Full text of Working Paper (Federal Reserve Bank of Cleveland) : The Effects of Inflation on Wage Adjustments in Firm-Level Data : Grease or Sand?, Working Paper 94-18

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

clevelandfed.org/research/workpaper/index.cfm

Workinn Paver 94 18

THE EFFECTS OF INFLATION ON WAGE ADJUSTMENTS
IN FIRM-LEVEL DATA: GREASE OR SAND?
by Erica L. Groshen and Mark E. Schweitzer

Erica L. Groshen and Mark E. Schweitzer are economists at
the Federal Reserve Bank of New York and the Federal
Reserve Bank of Cleveland, respectively. The authors thank
Kristin Roberts for excellent research assistance and
Edward Montgomery and Joseph Ritter for usehl
discussions. They also thank participants in workshops at
Columbia University, the Federal Reserve Banks of New
York and San Francisco, Lehigh University, American
University, MacMaster University, the City University of
New York Graduate Center, and Queens College.
Working papers of the Federal Reserve Bank of Cleveland
are preliminary materials circulated to stimulate discussion
and critical comment. The views stated herein are those of
the authors and are not necessarily those of the Federal
Reserve Bank of Cleveland, the Federal Reserve Bank of
New York, or the Board of Governors of the Federal
Reserve System.
December 1994

clevelandfed.org/research/workpaper/index.cfm

ABSTRACT
This paper studies wage changes in a 37-year panel of occupations and employers drawn
from the Federal Reserve Bank of Cleveland Community Salary Survey (CSS). Using an
institutional model of the wage-setting process as a guide, we 1) identifjl wage adjustments in
two embedded relative prices and 2) draw inferences about the costs and benefits of inflation
from the adjustments in these relative prices.
Typical institutional wage-setting policies manage employer-wide wage adjustments
(controlling for occupational wage changes) and interoccupational wage changes (controlling for
employer wage changes) separately. In the CSS, we are able to identifjl large independent
employer and occupation components of wage changes. While there is no a priori reason for
these adjustments to be altered by inflation (when the average change is subtracted out),
variation in both of these terms is positively correlated with inflation.

In the interpretation phase of the paper, we treat employer-wide wage deviations as
emphasizing forecasting errors and differences in the speed of adjustment to inflation. In
contrast, we argue that occupational wage deviations include a higher concentration of marketdriven relative price adjustments. This simple dichotomy, whose robustness we attempt to test,
yields two policy-oriented results: 1) Higher inflation and labor productivity appear to increase
the rate of occupational wage adjustments ("grease"), although these potential benefits taper off
after inflation rises to about 4 percent (assuming 1.5 percent average growth of labor
productivity); and 2) Potentially inefficient variations in employer wage adjustments ("sand")
continue to mount until inflation reaches rates of 7 to 10 percent (again assuming productivity
growth of 1.5 percent).

clevelandfed.org/research/workpaper/index.cfm

1. Introduction
How does inflation affect the labor market? This paper explores the effects of the level of
inflation on the dispersion of wage changes in order to expand our knowledge of the impact and
transmission of inflation in the labor market. Our findings add to the literatures on both wage
flexibility (or rigidity) and inflation's impact on price adjustments.
This paper's strength--the unusually tight link we forge between our analytic approach
and common compensation adjustment practices--is made possible by the data set we study. The
Federal Reserve Bank of Cleveland Community Salary Survey (CSS) from 1956 to 1992 offers
detailed data on employers' actual wage adjustments. Because the purpose of the data set is to
provide participating employers with information on market wage adjustments, it records wages
at the level of detail that compensation managers desire for assessing their market position.
Thus, the relative wages we consider are the margins of adjustment within which the firms
maintain comparability of their wage structure with competitors in their labor market. We find
that variability in both occupation- and employer-relative wages increases with inflation.
We draw inferences about the costs and benefits of inflation by examining the association
between inflation measures and the dispersion of occupation-wide and employer-wide wage
changes. Variation in these terms can be seen as desirable (increased occupational wage
flexibility) or undesirable (increased variation between employers). In keeping with the
exploratory nature of the exercise, we use statistical procedures to confirm the robustness of the
relationship between these terms and inflation, rather than imposing structural restrictions on the
associations we detect.
The paper proceeds as follows: The next section applies institutional wage-setting
procedures to decompose notional wage adjustments into the terms we analyze. Section three
reviews the two strands of relevant literature and contrasts our approach with those previously
taken. In the fourth and fifth sections, we describe our data and confirm that the nature of wage
adjustments observed is consistent with the model we advance. The sixth section analyzes

clevelandfed.org/research/workpaper/index.cfm

inflation's effect on the dispersion of occupational and employer wage adjustments and performs
several checks on the robustness of our findings. The final section summarizes our findings.

2. Institutional Wage Adjustment
Could an inflation-induced hike in the dispersion of wage changes be beneficial, or could
it reflect distortions in the labor market? The answer to this question depends on the unobserved
motivations of firms. We develop our statistical analysis in the context of the institutional wagesetting practices that the data were designed to inform. We base our institutional model on
discussions with personnel executives, compensation textbook descriptions of the process, and
compensation managers' responses in Levine's (1 993) and others' surveys.
a. Typical Compensation Policies

Fundamentally, observed salaries are bounded on the high end by workers' marginal
products and on the low end by employees' outside opportunities. However, these constraints
may not determine a unique wage in most corporate settings because both parties have limited
current information on individuals' productivity and labor market options. Since employers do
not observe labor supply and demand functions, they develop compensation policies to attract
and retain qualified employees. Although these policies differ across firms, large employers'
practices typically share the following common features: a job evaluation program to rate jobs;
salary grades or a wage line to assign earnings to jobs according to their evaluations; and a meritor seniority-based system to govern wage growth within salary grades.'
Annual compensation budgets, and therefore average pay increases, are determined by
top management, typically the chief executive officer (Freedman [1976]). After approval (two to
six months in advance of the actual salary adjustments), the budget provides the total "pie" for
wage increases to be split up among departments, and then within departments in accordance
with perceived merit and labor market conditions for particular workers or groups of workers.

Examples of compensation policy references that describe and recommend these practices include Hills
(1987), MiIkovich and Newman (1990), and Wallace and Fay (1988).

clevelandfed.org/research/workpaper/index.cfm

Although the degree of decentralization varies among companies, the basic mechanism usually
takes the form described above.2
In a 1976 Conference Board survey on corporate compensation setting, compensation

executives indicated that a diverse set of factors is important in determining the compensation
budget. Table 1 summarizes the prevalence of these factors for industries relevant to those
covered in the Cleveland Community Salary Survey (CSS), the source of the data analyzed in
this study. While the factors considered vary somewhat among worker categories and industries,
several conclusions can be drawn fiom the table. First, area wage surveys constitute the single
most influential factor in compensation budgeting for workers such as those typically covered in
the CSS.3 Second, this list of indicators clearly picks up labor supply and demand conditions as
well as the inflationary environment. Third, to an economist's eye, the list also emphasizes the
limited information available to firms as they set wages. In an uncertain environment,
interemployer variation either in the factors chosen for determining wages or in their reading of
these factors could contribute substantially to variation in wage growth rates.
8. Statistical Implementation

If firms foresaw all necessary adjustments and relied completely on wage scales specified
by job characteristics (or on a point system based on job characteristics)--adjusted to market
wage rates--then any individual's wage change could be decomposed as follows:
(1)

A(ln Wit) = whit = at + Fft + OOt+ E,

in each labor market,

For unionized employees, negotiations on more detailed terms and conditions of pay increases (or
reductions) take place further in advance because contracts typically last about three years. Nevertheless, the firm
completes a prospective compensation budget, similar to nonunion budgeting, prior to negotiations in order to
establish the acceptable range of wage adjustments. The data analyzed here include very few unionized
employees, because of the occupations surveyed. However, because many of the establishments included are
partially unionized, spillover effects are possible.
Area wage surveys are the most commonly mentioned factor for nonexempt salaried workers. In cases
where other factors were cited more--union hourly employees (union demands) or executives and officers
(companies' financial reports)--area surveys are still frequently cited factors in establishing compensation
budgets. See also Levine (1993), which reports that in a survey of 139 compensation executives, wage change
recommendations rarely reflect unemployment rates, quit rates, and corporate returns on assets.

clevelandfed.org/research/workpaper/index.cfm

where wfoit is defined as the change in log wages of worker i, in occupation o, in firmf, at time t.
While an unconstrained tit could obviously spec^ all wage changes, meaningful underlying
concepts are identified by location, firm, and occupation components. General wage increases
are picked up by at (change of the local log wage baseline). If wage inflation rates do not vary
by locality, then this term equals the national rate of wage inflation.

Fft represents the change in

firmf s "market position" at period t; a positive Ff, marks a decision to increase employers7
overall pay relative to the general market. In the compensation literature, firms are generally
viewed as maintaining their average wages at a fixed deviation from other local employers' offers
for a variety of reasons, i.e., Ff, is typically 0.4 Next, Oot is the change in the occupational
differential for workers in occupation o. Competition among firms for employees with specific
occupational skills tends to equalize both the levels and the changes in these differentials across
firms. Individual-specific adjustments (tit) include merit and longevity raises.
Lacking full information on the year's realizations of a, and O0, firms look primarily to
each other and to public measures of inflation for guidance, so they may make errors. If we
modifjr equation (1) to allow for mistakes, wage changes become more complicated:
(2)

wfoit = at +a>

+

Ffi+ Oot + O;ot

+ &it,

where a> and O;ot represent realized employer errors in determining the current local and
occupational wage adjustments. The timing of the payroll year may also result in firms leading or
lagging their desired market position at a particular date, an outcome that we consider simply
another form of error in the firm's attempt to match the local inflation rate. Note that all
equilibrium wage adjustments can still be described by varying ci,.
Ideally, we would use individual wage adjustments gathered from a large array of
employers to identifjr the components in equation (2); however, in most years, the CSS records
wages not of individuals, but as means or medians for "job cells" which specifjr the location,

Groshen (1991~)discusses the various explanations for observed wage variation among employers.
These include systematic human capital differences, compensating differentials, errors, efficiency wages, and
rent-sharing. AU of these reasons, except errors, are long-term strategies.

clevelandfed.org/research/workpaper/index.cfm

employer, and occupation. Aggregating the individual wages in equation (2) to job-cell averages
poses no inherent problem because it simply aggregates the individual-specific error by cell.
Thus, the structure of the CSS allows decomposition of wage changes into four terms at most:
employers, occupations, cities, and residuals. Specifically, we estimate these components via the
following fixed-effects regression:
(3) wfo = a + P D + y D o+ CLJ.,,for each locality and year,
where f3 and y are coefficient vectors for matrices of dummy variables (Dl and Do) referring to
the cell's firmand occupation, respectively.
The possibility of firm-specific errors for occupations that we highlight in equation (2)
(i.e., O;ot) means that we cannot confidently assume that the coefficient vector P provides
unbiased estimates of the FP's in equation (1). Furthermore, the lack of restrictions on the
individual-specific term (sit) will confound the direct correspondence between equations (1) and
(3) if correlations of the cell mean (or median) with firms or occupations exist. While we have
reason to believe that these biases are small, we need to clarifjr the nature of the potential misidentifications by the fixed-effects estimation in equation (3) in order to guide robustness checks
of our findings.
The primary concerns in our application are O;ot and tit Applying a hypothetical
regression of occupation and employer dummy variables on the unobserved term (Oyot) would
allow identification of linear employer and occupation components along with a residual. These
hypothetical terms (which will be identified by hats,

and 670co)allocate the misidentified

variation. Similarly, the individual differences term (sit) can be decomposed into the employer
and occupational terms. After we allocate and bracket these terms according to which
coefficient they would affect, equation (2) becomes the following:

clevelandfed.org/research/workpaper/index.cfm

Of the four influences on the

vector of firm effects, a> and

a&, represent firm errors

(or timing differences). We conjecture that the two other terms are small, and we attempt to
ferret out the robustness of our conclusions to this assumption. As we stated earlier, the
compensation literature argues that Ff, is small because employers make long-run decisions on
the quality of employee desired. The fourth term, sf,, , represents the bias due to the firm's
workforce composition. To the extent that employers report wages for many workers and that
changes in worker skill levels offset each other (i.e., have a sampling mean of zero), this term will
vanish.
The estimated occupational coefficients (the y7s)represent an agglomeration of market
responses (0,) and biases common to a particular occupation across firms

and B , , ) .

We expect both of the bias terms to be small when there are many independent employers for an
occupation and when the labor force within an occupation has changed minimally over the year.
Furthermore, the occupation-specific component of wage adjustments is, arguably, primarily an
intentional market outcome. We believe (but not strongly enough to forgo robustness checks)
that the intentional responses should predominate.
A priori, there is no reason to expect any particular relationship between variability in

these relative wage terms and inflation -- the scaling effect of inflation and real wage growth has
been removed by the intercept in the log wage specification.

3. Inflation and Wage Adjustment
Extensive literatures describe reasons why relative prices can be altered by purely
nominal shocks. However, no research has been applied to the relative wages we consider here.
We first review these literatures, then outline our strategy for interpreting hypotheses in this
context.

a. Wage Rigidity Studies--Inflation as Grease
Keynesian macroeconomics depends heavily on the assumption of downward nominal
price andlor wage rigidity; that is, recessions occur when such stickiness prevents markets fiom

clevelandfed.org/research/workpaper/index.cfm

efficiently allocating resources. Keynes explained the stickiness by asserting that workers'
notions of fairness make real wage erosion, imposed by idation, more acceptable than nominal
cuts. Thus, general wage and price inflation can be a mechanism to reduce cyclical
unemployment and raise economic efficiency.5 An important corollary of this reasoning,
developed by Slichter (see Slichter and Luedicke [1957]) and Tobin (1972), argues that even
without large shocks, moderate rates of inflation can "grease the wheels" of the economy,
facilitating downward real price changes in response to small shocks. While the neo-Keynesian
perspective appears to favor sticky goods prices over sticky wages as the explanation for
monetary non-neutrality, this is partly due to empirical concerns about the rigidity of wages (see
Ball and Mankiw [ 19941).
Within the wide variety of studies that look for empirical evidence of wage rigidity, the
largest group examines aggregate real wages for evidence of procyclicality and concludes that
real wages are indeed rigid downward (see review in Fischer [I98 11). Other studies examine
household or employer microdata and mostly reach opposite conclusions. Although Holzer and
Montgomery (1990) detect some downward rigidity, most recent micro studies (Bils [1985];
Solon, Whatley, and Stevens [1994]; McLaughlin [1991]; and Lebow, Stockton, and Wascher
[1993]) find evidence of substantial nominal wage cuts, which they take as proof that wages are

flexible downward.6 The discrepancy between aggregate and micro results is attributed to
composition bias, the impact of overtime and bonus pay, and worker mobility.
We argue that, more important, the existence of nominal wage cuts does not in itself
demonstrate that wages are flexible; meaningfbl wage rigidity occurs when wages do not adjust

adequately to ensure efficient allocation of resources. We seek to improve on the direct
observation of wage adjustments by looking for evidence of meanin@l wage rigidity. Hence,

Three theories of "fairness"have been advanced to explain why unemployed workers cannot bid down
wages in a Keynesian recession: implicit contracts, efficiency wages, and rent-sharing models. Haley (1990)
presents a modern review of the microeconomic theories that predict Keynesian-type wage rigidity.
Another group of empirical efforts takes the unusual approach of sweying employers directly about
compensation practices; see, for example, Blinder and Choi (1990), Kaufinan (1984), Bewley and Brainard
(1993), and Levine (1993). These studies uniformly suggest that "fairness" is an important governing principle in
wage-setting practices, and that employers refrain from nominal wage cuts except under extreme duress.

clevelandfed.org/research/workpaper/index.cfm

we identifl and test for an important implication of wage rigidity for the labor market--that is,
whether higher inflation facilitates the adjustment of interoccupational (relative) wages to
shocks.7 To state this another way, we look for limited relative wage adjustments during periods
of low inflation.
b. Relative Price Disruption Studies--Inflation as Sand

Whereas the wage rigidity story describes how inflation might facilitate necessary price
changes among different goods in the market, the relative price disruption story describes how
inflation and pricelwage rigidity may cause inefficientfluctuations in prices. In these stories,
inflation entails variation in agents' price adjustments, distorting relative prices. The sources of
pricelwage rigidities posited in the price dispersion literature are menu costs (i.e., expenses for
revising price lists, as in Sheshinski and Weiss [1977]) or consumer search costs (Stigler and
Kindahl [I9701 and Reinsdorf [1994]). Both imply that inflationary price changes are unlikely to
be transmitted uniformly and instantaneously. Such distortions cause market participants to
confbse adjustment lags with real shocks, and thereby to misallocate resources and increase risk
(Vining and Elwertowski [1976]). In this scenario, inflation acts like sand in the gears of the
economy, impairing the interpretation of price signals.
Price dispersion studies measure the extent to which inflation is unevenly distributed and
use this to gauge the costs of inflation. Early studies in this genre uniformly show that inflation
raises the dispersion of price change indices and industry wage change aggregates. Fischer
(1981) and Cukierman (1983) review and extend these studies. This literature is subject to some
important limitations. First, as Hartman (1991) shows, increasing price variability with inflation
could be an artifact of constant expenditure shares. Second, the sand theory is most compelling
in arguing that inflation distorts price relationships among similar or competing goods, rather
than among the dissimilar goods represented by price aggregates.

Lebow, Stockton, and Wascher (1993) also address this issue in their study.

8

clevelandfed.org/research/workpaper/index.cfm

Recent research on price adjustment delves into this relationship using product price
microdata. One group of studies considers price changes in a single class of goods, generally for
low-inflation countries. For example, Cecchetti (1986) studies magmines' cover prices. Other
research explores price changes in broader product categories in high-inflation environments.
For example, Lach and Tsiddon (1992) study the variation of adjustment to food prices in storelevel data fiom Israel. The proprietary nature of micro-level price data limits the broad
applicability of any particular study in this genre, since the results are only for high- or lowinflation countries and for unusual or regulated products. Nevertheless, on balance, the studies
suggest that higher inflation increases the variability of price changes. For the United States,
during the high and declining inflationary years (1980-82), Reinsdorf (1994) finds that the
variation of monthly actual prices within product category (rather than indices or price changes)
rose as inflation fell, due to negative inflation surprises. The variation of price changes, however,
was positively correlated with inflation.
With respect to wages, Hamermesh (1986), Drazen and Harnermesh (1986), and Allen
(1987) find that the cross-industry dispersion of wage-change aggregates falls as inflation rises.
They attribute this result to inflation-induced introduction of indexation, formal or informal.
Card (1990) reaches similar conclusions in a study of inflation's impact on wages set in longterm union contracts. Transaction-level analysis of adjustments is particularly important in labor
markets because the composition of the worldorce certainly varies over the business cycle.
This study explores the impact of inflation on the dispersion of a crucial price--labor. By
controlling for detailed occupation, we effectively replicate the comparability across goods
(intramarket variability) sought in the product price literature. Aside fiom adding a rnicro-level
wage study to the literature, we extend price dispersion analysis by covering a broad array of
prices across the varied inflation history of the United States fiom the 1950s to the 1990s.
c.

The Impact of Inflation on Wage Adjustment

Our estimates of the terms in equation (3) should yield direct information on whether
inflation is grease or sand (call these "Story G" and "Story S," respectively). Since these stories

clevelandfed.org/research/workpaper/index.cfm

are not mutually exclusive, it is possible for either, neither, or both to be true, or they could
operate over different levels of inflation.
Story G, which pertains to the wage adjustments of firms with limited downward
flexibility, can be described by the following firm's decision problem:
min, Ea(wi-wi*)
s.t. (1) Eiwi I W
(2) wi 2 c v
' i'

{budget)
{downward wage rigidity)

The firm's goal is to match the market's (or, more generally, some desired) wage movements.
We model this as minimizing the weighted sum of differences between wage change offers (wi)
and the desired wage changes (wi*), in the context of an overall wage budget (W) and a rigid
wage constraint (c). Without solving for first-order conditions, the two constraints are
potentially in conflict. However, when c is a nominal figure (such as 0) and the other parameters
respond to inflation, fewer individual wage changes are subject to rigid wage constraint. Thus,
inflation relaxes wage rigidity constraints.
Interestingly, downwardly rigid rules may also constrain wage raises during periods of
low inflation. When the compensation budget (constraint 1) binds, it limits wage adjustments to
those that can be balanced elsewhere. Thus, each occurrence of a wage constrained to exceed
wi* must be made up on other wages. While the traditional story of rigid wages stresses the
unemployment consequences, a firm might choose to limit higher-than-average desired increases
rather than lay off workers. The simple conclusion we note is that binding downward wage
rigidities reduce the variance of wage adjustments in two ways: first, by eliminating many wage
cuts and second, by restraining increases in order to balance the compensation budget. These
restrictions will be evident in intentional components of wages that require occasional,
substantial adjustments. An obvious candidate in equation (4) is occupational wage adjustment,
OOP

We propose a simple version of Story S: Employers offer different prices for similar
goods in high-inflation periods because they disagree on the expected rate of local wage

clevelandfed.org/research/workpaper/index.cfm

inflation.8 That is, firms' compensation administrators err more often in calculating the "correct"
adjustments as inflation rises because their uncertainty about inflation rises simultaneously.
Widespread reliance on employer salary surveys (rather than direct measures of inflation, such as
the CPI or GDP deflator) confirms compensation managers' concerns over matching
competitors' actions rather than matching an easily observed level of goods inflati~n.~
Uncertainty in market wage adjustments may well exceed that of the goods markets due to the
limited samples, retrospective nature, and infi-equency of salary surveys. Story S is indicated by
growing dispersion among employers' forecasts (i.e., larger a;t and 0;)as inflation rises.

d The Impact of Labor Productivity Increases on the Model
Finally, the analysis below incorporates the realization that general increases in labor
productivity have the same institutional impact on wage adjustments as inflation. Since broadbased productivity increases shift out the demand for labor, employers observe other companies'
productivity-based adjustments and include them in nominal firm-wide wage adjustments in the
same way as they do inflation adjustments. Thus, productivity increases can substitute for
inflation in both the grease and sand stories. In light of this, our independent measure of wage
change (dMRP, the aggregate increment in the marginal revenue product of labor) is the sum of
change in output prices (CPI-U) plus the general increase in labor productivity (outputlhour).
Ceteris paribus, this sum should approximate the average nominal wage growth in the economy
and leave relative wages unaltered.
This point has policy implications to the extent that the grease and sand relationships, or
their welfare costs, are nonlinear. Suppose, for example, that story G is true, and the beneficial
impact of rising inflation (plus productivity) has a negative second derivative. In that case, the
benefits provided by the additional grease due to inflation are diminishing in environments with

8 ~ contrast,
y
if employers were to agree on some expected inflation rate that proved incorrect, this rate
wodd effectively operate as the true rate and would not distort relative wages among the individual firms.
This focus makes sense because of regional divergence in wage levels and relativities (and the lack of
precision of local CPIs), and because goods price movements understate average nominal wage changes by the
growth of labor productivity. Indeed, a firm could well be worse off competitively if it were the only one to
correctly forecast and incorporate a higher- or lower-thanexpected inflation rate into its wage bill.

clevelandfed.org/research/workpaper/index.cfm

rapidly growing productivity. Since productivity growth is indisputably beneficial beyond the
factors considered here, we focus on the role of inflation while controlling for the productivity.
e. Hypothesis Testsfor Grease and Sand

In summary, the nature of the patterns of wage adjustment depend both on the validity of
the "story" and on employer reactions. A priori, though, we know little about the precise
functional forms exhibited by these two general relationships. They may be flat or steep; they
may accelerate or taper OK Using the standard deviations of estimated wage-change coefficients
from equation (3) to measure the dispersion of wage-change components, we test two
propositions:
1. If Story G is true, as inflation rises from zero, the ability of employers to adjust
occupational wage differentials (O,3 grows, so the dispersion of the measured

occupational adjustment coefficients (estimated in y) grows.
2. If Story S affects the labor market, the disagreement among employers (a> and 0;)
should be higher in years of higher wage inflation. Hence, the dispersion of employerwage adjustment coefficients (estimated in Ip) grows.
Other factors could also affect the standard deviations of these components. In
particular, large demographic shifts or more rapid difision of technology could alter the
intensity of pressures for interoccupational wage adjustment. The propositions described above
and the analysis presented below implicitly assume that the pressures for these changes are
uncorrelated with our measures of inflation.

4. Description of the Data
Most previous studies of inflation's impact on prices examine price indices or industry
aggregates rather than actual prices. This paper examines annual changes in mean wages for a
panel of occupations within firms, i.e., job cells. Only a few publicly available wage data sets
provide information on employers, and none of these offer occupational detail plus a long

clevelandfed.org/research/workpaper/index.cfm

period.'O This study uses a data set with both desired features, constructed fiom an annual
private wage and salary survey conducted in Cleveland, Cincinnati, and Pittsburgh by the
Personnel Department of the Federal Reserve Bank of Cleveland (FRBC) for at least 38 years.
The purpose of the survey is to assist in annual salary budgeting at the FRBC. In return for their
participation, surveyed companies receive result books for their own use.
Table 2 describes the basic dimensions of the CSS wage-change data set. The complete
CSS data set has 80,301 job-cell-years of mean wage observations." From these data, we
compute annual wage changes for each job cell observed in adjacent years, creating a total of
67,885 wage-change observations.12Each observation gives the change in the log of the mean or
median salary for all individuals employed in an occupation by an employer in the city.13 Cash
bonuses are included as part of the salary, although fiinge benefits are not. From 1956 though
1992, wages increased at the rate of about 5 percent per year, with a standard deviation of 0.083
log wage points.
Participants in each city are chosen by the FRBC to be representative of employers in the
area. The number of companies participating on an ongoing basis has grown over time fiom 66

losee Hotchkiss (1990) for a summary of data sets with information on employers. For example, the
microdata collected in Industry Wage Surveys and Area Wage Surveys by the Bureau of Labor Statistics have
occupational detail but are not easily linked over time or preserved for long periods. Unemployment Insurance
ES-202 data, when available, report individds' earnings, not wages, and lack occupational detail. The
Longitudinal Research Database, maintained by the Center for Economic Studies, goes back to 1972, but covers
only manufacturers and provides only mean establishment earnings for production and nonproduction workers,
with no occupationaldetail.
llUnfortunately, books for some cities in some years were not found. Thus, the data set does not include
observations on those cities in those years. No observations are available for 1966 and 1970.
12~ob-cell-year
observations where the calculated change in log wages exceeds 0.50 in absolute value are
deleted from the sample on the assumption that most of these arise from reporting or recording errors. This
eliminates 193 observations. It also considerably reduces the variance of wage changes without causing any
qualitative change in the estimated coefficients reported here. Approximately 1,000 observations are imputed
from cases where job-cells are observed two years apart. The imputed one-year changes are simply half of the
two-year differences. Many of the results reported here were also run without the imputed observations. Their
inclusion does not affect the results.
I3Medians were recorded from 1974 through 1990. Since medians should be more robust to outliers, our
results use means through 1974 and medians for the years thereafter. Comparison of the coefficients estimated
separately for means and medians for the years where both were available (1974 and 1981-1990) suggests that
they are highly correlated (correlation coefficients of .97 to .99). However, coefficients estimated on the medians
appear to show more variation than those estimated on means and are more highly correlated over time. The
latter two characteristics are consistent with medians being a more robust measurement of central tendency.

clevelandfed.org/research/workpaper/index.cfm

to 96 per year. Cincinnati companies usually make up about one-quarter of the sample, with
Cleveland and Pittsburgh evenly represented in the balance. Overall, 192 companies have
participated in the survey at one time or another, for an average of just under 13 years each (with
individual companies' participation ranging from one to as many as 35 years). The number of
participating employers per year is shown in table 3.
Each participating firm judges which of its establishments to include in the survey,
depending on its internal organization. Some include workers in all branches in the metropolitan
area; others report wages for only the office surveyed. The discussion below uses "employer," a
purposely vague term, to mean the employing firm, establishment, division, or collection of local
establishments for which the participating entity chooses to report wages.14
The industries included vary widely, although the emphasis is on obtaining employers
with many "matches," i.e., employees in the occupations surveyed. The employers surveyed
include government agencies, banks, manufacturers, wholesale and retail trade companies,
utilities, universities, hospitals, and insurance firms. These are generally large employers.
The number of occupations surveyed each year ranges from 43 to 100. (In this analysis,
each occupation in each city is counted as a separate occupation; thus, the total number of
"occupations" exceeds the number surveyed.) On average, each employer reports wages for 27
occupations per year. The surveyed occupations are almost exclusively nonproductionjobs,
because these are the jobs that can be found in all industries. They include office (e.g.,
secretaries and clerks), maintenance (e.g., mechanics and painters), technical (e.g., computer
operators and analysts), supervisory (e.g., payroll and guard supervisors), and professional (e.g.,
accountants, attorneys, and economists) occupations. Many of the occupations are divided into
a number of grade levels reflecting different degrees of responsibility and experience. Job
descriptions for each occupation are at least two paragraphs long.

14Since a participant's choice of the entities to include presumably reflects those for which wage policies
are actually administeredjointly, the ambiguity here is not particularly troublesome.

clevelandfed.org/research/workpaper/index.cfm

One reasonable concern is that the survey could be an unrepresentative sample of the
cities' employers. This was checked by comparing wages in the survey to the Bureau of Labor
Statistics' @LS) Area Wage Surveys (AWS) in the same years for the same cities. The AWS
also oversamples large employers. Movements of mean wages for similar occupations were
highly correlated across the two surveys, and levels were usually within 5 percent of one another.
Although the survey has been conducted annually, the month for which data are collected
has changed several times since 1955. Throughout this paper, we observe the following
convention: Results for any year refer to the period of time between the preceding survey and
the one conducted in that year. In most cases, this is a 12-month span, but occasionally the
period is less or more than a year. The appendix lists the periods included in each "year" of the
CSS. All data merged in have been adjusted (to the extent possible) to reflect time spans
consistent with those in the CSS.
Cleveland, Cincinnati, and Pittsburgh are more urban, have more cyclically sensitive
employment, and have undergone more industrial restructuring than the nation as a whole. Prior
to the 1980s, wages in these three cities were higher than the national average, but now they are
approximately average for the country.
We also use standard measures of inflation and national output per hour in our analysis.
As a measure of general inflation experienced in the country, we use percentage changes in the
monthly averages of the BLS consumer Price Index for all Urban Workers (CPI-Q.15 Our
productivity measure is the BLS nonfarm business sector output per hour worked.
Annual mean log wage changes for each city appear in table 3. Although some variations
are evident, mean wage changes among the three cities are highly correlated. But do they bear
any relation to national trends? Figure 1 plots the three-city mean log wage change over time,
along with a simple measure of wage flexibility derived from equation (2). This variable, labeled

dMRP, equals the sum of idation (CPI-U) and aggregate labor productivity. CSS mean wage

15Experimentswith the individual city CPIs yielded very similar results. For ease of exposition, we
report only the results obtained with the national CPI.

clevelandfed.org/research/workpaper/index.cfm

changes trended steadily downward from 1957 to 1966. During that time, CSS raises exceeded
increases in the CPI-U, but productivity was growing fairly rapidly. From 1966 through 1973,
wage and CPI-U increases both accelerated (with wages leading the way), despite the 1969-70
recession and a brief respite caused by the imposition of wage and price controls in 1971. The
relaxation of controls and the oil embargo in 1974 were followed by a dramatic spurt of wage
and price increases, which then subsided until 1978, when price increases again reached into the
double digits. Average productivity also dropped during four years of the seventies, and CSS
wage increases did not keep up with prices. The 1980 and 1981-82 recessions ushered in a
period of declining wage and price increases during which CSS wages grew faster than inflation.
In 1987, price increases reversed trend and jumped ahead of CSS wage gains, peaked in 1990,
and are now headed back down.
The institutional model presented above suggests that expected or perceived changes in
the cost of living and productivity are employers' primary considerations in the structuring of
annual wage increases. From a budgetary standpoint, both inflation and productivity increases
represent sources of revenue for compensation. Figure 1 confirmed the general synchronization
of CSS wage changes with general price increases and productivity gains. This observation can
be formalized with overall correlations among the indicators charted in figure 1 (plus expected
inflation), as shown in table 4. The correlations between mean CSS wage adjustments and the
CPI-U and dMRP (0.84 and 0.74, respectively) are quite high. But changes in labor productivity
are negatively, not positively, correlated with wage changes over these four decades. This
anomalous correlation has been noted before and is due to the high-inflation, low-productivity
recessions of the seventies and early eighties.16 Nevertheless, figure 1 and table 4 support the
characterizations made here about the process of wage setting during the period.
This figure demonstrates that wages in the CSS largely adhere to national trends, and
thus may enlighten us about the behavior of wages in the nation as a whole. It also puts the

16SeeEberts and Groshen (1991) for an example of similar results.
16

clevelandfed.org/research/workpaper/index.cfm

remainder of the analysis in some historical perspective, lest we forget the major influences on
wages and prices that underlie our research.

5. The Components of Wage Adjustment
In section 2, we argued that wage adjustments have significant, distinguishable employer
and occupation components. In the present section, we verie that assertion empirically, with an
analysis of variance (ANOVA) of wage adjustments of the CSS sample (see table 5) based on
equation (3). The first column lists sources of variation, and the second column lists each
source's degrees of fieedom. The data include three cities over 37 years. Thus, average annual
city-year wage changes absorb 103 degrees of freedom. Since three cities are represented in the
sample, occupation and city are interacted (accounting for 5,358 degrees of freedom) to avoid
restricting all three cities to have the same occupational wage movements. Employers' mean
annual wage movements absorb another 2,77 1 degrees of freedom.
The third column lists each source's marginal contribution to the model sum of squares
(over the contributions of the sources listed above it on the table). We choose this method of
presentation because of its parsimony when the data are unbalanced (i.e., the number of
observations in each group is not fixed). The results are similar to a stepwise regression. Ifjoint
effects are large (such as between occupation and employer in wage levels, as shown in Groshen
[I99 la, 1991b]), the order of presentation is crucial and a stepwise presentation can be
misleading. Surprisingly, estimates of joint effects among these sources of wage-change variation
(particularly occupation and employer) are minuscule; thus, the order of presentation is not
qualitatively important. Introduction of occupation after employer would change little in this
table.
All together, the model accounts for 27.9 percent of the variation in annual wage
adjustments. That is, the R~ for the regression shown in equation (3) is .279. The residual
variation in wage changes is presumably due to compositional changes and individual merit
raises. The fifth column of the table shows that slightly more than one-fifth of the equation's

clevelandfed.org/research/workpaper/index.cfm

explanatory power is due to changes common to all job cells in each year. Intercity differences,
while statistically significant, do not account for much of the variation here. Occupation-wide
changes, on the other hand, constitute more than one-quarter of observed variation. By far the
strongest effect is employer-wide changes, which account for almost half of the explained
variation and 13.1 percent of total variation. F-statistics for these five sources of variation are all
significant at the 1 percent level or less.
This decomposition suggests that the institutional model described above may provide a
usehl fiarnework for understanding wage adjustments. We do observe distinguishable
occupation-wide and employer-wide variations in wage changes. In particular, the firm-wide
wage movements are interesting because they are such a large component and because employer
wage differentials are generally quite stable (Groshen [1991b]), suggesting that these may be
errors and corrections.
To the extent that employers consider their company-wide and occupational wage
adjustments separately and the relevant information for the two come from independent sources,
the standard deviations of these two wage change components will be uncorrelated over time.
Table 6 presents correlations of the annual standard deviations of the three components of wage
changes, pooling the three cities together. These correlations are all positive, suggesting that
some factors affect the variation of the components similarly. However, as the model suggests,
the intertemporal patterns of dispersion of the employer and occupation components are only
moderately correlated. The higher correlations of the standard deviations of the residual and
occupational components suggest that some adjustment of occupational differentials may not
occur uniformly across employers.17

l7 Alternatively, such shifts in relative wages among occupations may stimulate simultaneous
compositional changes among job cells. That is, employers may accompany adjustments in relative wages with
some occupational reorganization of their work€orces.

clevelandfed.org/research/workpaper/index.cfm

6, Inflation's Impact on Employer and Occupation Wage Adjustments
With this evidence that the proposed framework is reasonably consistent with the CSS
data, we turn to asking how these sources of wage adjustment dispersion vary with inflation. All
specifications use the ANOVA results shown in table 5. In each city-year, the total variation of
wage changes was decomposed into three components, as shown in equation (3): occupationwide changes, employer-wide changes, and the residual. In this second stage, we regress the
standard deviation of the components on the level of general wage increases (the city-year mean
CSS wage adjustment) or the sum of inflation and productivity increases (dMRP). For brevity
and because we have few unambiguous predictions for the behavior of the composite residual
term, we report results only for the occupation and employer components of wage changes. To
confirm the robustness of our findings, we also perfbrm nonparametric, filtered, and panel
versions of these tests. Each of these enhancements is discussed in turn.

a The Basic Relationship
Table 7 presents the results of basic quadratic regressions of our two dependent variables
(the standard deviation of employer and occupation wage adjustments, whose means are shown
in table 6) on the two proxies for overall wage movement. To assist the reader with the slopes
over the relevant range, we report the implied value of the independent variable at the maximum
or minimum at the bottom of each table.
Column 4. of table 7 suggests a U-shaped (with a minimum at 2.2 percent) relationship
between the dispersion of employer wage adjustments and the city-year mean. In contrast,
column 2 suggests an inverted U-shape (peaking at 13.4 percent) between employer
disagreement and dMRP. Interestingly, mean CSS adjustment--an internal measure of wage
change--has less explanatory power (lower R-squared) than dMRP--the external measure.
While neither quadratic term is independently significant, in both cases the combination of
the two terms is significantly related, as indicated by the F test for joint significance. Thus, while
the exact specification is not strongly supported, there is a clear relationship between the level of
inflation and the standard deviation of employer wage adjustments. Indeed, plots of the

clevelandfed.org/research/workpaper/index.cfm

predicted relationships (as discussed below), over the range in which we actually observe mean
CSS adjustment and dMRP (3 to 10 percent), show that the shapes described in columns 1 and 2
track each other fairly well; both slope upward fairly steeply. These results suggest that
employer disagreement (including their errors--our measure of the sand story) rises substantially
as mean nominal wage increases rise from 3 to 10 percent. However, the lack of consistency
between the coefficients on internal (mean CSS change) and external (dMRP) measures of wage
increases tempers our ability to say whether the disagreement tapers off or accelerates at higher
rates of inflation.
Columns 3 and 4 apply the same analysis to occupational wage changes, estimating the
amount of grease added to the system by inflation. In this case, estimates based on the internal
and external measures agree on a statistically significant, inverted U-shaped relationship
(maximized at 10 to 11 percent) between mean wage changes and occupational wage
adjustments. Again, the external measure (dMRP) appears to have more explanatory power.
Indeed, both terms in the dMRP specification are significant, providing fairly strong, consistent
evidence that while inflation may grease the wheels of occupational adjustments, any benefits are
limited.
The relationship between dMRP and the variability of these two components of wage
aaustment is best shown graphically. The graphs also allow us to confirm that the quadratic
functional form imposed in the basic regressions is reasonable, because we plot nonparametric
estimates of the relationships alongside the predictions of the parametric regression.
Figures 2A to 2D plot the implied relationships shown in table 7, along with
nonparametric regression predictions. l8 These comparisons suggest that the standard deviations
of employer and occupational wage adjustment both increase with dMRP and the CSS mean, and
that the basic quadratic specification we employ describes the shapes of the functional

l8 We choose the LOWESS smoother with a bandwidth of one, proposed by Cleveland (1979), for its
robustness with respect to both axes. Various bandwidths for 0.2 to 1 were tried, with little variation in effect.
Cleveland recommends a bandwidth of 1, due to the tricube weighting already included in the LOWESS
technique. See H&dle (1990) for comparisons of nonparametric regression techniques.

clevelandfed.org/research/workpaper/index.cfm

relationships reasonably closely. The frequency of observations is partly indicated by the density
of tick marks for the nonparametric regression: Tick marks are plotted at each observation, but
some overlap. Generally, the nonparametric regressions confirm the parametric results; however,
the potential importance of outliers in this specification is made clear in figure 2D.
Over the observed range of dMRP and mean CSS changes (3 to 10 percent), each of the
plots indicates a positive relationship. The upward slope for employer variability appears
markedly steeper than that for occupational variability, particularly at higher rates of inflation.
We also note that although the standard deviation of occupational wage changes reaches a
maximum at mean wage changes of 10 to 11 percent, the curves' flatness suggests that little is
gained beyond rates in the neighborhood of 6 to 7 percent. Allowing for mean productivity
annual growth of about 2 percent, these results imply that any benefits conferred by inflation are
exhausted after rates of about 4 to 5 percent.
Under the model advanced above, our results suggest that the disruptive sand from
additional inflation (as measured by the standard deviation of employer wage adjustments)
increases rapidly as the level rises, while the potentially beneficial grease (as measured by the
standard deviation of occupational wage adjustments) shows a slower and even diminishing
relationship with nominal wage growth.
b. Filtered Results
A significant concern with the basic specification is whether the ANOVA in the first stage

correctly identifies the underlying factors we want. That is, are employer wage changes largely
short-term errors and corrections, while occupational movements are market-driven adjustments?
Equation (4) clearly indicates that undesired terms may creep into terms collected by the
ANOVA estimates. Though we give a number of specific reasons why we believe these
corrupting factors are small, we explicitly try to correct for them in this section.
We use the nature of the corrupting hctors to filter out their effects. Specifically, the
potential corruption to the employer component is the possibility that firms alter their long-term
"market position," a decision that is treated as uncommon in the compensation literature. This

clevelandfed.org/research/workpaper/index.cfm

suggests filtering out firms' long-term adjustments to emphasize their higher-frequency errors.
Similarly, occupationally correlated errors could corrupt the occupation components.
Eliminating high-frequency changes should leave a purer measure of the presumably longer-term
adjustments of the occupational wage structure to shifts in supply or demand.
We use the filters on the first-stage regression coefficients obtained fiom the ANOVA.
Then we calculate standard deviations for the filtered employer and occupation components and
run the same basic quadratic specifications. The results, shown in figures 3A to 3D, generally
confirm the results found for the unfiltered components. A minor exception is figure 3C, where
the filtered relationship turns down more steeply. The levels of variation in the filtered
components are lower, as would be expected. We take these results as confirmation of the
appropriateness of using the ANOVA procedure rather than replacing the unfiltered results,
because the filtering process undoubtedly eliminates much of the desired variation in the
components.
c. Panel Estimates

Alternatively, the skeptic may fear that the relationships we find stem from a spurious
correlation between inflation and wage-change variability, which could arise from some
employers adjusting wages biannually, or from sample drift. To address these issues, we obtain
panel estimates (rather than measuring associations between aggregates) because the panel
specification allows us to control for two classes of spurious correlation.
In contrast to the basic model, the panel estimates correlate the absolute deviations
(rather than standard deviations) of occupation and employer components of wage adjustments
with inflation. For the occupation regressions, we begin by calculating each occupation's
absolute deviation from the mean wage adjustment in the city and year. Employer absolute
deviations are constructed similarly. The mean of these terms (the mean absolute deviation) is
comparable, though not identical, to a standard deviation. We then regress the absolute
deviations of the cells on dMRP and two kinds of controls. The predictions of these regressions
are the conditional mean absolute deviations.

clevelandfed.org/research/workpaper/index.cfm

We estimate the following regressions:

I:[

abdev0ci.t =
abdevemjPt

+

]

8[abde~oci.t-~ gldMRPt
abdevemj,t-l
+

+

g2dMRPt2

,

where abdevocit and abdevemj, represent the absolute value of the occupation and employer
components for a given cell, the 6's are firm or occupation fixed effects, and t-1 indicates the
lagged dependent variables. The brackets indicate that those terms are included in only some of
the regressions.
In the simplest specification (without fixed effects or lagged terms), the panel estimates
roughly duplicate the basic results shown in table 7, because the inflation rate is the same for all
cells in a particular year. However, the panel data setting allows us to control for two key types
of extraneous covariation: correlation in firm decisions across adjacent years, and fixed
occupation or employer effects. Including fixed effects controls for some firms' or occupations'
long-run propensity to deviate more or less than others. Lagged terms control for the previous
period's adjustment in that occupation or firm. These controls should handle, for example, the
case of a firm that adjusts its wages only in alternate years--leading to an oscillation between
large positive and negative adjustments relative to other firms that adjust their wages more
frequently. The controls also account for sample drift in the survey's occupations or employers
over time.
Tables 8A and 8B show the results of these regressions for the occupation and employer
components, respectively. The reported regressions are for dMRP. Regressions using the
internal wage inflation variable are comparable. Specification 1 in tables 8A and 8B (the panel
equivalent of the regressions in table 7) provides a basis for identifying the impact of the
controls. Specification 2 includes lagged dependent variables. Specification 3 includes fixed
effects for the occupation or employers, in combination with the employer's city. Specification 4
includes both forms of controls.
Not surprisingly, since we are regressing a large cross-section of micro observations on a
single macroeconomic series, we obtain a very low R~ in specification 1. While the cell wage

clevelandfed.org/research/workpaper/index.cfm

components show tremendous heterogeneity, the aggregate relationships we detected earlier
between dMRP and the variation in employer and occupation wage adjustment components hold
in all specifications. From a statistical view, the correspondence between larger deviations
during periods of higher inflation and aggregate productivity is strongly confirmed at the firm
and occupation level.
Strikingly, even though adding controls improves the explanatory power of these
regressions, coefficients on dMRP and its square prove stable. While the coefficient estimates
vary somewhat between specifications, they are consistent with each other and with the previous
estimates. The qualitative impact of the specification changes can be noted in the bottom two
rows of each table, which show the implied slopes within the observed range of dMRP and

dMRP at the implied maxima. Table 8A shows that the slope of the predicted relationship falls
by less than 1 percentage point with the introduction of controls. The implied peak shifts back
slightly more, from about 7 percent to 5.2 percent. These results imply that the beneficial impact
of inflation may be exhausted at lower rates than those indicated in the basic model, but the two
sets of findings are otherwise consistent.
Similarly, panel estimates in table 8B support earlier indications that the employer
variation is even more strongly affected by inflation in the relevant range (implied slopes being
roughly twice those observed for interoccupational variability). Again, lags and employer fixed
effects have little qualitative impact on coefficient estimates. According to these results, the
disruptive sand caused by inflation continues to mount at least until dMRP levels of 8 to 12
percent--far beyond levels where the beneficial grease is maximized--and shows less evidence of
a turndown at high inflation levels.
In summary, the robustness of the results to these panel controls rules out a wide variety
of spurious correlations, increasing confidence in our basic results. We have tested more
explicitly whether job cell wage-change components deviate more when the level of inflation
allows more latitude for wage adjustments, and the results are affirmative.

clevelandfed.org/research/workpaper/index.cfm

6. Summary and Conclusions
This paper explores inflation's impact on the labor market with an eye toward
distinguishing positive effects (greasing the wheels by facilitating real wage adjustments to
shocks) from the negative ones (throwing sand in the gears by distorting relative wages). We
study wage changes in a panel of occupations and employers (from the Federal Reserve Bank of
Cleveland CSS) lasting from 1956 through 1992.
The analysis, governed by an institutional model of wage adjustments, focuses on
differences between the behavior of employer wage adjustments and occupation-wide
movements. We interpret the former as being more likely to include errors and corrections, or
deviations in speed of adjustment, while the latter has a higher concentration of market-driven
relative price adjustments. Relying on this distinction to interpret our results, we estimate the
relationship between the standard deviation of employer and occupation wage adjustments and
two measures (internal and external) of inflation. We also note that in this model, general
productivity increases play the same role as inflation, with the same costs and benefits.
However, since productivity growth, unlike inflation, has unambiguous benefits beyond the scope
of this exercise and is not a direct monetary policy target, we focus on the implications of our
research for inflation policy.
We examine the data in various ways to confirm the consistency of the model with
observables. In support of the model, we find that in the CSS: 1) As predicted by employers'
responses about how they determine wage levels, annual mean wage adjustments are highly
correlated with external measures of inflation and productivity growth. 2) An ANOVA of annual
wage adjustments among job cells suggests that employer and occupation components of wage
changes both play large, statistically strong, independent roles. 3) Over time, the standard
deviation of employer adjustments and occupation adjustments has a correlation coefficient of
0.475; this suggests that, while these two types of dispersion may have some common influences,

they often move independently of each other.

clevelandfed.org/research/workpaper/index.cfm

In our analysis of the relationship between inflation (along with labor productivity
increases) and these two kinds of wage change dispersion, we find the following:
(1) Is inflation grease? Higher inflation and labor productivity appear to increase the
range of occupational wage adjustments, although these potential benefits taper off
after inflation rates of about 4 percent (assuming labor productivity growth of 1.5
percent, the average rate over the period observed).
(2) Is inflation sand? Higher inflation and labor productivity are associated with higher,
potentially inefficient variation in employer wage adjustments. The variation between
employer wage adjustments rises about twice as quickly as occupational variation with
respect to inflation and shows less evidence of a turndown at high inflation levels.

Thus, we conclude by answering the question posed in the title with "yes, on both counts;
inflation can act as both grease and sand." Evidence from the CSS suggests that moderate
inflation (below about 4 percent) speeds the transmission of interoccupational wage adjustments.
But inflation also exacerbates potentially confbsing errors and corrections, or lagged
adjustments, in employers' wage policies. These costs of inflation have a steeper slope and a
later peak over the range observed in this study, suggesting that inflation's costs continue to rise
long after its potential benefits have been exhausted.
We think these findings add a unique micro-level perspective to aggregate-level research
on the relationship between inflation and productivity or income growth-studies that skip over
the mechanisms involved but presumably measure the net impact of grease plus sand on the
entire economy. Rudebusch and Wilcox (1994) review and extend these analyses on U. S. (and
international) time series data. They tentatively conclude that the level of inflation had a negative
correlation with productivity growth fiom 1954 through 1993, suggesting that the disruptive
impact of inflation outweighs benefits obtained from greasing the wheels.19
Since we do not consider impacts of inflation beyond the labor market, our study cannot
estimate inflation's net effect on overall productivity. However, within the labor market, our

l9 Interestingly, if monetary authorities acted as if they were aware of the relationships identified in this
study, they might be most likely to allow moderate inflation during periods of exogenously low productivity
growth. Such considerationswould also generate a negative correlation between inflation and productivity, as
observed by Rudebusch and Wilcox (1994).

clevelandfed.org/research/workpaper/index.cfm

study is the first to investigate the labor market mechanisms involved and to measure the relevant
ranges for the grease and the sand hypotheses simultaneously.
Suppose a monetary authority took our results at face value, neglecting other effects of
inflation.20 These findings suggest that optimal inflation targets depend on general labor
productivity growth. In times of high growth (say, over 4 percent), inflation's costs in the labor
market are virtually certain to outweigh its benefits, so inflation should be kept close to zero.
Only during periods of low productivity growth might the benefits of "greasing" the labor market
with mild inflation (5 percent or less) be supported.

20 We also finesse the problem of weighting costs and benefits in the welfare function in order to
determine a strategy. Arbitrarily, the rest of the paragraph assumes roughly equal weights between benefits and
costs, as measured by raising occupational and firm variability, respectively. Less symmetric weighting schemes
would shift the policy recommendations accordingly.

clevelandfed.org/research/workpaper/index.cfm

Appendix
Salary Survev Year
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992

Salarv Survev Coverage
March 1955 - March 1956
March 1956 - March 1957
March 1957 - March 1958
March 1958 - March 1959
March 1959 - March 1960
March 1960 -March 1961
March 1961 - March 1962
March 1962 - March 1963
March 1963 - March 1964
March 1964 - March 1965
March 1965 - March 1966
March 1966 -March 1967
March 1967 - March 1968
March 1968 - March 1969
March 1969 - March 1970
March 1970 - March 1971
March 1971 - March 1972
March 1972 - March 1973
March 1973 - September 1974
September 1974 - September 1975
September 1975 - September 1976
September 1976 - September 1977
September 1977 - September 1978
September 1978 - July 1979
July 1979 - August 1980
August 1980 - June 1981
June 1981 - June 1982
June 1982 - June 1983
June 1983 - April 1984
April 1984 - April 1985
April 1985 - April 1986
April 1986 - April 1987
April 1987 - April 1988
April 1988 - July 1989
July 1989 - July 1990
July 1990 - July 1991
July 1991 - July 1992

clevelandfed.org/research/workpaper/index.cfm

References
Allen, Steven G. "Relative Wage Variability in the United States 1860-1983," Review of
Economics and Statistics, vol. 69, no. 4 (November 1987), pp. 617-626.
Ball, Laurence and N. Gregory Mankiw. "A Sticky-Price Manifesto," National Bureau of
Economic Research Working Paper No. 4677, March 1994.
Bewley, Truman and William Brainard. "A Depressed Labor Market, as Explained by
Participants," Cowles Foundation, Yale University, mimeo (February 1993).
Bils, Mark J. "Real Wages over the Business Cycle," Journal of Political Economy,
vol. 93, no. 4 (1985), pp. 666-689.
Blinder, Alan S. and Don H. Choi. "A Shred of Evidence on Theories of Wage
Stickiness," Quarterly Journal of Economics, vol. 105 , no. 4 (November 1990),
pp. 1003-1015.
Card, David. "Unexpected Inflation, Real Wages, and Employment Determination in
Union Contracts," American Economic Review, vol. 80, no. 4 (September 1990), pp. 669-688.
Cecchetti, Stephen G. "The Frequency of Price Adjustment," Journal of Econometrics,
vol. 3 1 (April 1986), pp. 255-274.
Cleveland, William S. "Robust Locally Weighted Regression and Smoothing Scatter
Plots," Journal of the American Statistical Association, vol. 79 (1979), pp. 829-836.
Cukierman, Alex. "Relative Price Variability and Inflation: A Survey and Further
Results," Carnegie-Rochester Conference Series on Public Policy, vol. 19 (Autumn 1983), pp.
103-58.
Drazen, Allan and Daniel S. Hamermesh. "Inflation and Wage Dispersion," National
Bureau of Economic Research Working Paper no. 1811, January 1986.
Eberts, Randall and Erica L. Groshen. "Overview," in R. Eberts and E. Groshen, eds.,
Structural Changes in US. Labor Markets (ME. Sharpe: Armonk, NY, 1991), pp. 1-12.
Fischer, Stanley. "Relative Shocks, Relative Price Variability, and Inflation," Brookings
Papers on Economic Activity, vol. 12, no. 2 (1981), pp. 381-442.
Freedman, Audrey. The New Look in Wage Policy and Employee Relations, The
Conference Board, Report No. 865, 1976.
Groshen, Erica L. "Sources of Intra-Industry Wage Dispersion: How Much Do
Employers Matter?' Quarterly Journal of Economics, vol. 106, no. 3 (199 la), pp. 869-84.

clevelandfed.org/research/workpaper/index.cfm

Groshen, Erica L. "Do Wage Differences among Employers Last?'Federal Reserve Bank
of Cleveland, Working Paper 8906, June 1989 (revised 1991b).
Groshen, Erica L. "Five Reasons Why Wages Vary among Employers," Industrial
~elations,vol. 30, no. 3 (Fall 1 9 9 1 ~pp.
) ~ 350-8 1.
Haley, James. "Theoretical Foundations for Sticky Wages," Joumal ofEconomic
Suweys, vol. 4, no. 2 (1990), pp. 115-155.
Hamermesh, Daniel S. "Inflation and Labour-Market Adjustment," Economica, vol. 53
(February 1986), pp. 63-73.
Hiirdle, Wolfgang. Applied Nonparamefric Regression (Cambridge University Press:
Cambridge, 1990).
Hartman, Richard. "Relative Price Variability and Inflation," Journal ofMoney, Credit,
andBanking, vol. 23, no. 2 (May 1991), pp. 185-205.
Hills, Frederick S. Compensation Decision Making (The Dryden Press: Hinsdale, IL,
1987).
Holzer, Harry J. and Edward B. Montgomery. "Asymmetries and Rigidities in Wage
Adjustments by Firms," National Bureau of Economic Research Working Paper No. 3274,
March 1990.
Hotchkiss, Julie L. "Compensation Policy and Firm Performance: An Annotated
Bibliography of Machine-Readable Data Files," Industrial and Labor Relations Review, vol. 43,
no. 3 (February 1990), pp. S274-289.
Kaufhan, Roger T. "On Wage Stickiness in Britain's Competitive Sector," British
Joumal ofIndusfrial Relations, vol. 22, no. 1 (1984), pp. 101-112.
Lach, Saul, and Daniel Tsiddon. "The Behavior of Prices and Inflation: An Empirical
Analysis of Disaggregated Price Data," Journal ofPolitical Economy, vol. 100, no. 2 (1992),
pp. 349-389.
Lebow, David E., David J. Stockton, and William L. Wascher. "Inflation, Nominal Wage
Rigidity, and the Efficiency of Markets," Division of Research and Statistics, Board of
Governors of the Federal Reserve System, rnimeo, August 1993.
Levine, David I. "Fairness, Markets, and Ability to Pay: Evidence from Compensation
Executives." American Economic Review, vol. 83, no. 5 (December 1993), pp. 1241-1259.
McLaughlin, Kenneth J. "Rigid Wages?'unpublished working paper, University of
Rochester, April 1991.

clevelandfed.org/research/workpaper/index.cfm

Milkovich, George T. and Jerry M. Newman, Compensation (BPI-Irwin: Homewood, IL,
1990).
Reinsdorfj Marshall. "New Evidence on the Relation between Inflation and Price
Dispersion," American Economic Review, vol. 84, no. 3 (June 1994), pp. 720-73 1.
Rudebusch, Glenn D. and David Wilcox. "Productivity and Inflation: Evidence and
Interpretations," Federal Reserve Board, Washington, D.C., unpublished paper (May 1994).
Sheshinski, Eytan and Yoram Weiss. "Inflation and Costs of Price Adjustment," Review
ofEconomic Studies, vol. 44, no. 2 (June 1977), pp. 287-303.
Slichter, Sumner and Heinz Luedicke. "Creeping Inflation--Cause or Cure?" Journal of
Commerce, 1957, pp. 1-32.
Solon, Gary, Warren Whatley, and Ann Huff Stevens. "Real Wage Cyclicality between
the Wars: Evidence from the Ford and Byers Companies," University of Michigan, unpublished
paper, March 1994.
Stigler, George J. and James K. Kindahl. The Behavior ofIndustriaZ Prices, National
Bureau of Economic Research General Series No. 90 (New York: Columbia University Press,
1970).
Tobin, James. "Inflation and Unemployment," American Economic Review, vol. 62, no. 1
(March 1972), pp. 1-18.
Vining, Daniel R. and Thomas C. Elwertowski. "The Relationship between Relative
Prices and the General Price Level," American Economic Reviau, vol. 66, no. 4 (September
1976), pp. 699-708.
Wallace, Marc, J., Jr., and Charles H. Fay, Compensation Theory and Practice, (PWSKent: Boston, 1988).

clevelandfed.org/research/workpaper/index.cfm

Table 1

Factors Influencing Wage and Salary Budgets

Area surveys
Cost-of-living index
Corp. financial results
Corp. financial prospects
Internal equity among
employee groups
Worker productivity
Increases given by
industry leaders
Ability to hire
Nationally bargained
settlements
Union demands

Nonexecutive, Exempt
Manufacturing
Banks
Consumer Industrial
48%
41%
57%
24
39
30
31
45
50
41
37
30
27
15
27

Nonunion, Hourly
Manufacturing
Consumer Industrial
Banks
46%
40%
39%
26
25
18
21
19
30
16
18
30
24
10
8

15
30

16
35

34
13

9
23

5
11

20
3

15
6

19
5

11
1

8
20

11
17

6
3

10

4

-

15

17

-

Note: Multiple answers were allowed, so percentages do not sum to 100.
Source: Freedman (1976).

clevelandfed.org/research/workpaper/index.cfm

Table 2

Description of the Annual Wage Adjustment Data Set
Drawn from the CSS, 1956-1992

Total Number of Job-Cell Wage Adjustments Observed

67,885

Number of Years

36

Average Number of Observations Per Year

1,886

Mean Log Wage Adjustment

0.050

Standard Deviation of Log Wage Adjustment

0.083

Number of Occupations Ever Observed
Number of Occupation*City*Year Observations

166
5,27 1

Avg. No. of Occupation*City Observations Per Year

146

Number of Employers Ever Observed

192

Number of Employer-years
Average Number of Employers Per Year

Note: All numbers reported are for the first-differenced data set.
Source: Authors' calculations from the Federal Reserve Bank of Cleveland
Community Salary Survey.

27 16
75

clevelandfed.org/research/workpaper/index.cfm

Table 3

Description of Data by Year

1990
1991
1992
Total

2,505
2,536
2,187
68,839

222
223
222
5,462

84
89
80
2,875

0.052
0.038
0.039
0.050

0.0460.045
0.045
0.052

* In 1970-72, the CSS is missing Cincinnati; in 1970-73, the CSS is missing Pittsburgh.
Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community Salary Survey.

0.024
0.035
0.043
0.048

clevelandfed.org/research/workpaper/index.cfm

Table 4

Correlation Coefficients between CSS Wage Adjustments
and Relevant Economic Indicators

CSS Mean
Log Wage
Adjustment
0.839
(0.000)

Current
CPI-U

~ M R P *(CPI-U + Prod.)

0.737
(0.000)

0.86 1
(0.000)

Labor productivity*

-0.482
(0.003)

-0.60 1
(0.000)

-0.1 12
(0.5 10)

Mean
(Standard Deviation)

0.05 1
(0.022)

0.046
(0.035)

0.062
(0.028)

Current CPI-U"

dMRP

Labor
Productivity

0.015
(0.0 18)

*Percent change experienced during the period.
Note: Numbers in parentheses below the reported correlation coefficients are the probability
that the correlation coefficient equals 0. Total number of observations: 37.
Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community Salary
Survey.

clevelandfed.org/research/workpaper/index.cfm

Table 5

ANOVA of Annual Wage Adjustments
in the CSS, 1957-1992

Source of
Variation

Degrees
of
Freedom

Marginal
Contribution
to Sum of
Squares

Percent
of Total
Sum of
Squares

Percent
of Model
Sum of
Squares

Stepwise
F-Statistic

10.3
123.5
8.0
1.2
4.5

2
35
63
5,270
2,7 16

0.1
2717
3.2
37.5
60.4

0.0
6.0
0.7
8.1
13.1

0.1
21.5
2.5
29.1
46.9

Model
Residual

8,086
59,798

128.8
333.3

27.9
72.1

100.0

Total

67,884

462.1

100.0

City
Year
Year*City
Occ*Year*City
Employer*Year

*The three cities are Cleveland, Cincinnati, and Pittsburgh. The years are 1956-1957 through
1991-1992. Overall, 166 occupations are ever surveyed; in the ANOVA, each occupation is
counted separately for each city in each year. Similarly, a total of 192 employers are ever
observed.
Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community Salary
Survey.

clevelandfed.org/research/workpaper/index.cfm

Table 6

Standard Deviations and Correlations
of Components of Annual Wage Changes

Total

Occupation

Component:
Employer

Residual

Mean Standard
Deviation of Wage
Adjustments
(Std. Dev.)

0.0775

0.0273

0.0333

0.0670

(0.0 191)

(0.0 108)

(0.0 130)

(0.0 164)

Correlation of
Standard Deviation
with:
Occupation Std. Dev.
Employer Std. Dev.
Residual Std. Dev.

0.766
0.676
0.965

0.475
0.7 19

0.479

Note: All correlations are significant at the 0.0001 level. Total number of city-year observations: 104.
Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community Salary S w e y .

clevelandfed.org/research/workpaper/index.cfm

Table 7

Wage Inflation and the Standard Deviation
of Employer and Occupation Nominal Wage Adjustments

Model

Dependent Variable
Standard Deviation of Wage Adjustment
Components:
Employer
Occupation
2
1
4
3

Intercept

0.029
(0.006)

0.012
(0.007)

0.012
(0.004)

CSS Mean
Adjustment

-0.089
(0.176)

0.28 1
(0.122)

Squared CSS Mean
Adjustment

2.008
(1.328)

- 1.267

0.004
(0.005)

(0.917)

~MRP*

0.394
(0.198)

0.458
(0.136)

Squared ~ M W *

- 1.475
(1.227)

-2.293
(0.843)

Adjusted R-Squared
No. of Observations
F Stat. for joint test,
1% cutoff = 4.82
Implied Extrema
CSS Mean
Adjustment
~MRP*

0.107
101
7.01

0.138
101
8.97

0.111
101
7.22

0.15 1
101
9.86

Max:
11.1%

Min:

2.2%

ax:
13.4%

Max:
10.0%

*dMRP is the sum of the annual change in the BLS Consumer Price Index for all Urban
Workers (CPI-U) and the BLS Nonfarm Business Sector Output per Hour Worked.
Note: Numbers in parentheses are standard errors.
Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community
Salary Survey.

clevelandfed.org/research/workpaper/index.cfm

Table 8A

Wage Inflation and the Standard Deviation
of Occupation Nominal Wage Adjustments

Model
Intercept

Dependent Variable
Absolute Value of the Occupational Wage
Adjustment Term:
2
4
1
3

0.006
(0.0004)

0.006
(0.0004)

0.009
(0.0004)

0.222
(0.0037)

Lagged
Adjustment

0.01 1
(0.0004)
0.002
(0.0038)

~MRP*

0.245
(0.0120)

0.142
(0.01 19)

0.182
(0.0107)

0.1 17
(0.01 19)

Squared ~MRP*

-1.104.
(0.0749)

-0.578
(0.0745)

-0.797
(0.0662)

-0.4 10
(0.0680)

Fixed Effects Included

None

None

Adjusted R-Squared
No. of Observations
Implied Slope with
respect to ~ M R P *
Mean one STD
Min. and max. of data

0.0158
67,885

0.066
62,87 1

5.45%
6.99%

4.29%
5.10%

+

Occupation Occupation
x City
x City
0.254
0.250
67,885
62,871

4.45%
5.57%

4.62%
5.20%

*dMRP is the sum of the annual change in the BLS Consumer Price Index for all Urban
Workers (CPI-U) and the BLS Nonfarm Business Sector Output per Hour Worked.
Note: Numbers in parentheses are standard errors.
Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community
Salary Survey.

clevelandfed.org/research/workpaper/index.cfm

Table 8B

Wage Inflation and the Standard Deviation
of Employer Nominal Wage Adjustments

Model
Intercept

Dependent Variable
Absolute Value of the Employer Wage Adjustment
Term:
1
2
3
4

0.0 13
(0.0006)

0.012
(0.0005)

0.0 13
(0.0005)

0.180
(0.0036)

Lagged
Adjustment

0.012
(0.0005)
0.1 15
(0.0036)

~MRP*

0.148
(0.0 158)

0.086
(0.0 156)

0.159
(0.0 153)

0.1 16
(0.0 152)

Squared dMRP*

-0.225
(0.0986)

-0.028
(0.098)

-0.238
(0.0946)

-0.134
(0.0953)

Fixed Effects Included

None

None

Adjusted R-Squared
No. of Observations
Implied Slope with
respect to ~ M R P *
Mean f one STD
Min. and max. of data

0.0 185
67,885

0.054
62,553

10.94%
11.26%

8.07%
8.11%

Employer x Employer x
City
City
0.120
0.133
67,885
62,553

11.85%
12.18%

9.33%
9.52%

*dMRP is the sum of the annual change in the BLS Consumer Price Index for all Urban
Workers (CPI-U) and the BLS Nonfarm Business Sector Output per Hour Worked.
Note: Numbers in parentheses are standard errors.
Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community
Salary Survey.

clevelandfed.org/research/workpaper/index.cfm

Figure 1: CSS Mean Wage Change Versus dMRP

-

CSS Mean Wage
Change
-..--dMRP

End Year

Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community
Salary Survey.

clevelandfed.org/research/workpaper/index.cfm

Figure 2A: Standard Deviations of Occupational Adjustments Associated with dMRP: Nonparametric
and Regression Predictions
Smoothed S t d Dev o f Occ A d j

o

+

P r e d i c t e d S t d Dev o f Occ Adj

I

0

I

.05
INF(CP1-U)+CHANGE

I

.I
LABOR PROD

1

.I5

Figure 28: Standard Deviations of Employer Adjustments Associated with dMRP: Nonparametric
and Regression Predictions
o

1
0

Smoothed S t d Dev o f Ernp A d j

I

+

.05
INF(CP1-U)+CHANGE

P r e d i c t e d S t d Dev o f Ernp Adj

I

.I
LABOR PROD

I

.15

Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community Salary Survey.

clevelandfed.org/research/workpaper/index.cfm

Figure 2C: Standard Deviations of Occupational Adjustments Associated with CSS Mean Wage
Change: Nonparametric and Regression Predictions
Smoothed S t d Dev o f Occ Adj
-03

+

P r e d i c t e d S t d Dev o f Occ Adj

4

I

0

I

I

.I
MEAN CITY-YR CSS LOG WAGE CHANG
.05

I

.15

Figure 2D: Standard Deviations of Employer Adjustments Associated with CSS Mean Wage
Change: Nonparametric and Regression Predictions
Smoothed S t d Oev of Emp Adj

0

.0'5

+

P r e d i c t e d S t d Dev o f Emp Adj

.I

.15

MEAN CITY-YR CSS LOG WAGE CHANG

Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community Salary Survey.

clevelandfed.org/research/workpaper/index.cfm

Figure 3A: Standard Deviations of Long-Run Occupational Adjustnients Associated with dMRP
o

I

0

Smoothed STD o f LR Occ Adj

+

Reg P r e d STD o f LR Occ Adj

I

I

.05
.I
INF(CP1-UItCHANGE LABOR PROD

I

.I5

Figure 3B: Standard Deviations of Short-Run Eniployer Adjustments Associated with dMRP
Smoothed STD o f SR Emp Adj

I

0

I

+

Reg P r e d STD o f SR Emp Adj

I

.05
.1
INF(CP1-UItCHANGE LABOR PROD

I

.15

Sourcej Authors' calculations from the Federal Reserve Bank of Cleveland Community Salary Survey.

clevelandfed.org/research/workpaper/index.cfm

Figure 3C: Standard Deviations of Long-Run Occupational Adjustments Associated with CSS Mean
Wage Change
0

,015

i

0

Smoothed STD o f LR Occ Adj

+

I

Reg P r e d STD o f LR Occ Adj

I

.05
.I
MEAN CITY-YR CSS LOG WAGE CHANG

I

.I5

Figure 30: Standard Deviations of Short-Run Employer Adjustments Associated with CSS Mean
Wage Change
o

-05

Smoothed STD of SR Emp Adj

+

Reg P r e d ST0 o f SR Emp Adj

{

0

.05
.I
MEAN CITY-YR CSS LOG WAGE CHANG

.I5

Source: Authors' calculations from the Federal Reserve Bank of Cleveland Community Salary Survey.