So rather than ignore it, which is what the profession does, let's introduce it. And the idea here is to replace 24 unrestricted lag parameters by six. And that's shown in the slide here. We showed that every information set considered, PLL beats the Bayesian information criterion in terms of performance. And in our latest work, which is still ongoing, we find a fifth insight, which is that the pricing power affirms matters for price setting, along with the union power for wage setting. There's a great paper by Roulin and his colleagues showing what happened to the concentration ratio in the US industry back to the early 70s. And that measure is highly significant in our forecasting models, improves in-sample parameter stability and out-of-sample forecasting performance. So our model, the long-run drivers of core US inflation are union density, the unemployment rate, foreign prices, house prices, and the health and dollar concentration index. And under the compilation of union density, the unemployment seemed to do a good job of picking up unit labor costs. So the implication is that there is a, let's call it the Sargon Phillips Curve, where relative prices and equilibrium correction play an important role, is stable. And it turns out after three years of crisis, we're pretty much back to the model that there was before. Now here's a picture of what the model says about the impact of the concentration ratio, which is the red line, union density and the employment rate. You can see the explanation for one of the things that's puzzled central bankers, which is in the big, in the great recession, why did inflation not fall more? Well, our story is in the US. Part of it is because the concentration ratio rise rose. Most firms had more pricing power to offset the weakening power in the labor market. Well, stability is a big question. In our model, we can do recursive parameter estimates. And meaning from the top, we have the constant term, we have the integration term for import prices, then we have house prices, then the Huffendorf index and the unemployment rate, sorry, the unemployment rate and the union density. You can see recursively estimated from about 98, the parameters are really surprisingly constant. I mean, for an econometric model on inflation, I think that's rather good. So we conclude that the New Canes and Phillips curve is dead, but the Sargon Phillips curve is alive and well, given the right long-run controls, which include rise of price adjustment and including market power relationships. There are a few other insights that come from this as well. For example, the dynamics, I mean, you look at the details of the adjustment process, very consistent with what we know about from the exchange rate path through literature about pricing and local dollar markets. And for forecasting, once again, averaging of the best model with a relatively simple or progressive univariate model is improved robustness. So we can get a considerable improvement in the forecasting performance. Now here's a list of questions that come out of Janet Yellen's marvelous survey of issues and uncertainties connected with understanding inflation. Is there a stable relationship between unemployment and inflation? If it's stable, what's the lag? What measured labor market slack works best? Is the nairy or the natural rate a useful concept? Does the inclusion of private sector inflation forecast to improve the model? And is there a, well, and other questions, which I think we can all provide some answers to. So finally, to conclude, in macro heterogeneity rules, unemployment, for example, we know, is incredibly heterogeneous. It's one of the big inequalities in our society. But price-setting behavior is heterogeneous as well. And yet between these two heterogeneous objects, aggregations of heterogeneous objects, there is a relatively stable relationship. That's part of what macro is about. I made the point that profession often neglects longer lags. And our technique is a very simple method. Anybody can use anybody with a regression package can use to improve the trade-off between the range of variables considered and the range of lags. And then finally, I think the information economics revolution has highlighted the importance of credit and other liquidity constraints. And once you think about that, I mean, we know that the banking system has gone through a huge revolution. It's much more collateral-based than it was before. So once you acknowledge that credit is important, and that shifts in credit have taken place, surely you want to take into account the shifts in credit availability in the macro models, the policy models that you design to understand financial stability better. And this latent variable technique that we've been working with for some years, and it's quite helpful in making the link between the banking system, credit, and household behavior. Thank you. Thank you, John, for being exactly on time, which sets a very high standard for all next speakers. Where does the floor is yours? Thank you very much. So I'd like to echo John's thoughts about Vito and to thank him for including me. Like John, I've had many discussions or several discussions with Vito in which I would turn up in his office at four o'clock for a 15-minute discussion, and then an hour and a half we'd still be talking about economics. If I'd known that we could have also had a discussion about music, I'm sure we would still be there now. So let me say, to start out with, in 10 minutes, it's difficult to do justice to everything that's in John's very interesting paper. I'm gonna draw on what I take to be three themes. The first one that John did not say a huge amount about is that clearly some of the DSGE models we've been working with have not been particularly successful. Secondly, John draws and mentioned a couple of people that he's found very insightful. One is David Hendry and one is Joe Stiglitz, and I echo that. In fact, in my first job at the University of Toronto, I didn't really have a thesis, and then I went to see Joe Stiglitz give a talk, and my entire thesis ended up being inspired by that work, so I share the notion that there were some very important insights in what we call the information revolution in economics, and one of the things I'd like to talk about in this discussion is exactly what we can learn from that. My view is that what we learn is maybe even a little more radical than some of the things that John drew attention to. And finally, John provided some insights from his own work, his empirical work, and I'm going to, I hope, compliment that, agree with some of it and provide what I think are some important thoughts, particularly for policymaking, when we think about the relationship between inflation and unemployment. There are a couple of things, or three themes that I'll talk about. The first one I've alluded to, so exactly how should we think about introducing information theory. And there are really two ways, I think, in which it's important, and one's connected with the asset markets, and the other is connected with the labor market. And if you start thinking deeply about what we've learned about market failures, my view is that the lessons we should be taking away from information theory are a great deal beyond the idea that shocks become amplified. And in my own work, I've really tried to go back to what I take to be an important idea that was in Keynes's general theory, and which became forgotten. And that is that market economies are not self-correcting and can get stuck with unemployment rates, which could be 20% for decades or 5% for decades. And the way that we ought to think about that in the language of modern general equilibrium theory is that there are not just multiple equilibria, there's potentially a continuum of equilibria. And if you go away with that idea and you think about the progress we've made about the empirics of Phillips Curves, I'm gonna make the argument that there's a really important question we must ask ourselves because there's a huge amount of persistence in unemployment. In fact, I'll show you some evidence which suggests that it's difficult to distinguish the unemployment rate in the US from around and walk. And when you take that view, the question then becomes, is it on the supply side because some object that's useful called the natural rate of unemployment is moving around? Or is it on the demand side, in which case there's the potential for monetary policy not just to get us back to a better place more quickly, but to permanently influence the place that we're at? I'm gonna show you a little bit of evidence here from my own work on the connection between financial markets and labor markets. And what you're looking at here, the blue line is a measure of the real value of stock market wealth in the 1920s in the United States. The red line is the unemployment rate measured on the right axis on an inverted scale. So we're moving from zero to 30. And it's difficult to see an eyeballed chart which is as suggestive as that about a potential causal connection between the asset markets and the labor market. If you shoot ahead and you look at what happened in the most recent recession, the picture is similar. The magnitudes are not as great. We're moving from 4% to 12% rather than from zero to 30. But the notion of a causal mechanism between falls in wealth and increases in unemployment is I think a relatively easy one to take away from that chart. If you then ask, well, maybe those are special, this is a picture from 1945 up through 2011 showing you the connection in those same variables over that period of time. Again, the stock market is measured in real terms. The unemployment rate is, again, that's a transformation of the unemployment rate. And if you look at the time series properties of those two variables, you find that even though the unemployment rate is bounded, you can take a transformation of it that maps it into the real line. That transformation can be a random walk and looks like a random walk. The real value of asset markets looks like a random walk. Those two objects are co-integrated and the co-integrating relationship between them has been very, very stable over the entire post-war period. Now, my view is that that is a causal relationship that operates through a demand side mechanism and not a supply side mechanism. And now, I'm gonna show you some toy models to help you think about what that ought to entail. So this is the kind of model that all of modern DSGE theory works with. It's what I call a rocking horse model. And it has the property that after a shock, the economy would return back to its growth path. And the dynamics of a rocking horse model are a vector auto-aggression in which there's a stable point or a stable growth path which the economy is converging back to. The alternative is what I call elsewhere a windy boat model. So the economy is like a boat on the ocean with a broken rudder. And if the economy is like that, this is a description of what is called hysteresis. So if there's a shock, as there is in this picture here, instead of returning back to the same growth path, this economy returns back to a different growth path. So the interesting question is, is the economy more like the rocking horse or is it more like the windy boat? In the windy boat example, the dynamics of hysteresis are that instead of there being a point that the economy returns to, there's a set. So think of that as many, many potential equilibrium unemployment rates. And this is what happened in the US data after the Great Recession. And in my view, that picture is a lot closer to the windy boat view than the rocking horse model. And there's a saying I learned to pick up in my adopted country in the United States. If it looks like a duck, swims like a duck and quacks like a duck, it probably is a duck. The conclusion I take away from that is that, yes, there's a lot wrong with the kind of DSG models we've been looking at. Yes, we can learn from the information revolution, but the kinds of things we need to learn from the information revolution are likely to have more profound effects than the work of simple DSGE models. And it leads to a key question. And this is where John's work on the dynamics of unemployment and inflation come in. I personally, being quite critical of Phillips Curves, I've argued that they haven't really existed in data for really since Phillips wrote the first paper in 1958. On the other hand, on the other side, there are people like Bob Gordon. I'm not quite sure where John comes down, whether he comes down completely on the side of Gordon. Bob Gordon has been arguing that the Phillips Curve is alive and well, that you can estimate a stable Phillips Curve over almost all of the post-war period. But the way that he does that is by including the assumption that the natural rate of unemployment itself is a random walk. So the data has this non-stationary element in unemployment. The question we need to ask ourselves is that non-stationary movement, that very persistent movement in unemployment happening because of the structure of labor markets, which is causing natural rates of unemployment to go up, or is it due to some demand side variables that we can potentially influence through monetary policy? And they obviously have very different consequences for the way that we should think about operating, not just monetary, but also fiscal policy. So again, I enjoyed reading John's paper enormously. I've tried to give some complimentary ideas that I've talked about elsewhere, and I'm gonna end with a simple plug for where you might find them. And thank you.