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

Julen ESTEBAN-PRETEL (Assistant Professor, the University of Tokyo)

The Japanese economy, which GDP per capita prior to 1990 grew at a faster rate than the 2% of the world's leading economy, the U.S., entered a decade-long recession in 1991. Unemployment, which was fairly stable in the 1980s, began increasing in 1991 and more than doubled to 5.4% by 2002. This increase in unemployment could have been caused by different patterns in the worker flows. This study provides evidence on these flows of workers in Japan over the 1990s, and shows that the probability for a worker to leave unemployment went down, whereas the probability for a worker to lose his job increased. Despite the opposite trends in these probabilities, both the number of workers losing their job and the number of unemployed people finding job increased over this period. It was the larger increase in the amount of workers losing their job that increased unemployment.

At the same time that de-trended output dropped and unemployment increased, the government introduced a series of regulations increasing the number of holidays and closing government offices on Saturdays, which brought about a reduction in the average weekly hours of work from 42.7 to 38.5 over the course of the decade. Over the same period, total factor productivity (TFP), which had grown at a fast pace prior to 1990, suffered a growth slowdown after the turn of the decade, and it only started to recover after 2001.

This study assesses how much of the increase in unemployment and the labor market flows can be accounted for by the slowdown in the growth rate of TFP. It also explores the role of other factors, such as the decline in hours worked. A neo-classical growth model with search frictions in the labor market is built. It is calibrated to match the Japanese economy in 1990, and then it is simulated using the empirically observed movements in TFP, hours worked, and government expenditure as the driving forces of the economic changes. The results are then compared with actual data to assess the performance of the model and role of the different factors.

The model is a Cass-Koopmans type model with the labor market modeled in the style of the search and matching literature with endogenous job destruction. There are three types of agents: (1) households, who consume, save and supply labor to firms; (2) firms, which hire labor and rent capital to produce output; and (3) the government, who collects taxes to finance its own expenditures..

In terms of the household, the economy is composed of a big family. This assumption delivers perfect self-insurance in the model. Firms are owned by the Family, and their profits rebated back to the household. Every period, the Family chooses how much to consume and save to maximize lifetime utility subject to its budget constraints.

In the labor market, unemployed workers and vacancies meet randomly according to a CRS matching function, and try to form employment relationships, which consist of one worker and one firm. When matched, a firm produces output with the available technology, TFP, and using two inputs: labor (in the form of hours), and capital. Destruction of matches is endogenous and depends on the level of the non-productive intermediate input costs, which are idiosyncratic to each firm. There is free entry of firms, which in equilibrium determines the number of vacancies being posted. Wages are determined as the Nash solution to a bargaining problem, which implies that worker and firm get a constant fraction of the surplus.

The government taxes labor and capital income, as well as the general income of the Family in the form of a lump-sum tax. It uses its revenues to finance the government expenditures, and has a balanced budget every period

When determining the values of model parameters, a subset of them are chosen following other studies in the literature, or getting them directly from the data. The remaining parameters are calibrated to match the empirical evidence for Japan in the 1990s.

Once the model is parameterized, the exogenous movements of the three variables are fed in the model to perform the simulations. The exogenous variables are the growth rate of TFP, hours of work, and government expenditures. The evolution of these variables is as in the data from 1990-2002, and are kept constant at the 2002 level after that year. The solution method used is a perfect foresight shooting algorithm .

The model reproduces well the time path of unemployment and output from 1990-2002. It is found that TFP alone accounts for 50% of the change in unemployment by 2002. As for the flows of workers, the probability for an unemployed worker to lose his job increases in the model, as is observed in the data. The probability of finding a job decreases in both the model and the data, although the model underestimates the overall level decline. The model can also reproduce the empirically observed increase in the total number of workers flowing from employment to unemployment, and vice versa. Finally, de-trended wages, which in the data only started to decrease after 1996, stars decreasing in the model after the turn of the decade.

The intuition for the model results is as follows. The continuous drop in TFP growth, produces a decline in de-trended productivity, which, together with the fact that firms cannot make their employees work longer hours due to the government imposed decline in hours, reduces the output and profits of the firms. This in turn reduces the incentives for firms to post vacancies and increases the willingness to fire workers. These two effects imply a reduction in the probability for an unemployed worker to find a job, and an increase in the chances of an employed worker to lose his job. The increase in total unemployment driven by the previous changes implies that, even though the probability for an individual unemployed worker to find a job decreases, the total number of workers flowing into employment increases. The fact that wages start decreasing faster in the model than in the data is due to assumption of Nash bargaining for wage determination, which implies that wages are highly correlated with productivity and output, which in the model start decreasing right away.

In conclusion, this paper builds a neo-classical growth model with search frictions in the labor market.. It finds that a decrease in TFP growth accompanied with a decrease in hours can account for a large part of economic behavior in Japan in the 1990s with respect to output, unemployment, the creation and destruction of jobs, and to a lesser extent wages.

Q. People outside of Japan are not completely aware of the regulatory changes that have been made to working hours in Japan. It would be interesting to show how much can be achieved through changing the regulation of hours by holding TFP constant and observing the effect the change in hours can have on these patterns. This is important because many people are not completely aware of the changes made in workweek regulations.

While the study does not aim to do this, taking a look inside TFP to see what exactly happened would be interesting. The adoption of technology during the 1990s in Japan dropped dramatically relative to other Asian countries. Could there be some room in the model to give more structural interpretation of this TFP movement?

Julen ESTEBAN-PRETEL
The result of the change in hours was not explicitly examined. However, as stated before, TFP accounts for half of the increase in unemployment. Therefore, the decline in hours roughly accounts for the other half. In this model, hours and TFP complement each other.

The exercise we perform imposes the TFP changes as exogenous. As for the sources of the decline in TFP growth, there could be many. Technology adoption could have played an important role. In the credit crunch that followed the bubble burst, firms lost the ability to garner funds to invest in new technologies. Some authors have argued and changes in productivity in Japan are highly correlated with research and development in the U.S., but R&D in the U.S. did not decline over this period. However, in order to use these new technologies, Japanese firms require ample credit to import and implement them. Introducing technology adoption into the model would highly complicate the model, and since explaining the sources of the TFP decline was not the objective of the paper, it is left for future research.

Q. When reproducing the time path of output, if you feed in TFP and hours worked, is there much left to consider? How would the model look if the hours were not fixed and had to be chosen instead? That would be a much more revealing analysis.

Some of the literature uses cyclical variations in unemployment starting with the work of Shimer and Hall. There is a new paper suggesting that if the value of unemployment is very high, and the firm's bargaining power is very high, the cyclical features of unemployment can be reproduced. It would be interesting to see what happens and how that model performs in terms of reproducing the more long-run features that were assessed in this study. Furthermore, when the bargaining power of the worker equals the elasticity of matching with respect to unemployment, as is the case in your study, the decentralized equilibrium is efficient. When moving to the alternative calibration, this is no longer the case.

It would be interesting to look into the TFP a little bit more. In matching models, a key variable is the degree of goods market regulations. When firms are heavily protected by regulations, they choose to start updating at slow rates, which creates unemployment. The higher the growth rate of the frontier, the larger this problem becomes. Since the mid-1980s, the growth rate of the frontier has increased.

Julen ESTEBAN-PRETEL
A previous version of the model had endogenous hours, but the simulations delivered an increase in hours after the drop in technology, since within the model firms want to compensate for drop in technology by working their employees harder. However, we dropped the endogeneity of hours from the model since we believe that the empirically observed decline in hours was exogenously imposed by the changes in regulations by the government.

The way calibration is done, the fraction of the productivity of the worker that determines the value of unemployment is the same as in Shimer (American Economic Review, 2005), that is 40%. However, since our model has capital, this implies that the flow profit of the firm, net of wages, is much lower than that of Shimer (2005). In fact, it is very close to the alternative calibration proposed by Hagedorn and Manovskii (European Central Bank Working Paper, 2008) to make the model more responsive to productivity changes. When capital is removed from the model, the change in TFP does not produce much of a change in unemployment.

As stated before, it would be very interesting to know the forces underlying the movements in TFP over the Lost Decade, but it is beyond the scope of the paper.

Q. In the model, the employment rate is one minus the unemployment rate, but in the data, the labor market participation rate fluctuates over time. Was labor market participation included in this model? A starting point would be to introduce exogenous fluctuations in the labor market participation and later consider endogenous participation.

Julen ESTEBAN-PRETEL
In Japan, the labor force participation decreased by 2% over the decade to reach 61% in 2002. This small decrease in the labor force could be imposed, but our guess is that it would not change the results very much.

Q. Please explain why the model assumes that there was perfect foresight in 1990.

Regarding steady states, the equilibrium paths may vary depending on which steady states are applied. Given this possibility, how can steady states be determined?

Julen ESTEBAN-PRETEL
The solution method uses perfect foresight since this is the standard assumption when using shooting algorithms. Furthermore, Hayashi and Prescott (Review of Economic Dynamics, 2002) use this method, which makes our results are comparable to theirs. Other papers in the literature have studied the robustness of their model results to assuming uncertainty, and have found unless agents are extremely backward looking when setting their expectations, the results do not vary that much. This is due to the fact that what matters in the model is the current value of TFP and not so much its expected future value.

In terms of the way the solution method works, the initial steady state is calculated to reproduce the state of the Japanese economy in 1990, and the final one is calculated using the values of the exogenous variables after 2002 (when they are assumed to remain constant). Given the two steady states, the solution method calculates the path of the endogenous variables, given the path of the exogenous ones, which connects the two steady states. However, it should be noted that the model does not converge to the new steady state by the year 2002, but much later.