Читать книгу Out of Work - Richard K Vedder - Страница 30

In its simplest form, the hypothesized model is:

where U stands for the unemployment rate, W for the adjusted real wage (real wages divided by labor productivity), a is a constant term, and b is a coefficient measuring the degree to which unemployment changes with changes in the adjusted real wage.

From information on unemployment and the variables comprising the adjusted real wage, it is possible to perform a statistical evaluation of the neoclassical position. Generally accepted data on the aggregate unemployment rate are available from 1890 to the present.¹ Money-wage data are available in abundance, although there is less consensus on which series is appropriate. From 1947, the widely cited official Bureau of Labor Statistics data on money compensation per hour in the business sector are used.²

There is a problem with wage data for the period before 1947. Ideally, wages should be related to a specific period of work effort, so hourly wage data are most desirable. Unfortunately, hourly data are available only for some specific classes of workers, not the economy as a whole. Annual earnings data are available that are much more comprehensive in their coverage, although they suffer the drawback of representing varying amounts of time worked. The solution we arrived at was to combine the comprehensive Department of Commerce annual wage data with other estimates of the average work week to obtain estimated hourly wages for the entire work population.³

Productivity (output per hour) data are readily available; we employ the estimates in Historical Statistics of the United States for the years to 1946.⁴ These widely accepted data were originally derived by John Kendrick.⁵ Beginning with 1947, the standard Bureau of Labor Statistics data are our choice.⁶ For prices, we use the consumer price index.⁷ Dividing the hourly wage by the price index gives us a measure of real wages. Dividing real wages by productivity produces the adjusted real wage.

We can employ ordinary least squares regression procedures to examine the relationship between the adjusted real wage W and the unemployment rate U for the ninety years from 1900 through 1989. Doing so, we obtain the following results:

where the numbers in parentheses are t-values.

The results suggest that there is a positive relationship that is highly significant in a statistical sense between the adjusted real wage and the unemployment rate. Moreover, variations in the adjusted real wage alone can explain about five-sixths (83 percent) of the observed variation in the unemployment rate over time.⁸ The results are highly consistent with the neoclassical/Austrian view that unemployment is largely determined by adjusted real wages in excess of equilibrium (market-clearing) levels.

In the typical (median) year, the adjusted real wage moved by 1.85 points (ignoring the direction of the change). Thus the results suggest that in a typical year, the unemployment rate moves about 0.6 percentage points (say, from 5.0 to 5.6 percent) as a consequence of fluctuations in the adjusted real wage.⁹ Not all years are typical, of course. Indeed, in some twelve of the ninety years, the adjusted real wage rose or fell by more than five points, implying an unemployment-rate change of at least 1.6 percentage points (e.g., from 4.5 percent to 6.1 percent.) Thus fairly frequent relatively major changes in unemployment seem to be attributable to changes in the adjusted real wage.

The results reported in (2) are very supportive of the basic neoclassical/Austrian perspective, but they fail to fully describe the relationship between unemployment and changes in the various components making up the adjusted real wage—money wages, prices, and productivity. Also, it is possible that unemployment responds, at least in part, in a lagged fashion to changes in the adjusted real wage. This year’s increase in unemployment may partly have resulted from last year’s increase in the adjusted real wage.

To provide greater detail and to allow for some lag in the unemployment response to wage changes, we can expand the model:¹⁰

where W^t-1 represents the adjusted real wage one year earlier, and denote the change in percentage terms in money wages, consumer prices, and output per hour worked, respectively, over the past year, a is the constant term, and b, c, d, and e are coefficients measuring the extent to which the explanatory variables impact on unemployment. We can submit the expanded model to regression analysis:¹¹

The model confirms even more strongly the neoclassical/Austrian position. Every variable has the expected sign and is significant at the 99 percent level of confidence. Higher adjusted real wages last year are associated with more unemployment this year. Rising money wages in the past year are associated with more unemployment, while rising prices and productivity (both lowering the adjusted real wage) are associated with lower unemployment. The model explains an impressive 90 percent of the variation in unemployment over time.

It is interesting to see how the model performs in explaining the major cyclical episodes in modern American unemployment history. Figure 3.1 shows the actual and predicted unemployment rates by year from 1900 to 1989. The two lines in general move closely together, and there is not a single major downturn or upturn in unemployment that is not captured, albeit sometimes imperfectly, by the adjusted real wage model.

Because of the large amount of detail included in the small space of figure 3.1, it is difficult to ascertain with any precision the differences between the actual unemployment rate and the unemployment rate predicted by (4). In figure 3.2, the 90-year period is divided into approximately equal quartiles of 22–23 years each to allow a better look at the model’s forecasting ability. The model underpredicts the rise in unemployment at the beginning of World War I somewhat, but rather accurately forecasts the dramatic increase in unemployment in 1921. It does even better in predicting the Great Depression in the early 1930s. In the postwar era, the model accurately forecasts major upswings in unemployment (e.g., 1958, 1975, 1982), but often a year later than they actually occurred. Thus the lag between real-wage changes and unemployment may have changed over time. Later in the book, we will deal with this problem by using quarterly data that are available for the postwar period.

In summary, variations in unemployment throughout the century are well explained by changes in the adjusted real wage. How sensitive is unemployment to changes in the components of the adjusted real wage? In the typical year, money wages changed by 5.4 percent (median change, ignoring the direction of the change.) Multiplying that “typical” change by the coefficient from (4), 0.172, we see the effects of a typical year’s money-wage increase is to change the unemployment rate by about 0.9 percentage points (e.g., raising it from 5.5 to 6.4 percent.) Similarly, a typical productivity change of 2.88 percentage points (median change, ignoring signs) also had the impact of changing the unemployment rate by about 0.9 percentage points (2.88 × .302 = 0.870.)

The median price change was 3.26 percent (ignoring the direction of the change.) Multiplying that by the coefficient from (4), we observe that unemployment typically was changed by about 1.4 percentage points (3.26 times 0.437) because of changing price levels (e.g., from 5.5 to 6.9 percent.) All three components seem to have measurable impacts on unemployment, with the price variable being moderately greater than the money-wage or productivity factors in relative importance. In addition, of course, each component enters into the determination of the adjusted real wage, which, in lagged form, is also included in the regression equation. Therefore, the total impact of, for example, a 5.4 percent money-wage increase is greater than indicated above.

FIGURE 3.1 ACTUAL VS. PREDICTED UNEMPLOYMENT, 1900–1990

FIGURE 3.2 A CLOSER LOOK AT ACTUAL VS. PREDICTED UNEMPLOYMENT