The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
STOCK MARKET DISPERSION AND REAL ECONOMIC ACTIVITY: EVIDENCE FROM QUARTERLY DATA Prakash Loungani, Mark Rush and William Tave Working Paper Series Macro Economic Issues Research Department Federal Reserve Bank of Chicago September, 1990 (WP-90-15) Stock Market Dispersion and Real Economic Activity: Evidence from Quarterly Data Prakash Loungani University of Florida and Federal Reserve Bank of Chicago Mark Rush University of Florida and William Tave Brown University September 1990 Earlier drafts of this paper were presented at the NBER Summer Institute, the Federal Reserve Bank of Chicago and the University of Florida. We thank James Adams, Herb Baer, Bill Bomberger, Steve Davis, David Denslow, Martin Eichenbaum, Hesna Genay, Larry Kenny, Ken Kuttner, David Lilien, Jim Moser, Richard Rogerson, Steve Strongin, Mark Watson and other seminar participants for extensive comments. The Financial Institutions Center at the University of Florida supported this research. ABSTRACT We conduct an empirical investigation into the effects that stock market dispersion has on real economic activity. The results from fairly standard reduced-form equations suggest that, controlling for the effects of monetary and fiscal policy, stock market dispersion leads to a significant increase in unemployment and a decline in real GNP and investment. We also report results from including our stock market measure and a Lilien-type employment dispersion measure [see Lilien (1982)] in several VAR systems in which unemployment is used as the indicator of real economic activity. The performance of the employment-based measure turns out to be very sensitive to the ordering of the variables in the system. The stock market dispersion measure always explains a larger fraction of the variance of unemployment than does the employment dispersion measure, and the fraction explained is not sensitive to the ordering of the variables. Even after the inclusion of an interest rate variable and the Standard & Poor's 500 in the VAR system, stock market dispersion accounts for between 26% and 33% of the variance of unemployment at long horizons. Prakash Loungani Economic Research Federal Reserve Bank of Chicago 230 S. LaSalle St. Chicago, IL 60690 312-322-8203 Mark Rush Department of Economics University of Florida Gainesville, FL 32611 904-392-0318 William Tave Department of Economics Brown University Providence, RI 02912 1 1. Introduction David Lilien's (1982) paper has sparked a debate on the extent to which fluctuations in the aggregate unemployment rate may be attributed to the reallocation of labor across sectors. The voluminous literature that has followed Lilien can be divided into two groups: (1) time-series studies which test whether proxies for the amount of sectoral labor reallocation are correlated with the aggregate unemployment rate, and (2) studies which attempt to measure labor reallocation and its contribution to unemployment directly by using panel data sets. While this paper belongs to the first group, it is useful to briefly review the evidence from the second group of studies. Lilien appears to have had in mind a model— such as that of Lucas and Prescott sectors is (1974)— where fixed exogenously, the but time downturns required to switch are marked by an increase in the number of workers who experience unemployment as they switch between sectors. Using data from the Current Population Survey, Murphy and Topel (1987) present evidence against this early ("search") version of the sectoral shifts hypothesis. However, one can consider alternate models where the impact of sectoral shocks is not just on the number of workers who experience unemployment as they switch sectors, but also on the time it takes workers to switch sectors. This feature is likely to emerge in models that assign a prominent role to sector-specific human capital. For instance, Topel and Weiss (1985) present a model where some periods— such as the 1970's and early 1980's— are marked by 2 increased uncertainty about the relative returns to sector-specific human capital investment, leading to an increase in the time that displaced workers take to switch sectors. In Rogerson's (1989) model, the impact of sectoral shocks leads to very high durations of unemployment among older workers who are displaced from their jobs: The basic idea is that these workers are at an disadvantage relative to younger workers in that they do not have as long to reap the benefits of (new) human capital accumulation and hence require higher wages than do otherwise identical younger workers. Using the Michigan Panel Study of Income Dynamics, a longitudinal data set that enables researchers to observe workers' mobility and unemployment experience over several consecutive years, Loungani and Rogerson present (1989a) evidence and Loungani, consistent with Rogerson these and broader Sonn views (1989b) of the sectoral shifts hypothesis that stress the importance of sectorspecific human capital accumulation. Since most of the panel data sets start around the late 1960's or early 1970's, they do not offer any evidence on the contribution of sectoral reallocation to unemployment prior to that period. Hence, time-series in d u stry d isp e rsio n studies— which typically construct c r o s s indices to proxy for the amount of sectoral reallocation of resources— are a useful source of complementary evidence. As discussed in Davis (1985, p.32) and Barro (1986, p.138), the use of a dispersion index offers some advantages to researchers who are interested primarily in determining the impact of sectoral shocks on broad macroeconomic aggregates such as the 3 aggregate unemployment rate. Barro states that the use of the dispersion index circumvents "the need to isolate a detailed array of many— mostly preferences sectors." unobservable— disturbances (that) motivate reallocations to of technology resources and across Davis points out that "allocative disturbances from any particular source are likely to occur rather infrequently over available sample sizes," which makes it difficult to explicitly incorporate variables that capture the effects of allocative disturbances into an aggregate unemployment equation. In this paper we construct a measure of the cross-industry dispersion in stock price of growth capital to proxy for the and undertaken labor amount by of sectoral reallocation the economy. In a well-functioning stock market, the industry stock price represents the present value of expected future industry profits. An increase in the dispersion of stock prices across industries reflects the occurrence of shocks that are expected to have differential impacts on industries' profits. If these shocks are expected to be persistent, productive resources, such as capital and labor, will be displaced from the industries that are expected resources to be are adversely not affected. immediately To the absorbed extent into more that these profitable industries, the dispersion in stock prices will be followed by a decline in real economic activity. In Section 2 of the paper, we present a brief theoretical framework along these lines. We also present details on the construction of the stock market dispersion index. 4 While previous studies have focused on the impact of labor reallocation on unemployment, it is likely that the reallocation of capital across sectors is also fairly costly. plausible that the adjustment costs It is therefore associated with capital reallocation lead to declines in other macroeconomic aggregates. In Section 3 we show, using quarterly data for the period 1947 to 1987, that an increase in stock market dispersion leads not only to a statistically significant increase in unemployment but also to a decline in output and investment. The results for unemployment bolster our preliminary work on the relationship between stock market dispersion and unemployment. Loungani, Rush and Tave (1990) present evidence on the determinants of U.S. unemployment over a long time period, 1929 to 1987. Using annual data, we find that unemployment depends on up to three lags of a stock market dispersion measure. Loungani and Rush (1990) construct a stock market dispersion measure using British data for the period 1912 to 1938. This measure appears to reflect fairly well the decline of the traditional export industries and the rise of newer industries and turns out to explain a large fraction of British interwar unemployment. Our stock market dispersion index is clearly motivated by Lilien's use of cross-industry e m p lo y m e n t dispersion to proxy for the intersectoral flow of labor. Many researchers, most notably Abraham and Katz (1984, 1986), have questioned Lilien's use of employment dispersion as a measure of labor reallocation. Their basic point is that movements in employment dispersion may simply 5 be reflecting the well-known fact that the business cycle has non neutral effects across industries. The increase in the dispersion of employment growth rates could reflect, reallocation, not increased labor but simply the uneven impact of aggregate demand shocks on temporary layoffs in different industries. Hence there is an observational equivalence between the predictions sectoral shifts hypothesis and the more traditional of the "aggregate demand hypothesis." The main advantage of a stock market dispersion measure relative to Lilien's measure is that stock prices respond more strongly to disturbances that are perceived to be permanent than to temporary disturbances, which need not be true of employment changes. The industry stock price represents the present value of expected profits over a long horizon. The impact of innovations in industry profits on its stock price will therefore depend on how long the shocks are expected to be persist. If the shocks are purely temporary, the innovations will have little impact on the present value of expected profits and, hence, will have little impact on industries' shocks are fairly stock prices. persistent, the On the other hand, innovations will if the have a significant impact on expected future profits and will lead to large changes in industries' stock prices. Furthermore, it is these sorts of persistent shocks that will cause productive resources, such as capital and labor, to be displaced from the adversely affected industries. Hence, a dispersion index constructed from industries' sto c k p r ic e s automatically assigns greater weight to 6 permanent structural changes rather than temporary cyclical shocks. We conjecture, therefore, that a stock market based dispersion measure is less likely than an employment-based measure to reflect changes in temporary layoffs; this implies that our stock market dispersion variable is less sensitive than employment dispersion measures to aggregate demand disturbances that result in large swings in temporary layoffs. Rather than rely solely on these conjectures, we put them to the test in Section 4 of the paper. Abraham-Katz suggest two methods of resolving the observational equivalence problem that they identify. The first is to test whether the correlation between the dispersion index and the aggregate vacancy rate is positive or negative. Abraham-Katz argue that if the dispersion index is a good proxy for sectoral shifts this correlation should be positive, since the reduced labor demand in some sectors will be matched by increased hiring in other sectors. On the other hand, if dispersion is attributable to aggregate demand shocks, then this correlation should be negative since all sectors will reduce their hiring. The empirical relationship between dispersion and a proxy for the vacancy rate has been investigated in independent work by Brainard and Cutler (1989) using a stock market dispersion index similar to ours.1 They find that the impulse response of the vacancy rate proxy to innovations in their stock market dispersion consistently of one sign, and the standard errors "is not are large relative to the coefficients." This method of resolving the observational equivalence problem 7 suffers from the lack of availability of adequate vacancy data for the U.S. Instead, researchers are forced to use an index based on help-wanted advertising in newspapers in 51 cities. An additional problem is that recent work by Hosios (1988) implies that sectoral shifts models that allow for both capital and labor mobility generate a negative correlation between dispersion and vacancy rates. Hence in his model information on vacancy rates cannot be used to distinguish the aggregate demand hypothesis from the sectoral shifts hypothesis. The second method— which is essentially the one we follow in this paper— involves "purging" the dispersion index of movements that can be attributed to aggregate demand disturbances and then testing if the significantly residual correlated measure with of economic dispersion activity. is This still method requires a careful specification of a list of regressors that adequately capture aggregate demand. Recognizing that stock prices are forward-looking, we include in our list not only standard aggregate demand shifters such as money growth and government spending shocks, but also "information" variables, such as interest rate spreads and mean stock returns, that have emerged in recent studies as strong predictors of future economic activity. However, even after controlling for the effects of these current and potential aggregate demand shifts, innovations in our stock market dispersion index explain nearly 33% of the variance of unemployment at long horizons. dispersion measure On the other hand, explains less than a Lilien-type 5% of the e m p lo y m e n t variance of 8 unemployment once the aggregate demand shifters are included. To summarize, the empirical evidence strongly supports our conjecture that the stock market dispersion index is less susceptible to the Abraham-Katz critique than Lilien's measure. 2. Stock Market Dispersion and Economic Activity A. Theoretical Framework We begin by presenting a theoretical framework that is consistent with the key ideas in Lilien (1982), Black (1982) and Davis (1987) . For convenience we refer to this framework as the costly sectoral mobility model. Consider a n-sector economy with each sector producing a distinct product using a vector of productive resources or inputs, Zlt. Profits in each sector are given by, (1) 7iit = JC(Zl t ) e it where the ei 's are uncorrelated across sectors, with mean e and t (cross-sectional) standard deviation a. Not much significance should be attached to the particular way in which we specify the stochastic shocks to the profit function; this framework can be modified to distinguish among shocks to the sectoral price ("taste shocks"), shocks to the marginal physical product of inputs ("productivity shocks") and shocks to the cost function. The sectoral stock price equals the sum of discounted expected future profits over an infinite horizon, (2) S it = (1/p) ( S E ^ J W ) where P is the discount factor and Et is the expectations operator _x 9 conditional on information available in period t-1. Long-run equilibrium is characterized by the equality of stock returns across sectors, (3) Ri * = Rt t * for all i where Ri =log ( S Lt / S t i t _!) and ^ is a weighted average of the sectoral stock returns. We denote the allocation of inputs across sectors associated with this long-run equilibrium by Zl * Note that this t. target allocation of resources changes over time in response to realizations of the Bit's. In the short-run, productive resources move across sectors if Zj.t-1 The - & i ( z it ~ (Z it Z it - i) / with Z it ~ 0 partial-adjustment Hrt 1 *-» Zj.t-1 Z it - CS3 towards this target allocation as follows: < if a2 < > ^ it - l (Xi < reflects Z lt * < Z it* 1 the assumption that both capital and labor are partly specialized to a sector and hence the reallocation process is costly and/or time-consuming. The role of adjustment costs for capital is emphasized in early work by Eisner and Strotz (1963), while the quasi-fixity of labor was highlighted in seminal work by Oi above is the (1962) assumption asymmetric.2 In particular, reach their and Becker that the (1964) . Also reflected adjustment mechanism is contracting sectors are assumed to long-run equilibrium input levels faster than the expanding sectors, so that ax > a2. Two recent empirical studies provide indirect evidence of the sector-specificity of labor and capital. Topel (1990, p.17) states 10 that "when human capital is 'general' in the sense of being portable among activities, a job loss should imply fairly minor and transitory effects on earning capacity. But with specific capital, initial losses may be large and persistent." Using data from the PSID and the Displaced Worker Survey, Topel finds evidence of large short-run reductions in earnings— 40 percent for manufacturing worker— following job loss. Moreover, change industry or occupation following the the typical workers who job loss have atypically large short run reductions in earnings.3 Grossman and Levinsohn (1989) study the impact of exogenous changes in the prices of competing import goods on stock returns in six U.S. industries. They state (p. 1065) that "when factors are mobile, .. individual returns may respond little or even positively to adverse shocks to the particular sectors in which the factors are employed." They find however that for five of the six industries in their study, lower-than-expected import prices lead to substantial declines in stock returns, suggesting that capital is highly immobile between sectors in the short run. We next consider the changes in realizations Prescott o, of the the (1974) model impact on real economic activity of (cross-section) sector-specific a standard shocks. deviation In the of the Lucas and is assumed to be constant over time and hence the reallocation of product demand across sectors leads to a time-invariant natural rate of unemployment. In contrast, Lilien, Black and Davis suggest that a may vary over time, depending on the nature of the shocks to the economy. In the framework developed 11 above, an increase in a reflects the arrival of shocks that are expected to have differential impacts on sectoral profits. This leads to an increase in the stock prices of sectors that investors believe are going to expand and a decline in the stock prices of sectors that are expected to contract, thereby causing dispersion in the realizations of the stock returns. The greater the difference foreseen in the sectors' prospects, the larger is the dispersion in stock returns and the larger is the reallocation of productive resources across sectors that is required to attain the (new) long-run equilibrium. adjustment mechanism, Given our assumptions about the this reallocation involves an increase in unemployment, and a decline in aggregate output and investment. As discussed in the introduction, the evidence from panel data suggests that it is necessary to think of the reallocation process not just in terms of the am ount sectors but also in terms of the o f resources that have to switch it takes resources to switch tim e sectors. Topel relates activity. and Weiss the (1985) dispersion They assume, sector-specific. in present stock as we do, However, they an alternate theory which market returns to economic that human capital is partly interpret an increase in stock market dispersion as reflecting an increase in uncertainty about the relative returns to sector-specific human capital investment. In the face of this increased uncertainty about which sectors are going to prosper and which ones are going to decline, "individuals with less experience and those with greater costs of acquiring 12 sector-specific human capital will rationally and optimally postpone employment and human capital investment until uncertainty has been resolved." We refer to the Topel-Weiss framework as the sectoral uncertainty model. While the theory underlying their work is distinct from the costly sectoral mobility model outlined above, Topel and Weiss point out that it may be difficult to distinguish between the two empirically (p. 348): "In contrast to Lilien, who implies that the o c c u r r e n c e of a sectoral shock that requires labor to be reallocated raises unemployment, we argue that t h e p r o s p e c t o f f u t u r e s h o c k s is a likely candidate for explaining the observed rise in unemployment, especially among younger individuals. Of course, to the extent that the occurrence of sectoral shocks is correlated over time, a sectoral shock may increase expectations of future shocks, so it may be difficult to completely separate the two theories empirically. In this sense, models of costly sectoral mobility and sectoral uncertainty are complementary theories of rising unemployment." B. Construction and Properties of the Stock Market Dispersion Index This section of the paper describes the construction of the empirical analog to a. The basic data we used to construct our measure of the dispersion of stock prices were monthly average indices of various industries' Standard and Poors stock prices, as constructed by (1988). The industries, which are defined by Standard and Poors, range in size from 2 firms to 31 firms and the indices are computed by weighting each firm's stock price according to the firm's market value. Standard and Poors began compiling these data in 1926; at various times additional industries have been added (and others subtracted) so that currently Standard and 13 Poors compiles indices for about 85 industries. We used a sample of 60 indices, including most industries with a complete data series from 1947 through 1987 as well as a few shorter series deemed important. A list of the industries we used, together with their starting date, ending date (if relevant), and weight in our index is given in the appendix. In calculating the index, we first deflated each index using the GNP price deflator and then used quarterly averages of the monthly data. Then we calculated each indices' growth rate and defined our dispersion measure as (5) St = [£ wlt(ri - rt)2]12 t / where rl is the growth rate of industry i's stock at time t, rt is t the growth rate of Standard and Poor's composite listing, and wl t is a weight based on the industry's employment. Due to the changing number of available, industries the for which wi weight t given Standard an and Poor's industry changed data are as the industries included in our dispersion index changed. wl equals the t over-all weight for industry i, based on its share of employment from the entire sample, (called W±; see the Appendix) divided by the sum of the W± weights used in period t. Thus we compensated for the varying number of industries in different years and so S is an employment-weighted standard deviation of the growth rate of the industries' stock prices. 14 3. Empirical Results from Reduced-Form Equations To determine the role our dispersion index plays in affecting aggregate economic activity, we start by specifying a set of conventional reduced-form regressions of the type estimated by Lilien (1982). Our hypothesis is that the greater the difference foreseen in the industries' prospects, the larger will be the divergence in their stock prices, which will be reflected in an upward movement in the dispersion index. Moreover, the greater the difference foreseen in the industries' prospects, the more resources must be moved and so the larger will be the resulting unemployment and decline in real activity. Under both versions of the sectoral shifts hypothesis, there is reason to expect that an increase in dispersion will have a persistent impact on economic activity, i.e., that la g g e d values of dispersion will be correlated with economic activity. Under the costly sectoral mobility model, this reflects the fact that the reallocation of resources will be staggered over time due to adjustment costs. Under the sectoral uncertainty model, the lag length reflects the time it takes for the uncertainty about sectors' relative prospects to be resolved. We use changes in government spending and money growth to capture shocks to aggregate demand. To control for the effects of changes in government spending, the unemployment regression includes the ratio of federal government purchases of goods and services to trend GNP, called GY, while the output and investment equations include the log of federal purchases, called LF.4 We use the actual growth rate of the base money supply, called DB, as the 15 monetary variable.5 Unemployment rates trended upwards during the late 1960's and the 1970's and demographic changes are often thought of as an important factor in accounting for this rise. To capture this we include a variable DEMO, which equals the percentage of women in the total labor force, in the unemployment equation. To account for the trend growth in output and investment, we include a time trend, T.6 For all the variables, except the trend, we included lags. Clearly there is no theoretical basis for the number of lags to be included. The trade-off between more versus fewer lags hinges on the point that including more lags than justified lowers efficiency but including fewer biases the results. We expect that the relative price effects for which we are searching will occur with a fairly long lag, so at the risk of losing efficiency we included two years worth of lags for S. We also used eight lags for DB, one lag for the government spending variables and the demographic variable in the unemployment equation and, to capture any inertia that we failed to explicitly model, two lags of the dependent variables in each regression. Our main results are robust to several alternate lag structures.7 In summary, we estimated the following reduced form regressions: 8 8 1 2 LY = a x + Xb ( i ) DBt.i + I c U J S t . i + X d ( i ) L F t. i + eT + X f L Y ^ i i=0 i=0 i=0 i=l 8 LI = m 8 + p jjU JS ,., i 2 O + S tU JL F t.i + C T + X y L I ^ i=0 i=l 16 8 8 1 1 UN = (X + X p d JD B t.i + S y U J S t.i + X S d J L F t .i + I i i=0 i=0 i=0 2 kD O EM ,...,^ i=0 + SjlUNt.i 1 where UN = Log(U/[1—U ] ), with U being the unemployment rate, LY is is log of producers' real GNP and LI is the log of real investment in durable equipment and structures. We hypothesize that the P's and S's generally should be negative and the b's, d's, £'s, T's, and K's should be positive. More important, though, are the c's, < ) s and y's which indicate the effect from dispersion. Since J', we expect increased dispersion will lower output and investment, |' while raising unemployment, the c's and < ) s should be negative and the y's positive. For two reasons, though, we examine mainly the lagged values of the dispersion variables. First, the effects of the more contemporaneous dispersion variables may be reflecting effects from differentially other, affect omitted, industries. aggregate This is, variables of course, that the point made by Abraham and Katz. Second, as discussed above, dispersion in the stock market should lead movements in real economic activity. Unconstrained Equations We estimated these regressions for the period 1950-1 to 1987IV. The results from this are reported in Table 1. [In the table, S6 indicates the estimated coefficient for St _6. The other variables have similar interpretations.] The results from Table 1 show that the effect of dispersion on output, investment and unemployment is fairly clear cut. The stock dispersion variables are significantly 17 negative in the output regression at lags two, six and eight, in the investment regression at lags one and eight, and significantly positive in the unemployment equation at lags one, five, and seven. The only puzzle positive in is that the contemporaneous the investment regression at S is the significantly 10% level of significance. The failure of more individual coefficients to attain significance may well be because of collinearity because in all cases the sum of the coefficients is highly significant at over the 99% confidence level. These results, especially the significance of the longer lagged variables, provides evidence in favor of the sectoral shifts hypothesis. Constrained Regressions Because multicollinearity amongst the variables is clearly a problem, we re-estimated our regressions constraining the coefficients for DB and S to lie along a second order polynomial. The results from this estimation are reported in Table 2. Although coefficients this on procedure DB and S does by interpretation of the regressions. not change much, it the does For instance, sums of sharpen the our looking at the effects from changes in the base money supply, we see that all the coefficients the regressions have the expected sign and many are now significantly different from zero. Moreover, all lags of S now have the "correct" sign and most are significantly different from zero even up to lags of two years. It is particularly noteworthy that in the investment and unemployment regressions, the cumulative 18 effect from S lagged six, seven and eight quarters are larger than for any other three consecutive quarters. This large impact for what seems ex ante to be quite long lags appears to us as strong support for the sectoral shocks hypothesis. 4. Sectoral Shifts or Aggregate Demand? This section is devoted to determining the extent to which our stock market dispersion index is subject to the same criticisms that Abraham and Katz work. In the (1984, interests of 1986) aimed at Lilien's brevity we focus empirical largely on the unemployment equation, though similar considerations would hold for the output and investment equations. Our empirical work thus far rests on the assumption that the shocks to sectoral uncorrelated across profits— the sectors. eit's Hence, in equation movements (1)— are in the dispersion index are assumed to be driven by sectoral shocks alone. However, as Abraham-Katz satisfied in differential point out, practice. impacts on this assumption Aggregate sectoral demand profits is unlikely shocks will which also to be have lead to movements in the dispersion index. Under certain conditions— which are spelled out in their paper— aggregate demand shocks can also lead to a positive correlation between the dispersion index and aggregate unemployment. The Abraham-Katz critique points out that treating movements in dispersion as exogenously given— as was assumed in the reducedform equations estimated in the previous section— may be incorrect 19 under certain circumstances. In this section we show that by estimating VAR systems, and by imposing alternate orderings on the contemporaneous innovations, we can gauge the extent to which their critique is applicable in practice. A. Comparison with Employment Dispersion We begin by illustrating the Abraham-Katz critique in a VAR framework. We construct an alternate measure of denoted SIG; the difference between S and SIG is that the latter is a measure of the dispersion of employment growth rates across sectors. We then add SIG to a VAR system in which the other variables are unemployment (UN) and two aggregate demand proxies, the growth rate of the monetary base (DB) and the ratio of federal government purchases to trend GNP (6) (GY). That is we estimate a m-th order autoregression, Xt = A * . ! + ..... + + et where Xt is a vector of all the variables in the model (4x1 in this case) . As a first step, this allows us to ascertain if movements in SIG are Granger-caused by other variables in the system. The results of this estimation are contained in Table 3. The sample period is 1951:2 to 1987:4. The lag length is picked to be 8 quarters, which is a more generous lag length than that used in most VAR studies; however, pruning the lags does not affect our results in this table. Panel A shows that lags of SIG are highly significant in the unemployment equation. However, case that lags of the aggregate demand proxies, it is also the DB and GY, are fairly significant in the employment dispersion equation; the first 20 few lags of unemployment are also significant in this equation though the sum does not attain significance at conventional levels. Hence, there appears to be clear evidence of "reverse causality" running from the other variables in the system would be to employment dispersion. The Granger-causality tests sufficient in detemrnining the extent of the "reverse causality" problem if the contemporaneous innovations in different variables, i.e., the et's in equation (6) above, were independent. However, Panel B — which reports the contemporaneous correlation matrix of the et's— shows that there is that there is a strong, positive correlation between innovations unemployment and innovations in SIG. In light of this, Panel C reports results of the decomposition of variance for the unemployment and employment dispersion equations using the standard Choleski factorization under two alternate orderings. places SIG first in the system, Ordering 1 followed by GY, DB and UN. (This, of course, keeps SIG independent of the contemporaneous values of UN, GY and DB but allows UN to be affected not only by lags of SIG, GY, and DB but also by the contemporaneous values of these variables.) Hence, with only minor modifications, this equation is similar to the reduced-form equation reported earlier. Not surprisingly, the results support our earlier conclusions and the views espoused by Lilien. Employment dispersion explains close to 20% of the variance of unemployment, whereas unemployment explains only about 10% of the variance of dispersion. This pattern is dramatically altered when SIG is placed last 21 in the system, as shown in the results for Ordering 2. Now the results are closer to the Abraham-Katz view: SIG explains less than 5% of the variance of unemployment while nearly 25% of the variance of SIG is attributable to unemployment. To summarize, these results confirm the Abraham-Katz argument that it is difficult to distinguish the view that exogenous sectoral shifts cause some part of unemployment fluctuations from the view that unemployment causes increases in dispersion. Next, we consider the extent to which similar problems arise when our stock market measure, S, is used as the measure of dispersion. Once again, the sample period is 1951:2 to 1987:4 and eight lags of each variable are included. The results are reported in Table 4. Panel A shows that the sum of the lags of S is significantly different from zero in the unemployment equation; as we found in the reduced-from equations, it is the higher-order lags of S, particularly lags seven and eight highly significant. The evidence for "reverse causality" is much weaker, with equation. only Panel the B GY shows variable that in this being there case, significant there is that in very are the S little contemporaneous correlation between the residuals. Panel C presents variance decompositions for two different orderings, one in which S is placed first in the system and one in which it is placed last. The key finding is that the fraction of the variance of unemployment explained by S is not very sensitive to the ordering: S explains 32% of the variance (at step 20) if placed last and 38% if placed first in the system. Also, less than 2% of the variance 22 of S is attributable to innovations in unemployment. These results constitute preliminary evidence that S may be less vulnerable than employment dispersion to the Abraham-Katz critique. B. Results with Mean Stock Price Growth Stock prices are forward-looking and should respond to expected changes in aggregate demand that may not be reflected in current money growth or current government spending. Hence we cannot rule out the possibility that the stock market dispersion index is driven by imminent aggregate demand shocks that we have omitted that differentially affect industries' fortunes. To explore this possibility, we augment both our reduced form regressions and the VAR systems discussed above to include the real growth rate of the Standard & Poor's 500. stock market dispersion The idea is that if movements in the index are largely in expectation of imminent aggregate shocks, then those expectations should also be reflected in movements omitted index, aggregate the in the mean stock price growth. shocks inclusion of are the the factor mean driving stock price our Thus, if dispersion growth should eliminate the impact of stock dispersion on aggregate activity. Table 5 presents the results include results, mean stock price growth, from augmenting the system to DSP. Before several points should be noted. First, we discuss the it turns out that the government spending variable, GY, is no longer significant in the unemployment equation in the augmented system and hence we exclude this variable from the system. In any case, including GY 23 does not affect our main conclusions. brevity, Second, in the interest of we only report the results for a system in which both S and SIG are included simultaneously. Third, we continue to use the monetary base as the measure of money whereas many VAR studies use Ml; however, we obtain qualitatively similar results if we replace the base by Ml (a change which also involves starting the sample in 1959 rather than 1948). Hence the estimated system consists of unemployment, UN, the monetary base, DB, mean stock price growth, DSP, and the two dispersion measures, S and SIG. The lag length is set at eight for the S variable and four for all the other variables. Panel A shows that lags of stock market dispersion continue to be significant inclusion of at DSP about does a not 5% level. eliminate Panel the B shows importance that of S the for unemployment. While the stock market mean is fairly important at the shorter forecast horizons, stock market dispersion continues to account for between 34% and 39% of the variance of unemployment at the longer horizons. Barro between mean (1989) has recently stock price growth investigated the and aggregate relationship investment using reduced-from equations similar to those we use in Section 3. He finds that lagged stock price growth has strong explanatory value for the (growth rate of) investment and, moreover, that this variable dominates other predictors of investment such as q and measures of cash flow.8 In a companion paper, provides recent evidence Barro confirming the well-known (1988b) also link between 2 4 mean stock price growth and subsequent movements in output. In light of these results, it is interesting to briefly return to the reduced-form framework and see whether the inclusion of stock market dispersion has any impact of Barro's findings for output and investment. Table 6 presents the results from augmenting the unconstrained reduced form regressions to include the growth rate of the S&P 500; Table 7 presents similar results from a constrained system, where the coefficients for DB, DSP, and S are constrained to lie along a second order polynomial. In both Tables we see that mean stock price growth has a strong effect on output, and unemployment: Many of the individual investment coefficients significantly different from zero and, except for output, are so too are the sums of the coefficients. Including the growth rate of stock prices seemingly reduces the impact of our dispersion variable. In particular, for both the unconstrained investment and unemployment regressions the sum of our dispersion variables is no longer significantly different from zero at conventional levels. However, it is important to notice that this reduction takes place among the contemporaneous and first few lags of dispersion. If we examine only the last four lagged coefficients find we again that the sums are significantly different from zero: In the investment regression, the F-statistic for the sum of the last four dispersion coefficients is 2.43 and in the unemployment regression the F-statistic is 2.29. Given our emphasis on the lagged coefficients, we find the point that the lags Looking remain significant reassuring. now to the output 25 regression, well as we can see that the sum of all the coefficients— as the sum of just the last four coefficients— is significantly different from zero. Moreover, when we constrain the coefficients in Table 7, the sums as well as the last several coefficients again emerge as significant. C. The Role of Interest Rate Spreads Following the work of Sims (1980), who drew attention to the strong predictive power of the commercial paper rate for output, it has become customary to include some measure of interest rates in VAR systems that attempt to test whether movements in money affect real activity. Sims interest rates (1982) and McCallum rather than monetary (1986) suggest that it is growth rates that properly capture Federal Reserve actions, which may account for their being informative about the future of the real economy. However, a flurry of recent papers has shown that measures of interest rate spreads— differences between interest rates on alternative financial assets-dominate measures of the level of interest rates as robust predictors of economic activity.9 While the measure of the spread used differs across studies, the measure that appears to perform the best is the difference between the short-term commercial paper rate and the short-term Treasury bill rate. In prediction equations for real GNP, Friedman and Kuttner (1989) find that the sum of the interest rate spread variables is significant at the .001 level or better in all their specifications. Stock and Watson (1989)— who examined the information contained in a wide array of variables in 26 constructing a new index of leading indicators— find that the spread outperforms nearly every other variable in forecasting the business cycle. Bernanke (1990) provides preliminary evidence that the reason the spread works so well in predicting economic activity is that it combines information about the stance of monetary policy and, to a lesser extent, expected default risk. To the extent that the stock market dispersion index is also responding to information about the future course of monetary policy, including the interest rate spread in the VAR system should weaken its correlation with unemployment. Table 8 reports results obtained by adding the measure of the spread used by Friedman and Kuttner, the difference between the 4-to-6 month commercial paper rate and the 3 month Treasury Bill rate, which we call IRS, to the VAR system discussed earlier. Panel A reports the F-tests for the unemployment equation. All the variables included in the system are significant and, as in Friedman and Kuttner's work with output, the interest rate spread is significant at better than a .001 level. Panel B reports the variance decomposition of unemployment for two different orderings. Ordering 1 places the employment dispersion first in the system and the stock positions. dispersion last whereas Ordering 2 reverses Several conclusions are apparent. First, these the interest rate spread explains a much larger fraction of the variance than the monetary base. Second, the contribution of employment dispersion is relatively modest, ranging from 3% to 9% at step 20. The most important conclusion, from our perspective, is that stock 27 market dispersion continues to account for a large fraction of the variance of unemployment; at step 20, for instance, between 25% and 33% of the variance is attributable to movements in S. D. Have We Adequately Controlled for Aggregate Demand ? In the preceding monetary base growth stock price growth sections we have used four variables— (DB), government spending changes (DSP) and the interest rate spread (GY), mean (IRS)— to capture the state of current and future aggregate demand. In this section we conduct some tests suggested by Abraham and Katz (1984, pp. 17-20) to detemine whether these variables adequately control for the impact of aggregate demand fluctuations on sectoral stock price growth.1 0 As before, let Sl denote the stock price index for industry t i at time t and define rl = log t . We regress rl on the t aggregate demand variables: (7 ) rl = Yo + YiDBt + Y2GYt + y 3D SPt + Y4 IR S t + T i t lt where the T lt/s are residuals. | We then construct a stock market index based on the residuals from equation (8) (x): Sp urged, = [Z wit( i t - T t)2]1 2 t Tl j / where T t is a weighted average of the “ ] Hit's. We also estimate these equations allowing lagged values of the aggregate demand proxies— as well as current values— to affect sectoral stock price growth; (9 ) We r it = Yo + S y u D B t.i + SY2iGYt.i + ^ D S P ^ picked two alternate lag lengths, 4 + EY4iIRSt_i + T i'it and 8 . The two Spu g d re 28 measures corresponding correlated equation with the (7) . This to one is these lag lengths constructed shown using in Panel A of turn the to be highly residuals Table 9. from Hence the subsequent work only uses the estimated equations (7) and the Spu rged measure given in (8) . If the four variables— GY, DB, DSP and IRS— do a good job of capturing the common factors that underlie variations in stock market returns then the correlation among the T lt's should be much | lower than the correlation among the rit's. Whether or not this is indeed the case is investigated in Panel B of Table 9. The top number in each cell of the table gives the correlation between the r^'s for eight industries simple which were randomly chosen from our set of 60 industries. The bottom number gives the corresponding correlation between the T it's. As shown, the ) correlation between the rit's is uniformly positive— the average correlation is 0.42— which indicates that some common factors do underlie the variations in sectoral stock price returns. However the correlation between the T , ' s is almost always close to zero, h. suggesting that the four variables adequately control for aggregate demand. The cases where the correlation between the elt's is non zero tend to be cases where the two industries belong to the same broader industry group, e.g., the for residual returns "Aluminum" and "Copper." Note that "Auto" and "Oil" are negatively correlated, as one might expect in a period dominated by strong oil price shocks. Finally, we present results obtained from including Spu rged 29 instead of S in the VAR system. Panel A of Table 10 shows that the sum of the lagged values of the "purged" dispersion index is significant at a 2% level of significance. Panel B shows that Sp r a ugd explains 22% of the variance of unemployment if placed last in the system and 33% if placed first. Figure 1 plots the impulse response of unemployment innovations in the other variables of the system. As shown, to Sp r e ugd has a strong impact on unemployment with the peak occurring around lag 10. The impact is also fairly persistent; for instance, at lag 12 the impact of monetary base innovations is essentially zero but the impact of Sp r e is still at half its peak effect. ugd V. CONCLUSIONS A multi-sector economy is subject to a variety of shocks that- -initially at least— affect only one or a few sectors. Many recent papers investigate the impact of such sector-specific shocks, prominent example being Grossman and Levinsohn's (1990) empirical study in the capital in competing of import the goods impact on of variations returns to a careful prices six of U.S. industries. Their study complements Grossman's (1987) earlier work on the employment competition. sectoral While shocks, and wage our focus our goal in effects is also this of on paper variations the is in impacts different: import of We such are interested in determining the extent to which sectoral shocks can lead to changes in broad macroeconomic aggregates such as real GNP, aggregate investment and aggregate unemployment. Recent theoretical 30 work emphasizes two channels through which this can occur. First, if physical capital reallocation and human of resources affected by sectoral uncertainty about out shocks the capital of are sector-specific, industries can be costly. relative returns that are Second, to the adversely if there is sector-specific investment, firms and workers may delay making any investment until the uncertainty is resolved. Instead of explicitly modelling specific shocks, Lilien's we follow (1982) innovative use of a dispersion index to proxy for the intensity of sector-specific shocks. Unlike Lilien, however, we use the dispersion in stock price growth across industries— rather than employment growth dispersion— to measure the intensity of sectoral shifts. The results from fairly standard reduced-form equations suggest that, controlling for the effects base growth and fiscal policy, of monetary stock market dispersion leads to a significant increase in unemployment and a decline in real GNP and investment. While these initial results give strong support for a sectoral shifts explanation of unemployment, robustness, 1986) particularly critique of in light Lilien's it is necessary to test their of Abraham and Katz's employment dispersion (1984, index. Our principal empirical findings are as follows: (i) Using a VAR framework, contemporaneous we correlation find that between there the is a strong innovations unemployment and innovations in employment dispersion. in This makes it very difficult to distinguish empirically a model in 3 1 which exogenous unemployment shifts in from a model employment dispersion in which the causality cause runs the other way. Hence we confirm the basic Abraham-Katz finding, albeit in a different empirical framework. stock (ii) When sectoral market dispersion reallocation, there is used is as the little measure evidence of that unemployment Granger-causes movements in the stock dispersion index. On the other hand, after controlling for the effects of standard aggregate growth and changes demand shifters such as in government purchases, monetary base innovations in stock market dispersion account for between 32% to 38% of the variance of unemployment at long horizons. (iii) We recognize that stock prices are forward-looking and hence our dispersion index may be influenced not only by the current state of aggregate demand, as reflected in money growth and government spending, but also by the future state of aggregate demand. This leads us to expand our VAR system to include two "information" variables that have emerged in recent studies as robust predictors of economic activity. the mean return on the stock market Fischer and Merton (1984)] These variables are [see Barro (1988, 1989) , and an interest rate spread— the differential between the short-term commercial paper rate and the short-term month Kuttner Bernanke (1989), (1990), Stock Treasury bill and Watson rate (1989)]. [see Friedman As the spread appears to reflect discussed and in largely the stance of monetary policy. However, even after controlling for 32 the effects of these additional variables on unemployment, innovations in stock market dispersion account for between 25% and 33% of the variance of unemployment at long horizons. (iv) The set of four variables— DB, GY, DSP and IRS— does a good job of capturing the common stock price movements. factors that underlie sectoral Regressions of sectoral stock price growth on these variables yield residuals that are virtually uncorrelated across industries. (v) Finally, we construct a proxy for sectoral shifts, Spurged that is purged of the influence of aggregate demand. This measure continues to account for between 22% and 31% of the variance of unemployment. 33 REFERENCES Abraham, Katharine and Lawrence Katz, 1984, Cyclical Unemployment: Sectoral Shifts or Aggregate Disturbances?, NBER Working Paper no. 1410. Abraham, Katharine and Lawrence Katz, 1986, Cyclical Unemployment: Sectoral Shifts or Aggregate Disturbances?, Journal of Political Economy. 94, 507-22. Aiyagari, Rao S., Lawrence Christiano and Martin Eichenbaum. "The Output, Employment and Interest Rate Effects of Government Consumption." Federal Reserve Bank of Chicago Working Paper No. WP-90-10, June 1990. Barro, Robert J. "Unanticipated Money Growth and Unemployment in the United States." American Economic Review. March 1977, 101-115. Barro, Robert J. "Output Effects of Government Purchases." Journal of Political Economy. December 1981, 1086-1121. Barro, Robert J. Comment on "Do Equilibrium Real Business Theories Explain Postwar U.S. Business Cycles" by Martin Eichenbaum and Ken Singletion, NBER Macroeconomics Annual 1986. Barro, Robert J. "The Neoclassical Approach to Fiscal Policy." in Modern Business Cycle Theory, edited by Robert J. Barro. Cambridge: Harvard University Press, 1988. (1988a) Barro, Robert J. "The Stock Market and the Macroeconomy: Implications of the October 1987 Crash, Working Paper, Harvard University, February 1988. (1988b) Barro, Robert J. "The Stock Market and Investment." NBER Working Paper No. 2925, April 1989. Barro, Robert J. and Mark Rush. "Unanticipated Money and Economic Activity." in Rational Expectations and Economic Policy, edited by Stanley Fischer, Chicago: University of Chicago Press, 1980. Becker, Gary. Human Capital. NBER, New York, 1964. Bernanke, Ben. "On the Predictive Power of Interest Rates and Interest Rate Spreads." Working Paper, Princeton University, June 4, 1990. Black, Fischer. Business Cycles and Equilibrium. New York, N. Y.:Basil Blackwell, 1987. 34 Brainard, Lael and David Cutler. "Sectoral Shifts and Cyclical Unemployment." Working Paper, December 1989. Courtney, Hugh. "Asymmetric Responses in Factor Demands and the Business Cycle." Working Paper, MIT, November 1989. Davis, Steve. "Allocative Disturbances and Temporal Asymmetry in Labor Market Fluctuations." Working Paper, Chicago GSB, January 1985. Davis, Steve. "Fluctuations in the Pace of Labor Reallocation." Carneaie-Rochester Conference Series on Public Policy. 27, Spring 1987, 335-402. Denslow, Dave and Mark Rush. "Supply Shocks and the Interest Rate." Economic Inquiry. July 1989. Diebold, Francis and Glenn Rudebusch. "Long Memory and Persistence in Aggregate Output." Journal of Monetary Economics. 1989. Eisner, Robert and Robert Strotz. "Determinants of Business Investment," in Impacts of Monetary Policy. Commission on Money and Credit, Prentice-Hall, Englewood Cliffs, 60-338. Estrella, Arturo and Gikas Hardouvelis. "The Term Structure as a Predictor of Interest Rates." Working Paper, Federal Reserve Bank of New York, March 1989. Fazzari, Stephen, Glenn Hubbard and Bruce Petersen. "Financing Constraints and Corporate Investment." Brookings Papers on Economic Activity. 1988:1, 141-95. Fischer, Stanley and Robert Merton. "Macroeconomics and Finance: The Role of the Stock Market." Carnegie-Rochester Conference Series on Public Policy, 21, 1984, 57-108. Friedman, Benjamin and Kenneth Kuttner. "Money, Income and Prices after the 1980's." NBER Working Paper No. 2852, February 1989. Frydman, Roman and Peter Rappoport. "Is the Distinction Between Anticipated and Unanticipated Money Growth Relevant in Explaining Aggregate Output?" American Economic Review. September 1987, 693-703. Grossman, Gene and James Levinsohn. "Import Competition and the Stock Market Return to Capital." American Economic Review. 79, December 1989, 1065-87. Hamilton, James D. "A Neoclassical Model of Unemployment and the Business Cycle." Journal of Political Economy. 96, 1988, 35 593-617. Hosios, Arthur J. "Unemployment and Vacancies with Sectoral Shifts." University of Toronto Working Paper, August 1988. King, Robert and Charles Plosser. "Money, Credit and Prices in a Real Business Cycle Model." American Economic Review. 1984. Kowalczyk, Carsten and Prakash Loungani. "Energy Prices and Sectoral Reallocation." Working Paper, October 1990. Laurent, Robert D. "An Interest Rate-Based Predictor of Monetary Policy." Economic Perspectives. Federal Reserve Bank of Chicago, January/February 1988, 3-14. Laurent, Robert D. "Testing the 'Spread'." Economic Perspectives. Federal Reserve Bank of Chicago, July/August 1989, 22-34. Leban, R. and J. Lesourne. "The Firm's Investmant and Employment Policy through a Business Cycle." European Economic Review. 13, 1980, 43-80. Lilien, David. "Sectoral Shifts and Cyclical Unemployment." Journal of Political Economy. 90, 1982, 777-793. Loungani, Prakash. "Oil Price Shocks and the Dispersion Hypothesis." The Review of Economics and Statistics. August 1986, 536-539. Loungani, Prakash and Richard Rogerson. "Cyclical Fluctuations and the Sectoral Reallocation of Labor: Evidence From the PSID." Journal of Monetary Economics. March 1989. (1989a) Loungani, Prakash, Richard Rogerson, and Yang-Hoon Sonn. "Labor Mobility, Unemployment and Sectoral Shifts: Evidence From Micro Data." Federal Reserve Bank of Chicago Working Paper No. 89-22, December 1989. (1989b) Loungani, Prakash and Mark Rush. "Sectoral Shifts in Interwar Britain." Research Department, Federal Reserve Bank of Chicago Working Paper No. WP-90-7, June 1990. (1990a) Loungani, Prakash, Mark Rush, and William Tave. "Stock Market Dispersion and Unemployment." Forthcoming, Journal of Monetary Economics. May 1990. (1990b) 36 Loungani, Prakash, Mark Rush and William Tave. "Stock Market Dispersion and the Macroeconomy." Forthcoming, Economic Perspectives. Federal Reserve Bank of Chicago. (1990c) Lucas, Robert and Edward Prescott. "Equilibrium Search and Unemployment." Journal of Economic Theory. 1974, 188-209. McCallum, Bennett T. "On 'Real' and 'Sticky-Price' Theories of the Business Cycle." Journal of Money. Credit and Banking. November 1986, 397-414. Mishkin, Frederic S. "Does Anticipated Monetary Policy Matter? An Econometric Investigation." Journal of Political Economy. February, 1982, 22-51. Murphy, Kevin M. and Robert Topel. "The Evolution of Unemployment in the United States: 1968-85." in NBER Macroeconomics Annual 1987. edited by Stanley Fischer. Cambridge: MIT Press, 1987. Neumann, George and Robert Topel. "Employment Risk, Sectoral Shifts and the Geographic Distribution of Unemployment." Working Paper, 1985. Nickell, Stephen. "Fixed Costs, Employment and Labor Demand over the Business Cycle." Economica. 45, 1978, 327-345. Oi, Walter. Comment on the Relationship between Unemployment and Sectoral Shifts. Carneoie-Rochester Conference Series on Public Policy. 27, 403-406. Oi, Walter. "Labor as a Quasi-Fixed Factor." Journal of Political Economy. 70, 1962, 538-55. Rissman, Ellen. "What is the Natural Rate of Unemployment?" Economic Perspectives. Federal Reserve Bank of Chicago, September/October 1986, 3-17. Rogerson, Richard. "Sectoral Shocks, Human Capital and Displaced Workers." Working Paper, Stanford University, July 1989. Rush, Mark. "Unexpected Money and Unemployment: 1920 to 1983." Journal of Money. Credit, and Banking. August 1986, 259-274. Sims, Christopher A. "Comparison of Interwar and Postwar Business Cycles: Monetarism Reconsidered." American Economic Review. May 1982, 250-257. Sims, Christopher A. "Policy Analysis with Econometric Models" Brookings Papers on Economic Activity. 1982, 107-152. 37 Stock, James and Mark Watson. "New Indexes of Coincident and Leading Economic Indicators." NBER Macroeconomics Annual 1989. MIT Press. Sims, Christopher A. "Policy Analysis with Econometric Models" Brookings Papers on Economic Activity. 1982, 107-152. Strongin, Steve. "Macroeconomic Models and the Term Structure of Interest Rates." Federal Reserve Bank of Chicago Working Paper No. WP-90-14, September 1990. Topel, Robert. "Specific Cpaital and Unemployment: Measuring the Costs and Consequences of Job Loss." Working Paper, University of Chicago, February 1990. Topel, Robert and Laurence Weiss. "Sectoral Uncertainty and Unemployment." Chapter 16 in Employment. Unemployment and Labor Utilization, 1985. Weiss, Laurence. "Asymmetric Adjustment Costs and Sectoral Shifts," in Heller e t . a l . (eds.) Equilibrium Analysis: Essays in Honor of Kenneth J. Arrow, Volume II, 1986, Cambridge University Press. 38 Table 1: UNCONSTRAINED REGRESSIONS O U T PU T IN V E S T M E N T UNEM PLO YM ENT STANDARD COEFFICIENT ERROR C .83*** .16 STANDARD COEFFICIENTERROR C -.88*** .26 DB .15 DB1 .18 DB2 -.12 .11 DB3 DB4 -.05 .11 DB5 D B 6 -.02 DB7 .11 .13 DB8 .13 .13 .14 .14 .13 .12 .12 .12 .10 DB DB1 DB2 DB3 DB4 DB5 DB6 DB7 DB8 DB -2.29*** DB1 -.58 DB2 2 .12*** DB3 -2.63*** -.14 DB4 .47 DB5 .62 DB6 DB7 -3.62*** DB 8 .80 .25 I .30 .32 .33 .32 .31 .29 .29 .28 .25 -1.26*** • .62** I -.01 1 .20*** -.88** .15 .40 -.25 -.06 .96*** -.24 GO STANDARD COEFFICIENT ERROR C .82*** .22 .09 .10 .11 .11 .11 .11 .11 .10 .09 S SI S2 S3 S4 S5 S6 SI S8 .12 .46* -.11 .21 -.18 .52* -.19 .69** .15 .24 .27 .29 .29 .29 .30 .29 .28 .24 .60*** .2 1 I 1.67*** .54 -.014 -.050 -.097** .031 -.075 .034 -.078* .001 -.066* .037 .043 .045 .047 .046 .046 .045 .043 .038 S SI S2 S3 S4 S5 S6 SI S8 I -.312*** .086 I LF LF1 T .057 LF .014 .025 -.029 LF1 -.014 .056 -.035 .025 T .0012*** .0002 .0008*** .0002 1.14*** -.24*** R2 .9994 = .09 .08 LI1 LI2 S E = .0091 - 5.25*** 1 .40 .16* -.30*** .05 -.12 .05 -.12 -.06 -.06 -.19** S SI S2 S3 S4 S5 S6 S7 S8 LY1 LY2 I .81 .88 .86 .87 .84 .79 .80 .77 .66 1.30*** -.42*** .08 .08 R2 .997 6SE= .0217 = GY GY1 -•1.57 1.68 DEMO .045** DEMOl -.039** UNI 1.48*** UN 2 -.66*** 1 .14 1 .17 .020 .020 .07 .07 R:! 9708SE= .0584 =. *** indicates significance at the 1% level; ** at the 5% level and * at the 10% level. 39 Table 2: CONSTRAINED REGRESSIONS O U T PU T IN V E ST M E N T UNEM PLO YM ENT STANDARD COEFFICIENT ERROR .21 .87*** C STANDARD COEFFICIENT ERROR 94 * * * .17 C STANDARD COEFFICIENTERROR •1.04*** .27 C DB DB1 DB2 DB3 DB4 DB5 DB6 DB7 DB8 .16* .10* .05 .02 .02 .02 .05* .09*** .15*** DB DB1 DB2 DB3 DB4 DB5 DB6 DB7 DB8 I .66** S S4 S5 S6 S7 S8 -.047** -.040** -.035*** -.032** -.030** -.031** -.034*** -.039*** -.046** I - .3 3 4 * * * LF LF1 T LF .017 -.089 .058 .023 LF1 .034 .024 .058 -.040* .0002 T .0013*** .0003 .0008*** S3 1 .10*** -.20*** LY1 LY2 • I I CM 9994 .19 .11 .07 .08 .09 .08 .07 .08 .14 DB -•1.41*** DB1 -.91.*** DB2 -. 54*** DB3 -.31 DB4 -.22 DB5 -.26 D B 6 -.43** DB7 -.75*** DB 8 -■ 1.20 .52 .30 .21 .22 .25 .24 .21 .23 .38 I 1 .5 6 * * * .51 I - ■6.01*** 1 .4 7 .024 .016 .013 .014 .014 .013 .012 .014 .022 S SI S2 S3 S4 S5 S6 S8 -.061 -.052 -.049 -.052 -.061* -.076** -.098*** -.125*** -.159*** .063 .041 .033 .036 .037 .035 .031 .037 .059 S SI S2 S3 S4 S5 S6 S7 S8 .33** .22** .15*** .11 .11 .14 .22*** .33*** .48*** .16 .11 .09 .09 .09 .09 .08 .10 .16 .084 I - .7 3 2 * * * .2 1 9 2 .0 9 * * * .5 6 C M SI S2 .40** .25** .14* .07 .04 .05 .10 .19** .33** .08 .05 .03 .03 .04 .03 .03 .03 .06 .08 .08 SE=.0091 SI LI 1 LI2 1 .21*** - . 34 R2 .9970 = *** .08 .08 SE= .0234 I GY GY1 - 1.57 1.75 1.16 1.19 DEMO .029 DEMOl -.022 UNI 1.40*** UN 2 -.59*** R2 .9638 = .020 .020 .07 .07 SE =.0621 *** indicates significance at the 1% level; ** at the 5% level; and * at the 10% level. 40 Key: SIG = employment dispersion GY = govt, purchases/trend GNP DB =: monetary base growth UN =: unemployment rate Table 3 VAR System: SI 6 6Y DB UN Sample Period: 1951:1 to 1987 :4 F-TESTS : UN Panel A VARIABLE SIG GY DB UN F-STAT. 4.70 1.94 3.75 189.45 F-TESTS: SIG SIGN. : VL. L .00005 .05946 .00003 .00000 F-STAT. 3.18 2.71 2.63 1.62 SIGN. LVL. .0027 .0089 .0110 .1275 Entries are F-statistic values and significance levels of the hypothesis that 8 lags of the variable can be excluded. from the unemployment and employment dispersion equations. Panel B CORRELATION MATRIX OF RESIDUALS VARIABLE 1 . 000 SIG GY DB UN 2 4 8 12 20 DB UN - 0.114 - 0.005 0.026 # # 1.000 0.313 - 0.128 - 0.091 1.000 • Panel . C STEP GYI SIG • 1.000 • DECOMPOSITION OF VARIANCE: ORDERING 1 SIG 15.8 10.9 16.6 18.3 17.7 UN GY 1.4 4.0 7.5 11.6 13.5 DB UN 2.2 3.2 6.9 7.2 7.4 80.5 82.0 68.9 62.8 61.4 SIG 95.9 83.7 80.6 76.7 74.9 SIG GY DB 1.0 7.9 9.0 8.6 8.6 0.1 2.5 3.4 5.1 5.1 UN 3.0 5.9 7.0 9.5 11.3 DECOMPOSITION OF VARIANCE: ORDERING 2 STEP GY UN DB UN SIG 2 4 8 12 20 1.2 3.6 6.9 10.9 13.0 3.7 4.6 9.3 9.8 9.9 94.2 91.4 80.7 74.8 72.7 SIG 0.8 0.3 3.1 4.5 4.6 GY 0.8 7.4 8.6 8.2 8.2 DB UN 1.7 3.7 4.5 6.1 6.1 16.5 19.4 19.3 21.2 22.5 SIG 81.0 69.4 67.6 64.9 63.1 Entries show percentage of forecast variance of unemployment and employment dispersion at different horizons attributable to innovations in the variables of the system. Ordering is as shown 41 Key: S = GY = DB = UN = Table 4 VAR System: S GY DB UN Sample Period: 1951:2 to 1987:4 stock market dispersion govt, purchases/trend GNP monetary base growth unemployment rate F-TESTS: UN Panel A F-STAT. 2.61 1.24 3.04 165.29 VARIABLE S GY DB UN F-TESTS : S SIGN. L V L . .0116 .2807 .0038 .0000 F-STAT. 4.31 3.21 0.69 0.98 SIGN. LVL. .0001 .0025 .6942 .4522 Entries are: F-statistic: values and significance levels of the hypothesis that 8 lags of the variable can be excluded from the unemployment and stock dispersion equations. panel B CORRELATION MATRIX OF RESIDUALS VARIABLE 1.000 S GY DB UN Panel C STEP 2 4 8 12 20 GYI S - 0.022 # 1.000 DB UN 0.046 - 0.044 0.101 - 0.131 - 0.085 1.000 • • 1.000 • DECOMPOSITION OF VARIANCE: ORDERING 1 S 4.9 11.5 25.1 37.5 36.7 UN GY 1.0 1.6 1.9 3.9 7.5 DB 2.0 3.3 8.2 7.5 7.6 UN 92.1 87.8 64.8 51.1 48.1 S 97.5 92.9 84.6 80.9 75.2 S GY 1.2 4.9 10.3 13.9 16.6 DB 0.8 1.7 3.0 3.9 6.7 UN 0.5 0.5 0.9 1.3 1.5 S DB UN S DECOMPOSITION OF VARIANCE: ORDERING 2 STEP 2 4 8 12 20 GY 1.0 1.8 2.2 4.4 8.1 UN DB 1.7 2.8 7.0 6.2 6.4 UN 95.4 88.9 71.0 56.7 53.6 S 1.8 6.4 19.8 32.7 31.9 GY 1.5 5.0 10.4 14.0 16.5 0.7 1.7 2.7 3.8 6.2 0.9 1.0 1.2 1.7 1.8 97.0 92.3 85.6 80.6 75.5 Entries show percentage of forecast variance of unemployment and stock market dispersion at different horizons attributable to innovations in the variables of the system. Ordering is as shown. 42 Table 5 VAR System: S SI 6 DSP DB UN Sample Period: 1951:2 to 1987:4 Key: S SIG DSP DB UN Panel A VARIABLE s SIG DSP DB UN = = = = = stock market dispersion index employment dispersion index growth rate of S&P 500 monetary base growth unemployment rate P-TESTS: UN F-STATISTIC 1 .96 4. 96 4. 05 2 .58 496. 57 SIGNIF. LEVEL .0576 .0010 .0041 .0408 .0000 Entries are F-statistic values and significance levels of hypothesis that 4 lags of the variable (8 in the! case of ; be excluded from the unemployment equations. Panel B DECOMPOSITION OF VARIANCE: ORDERING 1 (SIG DSP1 DB UN S) STEP S SIG 2 4 8 12 20 1.7 4.4 11.4 27.1 33.8 12.0 5.3 5.7 6.2 5.6 DSP DB 4.0 25.2 32.2 28.8 27.5 0.8 1.6 5.5 4.6 4.4 UN 81.5 69.9 45.2 33.3 28.6 DECOMPOSITION OF VARIANCE: ORDERING 2 (S DSP DB UN SIG) STEP S SIG DSP DB 2 4 8 12 20 2.6 6.1 14.3 31.1 39.0 0.4 1.1 1.5 1.9 2.0 2.5 15.6 26.1 21.8 20.2 2.4 3.3 9.8 9.6 8.4 UN 92.2 73.8 48.3 35.5 30.4 Entries show percentage of forecast variance of unemployment at different horizons attributable to innovations in the variables of the system. Ordering is as shown in parenthesis (...). 43 Table 6: UNCONSTRAINED REGRESSIONS WITH STOCK PRICE GROWTH OUTPUT INVESTMENT STANDARD COEFFICIENT ERROR .23 .45** C STANDARD COEFFICIENT ERROR .17 .74*** C STANDARD COEFFICIENTERROR C ■ •07*** 1 .26 .13 .13 .13 .13 .13 .12 .12 .11 .11 DB DB1 DB2 DB3 DB4 DB5 DB 6 DB7 DB 8 -.11 1 .22*** -.70** .12 .36 -.24 -.02 1 .01*** -.16 .31 .31 .33 .31 .31 .29 .29 .28 .26 DB 2 .20** DB1 -.98 DB2 1.96** DB3 -•2.49*** DB4 -.44 DB5 .30 DB 6 .65 DB7 -•3.53*** DB 8 .42 .25 I 1.48*** .51 .004 DSP DSP1 .032*** DSP2 .034*** DSP3 -.002 DSP4 .007 DSP5 -.007 DSP6 -.001 DSP7 -.012 DSP8 -.004 .011 .012 .012 .012 .012 .012 .012 .011 .012 DSP -.019 DSP1 .035* DSP2 .087*** .037 DSP3 DSP4 .023 DSP5 .033 .028 DSP6 DSP7 .032 DSP8 .018 - 6.31*** 1.61 .02 .07 DSP DSP1 — .22 * * * .07 DSP2 -.18** .08 DSP3 -.19** .08 DSP4 -.09 .08 DSP5 -.11 .08 DSP6 -.03 .08 DSP7 -.10 .08 DSP8 -.08 .08 I .050 .041 I S SI S2 S3 S4 S5 S6 S7 S8 .006 -.060 -.093** .015 -.063 .028 -.090** .003 -.068* .037 .042 .043 .046 .045 .045 .044 .043 .037 S SI S2 S3 S4 S5 S6 SI S8 .208** -.256** .062 -.097 .079 -.056 -.073 -.040 -.166 I -.323*** .094 I -.340 LF LF1 T .036 -.061** .0004* .026 .026 .0002 LF LF1 T -.044 .058 -.001 .056 .0009*** .0003 DB .01 .22* DB1 DB2 -.12 .10 DB3 DB4 -.07 DB5 .11 .03 DB6 DB7 .06 .07 DB8 .39 I R2 9995 SE=.0087 : =. LI1 LI2 1 .22*** -.32*** I - 1 .00*** .30 .089 .101 .107 .107 .108 .107 .106 .101 .089 S SI S2 S3 S4 S5 S6 SI S8 -.08 .41 -.19 .14 -.33 .34 -.17 .53** .15 .24 .26 .28 .28 .28 .29 .29 .27 .24 .233 I .59 .098 .09 .09 R‘ 9980SE= .0209 ! =. GY GY1 • .09 .08 E o 1.08*** -.12 .84 .86 .84 .85 .82 .77 .79 .75 .69 0 0 LY1 LY2 .275*** .027 .027 .028 .029 .029 .028 .029 .028 .029 UNEMPLOYMENT -1.62 2.09* DEMO .042** DEMOl -.034* UNI 1.36*** UN 2 -.50*** 1.13 1.16 .019 .019 .08 .08 R 2 .9755SE =.0555 : = *** indicates significance at the 1% level; ** at the 5% level; and * at the 10% level. 4 4 Table 7: CONSTRAINED REGRESSIONS WITH STOCK PRICE GROWTH INVESTMENT STANDARD COEFFICIENT ERROR .63*** STANDARD COEFFICIENTERROR -1.23*** .27 C CM CM • OUTPUT STANDARD COEFFICIENT ERROR .92*** .17 C DB DB1 DB2 DB3 DB4 DB5 DB6 DB7 DB8 .07 .04 .03 .03 .03 .05* .08** .2 1* .08 .05 .03 .03 .04 .03 .03 .03 .06 DB DB1 DB2 DB3 DB4 DB5 DB 6 DB7 DB 8 I .55** .24 X .018** DSP DSP1 .016** DSP2 .014** DSP3 .011* DSP4 .007 DSP5 .003 DSP6 - . 0 0 2 DSP7 -.007 DSP8 -.013 .009 .006 .005 .006 .006 .005 .005 .006 .009 DSP DSPl DSP2 DSP3 DSP4 DSP5 DSP6 DSP7 DSP8 X .041 c .1 0 .046 .48*** .29** .16** .07 .04 .07 .15** 2 9* * * .48*** .20 .11 .07 .08 .09 .08 .07 .09 .16 DB DB1 DB2 DB3 DB4 DB5 DB6 DB7 DB 8 -1.79*** -1.16*** -.71*** -.43** -.32 -.38 -.61*** -1 .01*** -1.59*** 2.05*** .53 X - 8 .00*** 1.67 .002 .022* .038*** .048*** .053*** .053*** .047*** .036** .020 .022 .015 .013 .014 .015 .015 .014 .017 .024 DSP DSPl DSP2 DSP3 DSP4 DSP5 DSP6 DSP7 DSP8 X .32*** .10 -.017 -.006 -.002 -.006 -.019 -.039 -.067** -.102*** -.146*** .062 .042 .036 .038 .039 .036 .032 .037 .057 X s .242 S SI S2 S3 S4 S5 S6 S7 S8 -.041* -.036** -.033** -.031** -.032** -.033** -.037*** -.042*** -.049** .024 .016 .014 .014 .015 .014 .013 .014 . 0 2 2 S SI S2 S3 S4 S5 S6 S7 S8 X -.333*** .093 X -.403* LF LF1 T .026 -.051** .0005** .023 .024 LF LF1 T -.120** .057 .057 .064 .0011*** .0003 LY1 LY2 1.06*** -.12* R2 .9994 = . 0 0 0 2 .08 .08 SE= .0088 UNEMPLOYMENT LI1 LI2 1 .11*** -.24*** .08 .07 R2 .9973SE=.0226 = .56 .33 .21 .21 .24 .23 .21 .25 .43 -.08 -. 12*** -.14*** -.16*** -.16*** -.15*** — .14*** -.11** -.07 .06 .04 .04 .04 .04 .04 .04 .05 .07 - 1 .1 1 *** .30 SI S2 S3 S4 S5 S6 SI S8 .21 .08 .01 -.04 -.03 .02 .11 .25** ^44*** .16 .10 .09 .10 .10 .09 .09 .10 .15 X 1.06** .62 GY GY1 -1.29 1.70 .034* DEMO DEMO1-.024 UNI 1.27*** UN 2 -.46*** 1.13 1.16 .019 .019 .08 .07 R2 9674SE=. 0596 =. *** indicates significance at the 1% level; ** at the 5% level and * at the 10% level. 45 Table 8 VAR System: S DSP SIG IRS DB UN Sample Period: 1951:2 to 1987:4 Key: S DSP SIG IRS DB UN Panel A VARIABLE S DSP SIG IRS DB UN = = = = = = stock market dispersion index growth rate of S&P 500 employment dispersion index interest rate spread monetary base growth unemployment rate F-TESTS: UN F-STATISTIC 2.00 2.10 5.52 3.14 3.35 532.60 SIGNIF. LEVEL .0520 .0838 .0004 .0171 .0122 .0000 Entries are F-statistic values and significance levels of the hypothesis that 4 lags of the variable (8 lags in the case of S) can be excluded from the unemployment equations. Panel B DECOMPOSITION OF VARIANCE: ORDERING 1 (DSP SIG IRS DB UN S) STEP 2 4 8 12 20 S 1.2 2.7 8.9 20.3 24.6 DSP 3.6 18.1 29.6 26.8 26.2 SIG 10.7 4.6 5.4 5.6 4.9 IRS 5.9 12.9 12.9 14.5 16.1 DB 0.7 1.8 5.8 14.5 16.1 UN 77.9 59.9 37.4 27.9 23.5 DECOMPOSITION OF VARIANCE: ORDERING 2 (S DSP IRS DB UN SIG) STEP 2 4 8 12 20 S 2.0 4.9 12.8 27.0 33.1 DSP 3.1 16.4 26.1 22.5 21.5 SIG 0.4 0.7 1.2 1.8 1.9 IRS 4.5 10.7 9.7 9.6 10.4 DB 1.6 2.8 9.0 8.9 7.6 UN 88.3 64.4 41.1 30.3 25.5 Entries show percentage of forecast variance of unemployment at different horizons attributable to innovations in the variables of the system. Ordering is as shown in parenthesis (...). 46 Table 9 Panel A: Correlation matrix of alternate Spargsd measures Spurged (4 lags) C ‘ -'purged (8 lags) c ‘ -'purged 0.936* (no lags) 0.902* C ‘ -'purged (4 lags) •• 0.972* ) Panel B : Correlation matrix of rlt's and T lt's Entr. Auto. .47* .01 Copp. Alum. Drug .44* -.01 .53* .13 .21* -.41* .44* -.09 .30* -.07 .57* .11 .40* .05 .33* -.11 .47* .02 .39* .07 .63* .30* .63* .40* .42* .09 .52* .25* .23* -.03 .47* .11 .37* -.06 .42* .09 .24* -.07 .43* .00 .48* .25* .21* -.11 .29* -.20* •• .26* -.05 .47* .05 Copp. •• Alum. •• • # • • •• •• • • • Drug • • • • • • • • • • • • • Media •• •• Coal Coal .43* .12 Entr. Oil Oil .36* -.01 * denotes that the null hypothesis that the correlation is zero can be rejected at a significance level of .01 Kev to abbreviated industry names: A u t o .= Automobiles; Entr.= Entertainment; Copp. = Copper; Alum.= Aluminum; Oil = Domestic Oil; Media = Broadcast Media 47 Table 10 VAR System: Spurgad DSP SIG IRS DB UN Sample Period: 1951:2 to 1987:4 Key: Spurged = "purged" stock market dispersion index DSP = growth rate of S&P 500 SIG = employment dispersion index IRS = interest rate spread = monetary base growth DB = unemployment rate UN F-TESTS: O N Panel A VARIABLE Spurged DSP SIG IRS DB UN F-STATISTIC 2.39 3.30 5.52 2.78 5.16 426.53 SIGNIF. LEVEL .0207 .0137 .0004 .0307 .0007 .0000 Entries are F-statistic values and significance levels of the hypothesis that 4 lags of the variable (8 lags in the case of S) can be excluded from the unemployment equations. Panel B DECOMPOSITION OF VARIANCE: ORDERING 1 (DSP SIG IRS DB ON Sp r < l ugK) STEP 2 4 8 12 20 Spurqed 1.3 3.6 1 0 .6 20.5 2 2 .1 DSP 4.7 23.2 38.9 34.8 34.8 SIG 9.6 3.7 5.4 5.6 5.3 IRS 4.5 1 0 .0 8 .1 1 1 .0 1 2 .6 DB 3.1 5.3 9.5 7.7 6.9 UN 76.7 54.1 27.4 20.3 18.2 DECOMPOSITION OF VARIANCE: ORDERING 2 (SpU g < DSP IRS DB UN SIG) rBj STEP Spurqed 2 2.4 4 6 .8 8 12 20 16.0 28.9 31.4 DSP 3.4 20.3 33.1 28.2 28.2 SIG 0.5 0.7 2.3 3.1 3.1 IRS 3.3 7.7 5.3 6.3 7.4 DB 5.1 7.2 13.7 11.9 UN 85.2 57.3 29.7 1 0 .6 19.3 2 1 .6 Entries show percentage of forecast variance of unemployment at different horizons attributable to innovations in the variables of the system. Ordering is as shown in parenthesis (...). IMPULSE RESPONSE OF UN -.0 7 5 48 APPENDIX I : CONSTRUCTION OF THE D ISPE R SIO N INDEX To assemble our measure of the dispersion of stock market prices, we used 60 industrial indices compiled by Standard and Poor's. The following listing, arranged by length of the data series, gives the starting and, if relevant, ending dates as well as the employment weight for each industry used: INDUSTRY OIL-COMPOSITE MACHINERY (AGRICULTURAL) AUTOMOBILES COMPUTER SYSTEMS ENTERTAINMENT INVESTMENT COS (CLOSED END) RETAIL STORES (DEPARTMENT STORES) RETAIL STORES (FOOD CHAIN STORES) COPPER MACHINERY (CONSTRUCTION & MAT. HAND.) OIL (CRUDE PRODUCERS) BUILDING MATERIALS COAL DRUGS FINANCIAL (PROPERTY-CASUALTY INSURANCE) HOUSEHOLD PRODUCTS MACHINERY (DIVERSIFIED) MONEY CENTER BANKS PAPER RETAIL STORES (COMPOSITE) SHOES STEEL TIRES AND RUBBER GOODS TRANSPORTATION (RAILROADS) MACHINE TOOLS CHEMICALS CONTAINERS (METAL & GLASS) FOODS HEAVY DUTY TRUCKS & PARTS TEXTILE PRODUCTS TRANSPORTATION (AIRLINES) UTILITIES (ELECTRIC POWER COMPANIES) ELECTRONIC MAJOR COMPANIES AEROSPACE/DEFENSE BEVERAGES (SOFT DRINKS) TEXTILES (APPAREL MANUFACTURERS) BEVERAGES (DISTILLERS) FINANCIAL (PERSONAL LOAN) BEVERAGE S ( BREWERS) ALUMINUM DOMESTIC OILS INTERNATIONAL OILS STAR1’ END YEAR YEAR. . . w, .004614 1926 — 1926 1985 .007786 1928 — .048679 1930 — .026044 1930 — .008573 .001387 1930 — .044414 1930 1930 — .023748 1930 1986 .009005 1930 1985 .007786 1930 1985 .004614 1932 — .009658 1932 -- .000850 1932 — .032236 1932 — .000669 1932 — .032236 1932 — .007786 1932 — .021462 1932 — .012355 1932 — .044414 1932 — .002114 1932 — .009005 1932 — .019075 1932 — .017221 1933 .007786 1934 — .032236 1934 — .011508 .014427 1934 1934 — .048679 1934 — .010624 1934 — .013094 1934 .013124 1934 1986 .026044 1936 — .048679 .014427 1936 — 1936 — .010188 1936 1986 .014427 1939 -- .001387 1940 .014427 1941 — .009005 1943 — .031571 .031571 1943 — — — — — — 49 OIL WELL & EQUIPMENT SERVICE ELECTRICAL EQUIPMENT GOLD MINING HOUSEHOLD FURNISHINGS & APPLIANCES MAJOR REGIONAL BANKS METALS MISCELLANEOUS NATURAL GAS PIPE LINES PAPER CONTAINERS NATURAL GAS DISTRIBUTORS PUBLISHING BROADCAST MEDIA TRANSPORTATION (TRUCKERS) FINANCIAL (SAVINGS & LOAN HOLD. COS.) HOMEBUILDING TRANSPORTATION (AIR FREIGHT) ELECTRONICS (SEMICONDUCTORS) COMPUTER SOFTWARE & SERVICES HOSPITAL MANAGEMENT The 1943 — .007786 1945 — .026044 1945 .004582 1945 — .026044 — 1945 .021462 1945 .009005 1945 .013124 1945 .012355 1945 1984 .013124 1946 .011392 1947 .005688 1957 — .011936 .001387 1959 — 1965 — .001766 1965 — .013094 1970 — .026044 1978 — .025904 1978 .020130 — — — — — - - — weights used to construct S were derived from the Standard and Poor's Compustat II 1968-1987 Annual Aggregate Industrial File computer data tape. This tape lists, among other data, annual employment for each industry. The industries are organized by four-digit codes similar to the SIC codes, though the industry break-down is not exactly the same as in the Standard and Poor's Security Price Index, from which the stock data were obtained. However, the composition of these industries were the same for two-digit industries. Thus, we needed to make some approximations. We wanted weights based on data near the center point of our sample period. Thus, we started by using the four digit industries and calculated the industry's average employment figure using data between 1968 to 1972. If all of these years were missing data, we used the employment figure from the year closest to 1972. These four-digit industry weights were then grouped into the two-digit industry and the share of employment accounted for by each two-digit industry was calculated. Finally, to give our w1 this share was divided by # the number of our sixty industries that fell within each of the two-digit categories. Thus, similar industries that fall within the same two digit classification, eg FOOD and BEVERAGES, have the same employment weight. 51 FOOTNOTES 1. Brainard and Cutler (1989) regress industry stock growth on mean stock price growth and use the residuals from these regressions to construct their stock market dispersion index. However, despite this difference, the correlation between their index and ours is high: 0.66 in levels and 0.74 in logs. 2. For a discussion of asymmetries in adjustment costs of quasifixed factors, see Nickell (1978), Leban and Lesourne (1980), Weiss (1986) and Courtney (1989). 3. The losses are actually larger for occupational change than industry change, which is consistent with the comments of (1987) . See Loungani, Rogerson and Sonn for evidence on contribution of occupational mobility to total weeks unemployment. for Oi the of 4. Two points about our specification deserve mention. First, since we include time trends in the output and investment regressions, which is equivalent to detrending all the independent variables, the specification of the government spending variable is actually quite similar to that in the unemployment regression. Second, the distinction between permanent and temporary changes in spending is important in theory [see Barro (1981 and 1988a), Denslow and Rush (1989) and Aiyagari, Christiano and Eichenbaum (1990)] and empirical applications that include major wars. However, our sample period includes only the Korean and Vietnam wars; neither of these seemed sufficiently important relative to total output to enable us to distinguish between temporary and permanent government spending. 5. Our choice of the base as the measure of money is motivated by the possible endogeneity of broader monetary aggregates such as Ml. See King and Plosser (1984) and Rush (1986) for a further discussion of this issue. Studies that use quarterly data, starting with Barro and Rush (1980) and up to the more recent Frydman and Rappoport (1987), tend to find that all changes in the money supply— not just unexpected changes— matter for real activity. Hence we do not pursue a decomposition of base growth into expected and unexpected components. 6. There is still a lot of dispute over whether macro aggregates such as GNP are difference-stationary, as suggested by Nelson and Plosser (1982), or trend stationary, as suggested in many other studies such as Diebold and Rudebusch (1988). Faced with this uncertainty, we opted for the traditional approach of assuming 52 trend stationarity. 7. For instance, we increased the number of lags for S and DB to twelve and sixteen; increased the lags for government spending and DEMO to four; and increased the lags for the dependent variable to three and four. Individually and jointly the added lags rarely attained standard levels of significance. 8. For alternate views of the investment process that stress the role of cash flow variables, see Fazzari, Hubbard and Petersen (1988) . 9. In addition to the studies cited in the main text of the paper, the role of interest rate spreads is investigated in Laurent (1988, 1989), Estrella and Hardouvelis (1989) and Strongin (1990).1 0 10. It is quite likely that variables such as DSP are responding to events such as oil price shocks, which are not pure aggregate demand shocks. In fact, Davis (1985), Loungani (1986), Hamilton (1988) and Kowalczyk and Loungani (1990) provide theoretical and empirical evidence on the impact of oil price shocks on the sectoral reallocation of resources. However, in order to be as fair as possible to the Abraham-Katz view, we prefer to "over-control" for the effects of aggregate demand on sectoral stock returns by treating all movements in DSP as being "aggregate-demand-driven." Working Papers and Staff Memoranda The following lists papers developed in recent years by the Bank’s research staff. Copies of those materials that are currently available can be obtained by contacting the Public Information Center (312) 322-5111. W orking Paper Series A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics. REGIONAL ECONOMIC ISSUES Taxation of Public Utilities Sales: State Practices and the Illinois Experience WP-86-1 D ia n e F . S ieg e l a n d W illia m A . T esta Measuring Regional High Tech Activity with Occupational Data WP-87-1 A len k a S . G iese a n d W illia m A . T esta Alternative Approaches to Analysis of Total Factor Productivity at the Plant Level WP-87-2 R o b e rt H. S ch norbu s a n d P h ilip R . Isra ilev ic h Industrial R&D An Analysis of the Chicago Area WP-87-3 A len ka S. G iese a n d W illia m A . T esta Metro Area Growth from 1976 to 1985: Theory and Evidence WP-89-1 W illiam A . T esta Unemployment Insurance: A State Economic Development Perspective WP-89-2 W illiam A . T esta a n d N a ta lie A . D a v ila A Window of Opportunity Opens for Regional Economic Analysis: BEA Release Gross State Product Data WP-89-3 A len ka S. G iese Determining Manufacturing Output for States and Regions WP-89-4 P h ilip R . Isra ile v ic h a n d W illia m A. T esta The Opening of Midwest Manufacturing to Foreign Companies: The Influx of Foreign Direct Investment WP-89-5 A len ka S .G ie se l A New Approach to Regional Capital Stock Estimation: Measurement and Performance Alenka S. Giese and Robert H. Schnorbus WP-89-6 Why has Illinois Manufacturing Fallen Behind the Region? William A. Testa WP-89-7 Regional Specialization and Technology in Manufacturing Alenka S. Giese and William A. Testa WP-89-8 Theory and Evidence of Two Competitive Price Mechanisms for Steel Christopher Erceg, Philip R. Israilevich and Robert H. Schnorbus WP-89-9 Regional Energy Costs and Business Siting Decisions: An Illinois Perspective David R. Allardice and William A. Testa Manufacturing's Changeover to Services in the Great Lakes Economy William A. Testa Construction of Input-Output Coefficients with Flexible Functional Forms Philip R. Israilevich WP-89-10 WP-89-12 WP-90-1 Regional Regulatory Effects on Bank Efficiency Douglas D. Evanoffand Philip R. Israilevich WP-90-4 Regional Growth and Development Theory: Summary and Evaluation Geoffrey JD. Hewings WP-90-5 Institutional Rigidities as Barriers to Regional Growth: A Midwest Perspective Michael Kendix WP-90-6 ISSUES IN FINANCIAL REGULATION Technical Change, Regulation, and Economies of Scale for Large Commercial Banks: An Application of a Modified Version of Shepard's Lemma Douglas D. Evanoff, Philip R. Israilevich and Randall C. Merris WP-89-11 2 Working paper series continued Reserve Account Management Behavior: Impact of the Reserve Accounting Scheme and Carry Forward Provision W P-89-12 D o u g la s D . E v a n o ff Are Some Banks too Large to Fail? Myth and Reality WP-89-14 G eo rg e G . K au fm an Variability and Stationarity of Term Premia W P-89-16 R am o n P . D e G en n aro a n d J a m e s T. M o se r A Model of Borrowing and Lending with Fixed and Variable Interest Rates WP-89-17 T hom as M o n d sch ea n Do ‘'Vulnerable" Economies Need Deposit Insurance?: Lessons from the U.S. Agricultural Boom and Bust of the 1920s W P-89-18 C h a rle s W. C a lo m iris The Savings and Loan Rescue of 1989: Causes and Perspective WP-89-23 G eo rg e G. K au fm an The Impact of Deposit Insurance on S&L Shareholders' Risk/Retum Trade-offs WP-89-24 E lijah B r e w e r III Payments System Risk Issues on a Global Economy W P-90-12 H e rb e rt L . B a e r a n d D o u g la s D . E v a n o ff MACRO ECONOMIC ISSUES Back of the G-7 Pack: Public Investment and Productivity Growth in the Group of Seven WP-89-13 D a v id A . A sch a u e r Monetary and Non-Monetary Sources of Inflation: An Error Correction Analysis W P-89-15 K en n eth N . K u ttn er Trade Policy and Union Wage Dynamics WP-89-19 E llen R . R issm a n 3 Working paper series continued Investment Cyclicality in Manufacturing Industries W P-89-20 B ru ce C. P e te rse n a n d W illia m A . S tra u ss Labor Mobility, Unemployment and Sectoral Shifts: Evidence from Micro Data WP-89-22 P ra k a sh L ou n gan i, R ic h a rd R o g e rso n a n d Y ang-H oon Sonn Unit Roots in Real GNP: Do We Know, and Do We Care? WP-90-2 L a w re n c e J . C h ristia n o a n d M a rtin E ich en bau m Money Supply Announcements and the Market's Perception of Federal Reserve Policy WP-90-3 S teven S tron gin a n d V efa T arhan Sectoral Shifts in Interwar Britain W P-90-7 P ra k a sh L o u n g a n i a n d M a rk R u sh Money, Output, and Inflation: Testing the P-Star Restrictions WP-90-8 K en n eth N . K u ttn e r Current Real Business Cycle Theories and Aggregate Labor Market Fluctuations WP-90-9 L a w re n c e J. C h ristia n o a n d M a rtin E ich en bau m The Output, Employment, and Interest Rate Effects of Government Consumption WP-90-10 S. R a o A iy a g a ri, L a w re n c e J. C h ristia n o a n d M a rtin E ich en bau m Money, Income, Prices and Interest Rates after the 1980s WP-90-11 B en ja m in M . F ried m a n a n d K en n eth N . K u ttn er Real Business Cycle Theory: Wisdom or Whimsy? WP-90-13 M a rtin E ich en bau m Macroeconomic Models and the Term Structure of Interest Rates WP-90-14 S teven S tro n g in 4 Working paper series continued Stock Market Dispersion and Real Economic Activity: Evidence from Quarterly Data P ra k a sh L o u n g a n i , M a rk R u sh a n d W illiam T ave Term-Structure Spreads, The Money Supply Mechanism, and Indicators of Monetary Policy R o b e r t D . L a u ren t WP-90-15 WP-90-16 Staff Memoranda A series of research papers in draft form prepared by members of the Research Department and distributed to the academic community for review and comment. (Series discontinued in December, 1988. Later works appear in working paper series). Risks and Failures in Banking: Overview, History, and Evaluation George J. Benston and George G. Kaufman SM-86-1 The Equilibrium Approach to Fiscal Policy David Alan Aschauer SM-86-2 Banking Risk in Historical Perspective George G. Kaufman SM-86-3 The Impact of Market, Industry, and Interest Rate Risks on Bank Stock Returns Elijah Brewer, III and Cheng Few Lee SM-86-4 Wage Growth and Sectoral Shifts: New Evidence on the Stability o f the Phillips Curve Ellen R.Rissman SM-87-1 Testing Stock-Adjustment Specifications and Other Restrictions on Money Demand Equations Randall C. Merris SM-87-2 The Truth About Bank Runs SM-87-3 G eo rg e G . K a u fm a n On The Relationship Between Standby Letters of Credit and Bank Capital Gary D. Koppenhaver and Roger Stover Alternative Instruments for Hedging Inflation Risk in the Banking Industry Gary D. Koppenhaver and Cheng F. Lee SM-87-4 SM-87-5 The Effects of Regulation on Bank Participation in the Market Gary D. Koppenhaver SM-87-6 Bank Stock Valuation: Does Maturity Gap Matter? Vefa Tarhan SM-87-7 6 Staff Memoranda continued Finite Horizons, Intertemporal Substitution and Fiscal Policy SM-87-8 D a v id A la n A sch a u e r Reevaluation of the Structure-Conduct-Performance Paradigm in Banking SM-87-9 D o u g la s D . E v a n o ff a n d D ia n a L . F o rtie r Net Private Investment and Public Expenditure in the United States 1953-1984 SM-87-10 D a v id A la n A sch a u e r Risk and Solvency Regulation of Depository Institutions: Past Policies and Current Options SM-88-1 G e o rg e J. B en sto n a n d G e o rg e G . K au fm an Public Spending and the Return to Capital SM-88-2 D a v id A sch a u e r Is Government Spending Stimulative? SM-88-3 D a v id A sch a u e r Securities Activities of Commercial Banks: The Current Economic and Legal Environment SM-88-4 G eo rg e G. K au fm an a n d L a r r y R. M o te A Note on the Relationship Between Bank Holding Company Risks and Nonbank Activity SM-88-5 E lija h B r e w e r , III Duration Models: A Taxonomy SM-88-6 G. O . B ie rw a g , G eo rg e G. K au fm an a n d C ynthia M . L a tta Durations of Nondefault-Free Securities G . 0 . B ie rw a g a n d G e o rg e G. K au fm an Is Public Expenditure Productive? D a v id A sch a u e r SM-88-7 Staff Memoranda continued Commercial Bank Capacity to Pay Interest on Demand Deposits: Evidence from Large Weekly Reporting Banks SM-88-8 E lija h B rew e r, III a n d T h om as H . M o n d sch ea n Imperfect Information and the Permanent Income Hypothesis SM-88-9 A b h ijit V. B a n e rje e a n d K en n eth N . K u ttn er Does Public Capital Crowd out Private Capital? SM-88-10 D a v id A sch a u e r Imports, Trade Policy, and Union Wage Dynamics SM-88-11 E llen R issm a n 8