The Trend is the Cycle

Nir Jaimovich (Duke University) and Henry Siu (University of British Columbia) appear to have made a very interesting discovery. Evidently, there appears to be a very strong link between two much talked about phenomena: job polarization and jobless recoveries. Their paper is available here:  The Trend is the Cycle: Job Polarization and Jobless Recoveries. 

 

Job Polarization

 

Job polarization refers to the recent disappearance of employment in occupations in the middle of the skill distribution. To display this phenomenon, the authors decompose employment into occupational groups and then delineate occupations across two dimensions: cognitive vs. manual, and routine vs. non-routine. (These labels are largely self-explanatory, but refer to the paper for details.) 

 

Evidently, these classifications correspond to rankings in the occupational income distribution. Non-routine cognitive occupations tend to be high-skill jobs, and non-routine manual occupations tend to be low-skill jobs. Routine occupations--both cognitive and non-cognitive--tend to be middle-skill occupations. Here is what has been happening to employment shares across these categories:

Percent Change in Employment Shares by Occupation Group 

The figure above shows that across three decades, the share of employment in the middle of the skill distribution appears to be disappearing. Prime suspect: routine biased technological change (e.g., think of ATMs replacing bank tellers). 

 

Jobless Recoveries

 

Jobless recoveries refer to the unusually slow rebound in the employment dynamic following the end of a recession (when GDP is growing). Here is the typical pattern one would have observed 30 years ago (and before): 

The x-axis is centered at "0," which represents the trough of the recession (using NBER dating). The data is plotted for 2 years around the trough date. The shaded region represents peak-to-trough. The y-axis plots the percent change in employment relative to its value in the trough. The figure above shows the rapid recovery in employment following the trough of the recession.

The dynamic above is to be contrasted with what has happened in the previous 3 recessions (early 90s, early 00s, and most recent). Here is what the picture looks like following the most recent recession:

Yes, but what's the link?

 

So far, this all very interesting, but not very new. What is new is how the authors link the two phenomena. 

 

The following diagram depicts the same employment dynamic, except with employment decomposed along three dimensions: [1] non-routine cognitive, [2] routine, and [3] non-routine manual (same as the polarization graph above). Here is what we see for the 1982 recession:

 

In this episode, the recovery in employment was strong across all occupation groups. In fact, the non-routine  occupations appear to have grown throughout the recession! The key is the strong recovery in the non-routine class of occupations. Roughly the same pattern is evident in the 1970 and 1975 recession as well. 

 

But now let's take a look at the more recent employment dynamic: 

 

 

Here, we  see hardly any movement at all in the non-routine occupations, but a significant and persistent decline in routine occupations. The relative weakness in routine occupations is evident in the 1991 and 2001 recessions as well. 

 

The conclusion is that jobless recoveries are due entirely to jobless recoveries in routine occupations. In this group, employment never recovers beyond its trough level, nor does it come anywhere near its pre-recession peak. This is in stark contrast to earlier recessions. 

 

The Trend is the Cycle

 

Consider now how the employment ratio behaves across these 3 occupational groups over the sample period 1967-2011. Here is the non-routine cognitive group:

 

 Here is the non-routine manual group:

And here is the routine group:

 

This last figure is quite dramatic. It shows how, prior to 1990, routine employment rebounded strongly following a recession. But since 1990, it appears not to rebound at all. Indeed, the pattern appears to be one of a precipitous decline in recession, followed by a period of relative stability in the subsequent expansion. 

 

Moreover, because these routine jobs are associated with the middle of the income distribution, the data here suggest that job polarization is not a slow, secular phenomenon--it is intimately tied with the business cycle. 

 

Theory

 

The authors employ a Diamond-Mortensen-Pissarides model to show how routine biased technological change can lead to job polarization, and how recessions can accelerate this process. The modeling framework is a good choice, in my view. (In particular, I have a hard time imagining how an IS-LM or NK model can be used to understand this phenomenon--but maybe I just lack imagination!) 

The work here is still very preliminary, of course, but the results look promising. Needless to say, it is hardly the last word on the subject. But I am confident that talented young economists, like Nir and Henry, will continue to shed light on the matter. Well done, gentlemen. Keep up the good work! 

Bloggers in St. Louis

Another eventful week at work (last week). Two coauthors in town (Fabrizio Mattesini and Randy Wright) and three seminars (Nicolas Trachter, Mario Crucini, and Fatih Guvenen). Well, four seminars, I guess. Narayana Kocherlokota was in town to deliver the Hyman P. Minsky lecture at Washington University. Oh, and Mark Thoma also gave an interesting seminar on how bloggers have helped (or possibly harmed) the nature of economic discussions/debates (see here). Fascinating stuff all around.

Mark was visiting the St. Louis Fed all last week (at my invitation). Of course, I knew that keeping him away from Steve Williamson was going to be a problem. And I was right. Here they are at the Kocherlakota event, with me trying to break up their fight:

My two favorite bloggers coming to blows

Things calmed down after I agreed to buy them both a beer; see here:

Best Buds

Ah, good times. But now...back to work! 

Turning wine into liquidity

The bible credits Jesus with having once turned water into wine. Nowadays, we get to witness the "miracle" of seeing wine turned  into liquidity: Wine Cache Rescues Those Short of Cash.

Some pretty interesting tidbits of information here. For example,

"You'd be amazed by how many wealth individuals have terrible credit ratings. And besides, if you go to a bank, it can take weeks or months to get a loan. When we make a loan, it's usually the same day,'' said Joran Tabach-Bank, head of Beverly Loan Co. 

It seems hard to believe that the wealthy individuals he refers to apparently do not have good relationships with their local bank (he includes bankers in this set!). But there you have it.

"Most people have a vision of pawn shops as sad sites. But that's not the case here," Taback-Bank said. "I have a lot of people who come in who have a business opportunity and they need an infusion of cash for business purposes," he said. 

Like the banker who can't get the cash loan he needs from his own bank?!

Of course, the business of transforming "illiquid" assets into "liquid" securities is as old as...well, it's as old as banking; see here. And now that pawnshops are muscling into the shadow banking sector, I wonder how long it will be before they too will be subject to regulatory oversight? After all, standard monetary theory predicts that assets that suddenly emerge as good collateral objects will be valued above their "fundamental" value; i.e., they will trade a a liquidity premium (which resembles a price bubble).

And, lo and behold! Do I detect a wine price bubble emerging out there?! (source)

I wonder if the Fed might consider expanding the set of securities acceptable at the discount window to include...um, no...probably won't happen.

Anyway, just having a little fun here before cutting out for the weekend.
Cheers!

Fiscal Policy Ineffectiveness in the Interwar Period

Gregor Smith forwards me this paper (coauthored with Nicolas-Guillaume Martineau) that estimates the impact of government spending growth on real GDP growth, using data from a cross-section of countries during the interwar period 1920-1939. Here is their abstract:

Differences across countries or decades in the countercyclical stance of fiscal policy can help identify whether the growth in government spending affects output growth and so speeds recovery from a recession. We use the heterogeneity in the government-spending reaction functions across twenty countries in the interwar period to identify this effect. The main finding is that the growth of government spending did not have a significant effect on output growth, so that there is little evidence that this central aspect of fiscal policy played a stabilizing role from 1920 to 1939.

As usual, a lot depends on the plausibility of the identifying assumptions employed.

The limitations of the data, in frequency and coverage, may prevent us from reaching a precise answer about the efficacy of fiscal policy, but it is still of interest to know whether that is the conclusion. Of course, the answer and its precision depend on an identification scheme. This paper adopts a new one: the main identifying assumption is that countercyclical fiscal policy could have worked in any country but was not tried to the same extent in every country. Identification relies on differences across countries (or over time) in fiscal reaction functions that capture the response of government spending to national income. We use these differences to estimate the effect of this government spending on the growth of income in turn.

The authors conclude (in a rather provocative and un-Canadian manner, I might add), that the evidence over this period fits better the infamous "Treasury view."

If you have some thoughts to share on their identification scheme and/or interpretation of their results, please feel free to comment. I'm sure the authors would appreciate your feedback.

The trend is your friend (until it ends)

My previous post advertising Jim Bullard's recent speech seems to have generated a lot of discussion; see Mark Thoma, Scott Sumner, David Beckworth, Tyler Cowen, Noah Smith and, yes, even Princeton Charming himself.

Most commentators are rather negative on Bullard's view that a permanent (persistent) negative wealth shock should be associated with a permanent (persistent) decline in the level of real GDP, leaving it's long-run growth rate largely intact. Some question the logic of Bullard's explanation, but it is not inconsistent with what happens in a standard RBC model where productivity follows a random walk with drift. I'm not sure if that's what Jim had in mind (I will find out in due course), but just thought I'd put it out there. (Alternatively, see Steve Williamson.)

What I would like to talk about here is my own take on the matter, which is I think is subtly different from Jim's. In my previous post, I made the following comment:

I think that Bullard makes a persuasive case that the amount of household wealth evaporated along with the crash in housing prices should likely be viewed as "permanent" (highly persistent) negative wealth shock. Standard theory (and common sense) suggests a corresponding permanent decline in consumer spending (with consumption growing along it's original growth path). The implication is that the so-called "output gap" may be greatly overstated by conventional measures.

I think that the crossed out part above was a mistake in light of the theory I had working in the back of my head when I wrote that paragraph. But otherwise, what I said is fine. As long as you understand the theory; which, of course, I should have explained. Let me do so here.

It is probably fair to say that when most people take a look at a time-series for real GDP, in their mind's eye they decompose the time-series into a linear trend and deviations from linear trend. It is a perfectly natural thing to do. But that doesn't make it correct.

Implicit in any decomposition is a theory. The common decomposition assumes that trend (or potential) GDP follows a smooth upward path. Trend is labeled "supply." Actual GDP (the thing we observe) obviously fluctuates around trend (something we do not observe). And since trend is "supply," it follows that actual GDP must be "demand;" and that cyclical deviations from trend (the output gap) are caused by "demand shocks." A lot of people seem to take all this as self-evident truth. Unfortunately, the "right" data decomposition is not as obvious as people sometimes like to believe.

What is another way to decompose time-series data? Personally, I find Jim Hamilton's regime-switching model an interesting way to interpret the pattern of economic development. The basic idea here is that growth is driven by productivity, and that productivity growth is subject to infrequent, but random and persistent, regime changes. (Regime changes are possible along other dimensions, of course.) Sometimes we are in a high-growth regime, and sometimes we are in a low-growth regime--a view not inconsistent with Schumpeterian growth dynamics. And while these growth shocks are not likely the only reason behind our cyclical ups and downs, this is the type of shock I had in mind when I envisioned the large negative wealth shock mentioned by Bullard.

In particular, as Joseph Zeira has shown (Informational Overshooting, Booms and Crashes, JME 1999), the switch from a high to low growth regime generates equilibrium asset price dynamics that any econometrician is likely to interpret as a price bubble that booms and then crashes. The price crash (and consequent loss of wealth) however, is driven entirely by economic fundamentals (I suspect that one could generate something similar in a model with multiple equilibria and self-fulfilling prophesies). I discuss this possibility in a bit more detail here: The 2005 Real Wage Shock. In this latter post, I raised the question of whether the apparent slowdown in real wage growth may have led property owners to revise downward their estimates of future rental income, precipitating the crash in real estate prices.

Now, maybe all this sounds a little crazy to you and, of course, perhaps it is. But it is interesting to note that some recent evidence on U.S. productivity growth, reported by James Kahn and Robert Rich, seems to corroborate my hypothesis: The Productivity Slowdown Reaffirmed (Liberty Street Economics, Sept. 2011). Here is a snippet from their opening paragraph:

Economists generally agree that productivity is the primary ingredient for sustainable growth in GDP and wages. The August productivity data release provided some clarification regarding trend--or long-run--GDP growth, but the news was not good: Following a resurgence of strong productivity growth in the late 1990s and early 2000s after nearly a quarter-century of slow growth beginning in 1973, the latest reading from a trend tracking model now indicates that slow productivity growth returned in 2004.

OK, so I was off by a year. ;)

I would like to mention some related empirical work by Marco Lippi and Lucrezia Reichlin, Diffusion of Technical Change and the Decomposition of Output into Trend and Cycle (ReStud 1994). The authors argue against modeling productivity growth as a random walk, suggesting that it makes more sense to think of technology diffusing according to an S-shaped dynamic.  Of course, the S-shaped dynamic gives rise to the low/high/low growth regimes I mentioned above. The authors conclude:

Thus we have found what might be called a dynamic tradeoff: either we assume rich dynamics for the cycle and consequently a trivial trend, or else we assume more complicated dynamics for the trend, consequently impoverishing the dynamics of the cycle. All intermediate cases are rejected by the data. 

Interesting, don't you think? At the very least, I think it suggests there is some room for different interpretations of the cycle and that the relative merits of these different interpretations should be the subject of an open and respectful debate (I believe that this was the main motivation for Bullard's speech).

Why is understanding the true nature of the decomposition important for monetary policy? Kahn and Rich provide us with one answer to this question:

It is widely believed that the difficulty of detecting a change in trend growth contributed significantly to the economic instability of the 1970’s, as policymakers were unaware of the slowdown in productivity growth for many years, and only much later were able to date the slowdown at approximately 1973. This resulted in overestimating potential GDP (at least so the conventional wisdom goes) and setting interest rates too low, and double-digit inflation followed not long after.  

It is not surprising to discover that the memory of that event weighs on the mind of some Fed presidents. Now, it happens to be my personal opinion that the inflation threat this time around is overstated (largely because this time there is a huge worldwide demand for USD and US treasury debt that is keeping inflation and interest rates low; see here). But what I, or anyone else, thinks is beside the point I am trying to make here. One of  the Fed's most important jobs is to keep inflation expectations anchored. History shows that inflation expectations can change suddenly and capriciously. Whether one likes it or not, it is the job of Fed presidents to think about this possibility, and to voice concern if they see a danger of repeating past policy mistakes.

If we are indeed entering into 1970s era of relatively slow productivity growth, then current CBO measures of the output gap are likely overstated, and further LSAPs are probably not warranted. This does not mean, however, that there is no output gap, or that there should be no policies directed to those who are having a difficult time in the labor market. As I discuss here, there is considerable variation in regional labor market conditions and it is not at all clear that "looser" monetary policy is the tonic we want to employ (assuming that it will have any effect at all in current conditions). In particular, there may be ample scope for regional fiscal policies, education and retraining programs, or other more direct measures that are outside the realm of monetary policy.

Update: Related Links

Bleak Apologists (The New Arthurian Economics)
The Asset Price Decline IS a Negative Productivity Shock (Canucks Anonymous)
Chucking the Solow Growth Model Cont. (Noahpinion)
The Output Gap: Still Negative (hjeconomics)

What output gap?

James Bullard

In case you haven't seen it, you may be interested in this speech given recently by Jim Bullard, president of the St. Louis Fed: Inflation Targeting in the USA.

This speech is really about how to interpret the recent performance of the U.S. economy. Is the conventional interpretation, that we are far below "potential" GDP owing to "deficient demand," the correct view? Or should we instead be thinking in terms of a large negative shock to "potential" GDP, with unemployment returning slowly to its natural rate, according to its normal dynamic (see here)?

I think that Bullard makes a persuasive case that the amount of household wealth evaporated along with the crash in house prices should likely be viewed as a "permanent" (highly persistent) negative wealth shock. Standard theory (and common sense) suggests a corresponding permanent decline in consumer spending (with consumption growing along it's original growth path. This part is incorrect given the model I have in mind here.) The implication is that the so-called "output gap" (the difference between actual and "trend" GDP) may be greatly overstated by conventional measures.

The view that one takes here is likely to influence what one thinks about monetary policy. The conventional view seems to support the Fed's current policy of keeping its policy rate close to zero far into the future. In his speech, Bullard worries that this may not be the appropriate policy if, in fact, potential GDP has experienced a level shift down (or, what amounts to the same thing, if conventional measures treat the "bubble period" as the economy being at, and not above, potential). Among other things, he says:

But the near-zero rate policy has its own costs.  If we were proposing to remain near-zero for a few quarters, or even a year or two, one might argue that such a policy matches up well with the short-term business cycle dynamics of the U.S. economy.  But a near-zero rate policy stretching over many years can begin to distort fundamental decision-making in the economy in ways that may be destructive to longer-run economic growth.  

Precisely how such a policy "distorts fundamental decision-making" needs to be spelled out more clearly (though he does offer a couple of examples that hinge on a presumed ability on the part of the Fed to influence long-term real interest rates). I am sure that many of you have your own favorite examples.

At any rate, I think this is a nice speech because it challenges us to think about the recent U.S. recovery dynamic in a different way. And if recent history has shown us anything, it's shown that we shouldn't grow complacent over what we think we understand.

Update available here: The trend is your friend (until it ends)

Update: The "terrifying" James Bullard offers a reply to Tim Duy here.

Using Beveridge curve dynamics to identify cyclical and structural shocks

I recently gave a short presentation to the Board of Directors of the Louisville branch of the St. Louis Fed. Following my presentation (which stimulated a lively discussion), I had the opportunity to listen to each member report on local economic conditions from different parts of Kentucky. Two themes stood out. The first was how "an air of uncertainty" along a variety of dimensions had "frozen" investment plans (with the apparent exception of younger entrepreneurs, who probably do not know any better-jk). The second was the unfilled demand for highly skilled, specialized workers (primarily in manufacturing).

I want to focus on the second theme here. In some sense, it is really amazing that firms are struggling to find qualified workers in an era of 8% unemployment. The Financial Times recently ran a piece on the subject: Skills Gap Hobbles US Employers, and I have to say that Mr. Greenblatt below would have fit right in at my BOD meeting:

Drew Greenblatt has been looking for more than a year for three sheet-metal set-up operators to work day, night or weekend shifts. 

The president of Marlin Steel Wire Products, a company in Baltimore with 30 employees, Mr. Greenblatt says his inability to find qualified workers is hampering his business' growth. "If I could fill those positions, I could raise our annual revenues from $5m to $7m," he says.  

He is offering a salary of more than $80,000 with overtime, including health and pension benefits. Yet in spite of extensive advertising,  he has had no qualified applicants. He is trying to train some of his unskilled staff but says none has the ability or the drive to complete the training. 

This quote identifies two problems. The first is what economists call "skills mismatch" caused by a "structural" shock. The second, that some workers are unwilling and/or unable to upgrade their skills is another matter that deserves attention, but is something that I will leave aside here. 

  
Apart from anecdotal evidence, how does one go about measuring "skills mismatch caused by structural shock?" One idea, initially proposed by Abraham and Katz (JPE, 1986), is to use the comovement to in vacancy and unemployment rates to identify "cyclical" and "structural" shocks. 

I put those terms in quotes because there are no set definitions for them. I like to think of a cyclical shock as an event that makes it more or less profitable to find the same kind of worker for the same kind of job. And I like to think of a structural shock as an event that makes it more or less profitable to find a different kind of worker for a different kind of job. 

Anyway, the Abraham and Katz idea is that one would expect cyclical shocks to trace out a stable, negatively-sloped Beveridge curve. That is, one would expect the job-vacancy rate and the unemployment rate to move in opposite directions.  A structural shock, by contrast, is expected to move vacancy and unemployment rates in the same direction. The idea here is that it is now more difficult to find the right kind of worker, so that even greater levels of recruiting intensity is likely to be associated with higher unemployment rates. 

The FT article cited above uses this idea in the diagram to the right (together with the results of a Kaufman poll of entrepreneurs) to suggest that the high U.S. unemployment rate is primarily the consequence of "structural" factors. 

Here is what the U.S. Beveridge curve looks like from May 2005 - November 2011 (The vacancy rate is computed from the Conference Board's help-wanted-online data, which is available from 2005 only).

As the HWOL measure of job vacancies is available at the city level, Constanza Liborio and I thought it might be interesting to see how job availability varies across major U.S. metropolitan areas and how job vacancy rates correlate with regional unemployment rates before and after the beginning of the most recent recession.

Specifically, the exercise we perform is as follows. Consider a major U.S. metropolitan area. Compute the average job vacancy rate and unemployment rate for this metropolitan area over the prerecession period May 2005 – November 2007. Recalculate these averages since the beginning of the last recession, December 2007 – November 2011. Next, compute the change in the vacancy rate and unemployment rate across these periods. Perform this exercise for a set of the largest metropolitan areas in the U.S. 

The results are displayed in the following figure.

Not surprisingly, we see that the unemployment rate in all these metropolitan areas went up since the recession began.  However, the same is not true of job vacancy rates (that is, not all vacancy rates went down, as one might have expected). Specifically, while we observe the typical Beveridge curve dynamic in many jurisdictions (suggesting that cyclical factors are dominant), we also observe vacancy rates remaining relatively stable, or even rising, in several others (suggesting that structural factors are dominant). 

So the tentative conclusion here is that the relative importance of cyclical vs. structural factors appears to vary across regions. To the extent that monetary policy is an effective stabilization tool, it cannot be expected to impact all regions of the country equally. In many regions, localized fiscal policies (education and training subsidies, etc.) may prove to be a more direct and effective tool.
Related story:
More Workers Moving for Out-of-State Jobs