Peter A. Diamond (2014) - Unemployment

I guess it’s still good morning since we haven’t gotten to lunch. I want to start with the statement from the FOMC after their last meeting: However, a range of labour market indicators suggests that there remains significant underutilisation of labour resources." I want to do 3 things in my allotted time. First I want to touch briefly on the indicators that the statement is referring to. And how they factor into my thinking about the state of the economy. And critical there is the perspective, you can find it back in Okun 62 with the publication of Okun’s law, that the unemployment rate is a proxy for the state of the labour market. And as with any proxy the quality of the fit varies. And what the FED is doing, and what I do on my own, is dig into other aspects of the labour market to see how well the proxy is fitting in current circumstances. And in their case that feeds back into monetary policy. In my case it feeds back into talking to students. The second thing I want to do is talk about the Beveridge curve and its role in the discussion of policy. And some brand new research that I’m started on which casts doubt on everything I’ve written about the Beveridge curve. So that mostly will open up some questions which I think will be very interesting. And third I want to say a little bit about methodology. I’ll say a little bit more about it in the panel tomorrow. Because I think that’s an important issue that never gets addressed directly in graduate educations I’m familiar with. So let’s start with some of the pictures, detailed pictures, of the labour market in the US. Let me be clear, the US labour market is the only one I have studied in detail. And if you want to draw inferences about other countries, having looked at other countries data you’re welcome to. But I’m not going to. So what you see here very clearly is the labour market in the US has been slowly improving. Starting a little bit after the NBER dated trough. But if you look at the different colours, you can see that unemployment rate for short durations are back to normal. unemployment rate for long duration is not. And part of that is a natural consequence of dynamics when you approach the labour market, as search theory was designed to encourage, as a dynamic process. And then, obviously, if you don’t have a job and don’t find one, your duration of unemployment changes. And insofar as the hazard rate of finding a job varies significantly with duration, then it’s natural you get a pile-up that’s very slow in coming down. And that’s what we’re seeing: historically a stunningly high level of long-duration unemployment. And we know that’s terribly important because we’ve got lots of history of studies of displaced workers in the US and what happens when they come back. Those that get jobs on average have a 30% earnings cut and it’s long-lasting. So this is significant, obviously, for the people and their families but also for the productivity of the economy. Another element is that if labour force participation rate is down and together with that the ratio of employment to population is down. That has 2 parts to it: One is the trend demography of the aging of the US labour force. And the second is the impact of the great recession. To tell them apart a bit, here’s the same calculation just for 25 to 54 year olds. And we see the same kind of pattern there. In this case we’re seeing some improvement in the employment population ratio, but it’s still very low. So some of this is: hard to get a job, staying at school. Hard to get a job: you’re older, take your retirement then. If it’s hard to get a job, you’re eligible for disability benefits. You’d rather have a job but you’ll take the benefits rather than going hungry. Simple example there: In the US blind people are disabled. Lots of blind people work. If you’re blind and you lose your job you have this as an option. Will you come back to work? Well, by and large, the people who succeed in getting disability benefits don’t. But young people who dropped out of the labour market to go to school are obviously intending to come back. And with the hit that so many retirement asset accumulations took, when there are more jobs, I think, we’ll be seeing significant numbers of older people coming back. This is to my mind one of the more significant pictures of what’s a problem from the slack in the US labour market. The workers who were interviewed, they’re employed. They’re part-time employed and they’re asked, why are you part time? For some of them the answer is, "I want part-time work." That’s what I do since I retired from MIT: I’m part-time employed, coming and giving talks. But for others they’re part-time employed because they can’t find a full-time job that they want. And what we’re seeing here is just a huge number of that. And that’s connected to the very limited opportunities in the market. The other sign of the limited opportunities in the market is the quit rate. These are people - remember people get interviewed once a month. So one month they’re employed. And then the next month they may be employed again with a different employer and they may have quit the first job. Quits are a really important part of the function in the labour market. Because the allocation of different workers across different jobs includes what are called EE flows, employed to employed, as a big part of the process of the allocation of different workers to different jobs. And we’ve seen again the same picture as in the other things, that the economy is improving. People are quitting because they’re finding opportunities that are worth abandoning something they were in for something that is always going to be not quite sure how it will work out. And it’s slowly improving, but still way below what it was before the crisis. On the other hand layoffs which tend to spike in the recession and go back to normal fairly quickly are back to normal. That’s coming from a different place. Now here’s the way quits are a very good measure for the flow from employed one month to employed the next. Because these data come from different sources, the quit data presented here doesn’t come from interviewing workers but interviewing firms. Firms are asked at the end of the month: workers who left, workers who were added, net change in employment. So the quits data comes from there. The EE flow data - because a firm doesn’t know, if a worker quit, necessarily what happened – comes from the labour market surveys, the workers surveys. And as you can see there’s a stunningly high correlation. That’s a big part of what’s going on. The other element that’s a big part of it is, to a very large extent, when you quit one job for another you get replaced. You may be quitting because the firm is going down and you’re leaving before it actually happens. But the vast amount of quits are in places of ongoing businesses. And so that you get replacement hires afterwards. And it's replacement hires that are a large part of the job opportunities of workers. This, to my mind, is the most important picture for the functioning of the labour market from the perspective of the kind of dynamics I’m talking about. What you see here, month by month, is the number of workers who have been hired in firms – from a large survey of firms, incomplete and particularly missing out on start-ups - and the number of workers who are separated from the firm - quit, lay off, retired, what have you. And the net change in employment is the small difference between 2 very big numbers. So even running about 200,000 net increases in employment in the US in the steadily improving labour market, that comes about as you can see from 4.5 million departures and 4.7 million hires. The 200,000 net amount is very small relative to the gross flows. For a worker looking for a job it’s the gross hires that indicate the opportunities that are out there, not the net change. And what you can see in this picture is that’s very low. So put this together and it seems to me the FOMC statement, Janet Yellen’s speeches on the subject are right on target. This is a very soft labour market with a great deal of slack in it. And the opportunities are extremely limited for multiple reasons. Not just a low number of vacancies - I’ll get to that in a moment. But also a process where the mindset of workers and the mindset of employers about the future of the economy is hurting the growth of the economy and hurting individual experiences. That said let’s turn to the Beveridge curve. The Beveridge curve is the month-by-month unemployment rate and vacancy rate, which in this figure comes from the JOLTS data, the Job Opportunities and Labour Turnover Survey which only started in December 2000. And the Beveridge curve is meant to describe what happens in the business cycle. You go down a curve - vacancies go down, unemployment goes up. And in a recovery you go back the other way. Dynamic models differential equations - I’ll touch on that briefly coming ahead - tell us that you get a steady state curve if you were staying at that point. And you get loops around it. In the US vacancies get filled on average very quickly. In good labour markets workers find jobs very quickly, even in bad labour markets. On average workers find jobs very quickly. So the movement from, particularly the newly laid off. The deviations, the loops around the curve, are small. And what you see here is after the official NBER trough. The economy looked funny and vacancies started appearing. And the curve appears to have shifted. So the question is how to interpret that. And here’s an example of an interpretation from 2010. Which is if this shift is something structural - structural unemployment has the over tone - it's lasting, it’s not going to go away very quickly. And it’s a problem that monetary and fiscal policy is not addressing to. When the St. Louis Fed put this statement out, they also included in the report this picture of the Beveridge curve as it existed up to that point. The role of shifts in the Beveridge curve very much influence thinking about it. Here is the CBO. And notice the wording comes in part from a shift in what is known. What to my mind is totally appalling about the statement from the St. Louis Fed is a failure to realise that even if you get an increase in structural employment, you can still have an aggregate demand problem as well. So if the full-employment point you’re aiming for has moved and you’re still way above it, there’s an important role for monetary and fiscal policies. And to say, because it’s shifted there isn’t, is to fundamentally misuse the underlying theory. Now the underlying theory starts with this figure. You’ll notice unemployment is on the vertical axis, vacancies are on the horizontal. We now do it the other way around which is how the other figure was. But it doesn’t much matter, because this 1958 paper, drawing on Beveridge's discussion of how you use unemployment and vacancies to think about full employment, adopted his view that full employment is unemployed equals vacancy. We no longer think of that as a useful criterion. And so the picture here, given the relabelled axis: If you’re at point 5 and you’ve got this structural change, then 4 is their full employment point. And if you’re at 5 moving from 5 to 4 is an aggregate demand issue, not a labour market focused issue. Obviously, if you can do something to shift the curve down - that’s useful too. But it’s not a substitute for aggregate demand policies. So Dow and Dicks-Mireaux used the vocabulary 'labour maladjustment', 'degree of maladjustment' - that was their picture. And it’s now commonly referred to as mismatch. Shift in the curve means mismatch. Mismatch if it lasts, when there’s a presumption it will last, means the target for full employment is all different. And I’ve written that sort of thing myself as well. And then I started scrutinising some of the data. And now what I’m going to do is a full frontal attack on what I have written. So here the focus in my presentation will not be NBER dates, but the point of the highest unemployment rate, the point of the lowest unemployment rate. That’s what will determine a cycle. And so here is the point of highest unemployment rate. This is quarterly because it’s easier to see - 2009, quarter 4. And what you see is coming in to it. There was low vacancies and high unemployment. Vacancies started picking up, were moving vertically. But the unemployment is not responding, so we have a gap. And so I looked at the data and said, this is great recession. It will be different from before. But what sort of experience did we have before? So there are 4 of the previous recessions. Again, 6 months on either side of the point of the highest unemployment rate. And in those 4 every one of them at this point is showing the same kind of shift in the curve that we’re seeing now. Maybe, if it’s something that happens often, it’s not much of a signal of what’s available going ahead. So here it is, I’ve run it out a year. And now there are gaps in all of them although you see the 1961 one, the gap is starting to close. So that’s started to suggest looking at the others. Well, what about the ones that didn’t show a gap at 2 quarters? And there they are. But run those out to 4 quarters and with the delay, gaps reappear. It now appears as if the gap is happening more or less every time. And so there are all 9 recessions we have data for in the data set. Estimating vacancies coming from help wanted ads. And you can see, except for the very mild and rather unusual recession early in this century, a shift in the Beveridge curve is a standard part of what happens. And it appears to be bigger than would fit with the simple differential equation dynamics, although I’m not aware of simulations. What happens afterwards, if we’re seeing this many shifts? So here they are where we have run each of the curves from the previous minimum unemployment point to the next one. And some of the time you end up with a lower minimum unemployment than you had before. Some of the time you end up with a higher minimum unemployment than you had before. And it’s kind of 50/50. It’s clear that looking at shifts early on after the turning point is not terribly informative. What I’m talking about is work I’m doing by the way with Aysegül Sahin at the New York Federal Reserve Bank. So this table is in a paper that’s just been posted as in yesterday - I’m sorry I’m giving you such old stuff (laughter) – as a staff paper at the New York Federal Reserve Bank website. And as you can see shifts happen all the time. And as a signal of what will happen later: half the time it’s a yes, half the time it’s a no. The critical thing here is longer expansions lead us there. So let’s take a look. Here is the whole Beveridge curve making the point. This thing moves around a lot. And it’s not as if you can identify the reasons around these shifts in US labour market policy. Because the US policy unlike some places in Europe - primary example being Germany – have done significant changes in their labour market institutions. That hasn’t happened in the US. So let’s get to the theory. The simplest theory looks like this: The number of hires as a function of the unemployment rate and the number of vacancies is normally a Cobb-Douglas, gives you a pretty good fit. And then your dynamics or the change in the unemployment come from people leaving employment minus people being hired. This is a 2 statuses picture of the labour market, employed and unemployed. And the Beveridge curve is the steady state of that. The trouble with that is the actual employment picture. And there are multiple numbers on this chart. I’m going very fast because we’re getting very low on time. And there are widely recognised coding errors in answers to the question: And so people have done ways of trying to correct that to get a better view. But here are the flows. And the thing to notice is the hiring of the unemployed into employment on average over a long time. Is ballpark the same as the hiring of not in the labour force into employment? And if I had had the EE flows there, that’s also a ballpark the same number. And there are models that are more complicated, but then turn out to be much harder to work with. So what am I doing with this ongoing research? Econometrics of the level of showing you a picture of the static outcomes in the month is not the end of anybody’s sense of what good research should be. So the next question is what is it that’s happening around this turning point that’s generating the shift. And what we’ve started working with is the hiring function. We will also look at the mix of flows between unemployment and not in the labour force. And the first look at it, it appears as if the hiring function shifts down at the turning point pretty much all the time. We don’t have data quite as far back for this. We don’t have 9 recessions, I think we’ve got 4 or 5. Hiring is going down relative to unemployment and vacancies. And so the big question - we’ve gone as far as identifying that - is thinking through the models and trying to figure out why. What is going on? Thinking through the models on the different flows and then trying to put them together into some form of thinking about the labour market. A shortcoming here is this is partial equilibrium analysis. It’s the labour market taking vacancies as given. A lot of the literature takes productivity as an exogenous variable, explaining the value of creating a vacancy. That’s misguided for 2 reasons: One is productivity is an endogenous variable, not an exogenous variable, because labour is used in multiple ways in a firm. And secondly, the value to a firm of having more output is not measured the way productivity would measure it, unless you really do have a variation consumer market. Because it’s the market price times the quantity. But if to sell more you’ve got to cut your price, the familiar part. Or to sell more you’ve got to do more advertising. Or to sell more would have all sorts of game theoretic complications in the market. It’s not the right variable for what’s going on. So we need to be pushing this partial equilibrium model into general. Now let me say a few words on methodology. First I think it’s important to recognise - well, it’s common in graduate school to talk about a good model and a bad model. And what you want, if you’re a theorist, for your thesis is a good model. I think that’s basically misguided. What there are, are models that are good for a particular purpose, and models that are not helpful for that particular purpose. Let me give you an example. A very famous, wonderful article, Lucas Tree, in Econometrica, which we view as giving some insights into the determination of prices in asset markets. It’s only some insights, it’s not something you’d say this captures everything you’re interested in. But there’s another thing about stock markets that people are interested in: the quantity side. What determines how the volume of transactions changes over time? Is the Lucas Tree model going to be any help for that? Well, the level of transactions in that model, as all of you who have studied it know, is zero at all times. Zero at all times is not a wonderful insight into what’s happening to quantities in actual stock markets. So you’ve got to have - particularly if you’re moving as I do or trying to move as I do from basic research to policy recommendations – you’ve got to have a concrete question to think about: How to proceed and what you learn from different models. And this quote I’ll let you read from Marshall goes to the same place. Obviously, he’s talking in a world before computers, a world before econometrics. But I don’t think the message is any different. We use multiple models to gain insights in different aspects of how the economy works in a positive sense. And how policy will impact the economy in a normative sense. And to my mind there is no alternative to that. In part there are no policy changes that are Pareto improvements. Every policy change is going to help - I shouldn’t say that. Every good policy change is going to help some people and hurt others. You can set up a policy change that makes it worse for, if not everybody, almost there. So the point here is how do you pull together different sets of things you have information about in order to make choices. Both of the kind of policy you want and the numerical, the quantitative elements that are an essential part of that. And part of that goes back to thinking about how outcomes are determined in the market. I choose my words carefully staying away from the word equilibrium. Because that tends to mean different things to different people. And part of what drove me into working on search theory was a sense that there are a set of questions which the start of education - demand, supply, price clears the market or the price isn’t right to clear market and doesn’t – was an inadequate starting place for a whole range of questions. So here’s an example: Fenway Park is where the Boston Red Soxs play. And there are scalpers selling tickets outside all the time. There are individual game issues: What’s the weather like that day? Who happens to be pitching that day? And there are trend issues: Is the team doing well or badly this season? And so what happens to the price of scalpers? Well, it responds in part to what’s going on, but only in part. If you took out the standard model the price would fall to zero, because the scalpers end up not selling some of their tickets. But if you think about the game being played, the dynamic game of selling tickets over time - as this suggests-, it’s a more complicated story. And this then feeds back into how do you think about productivity, when selling is not easy. Anyone who has ever gone on the internet is aware that overwhelmingly the internet is driven by advertising. This is not unimportant in the functioning of the economy. It just happens not to exist in the Arrow-Debreu model or the models that basically build from that, unless they’ve put in explicit complications. So here’s Marshall again saying basically the same thing: The economy functions differently in good times and bad times. In good times the productivity measure is probably right, because you can sell all you can produce without a whole lot of change in prices. In bad times it’s a totally different story. Your focus is on the dynamics of the market in which you’re selling. And if you think it’s appropriate to use, as in many calibrated models, exactly the same calibration in good times and bad times, and if Marshall was right, then your model is leaving out something very important. The other thing that is left out are the things that the late Hyman Minsky talked about, which is the role of existing financial contracts. In the process of your incentives in the uncertain world of 'should you hire or should you produce more', And without tracking the financial position that firms and workers are in, you’re missing out on a big part of how the economy works. So I think the message here is a message that there’s lots of scope for very interesting work. And keeping an eye on what you want to learn from your research is an important part of how you design it. Thank you.

Peter A. Diamond (2014)

Unemployment

Peter A. Diamond (2014)

Unemployment

Abstract

Debates about higher structural unemployment occur when unemployment has stayed high. With monthly publication of the US Beveridge curve (the relationship between the unemployment and vacancy rates), the recent debate has focused on the shift in the Beveridge curve and whether the shift will be lasting long enough to move the full-employment point. The curve appears stable through the NBER identified business cycle through in June 2009 or possibly the month of the maximal unemployment rate in October 2009. This shift in the Beveridge curve, with the economy experiencing a higher level of unemployment than before for the same level of the vacancy rate, suggests a deterioration in the matching/hiring process in the economy. It is tempting to interpret this decline as a structural change in the way that the labor market works and thus assume that it is orthogonal to changes in aggregate demand. Indeed, an assumption that a shift in the curve is structural has been a staple of the academic literature since at least 1958. This interpretation has an obvious policy implication: however useful aggregate stabilization policies while unemployment is very high, they are likely to fail in lowering the unemployment rate all the way to the levels that prevailed before the recession, since the labor market is structurally less efficient than before in creating successful matches. This lecture reviews the theory underlying the Beveridge curve and US evidence on the ability to draw an inference of structural change from its shift or a shift in the hiring (matching) function

Content User Level

Beginner  Intermediate  Advanced 

Cite


Specify width: px

Share

COPYRIGHT

Content User Level

Beginner  Intermediate  Advanced 

Cite


Specify width: px

Share

COPYRIGHT


Related Content