Labor Markets and the Macroeconomy: Theory, Evidence and Policy Implications


  • Time and Date: 10:00-16:55, Friday December 19, 2008 (Open: 9:45 a.m.)
  • Venue: RIETI's seminar room (1121,11th Floor, METI ANNEX)
    1-3-1 Kasumigaseki, Chiyoda-ku, Tokyo
  • Language: English (no interpretation available)


Session 1: "The Ins and Outs of Unemployment: A conditional analysis"

Fabio CANOVA (ICREA Research Professor, Universitat Pompeu Fabra / Research associate, CREI / Research Fellow, CEPR)

This paper re-explores what causes movement in unemployment during recessions. The conventional wisdom holds that recessions are periods of high unemployment that begin with layoffs and persist over time due to difficulty in finding new jobs. Hall and Shimer go against this logic suggesting that while the job finding rate fluctuates over the business cycle, the job separation rate is acyclical.

Their analysis however has interpretation problems: What drives fluctuations in finding rates? What is the direction of causality? Could movements in separation rates drive fluctuations in finding rates? Do conclusions hold true for important business cycle shocks?

This paper brings new evidence on these issues. The dynamics of unemployment in technology-induced recessions are analyzed. Intensive margins and extensive margins are looked at. Also, unemployment dynamics in terms of job separation and job finding rates are characterized.

The two types of technology shocks considered in this paper induce very different dynamics in the labor market. Investment neutral shocks have a large effect on unemployment in the short term and affect labor markets primarily through the extensive margin. Investment specific shocks primarily affect aggregate hours and seem to go through the intensive margin of the labor market. As for finding and separation rates, their dynamics agree with the conventional wisdom. Neutral shocks can explain jobless economic expansions, such as the recovery of the early 1990s.

These results challenge the sticky price explanation for the relationship between technology shocks and hours. When technology improves and monetary policy is not sufficiently accommodative, demand is slow to respond and firms take advantage of technology improvements to economize on labor input. As changing labor is more costly than changing prices, this adjustment naturally applies to the intensive margin. However, the extensive margin is the big driver after a neutral technology shock. The fall in hours is related to the reallocation of workers across jobs. These dynamics are consistent with the Schumpeter's model of creative destruction where a neutral shock leads to a shedding of old technology and a reallocation of workers from old technology to new technology.

This paper is empirical and looks at U.S. aggregate data. A simple vector autoregressive model that includes six variables was used. The variables are (1) the price of investment relative to the price of consumption, (2) labor productivity, (3) total hours, (4) the unemployment rate, (5) the finding rate, and (6) the separation rate. Technology shocks are identified with the same long run restrictions as in Fisher and in Michelacci and Lopez-Salido. Approximate and exact rates are used for both the finding and separation rates. As the data spans 30-40 years, structural changes, low frequency movements and other noncyclical factors must be taken into account. Low frequency movements are particularly important in this analysis. If care is not taken to filter them out, the results will turn out to be biased.

Since the data runs from 1955 to the present decade, allowances have been made for breaks in the intercept in 1973 and 1997. This has been done while maintaining the dynamics of the model unchanged over the full sample. This strategy was chosen because using subsamples, for example, would have created identification and estimation biases.

It is found that when a neutral shock that increases productivity occurs, there will also be a reduction in hours. As a result of this, unemployment increases and remains high for a long period. The change in hours per employee is not significant and most of the dynamics are driven by the extensive margin.

The dynamics in finding and separation rates are important to identify the driver behind the changes in the unemployment rate. The separation rate is particularly important in the impact period, but it converges to normal very quickly. The finding rate, on the other hand, is responsible for the rise in unemployment over the medium run

Moving to investment shocks, a shock that causes a drop in the price of investments increases both hours and output, and this makes it look like a more traditional kind of technology shock. This shock has almost no effect on the unemployment rate. The jump in total hours is due to changes in hours per employee. Hence, these shocks tend to exert major effects on labor markets through the intensive margin. The insignificant effect on unemployment is due to the fact that both the finding and separation rates move approximately by the same amount but in the opposite direction. Shimer's approximate and exact rates are plotted. Whether approximate or exact rates are used makes very little difference on the results.

There are three main reasons why this analysis yielded such different results from Shimer's. First, this analysis considers dynamics conditional on certain shocks. Second, this analysis measures the effects on both finding and separation rates on impact and over the adjustment path, rather than on average over the business cycle. Third, feedback among the variables of the model is permitted.

The analysis gives to the separation rate an important role in the adjustments. It is therefore important to check whether the exclusion of other labor market flows could be important. For this reason, a simple, two-stage labor market model is employed. Using the responses of both finding and separation rates, the fictional unemployment rate can be constructed. This rate will allow us to study whether flows in and out of the labor market are important.

On impact, the responses of the actual unemployment and of the fictional unemployment differ when neutral shocks are present, and this indicates that there may be movement in and outside of the labor force. This is not the case with investment-specific shocks. Nevertheless, in both cases the separation rate is crucial to understand how technology shocks affect the unemployment rate.

It is important to ascertain whether neutral and investment-specific shocks are important sources of variability in the labor market. At almost all horizons, neutral shocks do not seem to be a significant driver of total hours, but they explain a sufficiently large percentage of the fluctuations in the unemployment rate. This is the opposite of what investment-specific shocks do. These shocks explain a large portion of the variability of total hours and hours-per-worker while the percentage of the fluctuations in the unemployment rate explained by these shocks is very small.. Taken together, technology shocks explain 20%-40% of the variability in labor market variables.

It is also interesting to see how important these shocks are for specific business cycle episodes, particularly the recession of the late 1980s and the slow labor market growth of the early 1990s. It turns out that the neutral shock of 1990 explains the jobless expansion of the early 1990s, and the component of the data due to technology shocks tracks quite closely to the raw data. This fact is consistent with a model of creative destruction, where technology shocks tends to shed jobs in low-productivity sectors.

A sensitivity analysis is undertaken to make sure the results were not dependent on auxiliary assumptions. The paper checks for omitted variables or omitted shocks, for VAR lag length, for alternative identification methods, including medium-run and sign restrictions, for the choice of price deflators for the price of investment and GDP, and for use of alternative data sets.

Several conclusions can be drawn from the analysis. First, mixing neutral and investment-specific shocks may be problematic. It is very important to keep them separate, especially to understand how technology shocks affect the labor market. The dynamics they induce are very different. Second, the distinction between the extensive and the intensive margin is important to understand how technology shocks are propagated in the economy. Third, the drivers behind movements in the unemployment rate are very much consistent with the conventional wisdom. In other words, finding and separation rates are both important and the tendency to use theoretical models with an exogenous separation rate should be avoided.

Finally, neutral shocks may explain the cyclical fluctuations in unemployment we have observed in the last few years. This may be especially useful when analyzing Japan in the 1990s and when trying to understand whether certain types of shocks could explain the dynamics of macroeconomic variables

Question and Answer Session

Q. Does the q you use in the model stand for Tobin's q ?

No, q in this case is the price of investment relative to consumption.

Q. How did you come up with this variable?

The problem in constructing this series is that investment changes over time. Adjustments must be made for the quality of investment. Several researchers have worked on this theory in its beginning and updating phases. We stopped the series in 2000 because there is no official adjustment for quality after that date.

Q. Are you suggesting that resources regarding the impact of the identified neutral shock can be explained in terms of the model of creative destruction? Why is the last response one gets after job separation in a neutral shock consistent with the sticky price model?

I agree with the use of the creative destruction model. Neutral shocks appear to have this kind of effect. A model which is consistent with the evidence presented in the paper is presented in complementary work (see Canova, Lopez Salido and Michellacci)

Q. How conditional are your results to labor market movements, especially in terms of analyzing Japan rather than the U.S.? And what was the productivity variable used?

The major problem with using data from other countries is that it is very difficult to find consistent series for the finding and separation rates. Most researchers look at U.S. data because it is very easy to find the appropriate numbers to study.

Regarding labor productivity, the use of total factor productivity (TFP) as a variable was considered. The problem with TFP was that typically capital series are reconstructed from investment. Therefore, results have to be taken with care when this variable is used.

Q. Would you agree that in order to assess the sticky price hypothesis, one could look at the response of either hours or employment in the capital goods sector to an investment-specific technology shock?

I do not have data on hours or employment by sectors. If these were available such a test could be undertaken,

Q. Why is the response of hours so strong relative to what others have found in the literature?

To measure the response of hours, it is important to filter out low-frequency movement. If low-frequency movements are not filtered out, the response of hours tends to be much smaller due to the co-movement of labor productivity and hours. Separating samples is something of a shortcut to accomplishing this task, but it shows clearly what happens.

Q. What relevance does this study hold for dealing with the current economic crisis?

This research tells us that if the current situation was driven by technological change, one of the two scenarios analyzed in the paper could apply. If, for example, the traditional automobile industry folds and a new fuel-efficient technology emerges, the results we obtain after a neutral shock may be more appropriate for predicting what may happen in the future.