Research in Economics 66 (2012) 149160

Contents lists available at SciVerse ScienceDirect

Research in Economics
journal homepage: www.elsevier.com/locate/rie

Income inequality and economic incentives: Is there an

equityefficiency tradeoff?
Lonnie K. Stevans
Hofstra University, Zarb School of Business, Department of IT/QM, 134 Hofstra University, Hempstead, NY 11549-1270, United States

article

info

Article history:
Received 18 January 2011
Accepted 24 October 2011
Keywords:
Cointegration
Vector Error Correction (VEC)
Income inequality
Equity
Efficiency
Unit root
Hysteresis
Incomplete markets
Political outcomes

abstract
What is the basis and direction of relationship between income inequality and economic
growth? The equity versus efficiency dictum which predicts a positive relationship
between inequality, capital formation, and real GDP growthemphasizes the importance
of economic incentives. Subsequently, this was challenged by the incomplete markets
and political outcomes theories, because of increasing empirical evidence of an inverse
relationship between income inequality and economic growth. In this paper, we offer a
further explanation of the basis and nature of the inequalitycapitalgrowth relationship
which emphasizes the divergence between savings and investment. For the United States
over the period 19702006, we have found no empirical evidence for the support of
the equity versus efficiency hypothesisthat economic incentives are necessary for capital
accumulation and growth. In fact, it was discovered that in most cases, inequality has had
little or no impact on movements in the US capital stock, net investment, and consequently,
economic growth. Another interesting finding of this study was that inequality exhibits
hysteresisimplying that any positive shock, such as the dot-com boom, can lead to
persistent and enduring increases in inequality.
2011 University of Venice. Published by Elsevier Ltd. All rights reserved.

1. Introduction
Can we have less inequality without reducing prosperity in the United States? In the US, the public finance literature has
primarily focused on the measurement of efficiency losses associated with government programs and policies. According
to Okun (1975), the efficiency cost of income redistribution or economic regulations may be large enough to result in less
national income. Thus, the argument is that although inequality may be reduced, everyone will be worse-off because there
would be less entrepreneurial-type or rent-seeking behavior and diminished labor/capital productivity-resulting in a lower
standard of living.
From 1990 to 2000, the United States has exhibited a high rate of economic growth (3.3%) as compared to other
industrialized nations, and contemporaneously the greatest increase in inequality since the late 1970s. In contrast, many East
Asian economies in the post-World War period experienced relatively low levels of inequality (for countries of comparable
income levels), yet grew at extraordinary rates and many Latin American countries had higher levels of inequality and
grew at a fraction of the average East Asian rate. These phenomena prompted an interest in the relationship between
inequality and growth, and in particular to a conundrum regarding the correlation between inequality and economic growth:
what is the direction of relationship between inequality and economic growth? There is ample lip service paid to the
disincentives and/or inefficiencies associated with redistribution and the resultant adverse effect on economic growth

Tel.: +1 516 463 5375 (Office), +1 631 598 8518 (Home).

E-mail addresses: acslks@hofstra.edu, Lonnie.K.Stevans@hofstra.edu.

1090-9443/$ see front matter 2011 University of Venice. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.rie.2011.10.003

150

L.K. Stevans / Research in Economics 66 (2012) 149160

in popular media writings.1 The notion that higher inequality is both a necessary and sufficient condition for increasing
economic growth appears to be an uncontested truth. On the other hand, in contrast to the positive relationship posited
by the equityefficiency approach, a number of studies in the academic literature have found an inverse and statistically
significant relationship between inequality and economic growth (Barro and Xavier, 1995). The theoretical construct behind
these approaches is grounded in the notion that greater inequality either stimulates or discourages productive investment
(depending on the policy involved) and ultimately GDP.2 In this paper, a Keynesian raison detre will be offered as an
alternative explanation of the relationship between inequality and a countrys capital stock. Following this, the association
between inequality and capital investment in the United States between 1970 and 2006 will be examined by using a time
series approach incorporating both the effect of inequality on net investment (short-run) and capital formation (long-run).
It is important to note that although policies changing inequality may also affect the labor market, only the impact on capital
productivity will be studied here.3
2. Theory
In a perfectly competitive market, there would be no impact of inequality on productivity. The only relationship that may
exist would result from policy attempts that influence inequality and also distort incentives. For example, a more progressive
tax system would reduce inequality, but may also create a deadweight loss and diminish work-effort.4 Okun (1975) has
discussed the tradeoffs associated with equity and efficiency a policy imposing more redistribution or less inequality may
generate less national income resulting in a positive relationship between inequality and economic growth.
In the 1990s, this view was challenged as a result of increasing cross-country empirical evidence of a negative relationship
between income inequality and economic growth (Persson and Tabellini, 1994; Alesina and Rodrik, 1994; Deininger and
Squire, 1996). The existence of this inverse relationship has led to the development of a number of theories to explain
the empirical evidence. One is known as incomplete markets, which affirms that an impact (not induced) of inequality on
productivity can only arise when there is market failure. This approach emphasizes the role of borrowing constraints and
externalities in generating the observed negative relationship between inequality and growth. When there are decreasing
returns to capital and credit rationing, the aggregate level of output may be affected by its distribution (Stiglitz, 1969).
Credit rationing occurs when there exist individuals who could profitably invest borrowed funds and repay with interest,
but lenders are unwilling to lend to them in full. When this market failure arises, it drives productive borrowers out of the
loan market leading to an inefficient allocation of resources, underinvestment, and reduced productivity. In this approach,
the poor are prevented from choosing the most productive activity available given their skills, because imperfect information
and incomplete contracts cause a credit market failure. Loans that would have been good are not made, and applicants that
are turned down remain poorer than they would otherwise be.
The political process also can explain the relationship among inequality, government policy, and economic growth.
Political outcomes determining government policy are endogenous to the distribution of income and rational economic
agents vote for or against tax policies which have redistributive consequences. Greater inequality would result in higher tax
rates since a larger proportion of voters will favor redistributive policies. As a result, the after-tax return of capital is reduced,
thus diminishing investment and economic growth (Bertola, 1991; Alesina and Rodrik, 1994; Persson and Tabellini, 1992,
1994). This approach also predicts a negative relationship between inequality and economic growth.
There is another way which changes in inequality can have little or no impact on economic growth through the capital
markets (other than the aforementioned trivial case of a competitive market without government intervention). The
equityefficiency approach emphasizes the importance of incentives. For example, according to this argument, if tax rates
for the rich are reduced, the argument is that this should create incentives for the rich to save and invest more, increasing
the demand for capital goods, and thus expanding economic growth. Of course, this trickle-down depends in large part
upon the linkage between inequality, savings, investment, the capital stock, and GDP. But a policy that augments inequality
and concomitantly the savings of the rich may indeed result in little or no increase in net investment spending. This would
result from any one (or both) of the following reasons: (1) since the decisions of savers and investors are essentially separate
and distinct from one another, there is no reason to expect that additional savings will generate the requisite amount of
investment spending in the economy, due to liquidity concerns5 ; and (2) even if the savings and investment of the rich
increased, the investment schedule may be interest inelasticresponding more to changes in income than fluctuations in
the interest rate. Consider the following empirical finding by Kopcke (1993),
Because all the models (in this study), either implicitly or explicitly, stress that investment is undertaken in anticipation of
profit, the prospect of a greater demand for output is a principal spur for capital spending.

1 See Roger Lowenstein, The Inequality Conundrum, How can you promote inequality without killing off the genie of American prosperity? The New York
Times Magazine, 10 June 2007, pp. 1114.
2 Productive investment is defined as real net investment in physical capital goods.
3 These would be the policies pronounced by the supply-siders in the 1980s, such as the Laffer Curve which professed increased work-effort (with an
upward-sloping and elastic labor supply) when tax rates were reduced.
4 Of course, this depends upon the elasticity of labor demand and supply.
5 John Maynard Keynes, The General Theory of Employment, Interest, and Money, First Harvard, Harcourt, Inc.: 1936.

L.K. Stevans / Research in Economics 66 (2012) 149160

151

Thus, inequality could have little or no effect on net investment in the short-run, little or no influence on capital formation
in the long-run, and therefore no discernible effect on economic growth. This result runs counter to the viewpoint that relies
upon a trade-off between equity and efficiency, since, as mentioned previously, the argument put-forth is that if a society
tries too hard at reducing inequality, there will be fewer incentives resulting in diminished productivity along with a
lower-standard of living.
Empirically, both the microeconomic and macroeconomic evidence concerning the relationship between inequality and
growth is far from conclusive. Forbes (2000) has found a positive relationship between inequality and economic growth.
Her estimates are based upon panel data (with country fixed effects) over a five year period. However, there were earlier
studies whose authors observed a negative impact of inequality on growth using 2530 years across countries (Persson and
Tabellini, 1994; Alesina and Rodrik, 1994; Perotti, 1993; Deininger and Squire, 1998). When Forbes (2000) and Barro (1999)
estimate their regressions over ten year intervals, the relationship became insignificant. The empirical evidence seems to
indicate that a positive short-run relationship becomes reversed over longer periods.
In sum, it is possible to identify three predictions from the literature as to the direction of the relationship between
inequality and investment/capital formation,

the equityefficiency view would predict a positive relationship between inequality and capital formation because of
incentives;

the incomplete markets and political outcomes theories would predict an inverse relationship between inequality and
capital formation due to incomplete information, and;

the Keynesian approach presented in this study would predict no perceptible relationship between inequality and capital
formation as a consequence of the interest inelasticity of investment spending.
The question as to which prediction dominates, is an empirical one that will be tested using quarterly time series data
for the US from 1970 to 2006the relationship amongst the US capital stock, income inequality, the rental price of capital,
the price level, and real GDP, will be examined. All of these variables are treated as jointly endogenous in the context of a
reduced form VEC/cointegration model. The advantage to using this approach is twofold: first, the statistical results are not
subject to endogeneity bias, since the models used have only predetermined or exogenous variables on the right-hand side.
Second, given testing for cointegrating relationships, there is little concern about the problem of spurious associations
among the variables that may exist when one simply correlates two or more random walks with each other (Enders, 2004).
The purpose will be to determine and test the direction of cointegrating relationships of income inequality measures with
the US capital stock in the long-run.
3. Empirical model
Following Beare (1978), the structural form of the demand for capital at time t , Kt , is specified as a function of the rental
price/user cost of capital, c, the price level, P, and output, Q ,
Kt = f (ct , Pt , Qt , t ),
ct = PKt (t + it P Kt ),

(1)

Kt demand for capital goods,

ct rental price/user cost of capital services,
PKt price of capital goods,
P Kt dPKt /dt,
t capital goods depreciation rate,
it nominal interest rate,
t random error.
Ignoring the question of what measure to use for the moment, the structural form of inequality may be expressed as,
INt = g (Pt , Qt , t ),

(2)

where INt is income inequality. This specification involving prices and output on the right-hand side of Eq. (2) has empirical
support dating back to Kuznets (1955). Considering a linear reduced form, Eqs. (1) and (2) may be expressed as,6
Kt =

a11i Kt i +

i=1
p

INt =

i=1

a12i INt i +

i=1
p

a21i Kt i +

i =1

a13i ct i +

i=1
p

a22i INt i +

i=1

a14i Pt i +

i=1
p

a23i ct i +

i =1

a15i Qt i + t

i =1
p

a24i Pt i +

(3)
a25i Qt i + t .

i=1

All variables are endogenous. If the variables are random walks, (integrated of order one), the reduced form can be
appropriately restricted. Using all of the variables and creating a Vector Error Correction (VEC) model,
6 It is important to note that this is a VAR model and the equations for c , P , and Q are omitted for reasons of brevity.
t
t
t

152

L.K. Stevans / Research in Economics 66 (2012) 149160

1 X t = X t 1 +

p1

i 1 X t i + t ,

(4)

i =1

where,

X t = (Kt INt ct Pt Qt )

= I

Ai

i =1

i =

Aj

j=i+1

Ai 5 5 matrix of the parameters as,k,i

s, k = 1, 2, 3, 4, 5 i = 1, 2, 3, . . . , p.

If the matrix has full rank (r = 5), then all components of X t are stationary or integrated of order zero. On the other
hand, if the rank of the matrix is less than five, then there are (5 r) common stochastic trends and r stationary relationships.

In this case, the transformation i X it is stationary and unique if r = 1. is the matrix of cointegrating parameters and
is the matrix of adjustment weights with which each cointegrating vector enters the five equations of the VEC. The can
also be considered as the matrix of the speed of adjustment parameters. Our interest lies with the unique case when r = 1.

X it may be written as,

t = 1 Kt + 2 INt + 3 ct + 4 Pt + 5 Qt ,

(5)

which is nothing more than a linear combination of the variables. One motivation for the VEC form is to consider X it as
defining the underlying economic relations and assume that the agents react to a shock through the adjustment coefficient
to restore equilibrium, that is, they satisfy the economic relationship (5) when t = 0. The econometric use of the
term equilibrium is any long-run relationship among nonstationary variables.7 The cointegrating vector, , is sometimes
referred to as the vector of long-run parameters. We can also normalize Eq. (5) with respect to Kt ,
Kt = 1 + 2 INt + 3 ct + 4 Pt + 5 Qt + t

i =

i
1

t =

t white noise.

t
1

(6)

Eq. (6) is nothing more than a linear (or logarithmic) form of the capital equation (1) with all of the endogenous
variables including inequality. Since all of the variables are in natural logarithmic units, the i and their estimates are
elasticity coefficients and may be interpreted as the percentage change in capital given a one percent change in the relevant
explanatory variable, ceteris paribus.
It is important to note that while Eq. (6) characterizes movement in the long-run capital stock, the first equation of the
system represented by the VEC (Eq. (4)) represents the period-to-period adjustment of changes in the capital stock, or net
investment. The j parameters and their respective estimates in the VEC represent the short-term effects (year-to-year) that
changes in inequality, the user cost of capital, the price level and output have on net investment.
4. Estimation results
A complete description of the data used for the above variables, Kt , INt , ct , Pt , and Qt is in Appendix, and the time plots
of Kt , Pt , ct , and Qt (from left to right) over from 1970 to 2006 may be found in Fig. 1.8 All are expressed in natural logarithm
units.
Four measures of inequality are used in this analysis,

Gini coefficient,
Ratio of income share of top five percent relative to income share of lower twenty percent,
Ratio of income share of top five percent relative to income share of lower forty percent,
Ratio of income share of top five percent relative to income share of lower sixty percent, and
Ratio of income share of top five percent relative to income share of lower eighty percent.

7 Cointegration does not require that the long-run relationship be generated by market forces or by the behavioral rules of individuals. In Engle and Granger
(1987) use of the term, the equilibrium relationship may be causal, behavioral, or simply a reduced form relationship among similarly trending variables
(Enders, 2004).
8 Although all of the analyses used quarterly data from 1970 to 2006, the figures (graphs) are expressed in annual averages.

L.K. Stevans / Research in Economics 66 (2012) 149160

153

Fig. 1. Real net investment, implicit price deflator, rental value of capital, and real gdp (annual and in natural log units).

The chained quantity index of the capital stock, Kt , is graphed along with each of the above inequality measures in
Fig. 2.9 It is important to note that there is a structural increase in each of the inequality measures beginning in 1992. Many
reasons have been given for this rise in inequality: immigration, outsourcing, rising executive compensation, but according
to University of Texas researchers James K. Galbraith and Travis Hale, much of the increase in income inequality in the
late 1990s resulted from large income changes in just a few locations around the countryprecisely those areas that were
heavily involved in the information technology boom.10 In addition to the level increase in 1992, it is also notable that the
share of income going to the upper five percent relative to the lower forty and sixty percent was higher (for just about the
entire period) than the income share of the upper five percent to the lower twenty percent. This lends support to the popular
notion that the middle class has been losing ground relative to the rich over the past thirty years, since these particular
inequality measures are essentially comparisons of the upper five percent tail relative to different size lower tails of the
income distribution. The lower forty and sixty percent tend to encompass more of the middle class than the lower twenty
or lower eighty percent of the distribution.
4.1. Unit root tests
Since each of the time series in a VEC/cointegration analysis must be integrated of order one, (I (1)), each series should
be tested for the presence of a unit root. However, it is well known that the usual unit root tests are biased toward accepting
the null hypothesis of a unit root in the presence of structural change.11 Perron (1989) has developed a formal procedure to
test the null hypothesis of a unit root with a one-time impulse change,
H0 : Yt = 0 + 1 DP92 + Yt 1 + t ,

(7)

9 Ibid.
10 http://www.ncpa.org/sub/dpd/index.php?Article_ID=12523. It has also been said that the increase in 19921993 was partially due to a change in the
Census Bureau data collection methods.
11 See Enders, Walter, Applied Econometric Time Series, Second Edition, John Wiley and Sons, Inc.: New York, NY, 2004, page 201.

154

L.K. Stevans / Research in Economics 66 (2012) 149160

Fig. 2. Capital stock vs. inequality measures (annual, 1970 = 100).

versus the alternative of a step level change in the intercept of a trend stationary process,
HA : Yt = 0 + 1 D92 + 2 t + t ,

(8)

where DP92 = 1 when t = Q 4 : 1991 + 1 = Q 1 : 1992 and zero (0) otherwise, and D92 = 1 when t > Q 1 : 1992 and
zero (0) otherwise. The question is whether in the presence of an exogenous shock (the dot-com boom), can inequality be
characterized as a process that is not mean reverting or is it trend stationary? Using Perrons (1989) method, we consider
the following regression for each inequality measure, INit ,
INit = i0 + i1 DP92 + i2 D92 + i3 t + i INit 1 +

ij 1INit j + it i = 1, 2, 3, 4, 5.

(9)

j =1

Under the null hypothesis of a one-time impulse change in the level of the unit root process, i = 1, i1 = 0, i3 = 0.
The results of this estimation and unit root tests are presented in Table 1. According to Perron (1989), the distribution of
depends upon , (found in Table 1), which is the proportion of observations occurring before the break. It should be noted
that the null hypothesis of a unit root could not be rejected for each of the five inequality measures. Thus, inequality exhibits
hysteresis, implying that any shock, such as the dot-com boom, can lead to persistent and enduring increases in inequality.12
Efficient unit root tests developed by Elliott et al. (1996) were run for the remaining variables, Kt (capital stock), ct (rental
price/user cost of capital), Pt (price level), and Qt (real GDP). In each case, the null hypothesis of a unit root could not be
rejected.13
4.2. Cointegration tests
In addition to modifying unit root tests because of the presence of structural change, cointegration tests have also been
developed for a system of variables with level shifts. Saikkonen and Lutkepohl (2000) propose to first adjust the time series
for deterministic terms and then apply the usual likelihood ratio tests for cointegration to the adjusted series.14 Suppose an

12 It is important to note that the Perron (1989) test allows for a known and exogenous structural break in the time series. Lee and Strazicich (2003)
(LS) have proposed a Lagrange Multiplier (LM) unit root test with endogenous breaks in which the alternative hypothesis implies that the series is trend
stationary. The findings from applying the LS test confirmed the Perron (1989) resultsthe inequality series all had a unit root with an endogenous break
in the second quarter of 1992. Results will be made available from the author upon request.
13 The results are omitted for brevity, but will be made available from the author upon request.
14 Because of these adjustments, there are no other cointegration tests available that have superior local power and size properties (Lutkepohl et al.,
2003).

L.K. Stevans / Research in Economics 66 (2012) 149160

155

Table 1
Structural change unit root test.
H0 : inequality measure has a unit root
Dependent variable: LOG_GINI
Sample (adjusted): 1971 2006
Included observations: 116 after adjustments
Variable

Coefficient

Std. error

t-statistic

Prob.

C
DP92
D92
@TREND
LOG_GINI(1)
D(LOG_GINI(1))
D(LOG_GINI(2))
D(LOG_GINI(3))
D(LOG_GINI(4))

0.617835

0.224467
0.005583
0.007483
0.000881
0.195190
0.126437
0.112583
0.101407
0.092698

2.752459
5.574280
1.454566
2.958171
2.709765
1.884834
0.724271
2.011069
1.687300

0.0113
0.0000
0.1593
0.0070
0.0722
0.4762
0.0562
0.1051

0.031120
0.010884
0.002607
0.471081
0.238313
0.081541
0.203936
0.156409

Perrons t-statistic

3.76

0.57

Perrons t-statistic

3.76

0.57

Perrons t-statistic

3.76

0.57

Perrons t-statistic

3.76

0.57

= 0.05

Dependent variable: LOG_U05_L20

Sample (adjusted): 1971 2006
Included observations: 116 after adjustments
Variable

Coefficient

Std. error

t-statistic

Prob.

C
DP92
D92
@TREND
LOG_U05_L20(1)
D(LOG_U05_L20(1))
D(LOG_U05_L20(2))
D(LOG_U05_L20(3))
D(LOG_U05_L20(4))

0.243914
0.120555
0.034696
0.005599
0.616914
0.171179
0.179293
0.205737
0.008166

0.117148
0.017947
0.020996
0.001824
0.142795
0.108568
0.084555
0.103652
0.104379

2.082089
6.717354
1.652524
3.070513
2.682769
1.576699
2.120445
1.984872
0.078235

0.0477
0.0000
0.1109
0.0051
0.1274
0.0441
0.0582
0.9383

= 0.05

Dependent variable: LOG_U05_L40

Sample (adjusted): 1971 2006
Included observations: 116 after adjustments
Variable

Coefficient

Std. error

t-statistic

Prob.

C
DP92
D92
@TREND
LOG_U05_L40(1)
D(LOG_U05_L40(1))
D(LOG_U05_L40(2))
D(LOG_U05_L40(3))
D(LOG_U05_L40(4))

0.065558

0.051027
0.026963
0.035996
0.002366
0.212230
0.081272
0.123667
0.092445
0.133617

1.284770
3.319969
1.645806
2.478462
2.272600
3.357367
1.598362
2.073797
0.114280

0.2106
0.0028
0.1123
0.0203

0.089517
0.059242
0.005864
0.517686
0.272861
0.197665
0.191712
0.015270

0.0025
0.1225
0.0486
0.9099

= 0.05

H0 : inequality measure has a unit root

Dependent variable: LOG_U05_L60
Sample (adjusted): 1971 2006
Included observations: 116 after adjustments
Variable

Coefficient

Std. error

t-statistic

Prob.

C
DP92
D92
@TREND
LOG_U05_L60(1)
D(LOG_U05_L60(1))
D(LOG_U05_L60(2))
D(LOG_U05_L60(3))
D(LOG_U05_L60(4))

0.198711

0.094462
0.025624
0.033957
0.001729
0.188057
0.080190
0.116001
0.059699
0.118587

2.103601

0.0456
0.0003
0.2133
0.0344

0.108690
0.043363
0.003868
0.649258
0.246053
0.238751
0.198791
0.010180

4.241681
1.276976
2.237269
1.865083
3.068388
2.058179
3.329891
0.085841

0.0051
0.0501
0.0027
0.9323

= 0.05
(continued on next page)

n-dimensional time series, X t , is generated by the following mechanism,

X t = 0 + 1 t + 2 DP + 3 D + Y t ,

(10)

156

L.K. Stevans / Research in Economics 66 (2012) 149160

Table 1 (continued)
Dependent variable: LOG_U05_L80
Sample (adjusted): 1971 2006
Included observations: 116 after adjustments
Variable

Coefficient

Std. error

t-statistic

Prob.

C
DP92
D92
@TREND
LOG_U05_L80(1)
D(LOG_U05_L80(1))
D(LOG_U05_L80(2))
D(LOG_U05_L80(3))
D(LOG_U05_L80(4))

0.273517

0.107251
0.021232
0.027834
0.001026
0.149826
0.100602
0.139080
0.092232
0.107734

2.550250

0.0173
0.0002
0.1288
0.0127

0.092729
0.043727
0.002756
0.656488
0.252079
0.205751
0.213103
0.035729

4.367401
1.570997
2.685132
2.292739
2.505706
1.479373
2.310514
0.331636

Perrons t-statistic

3.76

0.57

0.0191
0.1515
0.0294
0.7429

= 0.05

where DP is an impulse dummy and D is a step level dummy variable. Y t is a stochastic error which is assumed to have a
VAR process with the VEC representation,

1 Y t = Y t 1 +

p1

i 1 Y t i + t .

(11)

i =1

Saikkonen and Lutkepohl (2000) recommend forming the series,

Yt = X t 0 1 t 2 DP 3 D,

(12)

then performing the usual Johansen (1988) cointegration tests using the VEC,

1Yt = Yt 1 +

p1

i 1Yt i + t .

(13)

i =1

This adjustment was only made for the five inequality measures, since stability tests indicated that there were no
discernible breakpoints from 1970 to 2006 for each of the remaining series.15 The results of the cointegration tests are
in Table 2.16 It is important to note that in each case involving the maximum eigenvalue tests for the null hypothesis of
H0 : Cointegrating Rank = 0 versus the alternative HA : Cointegrating Rank = 1, the null hypothesis was rejected in favor of
the unique alternative. Thus, these single, cointegrating relationships may be represented by the normalized linear models
(Eq. (6)),
Kt = 1j + 2j INtj + 3j ct + 4j Pt + 5j Qt + tj

j = 1, 2, 3, 4, 5,

(14)

17

where the j represents each of the five inequality measures.

4.3. Vector error correction estimation

The estimation results of the single cointegrating equations (Eq. (14)) are presented in Table 3. In each of the equations,
the variables ct , Pt , and Qt all have the expected sign (negative, positive, and positive, respectively) and are statistically
significant at = 0.01 in every equation. The only two inequality measures that have a statistically significant influence
on the US capital stock are the ratio of the upper five percent to the lower forty percent and the ratio of the upper five
percent to the lower sixty percent (bolded in Table 3). Moreover, the sign of both coefficients are negative. As mentioned
previously, the lower forty and sixty percent tail of the income distribution includes more of the middle class than the lower
twenty or lower eighty percent of the distribution and the movement in these particular ratios represents the erosion in
the relative position of middle class household income. However, according to the empirical results, more inequality, as
measured by these increasing ratios, has served to reduce the nations capital stock and consequently economic growth in
the long-run (and decrease net investment in the short-run).18 This outcome runs counter to the hypothesized positive effect
of the equityefficiency tradeoff and appears to lend empirical support for the incomplete markets and political outcomes

15 Using EViews 6, the QuandtAndrews breakpoint test over the period 19702006 was performed on K , c , P , and Q and the null hypothesis of no
t
t
t
t
breakpoints within the trimmed data could not be rejected in each case. Results will be made available from author upon request.
16 The lag length for each model, p, was determined by using the sequential modified likelihood ratio test in EViews 7.1.
17 Remember that the inequality measures, IN , are essentially residualsthe effect of the deterministic variables have been removed.
tj
18 The effect of inequality on net investment will be presented shortly.

L.K. Stevans / Research in Economics 66 (2012) 149160

157

Table 2
Cointegration tests.
Inequality measure: Gini coefficient
Sample (adjusted): 1970 2006
Included observations: 118 after adjustments
Lags interval (in first differences): 1 to 2
Unrestricted cointegration rank test (maximum eigenvalue)
Hypothesized no. of CE(s)

Eigenvalue

Max-eigen statistic

0.05 critical value

Prob.

None
At most 1
At most 2
At most 3
At most 4

0.780867
0.522302
0.516642
0.392361
0.331507

51.61459
25.11841
24.71790
16.93792
13.69281

38.33101
32.11832
25.82321
19.38704
19.41798

0.0009
0.2795
0.0694
0.1095
0.1095

Eigenvalue

Max-eigen statistic

0.05 critical value

Prob.

0.756527
0.591848
0.480530
0.409836
0.371872

46.62078
29.57180
21.61322
17.40268
15.34535

38.33101
32.11832
25.82321
19.38704
19.42312

0.0045
0.0992
0.1634
0.0949
0.0951

Eigenvalue

Max-eigen statistic

0.05 critical value

Prob.

0.787474
0.576534
0.463000
0.357559
0.308806

51.10687
28.35631
20.51797
14.60186
12.18804

38.33101
32.11832
25.82321
19.38704
19.51798

0.0011
0.1346
0.2148
0.2162
0.2255

Max-eigenvalue test indicates 1 cointegrating eqn(s) at the 0.05 level.

Denotes rejection of the hypothesis at the 0.05 level.

MacKinnonHaugMichelis (1999) p-values.

Inequality measure: ratio of top five percent to bottom twenty percent
Sample (adjusted): 1970 2006
Included observations: 118 after adjustments
Lags interval (in first differences): 1 to 2
Unrestricted cointegration rank test (maximum eigenvalue)
Hypothesized no. of CE(s)

None
At most 1
At most 2
At most 3
At most 4

Max-eigenvalue test indicates 1 cointegrating eqn(s) at the 0.05 level.

Denotes rejection of the hypothesis at the 0.05 level.

MacKinnonHaugMichelis (1999) p-values.

Inequality measure: ratio of top five percent to bottom forty percent
Sample (adjusted): 1970 2006
Included observations: 118 after adjustments
Lags interval (in first differences): 1 to 2
Unrestricted cointegration rank test (maximum eigenvalue)
Hypothesized no. of CE(s)

None
At most 1
At most 2
At most 3
At most 4

Max-eigenvalue test indicates 1 cointegrating eqn(s) at the 0.05 level.

Denotes rejection of the hypothesis at the 0.05 level.

MacKinnonHaugMichelis (1999) p-values.

Inequality measure: ratio of top five percent to bottom sixty percent
Sample (adjusted): 1970 2006
Included observations: 118 after adjustments
Lags interval (in first differences): 1 to 2
Unrestricted cointegration rank test (maximum eigenvalue)
Hypothesized no. of CE(s)

Eigenvalue

Max-eigen statistic

0.05 critical value

Prob.

None
At most 1
At most 2
At most 3
At most 4

0.800354
0.572472
0.457606
0.373281
0.343586

53.16999
28.04123
20.18820
15.41949
13.89181

38.33101
32.11832
25.82321
19.38704
19.56799

0.0005
0.1453
0.2325
0.1719
0.1822

Max-eigenvalue test indicates 1 cointegrating eqn(s) at the 0.05 level.

Denotes rejection of the hypothesis at the 0.05 level.

MacKinnonHaugMichelis (1999) p-values.

(continued on next page)

theories outlined above. The rest of the inequality measures are found to have no statistically significant influence on the
US capital stockindicating corroboration of the aforementioned Keynesian approach.
As far as the short-run influence of inequality on net investment, the effects are as would be predicted by the incomplete
markets/political outcomes and Keynesian theories. The estimates of the short-run i parameters in the VEC of Eq. (14) are

158

L.K. Stevans / Research in Economics 66 (2012) 149160

Table 2 (continued)
Inequality measure: ratio of top five percent to bottom eighty percent
Sample (adjusted): 1970 2006
Included observations: 118 after adjustments
Lags interval (in first differences): 1 to 2
Unrestricted cointegration rank test (maximum eigenvalue)
Hypothesized no. of CE(s)

None
At most 1
At most 2
At most 3
At most 4

Eigenvalue

Max-eigen statistic

0.05 critical value

Prob.

0.791162
0.541585
0.478971
0.405936
0.373482

51.68455
25.73935
21.51437
17.18536
15.43008

38.33101
32.11832
25.82321
19.38704
19.62798

0.0009
0.2454
0.1676
0.1015
0.1108

Max-eigenvalue test indicates 1 cointegrating eqn(s) at the 0.05 level.

Denotes rejection of the hypothesis at the 0.05 level.

MacKinnonHaugMichelis (1999) p-values.

Table 3
Estimation results. Capital stock cointegration equations (t statistics in parentheses).
Dependent variable: US capital stock
Inequality measures:
Explanatory variables

Gini coefficient

Upper five to lower 20

Upper five to lower 40

Upper five to lower 60

Upper five to lower 80

Intercept
Trend

0.943

0.876
0.014
(3.49)**
0.027
(0.711)
0.253
(3.80)**
0.356
(3.41)**
0.158
(3.59)**

1.410
0.012
(3.76)**
0.067
(2.54)*
0.250
(4.97)**
0.434
(5.35)**
0.159
(4.71)**

1.680
0.010
(3.26)**
0.059
(2.06)*
0.277
(5.16)**
0.459
(5.46)**
0.180
(5.04)**

1.601
0.011
(3.28)**
0.049
(1.61)
0.275
(4.79)**
0.449
(5.32)**
0.176
(4.70)**

Inequality
Price
Real GDP
User Capital Cost
*
**

0.014
(3.80)**
0.180
(1.37)
0.257
(4.37)**
0.362
(3.69)**
0.164
(4.01)**

Statistically significant at 0.05 level.

Statistically significant at 0.01 level.

Table 4
Estimation results. Net investment VEC (t statistics in parentheses).
Dependent variable: net investment (1Kt -change in capital stock)
Inequality explanatory variables

Change in Gini at Lag 1

0.004
(0.051)
0.015
(0.204)
0.011
(0.562)
0.009
(0.463)
0.004
(1.98)**
0.002
(0.173)
0.003
(1.89)*
0.001
(1.78)*
0.004
(0.184)
0.002
(0.096)

Change in Gini at Lag 2

Change in upper five to lower twenty at lag 1
Change in upper five to lower twenty at lag 2
Change in upper five to lower forty at lag 1
Change in upper five to lower forty at lag 2
Change in upper five to lower sixty at lag 1
Change in upper five to lower sixty at lag 2
Change in upper five to lower eighty at lag 1
Change in upper five to lower eighty at lag 2
*
**

Statistically significant at 0.10 level.

Statistically significant at 0.05 level.

presented in Table 4. For reasons of brevity, only the first and second lags of the net investment (1Kt ) equation are displayed.
While changes in inequality as measured by the change in the ratios of the upper five percent to the lower forty percent

L.K. Stevans / Research in Economics 66 (2012) 149160

159

and lower sixty percent have an inverse effect on net investment, in the majority of the cases changes in inequality have no
perceptible influence on net investment which is what is envisaged by the Keynesian approach.
5. Conclusion
Can equality in the United States be promoted without eliminating the genie of prosperity? It is clear that for over a
quarter of a century, the higher the income quantile, the more income continued to grow and the rich-get-richer pattern
has continued to prevail. Many have asked why prosperity is not spreading more equally, but when it came to hard policy
decisions, the response has always been that there is a trade-off between equality and growthif a country tries too hard
to redistribute income to the lower quantiles, there would be fewer entrepreneurs, less capital investment, and therefore a lower
standard of living. According to the results of this study, there is no empirical evidence over the past 30 years in the United States to
support such a contention. In three of the five inequality measures, increases (decreases) in inequality have had no influence
on net investment, the capital stock, and consequently economic growth. The three remaining inequality measures have
had an inverse effect on capital formation positing the existence of market failure in the capital markets due to credit
rationing.
5.1. Study limitations
There are some statistical/estimation issues regarding the model used for this study which could tend to temper the
results. It is well known that although the Johansen (1988) estimators have less bias than other estimators, they exhibit
more variation. However, this does not appear to be a considerable problem in this analysis, given the degree of statistical
significance of the parameter estimators in Table 4. In addition, the following issues may be problematic and arise from
Monte Carlo studies on the Johansen (1988) cointegration tests,19
1.
2.
3.
4.

the tabulated critical values based on asymptotic distributions may be inappropriate if sample sizes are small;
all tests can be misleading if too few variables are included;
insufficient lag length can lead to substantial size distortions, over-specification leads to loss of power; and
if a low-order VAR model is used, both the trace and max statistics are biased toward finding cointegration.

This study could be criticized based upon any or all of these issues, but the use of detrending methods, as was done in
this analysis, has been shown to improve the power of the Johansen tests.20
Appendix. Variable descriptions (all data downloaded from http://www.haverselect.com)
Variable
Capital stock (Kt )
Inequality measures:
Gini (GINIt )
Upper five/lower
twenty (L20t )
Upper five/lower
forty (L40t )
Upper five/lower
sixty (L60t )
Upper five/lower
eighty (L80t )
Price of capital (PKt )

Depreciation rate (t)

Interest rate (it )

Price level (Pt )
Output (Qt )

Description
Net stock of fixed assets and consumer durables:
Chained quantity index (2000 = 100)
Gini coefficient
Ratio of income share of top five percent relative to income share of lower
twenty percent
Ratio of income share of top five percent relative to income share of lower forty
percent
Ratio of income share of top five percent relative to income share of lower sixty
percent
Ratio of income share of top five percent relative to income share of lower eighty
percent
Geometric average of: private fixed investment: chained price index
(2000 = 100) and personal consumption expenditures durable goods: chained
price index (2000 = 100)
Ratio of depreciation: fixed assets and consumer durable goods: chained
quantity index (2000 = 100) to net stock of fixed assets and consumer durables:
chained quantity index (2000 = 100)
Long-term treasury composite, over 10 years (%)
Implicit price deflator: gross national product (2000 = 100)
Real gross domestic product (billions of chained 000$)

19 G.S. Maddala and In-Moo Kim, Unit Roots, Cointegration, and Structural Change, Cambridge University Press, Cambridge, UK: 2004, pp. 219220.
20 Ibid.

160

L.K. Stevans / Research in Economics 66 (2012) 149160

References
Alesina, A., Rodrik, D., 1994. Distributive politics and economic growth. Quarterly Journal of Economics 109 (2), 465490.
Barro, Robert J., 1999. Inequality, growth, and investment. NBER Working Paper 7038.
Barro, Robert J., Xavier, Sala-i-Martin, 1995. Economic Growth. McGraw-Hill, New York.
Beare, John B., 1978. Macroeconomics: Cycles, Growth, and Policy in a Monetary Economy. MacMillan, New York.
Bertola, G., 1991. Market structure and income distribution in endogenous growth models. NBER Working Paper No. 3851. Cambridge, MA. National Bureau
of Economic Research.
Deininger, K., Squire, L., 1996. A new data set measuring income inequality. World Bank Economic Review 10 (3), 565591.
Deininger, K., Squire, L., 1998. New ways of looking at old issues: inequality and growth. Journal of Development Economics 57 (2), 259287.
Elliott, G., Rothenberg, T.J., Stock, J.H., 1996. Efficient tests for an autoregressive unit root. Econometrica 64, 813836.
Enders, Walter, 2004. Applied Econometric Time Series, second ed. John Wiley and Sons, Inc., New York, NY.
Engle, Robert F., Granger, Clive W.J., 1987. Co-integration and error correction: representation, estimation, and testing. Econometrica 55 (2), 251276.
Forbes, Kristin J., 2000. A reassessment of the relationship between inequality and growth. American Economic Review 90 (4), 869887.
Johansen, Soren, 1988. Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control 12, 231254.
Kopcke, Richard W., 1993. The determinants of business investment: has capital spending been surprisingly low? New England Economic Review 331.
Kuznets, Simon, 1955. Economic growth and income inequality. American Economic Review 128.
Lee, J., Strazicich, M.C., 2003. Minimum LM unit root test with two structural breaks. Review of Economics and Statistics 63, 10821089.
Lutkepohl, H., Saikkonen, P., Trenkler, C., 2003. Comparison of tests for the cointegrating rank of a VAR process with a structural shift. Journal of Econometrics
113 (2), 201229.
MacKinnon, James G., Haug, Alfred A., Michelis, Leo, 1999. Numerical distribution functions of likelihood ratio tests for cointegration. Journal of Applied
Econometrics 14, 563577.
Maddala, G.S., Kim, In-Moo, 2004. Units Roots, Cointegration, and Structural Change. Cambridge University Press, Cambridge, UK.
Okun, A.M., 1975. Equality and Efficiency: The Big Tradeoff. The Brookings Institution, Washington, DC.
Perotti, R., 1993. Political equilibrium, income distribution and growth. Review of Economic Studies 60, 755776.
Perron, Pierre, 1989. The great crash, the oil price shock, and the unit root hypothesis. Econometrica 57 (6), 13611401.
Persson, T., Tabellini, G., 1992. Growth, distribution and politics. European Economic Review 36, 593602.
Persson, T., Tabellini, G., 1994. Is inequality harmful for growth. American Economic Review 84, 593602.
Saikkonen, P., Lutkepohl, H., 2000. Testing for the cointegration rank of a VAR process with structural shifts. Journal of Business and Economic Statistics 18
(4), 451464.
Stiglitz, J.E., 1969. The distribution of income and wealth among individuals. Econometrica 37 (3), 382397.