1

Also in the European Journal of Political Economy, Vol. 6, No. 3, 1990, pp. 245–260.

Peacock and Wiseman’s Displacement Effect: A

Reappraisal and a New Test

by

Magnus Henrekson

January 1990

Abstract: Although Peacock and Wiseman's "displacement effect" is

frequently cited as an explanation of increases in government
spending over time, the theory has so far been inadequately tested. In
this paper four versions of the hypothesis are identified, and earlier
empirical tests are reviewed and found efficient. A new test focusing
in the time-series behavior of government spending is developed and
tested on data for Sweden and the U.K. The test examines upward
displacement after World War II, the only social upheaval which, du
to data limitations, allows for a test of displacement. Strong evidence
against the hypothesis is found.

Electronic copy available at: http://ssrn.com/abstract=2174657

 2

1. Introduction

The academic literature dealing with explanations for the growth of government spending has expanded
immensely in recent years and new theories have appeared1 at a rapid rate. Despite this, the so-called
“displacement effect” hypothesis expounded by Alan T. Peacock and Jack Wiseman in their well-
known 1961 monograph The Growth of Public Expenditure in the United Kingdom remains one of the
most frequently cited explanations, not the least in public finance textbooks (e.g. Musgrave and
Musgrave (1984), Brown and Jackson (1982)). According to Peacock and Wiseman’s hypothesis
government spending tends to evolve in a steplike pattern, coinciding with social upheavals, notably
wars.

Despite these frequent references to the theory, Bird (1972, p. 463) has claimed that

the final verdict on the “displacement effect” . . . cannot yet be handed down because an
appropriate hypothesis has not yet been rigorously formulated and tested.

Although this remark was made almost two decades ago, it is no less relevant today.

The purpose of this paper is to reappraise the displacement literature in order to (1) identify what
various scholars have really meant by displacement, (2) to review and assess previous empirical tests of
the hypothesis, and (3) based on reasonable interpretations of displacement, to devise and carry out a
test of the hypothesis, a test which will attempt to avoid the methodological shortcomings of previous
tests and is as closely as possible consistent with Peacock and Wiseman’s original writings.

In section 2, the displacement literature is surveyed with the purpose of identifying how displacement
has been interpreted in the literature. In section 3 and 4 we review and critically appraise previous
empirical tests of the hypothesis. Based on this review a new test is devised and implemented in section
5. The last section contains a brief summary and conclusions.

1See e.g. the extensive survey by Larkey, Stolp and Winer (1981) or more recently Henrekson (1988).

Electronic copy available at: http://ssrn.com/abstract=2174657

 3

2. The Displacement Hypothesis

Peacock and Wiseman (1961), hereafter referred to as P-W, adopt a clearly inductive approach to
explaining the growth of government expenditure. When P-W observed that expenditures over time
appeared to outline a series of plateaus separated by peaks, and that these peaks coincided with periods
of war and preparation for war they were led to expound the “displacement effect” hypothesis.

The three basic propositions underlying the P-W analysis are that (i) governments can always find
profitable ways to expend available funds, (ii) citizens, in general, are unwilling to accept higher taxes,
and (iii) governments must be responsive to the wishes of their citizens. From these basic tenets P-W
derive the key concept of a “tolerable burden of taxation”.

It is assumed that notions about taxation remain fairly stable in peacetime. As a consequence, the
limited revenue capacity of the government in peacetime prevents major increases in expenditures.
Therefore, in settled times the desired government expenditures and the limits of taxation are likely to
diverge. During periods of social upheaval such as war this divergence is likely to be narrowed,
permanently displacing the burden of taxation upward. The end result is the attainment of a new
expenditure plateau at a higher level than before the onset of the upheaval. In times of crisis formerly
unacceptable revenue-raising methods will be tolerated, and (it is claimed) the higher tax tolerance will
persist even after the crisis subsides, thus enabling the government to implement expenditure programs
that it previously desired but could not finance. Furthermore, P-W argue that a war brings into focus
problems that were not identified before. This is called the "inspection effect".

Although the displacement hypothesis was induced from a study of British data between 1890 and
1955, P-W claim that it “give[s] us an approach for the subject that might be equally fruitful in studying
other countries or periods” (p. 25). Despite their assertion of the validity of the displacement effect
across countries and time periods, the authors hold that they are not seeking to find universal laws, but
rather “a way of looking at year to year changes in government spending” (p. 25). But, as Rosenfeld
(1973) points out, this is not what they do, since there is no attempt to study cyclical variations around
underlying expenditure trends in their study. Thus, the displacement effect is undoubtedly an attempt to
explain why the (horizontal) trend line shifts upward in discrete steps over time. Therefore, the
displacement effect is a theory about the secular behavior of government spending, comparable, for
instance, to Wagner's famous “Law of expanding state activity”. Even if P-W make the reservation that
the displacement effect should not be assumed to govern the growth of public expenditures in all
countries at all times, one can safely conclude that the effect is assumed to apply to the two world wars.
 4

P-W do not deny that other more permanent factors may also influence the growth of government
spending. The effects of three such factors are examined by P-W, namely changes in population, prices
and unemployment (the business cycle). It is found to make little difference whether government
expenditures are measured per capita or not . When the effects of price changes are removed the time
pattern of spending growth remains virtually unchanged. When P-W discuss price changes their
primary interest is not relative prices, but rather the possible effect of general inflation (or deflation) on
public revenues through fiscal drag. Hence, the more recent emphasis in the literature on the
productivity lag in public production as one of the main causes of the growth of exhaustive
expenditures (see e.g. Beck (1981)) is of no concern to P-W. This is quite consistent, since they
presume that government expenditures are determined from the financing side, rather than via the
demand for public services by the citizenry. Relative prices may be of importance in determining the
composition of aggregate public spending, especially the division between exhaustive expenditures and
transfer payments. However, P-W assume that these have little or no bearing on the level of total
expenditures, which they believe is determined from the revenue side. Finally, although P-W find
support for a short-run positive relationship between government expenditures and the rate of
unemployment, they claim that “there was no permanent change in the level of expenditures following
upon periods of unemployment” (p. 50). Consequently, we may conclude that P-W find that the pattern
of expenditure development is mainly dictated by the displacement effect.

As noted, it is quite clear that P-W posit that the “tolerable burden of taxation” is the engine that runs
the displacement effect. Unfortunately, it is not quite clear how this concept should be defined.

In the introduction to their book (p. xxiv) P-W state that tolerable burden of taxation should be
understood as tolerable tax rates, and that these tax rates translate into a certain share of government
expenditure relative to GDP, which remains broadly constant in normal times. On the other hand, there
is also evidence that at times they have in mind the absolute level of expenditure per capita. As pointed
out by Bird (1972, p. 459) it is quite surprising that in none of P-W’s charts, which are their principal
tools of analysis, do they relate government expenditures to GDP or to some other appropriate measure
of national income. This ambiguity is intensified by the following statement by P-W (p. 27, fn 22):

A rising real GNP per head brings increasing tax yields with constant tax rates, so that if
people's ideas of tolerable burdens are concerned with tax rates rather than total payments, this
provides a reason why the peacetime plateau described by public expenditures may have an
upward slope.

As we can readily see from this quotation, P-W do not specify whether the tolerable burden should be
interpreted in absolute or relative terms.
 5

From the above review of the original P-W hypothesis, we may derive the following two testable
versions of the displacement effect hypothesis:

I The strong version: real absolute government expenditure per capita evolves in a steplike pattern,
where the movement from one step to another coincides with major social disturbances, such as wars.

II The semi-strong version: government expenditure as a share of national income evolves in this same
fashion.

In several subsequent treatments of P-W's writings, e.g. Bird (1972), Musgrave (1969), Herber (1975),
and André and Delorme (1978), the displacement effect has been significantly reinterpreted. Instead of
assuming a constant government share relative to GDP (the semi-strong version), this share is seen as
rising secularly over time as a result of growing income per capita, i.e. mostly as a result of Wagner's
“Law”. If there is a displacement effect it gives an upward shift of the trend line, which may already be
upward sloping. This takes us to the third version of the displacement effect:

III The weak version: the ratio of government expenditures to GDP follows an upward sloping trend in
normal times. This trend is shifted permanently upward following a social upheaval.

A possible variation may be that government expenditure per capita in normal times follows an upward
sloping trend that is permanently shifted upward as the result of an upheaval.

At this juncture it should be noted that this notion cannot be found in P-W (1961). Although they
concede that increasing income can lead to a proportional increase in government spending (if the
burden of taxation is perceived in relative terms), there is no claim that spending could start growing
more than in proportion to national income in peacetime. And as noted above, the change in total
income is not among the permanent factors considered by P-W as potential determinants of government
spending.

Gupta (1967), a student of Peacock, was the first to set out to make a statistical test of the P-W
hypothesis. Here we encounter the first attempt at a reformulation of the original thesis. Gupta
reinterprets the displacement effect so that it also covers the possibility for a changed rate of growth of
government expenditure after a social upheaval, despite the fact that there was no such notion present
in the P-W book. The lack of this notion is evidenced by the fact that the displacement hypothesis (in
its weak version) implicates that government spending would not grow faster than GDP. In the
introduction to the second edition of their book P-W (1967) shifted the focus of the original
displacement hypothesis to one dealing with a change in the character of public expenditures. However,
 6

they did this without specifying how and in what direction this change is expected to take place, and
thus they made the theory even more difficult to test (Bird (1972)).

A close associate to P-W, Diamond (1977) reformulated the displacement hypothesis thoroughly and
interprets it as a theory of structural break, the main claim being that the usual ceteris paribus
assumption of unchanged tastes, preferences and institutions after the upheaval is denied. This
reinterpretation is endorsed by P-W (1979) in a review article of different approaches to the analysis of
government expenditure growth. Hence, P-W and their followers have thoroughly revised their original
view of the tolerable burden of taxation and its upward displacement after major disturbances as the
chief determinant of the time pattern of public expenditure growth. This leads us to the fourth and final
version of the displacement hypothesis:2

IV The amorphous version: the values of the parameters of the relevant model explaining the
development of government expenditures will change following a social upheaval.

It is unclear whether it is the public spending share of GDP or spending per capita that is supposed to
change, but all tests have used spending per capita in some form as the dependent variable. More
importantly, as we will see below this version has been tested empirically but with income as the only
regressor. Thus, in these tests versions III and IV are equal save the fact that the income coefficient
may change between time periods.

The different versions of the hypothesis will be thoroughly commented upon in the following sections
when we consider different ways of testing for the presence of a displacement effect.

3. A Review of Previous Empirical Tests

Gupta (1967) was the first attempt to subject the displacement hypothesis to empirical testing. He fitted
the following equation for different subperiods separated by social upheavals:

lnGct = a + b lnYct + ct

where Gc is real per capita public expenditure (other than war-related but including defense), Yc is real
per capita GNP and ct is assumed to be a well-behaved error term The test was carried out for five
countries and in most cases a significant change in slope as well as intercept was found between

2Strangely enough, Wiseman together with Diamond (1975, p. 414) assert that the original 1961 thesis actually
was one of structural break. However, there is no mention of this concept in the first edition of the P-W book.
 7

contiguous subperiods. To test for displacement the fitted value of Gc was calculated using both
estimated equations for the first year following the upheaval. If the value calculated by the equation for
the later period significantly exceeded the value calculated with the equation for the earlier period
Gupta concluded that there had been an upward displacement.

Gupta found significant displacement after the world wars in all cases except for Sweden after World
War II. However, this finding seems to be due to an estimation error (Nagarayan (1983)). He also
found significant displacement caused by the Great Depression in the case of the U.S. and Canada.
Unfortunately, the study is awash with methodological shortcomings making the empirical findings
highly doubtful. Several subperiods only contain as few as five observations; in many cases
observations are arbitrarily dropped due to their alleged “abnormality”; the subperiods compared are in
some cases separated by as much as 18-21 years; neither standard errors nor Durbin-Watson statistics
are reported despite the fact that the author admits a likely autocorrelation of residuals; the Great
Depression is arbitrarily tested as an upheaval in the U.S. and Canada, 3 but not in any of the other
countries, and so forth.

It is evident that Gupta was testing for an upward shift in the underlying trend and in this sense was
testing version III of the hypothesis. On the other hand, he was also testing for a change in that trend
after the upheaval and therefore the test includes elements of version IV as well.4

The test of Bonin, Finch and Waters (1969) is similar in many ways to Gupta's. They test for
displacement in the U.K. after the two world wars estimating the following equations:

Gnt = a + bYnt + cD + dDYnt + t

Gxt = a + bYnt + cD + dDYnt + t

where

Gn = per capita public expenditures net of debt, war-related and defense

spending
Gx = Gn less replacement of nonmilitary goods sacrificed
during wartime
Yn = GNP per capita less total public expenditure per capita5

3It is claimed that this increase in G during the Great Depression is due to an inspection process, but a mere
c
visual inspection of the appended data shows that Gc decreased, although more slowly than Yc.
4In fact, Michas (1975) claims that what Gupta did was not a test of the P-W hypothesis, but a test of Wagner's
Law.
 8

D = 0 in prewar periods and 1 in postwar periods

In contrast to Gupta, Bonin et al concentrate on civilian public spending. In the second variant they try
to account for possible shifts in government spending caused by the need to catch up with the backlog
of foregone peacetime spending. Because all replacement (calculated as the cumulated difference
between the extrapolated prewar trend of Gn and actual Gn during the war) is assumed to take place in
the immediate postwar years, there is a considerable difference between the two series in the early
portion of the subperiods, something which shows up in large differences in the slope shift dummies for
Gn and Gx, respectively. In general, the slope changes are positive but insignificant. However, the
intercept shift, which is what the authors interpret as displacement, is always significant. There is
upward displacement after World War I for both series and for Gn after World War II. However, for the
Gx series there is a significant downward displacement after World War II. In testing for Great
Depression displacement similarly conflicting results are obtained: strong upward displacement for Gx,
whereas there is a weak downward displacement for Gn. Hence, the results seem clearly sensitive to the
definition of the dependent variable, which is no surprise given that the number of observations used in
each subperiod in no case exceeds ten and in the pre World War I period is as low as six. It is also
likely that the error terms are autocorrelated. Alas, no diagnostics are reported. In conclusion, the
arbitrary fashion used to account for replacement, the insufficient number of observations, and the
contradictory findings make this test of (a variant of) version III of the displacement effect highly
questionable.

As previously mentioned, Diamond (1977) explicitly restates the displacement hypothesis as a theory
of structural break. He tests this empirically with a Chow test comparing two time periods separated by
a social upheaval. If this shows significant structural change in the estimated parameters, he concludes
that there has been displacement. Three measures of government spending per capita are regressed on
GNP per capita. Of nine Chow tests only five were significant at the 5% level or better. 6 More import-
antly, as emphasized by Tussing and Henning (1979) the Chow test does not distinguish between
upward and downward displacement. In no case did slope as well as intercept terms increase. In eight
cases they changed in opposite directions and in one case they both declined. From the test we cannot
infer whether the slope or the intercept change is dominant. To do that would have required a
procedure that accounted for the change over time in the variables to which the coefficients apply.
Diamond's test may be seen as a test of version IV of the hypothesis, since it includes a test of changes
in the slope term. In light of the original P-W formulation we are inclined to agree with Tussing and

5This is an inconsistent measure of income since it also deducts the non-GNP part of government spending from
GNP.
6In common with the other studies this one suffers from a lack of degrees of freedom and a strong suspicion that
there is autocorrelation in the residuals (no diagnostics are reported).
 9

Henning's conclusion that “Diamond's Chow tests are virtually worthless as tests of the Peacock-
Wiseman displacement hypothesis” (p. 481).

A further critique of Diamond's analysis is that of Watt (1979). He shows that the assumption of equal
error variance across subperiods is incorrect in most cases, making the Chow test inappropriate to test
for structural break. Ironically enough, Diamond himself questioned the validity of the results obtained
by Bonin et al on exactly those grounds.

Other tests of the displacement effect, basically using the Gupta methodology are André and Delorme
(1978) for France and Nagarayan (1979) for India. In both cases support is lent to the hypothesis.

The above reviewed studies have one important thing in common — they all in some way relate
government spending to income. This may be somewhat surprising since in the original P-W hypothesis
government spending was taken to be a theory of the development of government spending over time.7
The only test using time as an explanatory variable is that of Pryor (1968), who estimates the following
two simple time relationships on U.K. data:

lnGt = a + bT + cWW1 + dWW2 + t

lnGnt = a + bT + cWW1 + dWW2 + t

where

G = total government expenditure per capita

Gn = per capita public expenditures net of debt, war-related and defense spending
T = time
WW1 = 0, 1890-1913; = 1, 1923-37, 1950-61
WW2 = 0, 1890-1913, 1923-37; = 1, 1950-51

Pryor finds statistically significant upward displacement for G (c, d > 0) but not for Gn.

The final significant methodology that has been applied for the testing of displacement is that of
Tussing and Henning (1974). In their study a comprehensive set of explanatory variables is applied
simultaneously as potential determinants of public non-defense spending growth. The presence of a
displacement effect was tested for by including intercept dummies, one for each postwar period in a

7This fact is even admitted by Diamond (1977, p. 391), who elsewhere argues that even the original P-W
hypothesis was one of structural break.
 10

way similar to that of Pryor. These tests failed to indicate upward displacement. The dummies were
either insignificant or negatively significant.

4. A Critical Assessment of the Empirical Tests

As the above review clearly points out, the empirical evidence on the displacement effect so far is
undoubtedly conflicting. There are a number of reasons for this. In this section we will attempt to
pinpoint the crucial differences in order to arrive at a more appropriate framework for the testing of the
P-W hypothesis.

Comparability of the various studies is impaired by the choice of data series and time periods. In
general, the estimated models are weakly related to the theory, and poorly specified. While P-W and
Gupta look at total public spending less war-related expenditures, Bonin et al exclude defense spending
from this same series but add replacement expenditures. Others, on the other hand, study either total
non-defense public expenditures or simply total expenditures. Given the few degrees of freedom even
small differences in the series will result in substantially different parameter estimates. This latter
problem in particular mars the attempts to test for displacement after WW I. For instance, when Gupta
tested for WW I displacement he had five prewar observations for Germany and six ditto for the U.K..
From this we may infer that there are not sufficient data available to perform a reliable test of the P-W
hypothesis for WW I.

The choice of time periods is greatly affected by the treatment of the Great Depression by different
researchers. Gupta (for the U.S. and Canada), Bonin et al, and Diamond all allow for possible
displacement due to the Great Depression, whereas André and Delorme and Tussing and Henning do
not. But can the Great Depression be justifiably treated as a social disturbance on par with, for instance,
WW II? P-W (1961, p. 50) do not think so. In addition, insufficient degrees of freedom make it
impossible to test for displacement caused by the two world wars and the Great Depression.

O'Hagan (1980) argues that the worldwide recession of 1974-75 might qualify as a major social
disturbance as well. But he concedes that the a priori arguments for a displacement effect after the
1974-75 recession are likely to have essentially short-term implications, leading to an initial rise in the
public spending ratio due to inertia in adjusting spending plans downwards. Bird (1972) has suggested
a “ratchet effect” explanation to the observed growth in the public spending ratio in times of slow or
negative economic growth. If there is a crisis that leads to a declining GNP, government spending also
declines, although less rapidly, causing the spending share to increase. Bird does not preclude the
possibility that the ratio temporarily declines in times of unexpectedly rapid growth in GNP. In
 11

conclusion, it seems precarious to treat the Great Depression or other less severe recessions as
upheavals comparable with the two world wars. Thus, we may argue with reasonable confidence that
both the lack of data for earlier periods and the literature indicate that WW II is the only social
upheaval to date that qualifies without reservation for testing the P-W displacement hypothesis.8

As already noted, all researchers with the exception of Pryor relate some measure of government
spending to some measure of income. Diamond (1977) and P-W (1979) have criticized this procedure
on the grounds that it concentrates on the demand side, to the relative neglect of supply, thereby
vitiating the gist of the original P-W hypothesis, which states that the notion of a tolerable burden of
taxation reflects both supply and demand influences as outlined above. However, in our view, this
criticism is ill-founded, because the aggregate in absolute or per capita terms can just as well be seen as
a proxy for taxable capacity (Afxentiou (1982, p. 34)). A minor criticism is that of Henning and
Tussing (1974), who argue that because (part of) government spending is a component of GNP, there is
a problem of simultaneity. Henning and Tussing solve the problem in part by using private GNP as a
regressor. But as Hadjimatheou (1976) emphasized, there is still a simultaneous equation bias since
private GNP is an endogenous variable; the only way to completely solve the problem is to specify a
more complete income determination model.

Since the original P-W hypothesis was expounded as an explanation of the development of public
spending as a function of time, an appropriate treatment of time is crucial in an empirical test. Those
researchers who treat government spending as a function of income suppress the time dimension
completely. For instance, no one heeds the fact that the observations preceding WW I are at best at
five-year intervals and often unequally spaced. Similarly, no account is taken of the fact that time
elapses during the disturbance or of how many years that have elapsed between the first and last years,
respectively of two "adjacent" subperiods.9 Since the concept of time is vital, an appropriate test
requires the modeling of the entire time pattern of public expenditure development. Otherwise there is a
great risk that what appears to be postwar displacement may be no more than the result of the workings
of the underlying trend mechanism.

5. The Empirical Test

8Bird (1972, p. 459) rightly points out that WW II in fact was too recent at the time P-W was writing to allow any
inference of displacement after WW II. Since most researchers who have tested for displacement have used the P-
W data the criticism is equally valid to their tests.
9Good examples are Gupta's exclusion of 21 years in the case of Germany when testing for WW II displacement,
and Diamond who defines his post WW II period to begin in 1950.
 12

From the above review of P-W’s original study as well as the subsequent displacement literature, we
may conclude that the two world wars are the only social upheavals that qualify as possible causes of
upward displacement. Moreover, due to lack of data for earlier periods, the test can only be applied to
World War II.

Because the original hypothesis focuses on the development of government expenditures over time,
time is the vital component of any model for testing the hypothesis. The most straightforward way to
model the time pattern of growth is by means of Box-Jenkins (1976) techniques. In order to account for
the effect of time, the war years should be modeled instead of being simply excluded from the data set,
as we have seen in the work of other researchers. These requirements limit the number of countries for
which the hypothesis can be adequately tested. First, it requires annual data for the relevant variables
from the beginning of the 1920s. Also, in order to qualify the country must have remained independent
during World War II as well as geographically unchanged afterwards in order to obtain a homogeneous
data series. The only two countries that we have found to qualify in all three respects are Sweden and
the United Kingdom.10

An alternative to the ARIMA-technique would be to formulate a structural model of government

spending that explicitly tests for postwar displacement. However, this approach is problematic in
several respects. First there is little consensus about what the model should look like. 11 Second, even if
one could manage to derive an acceptable model from theory, data on the necessary variables are rarely
available for periods before 1950, if at all. Furthermore, an ARIMA-analysis seems to be most
consistent with P-W (1961, p. 24), who assert that permanent influences are unlikely to give rise to
general and testable hypotheses about government spending.

Recently, following Nelson and Plosser (1982), many empirical studies have concluded that many
macroeconomic series are difference stationary, i.e. have a unit root, rather than being trend
stationary.12 However, as shown by Rappoport and Reichlin (1989) and Perron (1988), most
macroeconomic time series can be appropriately modeled as cyclical fluctuations around a
deterministic trend if infrequent changes in the intercept and/or trend is allowed. Indeed, the basis of
Perron’s methodology is to postulate that these changes occur at a known date. This is exactly parallel
to our own case, where it is hypothesized that there is a shift in the level of public spending after World
War II.

10For the U.S. for instance annual data of aggregate government expenditures is not available until 1929.
11See e.g. the various country studies in Lybeck and Henrekson (1988).
12For a good overview of this research, see Stock and Watson (1988).
 13

In the first stage of the test we identify an ARMA(p, q) model around a deterministic trend13 for the log
of real total government expenditures per capita (GR/N) and total government expenditures as a share of

GDP (G/Y) for the postwar period,14 1947-87. This simple exercise gives us a good representation of
the behavior of government expenditures in untroubled times, which may be checked for consistency
with the different versions of the displacement hypothesis. The results of these estimations are given in
Table 1.

To facilitate interpretation and to define the notation of Table 1 it may be noted that the following
models have been identified:

yt =  + T + yt- + yt- +t

where yt represents the relevant dependent variable, T a time trend and t a well-behaved error term.
For Sweden 2 = 0.

For Sweden, an ARMA(1, 0) model with trend adequately represents the behavior of GR/N as well as

G/Y. In the case of the U.K. an ARMA(2, 0) with trend is identified. In all four cases the residual
autocorrelation function has no spikes at key lags and the Ljung-Box Q-statistic is not significant.

Table 1. Identified ARMA Model with Trend for Sweden and the U.K., 1947-87

Independent Param- Sweden U.K.

variables eters GR/N G/Y GR/N G/Y
Constant  6.37** –.20** 8.77** .12**
(.039) (.084) (.19) (.077)
T  .039* .012** .039** .0051**
(.021) (.0014) (.0037) (.0014)
yt-  .93** .86** 1.24** 1.21**
(.077) .078 (.13) (.14)
yt-  –.39** –.35**
(.09) (.086)
Ljung-Box Q 25.7 13.0 21.1 22.6

13Hence, we assume that the dependent variable, y can be written as y =  + T + y ts where y s is stationary and
t, t  t
 is the constant annual growth of the deterministic trend.
14For a detailed definition of the variables and data sources, see the section Data Sources and Definitions on p. 18
below.
 14

d.f. 38 38 37 37
R2 .99 .99 .99 .97
SEE .034 .015 .037 .013

Note: Standard errors in parentheses, * and ** denote significance at the 5 and 1% levels, respectively.

The trend variable is positive and significant in all cases, leading to an immediate rejection of both
version I and II of the P-W hypothesis.

In order to test whether there has been upward displacement in the trend of GR/N or G/Y after World

War II (version III), we need to model the war years and test for an upward shift after the war. To do
this we apply Box-Tiao (1975) intervention analysis to the entire series. In order to test for a
displacement effect, the following dummy variable is used:

PW = 1, 1946-87; else = 0

In modeling the effect of the war on aggregate public spending, we must take into account that the two
countries’ involvement in the war was very different. On the assumption that the threat of a German
invasion into Sweden was the most important determinant of the Swedish war effort, we can safely say
that this threat decreased significantly towards the end of the war (starting in 1942). This fact should be
reflected in the coding scheme used to capture the effect of the war in the intervention analysis. 15
Account is also taken of the fact that 1939 and 1945 were only partly war years. For the war variable,
W, the following coding scheme has been used:

– 38 39 40 41 42 43 44 45 46 –
Sweden 0 .5 1 1 .5 .25 0 0 0
U.K. 0 .5 1 1 1 1 1 .5 0

Because it can be hypothesized that the effect of the war gained in momentum gradually, we use the
following first order transfer function:

15An alternative method to account for the fact that Sweden began to demobilize once the threat of a German
invasion disappeared is to code W identically for both countries and include an additional intervention variable. In
an earlier version of this paper the following variable was included in the regression for Sweden: D = 1, 1942-45;
else = 0. A preliminary analysis led us to use the following transfer function for D: f(D) = ( + L)Dt . These
results were almost identical to the ones presented here. Because of its greater simplicity we prefer to use different
coding schemes for W.
 15


f(W) = W
1 – L t

where L is the lag operator and is constrained to the interval –1 < < +1.

The result of the estimations for the entire period 1922-87 is shown in Table 2.

The estimation results are on the whole satisfactory. A close inspection of the autocorrelation and
partial autocorrelation functions indicates clearly that the residuals are white noise. This is also
confirmed by the low Q-statistics.

In no case is the P-W hypothesis of positive displacement after WW II supported by the regression
results. Rather, all estimated coefficients are negative. The point estimates indicate, that net of trend,
WW II led to a drop in the share of government spending to GDP. The estimate is highly significant for
the U.K. but not for Sweden. The negative effect for government spending per capita is significant at
the 10% level in both countries.
 16

Table 2. ARMAX estimations for the Whole Period, 1922-87

Independent Para- Sweden U.K.

variables meters GR/N G/Y GR/N G/Y
Constant  5.43** –.14 8.94** .16**
(.11) (.16) (.10) (.019)
T  .055** .012** .037** .0052**
(.0024) (.0025) (.0021) (.00054)
yt-1  .86** .93** 1.10** .98**
(.070) (.038) (.12) (.13)
yt-2  –.25* –.22(*)
(.12) (.12)
W  .33** .080** .57** .15**
(.053) (.015) (.050) (.018)
 .19 .47** .42** .54**
(.14) (.13) (.069) (.055)
PW  –.077(*) –.023 –.076(*) –.046**
(.047) (.017) (.046) (.018)
Ljung-Box Q 18.4 18.7 23.2 27.2
d.f. 60 60 59 59
R2 .99 .99 .99 .98
SEE .046 .015 .040 .015

Note: Standard errors in parentheses, (*), * and ** denote significance at the 10,
5 and 1% levels, respectively.

This means that our model indicates that World War II led to a permanent downward shift in G/Y by
4.6 percentage points for the U.K and 2.3 percentage points for Sweden (not significant). We consider
this as strong evidence against version III of the displacement hypothesis.

As regards the fourth version of the hypothesis it should first be noted that this version is so weakly
formulated that it is very difficult to test. A comprehensive critique of this version is presented by
Tussing and Henning (1979, 1986). In particular, they point out that Diamond’s (1977) test of
structural break is illegitimate, since the model he uses is grossly misspecified. Moreover, we have
already noted that the construction of a reasonable structural model explaining government spending is
still unavailable, making this version untestable. In the context of our own framework, we may note that
the ARMA model with trend identified for the postwar period also constitutes an adequate
 17

representation of the entire period once the effect of World War II has been modeled appropriately.
However, this finding does not allow one to infer anything about structural stability. That would once
again require the estimation of the original structural model.

Hence, the test developed here does not in any way give empirical support to any of the possible
testable versions of the displacement hypothesis.

6. Conclusion

Although Peacock and Wiseman’s displacement effect remains one of the most frequently cited
explanations for the growth of government spending, this hypothesis has so far been inadequately
tested.

In a survey of the displacement literature we have shown that one can identify at least four distinct
versions of the theory. Earlier empirical tests of displacement suffer from several methodological
shortcomings and comparability between the studies is impaired by different choices of time periods
and data series. Moreover, in all studies except that of Pryor (1968) the time dimension is completely
suppressed, despite the fact that the P-W hypothesis purports to explain the development of public
spending over time.

In this paper a test is devised that directly focuses on the time-series behavior of government
expenditure. This new test is restricted to post-WW II displacement, since this is the only upheaval for
which the necessary data is available. Due to the fact that government spending during the war years
has been explicitly modeled in these countries and the fact that data for a “long” period is required,
Sweden and the U.K. were the only countries found to qualify for the test.

In the first stage an ARMA(p, q) model with trend was fitted to the postwar data to obtain a suitable
time-series representation of the spending series. The interpretation of this representation allows us to
reject the first two versions of the displacement hypothesis. In order to test version III, i.e. whether
there has been upward displacement in government expenditure due to WW II as a share of GDP or per
capita, the war years are explicitly modeled and the postwar shift is tested for using Box-Tiao
intervention analysis applied to the period 1922-87.

The results clearly disprove the hypothesized upward displacement. Instead the model indicates that
WW II led to a permanent downward shift in government spending net of trend on the order of 4.6
percent of GDP for the U.K. The (insignificant) point estimate for Sweden indicates a downward shift
 18

of 2.3 percent. The effect of the war is remarkably similar in the two countries despite the considerable
difference in the nature of their war involvement.

Data Sources and Definitions

G includes all types of public spending at all levels of government. In the GR/N measure G is deflated

by the implicit deflator for private consumption. This is the correct price deflator, since any reasonable
interpretation of the “tolerable burden of taxation”, the chief engine of the P-W hypothesis, must be in
terms of private goods. In G/Y both government spending and GDP are in current prices. The data
come from the following sources:

Sweden: Data on total government expenditure for the period 1922-49 are from Höök (1962) and
Krantz (1987). GDP and price data for 1922-49 are from Krantz and Nilsson (1975). For 1950-87 all
expenditure and price data are from the Swedish National Accounts published by The National Central
Bureau of Statistics. Population data are from Statistiska Centralbyrån (1969) and Statistical Abstract
of Sweden.

United Kingdom: All data for the period 1922-60 are from Feinstein (1972). For 1961-87 the data are
taken from various issues of OECD National Accounts and OECD Labour Force Statistics.

References

Afxentiou, P.C. (1982), “Economic Development and the Public Sector: An Evaluation”, Atlantic
Economic Journal 10(December), pp. 32-38.

André, C. and R. Delorme (1978), “The Long-Run Growth of Public Expenditure in France”, Public
Finance 33(1-2), pp. 42-66.

Beck, M. (1981), Government Spending. Trends and Issues, New York: Praeger.

Bird, R.M. (1972), “The 'Displacement Effect: A Critical Note”, Finanzarchiv 30, pp. 454-463.

Bonin, J.M, Finch, B.W. and J.B. Waters (1969), “Alternative Tests of the ‘Displacement Effect’
Hypothesis”, Public Finance 24(3), pp. 441-456.

Box, G.E.P. and G.M. Jenkins (1976), Time Series Analysis: Forecasting and Control, 2nd revised ed.,
San Francisco: Holden-Day.

Box, G.E.P. and G.C. Tiao (1975), “Intervention Analysis with Applications to Economic and
Environmental Problems”, Journal of the American Statistical Association 70, pp.70-92.

Brown, C.V. and P.M. Jackson (1982), Public Sector Economics, Oxford: Martin Robertson.
 19

Diamond, J. (1977), “Econometric Testing of the ‘Displacement Effect’: A Reconsideration”,

Finanzarchiv 35(3), pp. 387-404.

Feinstein, C.H. (1972), National Income, Expenditure and Output of the United Kingdom 1855-1965,
Cambridge: Cambridge University Press.

Gupta, S.P. (1967), “Public Expenditure and Economic Growth: A Time-Series Analysis”, Public
Finance 22(4), pp. 423-461.

Hadjimatheou, G. (1976), “On the Empirical Evidence on Government Expenditure Development”,

Public Finance 31(1), pp. 144-149.

Henning, J.A. and A.D. Tussing (1974), “Income Elasticity of the Demand for Public Expenditures in
the United States”, Public Finance 29(3-4), pp. 325-341.

Henrekson, M. (1988), “Swedish Government Growth: A Disequilibrium Analysis”, in J.A. Lybeck and
M. Henrekson, eds., Explaining the Growth of Government, Amsterdam: North-Holland.

Herber, B.P. (1975), Modern Public Finance, 3rd ed., Homewood, Ill.: Richard D. Irwin.

Höök, O. (1962) Den offentliga sektorns expansion, Stockholm: Almqvist & Wiksell.

Krantz, O. (1987), Historiska nationalräkenskaper för Sverige: Offentlig verksamhet 1800-1980,

Lund: Studentlitteratur.

Krantz, O. and C.A. Nilsson (1975), Swedish National Product 1861-1970. New Aspects on Method
and Measurement, Lund: CWK Gleerup.

Larkey, P.D., Stolp, C. and M. Winer (1981), “Theorizing About the Growth of Government: A
Research Assessment”, Journal of Public Policy 1(2), pp. 157-220.

Lybeck, J.A. and M. Henrekson, eds. (1988), Explaining the Growth of Government, Amsterdam:
North-Holland.

Michas, N.A. (1975), “Wagner's Law of Public Expenditures: What is the Appropriate Measurement
for a Valid Test?”, Public Finance 30(1), pp. 77-84.

Musgrave, R.A. (1969), Fiscal Systems, New Haven and London: Yale University Press.

Musgrave, R.A. and P.B. Musgrave (1984), Public Finance in Theory and Practice, New York:
McGraw-Hill.

Nagarayan, P. (1979), “Econometric Testing of the ‘Displacement Effect’ Associated with a Non-
Global Disturbance in India”, Public Finance 34(1), pp. 101-113.

Nagarayan, P. (1983), “‘Displacement Effect’ in Government Spending in Sweden: A Reexamination”,

Public Finance 38(1), pp. 156-162.

Nelson, C.R. and C.I. Plosser (1982), “Trends and Random Walks in Macroeconomic Time Series”,
Journal of Monetary Economics 10(2), pp. 139-162.

O'Hagan, J.W. (1980), “Demonstration, Income and Displacements Effects as Determinants of Public
Sector Expenditure Shares in the Republic of Ireland”, Public Finance 35(3), pp. 425-435.

Peacock, A.T. and J. Wiseman (1961), The Growth of Public Expenditure in the United Kingdom,
Princeton: Princeton University Press.

Peacock, A.T. and J. Wiseman (1979), “Approaches to the Analysis of Government Expenditure
Growth”, Public Finance Quarterly 7(1), pp. 3-23.
 20

Perron, P. (1988), “The Great Crash, The Oil Price Shock and the Unit Root Hypothesis”, Econometric
Research Program, Princeton University, Research Memorandum No 338.

Pryor, F.L. (1968), Public Expenditures in Communist and Capitalist Nations, Homewood, Ill.:
Richard D. Irwin.

Rappoport, P. and L. Reichlin (1989), “Segmented Trends and Non-Stationary Time Series”, Economic
Journal 99(Supplement), pp. 168-177.

Rosenfeld, B.D. (1973), "The Displacement Effect in the Growth of Canadian Government
Expenditures", Public Finance, 28(3-4), pp. 301-314.

Statistiska Centralbyrån (1969), Historisk statistik för Sverige. Del 1: Befolkning 1720-1967,
Stockholm: Allmänna Förlaget.

Stock, J.H. and M.W. Watson (1988), “Variable Trends in Economic Time Series”, Journal of
Economic Perspectives 2(3), pp 147-174.

Tussing, A.D. and J.A. Henning (1974), “Long-Run Growth of Nondefense Government Expenditures
in the United States”, Public Finance Quarterly 2(2), pp. 202-222.

Tussing, A.D. and J.A. Henning (1979), “Econometric Testing of the Displacement Effect: A
Comment”, Finanzarchiv 37, pp. 476-484.

Tussing, A.D. and J.A. Henning (1986), “Measuring the Effect of Structural Change on Long-Term
Public Expenditure Growth: The United States, 1929-1981”, Syracuse University, mimeo.

Watt, P.A. (1979), “Econometric Testing of the Displacement Effect: A Note”, Finanzarchiv 36, pp.
445-448.

Wiseman, J. and J. Diamond (1975), “Comment on ‘Long-Run Growth of Nondefense Government

Expenditures in the United States’”, Public Finance Quarterly 3(4), pp. 411-414.