Household saving behavior and the rate of unemployment in the U.S.

A Structural VAR approach

Shuxian Zhu

Dissertation Supervisor: Wei Cui

Dissertation submitted in part-fulfilment of the MSc in Economics,

University College London

September 2014
Declaration

I, Shuxian ZHU, hereby declare that the work presented in this dissertation is my own
original work. Where information has been derived from other sources, I confirm that
this has been clearly and fully identified and acknowledged. No part of this
dissertation contains material previously submitted to the examiners of this or any

other university, or any material previously submitted for any other assessment.

Name: Shuxian ZHU

Date: 11/09/2014

Classification
This piece of research is primarily (please indicate as appropriate):
an analytical survey of empirical literature

Dissertation Library
I give permission for a pdf electronic copy of this dissertation to be made available to
future students by way of the Dissertation Library, should the Department of

Economics choose to use it in this way.

Yes
Abstract

The unemployment and savings are two main factors related to the business cycle,
which many Economists are concerned about. The U.S. has experienced significant
shifts in both the unemployment rate and household savings from 1980, especially

during the financial and debt crisis around 2008. Until now, many empirical studies
have focused on the relationship between the unemployment rate and the output or
inflation. However, in this paper, I aim to provide information on the transmission
mechanism between the rate of unemployment and household saving behavior.
According to the permanent income hypothesis and life-cycle hypothesis, it is implied

that households may save to smooth their consumption when facing uncertainties of
future income. This dissertation develops a structural VAR model to examine whether
the unemployment rate and the household saving behavior have effects on each other.
I used the U.S. quarterly data from 1980 to 2013 to seek the response of the four
relevant variables (inflation rate, inflationary expectations, personal saving rate and

unemployment rate) to simulate shocks imposed at the macroeconomic level. My

results indicate that the unemployment rate has a positive impact on the personal
saving rate and in turn, the personal saving rate affects the unemployment rate over
longer horizons, although initially there is no immediate link between shocks of both
the unemployment and the personal savings. After various explanations for this

result, implications and further research on this topic are discussed.

Acknowledgements
This dissertation is part of my MSc in Economics at the University College of London
and therefore I would like to thank Wei Cui for his guidance, support and supervision.

Moreover, I would also like to thank Boonpin Wong for his proofreading.
Table of Contents
Declaration ....................................................................................................................... 2
Abstract............................................................................................................................ 3
List of figures .................................................................................................................... 5
List of tables ..................................................................................................................... 6
Section 1 Introduction ........................................................................................................ 7
Section 2 Background and Literature Review ........................................................................ 9
2.1 Background .......................................................................................................... 9
2.2 Literature Review ................................................................................................ 10
Section 3 Model Specification and Data.............................................................................. 15
3.1 Econometric Model ............................................................................................. 15
3.2 Data and Variables ............................................................................................... 17
3.3 Empirical Strategy ............................................................................................... 20
Section 4 Econometric Estimations .................................................................................... 22
4.1 Unit Root Tests.................................................................................................... 22
4.2. Testing the Exogeneity on the Output gap .............................................................. 24
Section 5 Empirical Results............................................................................................... 25
5.1 SVAR Estimation ................................................................................................ 25
5.2 Granger Causality Test ......................................................................................... 29
5.3 Structural VAR Impulse-Response Analys is ............................................................ 29
Section 6 Conclusion and Further Discussion ...................................................................... 34
6.1 Conclusion and Implications ................................................................................. 34
6.2 Limitations and Further Research .......................................................................... 35
Bibliography ................................................................................................................... 37
List of figures

Figure 1: The personal saving rate and the unemployment rate in the U.S., quarterly
data, 1980-2013.
Figure 2: The actual output and the potential output estimated by HP filter,

1980-2013.
Figure 3: The measure of the output gap and the natural rate of output, 1980-2013.
Figure 4: Graphs of the unemployment rate, personal saving rate, inflation rate and
inflationary expectation, quarterly data,1980-2013.
Figure 5: Accumulated impulse response functions: response to the household saving

rate to one standard deviation innovations for 48 quarters.

Figure 6: Accumulated impulse response functions: response to the unemployment
rate to one standard deviation innovations for 48 quarters.
List of tables

Table 1: Unit root test statistics for other four variables, quarterly data, 1980-2013.
Table 2: Summary statistics of VAR model.
Table 3: Summary statistics of structural VAR model.

Table 4: Granger Causality test of the unemployment rate and the saving rate,
1980-2013.
Section 1 Introduction

Whether it was the Great Depression or it was in the stable developed Economy,
topics on the unemployment have always been one of the hottest ones that most
economists are interested in. The unemployment rate, measured as the prevalence of

the unemployment, is calculated using the unemployed workers divided by total labor
force. During the last three decades, the relation between the unemployment rate and
consumption is crucial among many theoretical and policy debates. It is known that
the employment status has a significant influence on the disposable income and thus
the household consumption. From the theory of precautionary saving, the risk of

uncertainty on income depresses household consumption and stimulates the

accumulation of savings as a type of insurance. Thus, an increase in the income
uncertainty is expected to have effects on household saving behavior.
In terms of household savings, there is a theoretical consensus that there are two or
three motives for household savings: precautionary savings which is due to the

unstable income source, life-cycle savings which can be used for purchasing a house
or retirement and the last possible one may be bequests (Mauro and Rob, 2013).
Using observed life cycle income variation, precautionary savings make up
approximately 30%-40% of all savings argued by many authors, including Carroll and
Samwick (1998) and Ventura and Eisenhauer (2006).

The precautionary savings usually occur in response to the uncertainty of future

income. Both the permanent income hypothesis and the life-cycle hypothesis imply
that households may save to smooth their consumption when facing uncertainties and
massive economic shocks during certain periods. For individuals or households, a key
determinant of uncertainty pertaining future disposable income and a potential

important factor affecting the household saving behavior is the risk of lost wage
resulting from the unemployment.
It is expected that the unemployment would affect personal savings since those
individuals who are out of work try to maintain consumption as much as feasible by
increasing personal savings during normal periods but decreasing savings during
periods when adverse economic shocks happened on income. Most economists are in
favor of the argument, which states that the unemployment reduces individual

consumption and increases precautionary savings, thus increasing household savings

in the short run. Other economists believe that the unemployment does not cause any
significant effects on the saving decisions over consumers’ lifetimes. In any case, the
effect of the unemployment on personal savings is unsolved and needed to justify
further empirical investigations.

The personal saving rate, the ratio of personal saving to personal disposable income,
can be an efficient proxy for household saving behavior. This paper examines whether
there is a link between the rate of the unemployment and the personal saving rate for
the U.S. with data from the last three decades using structural vector autoregression
(SVAR) models. SVAR model has been used extensively to explain the impact of the

unemployment on the output or investment, e.g. Groenewold and Hagger (2000),

Ramudo, Grela and Garcí
a (2007), but has not been applied to the impact on the
household saving behavior. One special aspect of this study is the addition of data on
the output gap, following the work of Basistha and Nelson (2006), who have a
research on the link of the unemployment rate and the output gap.

The layout of this paper is as follows. The next section includes background of the
trend in the unemployment and household savings in the last three decades in the U.S.
and the review of the literature related to both the unemployment and the savings.
Section 3 illustrates SVAR model specification and relevant variables as well as the
data source for the years of 1983 to 2013. After that, Section 4 describes multiple

econometric tests before the empirical estimation. In the next stage, Section 5 reveals
the results found using structural VAR approach and holds a short discussion based on
the results. Finally, section 6 sums up the conclusion of the paper and some further
discussion related will be given.
Section 2 Background and Literature Review

2.1 Background

From 1980 up to now, the U.S. has witnessed great development of economic growth
with a transformation from an economy mainly intervened by the government to an
economy oriented by the market. In 1986, the U.S. finally completed the interest rate
liberalization with the cancelation about the ceiling of the deposit rate on NOW

accounts. After that, the reforms on different welfare systems and the labor market
brought great uncertainty on both disposable income and employment status to
households. On one hand, unemployment became one of the most important social
and economic topics in the early-1990s and during great recession near 2008. From
the Federal Reserve Economic Data (FRED), the aggregate unemployment rate

fluctuated with an increase from 7.0 per cent in 1986 to 8.2 per cent in 2012. On the
other hand, since the economic reforms, although the U.S. has experienced rapid
growth in average household disposable income, at the same time it has also added
fluctuations as the economy undergoes massive structural shifts during last three
decades. During the period from 1986 to 2012, per capita real disposable personal

income (of chained 2009 dollars) grew up from 23632 dollar to 36815 dollar from
FRED, around 1.5 times increase and an average annual growth rate of about 1.67 per
cent. This comparison of available data implies that to some extent, the uncertainty of
disposable income is related to the unemployment. This result correspond s to the
argument discussed by Malley and Moutos (1996), which makes the assumption that

the aggregate unemployment rate can be viewed as a valuable signal for the aggregate
income uncertainty.
Hence, the link between the income uncertainty and household saving behavior can be
thought as the relation of the unemployment rate and personal saving rate. In
particular, Figure 1 shows that the trend of the household saving rate declines during

this period with the start point 9.7 in 1980 and the end point 6.7 in 2013. Additionally,
the mean of the unemployment rate was probably procyclical to the personal saving
rate from 1980 to the end of 2007. When there was Great Depression in the beginning
of 2008, the unemployment rates increased dramatically until the middle of 2010,

suggesting more uncertainty on the variance of aggregate income. From this figure, it
is found that over the long-run horizon, the fluctuations of the unemployment rate and
the personal saving rate co- moved. Logically, it is expected that the rate of
unemployment and the private saving rate is related to each other.
Figure 1: The personal saving rate and the unemployment rate in the U.S., quarterly

data, 1980-2013.

14.0
12.0
10.0
8.0
6.0
Personal saving rates
4.0
Unemployment rates
2.0
0.0
1984-05-01
1986-07-01

1993-01-01

2001-09-01

2008-03-01
2010-05-01
1980-01-01
1982-03-01

1988-09-01
1990-11-01

1995-03-01
1997-05-01
1999-07-01

2003-11-01
2006-01-01

2012-07-01

Source: Fred St. Louis

2.2 Literature Review

There are several theories which focus on the explanation of the relation between the
unemployment rate and the household saving behavior. As a starting point, we will
take our analysis in the permanent income hypothesis (PIH). Friedman (1957)
acknowledged the importance of the demand for precautionary savings stemmed from
the increasing variance of personal disposable income in the future by making the

intention of saving out of the permanent income to be a positive function of income

uncertainty. This is asserted as the permanent income hypothesis (PIH), which states
that changes in the permanent income, rather than changes in the temporary income,
are what drive the corresponding changes in a consumers’ consumption. Knudsen and
Paarnes in 1975 and Yotopoulos and Nugent in 1976 subscribed this hypothesis,
arguing that the effect of income uncertainty was to reduce the marginal propensity to

consume out of the permanent income. Hall (1978) conducted a statistical test of PIH
and found some evidence supporting this hypothesis. In the test of PIH, the permanent
consumption is addressed as a stable function of the permanent income,
Cp t = c (w, u) Yp t (1)
where the factor of proportionality c is approximately zero, depending on the ratio of

human to non- human capital w and the consumer’s propensity for

consumption-saving pattern u. Since the permanent income and the permanent
consumption are unobserved, correspondence can only be established between them,
Yt = YP t + YT t (2)
Ct = CP t + CT t (3)

where YT and CT are transitory income and transitory consumption, which are random
components of measurement in income and consumption. It is assumed that there is
no correlation between the permanent income, the transitory income and the
permanent consumption, the transitory consumption,
corr (YP , YT ) = corr (CP , CT ) = corr(YT , CT ) = 0 (4)

This non-correlated relation between the transitory income and the transitory
consumption suggests that the transitory income is used for private savings. As a
result, the marginal taste to consume out of the transitory income is zero. Hence, this
PIH saving function can be written as,
St = Yt - Ct (5)

Substituting (1), (2) and (3) into (5), we get,

St = (1-c) YP t + YT t - CT t (6)
where, S represents personal savings to form capital or spent on durable goods, (YT -
CT ) is the transitory saving, and (1-c) is the coefficient to save from the permanent
income. Equation (6) implies that the marginal rate of saving from the transitory

income is one while the marginal rate of saving from the permanent income is close to
zero because the factor c is approximately equal to 1. Beyond this point, the transitory
income is an important determinant of household savings. In the PIH, variance s of the
permanent income are predictable but the income uncertainty, the average deviation of
future income from a specific level, results from variations in the transitory income.

When the transitory income is extreme negative, savings will decline not only because
there is a positive propensity to save out of the transitory income but also because the
uncertainty on income is high. When the transitory income is e xtremely positive,
savings will go up by the changes in transitory income but decrease by the uncertainty
factor. Thereby, in short run, a fall in income has a much greater effect on personal

savings.
Similar to the permanent income hypothesis (PIH), the life-cycle hypothesis presents
that households both plan their savings and consumption behaviors over their entire
lifetimes and prefer to balance their consumption in the best possible manner in the
long term. This life-cycle consumption models were analysed by Modigliani and

Brumberg in 1954 and Modigliani in 2000. In these models, individuals take their
intertemporal decisions on consumption and savings not only as a function of current
income, but they also take into account the expected stream of income (from labor)
and their expected financial wealth.
Next, the link between the unemployment and the income uncertainty is derived.

Malley and Moutos (1996) pointed out that the aggregate unemployment rate, an
useful measurement of aggregate income uncertainty was attributed as an independent
role, affecting the transitory income and thus, the household saving decisions. Also,
they concluded that changes in aggregate uncertainty on income was a powerful
signal of changes in households’ consumption, if the uncertainty resulted from the

employment status fluctuated more on some individuals than others. As for those
individuals who are risk-averse, they are expected to add their personal savings more
than proportionally to preserve themselves against the rise in the size of income
variations. As a consequence, the aggregate impact on households’ expenditure and
savings will be greater if income fluctuates more often on some households.

From the view of Macroeconomics, the effects of the unemployment rate on

consumption and savings have been early introduced by Keynes in 1936 at first.
Keynes defined saving as the excess of income over expenditure on consumption and
discussed saving behavior as the desire ‘to build up a reserve against unforeseen
contingencies’. Under the pure theory of contract, it is argued by Rosen (1983) that

the economy could be treated as the complete insurance for both the employed and
unemployed workers, ensuring full consumption smooth and thereby the ident ical
individual savings. However, due to adverse selection and moral hazard, the
unemployment insurance could not be complete and income would not be identical.
Dynarski and Sheifrin (1987) estimated the changes in the consumption following the

period of unemployment. They found that the effect of the unemployment on the
consumption depended on different types of workers, with white-collar workers
reacting more to unemployment than the blue-collar workers. Specifically, for some
consumers such as the tenured professor, an increase in the rate of the unemployment
over the world does not have large impact on their perception of their future income

uncertainty. However, for other consumers such as construction workers, a rise in

aggregate unemployment rate does signify individuals’ income uncertainty. Thus, the
household savings followed the same path as the income uncertainty in response to
the employment status.
When it comes to the recent empirical works on the topic of the unemployment rate

and household consumption and savings, Carroll (1992) gave answers to survey
questions on savings, changes in the aggregate unemployment and changes in the
household income. The theory ‘Buffer stock theory of savings’, which was proposed
by Deaton in 1992, explained that the main reason for households holding assets was
to shield their consumption against the uncertainty of future income. In other

empirical studies, the strong correlation between the wage flexibility and the
unemployment was reviewed by Card (1995). In 2006, Moore and Pentecost had the
results about the labor market disequilibria using a structural VAR model for eight
European countries. They then concluded that there was a relation between the real
wage flexibility and unemployment conditions, finding that changes in the real wage

tend to experience low rate of unemployment.

Finally, the most recent study made by Ramudo, Grela and Garcí
a (2007), using
structural vector autorregressions (SVAR), reached an conclusion that permanent
changes in the consumption-saving pattern not only have a permanent influence on
the investment but also affect the rate of the unemployment subsequently in Spain. In

this paper, I will follow the same methodology of structural VAR model and focus on
the relations of household saving behavior and the rate of unemployment. Treating the
unemployment rate as an independent role to affect the private saving behavior, I also
include the variable, the output gap. In addition, to get explicit explanation of this
expected correlation, the structural VAR model is followed econometrically by both

multivariate Granger-causality tests and impulse response analysis.

Section 3 Model Specification and Data

3.1 Econometric Model

This article uses structural vector autoregression (SVAR) model to characterize the
mutual effect of the unemployment rate and household savings. The general
specification of the model is given by:

A0 Xt = A(L)Xt-1 + єt (7)
Xt is an n*1 vector of different variables. A0 is an n*n parameter matrix with ones on
its main diagonal and with off diagonal elements capturing dynamic contemporaneous
relationships between all other endogenous variables. A(L) is a polynomial matrix,
written as the following formula,

A(L) = A1 + A2 L + A3 L2 + … + AP Lp-1 (8)

here, L is the lag operator defined as LiXt =Xt-i.
єt is a n*1 vector of structural shocks that has zero mean with covariance- variance
matrix Σє, which assumes all shocks are time- invariant. In this model, structural
shocks are key elements, which are the inputs of this dynamic system generating from

the n-dimensional time vector Xt . Since the structural shocks are not observed, some
assumptions should be made to get the identification. One of the most used
assumptions is that structural shocks are orthogonal, which means they are mutually
uncorrelated.
The reduced form of (7) is given by standard vector autoregress (VAR) model:

Xt = B(L) Xt-1 + ut (9)

where B(L) = A0 -1 A(L) and ut =A0 -1 єt .
Notice that equation (7) is a system of simultaneous equation. Restrictions are needed
for (7) to be identified to get a proper economic explanation. When the system is
identified, it is possible to recover the parameters of structural equation (7) from the

estimation of reduced equation (9). In order to get identification, it is common to

impose restrictions on the coefficients of matrix A0 such that it becomes triangular.
This process, which was first proposed by Sims in 1980, leads to a recursive system.
Also, this process indicates that imposing certain exogenous restrictions on the
variables in Xt needs theoretical justification. To have the impulse-response analysis,

identification is important since it makes it possible to properly simulate the effects of

structural shocks єt on the dynamics of the system, which are different from ut in
equation (9). When the system is identified, the elements in the error vector ut are
actually the linear combinations of elements in the structural shocks in є t . According
to Enders (1995) and Lütkepohl and Kräzig (2002), by imposing n*(n - l)/2

restrictions on matrix A0 and setting matrix Σє diagonal, equation (7) can be just
identified.
To be more specific, following the same assumption proposed by Cermeño, Roth and
Villagómez (2008), it is suggested that B(L) =0 without loss of generality. Hence, the
structure of the model in the paper can be revised as the following,

ut O = a12 ut I + a13 ut E + a14 ut U + a15 ut S + b11 єt O

ut I = a21 ut O + a23 ut E + a24 ut U + a25 ut S + b22 єt I
ut E = a31 ut O+ a32 ut I + a34 ut U + a35 ut S + b33 єt E (10)
ut U = a41 ut O + a42 ut I + a43 ut E + a45 ut S + b44 єt U
ut S = a51 ut O + a52 ut I + a53 ut E + a54 ut U + b55 єt S

where O, I, E, U, S are the output gap, the inflation rate, the inflation expectation, the
unemployment rate and the personal saving rate respectively.
Since there are five variables in this model (n = 5), we need to impose ten restrictions
on matrix A0 . According to Arabinda and Charles (2006), the output gap is measured
using the deviation of the actual output from a statistical measure of trend, when the

assumption that the trend and the cycle shocks are uncorrelated is valid. Thus, the
output gap is an unobservable state variable to be obtained using the potential output.
This implies that a12 = a13 = a14 = a15 = a21 = a31 = a41 = a51 = 0, leaving only two
additional restrictions to be imposed. To this end, Howard (1978) pointed out that the
inflation had indirect effect on the real value of net liquid assets most generally. One

possible assumption is to set a25 = a52 = 0, which implies the household saving rate
does not cause any contemporaneous effect on the inflation rate. Therefore, our final
specification of the structural form parameter matrix is,
1 0 0 0 0
 
0 1 a 23 a 24 0 
A0 = 0 a 32 1 a 34 a 35 
 
0 a 42 a 43 1 a 45 
0 0 a a 54 1 
 53

These pervious restrictions together create a recursive system which was suggested by
Sims (1980). These identification restrictions imposed on matrix A0 will be tested
explicitly as discussed in the next few sections because they are compulsory

conditions for the rest of the analysis.

3.2 Data and Variables

As mentioned in the previous section, there are five variables included in Xt which Xt '
= [x1t , x2t , x3t , x4t , x5t ].
x1t is the output gap, which is defined by Gali (2003) as ‘the deviation of output from
its equilibrium level in the absence of nominal rigidities’.
In terms of the measurement for the output gap, the traditional approach is the

difference between the actual GDP and the potential GDP. The formula is given as,
GAPt = Yt - Yt P (11)
where GAPt is the output gap, Yt is the actual GDP and Yt P is the potential output of
the economy, which is an unobserved state variable. The potential output, determined
by the supply side in the long term, is consistent with full utilization of all factors

under conditions of stable inflation. It is obvious that the potential output is necessary
in the assessment of the economic situation and is an important element for
policy- making. For example, a positive output gap may develop if the output growth
is stimulated by an increase in demand in excess of the potential output, resulting in
the inflation rate to speed up. The potential output cannot be observed from available

date directly and a series of methods are suggested to figure out the potential output. It
is assumed that the output growth can be divided into two main components: the trend
growth and the cyclical growth. The statistical method is represented by:
yt = πt +ct (12)
where yt , πt , ct are the logarithm of GDP, the trend component and the cyclical
component respectively. The trend component is often thought to be a valuable

measure of the potential output, which reflects a broad long-term growth curve around
the fluctuations of the actual output, although this argument is not unanimously held
(Canova, 1998). There are many statistical approaches to measure the time series. One
of the main methods involved the estimation of trend GDP is to use the
Hodrick-Prescott (HP) filters (1997). To figure out the output gap, the logarithm of the

actual GDP is taken at first. The potential output is estimated using a standard HP
filter, which is demonstrated by the red line in Figure 2. The value of the smoothness
parameter, λ is set equal to 1600 during all periods. From the graph, the potential
output fluctuates slightly and co-moves with the logarithm of the actual output. The
associated output gap estimates are graphed in Figure 3. It can be seen in Figure 3 that

the output gap fluctuates around zero. Even though the various output gap estimates
suggest different levels of the output gap at each point during the period, there exist
periods of broad agreement. Figure 3 presents that during the mid-1980s, all the
output gap estimates implied a period of excess demand, which might be a reflection
of the turmoil of economic reform and the associated structural adjustment during

years around 1986. In the early 1990s, there was more agreement about the relative
incidence of shocks from both supply and demand sides, which implied an increase in
potential inflationary pressures at this time. While during the early 2000s, the great
changes in the cycle told broadly consistent stories, indicating great excess supply
since the rigors of fiscal and monetary reforms. From 2005 and afterwards, the

various output gap estimates returned to a period of excess demand again, consistent
with economic circumstance of the Great Depression around 2008. After a short
period of excess supply from 2009 to 2011, the actual output approached to the output
trend gradually until 2012. Visually, the shade area highlighted in Figure 3 refers to
the drops of cycle part, which coincide with the recession dates indicated on the

National Bureau of Economic Research (NBER).

Figure 2: The Actual output and the potential output estimated by HP filter,
1980-2013.

Figure 3: The measure of the output gap and the natural rate of output, 1980-2013.

Coming back to other observable variables, x2t is the inflation rate, which is widely
represented by the changes in Consumer Price Index (CPI). The inflation rate is the
percentage rate of a change in the price index during next year.

x3t is the aggregate inflationary expectation in period t+1 based on the information
given in period t. It is plausible to assume that the consumers take into account trend
and past inflation rates when forming the inflation expectation. This means that x3t is
a function of time and past inflation rates. In consideration whether the inflation
expectation has an influence on the saving behavior, it was pointed out by Howard
(1978) that inflation expectation was related to personal savings in some countries.

Moreover, unexpected changes in inflation signal the fluctuations in real disposable

income, causing uncertainty pertaining income. Although inflation expectation is
unobservable, it can be figured out using expected changes in price index with the
same methodology as the calculation of the inflation rate.
x4t and x5t are the rate of the unemployment and the personal saving rate respectively,

which summarize the relation we mainly focus on. The unemployment rate is
observed from available data. The personal saving rate is the ratio of personal savings
to personal disposable income and it is represented by the formula,
st = (yt – ct )/yt (13)
here, yt is real personal disposable income per capita and c t is real individual

expenditure on consumption per capita.

This study uses quarterly U.S. time series from the first quarter of 1980 to the end of
2013, a total of 136 observations. The database is a collection of seasonally adjusted
output, consumer price index (CPI), the unemployment rate, real disposable income
per capita and real expenditure per capita (all chained 2009 dollars), provided from

Federal Reserve Economics Database (Fred).

The inflation rates are regarded as the growth rate of the consumer price index for all
urban consumers (all items). Output is represented by Gross Domestic Product in
nominal terms. The quarterly inflation expectations are calculated using the
seasonally expected change in prices during the next year. The data of the expected

change in prices is available from Michigan Consumer Survey.

3.3 Empirical Strategy

First of all, a unit root test should be performed on all variables except the output gap
since the output gap is stationary by construction. As all other data collection is
represented by percentage, they are bounded between 0 and 100 per cent and hence
they are logically assumed to be stationary variables. When the variables are not
stationary, which means that they can be modeled as the unit root process, they would
not have intercepts or time trends. If our variables indeed can be viewed as the

random walk process, we then must estimate our SVAR model with the variables in
first differences.
Next, before going to the estimation of the model, we should focus on the valid ity of
the identification restrictions discussed in the previous section (regarding the output
gap as the state variable and making assumptions of no direct correlation between the

inflation rate and the personal saving rate). To test this validity, the Wald tests will be
carried out. Tests will be compared with critical values from a Chi-squared
distribution.
Furthermore, the SVAR model in unrestricted form will be estimated in the first case
while the restricted specification will be done respectively in the second case. The

only difference between the unrestricted and restricted model is that the unrestricted
model assumes that all coefficients in matrix A0 are different from zero.
Finally, the proposed SVAR model will be estimated if the identification restrictions
are right. For practical considerations, Granger causality test will be taken for
determining whether the unemployment (household savings) is useful in forecasting

another one. Additionally, to focus on the impact of structural shocks, figures of the
impulse-response functions will be analysed. The following section will discuss the
most relevant empirical findings.
Section 4 Econometric Estimations

4.1 Unit Root Tests

The unit root tests on the inflation rate, inflation expectation, unemployment rate and
personal saving rate are performed in the first place. It is known that the variables will
not have intercepts or time trends if they can be modeled as unit root process. From

Figure 4, both the unemployment rate and the expected inflation rate fluctuate around
certain value while the saving rate and the inflation rate move with drifts. To see
whether there exists unit root, preliminary tests for stationary are carried out.
In the Augmented Dickey-Fuller (ADF) test, the dependent variable can be estimated
using both raw material and first differences. When the explainable variable is

estimated with a first difference and the independent variable is estimated using the
lagged level, the null hypothesis is that the coefficient of this model is zero. In other
words, the null refers that the variable would have a unit root and the alternative
hypothesis assumes that the model is stationary. The results from unit root tests
always depend on the specification of the right-hand side. This study usually includes

a constant term in the equation when estimating, because some of the time-series
variables contain secular trend.
Another Phillips and Perron (PP) test can be also be computed using the same
specifications as in the ADF test. Thus, the null is also that there is a unit root against
the alternative of stationary. Perron (1988) suggested a sequential procedure to test if

the model has unit roots and whether it includes deterministic components or not. The
procedure of test starts with the more general case under the alternative hypothesis
with an intercept and a time trend. The model will be a unit root process without drift
under null hypothesis if no deterministic components are found. The aim of the PP
test is to improve the finite sample properties in addition to that in the ADF test.

Figure 5: Graphs of the unemployment rate, personal saving rate, inflation rate and
inflationary expectation, quarterly data,1980-2013.
 unemployment rate saving rate
.11 .16

.10
.14
.09

.08
.12

.07

.06 .10

.05
.08
.04

.03 .06
1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010

inflation rate expected inflation

.04 .8

.03 .6

.02 .4

.01 .2

.00 .0

-.01 -.2

-.02 -.4

-.03 -.6
1980 1985 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 2005 2010

Table 1 shows the results of unit root tests. Statistics reported in the ADF and PP
columns of the table respectively are the t-statistics for examining the presence of unit
roots in this model. In addition, a constant and a trend with ADF and PP adjustments

for residual autocorrelation are included, when doing the tests. For the unemployment
rate and the personal saving rate, the null hypothesis of non-stationary cannot be
rejected at the 5 per cent significance level while for the inflation rate and the
inflationary expectations, the statistics are less than the critical value, implying that
we reject the null hypothesis of the existence of the unit root. However, the first

differences for these four variables are clearly stationary based on the results from
both ADF and PP tests. This implies that all the variable included in this model is I(1)
and this result is what I have expected.
Table 1: Unit root test statistics for other four variables, quarterly data, 1980-2013.
Variables ADF PP 5% c.v
x2t -7.928365 -7.831605 -3.44
Δx2t -9.467738 -31.33837 -3.44
x3t -12.12572 -15.60300 -3.44
Δx3t -8.574529 -31.93620 -3.45
x4t -2.602094 -2.115484 -3.44
Δx4t -5.664997 -5.873947 -3.44
x5t -2.711381 -3.795483 -3.44
Δx5t -16.61665 -17.41154 -3.44

4.2. Testing the Exogeneity on the Output gap

It is discussed that in the previous section, the identification restrictions imposed on

the A0 matrix implies that the output gap is fully exogenous, i.e. it does not affect or is
not affected by all other variables. Before the estimation of the structural VAR model,
the previous identification restrictions imposed are tested to become strong suppo rt
for the validity.
For this purpose, the Wald test will be taken using the estimation by least squares. For

the exogenous restrictions of the output gap, given that the mean of the output gap
equals zero, the intercept is not included in this modified model as follows,
x1t = β1 x2t +β2 x3t + β3 x4t +β4 x5t (14)
Specifically, the Wald test statistics for β1 , β2 , β3 , β4 are 10.733, 0.414, 0.920 and
7.284 respectively. All of these values are insignificant compared with the critical

values from a Chi-square distribution at 5 per cent significance level (χc=11.143).

Thereby, the null hypothesis of β1 = β2 =β3 =β4 = 0 cannot be rejected, which implies
that the output gap variable can be regarded to be exogenous in this study.
Section 5 Empirical Results

5.1 SVAR Estimation

Initially, in terms of the lag selection of the VAR models, the optimal endogenous lag
selection from the information criteria results in the choice of the lag length equal to
three, given the values of the information criteria, AIC and SBC. Following the results

of unit root tests in the previous section, the structural VAR model was estimated
using first differences. Without the identification restrictions, the estimates of VAR
model are reported in Table 2. From the results, the lags of inflation rate have no
impact on the unemployment rate and the personal saving rate, except the inflation
expectations. However, the statistics are too small for the expected inflation to affect

all the other variables. It is interesting to find that there is no significant mutual effect
of the unemployment rate and the personal saving rate.
Table 2: Summary statistics of VAR model

Inflation Expected Unemployme Personal

rate inflation nt rate saving rate

Inflation rate (-1) 0.302321 5.121521 0.011098 0.068523

(0.10068) (3.13105) (0.05013) (0.16588)
[ 3.00270] [ 1.63572] [ 0.22139] [ 0.41310]

Inflation rate (-2) 0.010996 -7.283798 -0.012851 0.248984

(0.09559) (2.97268) (0.04759) (0.15749)
[ 0.11503] [-2.45024] [-0.27001] [ 1.58100]

Inflation rate (-3) 0.199087 -1.714745 0.017887 0.101354

(0.09103) (2.83096) (0.04533) (0.14998)
[ 2.18697] [-0.60571] [ 0.39464] [ 0.67580]

Expected inflation (-1) 0.002167 -0.399653 -0.000716 -0.006831

(0.00349) (0.10859) (0.00174) (0.00575)
[ 0.62047] [-3.68034] [-0.41206] [-1.18741]

Expected inflation (-2) -0.007258 -0.319957 0.002867 -0.008597

(0.00336) (0.10442) (0.00167) (0.00553)
 [-2.16151] [-3.06403] [ 1.71463] [-1.55399]

Expected inflation (-3) -0.000983 -0.037627 0.001712 -0.002228

(0.00312) (0.09711) (0.00155) (0.00514)
[-0.31480] [-0.38748] [ 1.10123] [-0.43304]

Unemployment rate (-1) -0.516332 -0.353539 1.489904 0.055829

(0.18777) (5.83933) (0.09349) (0.30935)
[-2.74979] [-0.06054] [ 15.9364] [ 0.18047]

Unemployment rate (-2) 0.936745 1.182851 -0.332108 -0.064515

(0.32188) (10.0098) (0.16026) (0.53030)
[ 2.91024] [ 0.11817] [-2.07228] [-0.12166]

Unemployment rate (-3) -0.428033 0.696151 -0.196640 0.084826

(0.18042) (5.61069) (0.08983) (0.29724)
[-2.37244] [ 0.12408] [-2.18902] [ 0.28538]

Personal saving rate (-1) 0.033180 -2.399801 0.118481 0.566348

(0.05724) (1.78009) (0.02850) (0.09430)
[ 0.57966] [-1.34814] [ 4.15723] [ 6.00552]

Personal saving rate (-2) -0.020649 1.540546 -0.055219 0.334757

(0.06436) (2.00146) (0.03204) (0.10603)
[-0.32084] [ 0.76971] [-1.72322] [ 3.15712]

Personal saving rate (-3) 0.027282 0.249147 -0.040760 0.010161

(0.05792) (1.80132) (0.02884) (0.09543)
[ 0.47099] [ 0.13831] [-1.41333] [ 0.10648]

Output gap 0.186341 3.112604 0.007060 0.079430

(0.09015) (2.80336) (0.04488) (0.14851)
[ 2.06710] [ 1.11031] [ 0.15729] [ 0.53483]

R-squared 0.466680 0.238208 0.983410 0.879858

Adj. R-squared 0.413348 0.162028 0.981751 0.867844
Sum sq. resids 0.002465 2.384321 0.000611 0.006692
S.E. equation 0.004533 0.140959 0.002257 0.007468
F-statistic 8.750458 3.126935 592.7743 73.23508
Log likelihood 535.8469 78.70658 628.5967 469.4457
Akaike AIC -7.862360 -0.988069 -9.257093 -6.863845
Schwarz SC -7.579845 -0.705554 -8.974578 -6.581329

After imposing identification restrictions discussed above, the long-run response

pattern is,
1 0 0 0 0 
 
0 1 C(3) C(6) 0 
A0 = 0 C(1) 1 C(7) C(9) 
 
0 C(2) C(4) 1 C(10)
0 0 C(5) C(8) 1 
The structural VAR is just identified with the statistics of estimates are shown in Table
3.
Table 3: Summary statistics of structural VAR model
Coefficient Std. Error z-Statistic

C(1) 1.528473 3.932577 0.388670

C(2) 0.733607 3.044560 0.240957
C(3) 0.595170 4.235288 0.140526
C(4) 0.581371 3.007093 0.193333
C(5) 0.200734 4.303621 0.046643
C(6) 0.685535 3.541184 0.193589
C(7) 1.058479 5.454346 0.194062
C(8) 0.731943 1.108952 0.660031
C(9) -0.066709 7.030193 -0.009489
C(10) 0.566891 0.680400 0.833174
Thus, the corresponding result of structural parameter for matrix A is:
1 0 0 0 0 
 
0 1 0.6 0.7 0 
Ã0 = 0 1.53 1 1.1  0.1
 
0 0.73 0.6 1 0.6 
0 0 0.2 0.7 1 
From the matrix, only the coefficient a35 is negative while the other nonzero

parameters are positive. This finding suggests that the household saving rate and the
inflation expectation are negatively correlated. When the inflation expectation
increases, the personal saving rate will decline, which is logical since the inflation
expectations encourages the holdings of real assets rather than the assets in nominal
value. The increase of the inflation expectation would be reflected in a decrease in the

personal saving rate where consumer goods in real terms are substituted for nominal
ones. This result corresponds to the argument that the inflationary expectation
discourages the personal savings in Japan in the research of relation between
household savings and the inflation rate by Howard in 1978.
It is clearly seen that the effect of the expected inflation on the inflation rate is greater
than the effect of the inflation rate on the inflationary expectation, which is in contrary

to the results from the VAR model. Thus, the inflation expectation can be a powerful
determinant affecting the inflation rate in this long-run response estimation.
The positive link between the rate of inflation and the unemployment rate contradicts
the Phillips curve, which reveals the inverse relationship of the unemployment rate
and the inflation rate. One of the main reasons to explain this phenomenon is that the

historical Phillips curve proposed by Friedman in 1968 is only applicable in the short
run under certain circumstance. Yet, this approach is estimated over long-run horizons
and the expected inflation is taken into account in this estimation as well.
In light of the Okun’s Law, it is described that the unemployment and the national
output are negatively correlated, in which the lowered unemployment leads to a

higher national output. That is, the "gap version" demonstrates that every additional 1
per cent increase in the rate of the unemployment will lead to an increase of roughly 2
per cent lower than its potential GDP for a certain country. Generally speaking, the
higher national output induced by the unemployment results in the increase of the
expected inflation rate and vice versa. Therefore, there exists an inverse relationship

between the unemployment rate and the expected inflation. However, we can see from
matrix A the influences of both the expected inflation and the unemployment rate in
which they highly affect one another in a positive manner. This weird finding can also
be interpreted as the same logic as discussed above about the positive link of the
inflation rate and the unemployment rate.

Focusing on the last two variables, the impact of the saving rate on the unemployment
rate is similar to the impact of the unemployment rate on the personal saving rate
from the estimation. To get an explicit analysis of this relation, the Granger Causality
test was done.
5.2 Granger Causality Test

The Granger causality test is defined as a hypothesis test to determine whether the
past value of one variable help forecast another one. It is concluded from Table 4 that
the saving rate does Granger-cause the unemployment rate since the p value of 0.0014

is so low that suggests the personal saving behavior will affect the future performance
of the unemployment significantly. Definitely, the p value of 0.9947 is higher than the
critical value at 5 per cent significance level. Hence, the coefficients of the
unemployment are all equal to zero when the saving rate is the dependent variable,
indicating that the unemployment rate offers no other information about the prediction

of the saving behavior in the long run. A likely possible explanation of this result
could be that the unemployment rate over the long-run horizons is determined by
demand-supply side of the labor market, nothing with the household saving behavior.
Table 4: Granger Causality test of the unemployment Rate and the saving rate,
1980-2013.

Null Hypothesis Obs F-Statistic Prob.

Saving rate does not Granger Cause unemployment rate 133 5.48772 0.0014
Unemployment rate does not Granger Cause saving rate 0.02473 0.9947

5.3 Structural VAR Impulse-Response Analysis

To analyse the specific relation between the personal saving rate and the
unemployment rate in the short term, the structural shocks can be explained via the
standard impulse response functions.

Figure 6 captures the impulse response functions (IRFs) of the expectation inflation
and the unemployment rate to the personal saving rate shocks. Every plot displays the
dynamic response of related variables to one standard deviation in permanent or
transitory shocks over forecast horizons from 1 to 48 quarters. It is obvious that the
expected inflation responds negatively to the shocks of the personal saving rate,
coincide with the same result got from the corresponding coefficient in the structural
VAR estimation. By the end of the sample period, the response of the inflation
expectation approaches to zero, implying that the direct effect of the expected

inflation shocks on the performance of personal savings dies out. In contrast to the
negative effect of the inflation expectation, a positive effect on personal saving rate
due to the unemployment rate can be observed during the period of 48 quarters. The
effect of this transitory shock demonstrates that the unemployment plays an essential
role in the consumption and saving behavior, which is behind the discussion by

Malley and Moutos (1996). At the end of their discussion, evidence provided suggests
that the precautionary saving is the primary reason that consumers cut their spending
on durables when the uncertainty pertain their personal employment status. The
tendency for the unemployment to increase personal savings can be explained in
several ways. Juster and Wachtel (1972) argued that the consumers are assumed to

prefer additions to savings in response to the unplanned withdrawals when there is a

rise in the variance of the expected real income. But at the beginning, there is no
response of the unemployment to the personal saving rate. Afterwards, this response
increases slightly in next year and reaching to around 1.5 per cent in two year horizon.
This result supports the allowance of precautionary savings by households. When the

income uncertainty from the employment status increases, individuals always adjust
the consumption-saving pattern based on their own characteristics after the occurrence
of losing jobs. That is, even though being unemployed and having more risk of
income uncertainty, households do not immediately change their saving plan with the
existence of precautionary savings. Acemoglu and Scott (1994) used the

representative agent framework to do research in the theory of the precautionary

saving, finding that the precautionary savings play an important role in explaining the
predictive ability of the consumer confidence.
Figure 5: Accumulated impulse response functions: response to the household saving
rate to one standard deviation innovations for 48 quarters.
When considering the factors affecting the shocks of unemployment, we find that
though the effect of the inflation rate on the unemployment is negative at the starting
point, the effect approaches to zero gradually and becomes positive five years later.
The initial inverse relationship reveals the Phillips curve in the short run. However,

after modifying the forms of the Phillips curve and taking other variables into account,
there is no direct tradeoff between the unemployment rate and the inflation rate. For
instance, when the inflation expectation is included, the negative effect of inflation
rate on the unemployment is influenced due to positive relation of the expected
inflation and the unemployment rate. The response of the expected inflation to the

unemployment rate undergoes the same path as the response of the inflation rate with
initially negative impact and afterwards positive effect during this period.
What’s more, it is shown in the last graph of Figure 6 that changes in the personal
saving rate does not cause any response to the unemployment rate immediately. After
one year, there exists slightly positive influence of the personal saving rate on the
unemployment rate. In five year horizon, additional one per cent increase in the

shocks of the personal saving rate can result in about 2.5 per cent rise in the changes
of the unemployment rate. As time goes on, this influence becomes higher. Thus, in
the Granger Causality test, the personal saving rate does Granger-cause the
unemployment rate in the long-run response.
Figure 6: Accumulated impulse response functions: response to the unemployment

rate to one standard deviation innovations for 48 quarters.

To sum up, for the impulse-response analysis, there is no immerdiate correaltion
between the personal saving rate and the rate of the unemployment. However, in the

short run, the response of one shock to another shock for household savings and the
unemployment rate is concluded to be positive, as illustrated in Figure 5 and Figure 6.
Section 6 Conclusion and Further Discussion

6.1 Conclusion and Implications

These time-series results indicate that although effects of shocks in the unemployment
and personal saving rate are zero at first, the unemployment rate has an important
impact on private saving behavior in the short term and the personal saving rate also

has an influence on the forecast of the employment status. This is because households
always plan their expenditures on the basis of intertemporal considerations. To be
more specific, in order to smooth their consumption, individuals tend to have
precautionary savings against the increase of income variance caused from the
unexpected unemployment. The tendency for the unemployment to increase personal

savings can be explained in several ways. The most common is the hypothe sis that the
current high unemployment rate affects consumers’ confidence of future income and
thus, they will augment personal savings in proportion to total disposable income. As
a result, this leads to an increase in the household saving rate. In turn, the household
saving rate has tiny impact on the unemployment at the beginning but over long-run

horizons this impact becomes so high that the household saving behavior would have
an explanation power in the forecast of the unemployment rate.
It is suggested by the Keynesian consumption function that the existence of a
precautionary saving motive is a reply of individuals to the changes in income
uncertainty. As far as some policies are concerned, the certainty equivalence PI model

proposed to have a prediction power that cutting current taxes but later increasing
beyond the normal level is expected to have no impact on consumption and saving
behavior.
From a social policy perspective, the significant effect of the unemployment on the
saving behavior is important to policy implications related to the social insurance. To

be more specific, if the unemployed households’ income is not greatly lower than the
employed households’ income, the effects of income uncertainty resulted from
unemployment on household savings will be quite small. Hence, it can be concluded
that the unemployment does not have a significant impact on private saving behaviors
in some countries which have generous social welfare systems. This finding suggests

that differences in the ratio of the employed individual’s income to the unemployed
individual’s income will be a response to the differences in the adjustment of
consumption and private savings (Malley and Moutos, 1996).

6.2 Limitations and Further Research

When previewing the results of the model, I have made the conclusion reached in this

paper depending on the limited availability of the data. The major problem is the lack
of data available for inflationary expectations. There are two expected changes in
prices with one change during the next year and another during the next 5 years. I use
the mean of expected change in prices during the next year when calculating the
expected inflation. Choices of different dataset have significant effect on the accuracy

of the tests and estimations. Given this limitation, it is important to notice that
different authors may find opposite results on the relation between the household
saving behavior and the unemployment rate if they use different data.
In this study, it is plausible to assume that the inflation rate and the personal saving
rate do not have effects on each other when it comes to the identification restrictions.

The reason why the effects are ignored is that the contemporaneous link between the
saving rate and the current inflation rate is much smaller, in contrast to the mutual
effect of other variables. Thus, I make the assumption that the coefficients of these
two variables are zero. However, in the research on the rate of inflation and personal
saving behavior, the inflation rate is concluded to have influences on the personal

saving rate in major industrialized countries (Howard, 1978). In the further research
on this topic, the restrictions of structural VAR model should be better addressed with
the full consideration of the mutual effects.
Another fact that needs to be taken into account is that some financial spreads have
significant impact on the household saving behavior. Take the interest rate as an
example. The interest rate is the opportunity cost of consumption and variations in the
interest rate do affect individuals saving behavior. It is known to all that when the
interest rate is high, households prefer to reduce their expenditure and save more to

get high interest. On the opposite side, the low interest rate discourages personal
savings. In addition, the interest rate can be used as an important tool in the decision
about monetary policy, which will result in the potential influence on the inflation rate.
Here, I did not include the interest rate mainly for two reasons. First of all, I mainly
focus on the household saving rate and the unemp loyment rate in America. It is

important to remark that the interest rate has less effect on the household saving
decisions and other economic indicators in the U.S. than any other countries,
especially the developing ones. Secondly, according to the Fishe r effect, the nominal
interest rate is defined to be determined by the expected inflation rate. To some extent,
the inflation expectations included in my study can be regarded as a proxy for the

interest rate. However, if the interest rate and other valuable financial variables are
included in this research, the result will be more efficient.
Bibliography

Acemoglu, D. and Scott, A. (1994), 'Consumer Confidence and Rational Expectations:

Are Agents Beliefs Consistent with the Theory?' The Economic Journal, 104, pp.1-19.

Arabinda, B. and Charles, R. N. (2006), ‘New measures of the output gap based on
the forward- looking new Keynesian Phillips curve’, Journal of Monetary Economics,
54, pp. 498–511.

Basistha, A. and Nelson, C.R. (2006), ‘New measures of the output gap based on the

forward-looking new Keynesian Phillips curve’, Journal of Monetary Economics,

Vol.54

Canova, F. (1998), ‘Detrending and business cycle facts’, Journal of Monetary

Economics, Vol 41, pp. 475-512.

Card, D. (1995), ‘The wage curve: A review’, Journal of Economic Literature, Vol.
33, pp.785–799.

Carroll, C. and Samwick, A. (1998), ‘How important is precautionary saving?’,

Review of Economics and Statistics, 80, pp.410–19.

Carroll, C.D. (1992), 'The Buffer-Stock Theory Of Savings: Some Macroeconomic

Evidence', Brookings Papers On Economic Acitivity, 2, pp.61-135.

Cermeño, R., Roth, B. D. and Villagómez. F. A. (2008), ‘Fiscal Policy and National
Saving in Mexico, 1980-2006’, Estudios Económicos, Vol. 23, No. 2 (46), pp.
281-312.
Deaton, A. (1992), Understanding Consumption, Claredon Press, Oxford.

Dickey, D.A. and Fuller, W.A. (1981), ‘Likelihood Ratio Statistics for Autoregressive

Time Series with a Unit Root’, Econometrica, vol.74, pp.1057-1072.

Dynarski, M. and Sheffrin, S. M. (1987), 'Consumption and Unemployment',

Quarterly Journal of Economics, 102, pp.411-28.

Enders, W. (1995), ‘Applied Econometrics Time Series’, John Wiley and Sons, New
York.

Friedman, M. (1957), ‘A Theory of the Consumption Function’, Princeton University

Press.

Friedman, M. (1968), ‘The role of monetary policy’, American Economic

Review, 68 (1), pp.1–17.

Fred St. Louis. (2014) Available at: http://www.research.stlouisfed.org/index.html

(Accessed: 3rd August, 2014)

Gali, J. (2003), ‘New perspectives on monetary policy, inflation and the business
cycle’, Advances in Economic Theory, Vol. 3, Cambridge University Press,
Cambridge, pp. 151–197.

Groenewold, N. and Hagger, A.J. (2000), ‘The natural rate of unemployment in

Australia: Estimates from a structural VAR’, Australian Economic Papers, Vol. 39,
pp.121-137.

Hall, R. E. (1978), ‘Stochastic implications of the life cycle-permanent income

hypothesis: theory and evidence’, Journal of Political Economy, 86, pp.971-987.
Howard, D. H. (1978), ‘Personal Saving Behavior and the Rate of Inflation’, The
Review of Economics and Statistics, Vol. 60, No. 4, pp. 547-554.

Lütkepohl, H. and M. Krätzig (2002). Applied Time Series Econometrics, Cambridge

University Press.

Jim, M. and Thomas, M. (1996), ‘Unemployment and consumption’, Oxford

Economic Papers, New Series, Vol. 48, No. 4 (Oct., 1996), pp.584-600.

Juster, F. T. and Wachtel, P. (1972), ‘Inflation and the Consumer’, Brookings Papers
on Economic Activity, no. 1 (1972a), pp.71-114.

Keynes, J. M. (1936), ‘The General Theory of Employment, Interest and Money’,

Macmillian, London.

Knudsen, O. and Paarnes, A. (1975), ‘Trade instability and economic development’,

Lexington, Lexington Books.

Mauro, M. and Rob, A. (2013) ‘The precautionary savings motive and

household savings’, Oxford Economic Papers, pp.164-187.

Malley, J. and Moutos, T. (1996), ‘Unemployment and consumption’, Oxford

Economic Papers, Vol. 48, pp.584-600.

Michigan Consumer Survey. (2014) Available at:

http://www.sca.isr.umich.edu/data-archive/mine.php (Accessed: 3rd August, 2014)

Modigliani, F. and Brumberg, R. (1954), ‘Utility Analysis and the consumption

function: an interpretation of cross-section data’, Post-Keynesian Economics, Rutgers
University Press, New Brunswick, pp. 388-436.

Modigliani, F. (2000), ‘Europe's Economic Problems’, Prepared for Testimony before

the Monetary Committee of the European Central Bank.

Moore, T. and Pentecost, E. J. (2006), ‘An investigation into the sources of

fluctuation in real and nominal wage rates in eight EU countries: A structural VAR
approach’, Journal of Comparative Economics, Vol.34, pp.357-376.

National Bureau of Economic Research (NBER). (2014) Available at:

http://www.nber.org/ (Accessed: 3rd August, 2014)

Perron, P. (1988), ‘Trends and random walks in macroeconomic time series: further

evidence from a new approach’, Journal of Economic Dynamics and Control, vol. 12,
pp.297-332.

Ramudo, R., Grela, M.F. and García, M.R. (2007), ‘Consumption, saving, investment,
and unemployment. SVAR tests of the effects of changes in the consumption-saving

pattern’, University of Santiago de Compostela.

Rosen, S. (1983), ‘Unemployment and Insurance', in K. Brunner and A. Meltzer (eds.),

Employment, Prices, and Money, Carnegie-Rochester Conference Series on Public
Policy, 15, pp.5-50.

Sims, C. A. (1980), ‘Macroeconomics and Reality’, Econometrica, Vol. 48(1),

pp.1-48.

Ventura, L. and Eisenhauer, J.G. (2006), ‘Prudence and precautionary saving’,

Journal of Economics and Finance, 30, pp.155–68.

Yotopoulos, P.S. and Nugent, J.B. (1976), ‘Economics of development: Empirical
investigations’, New York: Harper&Row.