Nobel Symposium Ellen Mcgrattan Modern DSGE models: Theory and evidence

thank you uh thanks for inviting me so I want to talk about modern DSGE models so let me start with the progress report we've made a lot when I started graduate school they were still teaching is LM barely got through the prelims really really close call John John tried to push the fast-forward button you know to get more you know there was this rational expectations revolution and we were kind of slow to you know get on that bandwagon but I made it through the prelims thank God and Here I am fast forward to today we got some DSG ease that's good news so optimizing households firms transparent easy to understand forward-looking agents market clearing prices so now we can do like serious serious stuff now I know a Manuel is here and no boosts here those guys would say to me you know we can take all this stuff and jam it into the old system they're really clever about that like they can take really complicated miles and stick it into the is-lm it's kind of like watching my kids pack a bag you know you put the bag down and yet they're so I'm so glad to be actually um studying Marty and Harold who went to Minnesota and they they're just not trained to do that so I don't have to like study any like complicated graphs or anything that I wouldn't understand alright so let me give as you can tell there's a live debate going on in this field you could just tell from from the talks here so I'm gonna give the progress report from the perspective of Marty and the perspective of Harold but don't think about Marty and Harold it's not Marty Harold it's it's like all of us so we're all in kind of the same boat asking what I think is the key question I'm gonna focus on the key question so Marty lissa billion achievements Harold lissa billion challenges I want to ask one question it's that question RDS cheeese useful not used used as low bar okay that's low bar that Michael Kiley is working on the Edo model in the back of the Fed whatever I mean no really I'm not even meaning that as a joke I'm like whatever aha central bankers using in a fruitful way these models are they using them to make predictions about serious policy or not Marty would say they are they are okay so he's taking a stance that's good I like when people take a stance Harold he's a maybe but really like if you read the chapter that he gave me it's a no so he's on the other side and I would say there's a lot of us in there debating this okay so I'm gonna try to give you the sense especially those of you who are more the corporate finance finance guys what is why are we at odds with each other okay it's gonna be surprisingly same answer for both so bear with me Marty would say I want me to use his words cuz he always get mad if I paraphrase him okay a key challenge was to develop an empirically plausible version of the New Keynesian model and he goes on to say you know the his work with Christiana and Evans met this challenge his work and so I view like you know the usefulness came because you know we we finally have an empirically plausible framework and he contrasts nowadays with the early days by the way whenever anybody ever does early RBC that means like 1982 they never talk about anything after that nothing happened but those early early early RBC models no money frictionless markets and yeah they missed some key properties and and I would say one of the key priorities one of the things that you know a lot of us focused on was actually employment and ours which you know it's part of the dual mandate just an FYI then were there were the early New Keynesian models and Bob and Neil are here I should have added no money there because they're always complaining like there's not even money in the New Keynesian models okay now Marty would say well it's really more those were more qualitative like you could get some idea of signs of things but now we have the core the quantitative okay challenge a bollock same thing okay notice watch this check it out same thing now they're empirical a plausible how could be the same thing how can be the achievement and challenge well there's a but of course there is the empirical fit is due to non micro founded frictions and non structural shocks okay so what we're effectively doing is taking those old newgate know the early I shouldn't say old early New Keynesian miles and sticking in effectively wedges in those first-order conditions that aren't micro founded that's the counter all right so he asks are these models then useful okay now let me focus I don't I only have a limited time and he's gonna start flashing signs at me so I want to focus on two of the shocks it was alluded to a little bit but two exogenous shocks make up most of the variation in these models both the early models and the current models okay they're given names these names sorta mean something but remember they're exotic that's key but they account for most of the variation in real activity not just nominal okay let me look at again I'm going to go back you know the labor market side I find the most interesting so let me look at key very key labor market variables I'll you I'll point to the work of golly Smits founders and look at their unemployment and then I'll look at ours from goosed Lopez Selita no herps to go okay I'll get to it when I get to it okay let me start with godly Smits founders and their picture of unemployment so here on the right axis is that's the unemployment rate that's this is unemployment okay and then you have a bunch of bars and then you have shocks with names to give names okay like productivity risk premium and so now let me do watch this this is really cool color okay so the wage markup and the risk premium are accounting for a lot of the graph so these when you sum them up because there's negatives and positives you get back the black line so then you can say okay how much of the variation in these series is due to these different shocks that have names okay well in the old days you used to be lots of people talking about wage markup shocks and now lots of people are talking about risk premium shocks and they account for if we take it you know say we take to two thousand nine quarter one I mean you know you got four percent and three and a half is accounted for the red and blue okay so that's a lot hours my gust Herbst Lopez cielito and Smith who actually are doing state-of-the-art nonlinear do the whole thing at the zero lower bound do some cartwheels and do everything they're doing it okay this is like serious stuff here's ours now here it's a different picture so it's just the times Cirie's if i put only one all if i put all shocks on that's the black line if i did only TFP only a shock to investment or only risk premium you get the other lines okay so you can see the blue line something called risk premium is making up a lot of the difference so before we had R is premium and wage markups by the way there's no wit they didn't put in wage markups here I'm not sure why money sits in this little gap here and so does exogenous spending okay but my question of course is are they shocks or are they wedges super important because in one case maybe they're primitive and we've got everything figured out and we can start doing policy if it's wedges well as Harold said we may be back into the conundrum of 1976 Lucas critique okay so wage markups are they preference shocks or these really monopoly rents matters risk premium and these are just exhaust and they're very volatile by the way okay you either way either way we got some explaining to do and the risk premia are these flight to quality shocks are the external financing costs are they capital quality shocks I don't know I don't know there are always narratives in the paper but they're exogenous shocks and what we really need to do is figure out you know are they invariant to policy without structure we're just summarizing the extent of our ignorance okay a way forward we could take the wedges okay that we have which will help us point to promising structural dsgs the step one and it's some an important step I don't want to I don't want to diminish it at all but we need to identify the promising dsgs with the micro founded frictions and primitive interpretable shocks then comes the super hard part where we now start going to microdata and and really backing it up I feel like we've only got this part people are doing this but their Lots getting skipped here I feel okay now I would be remiss as a longtime employee of the Minneapolis Fed to not give you the Minneapolis Fed view on a way forward I'll say the better way forward which is to put more emphasis well less emphasis on this quarter to quarter you know let's get it right immediately and more on designing rules and institutions okay we had a president Gary Stern and a research director art Rolnick that put a lot of weight on the long term not the high frequency but the low frequency I know that people in certain positions can't do that they have to talk to the press all the time but we have hundreds and hundreds of people in central banks around the world surely some of them could be devoting time to designing rules and institutions I'm giving my favorite list of some things done at the Minneapolis Fed that I think we're relevant for the quantitative easing the current euro crisis too big to fail things that are kind of like important okay but long-term and then these are the seeds of the you know like we've got the architects and then we got the engineers they the architects build a bedrock and then the engineers can come work with that thank you all right I see I guess we will I'll collect a bunch of questions and then get the so Robert I wanted to react to what Marty said about Calvo pricing which adds an aside how wonderful it is to have a pricing formula named after yourself Phillips curve Taylor rule Lucas critique Ricardian equivalence the barrel regression I'm sure I'm leaving something going back to Calvo pricing I think the problem with this formulation is not its simplicity it's the assumption that there's a lack of feedback from pricing errors to the hazard rate the probability of adjusting prices and that specification I think tends to generate fat tails would respect the pricing errors and sizes of price adjustments when they occur and overall I think it ends up overstating the implications of sticky prices for business cycles and welfare costs related to that and to inflation so the main alternative is what Marty referred to as menu costs and I certainly agree with them that in that formulation you don't want to interpret menu costs literally it's the cost of changing the prices that are being posted but the critical element of the menu cost view in contrast with Calvo pricing is the idea of feedback from pricing errors to the adjustment probability and that feedback is of course present in the menu cost models and that ends up producing much smaller effects from sticky prices business cycles and to Associated welfare costs and I think it also ends up generating a better fit to micro observations related to pricing adjustment and that was brought out in a 2008 paper by Burstein and Helbig and more recently in a current paper by Nakamura and Steenson Ricardo so it's a question to Marty as well Marty consistent with your desire to quantify and how the SES are extremely useful to that you described one strategy for DSG and I wanted you to ask you whether it's that one versus two alternatives I'm going to say one strategies the Wunder illustrate extremely well and has been very much your work with Larry and others which is I write a model I see this impulse post doesn't quite work and so hey I need an adjustment cost here in order to fit this that one doesn't work well I need some real wage rigidity and so it's a very inductive process of science which is very useful so far as allows you to then go and fit and learn along the line where is they need to work culminating more recent as you said with adding risk premium shocks and others now there's two alternatives to build DSG models that I would still call DSG modelling one is a long lens of what and Ray earlier this morning called portable models which is I'd like to write one type of idea that really permeates across all markets that usually push you towards behavioral or expectations theories that for instance the work that I was doing 15 years ago with Greg Mankiw I said well if you take this idea of planning costs and management that you were saying well that applies to wage decisions investment decisions and lots of others so let's just do that and nothing else on top of it and let's see how far one goes and you know one gets somewhat far there's then a third approach which is the one that again I guess this morning was in I exemplified by John's and a cop was when they said hey I'm a theorist here's a mechanism Marty now you take it and quantify near addysg and so a version of that would be I've learned about four mechanisms here in this conference maybe ten okay if I add it up maybe thirty five across all the presentations let's go bottom-up I have lots of microphones for this for this for this let me just add them all up and build them at the SG you I don't you described one given time but how do you see these three different approaches to VDS you because they're all about the SG and quantifying but they're very different strategies in an inductive deductive or ground-up thank you Mike I wanted to respond to some of the things Harold said about the challenges faced by the current generation of DSG models and of course I certainly agree that there are plenty of challenges and and you know many many things that we need to try to continue to extend but I didn't think the summary of what the obvious problems were was terribly fair to the to the existing literature one thing was this comment on the finding that monetary policy shocks aren't found to explain much of the variance decomposition of inflation as somehow suggesting that the models are are way off and I think Silvana already explained reasonably well why there's an identification problem there and that that doesn't necessarily mean that the that the Phillips curve mechanism isn't isn't operative but I think maybe another important point to point out is that these models that are giving that result about the variance decomposition aren't models that imply monetary policy doesn't matter for inflation and if anybody in the room got that impression that that you know these are somehow models that are saying inflation has a life unrelated to monetary policy that'd be way off I mean Marty was certainly right to say that the models are very much like Milton Friedman's view of the world in what they say about that and the average inflation rate is completely determined by the character of systematic monetary policy and the rules the variance decomposition that you are talking about is all about what accounts for the short-run variations and inflation around its average level which is not necessarily random shifts in shifts in monetary policy and another issue that you drew attention to was the the so-called Neo fish Aryan prediction of a simple version of the model right you did an exercise and shifting the Taylor rule and saying that a relaxation of the rule might be predicted to make nominal interest rates go up there I think it's important to point out I mean of course that simple exercise is a well-known one and in a lot of the expositions of New Keynesian models and also the ants various answers to it also in the quantitative model that Marty presented that feature was not there I mean Marty did the exercise of showing what the predicted response to a shift in the central bank's interest rate reaction function was and it didn't have that property the suggestion that sivanna made which is that maybe you need to back away from complete rational expectations is one response because the response in the simple experiment you are getting depended on people immediately for seeing what the subsequent inflation response is going to be and having that feed into nominal interest rates quickly but in the model that Marty presented there were rational expectations so that wasn't the answer the answer was that the inflation process itself has much more inertia in it and so the response of inflation is delayed therefore the rationally anticipated inflation expectations don't rise nearly as rapidly and that's that's how the model was not you know not in fact getting this kind of neo fishery and short-run response of course the idea that a more persistent shift in the monetary policy rule would move inflation and nominal interest rates in the same direction I think shouldn't be puzzling I mean that's what we we shouldn't believe we should believe was true the short-run effects are in fact not not ones that the standard quantitative the SG models actually have in them okay so we're out of times I want to give the author's last word sorry for those that oh I'll try and be sure it so first I want to thank Mike for those comments but I think both discussions were great I'm actually quite stunned that Allan and I actually agree on a major issue and that is there you know if you're gonna do macroeconomics you're gonna have shocks unless you want to lock yourself into an Einsteinian view of the universe and you want everything to come from nonlinear dynamics and so then the question is always going to be was it a shock or was it a wedge in Ellen's right and I alluded to this observational equivalence problem we now know how to match the macro aggregates and that's why I think the most exciting areas are exactly looking at micro data to sort of gain additional information and I think people like Arlene Wong other people that are working on I won't belabor the point I think we agree on that it's hard to disagree on that on Sylvanas comments were great I think they were really great I do want to put this puts on the in perspective there are parts of these models and I've talked about this a lot with John some things you want to take really seriously but if you find the minor deviations from rational expectations overturn the implications of the model you should change the model the classic example of that is forward guidance we know that that's quite dicey so gab X and then Manuel have done work on this Nakamura Knight lots of people have done work in saying shield deviations from rational expectations the representative consumer those strong implications go away well then of course you shouldn't take them very seriously and I'm quite sympathetic to moving in the direction of Manuel and and going to behavioral personally if to the extent that they're in there they're infeasible on the calvo tails I mean look I think sticky plans are what's really going on and Chris Sims once said I love maybe Olivia wrote this I take these models seriously in terms of their predictions for various observables the welfare implications I personally take much less seriously for some of the reasons you're talking about so I personally if I was gonna do welfare implications I would want to do the sticky plans to eliminate those tails these things only make sense in moderate inflation environments and if I'm gonna drive the welfare implications by three guys that haven't changed their prices in 10,000 years of course I'm not going to take that as a modeler now I've got it say you know for the purposes I'm using it do I want to go with the simplicity and elegance of the of the Calvo model for the purposes in hand or do I want to put in the sticky plans which is considerably harder you know to Ricardo there's not enough time to really respond that it's a great question I'm pursuing one approach I think people should pursue whatever approach they want but I do think the future is looking at micro data to get a finer feeling for what's going on but the mechanisms are they plausible or not we sort of hit the end of what we can learn from just the aggregate time series health I want you the last word yeah so so thanks for the concern these were great discussions discussions so let me go after the Phillips curve once more and so it may well be that the Phillips curve has gone away because monetary policy is now really optimal and therefore you don't see that trade-off in the rate anymore certainly too when I put up the Keynesian regression the XT is an order to dress on the right hand side so I'd certainly concede the point that if you be careful nonetheless it's that it's not that visible in the data anymore might give one cause for pause I what you know to identify whether there's really a Phillips curve trade-off requires identification of monetary policy shocks I think I've written paper in the Jamie in 2005 that some people may remember I don't want to rehash that debate instead let me just focus on one episode in particular that we all remember which is support locally so the Paul Volcker we often talk about the Paul Volcker disinflation you know inflation rose allotment Paul Volcker came in and he said he suddenly said beater and monetary policy around and then you know there was a deep recession and inflation came down sometimes people refer to that as a walk of recession as Paul Volcker disinflation episode but if you take this metaphor this model which has become key workers more many of these central banks that's the whole purpose of this exercise that I just did the Smith Wallace model just says inflation came down by itself it just it just was a bunch of negative shocks to to price bridge market shocks monetary policy you know that that model you know had monetary policy constant throughout that episode I mean maybe right it's possible that that the poor VOC I didn't do anything to monetary policy and the inflation came down on itself in the recession all by itself right it's possible but I so I am sympathetic to the perspective that monetary policy has to a lot to do with inflation I'm sympathetic with the view that the pole that entering for volca in 1980 wasn't end of 1979 should say and this change money to policy it was a shock but but that particular model that we picked and that's a prominent model doesn't give you that and that's something that you need to take seriously [Applause]

Channel: Swedish House of Finance

Views: 6,184

Rating: 5 out of 5

Keywords: Swedish House of Finance, Nobel, Money and banking, Modern DSGE models

Id: 69rLdiWRoUs

Length: 26min 58sec (1618 seconds)

Published: Fri Jun 08 2018