Believe those who are seeking the truth. Doubt those who find it. Andre Gide

Tuesday, November 23, 2010

The 2005 Real Wage Shock

In the course of preparing for my discussion of Rob Shimer's paper (see my post here), I had my RA (the tireless Constanza Liborio) dig up some aggregate wage data for the U.S. economy. Let me preface the discussion that follows by saying that I am wary of putting too much stock in aggregate wage data (the composition bias, in particular, is potentially a big problem; see here).  O.K., with this caveat in mind, let's take a look at some data.

As a measure of real wages, I use the BLS Employment Cost Index. Evidently, this measure is preferred by the likes of Bob Hall and others because it includes non-wage benefits. In what follows, I examine quarterly data for the sample period 1990:1 - 2010:3. The following chart plots the (annualized) rate of growth in nominal wages. The red line is a five-quarter rolling window I use to smooth out the series. The shaded areas represent NBER recession dates.

Prior to the most recent recession, nominal wages grew on average by about 3.5% per annum. The data shows a significant deceleration in nominal wage growth during the last recession. The composition bias suggests that actual wage growth displayed even greater "flexibility," as unemployment is typically concentrated among lower wage workers.

As I want to construct a measure of real wages, I need some measure of inflation. To this end, I use the GDP deflator, which is plotted in the next diagram.

Prior to the most recent recession, this measure of inflation averaged just above 2% per annum (the Fed's implicit inflation target). There was, however, a notable run up in the early 2000s, with (trend) inflation peaking at just over 3% in 2005 and early 2006. It is interesting to note that the rise in inflation over this period occurred while nominal wage growth decelerated. The next diagram plots the growth rate in real wages.

Now, the focus of Shimer's paper was apparent "ridigity" in real wage growth during the recent recession; a development that he interpreted as explaining the recent anemic behavior in employer recruiting intensity. As for myself, I was rather struck by the rapid deceleration in real wage growth in 2004, leading to falling real wages in 2005.

I want to take this data at face value for the moment and speculate a bit on what role these wage developments may have had in bursting of the U.S. home price "bubble" in 2006.

The story I have in my head revolves around an idea I first saw exposited by Joseph Zeira in his fine (and much under appreciated) paper: Informational overshooting, booms, and crashes

The basic idea in Zeira's paper is as follows. Imagine an asset whose dividend grows a H% per year. Everyone knows that this growth will one day come to an end. When this date arrives, the dividend grows at L% per year forever (a simplifying assumption), where L < H. The only uncertainty in this thought experiment pertains to the date of the "regime change."

Zeira demonstrates that the equilibrium (rational expectations) asset price rises over time, and continues to rise as long as dividend (read: real wage) growth expectations continue to be met. Then comes the shock.  I am tempted to call this a Wile E. Coyote moment, but of course, everyone in this model--unlike that hapless desert dog--knows that there is a date of reckoning. They just don't know beforehand when it will happen. So what happens?

Naturally, the asset price plummets like stone cast from heaven, before settling down along its new "fundamental" value (reflecting a new era of diminished expectations...gosh, I'm sounding a lot like PK these days). It appears as if asset prices "overshoot" their long-run fundamental value, before crashing.

To an outside observer, the asset price dynamics just described may be interpreted as a typical "irrational" boom and bust cycle (perhaps justifying some form of financial market regulation). In the context of the model world just described, this interpretation is completely wrong. This type of boom bust dynamic is, in fact, the natural consequence of how information is priced in an efficient asset market.

The picture I have in my head then is the following. Real wage growth appears relatively robust over the late 1990s and early 2000s. The return to labor, perhaps more than any other variable, measures the capacity for the average household to service debt. In the first half of the 2000s, creditors are looking at a recent history of relatively robust real wage growth, justifying credit expansion (even into subprime). By 2005, however, evidence of flagging fundamentals (anemic real wage growth) led to a (rational) revision downward in the real wage growth regime. Credit supply and real estate prices soon began to reflect this change in economic fundamentals.

Anyway, that's my crazy idea for the day. Feel free to share your thoughts...


  1. Hi David,

    I still wonder how much better (in levels of wage growth, and expected duration of the high growth regime) the late 90s-early 00s regime must have been to create such a large effect on credit supply and real estate prices. In your data it seems to have lasted at best 8 years. If one takes a longer time series and tries to pin down the probability of regime switch (both from bad to good and good to bad), I am not sure the resulting process explains the size of the changes in the credit and housing market. We might need an additional amplifying mechanism, that could be (more) important in itself...

  2. Anonymous: Yes, you are right to wonder about the quantitative significance of this story. One might need a very pessimistic expectation of the probability of returning to the high wage growth regime. And in any case, this is surely not the only thing that impacted on the housing boom and bust cycle we recently witnessed.

  3. Professor,

    First commenter, long-time reader. You write about the Wile E. Coyote moment when asset prices come tumbling down, and I can understand how the mechanism works and how there should clearly be a contingency for creditors for this type of situation. But, I think it a reasonable assumption for creditors to expect this shock to be asymmetric such that a negative shock in Florida would be balanced out by stability in the Dakotas. What do you think?

  4. Hamed: Good to hear from you. I do think that creditors got lulled by the belief that U.S. house price dynamics were largely a local affair, with the offsetting behavior you alude to. To the extent that this was the historical experience, it appears not to have played out this time around. It was a big shock.

  5. Hi David,

    Why do you choose to deflate wages by the GDP deflator? My understanding is that the deflator measures the inflation of prices of domestic components.

    Employers and workers do not consume only domestic products. If they were to purchase imported oil which may be inflating, then the GDP deflator would understate inflation. And on the other hand, if they are purchasing flat panels that are deflating, the opposite would be true. I don't say this to lecture you, just to give some background for my comment.

    Do you have any thoughts about whether your conclusion would change if the deflator was CPI/PPI?

  6. Chris: The thought of which price index to use had occurred to me. I asked my RA to compare the GDP Deflator, CPI, and PCE (not PPI) and she reported back to me that the three indices displayed roughly the same broad movements. So I doubt whether my real wage measure would change very much if I adopted a different index but, of course, I would have to check to make sure that this is true.

  7. I'm late to the party, following a chain of links from the current discussion of Bullard's view on potential output.

    David, the asset pricing phenom. you describe is well-understood and much older, going back to to at least the early 1990s. That literature, mostly on "peso problems" noted explicitly the observation equivalence between bubble-based and fundamental-based explanations of asset price dynamics. (Martin D.D. Evans and Enrico Perotti made important early contributions here, as I recall.)

    My reading of the empirical literature that followed was that it split into two camps. Camp A has a very low prior weight on the "bubble" model, and usually uses the dynamics of prices to infer the dynamics of unobservable expections. Camp B tries to directly model fundamentals or measure expectations, finds that they don't seem to have the dynamics that would support the fundamental interpretation, and identifies the price behaviour as a bubble.

    Would you say that characterization is about right?

  8. Simon,

    Thanks for alerting me to this. You wouldn't happen to have a handy reference describing the Peso problem and the subsequent division into camps A and B, would you? I don't know enough myself to say whether your characterization is right, but I am definitely going to do more reading to find out. Thanks again.