Tim Duy and Menzie Chinn have been trying to determine whether the recent employment report indicates that employment is growing above trend or whether it shows signs of weakness in the economy, a task that is complicated by uncertainties over how demographic changes affect labor force participation rates and about whether the recent decline in productivity growth is permanent or temporary:
Jobs, Productivity, and the Fed, by Tim Duy: Menzie Chinn at Econbrowser has an excellent piece taking another look at the employment report in which he presents a model to explain the consensus forecast for March job creation. The motivation for the model stems from a Deutsche Bank report detailing the lagged response of the labor market to economic activity:
Hence, I estimate a regression of the q/q change in quarterly payroll employment growth on one lag of the change in real equipment and software investment and three lags of real GDP over the 19671-07q1 period. This leads to the following regression result:
dnt = 0.0002 + 0.045(dequipt-1) + 0.197(dyt-1) + 0.173(dyt-2) + 0.126(dyt-3) + ut
Adj. R2 = 0.53, SER=0.0035, n=157
Where dn is the first difference in log payroll employment, dequip is the first difference in log real equipment investment, and dy is the first difference in log real GDP. All coefficients except the constant are statistically significant at the 10% MSL, using Newey-West standard errors for inference
Menzie uses this model to explain why the March forecast of 135k gain in NFP was reasonable, and why an average 70k monthly gain over the next three months is also reasonable.
Two things leapt to mind when I read this post. First, this is a neat idea. Second, can we use it to get at the question of the productivity slowdown? To simplify the analysis, I used 1983:1 as a start date, acknowledging a structural break in the data in the early 1980s reported by many practitioners. Then I create three models. The first is simply Menzie’s model over the period 1983:1-2006:4, which I call the productivity agnostic model. The second is a productivity optimist model that includes a dummy variable beginning in 1995:1 to account for the productivity boost delivered by the IT boom. The third is a productivity pessimist model, in which the dummy variable identifies the period 1995:1 – 2005:1. This assumes the productivity boom fades beginning in 2005. The resulting regressions are:
dnt = -0.0009 + 0.0241(dequipt-1) + 0.285(dyt-1) + 0.220(dyt-2) + 0.110(dyt-3) + ut
Adj. R2 = 0.63, SER=0.0023, n=96
dnt = 0.0001 – 0.002(dummy)+ 0.0289(dequipt-1) + 0.269(dyt-1) + 0.209(dyt-2) + 0.101(dyt-3) + ut
Adj. R2 = 0.67, SER=0.0022, n=96
dnt = 0.0 – 0.002(dummy)+ 0.0288(dequipt-1) + 0.270(dyt-1) + 0.209(dyt-2) + 0.108(dyt-3) + ut
Adj. R2 = 0.67, SER=0.0022, n=96
With the exception of the constant term, all coefficients are significant at the 10% level. Compared to Menzie’s long sample, the shorter sample shows a greater importance of lagged GDP on employment. The coefficient on the dummy variable is negative as expected; higher productivity growth suggests lower job growth holding output growth constant. The coefficients in the latter two models are virtually identical. This is not surprising; the dummy term turning off in 1995:2 does the work. I use Menzie’s estimates of 2% annualized 1Q07 GDP gain and real equipment and software decline of 2.5% y-o-y to generate a dynamic forecast of nonfarm payrolls for the first two quarters of 2007. Estimating the model through 2006:4, and then forecasting nonfarm payrolls for the first two quarters of 2007:
In 2007:1, the economy added 513k jobs compared to the previous quarter (note that I am using quarterly averages, not the average of the monthly gains). The expectation given the productivity agnostic model was a 310k gain; interestingly, this works out to 100k a month, or close to the Fed’s estimate of NFP growth consistent with a stable unemployment rate and similar to Menzie’s results.
The productivity optimist would be expecting a gain of 189k jobs as the economy had just experienced three quarters of growth well below the trend we became accustomed to in the late 1990s. Of course, this prediction could not be made given the stability on initial unemployment claims throughout the quarter and, more importantly, the pervious quarter. In contrast, the productivity pessimist would have been looking for a 403k gain, closer to reality but still 100k less than the actual gain.
The real story is the current quarter. With another weak GDP report anticipated in 2007:1, expected job growth is between a meager 98k gain to a 313k gain, with the productivity agnostic model in the middle at 229k, virtually the same as Menzie’s estimate of 70k per month. The implication is that solid job growth (roughly 100k per month) in the current quarter would suggest that potential GDP growth has shifted down substantially, to as low as 2.25%. Perhaps this is simply a reflection of a late-cycle productivity slowdown; perhaps an indication that trend productivity growth is slowing. Discriminating between these two options is the challenge for policymakers; from the minutes of the most recent Fed meeting:
Most participants continued to expect that core inflation would slow gradually, but the recent readings on inflation and productivity growth, along with higher energy prices, had increased the odds that inflation would fail to moderate as expected; that risk remained the Committee’s predominant concern….
…Participants expected that productivity growth would pick up as firms slowed hiring to a pace more in line with output growth but acknowledged that the improvement might be limited, particularly if business investment spending were to remain soft.
To me, the second quote sounds as if the Fed expected the March employment figures to come in considerably softer, closer to 100k. They also must be surprised that the unemployment rate dipped to 4.4% - the central outlook for the fourth quarter of this year is 4.5-4.75%. I think the Fed does see indications that the economy slowing. And I am sure they understand the implications of the drop in durable goods orders. They are trying to resolve this weakness in other areas with the relatively strong job market and persistent inflation. The problem the Fed faces is that any shift in the trend rate of productivity growth will only be known with certainty in hindsight, an issue reiterated by Governor Mishkin earlier this week:
In particular, over the past few decades the natural unemployment rate and the path of potential output have apparently moved around quite substantially. If we do not recognize the potential for such shifts, they can pose serious pitfalls for the conduct of monetary policy.
A slowdown in trend productivity growth helps explain the combination of low economic growth coupled with a stable to falling unemployment rate and persistent inflation. Moreover, the possibility of such a slowdown suggests a higher bar for a rate cut. Note, however, that a slowdown in productivity growth does not preclude an aggressive policy response to a substantial deterioration of activity. Mishkin also noted:
Finally, central banks should respond aggressively to output and employment fluctuations on those (hopefully rare) occasions when the economy is very far below any reasonable measure of its potential. In this case, errors in measuring potential output or the natural rate of unemployment are likely to be swamped by the large magnitude of resource gaps, so it is far clearer that expansionary policy is appropriate. Furthermore, taking such actions need not threaten the central bank's credibility in its pursuit of price stability.
This means that when we are in a recession, we are well below potential growth, regardless of the measurement uncertainty.
Regarding that recession, a productivity optimist would be expecting the labor market is going to show substantial weakening this quarter and, considering the reduction in GDP forecast this year, likely the next, consistent with my read on the yield curve. (Does this suggest that most market participants are productivity optimists?) In this case, one could expect a rate cut in the late summer or early fall. Watch initial unemployment claims; under this scenario, they should start rising rapidly. On that note, this week’s read was higher than expected, but we need to take a single data point with a grain of salt, as there are challenges seasonally adjustment weekly data. Be patient; when claims start moving, they move fast.