Printer Friendly

Structural unemployment in the United States: the effects of interindustry and interregional dispersion.


The effects of sectoral shifts, measured by dispersion in the growth rates of

employment or earning across industries or regions, on unemployment are tested in a

specification controlling for the effects of other labor-market variables and shifts in the

demographic composition of the labor force. Interindustry and geographical shifts in

labor demand have significant unemployment effects, with adult males the group most

strongly affected. The estimated equations imply that most of the fluctuation in

unemployment over the period 1956-87 was been due to microeconomic causes rather

than aggregate demand.


The unemployment rate in the United States increased from an average of 4.9 percent for 1956-73 to 7.2 percent for 1974-87. It is of considerable importance for purposes of economic policy to evaluate the contributions of various microeconomic and macroeconomic factors to this increase in the unemployment rate. If the primary cause is deficient aggregate demand, then the appropriate policy response may include stimulative monetary and fiscal policy. If microeconomic factors are most significant, then the appropriate policies would be microeconomic in nature--job training programs, restructuring of minimum-wage laws and unemployment-insurance programs, etc.

The period from 1974 to 1987 differed from the earlier postwar period in several widely recognized ways. Oil prices increased dramatically in 1974 and 1979 and decreased substantially beginning in 1982. The steady postwar increase in the female labor-force participation rate became more rapid in the 1970s. The large population cohort associated with the postwar baby boom began reaching maturity and entering the labor market in the early 1970s. Inflation in the United States was sustained at levels higher than ever before. The dollar began floating against the currencies of other industrial countries following the collapse of the Bretton Woods system in 1971 and fluctuated considerably in both nominal and real value. First West Germany, then Japan, and most recently Taiwan and South Korea developed impressive manufacturing bases in industries in which U.S. exports had formerly faced far less competition. Government expenditures on income-transfer programs accelerated dramatically in the 1970s, while regulatory policies were changed in many industries.

The influence of these disturbances on the natural rate of unemployment has been modeled through three main channels: (1) the effects of labor-market distortions such as minimum-wage laws, unemployment insurance, and unions, (2) the effects of changes in the demographic composition of the labor force, and (3) the effects of shifts in the sectoral composition of production and employment. A growing literature has emerged following David Lilien's [1982a] finding that sectoral shifts, measured by interindustry dispersion in rates of employment growth, explain much of the increase in unemployment in the late 1970s. However, the studies relating unemployment to dispersion measures have largely ignored the effects of demographic changes and other traditional labor-market variables. The present paper tests whether dispersion variables help to explain unemployment when the effects of several market-distortion variables and changes in the demographic composition of the labor force are taken into account. I also examine the relative explanatory power of several alternative measures of both interindustry and interregional dispersion.

The second section of the paper summarizes the literature on the effects of dispersion on unemployment. The results of my empirical analysis are presented in section III and applied to current economic policy issues in section IV. The final section of the paper summarizes my conclusions.


Search models of unemployment imply that workers having substantial industry-specific human capital or other kinds of high mobility costs may remain unemployed for relatively long periods of time while trying to find jobs similar to those they lost, which would enable them to avoid the cost of changing industries. In his seminal contribution to the empirical literature on the unemployment effects of dispersion, Lilien [1982a] uses as a dispersion variable the employment-weighted cross-sectoral standard deviation of employment growth rates of eleven sectors of the U.S. economy. For a 1948-80 sample period, this variable and its lag have significant positive signs when added to an unemployment equation whose other explanatory variables are current and lagged values of Robert Barro's [1977] measure of unanticipated growth in the M1 money supply, a time trend, and one lagged value of the unemployment rate. Lilien interprets the significant, positive coefficients of his dispersion measure as evidence that greater intersectoral inequality in employment growth associated with sectoral shifts in the economy leads to higher aggregate unemployment as workers (temporarily) resist moving from one industry to another. His estimated unemployment equation suggests that most of the rise in unemployment in the 1970s is explained by an increase in the natural rate due to increased dispersion, not to cyclical unemployment associated with aggregate-demand shocks.

Because the employment effects of changes in aggregate demand differ across industries, aggregate-demand fluctuations such as those associated with monetary policy may increase the measured degree of employment-growth dispersion in the economy. Using data on job vacancies, Katherine Abraham and Lawrence Katz [1986] rebut Lilien's conclusion, claiming that the positive coefficient of his dispersion measure picks up the effects of aggregate demand rather than interindustry supply disturbances, and that, therefore, aggregate-demand expansion was the appropriate policy for reducing unemployment in the 1970s. Stephen Davis [1987] attributes the contradictory vacancy-rate evidence of Abraham and Katz to their use of vacancy stocks rather than flows and provides a model in which their result is consistent with the sectoral-shifts hypothesis. Further, Lilien [1982b] shows that a dispersion measure that has been purged of intersectoral variation induced by monetary shocks affects unemployment with a significant positive sign. Davis also finds that interindustry variations in employment growth that reverse the pattern of the previous year's variations tend to reduce unemployment, supporting the sectoral-shifts hypothesis.

Kevin Murphy and Robert Topel [1987] provide a highly detailed study of the increase in unemployment during this period, based on a large cross-sectional sample. They find that most of the increase in unemployment has been in long-term unemployment spells and that intersectoral labor mobility has declined at times when the unemployment rate has risen. Based on these findings, they deny an important role for sectoral shifts in explaining the rise in unemployment.

Prakash Loungani [1986] examines the impact of changes in oil prices on employment growth in different sectors of the economy and concludes that most of the increase in interindustry dispersion in the 1970s can be explained by oil-price movements. In additional to oil-price changes, Shehadah Hussein [1988] finds that aggregate-demand fluctuations and movements in the U.S. real exchange rate have contributed to interindustry employment-growth dispersion, and that the components of dispersion associated with each of these sources contribute significantly in an equation explaining aggregate unemployment.

Lucie Samson [1985] and Janet Neelin [1987] have shown that dispersion helps explain the Canadian unemployment rate. Unlike Lilien, Samson includes a demographic variable (the percentage of women in the labor force) in her unemployment equation. Neelin examines both interindustry and interregional dispersion, concluding that interregional difference in employment growth have the more significant effect on Canadian unemployment.


Testing Dispersion Effects

In additional to updating through 1987 the sample period of previous studies, this paper seeks (1) to establish whether the effects of employment-growth dispersion on U.S. unemployment are robust to a specification in which several important labor-market variables are included, (2) to test the sensitivity of the evidence for the sectoral-shift effect to demographic factors by determining the degree to which dispersion measures affect the female, male, and teen unemployment rates, (3) to evaluate alternative specifications of interindustry dispersion: earnings-growth versus employment-growth variation and dispersion at a less-aggregated versus more-aggregated industry level, and (4) to examine the relative importance of geographical versus interindustry dispersion.

The theoretical basis for the effects of sectoral shifts on unemployment was developed by Lilien [1982a], Davis [1987], and others. The theory asserts that disparate growth in employment causes unemployment through workers' reluctance to incur mobility costs in moving from one "job situation" to another or through unavoidable time required for this movement. The crucial characteristic of "job situation" is that moving from one situation to another is time-consuming or inflicts costs on workers. In addition to explicit moving or retraining costs, these costs can include wages, working conditions, or living conditions in prospective new job situations that the worker perceives as inferior. To the extent that workers' skills are industry-specific, interindustry movements are costly and require time for retraining, and industry boundaries can define job situations.(1) The relevant boundaries could be quite broad sectors such as manufacturing, retail trade, and services, or much narrower industries, depending on whether workers are able to transfer easily between industries within the same broad sector of the economy.

Moving from one state or region of the country to another is also costly and time consuming, so job situations may also have a geographical dimension.(2) Thus, although dispersion has usually been measured as the interindustry variance of employment growth, the geographical measures used in this paper are also consistent with the hypothesis. Furthermore, in additional to employment reductions (presumably layoffs), reductions in average weekly hours worked or in wage rates could induce a worker to leave an established job situation and become unemployed while searching for a more suitable situation. Thus, interindustry dispersion in the growth of total compensation of employees--the product of employment, average hours, and wages--may incorporate relevant dispersion effects that are not captured in measures based strictly on employment growth.

Although these alternative measures are all consistent with the basic dispersion hypothesis, the theoretical foundations of the hypothesis have specific implications for the coefficients of the equations predicting the unemployment rates of females, males, and teens. Most fundamentally, greater dispersion in employment or compensation growth, whether across industries or regions, should increase the unemployment rate for all demographic groups. Furthermore, the greatest effect should be felt by those with the greatest mobility cost--the highest level of industry-specific skills or the longest expected retraining period. Because males on average have more work experience than females, and may therefore have acquired more job-specific skills, the male unemployment rate should be more sensitive to interindustry dispersion than the female rate. Because teenagers have little experience and are less likely to have dependents which complicate geographical moves, teenage unemployment should be the least sensitive to both interindustry and interregional dispersion. Women and teenagers are also more likely to respond to layoffs or wage/hours cuts by leaving the labor force (e.g., to pursue parental or home-production activities or to return to school) rather than by searching for a new job. This is an additional reason why female and teen unemployment may be less sensitive to dispersion than male unemployment.

Measurement of Dispersion and Other Variables

Annual data on employment and compensation of employees (earnings) by industry are published beginning in 1948 by the Bureau of Economic Analysis at a sixty-five industry level of aggregation.(3) The variable SE65 is defined for each year as the cross-industry standard deviation of employment growth rates (as an annual percentage change relative to the previous year) across these sixty-five industries, with each industry weighted by its current share in aggregate employment. This measure is similar to those used by Lilien and others, but uses a finer level of industry aggregation.

Because the relevant industry divisions (the boundaries of a "job situation") are unknown, two related dispersion measures were examined: SE13 measures weighted employment-growth standard deviation across thirteen more aggregated sectors,(4) and SE65X captures variation between industries within the same broad sector by summing across industries the weighted squared deviation of each individual subaggregate industry's employment growth rate from the current growth rate for the aggregated sector within which the industry lies.

To incorporate the possibility, mentioned above, that interindustry fluctuations in average hours and wages also lead to dispersion effects, a parallel set of measures (SCE65, SCE13, and SCE65X) were computed based on growth in industry employee compensation rather than industry employment.(5)

Data on employment by state are published by the Bureau of Labor Statistics. Data begin before 1950 for nearly all states, allowing dispersion measures to be constructed for 1950-87. Two measures of geographical employment-growth dispersion are used: SEST measures the cross-state standard deviation of employment growth rates, weighted by the states' shares in total U.S. employment, and SEREG is a similar measure defined over eleven regions corresponding to the Census Bureau's standard regional classification.(6) Figure 1 plots the behavior over the 1956-87 sample of two representative dispersion series: the series based on dispersion of earnings growth across broad sectors (SCE13) and the one based on interregional employment-growth dispersion (SEREG).

Previous studies, beginning with Lilien, have used Barro's [1977] original specification of unanticipated money growth, in current and lagged form, as the sole aggregate-demand variable. Most have also followed Barro's two-step procedure of estimating the money-growth equation prior to the unemployment equation and using the residuals from the former to represent unanticipated changes in aggregate demand in the latter. Adrian Pagan [1984] has shown that, because measurement error in the unanticipated money growth variables is neglected, the two-step estimator is consistent but inefficient, and that the estimated standard errors printed out by typical software packages are inconsistent. To avoid this problem, Leonardo Leiderman [1980], Frederic Mishkin [1983] and others have estimated the money-growth and unemployment equations jointly, imposing (and testing) the cross-equation restrictions implied by the rational-expectations hypothesis. The present study follows Mishkin in estimating the model jointly by iterated, non-linear, weighted least squares.(7)

The specification of the money-growth equation follows Barro [1977], except that a third lag of money growth proved to be significant for the sample period used here and was added to the equation. In addition to three lags of money growth (DM), a measure of federal expenditures relative to "normal" levels (FEDV) and a transformation of the lagged value of the unemployment rate (UALL) appear as explanatory variables. Because the money-growth equation was reestimated for each alternative variant of the unemployment equation, different estimates of the coefficients of the money-growth equation were generated for each variant. All of these were qualitatively similar to the estimates of equation (1) below, which was associated with the regression of the log of the aggregate unemployment rate on current and three lagged values of unanticipated money growth, the labor-market variables discussed below and two dispersion measures: earnings-growth dispersion across broadly defined sectors and interregional employment-growth dispersion.(8) Standard errors are shown in parentheses below each estimated coefficient. (1) [DM.sub.T] = 10.42 + 0.055 [DM.sub.T-1]

(3.43) (0.15)

- 0.092 [DM.sub.T-2] + 0.547 [DM.sub.T-3]

(0.16) (0.13)

+ 24.802[FEDV.sub.T]


+ 3.944 1n[[UALL.sub.T-1]/(100 - [UALL.sub.T-1)]]


+ [DMR.sub.T] The residuals from this equation (DMR) are interpreted as the unanticipated component of monetary growth. This is the aggregate-demand variable used in the unemployment equations.(9)

In addition to the aggregate-demand and dispersion variables discussed above, three indicators of government labor-market intervention are included in the unemployment specification. Increases in the ratio of the federal minimum wage to economy-wide average hourly earnings (MWRAT), by raising the cost of unskilled workers relative to their productivity (which is assumed to move over time with the aggregate average hourly wage), are expected to raise the level of unemployment, especially among teens, who participate most heavily in the unskilled labor market.

The unemployment equation also includes the ratio of unemployment insurance recipients' average weekly unemployment insurance check to average weekly earnings of employed workers (UIRATE) as a proxy for the replacement rate in state unemployment insurance programs. A higher replacement rate increases the subsidy to job-search activities of workers, lowering the private cost of search and raising unemployment. This variable is, at best, an imperfect measure of the effects of unemployment-insurance programs. Unemployment insurance is administered by the states, so replacement rates, limits on dollar benefits, and duration limits vary from state to state. This makes it difficult to construct an aggregate measure of the legal unemployment-insurance environment. At best, this variable may capture changes in the average replacement rate across states. It may also be affected spuriously by changes in the average income of the pool of unemployed recipients. For example, if the proportion of high-income workers among the unemployed were to rise, the average check (which usually depends on previous earnings of the unemployed) would increase, leading to an increase in the proxy variable UIRATE.

An increase in the size of the armed forces, whether through a draft or through enlistment, not only provides a source of employment among young men and women, but may absorb a relatively large proportion of workers who would otherwise be likely candidates for unemployment--those with little experience or education. Changes in military participation may have significant effects across the entire labor market, but should be felt particularly strongly in the market for young, unskilled or semiskilled workers. A variable (MILRT) measuring the number of active military personnel per 1,000 population is included in the unemployment equation. An increase in this variable is expected to lower unemployment.(10)

Estimated Unemployment Equations

As in most econometric applications, little is known from economic theory about the appropriate functional form, lag structure, and stochastic error in the relationship between unemployment and the explanatory variables described above. Because the unemployment rate is necessarily positive, I chose a semi-log specification with an additive error term. The choice of how many lagged values of explanatory variables to include was made on the basis of including one or more lags of a variable only when adding a lag resulted in a significant t-statistic. Only the aggregate-demand variable proved to have significant lagged effects. The time-series specification of the stochastic error was chosen in a similar manner, allowing for an autoregressive-moving-average structure.

Because monetary policy is only one dimension of aggregate demand, I tested several potential fiscal-policy variables in addition to the unanticipated money-growth variables, such as the ratios of real government spending and real government deficits to real gross national product. None of these variables approached statistical significance, so they are omitted from the results presented here. Additionally, the actual rate of money growth proved insignificant, confirming Barro's result that (in this specification) only unanticipated money growth affects unemployment.(11)

As discussed above, several alternative specifications of the interindustry and interregional dispersion variables were tested. Table I compares the estimated coefficients of aggregate-unemployment equations with several measures. Standard errors are in parentheses below the coefficients; the subscripts on the variable names indicate their time periods relative to the dependent variable, which is the log of the aggregate unemployment rate in period T. All equations include the current and two lagged values of the unanticipated money variable and were estimated jointly with the money-growth equation by iterated, non-linear, weighted least squares for a 1956-87 sample. A first-order autoregressive process proved to be an appropriate stochastic specification for all equations in Table I. The hypothesis of white-noise residuals could not be rejected using the Durbin-Watson statistic or the Box-Pierce Q-statistic at lags up to five years. The column headed "SEE" reports the standard error of the estimate for the equation. Adjusted [R.sup.2] statistics for the regressions ranged from 0.93 to 0.96.

Unanticipated money growth and the labor-market variables enter each equation with the expected sign and, with occasional exceptions, are statistically significant. Each of the dispersion measures has a significant and positive coefficient when entered singly. This verifies that dispersion measures affect unemployment even when such labor-market factors as minimum wages, unemployment insurance, and military participation are taken into account, corroborating the conclusions of earlier proponents of the sectoral-shifts hypothesis.(12)

Although the different measures of dispersion yeild quite similar results,(13) comparing the performance of alternative measures shown in Table I yields two broad conclusions. First, the measures based on the broader industry or regional definitions have somewhat larger t-statistics and yield slightly better-fitting equations than those based on narrower definitions. (i.e., SE13, SCE13, and SEREG outperform SE65, SCE65, and SEST, respectively). When measures based on finer aggregations are added to an equation along with those of the broader aggregations, such as in the second equation shown in Table I, the latter dominate with the former having insignificant marginal explanatory power. This suggests that mobility is considerably greater among jobs within these broad sectors/regions (e.g., durables manufacturing or the East North Central region) than between them. Second, the intersectoral measures based on earnings-growth dispersion (SCE13 and SCE65) lead to smaller residual standard errors than those based on employment-growth dispersion (SE13 and SE65). This result is consistent with the hypothesis that some dispersion-related unemployment is caused by quits in response to industry-specific reductions in wages and/or hours. On the basis of the regressions reported in Table I, the dispersion measures chosen for subsequent tests are the measure of earnings-growth dispersion for the broad, thirteen-sector industry classification (SCE13) and the measure of interregional employment-growth dispersion (SEREG).

Table II describes the results of regressions of the aggregate, female, male, and teen unemployment rates on these measures of interindustry and interregional dispersion, along with the unanticipated money-growth and labor-market variables. The aggregate equation suggests that intersectoral earnings-growth contributes more significantly to U.S. unemployment than interregional differences in employment growth,(14) though the latter contributes a significant positive effect. Note that the greater strength of interindustry dispersion variables relative to interregional measures found here is opposite to the result obtained by Neelin [1987] for Canada.

A comparison of the aggregate equation in Table II with those for the various demographic groups yields further support for the importance of sectoral and regional shifts in explaining unemployment.(15) Most crucially, dispersion measures are significant explanatory variables in the subaggregate equations, implying that the significance of these variables in the aggregate equation does not result from a spurious association with omitted demographic effects.

As hypothesized above, intersectoral dispersion affects male unemployment more strongly and significantly than it affects either female or teen unemployment. Male workers are likely to have greater industry-specific human capital because they are, on average, more experienced and better educated than females and teens. Males are also less likely than females or teens to leave the labor force upon leaving or losing a job, implying that job termination is more likely to lead to unemployment for males.

Among the groups shown in Table II, the effect of interregional dispersion is smallest and least significant (in the sense of the smallest t-statistic) for teenagers. If teenage workers have lower costs of geographical mobility (perhaps because of fewer dependents or lower likelihood of home ownership), this is strongly consistent with the dispersion hypothesis.

Thus, the evidence from Tables I and II strongly corroborates the conclusion that sectoral and geographic shifts in the distribution of labor demand raise the level of unemployment. Moreover, the pattern of effects among demographic groups in the labor force is consistent with casual evidence about the relative degrees of intersectoral and geographical mobility of members of these groups.


Estimating the Natural Rate of Unemployment

The distinction between microeconomic and macroeconomic causes of unemployment lies at the heart of the "natural-rate hypothesis." In his seminal statement of this hypothesis, Milton Friedman defines the natural rate of unemployment as

... the level that would be ground

out by the Walrasian system of

general equilibrium equations, provided

there is [sic] imbedded in them the

actual structural characteristics of the

labor and commodity markets,

including market imperfections,

stochastic variability in demands and

supplies, the cost of gathering

information about job vacancies and

labor availabilities, the costs of

mobility, and so on.(16) Thus, equations that estimate the microeconomic and macroeconomic causes of unemployment, such as those presented in Tables I and II, can be solved to yield estimates of the natural rate of unemployment--the rate attributable solely to microeconomic factors.

Figure 2 shows the values of the actual and estimated natural rates of unemployment for the total labor force over the 1956-87 period. The latter is calculated as the fitted values of the aggregate equation in Table II with unanticipated money growth and the residual set equal to zero.(17) Several implications emerge from Figure 2. Contrary to conventional wisdom, the equation implies that unemployment was considerably above the natural rate from 1956 through 1964. In the late 1960s, the decline in unemployment to below 4 percent was driven not by monetary policy but rather by microeconomic factors, mostly the increased size of the military associated with the Vietnam War. As noted by Lilien, the spike in unemployment in 1975 seems to be entirely attributable to natural (microeconomic) factors--partially sectoral shifts associated with the sudden rise in oil prices, but also due to the reduction in the armed forces after Vietnam. The monetary accommodation of the oil shock brought unemployment down below the natural rate during the Carter administration, with the natural rate rising in 1979 and 1980 both due to dispersion and labor-market factors. Contractionary monetary policy in the first Reagan administration moved unemployment well above the natural rate, which began to recede substantially in 1982. The rise in the natural rate in 1986 and 1987 was substantially due to increases in dispersion, probably caused by falling oil prices and changes in the real foreign-exchange value of the dollar. The quantitative contributions of the variables in the equation to unemployment in selected years are examined in the next section.

The natural-rate hypothesis of Friedman [1968] and Phelps [1970] implies that inflation will accelerate--or, more precisely, rise above the expected rate--when the actual unemployment rate falls below the natural rate. According to this hypothesis, the gap between actual and natural unemployment in 1986 and 1987 should have stimulated a rise in inflation during the late 1980s. Although some acceleration occurred, its magnitude seems smaller than that suggested by the rather large unemployment gap shown in Figure 2.

Using the natural-rate hypothesis, some authors have estimated the natural unemployment rate by estimating the rate that would be consistent with an unchanging inflation rate--the nonaccelerating-inflation rate of unemployment (NAIRU). For example, David Stockton [1988] estimates equations representing the price-setting behavior of firms that are subject to microeconomic and macroeconomic shocks. He then infers a series for the NAIRU from 1957 to 1980 from these price equations. Stockton's estimates of the NAIRU are much more volatile in the 1970s than the estimates of the natural rate shown in Figure 2 (reaching a high of nearly 10 percent in 1974 and a low of just over 1 percent in 1977), but are more stable in the 1960s, hovering between 5 and 6.5 percent. Although there are sizable differences in the magnitude of fluctuations, the broad movements in the two series have similar shapes from 1965 through 1980, lending some support to the natural-rate hypothesis linking the natural rate in the aggregate labor market to the nonaccelerating-inflation rate of the aggregate product market. Unfortunately, Stockton's estimates end in 1980, making it impossible to ascertain whether the increase in the natural rate shown in 1986 and 1987 would correspond to an increase in his estimated NAIRU.(18)

Decomposition of Changes in Unemployment

Another application of the estimated aggregate unemployment equation is an examination of the contributions of various factors--labor-market variables, dispersion variables, aggregate demand, and the unexplained residual--to the deviations in various periods of the unemployment rate from its sample mean of 5.9 percent. Figure 3 shows a decomposition of the effects on unemployment attributable to these four sources for six selected years.

In 1967, the unemployment rate and the natural rate were very low--two percentage points below the sample mean. Nearly all of this low unemployment was caused by labor-market variables. Military participation was high as the Vietnam War approached its peak and the unemployment insurance replacement rate was quite low relative to later years. Both of these effects tended to reduce the aggregate unemployment rate. By 1971, the unemployment rate had risen to near its sample mean, again due to "natural" causes--a reversal of those labor-market factors depressing unemployment in 1967 coupled with a contribution from above-normal dispersion.

The high unemployment of 1975 is caused by a combination of labor-market factors (mainly the end of the war) and extremely high dispersion associated with the oil-price shock. Aggregate demand is also relatively contractionary in this period and there is a sizable negative residual--the equation overpredicts the rise in unemployment in 1975! Unusually low dispersion in 1976 through 1978 is reversed with the second oil shock in 1979 (though the dispersion variables surge up only to their sample mean values), but a high real minimum wage and continuing demilitarization push the natural rate upward.

In 1982, as in 1975, all three explanatory sources contribute to the upward spike in unemployment, with dispersion playing an important role as the dollar rises in foreign-exchange markets. The contractionary monetary policy of the Volcker regime is a major cause, but not the sole cause, of the rise in unemployment. As shown in Figure 2, the greatest contributors to the increase in unemployment are the labor-market variables. By 1987, both the dispersion and the labor-market variables have lessened their positive impact on unemployment, but the driving force in the decline in actual unemployment is the rapid monetary expansion of the two preceding years, pushing the actual rate well below the natural rate.

The results reported here verify a significant contribution of sectoral and regional shifts to unemployment, as shown initially by the natural-rate series constructed by Lilien [1982a]. However, a large share of the natural-rate fluctuations in the present model are caused by changes in the labor-market variables rather than by dispersion. In particular, the military-participation and unemployment-insurance variables seem to explain most of the rise in the unemployment rate from the 1950s to the 1980s. Because Lilien did not include these variables in his analysis (relying instead on a time trend to pick up these effects) he may have somewhat overestimated the quantitative impact of sectoral shifts on unemployment.


This paper attempts to extend testing of the effects of sectoral shifts on unemployment to control for the effects of other labor-market variables and of shifts in the demographic composition of the labor force. It also explores several alternative measures of dispersion, including interindustry dispersion in earnings growth rather than employment growth, dispersion across finely defined industries versus more aggregated sectors, and interregional dispersion in employment growth. Significant unemployment effects result both from interindustry shifts and from geographical shifts in labor demand, and these effects appear to be more significant when sectors and regions are defined broadly rather than finely. Moreover, these results hold for major demographic groups in the labor force and complement the effects on unemployment of such labor-market phenomena as minimum-wage laws, unemployment-insurance programs, and military participation. The estimated equations imply that most of the fluctuation in unemployment over the 1956-87 period has been due to microeconomic causes rather than to aggregate demand, and that the actual unemployment rate dipped considerably below the microeconomic natural rate in 1987. [Figures 1 to 3 Omitted] [Tabular Data 1 to 2 Omitted]

(1)It is quite possibe that there are sufficient firm-specific skills to make the individual firm the relevant boundary for the job situation. However, data on employment growth disaggregated to the firm level are not readily available. (2)An additional possibility is that job situations could be defined by occupations, in which case dispersion in growth of employment across occupations would be a relevant measure. This promising possibility is not explored in this paper. (3)The data are taken from Table 6.6B of the National Income and Product Accounts of the United States, 1929-1982, updated in July issues of the Survey of Current Business. Most other studies have used Bureau of Labor Statistics (BLS) data rather than the series from the Bureau of Economic Analysis (BEA). The BLS data are available at a slightly finer level of industry detail, but series for many industries begin in the mid-1950s or later, making it difficult to construct consistent dispersion measures over a long sample period. Measures were constructed based on the ninety-eight industries for which employment data are published back to 1948. The results of the equations using these measures were qualitatively identical to those described in the paper for the BEA measures based on employment-growth dispersion. (4)The thirteen sectors are (1) agriculture, forestry, and fisheries, (2) mining, (3) construction, (4) durable goods manufacturing, (5) nondurable goods manufacturing, (6) transportation, (7) communications, (8) utilities, (9) whosesale trade, (10) retail trade, (11) finance, insurance, and real estate, (12) services, and (13) government. (5)Compensation of employees is given in Table 6.4B of the National Income and Product Accounts. (6)The Census Bureau uses a nine-region breakdown. However, for this study, Alaska and Hawaii are treated as separate regions rather than included in the Pacific region because their geographic isolation implies high labor-mobility costs of moving into or out of these states. (7)The model was estimated using the SYS(H) command of MicroTSP 6.5. (8)This unemployment equation is reported below in Table II. (9)Changes in the Fed's operating procedure, in its objectives, or in its perception of the reaction of the economy to monetary policy should cause changes in the way that agents having rational explanations forecast future money growth. However, there are insufficient annual observations to permit different equations to be estimated over several subsample periods. Because the effects of aggregate demand on unemployment are of secondary importance to this study, subsample regressions with monthly or quarterly data were not performed. The choice of M1 is also potentially problematic. Recent volatility in the velocity of M1 (see Darby et al. [1987], for a discussion of the evidence and issues) casts suspicion on the stability of the linkage between M1 and aggregate demand. This study uses M1 to retain symmetry with previous related work; using M2 or a Divisia index number of monetary assets may be more appropriate for the 1980s. (10)The variable MILRATE used here is not the same as the MIL measure used by Barro [1977]. Barro's variable specifically attempted to measure the effect of the military draft, and was set to zero after 1969 to reflect the end of the draft. Barro's estimates of the effects of the sudden drop in his MIL variable on unemployment in the early 1970s were shown by Small [1979] to be implausibly large. Because the variable used here is not subject to this discontinuity, its presence in the equation does not imply a large jump in natural unemployment in 1970. Another method for dealing with the effects of the military on the labor market is to use the ratio of real federal expenditures to GNP. When added to the unemployment equation in place of MILRATE, this variable had an insignificant coefficient and resulted in a poorer overall fit for the equation. The principal coefficients of interest (those on the dispersion measures) were not significantly affected by this change in specification. (11)There is a long literature testing the ineffectiveness of anticipated money growth. Among many other studies, [1983] finds that in a more general specification of the money-growth equation, anticipated money growth does affect real variables such as unemployment significantly; this study does not pretend to offer new evidence to this debate. Since the contribution of aggregate demand is not the central issue here, relatively little experimentation was done with alternative specifications of the money-growth equation. Several minor changes to the chosen specification for the money-growth equation failed to alter any of the conclusions. (12)Moreover, when these labor-market variables are included, a time trend added to the equation is not only insignificant, but its coefficient has a negative point estimate. Thus, the variables included in the equation provide an adequate statistical explanation for the increase in unemployment from the earlier to the later years of the sample. (13)As one might expect, the dispersion measures are positively correlated with one another. Correlation coefficients between measures based on broadly vs. narrowly defined industries are in the range of 0.89 to 0.93; the state and regional dispersion measures have a correlation of 0.93. The correlations between the employment-growth and earnings-growth measures are in the range of 0.81 to 0.86, while regional dispersion and earnings-growth dispersion have a correlation of 0.53. (14)All of the results reported in Table II were quite insensitive to the choice of intersectoral or geographic dispersion measure. (15)In Table II, the specification of the equation for each demographic group was kept symmetric to that chosen for the aggregate equation. In some cases, this implied retaining an insignificant coefficient, such as the autoregressive error term in the equation for female unemployment. (16)Friedman [1968, 8]. It is worth noting that the natural rate need not correspond to the socially optimal rate because government policies (such as unemployment compensation) and imperfect capital markets or other market failures may cause divergences of the private costs and benefits of search from the social costs and benefits. (17)This implicitly assumes that the residual is part of the deviation of actual from natural unemployment, rather than error in estimating the natural rate. This seems to me to be more natural than the opposite assumption. In any case, because the residual accounts for less than 3 percent of unemployment variance (the unadjusted R-squared value is higher than 0.97), the estimates of the natural rate are not very sensitive to this assumption. The next section examines the contributions of various factors (including dispersion and the residual) to unemployment for selected sample years. (18)The unstable demand for MI in the 1980s, mentioned earlier, makes the linkage between aggregate demand and any unanticipated money measure somewhat suspect over this period. M1 growth increased from 5.7 percent in 1984 to 12.4 percent and 17.0 percent in 1985 and 1986, then felll to 3.5 percent in 1987. The estimated equation translates this erratic behavior in M1 into large movements of the actual unemployment rate relative to the natural rate. The 1987 number in particular may well overstate the effect of the rapid expansion in the two previous years on unemployment.


Abraham, Katherine, and Lawrence Katz. "Cyclical Unemployment: Sectoral Shifts or Aggregate Demand?" Journal of Political Economy, October 1986, 507-22. Barro, Robert J. "Unanticipated Money Growth and Unemployment in the United States." American Economic Review, March 1977, 101-15. Darby, Michael, William Poole, David E. Lindsey, Milton Friedman, and Michael J. Bazdarich. "Recent Behavior of the Velocity of Money." Contemporary Policy Issues, January 1987, 1-33. Davis, Stephen J. "Fluctuations in the Pace of Labor Reallocations," in Empirical Studies of Velocity, Real Exchange Rates, Unemployment, and Productivity edited by Karl Brunner and Allan H. Meltzer. Carnegie-Rochester Conference Series on Public Policy, vol. 27, 1987, 335-402. Friedman, Milton. "The Role of Monetary Policy." American Economic Review, March 1968, 1-17. Hussein, Shehadah. "Sectoral Shifts: Sources and Effects on Aggregate Unemployment and Output in the United States." Ph.D. dissertation, University of Houston, 1988. Leiderman, Leonardo. "Macroeconomic Testing of the Rational Expectations and Structural Neutrality Hypotheses for the United States." Journal of Monetary Economics, January 1980, 69-82. Lillien, David M. "Sectoral Shifts and Cyclical Unemployment." Journal of Political Economy, August 1982a, 777-93. __. "A Sectoral Model of the Business Cycle." Working Paper, University of Southern California, 1982b. Loungani, Prakash. "Oil Price Shocks and the Dispersion Hypothesis." Review of Economics and Statistics, August 1986, 536-39. Mishkin, Frederic A. Rational Expectations Approach to Macroeconometrics. Chicago: University of Chicago Press for National Bureau of Economic Research, 1983. Murphy, Kevin J., and Robert H. Topel. "The Evolution of Unemployment in the United States: 1968-1985," in Macroeconomics Annual 1987, edited by Stanley Fischer. Cambridge, MA: MIT Press for National Bureau of Economic Research, 1987, 11-58. Neelin, Janet. "Sectoral Shifts and Canadian Unemployment." Review of Economics and Statistics, November 1987, 718-23. Pagan, Adrian. "Econometric Issues in the Analysis of Regression with Generated Regressors." International Economic Review, February 1984, 221-47. Phelps, Edmund S., ed. Microeconomic Foundations of Employment and Inflation Theory, New York, W. W. Norton & Co., 1970. Samson, Lucie. "A Study of the Impact of Sectoral Shifts on Aggregate Unemployment in Canada." Canadian Journal of Economics, August 1985, 518-30. Small, David H. "Unanticipated Money Growth and Unemployment in the United States: A Comment." American Economic Review, December 1979, 996-1003. Stockton, David J. "Relative Price Dispersion, Aggregate Price Movement, and the Natural Rate of Unemployment." Economic Inquiry, January 1988, 1-22.

JEFFREY PARKER, Associate Professor of Economics, Reed College, Portland, OR 97202. I wish to thank Paul Evans, James Hagerman, Carl Stevens, Jeffrey Summers and an anonymous referee for their helpful comments on this paper. Shehadah Hussein stimulated my interest in and increased my knowledge of this subject through his dissertation research at the University of Houston. My-Linh Ha provided valuable research assistance. Remaining errors and shortcomings are, of course, my sole responsibility.
COPYRIGHT 1992 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1992 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Parker, Jeffrey
Publication:Economic Inquiry
Date:Jan 1, 1992
Previous Article:Racial differences in homeownership and housing wealth, 1970-1986.
Next Article:Sectoral shifts and cyclical unemployment: a reconsideration.

Terms of use | Copyright © 2016 Farlex, Inc. | Feedback | For webmasters