The employer of last
resort (ELR) policy proposal, also referred to as the job guarantee or
public sector employment, is promoted by its supporters as an alternative to
unemployment as the primary means of currency stability (see Forstater 1998;
Mitchell 1998; Wray 1998, 2000; Mitchell and Wray 2004; Tcherneva and Wray
2005; and numerous publications available at the Center for Full Employment
and Price Stability and the Centre of Full Employment and Equity). The core
of the ELR proposal is that a job would be provided to all who wanted one at
a decent, fixed wage; the quantity of workers employed in the program would
be allowed to rise and fall counter to the economy’s cycles as some of the
workers moved from public to private sector work or vice versa depending
upon the state of the economy. Supporters have played an important advisory
role in Argentina’s Jefes de Hogar (hereafter, Jefes) jobs
program that has provided jobs to over two million citizens—or five percent
of the population; though there are some important differences, the Jefes
program has many similarities with the ELR proposal (Tcherneva and Wray
2005).
While ELR proponents argue the
program would not necessarily generate budget deficits (Mitchell and Wray
2004), the program is based upon Abba Lerner’s (1943) concept of functional
finance in which it is the results of the government’s spending and taxing
policies in terms of their effects upon employment, inflation, and
macroeconomic stability that matter (Nell and Forstater 2003). This is in
contrast to the more widely promoted concept of “sound” finance, in which
the presence of a fiscal deficit is itself considered undesirable. Rather
than not being able to “afford” an ELR program, ELR proponents argue that
societies would do better to consider whether they can “afford” involuntary
unemployment.1 The proposed ELR’s approach of hiring “off the
bottom” is argued to be a more direct means for eliminating excess, unused
labor capacity than traditional “military Keynesianism” or primarily
“pump-priming” fiscal policies, particularly given how the U.S. economy
struggles to create jobs for the poor even during economic expansions
(Pigeon and Wray 1998, 1999; Bell and Wray 2004). As Wray (2000) notes,
“How many missiles would the government have to order before a job trickles
down to Harlem?” (5). More traditional forms of fiscal stimulus or
stabilization are still useful and complementary to an ELR program, though
proponents argue that only the latter could ensure that enough jobs would be
available at all times such that every person desiring a job would be
offered one while also potentially adding to the national output.
Regarding macroeconomic
stability, it is the fluctuating buffer stock of ELR workers and the fixed
wage that are argued by proponents to be the key features that ensure the
program’s impact would be stabilizing. With an effectively functioning
buffer stock, the argument goes, as the economy expands ELR spending will
stop growing or even decline—countering the inflation pressures normally
induced by expansion—as some ELR workers take jobs in the private sector.
Regarding the fixed wage, traditional government expenditures effectively
set a quantity and allow markets to set a price (as in contracting for
weapons); in contrast, the ELR program allows markets to set the quantity as
the government provides an infinitely elastic demand for labor, while the
price (the ELR base wage) is set exogenously and is unaffected by market
pressures. Together, proponents argue, the buffer stock of ELR workers and
the fixed wage thereby encourage loose labor markets even at full
employment. Aside from an initial increase as the program is being
implemented (the size of which will depend upon the wage offered compared to
the existing lowest wage and whether the program is made available to all
workers), proponents suggest the program would not generate inflationary
pressures and thus would promote both full employment and price stability.
The purpose of this paper is
to model quantitatively the potential macroeconomic stabilization properties
of an ELR program utilizing the Fairmodel (Fair 1994, 2004). The paper
builds upon the earlier Fairmodel simulations of the ELR in Majewski and
Nell (2000) and Fullwiler (2003, 2005). Here, a rather simple version of
the ELR program is incorporated into the Fairmodel and simulated. The
quantitative effects of the ELR program within the Fairmodel are measured
via simulation within historical business cycles and in comparison to other
policy rules for both fiscal and monetary policies through stochastic
simulation.
The Fairmodel and
Macroeconometric Simulation
The Fairmodel is a well-known,
large macroeconometric model of the U.S. economy developed in the 1970s by
Ray Fair. The model is dynamic, nonlinear, and simultaneous and it
incorporates household, firm, financial, federal government, state and local
government, and foreign sectors of the economy. The model combines 30
stochastic equations that are estimated using the two stage least squares
method with another 100 identity equations. National Income and Product
Account (NIPA) and Flow of Funds data are completely integrated into the
model within the identity equations; balance sheet and flow of funds
constraints are thus fully accounted for. There are 130 endogenous
variables and over 100 exogenous variables.
The overarching intellectual
tradition of the Fairmodel is the Cowles Commission approach to econometric
modeling, which is strongly empirical but nonetheless relies heavily on
theory—and in the case of the Fairmodel, particularly on an acceptance of
the possibility of market disequilibrium—in specifying the stochastic
equations that are the model’s core (see Fair 1994, chapter 1, for further
discussion).
As a structural econometric
model, the Fairmodel is admittedly subject to Robert Lucas’s (1976)
critique, which suggests that estimated coefficients from structural models
may not be consistent across policy regimes. Fair (1994) answers that “the
logic of the Lucas critique is certainly correct, but the key question for
empirical work is the quantitative importance of this critique” (13). Alan
Blinder notes similarly that
while Lucas’s conceptual
point is valuable and indubitably correct, so are the well-known points that
heteroskedastic or serially correlated disturbances lead to inefficient
estimates and that simultaneity leads to inconsistent estimates. But we
also understand that small amounts of serial correlation lead to small
inefficiencies and that minor simultaneity leads to only minor
inconsistencies; so suspected violations of the Gauss-Markov theorem do not
stop applied econometrics in its tracks. In the same spirit, the
realization is now dawning that the Lucas critique need not be a show
stopper. Indeed, evidence that it is typically important in applied work is
lacking. (1989, 107)
Fair and others have further
pointed out that attempts to generate tests and reliable predictions from
models based upon the “deep structural parameters”—such as in real business
cycle models or models employing rational expectations—Lucas prefers have
not been overly successful:
When deep structural
parameters have been estimated from the first order conditions, the results
have not always been good even when judged by themselves. The results in
Mankiw, Rotemberg, and Summers (1985) for the utility parameters are not
supportive of the approach. In a completely different literature—the
estimation of production smoothing equations—Krane and Braun (1989), whose
study uses quite good data, report that their attempts to estimate first
order conditions were unsuccessful. It may simply not be sensible to use
aggregate data to estimate utility function parameters and the like. (Fair
1994, 15)
On the other hand, Fair (1994,
2004) argues that the economic significance of the Lucas critique can be
tested, and thus models that suffer in important ways from the critique can
be “weeded out.”
For its part, the design of
the Fairmodel’s stochastic equations—following the Cowles Commission
approach—is more driven by data and statistical testing than so-called
“modern” macroeconomic models that are based upon rational expectations and
calibration. For each of the model’s stochastic equations, Fair has
performed statistical tests of dynamic specification, spurious correlation,
serial correlation of errors, rational expectations, structural stability
where the date of potential structural change is not known a priori,
end-of-sample structural stability, and over-identifying restrictions.
Consequently—and importantly—the Fairmodel’s basic structure has changed
surprisingly little over time and according to statistical tests
demonstrates structural stability across several business cycles and policy
regime changes. Indeed, 29 of the 30 stochastic equations passed structural
stability tests even in the face of the so-called “new economy boom” during
the mid-to-late 1990s; since stability was rejected only for the stock
valuation equation, these results led Fair to predict early on that the
so-called “new economy” was driven primarily by a stock market bubble as
none of the other equations provided rationale for the increased equity
values (see Fair 2000b, 2004 (chapter 6)). His related estimates of
accelerations in trend and cyclical productivity during the 1990s from the
model (reported in Fair 2004, chapter 6) were nearly identical to those
reported in the influential study by Robert Gordon (2003). Furthermore,
Fair (2004) shows that wealth effects (chapter 5) and interest rate effects
of monetary policy (chapter 11) in the model are consistent with generally
accepted empirical evidence; his research also suggests that assumed
asymptotic distributions of the model’s dynamics (as in modeling multiplier
effects of policy actions, for example) are reasonable and not biased
according to bootstrapping and stochastic simulation procedures (chapter
9). Finally, Fair (1994 (chapters 8 and 9), 2004 (chapter 14)) demonstrates
that the Fairmodel’s predictive abilities “dominate” vector autoregressive
(VAR) models (using the same variables that Sims (1980) uses, which are also
similar to those in the Atlanta Fed’s VAR forecasting model (Robertson and
Tallman (1999)) and autoregressive components models in terms of root mean
squared errors (RMSE) and other standard tests of competing models.
Due to the
statistical rigor of Fair’s methodology, the Fairmodel has found support
among many orthodox economists even as its theoretical approach is counter
to much of the modern macroeconomic research program. Indeed, going back
only to the mid-1990s, Fair has published numerous articles incorporating
the Fairmodel or econometric issues closely related to the Fairmodel’s
structure in many of the highly-ranked orthodox journals; as a
representative though not exhaustive list, see Fair (2005, 2001, 2000a,
1999a, 1999b, 1996, 1993) and Fair and Howrey (1996). Laurence Klein’s
(1991) frequently-cited, standard text on comparative macroeconometric
models presents Fairmodel simulations alongside models used by the Federal
Reserve, the University of Michigan, the Bureau of Economic Analysis, DRI,
and WEFA. More recently, Seidman and Lewis (2002) and Seidman (2003) make
extensive use of the Fairmodel for simulations of alternative macroeconomic
policies. Both also show that the Fairmodel’s structural approach to
modeling consumer behavior is more consistent with generally accepted
empirical evidence than the intertemporal consumption smoothing approach
prevalent in modern macroeconomic theory.
Nevertheless, all models are
mere representations of the economy and thus all of them have various
shortcomings. The Fairmodel is no exception in this regard, but the fact
that it differs from the dominant research program of orthodox
macroeconomics during the past few decades perhaps says more about the
shortcomings of a research program driven by a desire to found modern
macroeconomics strictly on neoclassical microeconomic theory than it says
about problems inherent to the Fairmodel. Fair, given his unapologetic
support of the Cowles Commission approach, has argued that both the New
Classical and New Keynesian literatures do not sufficiently test their
models against either actual time series or competing models:
The currently popular
approach in [New Classical] macroeconomics of working with calibrated models
does not focus on either single-equation tests or complete model tests,
which leaves the field somewhat in limbo. . . . If in the long run the aim
is to explain how the macroeconomy works, these models will need to become
empirical enough to be tested, both equation by equation and against
time-series models and structural models like the [Fair] model. (2004, 176)
As is also true in the
RBC [i.e., real business cycle] literature, one does not see, say,
predictions of real GDP from some New Keynesian model compared to
predictions of real GDP from an autoregressive equation using a criterion
like the RMSE criterion. (1994, 15)
The RBC literature
should entertain the possibility of testing models based on estimating deep
structural parameters against models based on estimating approximations of
decision equations. Also, the tests should be more than just observing
whether a computed path mimics the actual path in a few ways. The New
Keynesian literature should entertain the possibility of putting its various
ideas together to specify, estimate, and test structural macroeconometric
models. (1994, 16)
From a heterodox perspective,
being out of step with the modern counter-revolution in orthodox
macroeconomics is not in itself a shortcoming; indeed, several
characteristics of the Fairmodel are consistent with heterodox macroeconomic
approaches. These include the following, some of which were noted by
Majewski and Nell:
·
Expectations are
important in the Fairmodel, but they are generally adaptive in nature. Fair
(2004) writes that “there is . . . no strong evidence in favor of the
[rational expectations] assumption (and some against it)”; thus, in the
Fairmodel, “Agents are assumed to be forward looking in that they form
expectations of future values that in turn affect their current decisions,
but these expectations are not assumed to be rational (model consistent).
Agents are not assumed to know the complete model” (4).
·
Equations for
household spending on non-durable goods, services, durable goods, and
residential investment are each affected by a nominal interest rate rather
than a real interest rate (see Fair 2004, chapter 3). Household durable and
non-durable consumption spending are also influenced by current disposable
income and a wealth effect.
·
Production—which
here refers to “output” of both industrial and services sectors—depends upon
lagged production, current sales, and the lagged change in inventories, all
of which essentially generate an expected level of production consistent
with sales and desired inventories. As Fair put it, in the production
equation “production is smoothed relative to sales.” Private sector
employment and hours worked are then heavily influenced by changes in firms’
production. Like production, productivity is not purely determined on the
“supply side,” but is rather a result of the interaction between aggregate
demand and aggregate supply. Fair’s own empirical research has found
significant support for this “production-smoothing” approach (e.g., Fair
1989).
·
The main capital
investment equation—non-residential fixed investment—is a capital stock
adjustment accelerator in which the current capital stock is set largely by
an estimate of excess capital on-hand and expected production to meet sales
demand. Costs of debt and equity capital are statistically significant
determinants of investment spending. In the Fairmodel, saving equals
investment due to national income accounting identity; saving does not
“fund” investment.
·
The monetary
policy tool is a short-term interest rate, the 3-month T-bill, while the
money supply is endogenously determined. (While the Fed actually sets the
federal funds rate, the two are closely related through arbitrage; Fair
(2004, 2005) notes that it makes virtually no difference econometrically
which rate is used.) The short-term rate responds, similar to Taylor’s
rule, positively to higher inflation and lower unemployment. The policy
responses are fitted to the actual historical interest rate strategy of the
Federal Reserve, which is different from the Taylor’s Rule approach of
modeling an “optimal” monetary policy feedback rule (Fair 2001).
Nevertheless, the comparison tests by Fair (2004, chapter 11; 2005) show
that the Fairmodel’s estimated rule performs as well as an optimal control
procedure. Fair’s estimated feedback rule for the short-term interest rate
is discussed more completely in a later section of this paper.
·
Fair’s tests of
unemployment and inflation data reject the NAIRU dynamics in which inflation
spirals out of control if unemployment falls below a certain level (Fair
1999b, 2000, 2004 (chapter 4)). He found that there are no low rates of
unemployment associated with high and spiraling inflation over the model’s
estimation period. Rather, his research suggests that the U.S. economy has
normally operated below full capacity utilization. According to his results
from econometric testing, a NAIRU or a natural rate of unemployment can only
be assumed and imposed on a large macroeconometric model, as is done in the
Federal Reserve’s FRB/US model (see Fair 2004, chapter 7). As discussed
below, the inflation dynamics in the Fairmodel have much in common with
recent Post Keynesian critiques of the orthodox “New Consensus” view (which
Fair refers to as the “modern view” or modern macroeconomics; the three
terms are used interchangeably in this paper).
·
Long-term interest
rates in the Fairmodel are determined as a markup over short-term rates
based upon the trend in short-term rates and the trend in the markup.
Interest rates in the Fairmodel, like the capital stock above, are not
determined in a loanable funds market. Contrary to current orthodox
thinking (particularly that related to the “generational accounting”
literature) or the FRB/US model, expectations of future federal deficits do
not lead to higher long-term interest rates today in the Fairmodel.
Orthodox researchers have had little success in empirical studies finding
consistent, economically significant effects of rising deficits or debt on
interest rates (Engen and Hubbard 2004, Galbraith 2005), which is consistent
with the endogenous money/horizontalist view of many Post Keynesians.
·
As mentioned, the
Fairmodel integrates complete consistency with NIPA and Flow of Funds
identities, which is also emphasized in the related “stock-flow consistent”
modeling and “social accounting matrices” literatures currently popular
among heterodox economists (e.g., Dos Santos 2002, Taylor 2004, Godley and
Lavoie 2005). Not surprisingly, the Fairmodel is frequently referenced in
this literature. Indeed, Godley and Lavoie have labeled Fair’s research one
of the “outstanding individual contributions to the stock-flow consistent
approach” (2).
An understanding of the tools
being used in empirical analysis reduces the likelihood that the evidence
gathered will be misused or misinterpreted. Accordingly, what the
simulations reported here demonstrate is the logic of an ELR program
given historical relationships among macroeconomic variables implied
by coefficients of the Fairmodel’s stochastic equations and given
constraints provided by NIPA and Flow of Funds accounting identities. It is
certainly possible that a policy such as the ELR would alter some of the
historical relationships—though it would not alter NIPA and Flow of Funds
account identities—but it is essentially impossible to know how much (in
quantitative terms) the relationships would be altered. As is shown below,
the simulated ELR program is not an expensive program and thus structural
changes in coefficients that could occur arguably might not be of tremendous
economic significance; the fixed wage/hiring off the bottom nature of ELR
and the stability of the Fairmodel’s structural equations through time and
across alternative policy regimes would both seem to support this to some
degree. Most importantly, economists desiring to provide advice to
policymakers recognize that some estimate regarding the impacts of a policy
proposal in terms of predicted outlays, benefits, and impacts upon the
broader economy is absolutely essential. The simulations here are one
possible source of such information regarding an ELR policy.
ELR in the Fairmodel
For simplicity the ELR program
is modeled as a purely federal program, though it is computationally similar
in terms of costs to a federally funded but state or locally administered
program. It could also provide funding for targeted job creation in the
private sector, particularly at not-for-profit businesses, if desired (Wray
2000). Both state/local administering and not-for-profit sector job
creation have been common in Argentina’s Jefes program (Tcherneva and
Wray 2005). The program is simulated from the first quarter of 1985 through
the third quarter of 2005, which enables simulation of roughly two complete
business cycles following the initial implementation period (discussed
below). The rest of this section discusses the equations added to or
changed in the Fairmodel in order to simulate the ELR program. The
simulations are carried out using the Fair-Parke program (Fair 2003).
ELR Wage and Jobs Equations
The basic wage for ELR workers
(hereafter, WELR) is set to reach $6.25 in 2005, which had been proposed as
a “starting point” for discussion in some of the ELR literature. There is
no particular WELR necessarily most consistent with the ELR proposal; there
are obviously more generous versions of an ELR program that one could
envision and that proponents have in fact proposed. For instance, Wray
(2000) suggests the program could offer “a package of benefits [including]
healthcare, child care, sick leave, vacations, and contributions to Social
Security so that years spent in ELR would count toward retirement” (4).
More recently, Tcherneva and Wray (2005) propose that ELR workers be paid a
“living” wage. As Wray (2000) confirms, “On one level, it does not matter
where the [ELR] wage is set” (4). However, WELR becomes the effective
minimum wage in society, and in that sense the level at which it is set
does matter. “Setting [WELR] well below current market or minimum wages
would require massive deflation of the price level in order to generate a
pool of workers willing to work for that wage. On the other hand, setting
[WELR] well above the going wage would generate a large increase of wage and
price levels as firms would have to compete with [WELR]” (Wray 2000, 4).
The choice of WELR for the simulation is based on the criterion of
consistency with the lower end of the wage structure for the actual time
period being simulated, which a WELR of $6.25 in 2005 achieves.
Since there is modest
inflation during 1985-2005, the WELR should be adjusted periodically. For
purposes of price stability, however, indexing the WELR to the inflation
rate may not be desirable. Again, there are countless possibilities
consistent with the ELR proposal (e.g., indexing to growth in the calculated
living-wage level, to trend productivity growth, and so forth). The
approach here, consistent with modest average wage and inflation growth
during the simulation period, is simply to index WELR to an inflation target
of, say, 2.5 percent. This ensures that WELR is consistent with the
inflation target and that the effect of the WELR on private sector wages is
consistent with the target. In other words, if the economy were in an
expansion, indexing WELR in this manner would not contribute to additional
cost-push inflation or to the sort of wage-price spirals hypothesized in the
natural rate of unemployment or NAIRU theories; if the economy were in
recession, such indexation would reduce the threat of deflation. WELR is
set at $3.81 in 1985 and grows at 2.5 percent annually, rising to $6.25 in
2005. Indexation is assumed to be fully implemented in the first quarter of
a new year, as is done for government transfer programs such as Social
Security; there is no change in WELR during the 2nd, 3rd,
and 4th quarters of a year.
In the Fairmodel, let WELR-1
refer to the value of WELR in the previous year; the current WELR is then
set by the following equation:
(1) WELR = 1.025 • WELR-1
Regarding the number of ELR
employees, all unemployed workers are included in order to avoid
understating the costs of the program, though the feedback from larger
assumed ELR expenditures to private sector hiring will offset some of this.
It is obvious that not everyone counted as unemployed will take an ELR job
since the WELR will be below the reservation wage of some; at the same time,
it is expected that many from outside the labor force will join the ELR
workforce.2 While it is difficult to know how many of the
unemployed will not take ELR jobs versus how many from outside the labor
force will, this treatment assumes that the magnitude of both roughly offset
one another; more importantly, it generates a rather large ELR program. The
point here is to examine “what if” several million people were to take ELR
jobs, not to provide a micro-theoretic model explaining why they would do
so. This also is perhaps the most “theoretically pure” version of the ELR
buffer stock, since changes in non-ELR employment are matched one-for-one by
changes in the ELR workforce and because the policy accepts all those
looking for work.
While the simulated ELR program
begins in 1985, it is assumed to be phased in during 1985-1987, employing
8.33 percent of the total unemployed to begin and then adding another 8.33
percent each quarter until fully implemented in the last quarter of 1987.
In the Fairmodel, let U be the
number of unemployed according to the Bureau of Labor Services survey and
let PHASE be the variable that phases in the program (set to 0.0833 to
begin, and then rising by 0.0833 each quarter until reaching 1.0 in the last
quarter of 1987 and remaining at 1.0 thereafter); the number of ELR jobs
(JELR) is given by
(2) JELR = U • PHASE
ELR Spending Equations
According to the
NIPA and Flow of Funds data used by the Fairmodel, average hours worked in
the private sector fluctuates cyclically between 32.5 and 34 hours per week
during 1985-2005; in the government sector average hours worked ranged from
34 to 37 hours per week, with little discernable pattern. In this paper,
ELR workers are assumed to work an average of 34 hours each week, which is
an average of hours worked in the two sectors and also avoids cyclical
fluctuations.
In the Fairmodel, let HELR
denote average hours worked per quarter per ELR employee and be exogenously
set to 442 (34 hours per week times 13 weeks). Given (1) and (2) above, the
aggregate income earned by ELR workers (YELR) is given by
(3) YELR = WELR • JELR • HELR
Following Majewski
and Nell, whose treatment is consistent with earlier CETA experience, it is
assumed that there will be additional costs to the ELR program for
production supplies totaling 15 percent of the income paid to ELR workers.
The treatment here, again, errs on the side of overstating the costs of the
program, given that it would not necessarily be the case that significant
additional costs would be incurred if, for instance, ELR workers were
employed at qualifying non-profit organizations and the organizations were
simply given a subsidy equal to the WELR. Such additional costs are
frequently absent in Argentina’s Jefes program (Tcherneva and Wray
2005).
In the Fairmodel,
the total non-labor costs of the ELR program (COSTELR) are determined by the
following equation:
(4) COSTELR = 0.15 • YELR
Given (1), (2), (3), and (4)
above, the total spending on the ELR program (ELRSPEND) within the Fairmodel
is given by
(5) ELRSPEND = YELR + COSTELR
Lastly, it is useful to also
discuss the impacts of the ELR program on unemployment benefits.
Unemployment benefits are estimated in equation 28—a stochastic equation—of
the Fairmodel, which takes the following form:
(Fair 28) log(UB)=a+b1log(UB-1)+b2log(U)+b3log(WF)+e
where,
·
UB is the nominal
dollar value of unemployment benefits paid
·
a is a constant
term
·
bi are
coefficients from two-stage least squares estimation
·
WF is the average
wage in the firm sector and accounts for the fact that unemployment benefits
are frequently increased through legislation as average wages rise
·
e is an error
term.
The response of log(UB) to a
decrease in the number of unemployed workers is through the coefficient b2.
The estimated value of b2 is not large in the economic sense;
decreasing U by 500,000 or even 1,000,000 workers does not reduce UB by more
than, say, $5 billion annually while total UB itself was in the $60 billion
range during the most recent recession. There is no other cyclical variable
in the equation such as jobs or real GDP in the equation. In the
simulations here, then, effects on UB are limited to the rather minor direct
effects of falling U since any other treatment would be rather ad hoc.
While one could envision
policy regimes in which ELR employment directly replaced some unemployment
benefits, any level of unemployment benefits would be consistent with the
ELR proposal. Wray (2000) confirms that “no matter what social safety net
exists, ELR can be added to allow people to choose to work over whatever
package of benefits might be made available to those who choose not to work”
(4). By the same token, since well over half of those unemployed are not
eligible for unemployment benefits, and since many others employed by ELR
would likely come from outside the labor force altogether, it is not
necessarily the case that those eligible for unemployment benefits would be
the same people accepting ELR jobs particularly since WELR is set at a
fairly low value. The approach here is to leave Fair’s equation 28 above
intact, which likewise suggests that some receiving unemployment benefits
will take ELR jobs while other ELR jobs will be taken by those receiving no
benefits or coming from outside the labor force.
Changes to Fairmodel Identity
Equations
In order to simulate
the ELR policy, equations (1) through (5) above, and parts of these
equations in some cases, are incorporated into the following NIPA and Flow
of Funds identity equations within the Fairmodel:
·
Equation 43:
Average nominal hourly earnings excluding overtime of all workers
·
Equation 53:
Employee social insurance contributions to the federal government
·
Equation 60:
Total real sales of the firm sector
·
Equation 61:
Total nominal sales of the firm sector
·
Equation 65:
Total nominal saving of the household sector
·
Equation 76:
Nominal saving by the federal government
·
Equation 82:
Nominal GDP
·
Equation 83: Real
GDP
·
Equation 95:
Total worker hours paid divided by population over 16
·
Equation 104:
Nominal purchases of goods and services by the federal government
·
Equation 115:
Nominal disposable income in the household sector
Changes made to the above
equations are shown in Appendix A. One point of note is that ELR wages
earned are applied to the workers’ portion of payroll taxes but not to
regular income taxes. The ELR proposal is not necessarily more consistent
with any particular method of taxation of ELR earnings, though proponents
appear to prefer that ELR income be subject to as little tax as possible
(since the program hires “off the bottom”). The treatment here is simply in
order to be consistent with actual tax treatment of lower-end wages during
1985-2005. While much of that income would not rise above standard
deductions for income taxes, the treatment here will tend to slightly
overstate the net outlays of the program since it is likely that some
portion of ELR earnings would be subject to income taxes (for instance, in
the case of a second family income). Another noteworthy point is that the
changes made to real GDP in equation 83 assume that ELR workers are
unproductive; that is, they receive an income but do nothing to add to
the national output directly (this assumption is relaxed, however, for some
of the stochastic simulations reported later in the paper).
These identities
themselves—except for equations 43 and 95, which simply track average wages
and hours worked, respectively, but do not impact any other equations in the
model—affect directly and indirectly the determination of variables in many
other stochastic equations and NIPA/Flow of Funds identities during
simulation. For instance, ELR workers receive an income that is
subject to payroll taxation, so this must be added to government
spending/receipts/saving equations and to household income/saving/taxation
equations in order to actually simulate the program.
Supply-Side Effects
On the supply side,
two important variables are the price level and the wage rate. As is
standard practice in Fair’s research using the Fairmodel (e.g., Fair 2001,
2004, 2005), PF—the price level in the firm sector which measured the
average price of finished goods—is the measure of the price level used
here. The inflation rate as measured by the growth in PF moves closely with
the PCE Deflator. The two are shown on a year-to-year basis in Figure 1
during 1985:1-2005:3, for which time their correlation was 0.87.
As Fair (2004) explains,
The price level [PF] is
a function of the lagged price level, the wage rate inclusive of the
employer social security tax rate, the price of imports, the unemployment
rate, and the time trend. The unemployment rate is taken as a measure of
demand pressure. The lagged price level is meant to pick up expectational
effects, and the wage rate and import price variables are meant to pick up
cost effects. The log of the real wage rate has subtracted from it LAM,
where LAM is a measure of potential labor productivity.3 (32)
Fair has tested the unemployment
rate against other output gap variables for the equation and has found that
the unemployment rate dominates the others statistically (2004, 32).
Another way of stating Fair’s PF equation is that prices of finished goods
are set as a markup on wages and import prices—such as the price of oil and
of other primary commodities—and also has a pro-cyclical component to
account for labor market and/or bottleneck effects. Also, note that the
unemployment rate affects the level of prices, which means
that a fall in the unemployment rate creates a one-time increase in
inflation, rather than spiraling inflation as in the NAIRU view of modern
macroeconomics. Overall, the approach has much in common with recent theory
and empirical evidence published by Post Keynesians (e.g., Arestis and
Sawyer 2005, Bloch et al 2004, Kreisler and Lavoie (forthcoming)).
The average wage in the firm
sector, WF, is also determined in a stochastic equation.
The wage rate is simply
taken to be a function of the constant term, the time trend, the current
value of the price level, the lagged value of the price level, and the
lagged value of the wage rate. Labor market tightness variables like the
unemployment rate were not significant in the equation. The time trend
variable is added to account for trend changes in the wage rate relative to
the price level. The [dependent variable is the] potential productivity
variable, LAM, . . . subtracted from the wage rate. (2004, 39)
In the Fairmodel, therefore,
when unemployment is reduced, higher prices and wages can result as in any
macroeconomic model.4 As a result, when ELR-related spending
automatically rises and falls countercyclically, the potential for
demand-pull inflation or, as the case may be, greater price stability as a
result of the program can be simulated.
Regarding cost-push
effects of the ELR program, if WELR is set above the minimum wage—as is the
case here—then it would be reasonable to expect that there would be some
pass-through effect to average wages and prices given that WELR becomes the
effective minimum wage. At the same time, the actual effect on the overall
wage structure could be far less than the rise in the effective minimum wage
since WELR affects primarily the low end of the wage structure. While there
is no mechanism within the Fairmodel to account for such a pass-through
effect, the Federal Reserve’s FRB/US model does incorporate such an effect
in its dynamic wage-adjustment equation.5 According to the
approach of the FRB/US model as described in Appendix B, an increase in the
current minimum wage from $5.15 to $6.25 raises WF by around 12 cents. The
effect on WF from Appendix B is increased in the simulations (in order to
err on the side of a less stabilizing ELR program) such that at least
one-third of an increase in WELR (or, to begin the simulation, the increase
from the minimum wage in 1984 to the beginning level of WELR in 1985) is
passed through to WF—that is, the rise from $5.15 to $6.25 would raise WF by
around 38 cents in these simulations—by adding this amount to the constant
term in the WF equation during simulation (but following estimation of
coefficients).
Also, because ELR
will generate de facto full employment and permanently reduced
“slack” in the economy, the trend variable in the WF equation was replaced
with the lagged natural logarithm of real GDP and re-estimated (some
first-stage regressors were replaced or added, as well). The other
coefficients for variables in the equation were not altered to a
statistically or economically significant degree; there was also a slight
increase in the equation’s R-squared value. Without this change, an ELR
induced, permanent increase in real GDP would not affect WF except to the
degree that increased aggregate demand affected the general price level of
output (that is, the effect would be indirect only). This change enables
the ELR program to directly influence wages as aggregate spending raises
capacity utilization.
There are two
potentially stabilizing effects of the ELR program in the Fairmodel worth
discussing. The first effect is related to another important supply-side
variable, the labor force. Labor force participation in the Fairmodel is
determined by stochastic equations; according to these equations, labor
supply responds negatively to the unemployment rate, positively to aggregate
wages and salaries, and (modestly) negatively to aggregate household
wealth. These effects are unchanged in the simulations. There is thus the
possibility of a modest stabilizing effect if increased employment and wages
raise labor supply, since they would encourage greater labor force
participation and thereby soften a decline in the unemployment rate and then
also soften a rise in PF. The effect upon PF would then feed into the
determination of WF, softening its rise, as well. In other words, according
to the Fairmodel’s stochastic equations—and like most any macroeconomic
model—an increase in labor supply can suppress wage demands by slowing the
fall in the unemployment rate. Likewise, the simulated ELR program may
raise labor force participation (as has been the case in Argentina’s
Jefes program; see note 2) as average wages rise—since WELR is set above
the minimum wage—and as “slack” in the economy is permanently reduced, which
will soften somewhat potential increases in PF and in WF. A related, more
traditional, impact of the ELR program in the Fairmodel results from the
countercyclical nature of ELR-related spending. As the economy expands, the
concurrent reduction in ELR-related spending will offset to some degree a
falling unemployment rate. Since, again, the unemployment rate has a direct
negative effect on PF, the former’s slowed or muted reduction during an
expansion will modestly offset the procyclical behavior of PF. The addition
of real GDP to the WF equation discussed above will also have a similar,
albeit smaller, effect if the ELR buffer stock’s fluctuations soften
swings in the level of real GDP.
ELR proponents have long
argued that the pool of ELR workers will suppress excessive wage demands,
even during economic expansions6 (e.g., Wray 1998, 2000). Using
the Fairmodel, the simulated impact on PF of both increased labor supply and
countercyclical ELR spending during an expansion provide at least some
quantitative estimate of the anticipated (by ELR proponents) ability of the
fluctuating buffer stock of ELR workers—who are employed at a fixed wage
that does not respond to market pressures—to promote price stability.
Nevertheless, the total quantitative impact in these simulations is probably
smaller than ELR proponents expect would occur in reality; for instance,
throughout the simulations the labor supply increase has a total effect of
only 0.1 to 0.3 percentage points on the measured unemployment rate.
To conclude this section, even
with the potential stabilizing effects described here, the stabilizing
properties of the ELR policy are possibly understated in the
simulations. As mentioned, ELR workers are assumed to contribute nothing
directly to the national output (except in some of the stochastic
simulations reported later). Also, potential productivity-enhancing
externalities from ELR employment such as education, training, and reduced
skills depreciation from unemployment—not to mention employers’ reduced
search time when looking for available workers—are omitted here, even as
they are common within the ELR literature and are reported to be important
aspects of Argentina’s Jefes program (Tcherneva and Wray 2005).
Given the structure of the Fairmodel, these other benefits could only be
incorporated in a rather ad hoc manner and would, in a similar ad
hoc manner, bias the simulations toward greater macroeconomic
stability. On the other hand, the primary labor demand influence on prices,
the demand-pull effect on production due to increased aggregate income, is
already present in the Fairmodel’s supply-side equations that are
statistically robust across several business cycles and policy regime
changes. And less favorable—to the ELR’s ability to stabilize the
economy—impacts upon the supply side, namely the pass-through effect of an
increased effective minimum wage and a direct influence of reduced “slack”
in the economy upon labor costs, are added here as well.
ELR Simulations within
Historical Business Cycles
As mentioned, the above ELR
program is simulated for the period 1985:1-2005:3. Error terms (residuals)
from stochastic equations were used in the simulations in order to generate
business cycles. This also generates a perfect tracking solution for the
period, as the “base” data are the actual data for the period and are shown
in Table 1. In order to isolate the stabilization potential of ELR, the
short-term interest rate is set to be exogenous and remain at the “base” or
actual levels for the period; there is thus no feedback from the effects of
ELR as an influence upon monetary policy in the simulation in this section.
There are various “shocks”
coming into contact with the ELR program here. From Table 1, the economy is
in an expansion through the first half of 1990, though unemployment rates
are high according to recent standards. A recession ensues driven largely
by oil price increases and falling consumption. While the recession ends in
1991, unemployment does not peak until late 1992. Unemployment drops
thereafter and there is economic expansion throughout the decade; the period
1996-2000 is notable for both low unemployment and low inflation; it is also
notable for the exceptional stock market “bubble” and federal budget
surpluses. Another recession occurs in 2001 due to falling investment
spending and a stock market correction; another sluggish recovery until late
2003 follows. In 2004-5, oil prices rise significantly and economic
recovery appears to be in full swing to the degree that the Fed raises
interest rates at a “measured” pace.
Macroeconomic Effects of the
Simulated ELR Program
Figure 2 shows the
workers employed in the simulated ELR program (JELR). Aside from the phase
in period during 1985-1987, JELR moves with the unemployment rates in Table
1 since JELR is assumed to be the same size as the number of measured
unemployed. From the business cycle trough in 1992 to business cycle peak
in 2000, there is a reduction in JELR of around 3.5 million; from the two
business cycle peaks (1990 and 2000) to the respective troughs, the increase
is a bit over 2.5 million. These results are consistent with Mitchell and
Wray’s (2004) contention that a minority of total ELR workers moving between
ELR and other jobs would suffice for the buffer stock effect of the ELR
program to be effective.7
Figure 3 shows the annualized
level of real GDP during the period compared to the simulated value. From
this graph, the ELR program moves the economy to a permanently higher level
of real GDP, while the swings in the level of real GDP due to exogenous
shocks are less pronounced. Figure 4 shows the non-annualized differences
between the simulated and base levels of real GDP. Again, the ELR program
both permanently raises the level of real GDP while this increase is greater
when the economy slows in the early 1990s and 2000s and is reduced
significantly as the economy expands during the mid-to-late 1990s. The
reason for the permanent rise in real GDP within the simulation is
straightforward: individuals previously unemployed are now earning incomes
in the ELR program; as they spend their incomes, this in turn begets more
aggregate income as firms raise production levels, hire more workers, and so
forth, to meet increased sales.
Figure 5 shows the
unemployment rate in the simulation less the base rate. Note from the
previous section that JELR is assumed equal to all those in the labor force
but not employed in non-ELR jobs. Thus, the measured unemployment rate is
identical to JELR divided by the labor force. By itself, then, the
unemployment rate in the ELR simulation is uninteresting, since the measured
unemployment rate is simply the measured ELR employment rate. Compared with
the base rate of unemployment, however, Figure 5 shows what additional
percentage of the labor force was able to find non-ELR jobs in the ELR
simulation. As the program is implemented during the relatively high
unemployment of the late 1980s and continuing into the recession of 1990-91,
the stimulus from the program and thus the reduction in the unemployment
rate (i.e., increase in non-ELR employment) is the largest. During the
expansion of the 1990s, simulated unemployment moves closer to the base
value as the economic stimulus it provided by the ELR policy is reduced. As
the economy moves to recession again in 2001, simulated unemployment is
again reduced further below the base level (i.e., non-ELR employment
rises). Again, the unemployment rate—like real GDP—is altered as ELR
employees receive incomes, spend these incomes, and the resulting increased
production leads to greater private sector hiring; when the economy expands,
workers leave ELR for newly created, higher paying jobs and the stimulus
provided by the program is thus reduced.
The goal of many ELR workers
would be higher paying, private sector employment rather than ELR work.
Whereas Figure 5 showed the percentage of the labor force leaving the
unemployed ranks (i.e., leaving ELR employment in the simulation) and taking
non-ELR jobs, Figure 6 shows the actual number of non-ELR jobs created in
the private sector. As the spending of ELR workers encourages greater
production and hiring (and, during recessions, reduced layoffs) by firms,
around 1,300,000-2,100,000 additional private sector jobs are permanently
created (or not lost during downturns) beyond the base level of jobs (which
itself increases through time except during recessionary periods). As
above, this effect is stronger during recessionary periods and weaker during
expansions.
As noted in the introduction,
those promoting ELR argue that it would not create significant inflationary
pressures, aside from modest initial impacts. The year-to-year inflation
rates in the simulation are shown in Figure 7. As they predicted, a modest
increase in inflation occurs at first, peaking at 0.44 percentage points
greater than the base inflation rate; this initial effect evaporates over
time even as the economy expands in the late 1980s. During the recessions
of the early 1990s and early 2000s, there are increases in inflation
generally peaking at 0.10 and 0.13 percentage points above the base value,
respectively, due to the additional ELR spending. Note that the simulated
inflationary effects of the ELR program during 1990-1991 were slight even as
the economy was experiencing an oil price shock. Most interesting, however,
is that during the expansion of the 1990s simulated inflation rates are
actually slightly below base values as ELR workers take private
sector jobs and thereby reduce ELR-related government spending; likewise,
after the slowdown and sluggishness of the early 2000s, inflation in the ELR
simulation falls slightly below the base level by 2005 as the economy
recovers and expands (and yet another oil price shock is in force).
Regarding stabilization properties, therefore, the simulated ELR program
provides modest countercyclical stabilization to inflation rates, raising
them slightly as ELR payrolls rise during recessions when deflation could
otherwise threaten, and actually reducing inflation pressures slightly
during expansion as workers leave the ELR program. Consistent with what ELR
proponents have long argued, the ELR program simulated here is not
inflationary.
Note that during the early
stages of the simulation that a permanent increase in the price level
did occur due to a permanent reduction in economic “slack” and a higher
effective minimum wage in the economy. But, again, the effect upon the
growth rate of the price level (i.e., inflation) was temporary. In
other words, during the implementation period the simulated ELR program
raises aggregate demand and thereby raises production (compared to without
the ELR program). Thereafter, the simulated program does not raise
production any further beyond the non-ELR level. Rather, the buffer stock
of ELR workers—and thus the macroeconomic effects of the program—adjusts
counter to the business cycle to sustain and stabilize this
raised level of production (compared to without the ELR) while also
countering modestly the business cycle’s influence upon inflation.
The effect of the simulated
ELR program upon inflation is consistent with recent heterodox research
suggesting that a Phillips curve relating actual inflation and unemployment
rates can in fact be essentially horizontal across a wide range of capacity
utilization. According to these views, changes in employment lead to
significant accelerations or decelerations in inflation only when capacity
utilization moves significantly outside these ranges (Bloch et al 2004,
Kreisler and Lavoie (forthcoming), Palacio-Vera 2005). An empirically-based
Phillips curve estimated by Kansas City Fed Economist Andrew Filardo (1998)
was also supportive of these views. All of these authors, and Fair’s
research as well, have suggested that the U.S. economy has historically
operated well within this theorized horizontal range and below full
utilization of capacity. Furthermore, in the simulations it need not be the
case that productive capacity has not increased with sales and production;
in the Fairmodel and consistent with most heterodox analyses, rising sales
demand can encourage greater capital accumulation (see the discussion and
literature reviews in Palacio-Vera (2005, especially 754-758) and in Fontana
and Palacio-Vera (2005)).
Budgetary Impacts of the
Simulated ELR Program
The simulated ELR
program has modest budgetary impacts. Figure 8 shows that the total
ELR-related spending (ELRSPEND) in the simulation is generally between 0.6
and 1.25 percent of GDP. The percentage is smallest toward the late 1990s
after economic expansion had continued for several years and the quantity of
ELR workers is at its lowest level for the simulated period. Figure 9 shows
the federal budget (calendar year) as a percent of GDP for both base and
simulation. As with Figure 8, the budgetary effects are modest and a bit
smaller than in Figure 10 due to greater tax revenues as a result of the
higher level of real GDP. Figure 9 also demonstrates, as Mitchell and Wray
(2004) argue, that ELR does not necessitate government deficits; in the
simulation, there are federal surpluses in both 1999 and 2000. In Figure
10, simulated state and local budgets are improved by the ELR program due to
the economy’s enhanced stability and raised level of real GDP; notably, the
fiscal crises encountered by states in the early 2000s is significantly less
severe in the simulation, which itself would have countercyclical benefits
as fewer spending cuts and tax increases would be necessary where balanced
budgets are mandated by law or constitutional amendment.
Comparison of
Macroeconomic Policy Rules with ELR
This section
utilizes stochastic simulation to compare the stabilization properties of
the ELR policy with three other macroeconomic policy rules that have
appeared in Fairmodel-related literature: an estimated interest rate rule,
a sales tax rate rule, and a transfer payment rule. Each of the three
alternative rules are discussed in turn, followed by explanation of the
stochastic simulation procedure, and then discussion of simulation results
and implications.
The Fed’s Interest Rate Reaction
Function in the Fairmodel
The short-term
interest rate reaction function in the Fairmodel is thoroughly discussed in
Fair (2001, 2004, 2005). The structure of the rule is the following:
Where
·
t-i
indicates the lag of i quarters
·
Δ is the level
change from the previous quarter
·
%Δ is the
percentage change from the previous quarter
·
π
is the inflation rate
·
u
is the unemployment rate
·
M1 is the M1
measure of the money supply
·
δ79-82
is 1 during the
period of 1979:4-1982:3 and 0 otherwise
Fair’s goal was to model actual
Fed behavior, rather than “optimal” behavior often modeled in the modern
macroeconomics literature. Fair has thus fit the equation to data for the
period 1952:1-2005:3, estimating the coefficients via two-stage least
squares regression. The R-square is 0.973.
Aside from the
effects of lagged interest rates—which aid the equation’s fit to the
dynamic, “gradualist” nature of the Fed’s interest rate target
changes—according to the Fairmodel’s estimated interest rate reaction
function is a “leaning against the wind” rule that is most influenced by a
negative response to a rise in the unemployment rate (that is, coefficients
α3 and α4 are less than zero) and a positive response
to the inflation rate (α5 is greater than zero). Fair found the
lagged growth rate of the money supply to be statistically significant (and
positive)—though economically less important than other terms in the
equation—in explaining actual Fed behavior for the entire estimation
period. The δ79-82 term is a dummy variable for the monetary
targeting experiment during the early Volcker years.
Fair (2001) finds
that once δ79-82 is added, coefficients for the other variables
are structurally stable for the entire post-1952 period. This is
significant, since the widely held view among orthodox economists is that
the post-1979 Federal Reserve—and especially the Greesnpan-era Fed—has
carried out a much more “optimal” strategy than the pre-1979 Fed, namely by
responding more strongly to inflation than previously (e.g., Clarida, Gali,
and Gertler 2000, 177-8). To the contrary, Fair has found little difference
between the two periods aside from the 1979-1982 period. Interestingly,
Gordon’s (2005) recent study of postwar Fed strategy and its effects is
supportive of Fair’s findings:8
Perhaps the most
surprising finding of this paper is that there has been no change in
monetary policy after 1990 compared to the policies pursued before 1979,
taking a narrow view of policy as the response coefficients in a Taylor Rule
monetary policy reaction function. . . . We show that previous estimates of
Taylor Rule reaction functions are plagued by serial correlation. Once an
autoregressive correction is applied to the Taylor Rule equation, the
post-1990 “Greenspan” policy turns out to look much the same as the pre 1979
“Burns” policy . . . . (8)
Fair (2005) explains
that the total response of the interest rate target to inflation in the
“long run” is roughly 1.0, as given by (α2 + α7) / (1
– α1) and based upon the assumption that growth in the money
supply eventually converges to the inflation rate. For the period
1952:4-2005:3, the response is estimated to be 1.0375 [= (0.072 + 0.011) /
(1 – 0.92)]. According to modern-view (or New Consensus) models based upon
neo-Wicksellian notions of a “neutral” rate of interest, a coefficient of
1.0 would be insufficient for macroeconomic stability; rather, the
coefficient on inflation must be greater than 1.0 such that the real
interest rate rises more than inflation or else the interest rate target
remains “accommodative” (e.g., Taylor 2000; Clarida, Gali, and Gertler
2000). Both Fair (2004 (chapters 7 and 11), 2005) and Giordani (2003)—using
the Fairmodel and a VAR model, respectively—suggest that empirically this is
not the case and that the coefficient on inflation need not be greater than
1.0. Fair (2004, 107-8) argues that modern-view models require larger
coefficients on inflation simply because they impose significant real
interest rate effects on aggregate expenditures; numerous econometric
studies by Fair (e.g., Fair 2004, chapter 3) have found scant empirical
evidence of such real interest rate effects (and tests in Giordani (2003)
support Fair’s results).
Following Fair
(2005), in the simulations below three versions of Fair’s estimated interest
rate reaction function are analyzed:
·
The estimated rule
as shown above.
·
A modified version
of the estimated rule in which the coefficient on inflation (α2)
is increased from 0.072 to 0.109, yielding a “long-run” coefficient on
inflation of 1.5 [= (0.109 + 0.011) / (1 – 0.92)].
·
A modified version
of the estimated rule in which the coefficient on inflation (α2)
is increased from 0.072 to 0.189, yielding a “long-run” coefficient on
inflation of 2.5 [= (0.189 + 0.011) / (1 – 0.92)].
Fair’s Indirect Business Tax
Rate Rule
Fair (2004, chapter
11; 2005) analyzes the stabilization effects of a tax rate rule to see
whether it aids monetary policy.
The idea is that a
particular tax rate or set of rates would be automatically adjusted for each
quarter as a function of the state of the economy. Congress would vote on
the parameters of the tax rate rule as it was voting on the general budget
plan, and the tax rate or set of rates would then become an added automatic
stabilizer.
Consider, for example, the federal gasoline tax. If the short run demand
for gasoline is fairly price inelastic, a change in the after-tax price at
the pump will have only a small effect on the number of gallons purchased.
In this case, a change in the gasoline tax rate is like a change in
after-tax income. Another possibility would be a national sales tax if such
a tax existed. If the sales tax were broad enough, a change in the sales
tax rate would also be like a change in after-tax income. (Fair 2005, 655)
Within the Fairmodel, let τ*
be the base indirect business tax rate. Fair then set a new rate each
quarter, τ, according to previous quarters’ levels of real GDP and inflation
rates versus targeted values of each in the same quarters. The adjustment,
intended as an example of a tax rate rule rather than an “estimate,” is as
follows:
where,
·
t-i
indicates the lag of i quarters
·
*
indicates a variable’s targeted value (by policymakers)
·
y
is real GDP
·
π
is the inflation rate
According to the rule, the
existing indirect business tax rate is thus raised (lowered) 12.5 percentage
points for every 1 percent excess (deficiency) of real GDP from the target
(i.e., full employment or “potential”) level of real GDP averaged across the
past two quarters, and another 12.5 percentage points for every 1 percentage
point excess (deficiency) in the inflation rate from the targeted rate
averaged across the past two quarters. (Presumably a more broad-based
national sales tax would enable percentage point changes to the tax rate
that were smaller in size.)
In the simulations
below, the suggested tax rate rule is analyzed both with and without Fair’s
estimated interest rate rule.
Seidman and Lewis’s Asymmetric
Transfer Payment Rule
Seidman and Lewis
(2002) propose an asymmetric fiscal policy rule which triggers transfer
payments “in response to a decline in the output of the economy of
particular magnitude” (262). The policy rule is asymmetric because it “does
not attempt to restrain demand when demand is excessive. That task is left
to monetary policy because we suspect that Congress may be more willing to
pre-enact fiscal stimulus than to pre-enact fiscal restraint” (262). The
basic rule, which determines an aggregate transfer payment, tr, is
the following:
where,
·
t-i
indicates the lag of i quarters
·
*
indicates a variable’s targeted value (by policymakers)
·
y
is real GDP
·
α
is a power coefficient
·
χ
is a threshold that must be
cleared before transfer payments are triggered.
Thus, once real GDP is reduced
below the target (i.e., full employment or “potential”) level, if the
percentage point deficiency is larger than χ, then aggregate transfer
payments are triggered as a percent of target real GDP and according to the
power coefficient, α. If the deficiency does not exceed χ then no transfer
is triggered. Seidman and Lewis’s proposal requires that Congress pre-enact
the values of α and χ. They considered three different pairs that were
labeled policies F1, F2, and F3; these were illustrated by assuming a deep
recession in which the previous quarter’s real GDP was 4 percent
below its target or full employment level.
·
For policy F1, α =
0.5 and χ = 2%; tr = 0.5 (4% – 2%) = 1% of the previous quarter’s
level of y*.
·
For policy F2, α =
1.5 and χ = 2%; tr = 1.5 (4% – 2%) = 3% of the previous quarter’s
level of y*.
·
For policy F3, α =
1.5 and χ = 0%; tr = 1.5 (4% – 0%) = 6% of the previous quarter’s
level of y*.
Obviously, with policies F2
and—especially—F3, transfer payments triggered by the rule can be very
large. Recognizing this, Seidman and Lewis also proposed a possible
modification in which the transfer, tr, is reduced if the inflation
rate is above the targeted level:
Here, the reduction in the tr
is enacted only if actual inflation in the previous quarter was
larger than the targeted rate. As an example, they proposed β = 0.8; if
inflation in the previous quarter was 5 percent higher than its targeted
rate, then tr is reduced by 60 percent (since 0.6 = 1 – 0.8 x 5%).
The calculated value
of tr, whether with the modification for inflation or without it, is
multiplied by a price deflator (ph, which is the deflator for the
household sector in the Fairmodel) and this amount is added to the level of
transfer payments that would have occurred without the rule. In the
Fairmodel, trgh denotes transfer payments from the federal government
to households. If trgh* is used to denote the level of transfers
absent the transfer rule, the new level of transfers, trgh, is given
by:
Since trgh is untaxed,
the additional transfer payment adds directly to disposable income and
impacts all household spending and sales equations.
Seidman and Lewis
simulated the transfer rules (F1, F2, and F3) using the Fairmodel and found
all three rules to be helpful in mitigating the effects of deep recessions.
In the simulations below, the three proposed transfer rules are analyzed,
both with and without Fair’s estimated interest rate rule. Also, each
transfer rule is modified as above for deviations in inflation from its
targeted rate (assuming β = 0.8) and analyzed both with and without Fair’s
estimated interest rate rule. In sum, twelve total versions—four versions
of each of the three proposed transfer rules—are simulated and analyzed.
The Stochastic Simulation
Procedure
The stochastic simulation
procedure (Fair 2004, 2005) is carried out here using 1993-1997 as the base
period, though the choice of base period is actually unimportant. The
procedure involves (1) adding historical residuals to the stochastic
equations during 1994-1997 to create a base path that is the identical to
the historical path, (2) drawing residuals randomly for one quarter from a
selected set of past quarters (in the manner discussed below) and adding
these to existing residuals for the given quarter, (3) solving the model’s
130 equations for the given quarter, and (4) as steps 1-3 makeup one trial,
these steps are repeated this procedure many times. In this case, 100
trials are simulated for 1993-1997; since each trial is 20 quarters long,
this means a total of 2000 quarters are simulated. The period for
historical error draws is 1954:1 to 2005:3, or 207 quarters in length, which
means there are 207 vectors of 30 residuals from which to draw.
For each quarter, an integer
between 1 and 207 is randomly drawn with probability 1/207; this draw
determines which of the 207 vectors of residuals is used for that quarter as
discussed in (2) and (3) above. The solved model is an estimate of how the
economy might have performed had the particular “shocks” as represented by
the draw of residuals actually occurred. The advantage of using historical
error terms is that no distributional assumption has to be made and no zero
restrictions have to be imposed; further, the “shocks” simulated are
realistic since they actually have occurred in the past.
As mentioned
earlier, Fair’s estimated short-term interest rate reaction function is
incorporated into some of the simulations. As in Fair (2004, chapter 11;
2005), estimated residuals are added to the interest-rate equation, but no
errors are drawn. As with other variables, the base case for the Fed’s rule
is a perfect tracking solution for the actual short-term rate. Not drawing
errors, however, ensures that the Fed does not behave randomly but rather
simply follows the rule in response to shocks to the other stochastic
equations. Following Fair (2004, 161), let
be
the simulated value of an endogenous variable i (e.g., real GDP,
inflation, unemployment) in quarter t on trial j, and let
be
the historical or base value of variable i for quarter t. One
could simply compute the standard deviation for the variable after several
trials, but the problem is that there are 20 quarterly values simulated here
per variable, which makes comparison across quarters difficult. Instead,
Fair suggested calculating a value, Li, for each variable,
i, using the following two-step procedure:
First, allowing T to
denote the length of the simulation period (T=20 quarters here), let
be
.
is
the average value of the squared quarterly deviations of the 20 simulated
values of variable i for a given trial j from the 20 quarterly
base values. Second, recalling that J=100 here, let Li
be
.
Li
is then a measure of the average
within quarter deviation of variable i from quarterly base
values in a simulation of length T quarters and averaged across all
J trials. In essence, the lower Li is, the more
effective a given policy was at offsetting or neutralizing the shock by
returning the economy back toward its base values. In other words, Li
= 0 implies that the policy was able to completely offset all shocks and
return the economy to the base value of variable i (i.e., the real
GDP or inflation “gap” was zero).
As mentioned already,
stochastic simulation was carried out for the three previously discussed
policy rules (interest rate rule, tax rate rule, and transfer rule),
including some variations of each. A significant concern for policy
implementation would be how real GDP and inflation targets were determined
in the tax rate and transfer rules, but for stochastic simulation this
concern is easily sidestepped since it simply considers how well the policy
rules offset deviations from base values brought on by the randomly drawn
“shocks.” Thus, for the tax rate and transfer rules, the actual base values
of real GDP and inflation serve as proxies for y* and π*,
respectively, and Li measures how well these policies
offset shocks to keep the economy at or close to these base values. This
explains why it is unimportant which time period is selected to be used as
base values for the stochastic simulation (Fair 2004, 2005).
Each of the three policy rules
targets an inflation rate and a level of real GDP. Note that
a policy could significantly raise Lprice level if a
one-time jump in the price level occurred even if the price level remained
very stable thereafter. As such, it is Linflation that is
more important than Lprice level. Similarly, each of the
policy rules theoretically targets full employment real GDP (i.e., no real
GDP “gap”) rather than a growth rate of real GDP; indeed, a policy with a
low value for Lreal GDP could have a high value for Lreal
GDP growth, particularly if alternating positive/negative shocks
are occurring. As such, it is Lreal GDP that is more
important than Lreal GDP growth. However, if two policies
are fairly equally successful at reducing Lreal GDP, if
one also is more successful at reducing Lreal GDP growth
(that is reducing real GDP variability across quarters) then this
would seem to be preferable t
