Professional Documents
Culture Documents
CENTRAL BANK
OF SWAZILAND
RESEARCH BULLETIN
VOLUME 2
RESEARCH BULLETIN
VOLUME 2
MARCH, 2018
FOREWORD
The Central Bank of Swaziland welcomes you to yet another rendition of its Research
Bulletin. This publication forms part of the Bank’s endeavor to provide voice to research
work conducted within the institution. We hope that a reading of this document ignites new
discussions, revives unfinished deliberations and reorganizes existing policies in Swaziland.
Encompassed in this document is the research work that has been undertaken during the
2017/2018 financial year.
It has to be amplified that whilst care has been given to ensure accuracy in terms of
methodological application and use of analytical tools available, errors and omissions are
those of authors. As such should a reader identify such an error and/or an omission, please
forward these to the General Manager, Economic Policy Research and Statistics Department
who is reachable at sikhumbuzod@centralbank.org.sz. Further, each of the published
material contains contact details of the authors. Readers are encouraged to interact with
authors to enable the Bank to improve the quality of its research work.
Disclaimer: The ideas expressed in the papers should not be reported as representing the
views of the Central Bank of Swaziland. The views expressed in this papers are those of the
authors and do not necessarily represent those of the Central Bank of Swaziland and Central
Bank Policy.
TABLE OF CONTENTS
Zana S. Mabuza and The Relationship between Household Debt and Economic 71
Ntobeko S. Dlamini Growth in Swaziland.
Ntobeko Dlamini Rand-Dollar Price Fixing Effect and Policy Recommendation. 122
increased from a low of 1.5 per cent of GDP (2011) argue that for high level of debt it is a
in 2009 to 10.0 per cent of GDP in 2016. As challenge to know if debt has reached levels
noted by Semmler and Sieveking (2000), an that are suppressing to growth. The debt
increase in debt above a critical level can threshold therefore provides a quantifiable
result to unsustainable debt and insolvency. level at which additional debt lowers
Consequently, creditworthiness may be growth by channeling resources to debt
lost leading to capital outflows causing a repayment. Unlike the critical debt ratios
currency and financial crisis and a huge set by the World Bank, the study determines
decline in output. a country-specific debt threshold supported
by econometric models. For example, the
Against this background, the purpose of this 40 per cent to GDP debt critical ratio for
study is to establish an optimal level of public developing countries may not be applicable
debt given the growth rate for Swaziland in Swaziland.
using both quantitative and qualitative
analysis. More specifically, the study intends To the best of our knowledge, no study has
to achieve the following objectives. Firstly, been done in relation to the determination
to ascertain if there is evidence in support of an optimal public debt threshold beyond
of a non-linear quadratic relationship which additional debt impedes economic
between the public debt-to-GDP ratio and growth in Swaziland. Unlike existing studies
growth in Swaziland. Secondly, to determine which investigate the relation between debt
the optimal debt-to-GDP ratio for Swaziland and growth within growth models, this study
above which increases in public debt has a examines this relation using the aggregate
negative effect on growth. Thirdly, to find out demand specification. In this regard, we
if there are significant differences between build on the work by Moron and Winkelried
the estimated optimal debt-to-GDP ratio (2005), who formulate an aggregate demand
for Swaziland and the 35 per cent of GDP function for Latin American countries. By
debt limit set by the Swaziland government adding the linear term and squared term
in its debt strategy3 and the 40 per cent of of the public debt-to-GDP to the aggregate
GDP4 debt limit set by the World Bank for demand equation, we formulate a non-linear
developing countries respectively. Lastly, debt-growth nexus. This formulation allows
we intend to verify if the estimated optimal us to investigate whether public debt has a
debt threshold is in line with the SADC debt positive short-run and a negative long-run
limit of less than 60 per cent of GDP. effects on economic growth.
The significance of the study is that the The remainder of the paper is organized as
public debt threshold will guide policy follows: In section 2, we review previous
makers in designing appropriate optimal related literature on the relationship
public debt strategies for the country. In this between public debt and economic growth,
regard, the debt threshold acts as a debt and on the determination of threshold
ceiling to ensure positive economic growth levels. Section 3 presents developments of
rates and debt sustainability. Pattillo et al. the public debt and growth in Swaziland
in relation to its counterparts in the
SACU region. In section 4, we present the
theoretical public debt threshold models.
Data analysis and the estimation technique
3
Source: Budget speech presented by the Minister of are presented in section 5. In section 6,
Finance, 21 February 2014 pp.27.
the empirical results are presented and
4
These debt-to-GDP ratios are based on the standard
analyzed. Section 7 concludes with some
critical value of debt ratios sourced from the World Bank,
World debt Tables: 1989 - 91.
policy recommendations.
Pattillo et al. (2011) use a quadratic region, Swaziland’s public debt as percentage
function to examine the nonlinear linkage of GDP remains low over the period 2006
between public external debt and growth to 2016. The current debt-to-GDP ratio is
for 93 developing countries. Their findings about 20 per cent and below the 35 per cent
confirm the existence of a nonlinear and a limit set by government. The debt limit set
hump-shaped relationship between debt by government is within the convergence
and growth. The results indicate debt-to- criteria set by SADC, which stipulates a
GDP threshold levels ranging from 35 to public debt as percentage of GDP of less than
40 per cent for these countries. Similarly, 60 per cent. Over the period under review,
Wright and Grenade (2014) examine the Swaziland’s public debt averaged 14 per
relationship between public debt and growth cent of GDP. This is lower than the averages
and the nonlinearity issue using panel of 15 per cent recorded in Botswana, 42 per
dynamic OLS and threshold dynamics in 13 cent in Lesotho, 24 per cent in Namibia and
Caribbean countries. Their findings indicate 38 per cent in South Africa. For Swaziland,
that beyond the debt-to-GDP ratio of 61 per public debt started growing at a faster pace
cent, debt negatively affects investment during and after the global financial crisis.
and economic growth. For Nigeria, Bawa et This was due to the crisis induced decline in
al. (2016) find a total public debt-to-GDP SACU revenue, which was reflected in trends
ratio of 73.7 per cent, an external debt ratio in fiscal deficits.
of 49.4 per cent and a domestic debt ratio of
30.9 per cent above which growth begins to As shown in Table 1b economic growth in
fall, respectively. Swaziland considerably slowed down in 2008
to 0.8 per cent. On average, Swaziland’s
3.0 PUBLIC DEBT AND ECONOMIC economic growth remains relatively low at
GROWTH IN SWAZILAND 3.4 per cent compared to 4.6 per cent in
Tables 1a and 1b show SACU revenue growth Botswana, 4.5 per cent in Lesotho and 4.7
rates, GDP growth rates and public debts per cent in Namibia over the period under
as percentage of GDP in SACU countries. review.
Compared to its counterparts in the SACU
Table 1a: SACU Revenue Growth (SR), GDP Growth (GDP) and Public Debt as % of GDP (PD) in
SACU Countries 2006-2016
Botswana Lesotho Namibia
SR GDP PD SR GDP PD SR GDP PD
2006 107 8.4 6.2 112 4.2 54 167 7.1 24
2007 8.3 8.3 8.3 -2.6 4.8 50 -23 6.6 19
2008 5.2 6.2 7.7 20 6.7 44 28 2.6 18
2009 -30 -7.7 18 -27 2.2 33 -28 0.3 16
2010 -14 8.6 19 -39 6.5 30 -16 6.0 16
2011 58 6.0 20 27 6.6 32 39 5.1 23
2012 71 4.5 19 117 5.9 36 93 5.1 24
2013 0.3 11.3 17 1.5 2.2 38 6.7 5.7 24
2014 26 4.1 17 18 2.3 43 25 6.5 25
2015 4.0 -1.7 16 -11 5.6 50 -6.6 5.3 38
2016 15 2.9 14 -2.4 2.5 48 14 1.2 41
Avg. 23 4.6 15 19 4.5 42 27 4.7 24
Sources: Central banks of SACU members and Swaziland Revenue Authority
limited
C E N T R A L B A N K O F S WA Z I L A ND | expenditure
R E S E A R C adjustments
H B U L L E T I N led
V O Lto
U Mextreme
E 2 non-payment
fiscal stress. This is evident in Figure 1 where private secto
fiscal balances deteriorated from surpluses sector contr
Table 1b: SACU Revenue Growth (SR), GDP Growth (GDP)
recorded and
prior to Public Debt
the global as % ofcrisis
financial GDP (PD)
to economic gr
in SACU Countries 2006-2016 register high deficits in 2009 and 2010 in line during and af
RSA Swaziland
with a significant decline in SACU revenues. The
SR GDP PD SR GDP PD DD 4 Theoretica
situation was further exacerbated by Swaziland's
2006 41 5.6 26 105 6.0 16 2.6 To determine
failure to secure international funding to close the
2007 16 5.4 24 -13 4.4 18 2.0
between pub
2008 14 3.2 financing
26 gap.20 Domestic
0.8 payments
16 arrears
1.6
the optimal d
2009 -9.4 -1.5 accumulated
31 to-37
more than
1.65 per cent
12 of GDP1.5 in
threshold ty
2010 6.3 3.0 201135due to the
-48 government's
3.8 12
inability to 4.7
pay
2011 1.0 3.3 38 in time.
46 The fiscal
2.2deficit also
11 increased
4.8 allows the
suppliers
2012 30 2.2 41 13 per
145 4.7 in 2011/12.
12 Without
5.5 according to
to almost cent of GDP
2013 10 2.5 44 1.3 6.9 13 5.8 below a thr
adjustments to expenditure to match the declined
2014 9.4 1.7 47 6.3 4.2 13 5.8 been applied
revenue flows, public debt, particularly domestic
2015 4.8 1.3 49 -10 1.9 15 5.9 Proano et al.
debt rose on the back of high fiscal deficits.
2016 13 0.3 52 4.3 0.6 19 10.0 the debt-to-G
Avg. 13 2.4 Domestic
38 debt continued
20 3.4rising since
14 2010 to
4.6
reduces outp
Sources: Central banks of SACU members and Swaziland Revenue
register 10 per cent in 2016.
Authority
Grenade (2
Swaziland’s SACU revenue declined Figure
Figure 1: 1: Fiscal
Fiscal Balances
Balances andRevenue
and SACU SACU Revenue
Growth
between pub
significantly by 37 per cent in 2009 and Growth Rates
Rates
48 per cent in 2010 compared to those of dynamics. Th
the other SACU members. As indicated 200 15
is specified a
by Mafusire (2015), the sharp decline in 150 10
Per cent
expenditure adjustments led to extreme 50 0
2007
2008
2009
2011
2012
2013
2014
2015
2016
Per cent
t
economic growth, hence the optimal debt
Per cent
according to whether public debt is above or 𝑦𝑦 = 𝛼𝛼 + 𝛼𝛼 𝑥𝑥 + 𝛼𝛼 𝑑𝑑(𝑥𝑥 − 𝑘𝑘) + 𝛽𝛽 ′ qu
𝑍𝑍
Uditure 100 adjustments
limited expenditure
200 led to extreme adjustments
accumulated non-payments 5to led more to extreme
than to5 per
15 suppliers cent non-payments
of GDP negatively in toaffected supp
SACU Growth
Per cent
stress.
nd after the This threshold,
global is evident
economic 4 thecrisis.
Theoretical study
in Figure employs
Model 200
1 threshold sector activities. is Given
specified thatas followthe
nd. The 2011/12. 50fiscal below Without
deficit a threshold 0also increased
value. This framework between 0 has 𝑦𝑦 ,
interpreted
𝑡𝑡-5 and 𝑥𝑥 𝑡𝑡 10 is nonlinear
as inability
linear andtononlinear of an inverted threshold
terms U- t
eclined by
0 Swaziland's type nonlinear 150 model. This 2011
150model due
allows tothe the government's pay Th
This 200 is evident fiscal instress. Figure This
below 1 where
is a evident
threshold
-5private
according in
15 Figure
linear
shape.
value. of
to sector
is the1
and
Hence, This
whether where activities.
nonlinear
the
threshold
specified coefficient
framework variable.
public
as
10private
terms
follows: Givendebt of Eq.
has of
𝑥𝑥 1, sector
that
(2) the
is
is expected
𝑖𝑖𝑖𝑖
indicates
above the activities
threshold (𝑥𝑥 public
orto> 𝑘𝑘)
2013 2010
2006
2007
2008
SACU Growth2009
2014 2011
2015 2012
2013
2014
2015
2016
balances been existence
deteriorated
applied -50 of
To in atwo
determine number policy
from regimes
theofpotential surpluses
studies. according
nonlinear
For shape.
instance, sector
-5 relationship
Hence, -10 the contributes
coefficient 1,ofabout 𝑖𝑖𝑖𝑖 𝑡𝑡
is{ expected
𝑑𝑑𝑥𝑥𝑦𝑦𝑡𝑡=Volume 40(𝑥𝑥 𝑡𝑡𝛼𝛼allows
per >+to 𝑘𝑘)cent +𝑥𝑥𝑡𝑡the
𝑡𝑡
o
2010
2006
2007
2008
2011
2012
2013
2014
2015
2016
etical
cent match
ding
mestic of0close
Model
to the
GDP declined
the inwhether
2011/12. 100 Without 100suppliers -10in time. The that fiscal 5Research
deficit 𝑑𝑑 =also {5Bulletin
increased 𝑡𝑡 =0, 𝑦𝑦𝛼𝛼𝑡𝑡0= ,2+and 𝛼𝛼 1 𝑥𝑥𝑡𝑡𝛼𝛼}
0 𝑂𝑂𝑂𝑂ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝛼𝛼+ap
SACU Growth
Per cent
es deteriorated
150 fiscal balances
from surpluses deteriorated sector 10 from
be variable. contributes
positive Eq.
surpluses and (2) that indicates
aboutof the sector 40
squaredthat per the
contributes cent
variable relationship of
to be GDP
abou
Per cent
0,= { U-shape. 𝑂𝑂𝑂𝑂ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
2010
2006
2007
2008
2009
2011
2012
2016
-100
et between public debt -15 𝑡𝑡
-50 Proano al. (2014) beenestimate 50aand
applied economic
threshold in -15
below aadjustments
benumber
level growth,
a ofthreshold
positive
-10 is of hence
and studies.
nonlinear thatvalue. of an
0of For
theinverted instance,
𝑑𝑑
0This
squared framework
variable according
Hence,to be has }t
ed
mine
yments
100eficits.
articularly
xpenditure 100 prior
the to
potential
arrears a the
domestic
to
threshold
nonlinear
match global 50 value.
SACUthe financial
relationship
limited
declined
Growth
This
rate expenditureto crisis
framework
almost to
Deficit/surplus135
has
per as
economic
cent% GDP ofled
𝑦𝑦GDP =
Researchto in
𝛼𝛼 growth
Where
extreme2011/12.
+
Bulletin 𝛼𝛼 𝑦𝑦𝑥𝑥 𝑡𝑡
Volumeslowed
is
+ Without the
non-payments
0,
𝛼𝛼 2 𝑑𝑑(𝑥𝑥 even
dependent − 𝑘𝑘) + further
to
𝑂𝑂𝑂𝑂ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝛽𝛽variab
′ supp
𝑍𝑍 si
+
between
negative. the coefficient
Where 𝑦𝑦𝑡𝑡𝑡𝑡a, and
Differentiating
𝑦𝑦 0𝑥𝑥𝑡𝑡 is of nonlinear
1-5eq. is(2)
𝑡𝑡dependent 2 of
expectedwith an inverted
𝑡𝑡 respect to be to U-𝑡𝑡 𝑥𝑥
or to -100thethe global
recorded
been debt-to-GDP the
applied optimal
financial prior
in aStates
ratio debtto threshold,
crisis
number
beyond the 0whichto
ofglobalstudies. the
a GDP
rise economic study
financial
in For
debt employs crisis
growth -5isslowed
𝑡𝑡 to the economic even further
growth variable, down slow
cent
0growth, -15
negative. Differentiating eq. (2) with respect below to a 1,th
rpublic
010 r centSACU
to
high of
debt Growth
GDP and
deficits in rate SACU
economic
Source:
in
Member
2009 Proano
Deficit/surplus
fiscal and hence
al.2010
et stress. et al.
Thisas %
in
adjustments (2014)
been line
is applied
evident estimate
0 toaexpenditure positive
during in
intoFigure a
Where aand threshold
number
andto threshold
1the that
after
where
match 𝑦𝑦 of of
the
is the level
studies.
variable
the
the global of
squared
private
declined dependent For and
economic
sector instance,
variable 𝑘𝑘 isactivitie a= {give
crisis
variabl (2
2010
2006
2007
2008
2009
2011
2012
2013
2014
2015
2016
50 instance, Proano (2014) estimate 1,
high
ublicMember fiscal
debt, deficits.
particularly
2010 limited domestic
expenditure adjustments shape.led extreme
Hence, coefficientnon-payments of 𝑥𝑥 is to
expected suppliers𝑑𝑑 to
2006
2007
2008
2009
2011
2012
2013
2014
2015
2016
SACU SACU
reduces
StatesGrowth threshold
outputrate growth. typeDeficit/surplus-50
nonlinear Wright
Similarly, model.
as % GDP the
andThis threshold model variable 𝑡𝑡 and -10 equating𝑡𝑡 to zero 𝑑𝑑 yields
= {
Per
deficits toinpay register
2009 theand -50
high 2010ofdeficits in-100 line in 2009 during and to
threshold
2010
and be
afterin negative.
line
the -10variable Differentiating and 𝑘𝑘zero isaftereq. a given
been (2) applie 0, th
the threshold variable
aglobal
and during
𝑑𝑑equating aeconomic and
to crisis.
yields the glob
mal debt threshold,
inability threshold
Mafusire level
study employs the adebt-to-GDP -5ratio
urce:0SACU Member States (2015) the
fiscal further
debt-to-GDP
balances notes
revenue Proano that ratio
deteriorated
flows, the
et beyond
publical.higher (2014)
from debt, which value.
estimate
surpluses
particularly rise is
-15 ina debt
dummy
threshold
sector
domestic variable
contributes level that 0,of ass
abo de
he significant
ing backsince decline
beyond
Grenade
of2010 high fiscal
allows
which
(2014)
to in
fiscal SACU
astress.
the
riseexamine revenues.
This
existence
in debt is
reduces
the evident
of The
two output
relationship in
policy beFigure with
positive
the following; -15 regimes 1 respect
threshold and
where thatto thevariable
of the
private threshold squared sectorand variable
variable 𝑘𝑘
activities.
(𝑥𝑥 is and a
to give
be Giv
-100
Thisdeficits. 1, 𝑖𝑖𝑖𝑖
Where 𝑡𝑡𝑡𝑡 >
𝑦𝑦Proano is 𝑘𝑘) the da
2010
2006
2007
2008
2009
2011
2012
2013
2014
2015
2016
dire it -50(2015)
type
also nonlinear
increased further
interest model. notes
costs associatedthat
model theSACU with higher
Growth high theratefollowing;
borrowing 4 Theoretical
value. 𝑑𝑑
Deficit/surplus is a asModel
dummy 𝑑𝑑
% GDP = { variable that assumes et
} qu
cant
Growth decline with
growth.in SACUa significant
Similarly, revenues.
recordedWright
reduces decline
and The
prior Grenade
debt
output to in
rosethe
theStatesSACU -10
(2014)onglobal
growth.
debt-to-GDP revenues.
the equating
financial
back
Similarly,
ratio of toThe zero
of
crisis
highbeyond one yields
to
fiscal
Wright for the
Where
which
following;
observed
economic
deficits.
and
0, contributes awith𝑦𝑦
rise is
values
growth
in
𝑂𝑂𝑂𝑂ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 the
debtto dep
ofslow th
on afusirewas (2015)
betweenexacerbated
further further
according
public
fiscal debt
SACU notes
to
balancesand
Growth
Source:whether
by growth that
rate
Swaziland's
SACU using
deteriorated
Member the
public 4 higher
debt
threshold is
negative. above
from surpluses
Deficit/surplus
Theoretical or
Differentiating
asModel
value. % GDP 𝑑𝑑 is a sector 4
dummy eq.
Theoretical (2)threshold
variable 𝑡𝑡 respectModelabout
that
variable assu40
he
st ontinued
011/12. -100 existenceWithout
costsexacerbated rising
of
examine
associated two
requirements since
thepolicy
with 2010 regimes
relationship
spilled high to borrowing
over
between
into the
public
-15
private sector. 𝜕𝜕𝑦𝑦
𝜕𝜕𝑦𝑦𝑡𝑡 the debt-to-
further situation
dynamics.
debt
Source:
and below
The SACU
growth
recorded was
by a
general Swaziland's
Member
register further
threshold
using
Grenade
States
empirical
prior high
threshold
to exacerbated
deficits
Domestic
value.
threshold
the (2014) This
dynamics.
global in
debt 2009
framework
model by
examine
the
financial of Swaziland's
continued
To and
threshold one𝑡𝑡2010
determine
Where has
crisis thefor
=
rising variable
𝑦𝑦
variable
to in𝛼𝛼 1 observed
the
is −
line
since 2𝛼𝛼
the
relationship
and
greater
potential
economic 2 𝑥𝑥2010
threshold =
during
dependent
equating
𝑡𝑡 0values
than
to
growth and
nonlinear
to 𝑑𝑑zero
theafter
variableof given
(3the
variable, the
yields
slowed
) relat thr
glob
an th
Sw
e
gterest
atch to to whether
the
secureSACU costs
declined public
international
Growth associated
rate debt is above
funding Mafusire
with or
Deficit/surplus high (2015)
to closemodel reduces
borrowing
as %the GDP further output notes of growth.
=
that
𝜕𝜕𝑥𝑥 one
𝑡𝑡 𝛼𝛼 1 the
𝑡𝑡− 2𝛼𝛼
for To Similarly,
higher
2 𝑥𝑥 =
observed
𝑡𝑡 0 value. Wright
values (is3
reduces
)a and
dummy of ou th
nt
ements15in 2016. The
spilled The general
increase
overbeen empirical
into in
with
applied the
the aclose
in threshold
government
private
significant
a number register sector.
ofdecline To
borrowing
10funding
studies. determine
perand is
in cent
For SACU from
inthe
instance,
𝜕𝜕𝑥𝑥 𝑡𝑡 the
revenues.
2016. potential
values The and nonlinear
determine
zerothe otherwise. relationship
the potent
𝑍𝑍𝑡𝑡 is ath
ACU e:re international
SACU Revenue
threshold
cularly Member is specified
domesticvalue. Mafusire
failure
Growth
States This
as
fundingto
register follows:
secure
framework (2015)
to
highinterest further
international
deficits
has thecosts inGrenadenotes
2009
associated the that variable
betweento
following;
2010
with close
threshold inhigh higher
line greater
the
public borrowing
variabledebt
during than and
value. and
and economic
𝑑𝑑
after given
is
𝑘𝑘 a the
is dummy a thresho
growth
global given va(
eco
equirements
ng gap. the
10
specified
spilled
Domestic domestic
as follows:
over between
into
payments
markets thecaused
public
privatearrears
the
debtsector.
lending The
and (2014) growth
optimal
rates variable
to
using
examine
threshold greater threshold
level
4
the Theoretical
than
𝑥𝑥 ∗ of
is relationship
one
the
obtained
for
Grenade
given
Model observ
by thr
ncrease in the Proano
government et al.
situation (2014)
borrowing was estimate
further from abetweenthreshold
exacerbated
The optimal The public
level by of debt
Swaziland's
threshold
optimal contains
threshold and level between
economic
control
𝑥𝑥 ∗Growth
level isx* is public
obtained growth,
explanatory
obtained debt
byby hence
variaband pu
hlied
usirep.fiscal
alances 5 inDomesticdeficits.
a(2015)
number
and 𝑡𝑡 =
SACU interest
offurther
𝑦𝑦financing with
𝛼𝛼studies.
0payments
+
Revenue𝛼𝛼1 𝑥𝑥𝑡𝑡acosts
For+ significant
gap.
notes 𝛼𝛼instance,
𝑑𝑑(𝑥𝑥
Growth associated
arrears
Domestic
𝑡𝑡that
requirements decline
− 𝑘𝑘)Figure +the +
1: in
𝛽𝛽 ′ 𝑍𝑍𝑡𝑡spilledwith
higher
𝜀𝜀Fiscal SACU
𝑡𝑡paymentsover
(1) high values
revenues.
the
Balances into value. borrowing
arrears
the
optimal
and andThe
private
𝜕𝜕𝑦𝑦 SACU zero
debt sector.
Revenue otherwise.
threshold,
of one variable the
for 𝑍𝑍 isstudy a
observed
greater vecto em tha
dynamics. The general empirical eq.𝑑𝑑 is
𝑡𝑡threshold a𝛼𝛼 dummy model Tovariable 𝑡𝑡 that the(assum
2
he 0increase in the thegovernment borrowing between from public debt and growth
−4zero Theoretical using
determine threshold
Model between poten
) apu
Per cent
ulated to estimate
more rise, hence
than 5crowding
debt-to-GDPper cent out
ratio the
ofbeyond GDP privatewhichin solving
sector.
a rise solving
in eq.
The debt(3)
values as (3) andas
follows; follows; otherwise. 𝑍𝑍 3is
t since
al. (2014) 2010 to a threshold failure level to secure
of Rates international
the solving optimal funding
eq. (3) debt
as tofollows; =
corresponding
close
threshold, 1the 2𝛼𝛼
the 2 𝑥𝑥the =
𝑡𝑡optimal 0study
vector debt of
employs 𝑡𝑡 coefficie
thresh mv a
omestic
to -5more
rest costs markets
than accumulated
5
associated caused
requirements
persituation cent the
withto was
The
of lending
spilled
moreGDPfurther
increase
high over
than
in rates
exacerbated
borrowing in5into
the togovernment
per the cent by privateSwaziland's
contains
threshold of GDP 𝜕𝜕𝑥𝑥
borrowing
sector.
𝑡𝑡 control
in
type from explanatory
nonlinear values model. variables.
and zero This oth
15 reduces isborrowing
output specified growth. as follows:
Similarly, Wright of and one for observed variablebetween values greater
public dynamics. of debt than
the and T
he
due to-GDP domestic
toratiothebeyond increase
markets
government's which
failure
incaused
the
afinancing
rise
1,to in
secure𝑖𝑖𝑖𝑖the
inability
debt(𝑥𝑥 lending
gap. costs
𝑡𝑡 > 𝑘𝑘)
international todynamics.coupled
payrates towith
Domesticthreshold funding
The
payments
∗to
the
𝛼𝛼
general
𝛼𝛼1 explanatory
contains
type
close the
empirical
arrears
nonlinear controlTo determine
threshold
threshold
variables.
explanatory
model.
the𝜀𝜀𝑡𝑡potential
type This ismodela variab
random
nonline
mode
no re
2016
%
increase
GDP
(2014)
crease in
examine
inthreshold
the the government
the relationship
2011 due borrowingto the 100 from
government's Graphically according inability
values
Graphically the above and to
the to whether
pay
zero
relation
above the optimal
contains
otherwise.
between
relation public
𝑡𝑡
5 debt
between 𝑦𝑦 control 𝑍𝑍
anddebt
threshold,
𝑡𝑡
is
𝑥𝑥
𝑦𝑦 = a is 𝑡𝑡 ab
expla 0 de
vec
𝛼𝛼 +th
centborrowing
threshold variable costs
and k coupled
is a given with
threshold
Research the
Bulletin Volume 2
SACU Growth
Per cent
variable and
The 𝑘𝑘
general is a given
empirical threshold
𝑦𝑦according threshold model
𝛼𝛼1to whether public
according 𝛽𝛽asdebt
′ explanatory 𝑡𝑡to 𝜀𝜀is whether
above
(1) variable or
lpublicBank of Swaziland
debt andvalue. growth
rise, d©is 2018
hence a dummy
using threshold
crowding
5
variable that50the
out assumes𝑡𝑡 =private 𝛼𝛼0 +
can be and 𝑥𝑥𝑡𝑡 +
represented 𝛼𝛼
can 𝛼𝛼be 2 𝑑𝑑(𝑥𝑥 as𝑡𝑡follows:
represented − 𝑘𝑘) + allows 0𝑍𝑍 𝑡𝑡 +the
follows: 𝑡𝑡nonlinearexistence
perdomestic cent ofvalue.
GDP
to
markets almost
in 2011/12.
caused 13 per
suppliers the Without
cent lending
in of
time. GDP rates
The in 2011/12.
to
fiscal deficit ∗sector.
Without
also 1increasedThe threshold type (2 m
Per cent
higher
ntral term
Bank with
15
ofa 𝑑𝑑zero
Swaziland
value -5
is is2011
aof mean
dummy
specified
one © due and
2018
for as to
variable a
observed
Central constant
the
follows: that
Bank government's
assumes
values
of variance.
Swaziland of
can
a
be
value
the © below
represented
inability
2018Fig. 𝑥𝑥contains
2. A a
quadraticthreshold
as
= to paycontrol explanatory follows: relation value.
corresponding
between This 𝑦𝑦 and framewo
variables
𝑥𝑥 ( 4vector
)
ments to expenditure to match 0 the declined 0
below 1, 𝑖𝑖𝑖𝑖
aFigure (𝑥𝑥2𝛼𝛼
threshold 𝑡𝑡 > 2 𝑘𝑘) value. below according
This -5 to whether
s. The general empirical threshold model theabetween framework
threshold ofvalu
has
2008 2012
2009 2013
2010 2014
2011 2015
cent 2016
a hencegexpenditure
e Estimating 10threshold
increase variable into the P greater
almost a g e 13 than
borrowing per cent
𝑑𝑑 the
= costs {of GDP
given coupled in 2011/12. with2. A Without
quadratic
the } allows relation existence tw
pr
o
rowing crowding adjustments
to -10
eq. match
out (1) the the
to
repeatedly expenditure
declined
private withsector. to
different match
The the declined
2010
2006
2007
2008
2009
2011
2012
2013
2014
2015
2016
2013
2014
2015
2016
2014
2015
2016
ates
tic
th to
ratedebt
-10
that minimizes
continued
contains the residual
control
Deficit/surplus rising
debt
threshold
explanatory
as %
𝑑𝑑sum= { of squares
since
rose
GDP 0, on 2010variable
the
variables. Where 𝑡𝑡
(RSS).
back to 𝛽𝛽
𝑂𝑂𝑂𝑂ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 ′ and
is
}𝑦𝑦𝑡𝑡 is𝑘𝑘𝑦𝑦the
of
the high
isfiscal adependent
given
deficits.
threshold
been Proanoin
variable,
applied et𝑥𝑥 a al.
𝑡𝑡number
(2014)
is theofest st
e ε
-15
t is a random disturbance
revenue flows, public term
Mafusire debt,the
with zero debt-to-GDP
particularly
(2015) further domestic notesratiothat beyond the
©the
debt-to-GDP
which a rise
higher value.
ratio
in𝑑𝑑debt is
be a
bt
or.es ates continued
Thethat 1, the Domestic
𝑖𝑖𝑖𝑖
mean
𝑑𝑑 = {as % corresponding
rising
higher (𝑥𝑥
and > 𝑘𝑘)
a
Where}vector
𝑡𝑡 since
debt
constant
𝑦𝑦𝑡𝑡value.
Domestic is the2010
continued
variance.
of 𝑑𝑑dependentis to a
coefficients
debt
rising
Estimating
dummy
threshold
continued of variable
variable, since
Central
the risingreduces
variable
𝑥𝑥 2010
Bank
𝑡𝑡e is since that
the
of to
Swaziland
output
assumes
and
2010 𝑘𝑘to growth.
is
Proano
2018
a a the
value given
et debt-to-GDP
al. Similarly,
(2014) threshold ratio
estimate Wrig be
r 10 per
cit/surplus
0, cent
GDP
eq.in 2016. debt rosewith
𝑂𝑂𝑂𝑂ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
(1) repeatedly on different
theinterest back values ofreduces
costs high P a
of k,associated
g
fiscal output deficits. with growth. high reduces Similarly,
borrowing output Wright growth and
ith the
r further
cent in 2016. register 10 per cent in 2016. reduces output of one growth for
th is high
the
However,borrowing
notes
explanatory
chosen
dependent
some that
in studies
threshold
ascending
variable, the
variables. employ
of
register
𝑥𝑥
variable
higher
𝜀𝜀 one
order,
is𝑡𝑡 is 10
the
a quadratic
and
random
the
per for 𝑘𝑘
optimal
cent is model
observed
value.in a
disturbance given
value
2016.𝑑𝑑 is threshold
Grenade
avalues
dummy of (2014)
variablethe the
thresholdthat examine
debt-to-GDP assumes ratiothe
a valuebeyond relat
𝑡𝑡 Domestic 𝑡𝑡 debt continued requirements rising 9 | spilled
sinceover 2010 into tothe private sector.
1:thatFiscal the higher
to obtain of the
Balances k isthresholdobtained
value.
and 𝑑𝑑
SACU abydummy
islevel. finding
For example,
Revenue theGrowth
variable valuethatGrenade
Pattillo that
assumes a value(2014) examine Grenade Grenade the (2014) relationship
variableexa
(2014) gre
exa
odssociatedthe private with sector.
ishigh borrowing between public reduces
debt and output growth. Sim
lhigh
variable
Balances
and 𝑘𝑘
minimizes a giventhe
register threshold
variable
residual
10 per sumcent ofgreater
The squares
invalues of SACU
2016.
increase than
oneand
(RSS). for
inRevenue the
the given
observed
government threshold values
borrowing level
of from the threshold th
growth using
et al. and
borrowing Figure
(2011) SACU of 1:one
examine Fiscal
Revenue fora Balances
Figure 1:Growth
observed
nonlinear Fiscal and
Balances
and hump- of the
between public debt and
SACU
threshold Growth
Revenue Growth
between between
growth(2014) using
public public values
threshold
debt debt and an
and
ent s a dummy variable
borrowing fromthat assumes Rates a value Grenade examine
lled
e private over into
sector. Rates
However, the private
some1:studies sector.
values and thethe zero
avariable
domestic otherwise.
markets dynamics.
greater 𝑍𝑍Growth
caused𝑡𝑡than is The a the
the vector general
lending given which
rates empirical
thresholdto The threshold
level
shaped linkagevariable between
Figure Fiscal employ
greater
external thandebt
Balances and quadratic
given
andgrowth SACU threshold Revenue level dynamics. 𝑥𝑥contains general coe
for observed model valuesto ofobtain the threshold the threshold level. dynamics. For The general 𝑥𝑥 empirical
∗
betweendynamics. public thresholdThe debt general
and mode grow e
the
00he
borrowing lending
government
by using from rates
a quadratic to
borrowing
Rates
values and zero
contains
function from
et200 rise,
otherwise.
al.model. control hence
The
15
𝑍𝑍
values 𝑡𝑡 non- is crowding
a vector
explanatory
and zero is specified
outotherwise.
which the private
variables. as follows:
15 ′
sector.
𝑍𝑍𝑡𝑡isis
𝛽𝛽 isthe aThe
specifiedvectoras which
correspondi
follows:
greater than example, the given 200Pattillo
threshold level 15(2011) examine ∗ 15 dynamics. The general empiri
50 lending rates a to nonlinear contains and control hump-shaped
150 explanatory 10 is
linkage specified
variables. Where
𝛽𝛽 ′
is asthe follows:
𝑥𝑥 is the 10 threshold is specified level as
beyond follows: which
rkets
nd private linear
zerocaused sector.
otherwise. model the𝑍𝑍The assumes
lending
𝑡𝑡 150is a vector arates quadratic
200 corresponding
which to10 increase
form of
contains in the
the
vectorcontrol borrowingWhere
of𝑦𝑦 =coefficients x*costs
explanatory
1510
is coupled
the threshold with
of𝛼𝛼variables. the thelevel beyond ′
𝛽𝛽+explanatory
is𝑥𝑥′ 𝑍𝑍the
00vate sector. The between external debt 100 and growth 5by using increasing
which 𝛼𝛼
increasing 𝑥𝑥+ 𝛼𝛼
reduces 5 𝑥𝑥 is + specified
reduces 𝑦𝑦 . On 𝑑𝑑(𝑥𝑥𝑦𝑦the .= as −
On follows:
𝑘𝑘)
other
𝛼𝛼 the + 𝛼𝛼𝛽𝛽
hand,
other 1 𝑡𝑡 + 𝑡𝑡if+𝛼𝛼 𝑥𝑥𝑡𝑡2𝜀𝜀𝑑𝑑𝑡𝑡
2015 SACU Growth
Per cent
a quadratic function
150 𝛽𝛽 ′ ismodel. Central
The Bank of Swaziland © 2018
non-linear ′
ts coupled
ding control out the with
explanatory private the sector.
variables.
100
explanatory the
The P a g evariables.
5
corresponding 𝑦𝑦𝑡𝑡 =𝜀𝜀 𝛼𝛼is is+
𝑡𝑡hand, a 𝛼𝛼 1 𝑥𝑥𝑡𝑡𝑥𝑥+
random
vector
if
105
the𝛼𝛼20disturbance
∗is below
of𝑑𝑑(𝑥𝑥x* − 𝑦𝑦𝑡𝑡𝑘𝑘)
𝑡𝑡coefficients
the =increase
+𝑥𝑥𝛼𝛼𝛽𝛽0 + 𝑍𝑍𝑡𝑡 in 𝛼𝛼+of 1 𝑥𝑥 𝜀𝜀𝑡𝑡 + isthe (1)𝛼𝛼to
2 𝑑𝑑(
SACU Growth
50 0 below increase in
50
coupled with the 0 𝑡𝑡 is expected
Per cent
Per cent
model explanatory
assumes 100 a variables.
quadratic 𝜀𝜀𝑡𝑡form of thedisturbance 5
is a random 𝑦𝑦𝑡𝑡 .= 𝛼𝛼0 + 𝛼𝛼1 𝑥𝑥𝑡𝑡 + 𝛼𝛼2 𝑑𝑑(𝑥𝑥𝑡𝑡 − 𝑘𝑘
nding vector of coefficients of 0the 0 expected to 0 increase
Growth
50 -5
cent
2010 2007
2008
2009
2011
2014 2012
2015 2013
2016 2014
2015
2016
2007
2008
2009
2011
2012
2006 2013
2014
2015
2016
-5 1, 𝑑𝑑 = 𝑖𝑖𝑖𝑖 { 1,}
50 𝑡𝑡 -10 0,
SACU
zilandThe
© 2018 0
-15 0, (𝑥𝑥𝑡𝑡 >𝑂𝑂𝑂𝑂ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
𝑘𝑘) 𝑖𝑖
9
2010
2007
2008
2009
2011
2012
2013
2006
2014
2007
2008
2016
2011
2012
2013
2014
2008
2009
2011
2012
2013
2015
2016
𝑡𝑡 𝑡𝑡 𝑡𝑡
Bank Of Swaziland Where© 2018 𝑑𝑑 𝑦𝑦 is the
00 -50
SACU Growth rate
-15
Deficit/surplus as 0,% GDP 𝑂𝑂𝑂𝑂ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒=𝑡𝑡 {0, 0, depen
𝑂𝑂𝑂𝑂ℎ
(1). However,
SACU Growth rate 𝑥𝑥
-100𝑡𝑡 and 𝑥𝑥 2
are now
Deficit/surplus
-100
Source: SACU Member interpreted
-15 as States
% GDP as Where 𝑦𝑦
Whilst the-15 𝑡𝑡two approaches are equally important,𝑥𝑥𝑡𝑡
-15 is the dependent variable,
threshold variable and
debt is deterrent to economic growth. debt v
Consequently, the nonlinear quadratic the p
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
relationship between public debt and economic respe
growth can be represented as follows; perce
4.1 Public Debt Threshold Model 𝑦𝑦̂𝑡𝑡 = 𝜃𝜃𝑦𝑦̂𝑡𝑡−1 + 𝜑𝜑𝑞𝑞̂𝑡𝑡 + 𝛼𝛼𝑟𝑟̂𝑡𝑡 + 𝛿𝛿𝑦𝑦̂𝑡𝑡∗ + 𝜌𝜌𝑑𝑑̂𝑡𝑡 + 𝜎𝜎𝑑𝑑̂𝑡𝑡2 5. Est
Whilst the two approaches are equally
+ 𝜇𝜇𝑡𝑡 (5)
important, our study adopts the nonlinear The s
model of the quadratic form in line with Where Where 𝑦𝑦𝑡𝑡 ,yt𝑞𝑞, q𝑡𝑡 ,t, r𝑟𝑟𝑡𝑡t and
andyt𝑦𝑦* 𝑡𝑡∗are
arecontrol
control explanatory
explanatory
Pattillo et al. (2011). This approach is variables, which are explained above. The estima
variables, which are explained above. The public
preferred compared to the other approach public debt variable dt is the threshold robus
as represented in Eq. (1) because of its debt variable
variableand 𝑑𝑑𝑡𝑡 isa the crucial variable
threshold variablein and
our a
simplicity. However, unlike in Pattillo et al. analysis. The following θ, φ, α, δ, ρ and σ possib
crucial variable in our analysis. The following
(2011) who employ a panel data set of 93 are parameters to be estimated and μt is impac
developing countries, our study estimates 𝜃𝜃, 𝜑𝜑, the𝛼𝛼, error
𝛿𝛿, 𝜌𝜌 and term andparameters
𝜎𝜎 are is assumed to estimated
to be have a
a quadratic equation for one country, zero mean and a constant variance. The affecte
and 𝜇𝜇𝑡𝑡 is the error θ of the term and is assumed
lagged to have a
which is Swaziland. While a number of parameter one-period output
(Patti
studies investigate the nonlinearity of the zero gap meanis expected and ato constant be positive indicating
variance. The
relationship between public debt and growth output persistence. The real exchange rate Gener
by augmenting growth models with public parameter parameter θ ofφthe is one-period
expected to lagged output gap
be positive.
estima
debt, our study examines this relationship is This is the to
expected casebebecause positivean indicating
increase inoutputthe
from the demand side of the economy. real exchange rate, a real depreciation, that
Hence, the study uses the aggregate demand persistence.
improves The the external real exchange rate parameter
competitiveness of
hetero
equation as in Morón and Winkelried (2005) 𝜑𝜑 isdomestic expected exports to be leadingpositive. to This
an increase in
is the case
for Latin American countries. The proposed output growth. values
aggregate demand relates the output gap yt because an increase in the real exchange rate, a
instru
as a function of the one-period lagged output realThe depreciation,
coefficient α of improves the real interest the rate is
external
gap yt _1, the real interest rate rt, the foreign expected to assume a negative sign implying enable
output gap yt* and the real exchange rate qt. CentralthatBanka rise in the real
of Swaziland © 2018interest rate reduces
investment,
Page hence a reduction in economic
The nonlinearity relationship between growth. The sign of the coefficient δ of the
output growth and public debt is introduced foreign output gap is expected to be positive.
in the above aggregate demand model by An increase in the foreign output gap leads
including the squared term of the public to an increase in demand for domestic
debt d_t^2 and the linear term of the public goods and services, which in turn causes
debt dt.The combination of the linear and domestic output to increase. For a nonlinear
nonlinear terms of the public debt implies relationship to exist between the public debt
a hump-shaped relationship between and growth, hence an optimal threshold debt
public debt and growth. The turning point level, the signs of the parameters of the
of the quadratic equation determines the linear debt variable and the squared term
optimal threshold level of the public debt variable of the public debt must be positive
above which debt is deterrent to economic and negative, respectively. The hat above
growth. Consequently, the nonlinear the variables denotes percentage deviation
quadratic relationship between public debt from the steady state.
and economic growth can be represented as
follows;
quadratic form between debt and growth. one-period lagged depreciation increases
In this regard, we extend the aggregate exports leading to a 0.03 per cent increase
demand function by Moron and Winkelried in output. The coefficient of the interest
(2005) for Latin American countries by rate is negative at 0.35. This suggests that
adding the linear and squared terms of the a 1 per cent rise in the interest rate reduces
public debt-to-GDP. This formulation allows aggregate demand, which causes output
us to investigate the existence of a quadratic to fall by 0.35 per cent in the following
nonlinear relationship between public debt period. The parameter of the foreign output
and economic growth. Hence, the study as proxied by South Africa’s output has a
employs a quadratic nonlinear threshold positive value of 0.12. This means that a
model as in Pattillo et al. (2011). To control 1 per cent increase in the foreign output
for the possibility of endogeneity, the paper causes a 0.12 percent rise in the domestic
uses the Generalized Method of Moments output by increasing demand for domestic
(GMM) estimation technique. exports.
The GMM results confirm the existence Our findings, particularly with respect to
of a nonlinear hump-shaped relation the optimal public debt threshold level,
between public debt and economic growth suggests that Swaziland needs to adopt
in Swaziland. This is consistent with the debt-management policies that will allow
findings of Pattillo et al. (2011) for 93 the country to gradually move its public debt
developing countries and Bawa et al. (2016) level towards the estimated debt threshold
for Nigeria. Our empirical results suggest to support growth while also ensuring debt
that a 1 per cent increase in public debt-to- sustainability. However, this will yield
GDP has a short-run positive impact of 0.22 positive results provided that the debt funds
per cent and a long-run negative effect of are used to fund viable capital projects. The
0.24 per cent on growth. The optimal level preferred capital projects are those that
of public debt above which an increase in will have greater growth multiplier effects
debt reduces economic growth is estimated through usage of mainly local inputs, or at
at about 46 per cent of GDP. This threshold worst imports from SACU countries in order
is higher than the 35 per cent limit set by to earn the country higher SACU receipts.
government and the World Bank debt critical This would call upon government to identify
ratio of 40 per cent for developing countries. such projects in consultation with relevant
The estimated debt threshold for Swaziland stakeholders.
is also within the SADC convergence criteria
of a public debt limit of less than 60 per cent We therefore recommend that government
of GDP. Using the quadratic form relation, put in place policies that will ensure that
Mupunga and le Roux (2015) find an optimal debt accumulation, be it external or
growth-maximising public debt threshold of domestic, is consistent with the country’s
about 48 per cent of GDP for Zimbabwe. growth objectives. The recently established
debt management unit, which is highly
The coefficients of the control variables commendable, should spearhead and
exhibit the correct signs in line with theory expedite the formulation of such policies.
and highly statistically significant. The Sound communication and cooperation
parameter of the one-period lagged output between the ministry of finance and
is positive indicating a 0.81 per cent degree the central bank should be maintained
of output persistence. The exchange rate in managing public debt. In this regard,
coefficient has a positive value of 0.03. the central bank may have to maintain a
The results show that a 1 per cent of a positive interest rate differential with its
Bakar, A., Hassan, S. (2008). Empirical Eberhardt, M., Presbitero, A. (2015). Public
evaluation on external debt of debt and growth: Heterogeneity and
Malaysia. International Business & non-linearity, Journal of International
Economics Research Journal. Economics. 97, 45-58.
Mupunga, N., le Roux, P. (2015). Estimating Proano, C., Schoder, C., Semmier, W.
the optimal growth-maximizing (2014). Financial stress, sovereign
public debt threshold for Zimbabwe, debt and economic activity in
Southern African Business Review, industrialized countries: Evidence
Vol. 19 No. 3. from dynamic threshold regressions,
Journal of International Money and
Padoan, P., Sila, U., van den Noord, P. Finance, 45, 17-37.
(2012). Avoiding debt traps: financial
backstops and structural reforms, Reinhart, C., Rogoff, K. (2010). Growth in
OECD Economics Department Working a time of debt, American Economic
Paper, No. 976. Review: Papers and Proceedings,
100(2): 573-578.
Palley, T. (1994). Debt, aggregate demand,
and the business cycle: An analysis Semmler, W., Sieveking, M. (2000). Critical
in the spirit of Kaldor and Minsky, debt and debt dynamics, Journal of
Journal of Post Keynesian Economics, Economic Dynamics & Control, 24,
Vol. 16, No. 3, 371-390. 1121-1144.
Pattillo, C., Poirson, H., Ricci, L. (2002). Saint-Paul, G. (1992). Fiscal policy in an
External debt and growth, IMF endogenous growth model, Quarterly
Working Paper. Journal of Economics, 107 (4), 1243-
1259.
Pattillo, C., Poirson, H., Ricci, L. (2011).
External debt and growth, Review of Wright, A., Grenade, K. (2014). Determining
Economics and Institutions, Vol. 2, No. optimal public debt and debt-growth
3. dynamics in the Caribbean, Research
in Applied Economics, ISSN 1948-5433,
Vol. 6, No. 2.
currencies to the South African Rand. as the sum of private short-term capital
Lesotho phased out the discount rate and outflows and errors and omissions.
adopted the Lombard rate pricing it above
the 91-day TB rate and the lending rates in Section1 is the brief introduction, section
Lesotho respond very strongly to changes 2 is the literature review, section 3 is the
in the repo rate in South Africa (Sylvanus model, data and methodology. Section 4 are
Ikhide, 2010). Lesotho central bank has just the empirical results and section 5 is the
recently reverted back to setting interest conclusion.
rates with reference to South Africa’s repo
rate. Thus due to the lack of data where 2.0 LITERATURE REVIEW
monetary policy is actively set by the central Reviewed literature suggest that interest
bank of Lesotho, Lesotho is not included in rates are an important determinant of
the study. capital flight which is often measured as
portfolio investment assets Puah etl. (2012),
Sylvanus Ikhide, (2010) observed that lending Folorunso S. Ayadi (2008) and Dooley (1998).
rates, level of prices and money supply in The are many determinants of capital
the monetary union respond instantaneously outflows in literature and the following
to changes in the discount rate by the South critical factors will be discussed for the
African reserve bank with Namibia’s repo rate cases of Namibia and Swaziland: (1) the GDP
responding sharply compared to Swaziland’s growth rate differential (2) the fiscal deficit
as their central banks actively set interest (3) financial development (4) interest rate
rates. He confirms that South Africa discount differential.
rate is the relevant policy instrument for the
LNS countries. He concludes that due to the The following authors discuss these factors
asymmetry of shocks the LNS countries might in the context of capital flight; where capital
be able to undertake independent monetary flight is measured by portfolio investment
policy by setting their central bank discount assets as obtained in the balance of payments
rate lower than South Africa’s to stimulate statistics.
the economy.
Puah and et al (2012) employed a vector
The question that this study will therefore error correction model (VECM) to measure
address is whether the discount rate capital flight on macroeconomic variables
differential pursued by Namibia and for Malaysia using time series data from
Swaziland in their conduct of monetary policy first quarter of 1991 to the fourth quarter
do impact on international flow of portfolio of 2008. He explained capital flight as
investment assets. Many studies in literature errors and omissions and short-term capital
have supported the inflow of capital as a pre- outflows. The results showed that capital
requisite for economic growth with Lisa M. flight is determined by real GDP, the budget
Schineller (1997) extending the concerns to deficit, the treasury bill rate, foreign direct
the degree of economic, social and political investment and the stock market. All the
fracture. Most researchers address the variables were found to have the right signs
importance of capital inflows for economic and statistically significant. The treasury bill
growth under studies on capital flight. The rate was found to be negatively related to
current research therefore borrows most capital flight. He concluded that capital is
of its literature review from studies on expected to flow to developing countries
capital flight in developing countries. The in being attracted by a positive rate of
balance of payments approach popularised return on capital. His paper also finds that
by Cuddington (1987) define capital flight improvements of in the real GDP and budget
deficit stemmed capital flight. He linked real inflation and political stability. He finds
GDP growth to investors’ confidence and the the Real GDP to be negatively related to
improvement in the budget deficit to less capital flight and GDP growth to significantly
distortion in the economy as a conditions reduce capital flight both in the short and
to stall capital. The study shows that the long run. High economic growth brings about
financial markets liberalization and the opportunities for investment that discourage
deregulation of the international movement capital flight.
of capital in Malaysia resulted in an improved
stock market causing capital flight. 3.0 MODEL, DATA, AND
METHODOLOGY
Boyce and Ndikumana (2012) also measures The time-series data used in this study are
capital for 30 sub-Saharan African from 2010 Q1 to 2015 Q4 and sourced from
countries, including 24 countries classified the Central bank of Swaziland Quarterly
as severely indebted low-income countries Bulletin and the Bank of Namibia Quarterly
for the period 1970-1976. He finds that the Bulletin. Because the time-series is short the
difference between domestic interest rates data is pooled to solve the problem of a few
and US interest rates to have the expected degrees of freedom and to also allow for the
negative sign in both a pooled and cross- analysis of lagged values (Leonce Ndikumana,
sectional regression but in neither cases was 2010). Quarterly GDP data are not available
it significant at 10 per cent. Higher interest and are obtained by interpolating the yearly
rates in the sub-Saharan African countries GDP data.
than in the US curb capital flight. He found
the fiscal position to have an ambiguous The generalised least squares cross section
relationship with capital flight. Citing the weights is used to estimate the parameters
lack of accurate fiscal position statistics, of the model in order to solves the problem
he could therefore not reach any firm of heteroscedasticity, serial correlation and
conclusions. Boyce and Ndikumana (2010) normality in the data to avoid obtaining
found that an unfavourable foreign and parameters that are not best linear and
fiscal position are among causes of capital unbiased (BLUE) (Marius Ooms, 2007;
flight. FitzGerald (1997) also argues that Ruppert D and Carrol R,1998; Wooldridge,
panic sets in when foreign exchange reserves 2002).
levels fall in periods of high fiscal deficits
and investments move to countries with The foregoing literature review measures
better fiscal and foreign exchange reserves capital flight as portfolio investment assets.
positions. Most of the variables used in the literature
to explain capital flight are used in this
Folorunso S. Ayadi (2008) when employing study to explain portfolio investment assets.
ordinary least squares (OLS) and error The variables to use in the model are; (i) the
correction model (ECM) for Nigeria supports GDP growth differential (ii) the discount rate
the argument that higher domestic interest differential (iii) financial development and
rates than foreign interest rates reduce the (iv) the fiscal position. Portfolio investment
outflows of portfolio investment assets. In assets are expressed as a proportion of
fact, he finds interest rate differential to be GDP. The GDP growth differential and the
the most significant determinant of capital discount rate differential is expressed in per
flight in Nigeria. The other explanatory cent form, financial development is defined
variables he uses in the estimation of capital as M2 divided by GDP and the fiscal deficit is
flight are the total debt stock, exchange expressed as a per cent of GDP. The following
rate, real GDP growth, the trade balance, is the model:
Table 2. Panel Regression Model Results/ fiscal position are found to be statistically
Corrected for Heteroscedasticity and significant. The GDP growth rates for
Autocorrelation Namibia and Swaziland increase above that
Dependent variable: GDPP (Portfolio Investment Assets) of South Africa leads to an improvement in
Independent variable Coefficients t-Statistics Prob. the level of portfolio investment assets in
Constant -0.000138 -0.035593 0.9718 the respective economies. Low GDP growth
GDP growth (-2) -0.002296 1.82193 0.0659 encourages portfolio investment assets
Discount rate differential -0.00390 0.386676 0.7011 outflows (Ndikumana and Boyce 2000; Pastor
Financial development -0.027398 -2.068415 0.0453
1990; Nyoni, 2000).
Deficit -0.001073 -2.514358 0.0162
Observations; balanced 44
panel
The discount rate differential increase i.e.
R squared 0.22386 when the Namibia and Swaziland discount
DW 1.547792 rates are higher than South Africa’s discount
F-statistics 2.804493 rate the level of portfolio investment
assets outflows reduces but insignificantly.
Table 3 presents the final model arrived at In 10 attempts only 3 are likely to attract
by dropping the insignificant discount rate portfolio investment assets in a sample of
differential which is shown in table 2. The 44 observations. Investors are assumed
chances of the discount rate differential to seek to maximise profits and would
to influence portfolio investment assets is allocate funds between domestic and
merely 30 per cent or rather 3 out of 10 foreign financial markets, placing the funds
chances. to where the returns are high. Risks and
returns to investment based on portfolio
choice determine the outflow of funds. The
Table 3. Panel Regression Model without higher the discount rate as a proxy for the
the Discount Rate Differential returns on financial assets should attract
Dependent variable: GDPP (Portfolio Investment Assets)
more investment.
Independent variable Coefficients t-Statistics Prob.
Constant 0.000798 -0.232295 0.8175
The low level of development of the financial
GDP growth rate (-2) 0.002136 1.899524 0.0647
sector in particular in terms of diversification
Financial development -0.023741 -2.615909 0.0125
in the choice of financial products compared
Deficit -0.001024 -2.547399 0.0148
to South Africa in both Namibia and
Observations; balanced 44
panel Swaziland discourages portfolio investment
R squared 0.221093 in two countries vis-à-vis South Africa. Risk
DW 1.512939 aversion has driven portfolio investment
F-statistics 3.784663 assets in Namibia and Swaziland with fund
managers indicating that the discount rate
Most of the independent variables differential has some-what affected their
significance improves under the panel decision to either place the funds locally or
regression model. The final model in table 3 internationally particularly in South Africa.
with a larger F-statistics is selected (Kadane This happens within the confines of the
and Lazar, 2004) after the correction of serial local compliance with legislation by fund
correlation, heteroscedasticity, normality managers, the pension fund managers and
and the dropping of the insignificant variables other financial institutions. The insurance
to remain with the final specification of the companies such as the Swaziland Royal
model in table 3. As opposed to the discount Insurance Corporation, Aon and Metropolitan
rate differential the GDP growth rate mostly engage fund managers for the
differential, financial development and the investment of their funds.
Commercial banks besides interest rate defended in the spirit of Gelbard and Leite
risks which they manage within defined (1999).
risk limits while integrating the impact of
liquidity risk maintain current accounts in The effects of financial development in
South Africa for the purpose of facilitating literature has been mainly found to be mixed
customer demand for South African Rands. owing to the challenged definition of financial
The commercial banks have been influenced development. Gelbard and Leite (1999) in a
by the discount rate differential between study on measuring financial development
South Africa and Swaziland and excess funds in 38 Sub-Saharan Africa, found that even
are naturally placed in the accounts giving though there has been improvement in
better yields while taking into account the financial development in Sub-Sahara Africa,
need to meet daily customer obligations. The in many countries the range of financial
money market portfolio consists mainly of products remains extremely limited. Interest
Swaziland Treasury Bills for both compliance rate spread are wide, capital adequacy ratios
and investment purposes and the balanced are insufficient, judicial loan recovery is a
fund is then split between the Rand current problem, and the share of nonperforming
account and central bank call account. The loans in large. The composition of M2 is
fund managers and commercial banks also critical in determining the influence of
prefer to invest in in South Africa because the financial development when it is measured
market has a variety of highly liquid financial as M2/GDP. In the same study Gelbard and
products which can be readily tapped into Leite (1999) further note that financial
when the need arises. These allocation development is associated with increased
needs shroud the effect of the discount rate importance of term deposits and the ratio
differential on portfolio investment assets. proportion of cash in M2 is dominant which
assets are susceptible to be invested across
Financial development is found to encourage the border. The commercial banks naturally
portfolio investment assets outflows, which place excess funds in better yielding accounts
could be least expected. The development and to meet daily customer obligations.
of the financial sector has to present
expanded and varied opportunities for The fiscal deficit is found to encourage
domestic portfolio diversification (Puah and portfolio investment assets like in the
Liew, 2012). The financial development in studies by Puah and Lewis (2012) and Leonce
Namibia and Swaziland are not diversified and Boyce (2002) as earlier mentioned. A
enough to withstand the attraction of funds percentage point improvement in the fiscal
to the South African financial sector. Collier, deficit results in an outflow of 0.102 per
Hoeffler and Pattillo (2001) found that cent of GDP in portfolio investment assets
financial development is not statistically and a per cent improvement in financial
significant in driving portfolio investment development leads to an outflow of 0.024
assets when using M2/GDP. They also find per cent of GDP in portfolio investment
financial development to have an incorrect asset. For fear of future taxation due to
expected sign hence they drop it out of the central government running successive
the model specification. But due to the deficits economic agents change their
ambiguities of the influence of financial holding of local financial assets to foreign
development on portfolio investment exchange denominated financial assets by
assets flows the variable is not dropped but transferring their funds to South Africa.
Michael P. Dooley., 1998, Capital Flight A Washington DC: The IMF Institute, 256-
response to Financial Risks, External 99. Olopoenia, R, (2000).
Adjustment Division IMF working
paper. Paul Collier, Anke Hoeffler and Catherine
Pattillo., 1999, Flight Capital as
Manuel Pastor., 1990, Capital Flight in Latin a Portfolio Choice; IMF working
America, World Development, 1990, Paper/99/177.
vol. 18. Issue 1, pp 1-18.
Sylvanus Ikhide and Ebson Uanguta., 2010,
Nyoni. T., 2000, Capital flight from Impact of South Africa’s Monetary
Tanzania, in I. Ajayi and M.S. Policy on LNS Economies, Journal of
Khan(eds), External Debt and Economic integration, 25(2), June
Capital Flight in sub-Saharan Africa. 2010; 324-352.
25000000 15.00
in the above
standing primary and secondary
60 percent sectors.
of total exports (CBS The
projected deceleration in output is mainly
20000000 Quartely, December 2016).
informed by the drastic effects of the
% growth rates
10.00
15000000 drought
Going forward,experienced by the iscountry
real GDP growth expected into2016,
E'000
60
performance
The of the economy continued
poor performance of the into the
economy has always been lower than recurrent 2014/
40
continued
turn of the into the turn with
millennium, of thegrowth millennium,
rates expenditure over the years, and has never reach
with growth rates averaging
averaging 2.4 percent between 2000 and 2016,
2.4 percent 20
exceeded 40 per cent, as shown in the figure budge
between 2000 and 2016, and reached a high 0
and reached a high of over
of over 3.5 per cent in 2007. This growth 3.5 per cent in 2007. below. This t
2013
1980
1983
1986
1989
1992
1995
1998
2001
2004
2007
2010
2016
was
This driven
growth was by driven
the manufacturing
by the manufacturing sector, Figure 2: Capital and Recurrent Expenditure as a financ
which is the main growth
sector, which is the main growth engine and in engine and in Percentage of total Expenditure
Capital Expenditure (% of Total
turn encouraged rapid growth in supporting expenditure)
1.1
turn encouraged rapid growth in supporting
sectors such as construction. Manufacturing Gener
sectors rangesuch as construction. Manufacturing Recurrent Expenditure (% of Total
entities from small factories engaged
Expenditure) have a
inentities
light rangeindustry to large
from small factories ones endowed
engaged in
with the latest technology and producing Source: Ministry of Finance
Source: Ministry of Finance of th
light industry to large ones endowed with the
highly sophisticated goods which, given Just like other countries, Swaziland experienced a conce
latest technology and producing highly
the small size of the domestic market, are Just like other countries, Swaziland chann
sophisticated goodsfor which, severe financial crisis in 2009/10, which had
destined mainly thegivenexport the small
market.size of
The experienced a severe financial crisis in
been looming since 2008 with massive declines in which
sector’s
the domestic contribution
market, areto exportmainly
destined earnings
for the has 2009/10, which had been looming since
improved recently, there
export market. The currently standing above
sector’s contribution to SACUwith
2008 revenues
massive after declines
the global financial
in SACUcrisis. To
revenues
date the country has not fully recovered from the are ch
export earnings has improved recently, currently
25
crisis. At the onset of the financial crisis in 2010 social
Central Bank Of Swaziland © 2018
when SACU revenue declined by over 6027percent subca
Central Bank of Swaziland © 2018 |
Page
the country has had to undertake some major
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
after the global financial crisis. To date the Figure 3: Sub-categories of Expenditure in
country has not fully recovered from the Swaziland
crisis. At the onset of the financial crisis in
2010 when SACU revenue declined by over
60 percent the country has had to undertake
some major reforms (both structural and
institutional) to ensure financial stability.
These reforms have been stipulated in
the Fiscal Adjustment Roadmap (FAR,
2010) document which was produced by
the International Monetary Fund (IMF) in
consultation with the Government of the There is currently no general consensus
Kingdom of Swaziland. on the share of the budget which can be
directed to each of the sub-sectors, save for
Amongst others, reforms on the expenditure bits and pieces of declarations which were
side included a wage freeze, a slowdown in signed at regional and international levels
the implementation of capital (investment) which were meant to guide budgeting in
projects, the Voluntary Exit Strategy to some components of the sub-categories. An
reduce the wage bill, and the introduction example is the Maputo declaration, which
of the value added tax (VAT). These reforms states that at least 10 per cent of a country’s
were for the short to medium term and were national budget should be allocated to
intended to reduce government expenditure agricultural activities. As a result, budgeting
and bring the budget deficit to levels of about in Swaziland is based on needs basis, with
3 percent of GDP from 2014/15. Although the general public services getting a fare
a budget of 1.2 per cent was reached in share of the budget. It should be noted at
2014/15, the challenge remains as the this point that this category also caters for
budget deficits are spiralling in the later the wage bill, which is the largest stand-
years. This therefore means that there is alone component of the budget.
still a huge financing gap for the government
budget. Figure 4 shows the allocations to the three
components over the years, for both capital
1.1 The effect of fiscal policy on output and recurrent expenditure. From the figure,
Generally, a country’s fiscal policy is expected during the times of economic boom in the
to have a positive effect on output. In the 1980s, general public services has always
utilization of the scarce resources (revenue), been the lowest, at less than 30 per cent
a major concern is where exactly the of total expenditure. That scenario has
resources should be channelled to achieve changed since the 90s, with the component
the desired output, and by which variations. toping the others at over 40 per cent.
In the country’s budget system there are
three broad categories where resources are
channelled, which are general, economic,
and social services. Each of these categories
has subcategories as shown in the figure
below.
ch Bulletin Volume C E N T R A L B A N K O F S WA Z I L A N D
2 result in revenue | RESEARCH BULLETIN VOLUME 2
et policy could multipliers that are
et
ed policy couldone
larger than result
andineven
revenue multipliers
positive in some that are
cases.
er
ed larger than one and even positive in some cases.
and
At itseven positive
simplest in some
level, fiscal cases.is the change
multiplier when the economy has slack, expansionary
er
he government spending shocks are less
At its simplest
in output level,
arising fromfiscal multiplier
a change in aisfiscal
the change
policy
he
th At its simplest level, fiscal multiplier is the likely to crowd out private consumption or
instrument.
in
change For instance;
outputinarising
outputfrom a change
arising frominaachange
fiscal policy
in a investment. To the extent that discretionary
th
al
fiscal policy
instrument. instrument.
For instance; For instance; fiscal policy is heavily used in recessions
𝑑𝑑𝑌𝑌 𝑡𝑡
al
nd to stimulate aggregate demand, the key
𝑑𝑑𝑍𝑍
𝑑𝑑𝑌𝑌𝑡𝑡𝑡𝑡
nd
he empirical question is how the effects of
𝑑𝑑𝑍𝑍𝑡𝑡 fiscal shocks vary over the business cycle.
he
ty where 𝑌𝑌𝑡𝑡 is output (or some other activity
where Yt is output (or some other activity The answer to this question is not only
y;
ty where
variable)𝑌𝑌𝑡𝑡 and
is output
𝑍𝑍𝑡𝑡 is a (or some
fiscal
variable) and Zt is a fiscal instrument,other activity
instrument, either interesting to policymakers in designing
y;a either government
variable)
government 𝑍𝑍𝑡𝑡 is aspending
andspending onfiscal on services,
goodseither
instrument,
goods and and
on stabilisation strategies but it can also help
services, ontransfers,
government transfers, or taxes the economics profession to reconcile
a
l., government spending on goods
or taxes orand
tax services,
rates. on
Since conflicting predictions about the effects
or tax rates. Since there are typically
l., government
there arethe
lags in transfers,
typically lagsorin
effects, taxes
one or tax distinguish
theshould
effects, rates. Since
one should of fiscal shocks across different types of
between impact multipliers (above) macroeconomic models (Auerbach and
there are typically
distinguish between lags in themultipliers
impact effects, oneand the
should
(above)
al cumulative multiplier: Gorodnichenko, 2012).
and the cumulative
distinguish betweenmultiplier:
impact multipliers (above)
ala
and the cumulative multiplier: 1.3 The Business Cycle
a ∑𝑛𝑛 𝑑𝑑 𝑌𝑌
ng 𝑗𝑗=0 𝑡𝑡+𝑗𝑗 The changing states of the economy during
∑ 𝑛𝑛
𝑛𝑛 𝑑𝑑 𝑍𝑍𝑡𝑡+𝑗𝑗
ng
te ∑𝑗𝑗=0
𝑗𝑗=0 𝑑𝑑 𝑌𝑌𝑡𝑡+𝑗𝑗 times of recession and expansion is referred
𝑛𝑛
∑𝑗𝑗=0 𝑑𝑑 𝑍𝑍𝑡𝑡+𝑗𝑗 to as the business cycle. Many variables are
te
er The interpretation
The interpretationofofthe
thefiscal
fiscalmultiplier
multiplieris used to measure the business cycle, which
is complicated by the fact that it is not can be capacity utilization, unemployment
m
er The interpretation
complicated by the of that
fact the itfiscal
is notmultiplier is
a structural
a structural parameter. Rather, in most gap, output gap, etc. This study limits
m
in relevant Rather,
parameter.
complicatedcontexts,
by the in therelevant
most
fact that it multiplier is the
is notcontexts, a
a structural itself to the output gap as a measure of the
in
ic function of structural parameters and policy business cycle. Output gap can be defined
parameter.
multiplier isRather, in most
a function relevant contexts,
of structural the
parameters
reaction parameters (Chinn, 2013). However, as potential output and its corresponding
ic
an multiplier
and the isyears,
overpolicy a function
reaction ofsimple
structural
parameters
this parameters
(Chinn,
estimation 2013).of
deviations from actual output. The most
an
es fiscal
and multipliers
policyover
However, reaction has shifted
parameters
the years, this simple to the
(Chinn, use
estimation2013).of
of
common measure of output gap which is also
es
an
econometric models, resulting in varying used in this paper is the Hodrick-Prescott
However,
fiscal
multipliersoveracross
the years,
multipliers has this simple
shifted
countries. to
For estimation
the use of
example, (HP) filter.
an
or Barro multipliers
fiscal and Redlick
econometric (2009)
has
models, find atomultiplier
shifted
resulting use of
the Research
in varyingofBulletin Volume 2
al
or 0.6 to 0.8 when using data on US defence Figure 5: US and Japan Recession Shadings
econometric
multipliers
zero bound,across models,
which occurs resulting
countries. in only
For example,
rarely and varying
Barro
in Figure 5: US and Japan Recession Shadings with
spending while Almunia et al. (2009) find a Swaziland’s with Swaziland’s
Output Output
Gap Gap
al
ce and Redlickacross
multipliers (2009) find than
a multiplier
countries. For of 0.6 to 0.8
recessions.
multiplier of greater one example,
when Barro
looking
Output_Gap
ce
te at the
and
when Great
Redlick
using dataDepression.
(2009) onfind
US adefence Theoretical
multiplier of 0.6 to
spending work0.8
while .05
when
Almunia using data
et al. on US
(2009) finddefence spending
a multiplier while
of greater
and others
clearing modelsemphasise
echo earlierthat government
Keynesian
.03
gh
nt spending
Almunia
than may
oneetwhen have find
al. (2009)
lookinga large
at athe multiplier
multiplier in the
of greater
Great Depression.
.02
nt
ax US when
than the
one when
Theoretical nominal
worklooking interest
at the Great
by Christiano rate al.is (2009),
at the
et Depression. .00
have
zero larger
bound,expansionary
which occurs effects
rarelyin and
recessions
only in -.01
in
ax Theoretical
Woodford work by
(2010), andChristiano et al. (2009),
others emphasise that -.02
recessions.
than in expansions. Intuitively, when the -.03
in
ax government
Woodford spending
(2010), mayothers
have aemphasise
large multiplier
slack,and that -.04
economy has expansionary government 1980 1985 1990 1995 2000 2005 2010 2015
recession shadings from the National Berea of fiscal expansions during recessions as a
Economic Research (NBER). The figure shows means of stimulating economic activity.
that most recessions in the country follows However, modern business cycle models, and
those of the US, just like most countries most empirical evidence suggests that these
in the world. This figure also incorporates policies are ineffective. The theoretical
Japan recession shadings. argument is that an increase in government
Research Bulletin Volume 2
spending raises consumers’ expected tax
The Fiscal Multiplier over the Business Cycle burden, and this negative wealth effect
n Blanchard andthese
Despite Perotti (2002)
important theoretical business
and insights cycle,
largely which
curtails the implies
expansionthat fiscal policy i
of aggregate
and strong demand by the policy process for demand. The multipliers generated by these
t studies, estimates
their paper
of fiscal extends
multipliers,the ineffective
no empirical models even
are during very severe
small, hovering at most downturns.
around
research trying to assess how the size of fiscal one. Moreover, their size does not vary over
terature inmultiplies
three varies
ways.over First, using
the business cycleTraditional
has the business cycle, which implies models
that fiscalusuall
Keynesian (IS-LM-AS)
been carried out in Swaziland,
itching SVAR models, they estimated but elsewhere policy is ineffective even during very severe
in the world. Most of empirical research have downturns.
large multipliers since the size of th
in this policies
ax and spending area is based
that canon linear
vary structural
vector autoregressions (SVARs) or linearised multiplier (when
Traditional accommodating
Keynesian (IS-LM-AS) models monetar
business dynamic
cycle. Theystochasticfound largeequilibrium usually have large multipliers since the size
general
(DSGE) models, which by construction policy rule of keeps
the the interest(when
multiplier rate from rising) is give
accommodating
s in the size of fiscal multipliers
out state-dependent in
multipliers. monetary policy keeps the interest rate
by from rising) is given by
and expansions, with
In their work, fiscalandpolicy
Auerbach Gorodnichenko,
(2010) addressed this issue by providing
siderably more effective instate-dependent
recessions 1
estimates of fiscal
1 − 𝑀𝑀𝑀𝑀𝑀𝑀
pansions. multipliers. Building on Blanchard and
Perotti (2002) and subsequent studies, their
paper extends the existing literaturewhere in whereMPCMPCis is the the marginal
marginal propensity to
propensity t
of the paper is ways.
three organised
First, asusingfollows;
regime-switching consume which is typically quite large (about
SVAR models, they estimated effects ofconsume which
tax 0.5–0.9). To is typically
that extent, quite
the AS large
curve in(about
the 0.5
is the review of literature,
and spending section
policies that 3 over the
can vary IS-LM-AS model is upward sloping. Therefore
business cycle. They found large differences 0.9). Tothethat extent,
multiplier canthe varyAS curve
from in thelarge
relatively IS-LM-A
he methodology, section
in the size 4 presents
of fiscal multipliersthe
in recessions (the AS curve is flat and there is a great deal
model is upward sloping. Therefore the multiplie
and expansions, with fiscal policy being of slack in the economy; i.e., in a recession)
sults, multipliers for more
considerably aggregated
effective and
in recessions to relatively small (the AS curve is steeply
can vary from relatively large (the AS curve is fla
upward sloping and the economy operates
than in expansions.
ted data, as well as multipliers in at full capacity; i.e., in an expansion). In
and there is a great deal of slack in the economy
The rest of the
s and recessions, paper is
while organised5as follows; contrast, an increase in government spending
section
section 2 is the review of literature, sectioni.e., inina modern
recession) business to cycle modelssmall
relatively usually
(the A
3 presents the methodology, section 4 leads to a large crowding out of private
presents the models results, multiplierscurve for is steeply upward
consumption sloping
in recessions andand the econom
expansions
aggregated and disaggregated data, as well and correspondingly the typical magnitude
ture Review as multipliers in expansions and recessions, operates at full
for the capacity;
multiplier is lessi.e.,
thanin0.5
an(Auerbach
expansion). I
while section 5 concludes. and Gorodnichenko, 2012).
contrast, an increase in government spending i
dvocated a fiscal stimulus during the The majority of studies on fiscal multipliers
2.0 LITERATURE REVIEW modern business
Keynes advocated a fiscal stimulusduring have focused cycle
on themodels
advancedusually leads to
economies.
ression, and since then governments
large Acrowding
the Great Depression, and since then comprehensive literature review on fiscal
out of private consumption i
multipliers in advanced economies can
imes implemented fiscal expansions
governments have at times implemented
recessions and expansions and correspondingl
cessions as a means of stimulating
activity. However, modern business
the typical Central
magnitude for ©the
Bank Of Swaziland
29
2018 multiplier is les
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
be found in Baunsgaard et al. (2012), who in periods with a binding zero lower bound
extended and updated work undertaken by (ZLB) on nominal interest rates (which are
Spilimbergo, Symansky, and Schindler (2009). recessionary times) could be somewhere
Baunsgaard et al. (2012) review a total of 37 between 3 and 5. Intuitively, with the
studies including both model based (DSGE) binding zero lower bound, increases in
and vector autoregressive (VAR) approaches. government spending have no effect on
For those studies government spending interest rates and thus there is no crowding
multipliers range between 0 and 2.0, with a out of investment or consumption, which
mean of 0.8 during the first year after fiscal leads to large multipliers. A key issue coming
measures are taken. Government revenue out of recent economic events is the size of
multipliers range from about –1.5 to 1.4, fiscal multipliers when the economy is in
with a mean of 0.3. Coenen et. al. (2012) recession.
compared seven models used at policy making
institutions, and found that most models In a paper by Auerbach and Gorodnichenko
(of which six are DSGE models, including (2011), they extended the standard Structural
GIMF) have similar short run multipliers for Vector Autoregression (SVAR) methodology
temporary changes in government spending in three ways to shed light on this issue.
(roughly 0.6 to 1.5 under the normal conduct First, using regime-switching models, they
of monetary policy) and revenues (roughly estimated effects of fiscal policies that can
0.1 to 0.5 under the normal conduct of vary over the business cycle, and found
monetary policy). large differences in the size of spending
multipliers in recessions and expansions
In spite of an extensive literature, there is with fiscal policy being considerably more
still no consensus regarding the size of fiscal effective in recessions than in expansions.
multipliers, even in advanced economies. Second, they estimated multipliers for more
They tend to be smaller in more open disaggregated spending variables that behave
economies and in countries with larger differently in relation to aggregate fiscal
automatic stabilizers, but as the theoretical policy shocks, with military spending having
and empirical literature suggest, they differ the largest multiplier. Third, they showed
widely across countries. For the advanced that controlling for real-time predictions of
economies, Spilimbergo et. Al., (2009) fiscal variables tends to increase the size of
suggest that as a rule of thumb, government the multipliers in recessions.
consumption multipliers are 0.5 or less in
small open economies, with smaller values Blanchard and Perotti (2002), estimated
for revenue and transfers and slightly larger multipliers for government purchases and
ones for investment. Moreover, recent taxes on quarterly US data. They applied four
studies have concluded that multipliers are identifying assumptions; (i) discretionary
significantly larger when the economy is policy does not respond to output within a
undergoing a recession than when it is in an quarter, (ii) nondiscretionary policy responses
expansion (Auerbach and Gorodnichenko, to output are consistent with auxiliary
2012, Batini and others, 2012, Baum and estimates of fiscal output elasticities, (iii)
others, 2012). innovations in fiscal variables not predicted
within the VAR constitute unexpected
Some findings from DSGE models with fiscal policy innovations, and (iv) fiscal
some Keynesian features (e.g., Christiano, multipliers do not vary over the business
Eichenbaum, and Rebelo 2011; Eggertsson cycle. These multipliers are still commonly
2008; and Woodford 2011), however, suggest cited, although subsequent research has
that the government spending multiplier questioned whether the innovations in these
for fiscal policy being more effective in because the ADF statistic is smaller than the
downturns than in expansions is that under MacKinnon critical values for the rejection
a negative output gap, excess capacities of hypothesis for unit roots and therefore
are available in the economy, making the had to be differenced to induce stationarity.
crowding out of private investment lower. This means that the null hypothesis for unit
This argument is expected to hold as long root is not rejected for these variables.
as the output gap is negative, which can
hardly be captured by other variables. These 4.1 Cointegration Results
equation specifications will be estimated After determining the order of integration of
in two fold; aggregated and disaggregated the variables, the next step is to determine
components of spending. This study used whether there is cointegration between the
annual data from 1980 to 2016 to examine variables. This is to establish if the linear
the effect of fiscal multipliers both in times relationship of the variables is stationary.
of recession and expansion in Swaziland. If the null hypothesis of no cointegration is
Data was collected from the Central Bank of rejected then the linear combination of the
Swaziland, the Central Statistics Office, and variables is stationary, hence a non-spurious
the Ministry of Finance. long-run relationship exists between the
variables and as such, consistent estimates
4.0 FINDINGS AND DISCUSSIONS of the long run relationship is evident. To test
Stationarity tests are the pre-tests for for cointegration between these variables,
avoiding spurious regressions. They are the the Johansen approach was employed. Using
starting point in any cointegration analysis as the lag length selection criteria, a lag length
well as regression analysis. In non-stationary of 3 was selected. Tthe results, it is evident
series, the order of integration is therefore that the variables are cointegrated (there is a
determined by the number of times it has to long-run relationship amongst the variables),
be differenced to attain stationarity. All the as there are cointegrating equations in both
variables were found to be nonstationary at the trace statistic and maximum eigenvalue
their levels, and had to be differenced. The statistic. In that regard, we proceed to
ADF test results for the first differences are estimate the SVAR for both aggregated and
presented in the table below. disaggregated data. Diagnostic tests were
carried out and the model passed all the
Table 2: Unit Root Tests Results (first difference) diagnostic tests as shown in the table below,
Variable Intercept Trend and None
where a rule of thumb is that the probability
Intercept should be greater than 0.05.
Log RGDP -4.239*** -3.693** -2.933***
Table 3: Results of Diagnostic Tests
Log -6.065*** -4.498*** -3.660***
Revenue Significance X2 statistic Probability
Log Expend -6.145*** -6.012*** -0.612 Breusch–Godfrey serial 2.98 0.24
Note: ***,**,*= significant at the 1%, 5%, 10% level. correlation LM test
White 0.53 0.63
A variable is stationary if the ADF statistic is Heteroskedasticity test
greater than the MacKinnon critical values Jarque–Bera test 0.65 0.56
for the rejection of hypothesis for a unit Ramsey RESET test (log 0.14 0.77
root. From the table above, all the variables likelihood ratio)
were previously not stationary at their levels Source: Own estimations
displays
to a shock
increase
aonsummary
in government revenue. The of the
spendingimpulse impulse
asresponses
response
depictedofin This.10 is in line with the Keynesian
.03
Response of LREV to LEXP
theory that an
Response of LRGDP to LEXP
to of GDPrevealing
GDP, and government that the spending revenue has to abeen shock
in revenue significantyields different effect onresults, with a
-.05 .00
pronounced expenditure
generated by the economy, hence a downward Figure 7: Response of Revenue and GDP to a
pronounced
compared to almost significant effecteffect
insignificant on expenditure
to GDP. A
-.10 -.01
Spending Shock
trend. However, the
compared to almost insignificant impact is felt in the effect laterto -.15
1
-.02
2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
positive shock in revenue (taxes) results to an Response to Cholesky One S.D. Innovations ± 2 S.E.
GDP.
years A upward
as an positivetrajectory shock in revenuein (taxes)
is observed GDP. Source: Authors Calculations
Response of LREV to LEXP Response of LRGDP to LEXP
Source: Authors Calculations
results to an increase inas depicted
increase in government spending government in .10 .03
spending as depicted in the firstand diagram, 4.3 Effects of revenue and components of
the first diagram, but is not generally GDP tobut
significant
.05 .02
Figure 6: Response of Expenditure
is noton generally 4.3 Effects (capital
of revenue and components
Revenuesignificant to GDP, revealing expenditure and recurrent) on output
.00 .01
a Shock
to GDP, revealing that the revenue has been
that the revenue has been generated by of expenditure (capital and recurrent) on
-.05 .00
Response to Cholesky One S.D. Innovations ± 2 S.E.
generated
the economy, by the economy,a hence
Response of LEXP to LREV hence downward a downward trend.
Response of LRGDP to LREV
output
4.3.1
-.10
Shock on revenue -.01
However,
trend. However,
.16
the impact the impact is felt is infeltthe in later
the later
.02
years -.15
1 2 3 4 5 6 7 8 9 10
-.02
1 2 3 4 5 6 7 8 9 10
as
.12
an upward trajectory is observed in GDP. 4.3.1
In this
Shock
Authorssection,
on revenue
Calculations expenditure was disaggregated
years as an upward trajectory is observed in GDP. .01
Source:
.08 In this section, expenditure was
Figure 6:Response
Responseofof Expenditure
.00
into
4.3 Effectscapital
of (investment)
revenue and and
components
disaggregated into capital (investment) and recurrent
of
Figure 6:
.04
Expenditure andand GDPGDP to to
aa Shock
Shockon
.00 onRevenue
Revenue -.01
recurrent(capital
expenditure
(consumption) (consumption)
to and totheir
recurrent)
ascertain ascertain
on output
impacts their
to
Response to Cholesky One S.D. Innovations ± 2 S.E. impacts to revenue and GDP. As shown in
revenue and GDP. As shown in Figure 8,
-.04 -.02
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Response of LEXP to LREV Response of LRGDP to LREV Figure 8, whenever there is a positive shock
4.3.1 Shock on revenue
Source: Authors Calculations.
.16 .02
Research Bulletin Volume 2
whenever
in revenue, therecapital
is a positive shock inresponds
expenditure revenue,
.12
goes towar
GDP on the other hand does not respond
Source: Authors Calculations.
GDP to an increase in government spending. The
Source: Authors Calculations whenever there is a positive shock in revenue,
Figure 8: Response of Expenditure The positive
4.2.2 Response of GDP and revenue to an Figure
capital 8: Response of Expenditure Components
4.2.2
Central Response
Bank of©GDP
of Swaziland 2018and revenue to an Components and GDP to a Revenue however
expenditure responds positively, Shock 38 |
P a expenditure
ge and GDP to a Revenue Shock not result to
expenditure shock
shock to a lesser extent Responsethanto Cholesky
recurrent
One S.D. Innovations ± 2 S.E. expenditure
Figure 7 displays the responses of revenue where the positive shock is largely pronounced. Response of LCAP_EXP to LREV Response of LREC_EXP to LREV Response of LRGDP to LREV and GDP. T
.3 .12 .03
FigureGDP
and 7 displays
to an the increase
responses of
inrevenue and
government .2 .02 recurrent e
to an GDP on the other hand does not respond
.08
34
.08
Central Bank Of Swaziland © 2018
Whenever there is a positive shock on capital .04
.00
.08 .005
.1
4.3.2 Shockthere
expenditure, on Capital Expenditure
is no immediate significant 4.4 Response of disaggregated expenditure
.00
-.1
-.005
impact
capitalon revenue and there
expenditure, recurrent
is noexpenditure,
immediate
-.08
revenue shock
1 2 3 4 5 6 7 8 9 10
-.2
1 2 3 4 5 6 7 8 9 10
-.015
1 2 3 4 5 6 7 8 9 10
significanta impact
although fall in on revenueexpenditure
recurrent and recurrentis Government spending was further
Source: Authors Calculations
expenditure,
observed in Figurealthough a fall Notable
9 as expected. in recurrent
is the disaggregated to three components; general
expenditure is observed in Figure 9 as services (general administration, public
positive response in GDP, which is significant 4.4 Response of disaggregated expenditure
expected. Notable is the positive response order and safety), social services (education
(economic, social, and general) on a revenue
after three
in GDP, years.isThat
which shows that
significant expenditure
after in
three years. and health), and economic services
shock
That shows
investment bythat expenditure
the government hasinaninvestment
impact on (agriculture, industry, and mining). The idea
by the government has an impact on GDP, is to identify
Government the impact
spending of each
was further component
disaggregated
GDP, compared to consumption expenditure.
compared to consumption expenditure. to revenue and GDP. First to be examined are
to three components; general services (general
the response of the expenditure components
Figure 9: Response of the Variables on Capital administration,
Figure 9: Response of the Variables on Capital
Expenditure to a positive public shock order and safety),
on revenue. The social
results
Expenditure in Figure 11 shows that the
services (education and health), and economicresponse is not
Response to Cholesky One S.D. Innovations ± 2 S.E.
Response of LREV to LCAP_EXP Response of LREC_EXP to LCAP_EXP Response of LRGDP to LCAP_EXP very significant
services (agriculture, in all the three
industry, components,
and mining). The
.10 .08 .04
however social and economic servicesResearch showsBulletin V
.05 .04
.03
idea is to identify
a positive responsethe impact
in theoffirst
eachyears,
component
which
.02
.00 .00
the first years, which shows that they do get mul
-.05 -.04
.01
shows
to revenue thatandtheyGDP. doFirstget positive
to be examinedspill
are overs
the
.00
positive
from the spill overs from the improved revenue. mea
-.10 -.08
-.01 response ofimproved revenue.
the expenditure components to a
-.15 -.12 -.02
Lila
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
positive
Figure shock
Figure 11:11: on Response
revenue.
Response The of
results
of Expenditure in Figure
Expenditure
Source: Authors Calculations
Source: Authors Calculations
Components to fisca
Components
11 toRevenue
shows that the Revenue
response is not very significant
4.3.3
4.3.3Shock
Shockonon
recurrent expenditure
recurrent expenditure Response to Cholesky One S.D. Innovations ± 2 S.E. imp
in allResponse
theof LGEN_SERV
three to LREV
components, howeverResponse
Response of LSOC_SERV to LREV
social and
of LECON_SERV to LREV
Research Bulletin
A Volume 2
positive shock to recurrent expenditure .3
economic services shows a positive response in
.12 .20 con
A positivetoshock
results to recurrent expenditure
a positive results
as shown response toThat
revenue, .15
e ten-year very significant, in Figure 10. could .2 .08
.10 the
although
to a positivenot very significant,
response as shown
to revenue, although notin
be a result of taxes to the government employees .1 .04 .05
inte
Figure 10. That could be a result of taxes .00
component
goes towards oftherecurrent expenditure
compensation goes
of employees.
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
outp
towards the compensation of employees. Source:
Source:Authors
AuthorsCalculations
Calculations
penditure The positive shock on recurrent expenditure does cum
hock The positive shock on recurrent expenditure 4.4.1
not
does result
nottoresult
a significant
to aresponse to investment
significant response 4.4.1 Response
Responseofof GDP
GDP to
to components
components ofof
expenditure fisca
expenditure
and GDP. That is because a positive shock toa
to investment and GDP. That is because
Whenever tota
Wheneverthere is aispositive shockshock
in eachinofeach
the
onse of LRGDP to LREV
on capital .04
.0
.000 government to increase its expenditure in perc
.00 -.005
significant -.04
-.1
-.010
economic services. perc
penditure, -.08
1 2 3 4 5 6 7 8 9 10
-.2
1 2 3 4 5 6 7 8 9 10
-.015
1 2 3 4 5 6 7 8 9 10
con
Figure 12: Response of GDP to Disaggregated
diture is Source: Authors Calculations Expenditure Shocks corr
Source: Authors Calculations
able is the ave
significant 4.4 Response of disaggregated expenditure GDP
nditure in
(economic, social, and general) on a revenue
shock
Central Bank Of Swaziland © 2018
35 elas
impact on
Figure 12:expenditure,
Response of GDP to Disaggregated
convert these
measured elasticities the to multiplier units, we
economic services
Figure 12: Response convert
Expenditure of GDPthesetoelasticities
Shocks Disaggregated
albeit
to GDP
multiplier
overunits, the inwe logarithms,
estimation
correct impulse period. obtained
It is known
response impulse
that by the
functions
GDP to Disaggregated correct impulse response functions by the
Expenditure Shocks response by functions are elasticities
insignificant. That correct shows impulse the needresponse for the functions
C E N T R A L B A N K O F S W A elasticity ZILAND |𝛽𝛽 can
the average
R E Sbe
E A Rdefined
C H B U L Las
ratios of themeasuring
ETIN VOLUME 2
respectivethe fiscal variable and
government to increase its expenditure in average
percentage ratios of
change thein respective
output in fiscal
response variable
to and
one
average ratios of the respective fiscal variable and GDP over the estimation period. It is known that
GDP over the 𝑋𝑋𝑡𝑡 Δ𝑌𝑌𝑡𝑡inperiod.
estimation It isvariables.
known that
economic services. GDP over the estimation period. Itpercentage is known that point change
elasticity 𝛽𝛽= the𝛽𝛽fiscal
can be defined as To
elasticity 𝛽𝛽
convert these elasticitiescan be𝑌𝑌 Δ𝑋𝑋
defined
𝑡𝑡 as
𝑡𝑡 to multiplier units, we
Figure elasticity 𝛽𝛽 can be defined as Where β is a percentage. Hence𝑋𝑋𝑡𝑡 Δ𝑌𝑌𝑡𝑡 = 𝛽𝛽
Figure12: 12: Response
Response of of GDP
GDP to to Disaggregated
Disaggregated 𝑋𝑋𝑡𝑡 Δ𝑌𝑌Hence
Expenditure
ExpenditureShocks
thors Calculations Shocks 𝑋𝑋𝑡𝑡 Δ𝑌𝑌𝑡𝑡 Where 𝛽𝛽
correct is a percentage.
impulse response 𝑡𝑡
= 𝛽𝛽functions 𝑌𝑌𝑡𝑡 Δ𝑋𝑋𝑡𝑡 by the
ulletin Volume 2 = 𝛽𝛽 𝑌𝑌𝑡𝑡 Δ𝑋𝑋𝑡𝑡
𝑌𝑌𝑡𝑡 Δ𝑋𝑋𝑡𝑡 average ratios ofWhere Δ𝑌𝑌 𝑋𝑋fiscal
𝛽𝛽 is a percentage.
the𝑡𝑡respective Henceand
variable
Source: Authors Calculations
Where 𝛽𝛽 is a percentage. Hence 𝑡𝑡
calmultipliers.
Multipliers When put
Source: Authors Calculations
Where 𝛽𝛽 is a percentage. Hence
= 𝛽𝛽 ÷
simply, fiscal multipliers GDP over the estimation Δ𝑋𝑋Δ𝑌𝑌𝑡𝑡 period. 𝑌𝑌 ItΔ𝑌𝑌
𝑋𝑋𝑡𝑡𝑡𝑡
is𝑡𝑡 known 𝑋𝑋that
= 𝛽𝛽 ÷
𝑡𝑡
4.5 Fiscal Multipliers 𝑡𝑡
= 𝛽𝛽 ÷ Δ𝑋𝑋 𝑌𝑌
multipliers Δ𝑌𝑌𝑡𝑡 𝑋𝑋𝑡𝑡
measureinthe thischange
study are different from
4.5 Fiscal Multipliers 𝑡𝑡 𝑡𝑡
Fiscal in outputinarising
multipliers this
Δ𝑋𝑋
= 𝛽𝛽from
study ÷are
Where
𝑌𝑌
aelasticity
𝑋𝑋𝑡𝑡
different is
from
𝛽𝛽 can be defined
expressed Δ𝑋𝑋𝑡𝑡
as
as 𝑌𝑌𝑡𝑡
𝑋𝑋𝑡𝑡 a constant and it is the
Fiscal multipliers in this study are different 𝑡𝑡 from 𝑡𝑡 𝑌𝑌𝑡𝑡 𝑋𝑋 Where𝑋𝑋𝑡𝑡 Δ𝑌𝑌 is expressed as a constant and it is the
estudy
response functions
Lilangeni change in
are different from in the
expenditure
impulse response
𝑋𝑋 𝑡𝑡
sense or that
revenue.
functions in The
the Where 𝑌𝑌𝑡𝑡 isisexpressed
Where
sense that expressed 𝑡𝑡𝑡𝑡 =
𝑌𝑌as a as
constant
𝛽𝛽 a constant and it is andthe
impulse responseWhere is expressed as a that
constant and it is the 𝑌𝑌 Δ𝑋𝑋
𝑡𝑡
functions in the sense average is proportion
it averagethe proportionaverage of
average 𝑡𝑡 variable
proportion proportion𝑋𝑋𝑡𝑡ofvariable
𝑡𝑡 to output
variable 𝑋𝑋𝑡𝑡 to𝑌𝑌𝑡𝑡output
. 𝑌𝑌𝑡𝑡 .
of variable 𝑋𝑋of
𝑌𝑌𝑡𝑡
eons
responses
in the sense
fiscal measure
multipliers can a percentage
that impulse
be responseschange
categorised measure
into a percentage
three: (i)Where change 𝛽𝛽 is a percentage. Hence 𝑡𝑡 to output 𝑌𝑌𝑡𝑡 .
impulse
Source: Authors responses
Calculations measure
average a percentage
proportion change 𝑋𝑋𝑡𝑡toto output
of variable output 𝑌𝑌𝑡𝑡 .. These These averages for revenue
Source: Authors Calculations
ure a percentage change compared to a unit change measured TheseThese averages
by fiscal averagesfor for revenue
averages
revenue and
for
and expenditure
revenue
expenditure
and expenditure
edimpact
to a unitcompared change to a unit
multipliers, measured
These change
(which
averages by fiscal
measured
forshow by fiscalthe
revenue and expenditure
and expenditure
Δ𝑌𝑌 𝑡𝑡 (including 𝑋𝑋 𝑡𝑡 expenditure
4.5Fiscal
Fiscal Multipliers = 𝛽𝛽 ÷
4.5
nge measured Multipliers
by fiscal Central Bank of Swaziland © 2018 components) forΔ𝑋𝑋 the 𝑡𝑡 estimation 𝑌𝑌𝑡𝑡 period from 40 |
Central Bank of PSwaziland © 2018 40 | 40 |
Bank contemporaneous
of Fiscal
Swaziland
Fiscal © 2018
multipliers
multipliers effect
gin
ain this
ethis of
studystudyone are unit
are
different increase
differentfrom in
1980 to 𝑋𝑋 2016 are shown in table 5.
2018 Page Where 𝑡𝑡40is| expressed as a constant and it is the
from impulse
impulse response
response
functions
functions
in the
in
sense that
the 𝑌𝑌𝑡𝑡
the sense
respective fiscal variable
that impulse responses measure a Table on output), (ii) the 5: proportion
Ratio of Each Component to GDP𝑌𝑌𝑡𝑡in
average of variable 𝑋𝑋𝑡𝑡 to output .
impulse
percentage responses
change measure compared a percentage
to achange
unit the Estimating Horizon (1980 to 2016)
interim multipliers (show the effect of one unitThese averages for revenue and expenditure
change
compared measured
to a unit bychange
fiscal multipliers.
measured by When fiscal REV EXP CAP_ REC_ ECON_ GEN_ LSOC_
increase in thefiscal
put simply, respective fiscal
multipliers variable
measure the on EXP EXP SERV SERV SERV
Central Bank of Swaziland © 2018 40 |
change in output arising from a Lilangeni 0.180 0.199 0.043 0.156 0.0847 0.066 0.034
output
P a g ein the medium-term). Finally, theSource: Own calculations
change in expenditure or revenue. The fiscal
cumulative multipliers
multipliers is defined into
can be categorised as the sum ofFor example, the ratio of average revenue
three:
(i) impact multipliers, (which show the to average GDP for the period 1980 to 2016
fiscal multipliers through
contemporaneous effect aof horizon (shows the
one unit increase is 0.18 as shown in Table 5. Dividing the
in the respective fiscal variable on output),
total effect of fiscal policy changes to output overimpulse response value of 0.0019 in Table 6
(ii) the interim multipliers (show the effect by this ratio gives an impact multiplier (first
a pre-defined
of one unitperiod.
increase in the respective fiscal year) of 0.01055 in the interim multiplier
variable on output in the medium-term). column. Summing the interim multipliers
Finally, the cumulative multipliers is defined cumulatively gives the accumulated
For as estimation
the sum of fiscal multipliers
purposes, giventhrough
that a themultipliers as shown in the fourth column.
horizon (shows the total effect of fiscal
endogenous variables
policy changes used in
to output thea estimations
over pre-defined areTable 6: Revenue and Expenditure Multipliers
period. in logarithms, the obtained impulseto GDP
measured
Period REVENUE
response functions are
For estimation elasticities
purposes, givenmeasuring
that the the (Yr) IMPULSE INTERIM ACCM MLTP
endogenous variables used in the estimations
percentage changeininlogarithms,
are measured output in the
response
obtainedto one 1 0.00190 0.01055 0.01053
2 0.00097 0.00537 0.01591
impulse point
percentage response functions
change are elasticities
in the fiscal variables. To 3 0.00189 0.01047 0.02638
measuring the percentage change in output
convert these elasticities to multiplier units, we 4 -0.00368 -0.02041 0.00597
in response to one percentage point change
in the fiscal variables. To convert these 5 -0.00254 -0.01407 -0.00810
correct impulse response functions by the 6 0.00013 0.00070 -0.00740
elasticities to multiplier units, we correct
average ratios
impulse of the respective
response functions byfiscal
thevariable
average and 7 0.00300 0.01662 0.00922
ratios of the respective fiscal variable and 8 0.00530 0.02938 0.03860
GDPGDPoverover
thethe
estimation
estimationperiod. It Itis isknown
period. knownthat 9 0.00703 0.03898 0.07758
elasticity 𝛽𝛽 can beβdefined
that elasticity can be as
defined as 10 0.00614 0.03401 0.11160
𝑋𝑋𝑡𝑡 Δ𝑌𝑌𝑡𝑡
= 𝛽𝛽
𝑌𝑌𝑡𝑡 Δ𝑋𝑋𝑡𝑡
Where 𝛽𝛽 is a percentage. Hence
36 Δ𝑌𝑌
Δ𝑋𝑋𝑡𝑡
= 𝛽𝛽 ÷
𝑋𝑋
Central𝑡𝑡Bank Of Swaziland
𝑌𝑌𝑡𝑡
𝑡𝑡 © 2018
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
hand consists of agriculture, industry, and Shin, 1997). The test finds parameter
mining, transport and communications. instability if the cumulative sum goes
That means the government should invest outside the area between the two critical
more on the capital economic services, lines. The figure below plots the results
which were found to be the most drivers of for CUSUM and CUSUMSQ tests. The results
the economy. Although low, social services indicate the absence of any instability of the
are also very important as they consists of coefficients because the plot of the CUSUM
building schools and hospitals, but their and CUSUMSQ statistic fall inside the critical
economic significance is low in the medium bands of the 5 percent confidence interval
term. for parameter stability.
4.6 Fiscal Multipliers in times of recessions Figure 16: Plot of the CUSUM and CUSUMSQ
Research Bulletin Volume 2
and expansions Statistic
One ofLREV(-3)
the major objectives of the study
0.044898 0.014208 3.160071 0.0039 is 16 1.4
1.2
to findLEXP(-3)
the size of fiscal 0.014194
multipliers in times
12
0.6
(1998) andOUTPUT_GAP
estimated it using the Threshold
<= 0.015149999 -- 27 obs
-8 0.0
-16
-0.2
-0.4
LREV(-3) 0.144190 0.041287 3.492352 0.0017 CUSUM 5% Significance CUSUM of Squares 5% Significance
After estimating the model, the cumulative higher during times of recession than
stability. These results shows
sum of recursive residuals (CUSUM) and expansion, with that multipliers are
a Lilangeni higher in
increase
the CUSUM of square (CUSUMSQ) tests are expenditure
during bringingthan
times of recession 89expansion,
cents with
to aGDP
applied to assess parameter stability (Pesaran Lilangeni increase in expenditure bringing 89 in
in recessions compared to 39 cents
cents to GDP in recessions compared to 39 cents
in expansions.
recent
These results are in line with
Central Bank Of Swaziland © 2018
studies that have concluded
39 that
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
expansions. These results are in line with The results shows that expenditure on capital
recent studies that have concluded that projects far exceed one during recessions,
multipliers are significantly larger when which shows that the magnitude of the gains
the economy is undergoing a recession than on government investment is huge. From
when it is in an expansion (Auerbach and the results, an additional Lilangeni used in
Gorodnichenko, 2012, Batini at al. 2012, capital expenditure results to E1.60 increase
Baum at al. 2012). in GDP during recessions, compared to 51
cents during expansions. Therefore, since it
The nice point to emerge out of this exercise has been found that increasing government
is that during recessions policymakers are spending in periods of recessions would
trying to stimulate the economy in order to stimulate output, much of the expenditure
raise output and employment. The policy should be directed to capital projects.
implication is that increasing government
spending in periods of recessions (as 5.0 CONCLUSION AND
Keynesian considerations would call for) RECOMMENDATIONS
would stimulate output whereas increasing it The major objective of this study was to
in times of expansion would have essentially assess the impact of fiscal policy on output
no effects. Further, since we have made no in Swaziland, with emphasis on the size
distinction between increases and decreases of the fiscal multipliers, including times
in government spending, it would also of recessions and expansions. The SVAR
follow that reducing government spending modelling approach was used to determine
in recessions, as many developing countries linear multipliers, while MTAR was applied
have historically done would be quite for non-linear multipliers (recessions and
contractionary whereas reducing spending expansions). The first step therefore was
in good times would have little effect. to test for stationarity of the variables.
Since the study has found that it is wiser Results of the ADF stationarity test show
to increase spending during recessions, the that the variables are not stationary. The
issue is financing the budget, whereby the Johansen and Juselius cointegration test
country might find itself with a high debt was performed to establish the existence of
to GDP ratio. Thus, policy makers should a long run relationship among the variables,
be careful on the magnitude in which the results of which show the existence of such
spending should be increased. The results a relationship.
are further confirmed when expenditure is
disaggregated into capital and recurrent as Once cointegration was established, the
shown in the table below. SVAR model was estimated and the impulse
response analysis was undertaken. The
Table 10: Fiscal Multipliers in Times of
results shows that the response of GDP is
Recessions and Expansions for Capital and
not significant for revenue but significant
Recurrent Expenditure
to expenditure. When expenditure was
Variable Expansion Recession
disaggregated to capital and recurrent, the
Revenue 0.2645 0.7487 results shows that GDP response is significant
Capital 0.5129 1.6154 for capital than recurrent expenditure.
Expenditure On further disaggregation of capital
Recurrent 0.3394 0.6058 expenditure, the results show that economic
Expenditure services followed by general services
Source: Authors Calculations
expenditure drives capital expenditure.
Ilzetski, E., Mendoza, E., and Vegh, C., Persaran, MH & Shin, Y 1997, ‘An
2011, “How Big (Small) Are Fiscal Autoregressive Distributed Lag
Multipliers?” IMF Modelling Approach to cointegration
Working Papers 11/52 (Washington: Analysis’, Cambridge Working Papers
International Monetary Fund). in Economics No. 9514, Faculty of
Economics, University of Cambridge.
International Monetary Fund, 2008,
World Economic Outlook, October, Spilimbergo, A., Symansky, S., and
Chapter 5, “Fiscal Policy as a Schindler, M., 2009, “Fiscal
Countercyclical Tool” (Washington). Multipliers,” IMF Staff Position
Note, SPN/09/11, May (Washington:
Mendoza, E. G. and M. E. Terrones (2012). International Monetary Fund).
An anatomy of credit booms and
their demise. NBER Working Papers Tsay, R. S., 1998,”Testing and Modelling
18379, National Bureau of Economic Multivariate Threshold Models,”
Research, Inc. Journal of the American Statistical
Association, Vol. 93, pp. 1188-1202.
Mineshima, A., M. Poplawski-Ribeiro, and
A. Weber, 2014, “Fiscal Multipliers,” Woodford, Michael. 2011. “Simple Analytics
in Post-Crisis Fiscal Policy, ed. by C. of the Government Expenditure
Cottarelli, P. Gerson, and A. Senhadji Multiplier.” American Economic
(Cambridge: MIT Press, forthcoming). Journal: Macroeconomics 3(1): 1-35.
The Efficacy and Potency or Key Words: CMA, Optimum Currency Area,
Paralysis of Monetary Policy in Macroeconomic Convergence, Monetary
Policy, Structural VAR.
Swaziland under the Common
Monetary Union
1. INTRODUCTION
The paper argues that Swaziland’s
Sikhumbuzo S. Dlamini7 ,
membership to the Common Monetary Area
Ntobeko S. Dlamini8 and Sive Kunene9
(CMA) together with Lesotho, Namibia and
South Africa has paid huge dividends for the
Abstract
country over the years. With macroeconomic
stability the principal outcome over the
This paper is a twofold analysis of the
years the country has achieved a stable
effectiveness of monetary policy in Swaziland
currency with the Lilangeni pegged one-to-
and an assessment of the Common Monetary
one with the South African Rand. Under this
Area (CMA) as an optimal currency area
arrangement, the economy has also been
(OCA) using structural vector autoregression
stable. The interlinkages between fiscal
(VAR) models. In assessing the optimality,
and monetary policies have been strong.
the paper identifies the underlying shocks
Monetary policy implementation has yielded
and correlation in the shocks to real gross
positive reviews over the years. The country
domestic product, consumer price index and
has maintained relatively stable levels of
real exchange rate. The results indicated
increase in prices (inflation) as well as a
that demand shocks proxied through the
stable financial sector.
consumer price index are less correlated
among the CMA member states; instead,
Monetary policy has had a measurable and
they have converged around the South Africa
foreseeable impact on demand and inflation
as an anchor economy. The results therefore
over time. Within the confines of lose of
support the Central Bank of Swaziland’s
monetary policy autonomy in a currency
drive to align policy rates with that of South
union, monetary policy may not always be
Africa, but dependent on data outcomes.
effective. This requires central banks to
The impulse response results revealed
continuously innovate in areas such as risk
that a shock on the discount rate only has
management and developing analytical
a negative significant effect on the real
tools to deal with adverse developments,
gross domestic product (GDP), inflation and
including the inherent costs of monetary
money supply (M2), indicating the efficacy
policies, especially those applied in quasi-
of monetary policy in Swaziland.
currency unions such as the CMA.
per cent
difficult external environment, especially from 6
4
C E N T R A L B A N K O F S WA Z I L A N D | South
RESEAR C H B Africa,
U L L E T Ileading
N V O L Uto
MEa sharp
2 decrease in South
2
African Customs Union (SACU) revenues. Such a 0
decrease in revenue, combined with increased
independence or autonomy is inherent increased
public publichasspending,
spending, generatedhas generated
higher fiscal
by jurisdiction or national sovereignty. higher fiscal deficits and a growing public
deficits and a growing public debt.
Remaining within the CMA is one such debt. Source
conscious decision in the implementation Figure 1: GDP Growth Rates in Swaziland
of the country’s monetary policy strategy. It Figure 1: GDP Growth Rates in Swaziland Sout
notes various developments in South Africa 8.00% 6.86%
5.99% dete
that had devastating effects to the other 6.00% 4.44% 4.74% 4.25%
3.79%
4.00% (Ndz
economies in the CMA, especially those 1.57%
2.25%
1.46%1.31%
2.00% 0.82% betw
associated with the collapse of the exchange
0.00%
rate. 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 thro
Real GDP Growth extr
1.1 Stylised Facts about the Swazi
Economy Source:
Source: Central
Central Statistical
Statistical Office Office Furt
inve
1.1.1 Economic Growth 1.1.2Inflation
1.1.2 Inflation Developments
Developments
finan
During the 1980s Swaziland recorded high The Central Bank of Swaziland has as its
The Central Bank of Swaziland has as its ultimate
economic growth rates, driven by an influx of ultimate goal price stability to create Thes
foreign direct investment (FDI) arising from goal price stability conducive
an environment to create anfor environment
economic and
sanctions imposed on South Africa, which conducive growth. The monetary
for economic authorities
growth. in the
The monetary
propelled the relocations of enterprises into country basically uses monetary policy tools
authorities in the country basically uses monetary
Swaziland. The high levels of foreign direct (namely; the discount rate, reserve ratio and Imp
investment caused an economic upturn in policy liquiditytools (namely; the discount rate, reserve
requirement) to control the level thro
the manufacturing sector, which became the ratio of money supply and
and liquidity consequently
requirement) inflation.
to control the
Research Bulletin Volume 2 at
main growth engine, which in turn encouraged The Bank mainly uses the discount rate
level of money supply and consequently inflation.
rapid
export growth in sugar,
prices for supporting sectors
reinforced such
by the realas largely moves
which largely in harmony
moves in with the movements
harmony with the in cont
construction. The Bank mainly uses the discount rate which
depreciation of This also generated
the Lilangeni. However,additional
in recent movements
the repurchaseinrate theinrepurchase
South Africa. rate
This isinmainly
South inde
revenue which permitted the consequent Central Africa.Bank This is mainly
of capital
Swaziland done to curb capital
© 2018
years, that growth
expansion stimulus subsided.
of government services. Apart Pflight done to curb flight as there is free flow of
a g e as there is free flow of capital in the
from the inflows into the manufacturing CMA. capitalThein the CMA.below
graph The graph
showsbelow showsseries
the time the
Swazilandthe
sector, has growth
suffered sluggish growth was
performance since also
the data for the discount rate and inflation from
time series data for the discount rate and inflation
aided by more conventional external
turn of the century, averaging 3.2 per cent stimuli, 2006 to 2017.
such as improved export prices for sugar, from 2006 to 2017.
between the years 2001 to 2016. Economic
reinforced by the real depreciation of the Figure 2: 2: The
The Monetary
Monetary Policy
PolicyRates
RatesBetween
Between
growth has been
Lilangeni. slowingin
However, down since 2013,
recent years,falling Swaziland and
that Swaziland South Africa
and South Africa
growth stimulus subsided.
from 4.6 per cent to 1.3 per cent in 2016, 14
12
averaging 2.5 per cent over this period. The
Swaziland has suffered sluggish growth 10
slowdown
since the isturn
due of
to continued drought
the century, and a
averaging 8
per cent
45
6.00% 4.44% 4.74% 4.25%
3.79%
4.00% 2.25%
(Ndzinisa, 2008).
Central Bank Of With free
Swaziland flow of goods
© 2018
1.57% 1.46%1.31%
2.00% 0.82% between South Africa and Swaziland, as well as
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
determining the inflation level in Swaziland and this had a once-off effect especially
(Ndzinisa, 2008). With free flow of goods during the period in which it was introduced.
between South Africa and Swaziland, Research as wellBulletin Volume 2
as through close proximity, the two countries Figure 3: Inflation Movements in the CMA
fuel, electricity
have extremelyand food.trade
strong With andfuel investment
having the 14
links. Further,
strongest importthe two countries
inflation effect.enjoy strong
Swaziland 12
investment links with South African
imports over 80 per cent of its electricity fromretailers
10
and financial institutions established in
South Africa. These
Electricityaretariffresponsible
increases have
per cent
Swaziland. for 8
Such imports
Imported therefore
inflation transmit
effects arethe effects of
transmitted 2
South Africa,
Swaziland among especially fuel,
the highest andelectricity
Namibia withand Lesetho Namibia
46
Central Bank of Swaziland © 2018 51 |
Page Central Bank Of Swaziland © 2018
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
The influence of exchange rate towards 1.2 The importance of the Paper
inflation itself depends on the level of Monetary integration, to varying degrees
imports denominated in foreign currencies. has become an increasingly important
As the Lilangeni/Rand freely floats, it has phenomenon within the modern global
over time been extremely volatile, on the political economy. The global political
weaker side. A depreciation (devaluation) economy is characterized by regions
of the domestic currency can affect the developed into building blocks to develop
price level directly through imported goods economies of scale to take advantage of trade
that domestic consumers pay. However, agreements as well as policy harmonisation
this condition occurs if the country is the for quasi-homogeneous economies. Such
recipient of international prices. This is homogeneity is mainly on the structure
more pronounced for fuel and some food of the economies and historical political
items, such cereals linked to wheat prices. links, such as is the case for the Southern
That is on the demand side. On the supply African Customs Union (formed in 1969).
side, the transmission of the effects of Due to such demand, the countries agreed
the exchange rate is through intermediate that in order to foster stronger trade ties,
goods, which are inputs to the production a currency union, CMA, was an imperative.
process. With trade liberalization, and The CMA offers the opportunity for countries
global interconnectedness, it has become such as Swaziland to enjoy macroeconomic
easy to source products globally. However, stability as well as price and financial sector
the depreciation of the currency would tend stability through exchange rate stability.
to increase production costs and such be The Lilangeni is pegged one to one to the
inflationary as manufacturers will transfer South African Rand. This allows the country
costs increases to consumers. to “import some” price stability as well as
any ailments into the domestic economy.
Figure
Research Bulletin Volume4:
2 Exchange Rates Movements in the This was evident in Rand crises in 1985, 1998
CMA and 2002.
in South 25
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
the structural vector autoregressive (SVAR) the Nigerian economy using data spanning
model approach to investigate the impact of from 1981 to 2008. The results revealed
monetary policy changes on macroeconomic that monetary policy represented by money
variables in Swaziland. The results revealed supply had a positive impact on GDP growth
that a shock on the discount rate had a and Balance of Payments (BOP), but had a
negative impact on inflation, even though negative impact on inflation.
the response was not statistically significant.
The results further revealed that a positive Peersman and Smets (2003) applied a
shock on the discount rate had a negative standard VAR for the Euro area using data
effect on business credit and GDP. These spanning from 1980 to 1998 to investigate
variables respond to the shock with different the response of main macroeconomic
lags. variables to an unexpected monetary policy
shock. The results revealed that a positive
Results by Simatele (2003), revealed that shock to the interest rate was followed by a
in Zambia bank lending is not driven by fall in output and a real appreciation of the
monetary policy but rather by demand. exchange rate. The results also revealed that
The results further revealed that a prices appear to be sticker and therefore
contractionary monetary policy had a respond with a lag and a decline several
negative and significant effect on inflation quarters later. Popescu (2012) employed
and GDP. Kalikeka and Sheefeni (2013) a VAR approach to analyse the effects of
investigated monetary policy transmission monetary policy in Romania using quarterly
in Zambia focusing on the interest rate data spanning from 2005 quarter three to
channel using data from the year 1980 to 2012 quarter one. The results revealed a
2011. The granger causality results showed negative response of inflation, GDP and
a bidirectional causality relationsionship Money supply, and a positive response by the
between inflation and interest rate, and a nominal exchange rate.
unidirectional causality from GDP to interest
rates. The impulse response results were Cheng (2006) applied both recursive and
similar to those of Simatele (2003) in which non-recursive SVAR to monthly data in Kenya
a positive shock on monetary policy had a for 1997–2005. The results revealed that an
negative impact on GDP and inflation. On increase in interest rate led to an increase in
the overall, the results provided evidence of the price level followed by a falling price level
a functional interest rate channel existence that was statistically significant for about
in the Zambian economy. two years following the shock. In response
to a contractionary monetary policy, output
Ikhide and Uanguta (2010) used a vector rises initially but falls eventually, though the
autoregressive (VAR) approach to study the decline was not statistically significant.
impact of South Africa’s monetary policy
impact on Lesotho, Namibia and Swaziland Maturu, Maana and Kisinguh (2010) applied
(LNS) and found that monetary policy in the the use of a SVAR and a VAR to study
LNS responded instantaneously to changes in monetary policy transmission in Kenya using
monetary policy in South Africa. Ikhide and quarterly data from 2000 to 2010. Maturu,
Uanguta (2010) concluded that the South Maana, and Kisinguh (2010) used M3 as the
African repo rate was the most relevant monetary policy instrument and the results
policy instrument than the bank rates for the revealed that a positive shock on M3 had no
LNS countries. Onyeiwu (2012) employed effect on real output but had a statistically
the Ordinary Least Squares (OLS) method to significant positive effect on prices for
examine the impact of monetary policy on almost 18 months. A positive shock on the
arameter
us
that mostmatrix
variables. The
of the
A . Thus,
matrix the order
form
studies of theSims
of by
this variables
nature (1980), whereby Based
have used the on Cholesky
the Cholesky
se similar to many
VAR model studies as
is selected used indecomposition
the context ofis applied to the contemporaneous
3.2.2 The Structural Vector Autoregression residuals (et) and structural shocks (Et) can
VARs, especially recursive VARS. On this basis, this reduced form VAR, the stu
ARs in advanced Model (SVAR) for Efficacy
economies, including
parameterof Monetary Sims A. Thus,
matrix be expressedthe order as follows:
of the variables
study uses the standard recursive structural
Policy VAR. Bulletin
Research thatVolume define2 matrix A0 as a l
1992), and Fry-McKibbinis as follows; andoutput
Zheng is (GDP),
similar
(2012) and inflation
to many studies used in the context of
Naceur,
The
that
ate (CPI), basic
most money structural
Boughrara
of the
supply (M2), VAR
and Ghazouani
studies of
private model
(2009) assert
this
VARs sector nature
in in
credit this
that
advanced have study used
economies,
Based
The
on 2 the Cholesky Basedidentification
including on Simsthescheme
decomposition
fo
Cholesk
of
empirical analysis of the effects of monetaryResearch Bulletin Volume
zz Amongst other things, the Central identification scheme and identified a series
Bank of Swaziland before deciding on of demand and supply shocks for economies
its monetary policy stance considers in different regions to examine whether
developments in the domestic economy forming an optimum currency area was
and this include output, inflation, viable.
money supply and private sector credit.
In this regard, the study orders the These authors tested for symmetry
discount rate last since its level is, in (correlation) of the underlying structural
part, influence by the above mentioned shocks between countries in Western
factors. Europe, East Asia and the Americas. Coco
and Silvestrini (2017) also used a SVAR model
The impulse response and variance approach to determine the viability of the
decomposition analysis are carried out to European Monetary Union, after its long
determine the effects of the discount rate existence. As in Coco and Silvestrini (2017),
shocks in the estimated model. The impulse this study adopts a SVAR representation to
response function in a VAR analyses dynamic investigate whether the CMA is an optimum
affects on the system when the model currency area or not. The study uses three
received the impulse of say one standard variables (consumer price index, real GDP
deviation shock. The variance decomposition and real exchange rate) to construct the
reveals some amount of vital information VAR model required to examine the shocks
and explanation of the contribution of each according to the OCA criteria and to establish
variable to other variables in the system. whether countries in the CMA do exhibit
some level of convergence or not.
3.3 Is the CMA an Optimum Currency Area?
The second objective of the study is to Following Fielding and Shields (1999), the
determine whether the CMA is an optimum estimated innovations in the SVAR model are
currency area or not and the study uses the referred to as the macroeconomic shocks
OCA criteria to asses such. wherein conclusions about the degree of
similarity of shocks between countries, are
3.3.1 OCA Criteria based on the pairwise correlation of the
The paper builds on the two-country model innovations in their respective models. This
developed by Ricci (1997). Ricci (1997) paper thus focuses on shocks to real gross
asserts that converging to the anchor domestic product (q), consumer price index
economy accrued certain costs to the smaller (p) and the foreign price (xr) shock common
economy. Any volatility of the exchange to all member countries.
rate (of the anchor economy) would be
transferred fully to the smaller countries. 3.3.2 The Structural Vector Autoregression
(SVAR) for OCA
Generally, the OCA analysis focuses on how In this paper’s SVAR OCA criteria, we let
various macroeconomic variables such as GDP
growth, inflation, exchange rates, interest
rates and stock prices are correlated across where Δ represents the first difference
economies or regions. Among the first to operator, y is the vector of the endogenous
identify the fundamental structural shocks variables and denote the
were Bayoumi and Eichengreen (1992), and orthogonal shocks of supply, demand and
Bayoumi and Eichengreen (1994) who used exchange rate, respectively. The structural
a structural vector autoregression (SVAR) VAR model is thus presented as follows;
.000
.24
.20
-.004 .12
.00
-.016 -.08
0.0
and 36 months, respectively. This indicates
.005
.00
.000
-.01
-0.5 -.010
-1.0
smaller portion of the variation in inflation.
.16
Response of DR to DR
-1.5
A variation in money supply is explained by
.12
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 1.6 per cent of variation in the discount rate
.08
.04
Source: Own Calculations in 6 months and 1.47 per cent in 36 months
.00
Source: Own Calculations
whiles private sector credit variation is
-.04
5 10 15 20 25 30 35
4.1.3 Impulse Response Functions explained 0.2 per cent and 2.58 per cent
4.1.3 Impulse Response Functions
The impulse response results presented on Source: own calculations
variation in the discount rate in 6 and 36
The impulse
Figure response
8 indicate results
that presented
a shock on theon Figure
discount months, respectively. Lastly, the results
4.1.4 Variance Decomposition Analysis
rate only that
8 indicate hasa ashock
negative
on the significant
discount rateeffect
only indicate that discount rate shocks account
on the real gross domestic product (GDP), Table
for about 1 shows0.15 per the cent
variance decomposition
and 15.18 per cent
has a negative significant effect on the real gross
inflation and money supply (M2). Inflation variation
analysis for in
realreal
GDP,GDP in 6money
inflation, and 36 months,
supply, and
domestic product
responded with a(GDP),
lag andinflation
starts and money
decreasing respectively.
the private sector credit in response to shocks in
after
supply 6(M2).
months whiles
Inflation real GDP
responded with aresponds
lag and
significantly after 18 months. The results the monetary
Table policy rate
1: The Variance (discount rate).
Decomposition The
Results
starts decreasing after 6 months whiles real GDP
also revealed that private sector credit Period reveal
results DR tothatDR to DR
shocks on tothe DR to DR
discount to
rate
responds
(PSCR) significantly
responds after 18
negatively to amonths.
shock onThethe GDP CPI LM2 LPSCR DR
account for a 4.4 per cent and 6.96 per cent
discount andrevealed
results also slightlythat
trends down
private overcredit
sector time. 1 0.00 0.00 0.00 0.00 93.03
variation
6 in0.15
inflation4.44
in 6 months
1.611 and0.20
36 months,
78.36
arch Bulletin Volumeresponds
(PSCR) 2 negatively to a shock on the
Figure 8: Impulse response functions for the respectively.
12 1.53This indicates
6.048 1.57 that
0.58 inflation
58.05
mial discount
SVAR and slightly trends down over time.
model
Response to Cholesky One S.D. Innovations ± 2 S.E.
18
Response of LGDP to DR Response of INFL to DR
.004 .24
.20
dynamics in4.72 5.72 are1.43
the country 0.85 46.89
largely explained by
.000
.16
24 8.85 5.89 1.51 1.17 42.76
-.004 .12
other variables and not the discount rate as it
-.008
.08
.04
30 12.62 6.52 1.44 1.73 41.52
-.012
.00
-.04
accounts
36 for a smaller
15.18 6.97portion of the2.58
1.47 variation in
40.54
Source: own calculations
-.016 -.08
5 10 15 20 25 30 35 5 10 15 20 25 30 35
inflation. A variation in money supply is explained
Figure 8: Impulse response functions for the
Response of LM2 to DR Response of LPSCR to DR
SVAR model
.010 .01
by
4.21.6Optimal
per cent Currency
of variationArea
in the(OCA)
discount rate in
Model
.005
.00
.000
-.01 6 months and 1.47 per cent in 36 months whiles
4.2.1 Pairwise Correlation
-.005
-.02
-.020 -.04
5 10 15 20 25 30 35 5 10 15 20 25 30 35
cent and 2.58
measuring thepercorrelation
cent variation in the discount
of demand shocks
Response of DR to DR
.16
rate in 6 and 36 months, respectively. Lastly, thein
across the CMA countries are presented
.12
Table 2.
results The that
indicate results revealed
discount that account
rate shocks demand
1.5
.08
Central Bank
Source: of Swaziland © 2018
Own Calculations 61 |
P a g eown calculations
Source: economy.
ure
4.1.4 Variance Decomposition Analysis
nly
oss
54 Central Bank Of Swaziland © 2018
Table 1 shows the variance decomposition
analysis for real GDP, inflation, money supply, and
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
The results reflect the effect of South Africa Table 3: Correlation of Structural Supply
as an anchor economy on CMA monetary policy Shocks across CMA countries, 2002Q1-2016Q4
implementation and how developments in Country SA SD NAM LES
this economy filter through to the other SA 1.00
member states. Even though the structural
SD 0.05 1.00
shocks are insignificant between Swaziland
and Namibia, and Lesotho and Namibia, the NAM -0.85 -0.46 1.00
correlation coefficients of the shocks are LES -1.56 0.12 0.17 1.00
Note: SA: South Africa, SD: Swaziland, NAM: Namibia, LES:
positive, indicating that they are symmetric.
Lesotho.
Exchange Shock
Exchange Shock
Exchange Shock
Demand Shock
Demand Shock
Demand Shock
Supply Shock
Supply Shock
Supply Shock
4.2.2 Variance Decomposition Analysis
The variance decomposition analysis
measures the proportions of forecast error
variance in a variable that is explained by
SA 76.2 14.0 9.8 17.5 74.5 7.96 14.2 18.6 67.2
impulses in it and by the other variables
SD 77.3 3.43 19.0 22.1 66.4 11.5 46.9 9.06 43.9
in the system. In a statistical sense, if a
NAM 73.1 18.6 8.3 8.7 80.7 10.6 21.1 18.1 60.8
variable explains most of its own shock, then
LES 69.1 26.3 4.6 6.1 81.5 12.4 15.7 40.5 43.8
it does not allow variances of other variables Source: own calculations
to contribute to it being explained and is
therefore said to be relatively exogenous.
5.0 CONCLUSION AND
Kazerooni and Razzaghi (2014) asserts
that if the variability of the forecast error RECOMMENDATIONS
variance is different between countries
then those countries would have to follow 5.1 Conclusion
different policy strategies and the chance of The study had two main objectives, namely;
establishing a common currency area should to determine the efficacy of the monetary
be put on hold. policy in Swaziland and secondly, to
determine if the CMA is an optimal currency
The results for the variance decomposition area. The study employed the use of two
analysis of real GDP, the consumer price SVAR models, one to determine the impact of
index and real exchange rate at the 12th monetary policy on selected macroeconomic
quarter are presented on Table 4. The supply variables, including inflation and the other
and demand shocks were found to be the to determine optimality of the CMA.
most predominant accounting for a larger
variability in their own variables in all the The impulse response results on the efficacy
CMA economies. At most, the own shocks to of monetary policy in Swaziland indicate that
the consumer price index (demand shocks) a shock on the discount rate has negative
and real GDP (supply shocks) account for effect on real GDP, inflation and private
almost 70 per cent of total variation in the sector credit. The variance decomposition
last quarter of the sample period in all four results indicate that the variability in GDP,
countries of the CMA. The similarities in these CPI, money supply and private sector credit
shocks accounting for larger variations in was explained by marginal variation of
these economies indicate that the countries the discount rate in the sixth month. The
can pursue similar policy strategies and results revealed that the effect of a shock
therefore are in support of the formation of on the discount rate on the other variables
the OCA. An own shock to the real effective increases with time. In 36 months, a shock on
exchange rate accounts for a combined 50 the discount rate accounts for 6.97 per cent,
per cent of total variation on average over 1.47 per cent and 2.58 per cent in variation
the 12th quarter period. in inflation, money supply and private sector
credit, respectively.
For the SVAR model the study considered observed between South Africa and Namibia,
three variables, namely; GDP as a supply South Africa and Lesotho, and Swaziland and
shock, real effective exchange rate as Namibia whilst South Africa and Swaziland,
monetary policy shock and the CPI as a Swaziland and Lesotho, and Namibia and
demand shock. In this model, the study Lesotho exhibited symmetric shocks. The
used quarterly secondary data spanning results for the pairwise correlation of real
from 2002 to 2016. All variables were non- effective exchange rate shocks across
stationary at levels and were transformed the CMA countries exhibit some level of
and differenced once to achieve stationarity. symmetry between Swaziland, South Africa
The pairwise correlation coefficients results and Namibia. The significant correlations are
for the demand shocks across the CMA synonymous with the pegged exchange rate
countries revealed that demand shocks are policy as well as the close coordination of
less correlated among the CMA member the monetary policy in the CMA. Generally,
states; instead, they have converged around the overall OCA criteria results are in favour
the South African economy. These results of the existence of the CMA as an optimal
reflect the effect of South Africa as an anchor currency area.
economy in the CMA and how developments
in this economy affect the other member 5.2 Recommendations
states. Even though the structural shocks The findings of the study supports the
are insignificant between Swaziland and Central Bank of Swaziland’s drive to align
Namibia, and Lesotho and Namibia, the policy rates with that of South Africa,
correlation coefficients of the shocks are but dependent on data outcomes. Since
positive, indicating that they are symmetric. the shocks of the LNS country are mainly
symmetric with that of the anchor economy
Mixed effects were observed for the supply (South Africa), the country’s membership
shocks correlation coefficients but all were to the CMA is therefore justified. Swaziland
insignificant. Asymmetric shocks were should therefore remain within the CMA.
Kazerooni A. and Razzaghi S. (2014). Paolo F. (2002). New views on the Optimum
Assessing the Feasibility of Common Currency Area Theory: ‘What is the
Currency Area among D-8 Group EMU telling us?’ European Central
Members: Structural VAR Model Bank Working Paper.
Approach. The International Journal
Applied Economics and Finance, 8: Peersman, G. & Smets, F. (2003). The
17-28. Monetary Transmission Mechanism in
the Euro Area: More Evidence from
Lee Y. and Azali M.(2012). Is East Asia an VAR Analysis. E u r o p e a n C e n t r
optimum currency area? Economic a l B a n k, Working Paper no. 91 .
Modelling, 29(2): 87-95.
Popescu I. V. (2012). Effects of monetary
Maturu, B., Isaya M. and Kethi K. (2011). policy in Romania: A VAR approach.
Monetary Policy Mechanism in Kenya. Munich Personal RePEc Archive.
Central Bank of Kenya Working Paper.
Ricci R., A. (1997). Exchange Rate Regimes
Mishra P., Montie P.J. and Spilimbergo A. and Location. IMF Working Paper
(20101). Monetary Transmission in Low No.97/69.
Income Countries. IMF Working Paper,
WP/10/223. Simatele, M. (2003). Financial Sector
Reforms and Monetary Policy in
Ndzinisa P. (2008). The Efficacy of Monetary Zambia.
Policy on Economic Growth in
Swaziland. Central Bank of Swaziland. Sims A. (1992). Interpreting the
Macroeconomic time Series Facts:
Omotor G. and Niringiye A. (2011). The Effects of Monetary Policy. Yale
Optimum Currency Area and Shock University
Symmetry: A Dynamic Analysis of
West African Monetary Zones (WAMZ). Zhang Z., Sato K. and McAleer M. (2003).
Romanian Journal of Economic Asian Monetary Integration: A
Forecasting, 3: 71-82. Structural VAR Approach. http://
www.cirje.e.u-tokyo.ac.jp/research/
Onyeiwu C. (2012). Monetary Policy and dp/2003/2003cf212.pdf
Economic Growth of Nigeria. Journal
of Economics and Sustainable
Development, 3 (7): 62-61.
The Impact of Monetary Policy and 0.1 per cent on domestic public debt
Changes on Macroeconomic over twenty-four months. The insignificant
effect of the reserve requirement and
Variables in Swaziland: A special
the liquidity requirement on the fiscal
focus on Fiscal Variables variables, specifically domestic public debt,
suggest minimum monetary and fiscal policy
Sive Kunene10 and coordination between authorities. The
Thandeka Mdladla11 study therefore recommends that the two
monetary policy instruments be reviewed
Abstract more regularly to enhance coordination
between monetary and fiscal policies.
The study investigates the impact of
monetary policy changes on budget deficit, Key words: Monetary policy, fiscal variables,
tax revenue and domestic public debt in Structural VAR
Swaziland. Time series data obtained from
the Central Bank of Swaziland and the 1.0 INTRODUCTION
Ministry of Finance was used to run three The 2008/2009 financial crisis inspired
Structural Vector Autoregressive (SVAR) debates on the role of central banks and their
models where the discount rate, liquidity ability to efficiently manage the economy.
requirement and reserve requirement were Generally, there is consensus in literature on
used as policy rates. In all three models, price stability being the primary objective
impulse response functions as well as variance of monetary policy (Hilbers, 2005); however,
decompositions were obtained. The impulse arguments are being raised on whether
response results revealed that all the policy Central Banks should focus on price stability
rates did not have a significant effect on the as their only objective (Bhattacharyya,
fiscal variables over a 24-months period. 2012). Reis (2016) asserts that one of the
The variance decomposition results indicate common discussions surrounding central
that a shock on the discount rate can banks is their ability to ease the fiscal burden
cause a 2.9, 1.4 and 1.5 per cent variation for a country. After investigating ways
on government deficit, tax revenues and through which central banks can alleviate
domestic public debt over a period of 24 fiscal burdens, Reis (2016) concluded that
months, respectively. A shock on the reserve there is limited scope for the central bank
requirement would cause a 0.4, 0.5 and 0.1 to lower or ease the fiscal burden.
per cent variation on government budget
deficit, tax revenues and domestic public Kaplan (2016) argues that even though
debt in twenty-four months, respectively. monetary policy is highly important, by itself
A shock on the liquidity requirement would it is not meant to address key structural
cause variation of 0.5 per cent on government issues. Kaplan further warns that monetary
budget deficit, 0.4 per cent on tax revenues policy should not be treated as a substitute
for actions, which should address structural
issues and other economic challenges.
10
Sive Kunene is an Economist, Policy Research, in
the Policy Research and Macroeconomic Analysis at Hilbers (2005) and Reis (2016) posit that
the Central Bank of Swaziland.,reachable at Sivek@ while different bodies implement monetary
Centralbank.org.sz. and fiscal policies, the two policies are not
11
Thandeka Mdladla is an Economist, Policy Research, independent of each other. A change in one
in the Policy Research and Macroeconomic Analysis policy affects the effectiveness of the other
at the Central Bank of Swaziland.,reachable at
and hence the overall impact of any policy
ThandekaM@Centralbank.org.sz.
change.
The Southern African Customs Union (SACU) these figures are below the 60 per cent
receipts, which have become the most debt-to-GDP ratio threshold stipulated in
important source of revenue for the Swazi the Protocol on Finance and Development
economy, averaging 48.6 per cent of Swazi for SADC members, the country’s ability to
Government’s total expenditure between service its debts has come under serious
the years 1999/2000 and 2016/2017, have scrutiny, as the most important element of
become increasing volatile (MoF, 2017). The debt management is the ability to service
percentage of the SACU receipts contribution the incurred debts to the satisfaction of the
to government expenditure has decreased creditors.
from 66.9 per cent in 2012/13 down to 24.8
per cent in 2016/17 as shown on Figure 1. The year 2016/17 and the first two quarters of
The volatility in SACU receipts over recent the year 2017/18 saw the government facing
years has put a strain on the fiscus (MoF, serious cash flow challenges and questions
2017), resulting in the government running were raised on the role of the Central Bank
budget deficits since 2014, which are above in ameliorating the situation. In light of
the 3 per cent recommendation of the these concerns, the departure point would
Protocol on Finance and Development for be to empirically establish how monetary
SADC which was stipulated in 2012. From a policy changes affect macroeconomic fiscal
budget surplus of 3.7 per cent in 2012/13, variables, hence this study. To the best
the government ran a deficit of 1.2 per cent knowledge of the authors and information
in 2014/15, shooting up to 12.4 per cent available, there is no published study in
in 2016/17 in pursuit of an expansionary the country, which has tried to establish
stance. This has led to an increase in public the relationship between monetary policy
debt levels. changes and fiscal variables. Specifically,
the study seeks to determine the effects of
Figure 1: SACU receipts as a Percentage of monetary policy shocks on macroeconomic
Total Expenditure and the Budget Surplus as aBulletin
Research fluctuations
Volume 2 with a special focus on fiscal
Percentage of GDP variables, namely; budget deficit, tax
15 100 published study in the country, which has tried to
revenue and domestic debt.
10 establish the relationship between monetary
80
The
policysignificance
changes and of the
fiscal study is
variables. that a
Specifically,
5
60 relationship between monetary policy
Per cent
Per cent
0
the studyand
changes seeks to determine
its effects the variables
on fiscal effects of
40
2000/01
2002/03
2004/05
2006/07
2008/09
2010/11
2012/13
2014/15
2016/17
-5
will be established
monetary and hence
policy shocks add to the
on macroeconomic
body of knowledge,
fluctuations with a which
specialwill assist
focus onpolicy
fiscal
-10 20
makers to better engage especially during
variables,
tough namely;times.
economic budget deficit, tax revenue and
-15 0
domestic debt.
The rest of the study is subdivided into four
Surplus/(deficit) (% of GDP)
The significance
sections. of the study
In section 2, theis that a relationship
paper presents
Source: Ministry of Finance
the review
between of literature
monetary between
policy changes andmonetary
its effects
Source: Ministry of Finance
policy shocks and its effects on fiscal
On average, total public external debt for on fiscal variables
variables whiles will
the bemethodology
established and of hence
the
On average, total public external debt for
Swaziland
Swazilandaveraged about
averaged 15.1 per
about 15.1cent
perof cent
GDP study is presented in section 3. The results
add to the body of knowledge, which will assist
of GDP and analysis
makersare presented
engageinespecially
section 4during
and
for the for the 2000
period periodto 2000
2016,to whilst
2016, public
whilst policy to better
public domestic debt averaged 3.7 per cent section 5 constitutes the conclusion together
domestic debt averaged 3.7 per cent of GDP for toughpolicy
with economic times.
recommendations.
of GDP for the same period. Even though
the same period. Even though these figures are
below the 60 per cent debt-to-GDP ratio The rest of the study is subdivided into four
threshold stipulated in the Protocol on Finance sections. In Bank
Central section 2, the ©paper
Of Swaziland 2018
61
presents the
review of literature between monetary policy
and Development for SADC members, the
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
Bernake, Boivin and Eliasz (2005) developed Muscateli et al. (2004) reported results
a Factor Augmented Vector Autoregression similar to Fry-McKibbin and Zheng (2012)
(FAVAR) model in order to minimize the for the US and further stated that fiscal
limitations of the VAR which have been variables adjust with a lag to the output
widely raised in most papers. The FAVAR effects caused by a contractionary monetary
combines the structural VAR and factor policy. Fry-McKibbin and Zheng (2012)
analysis for large data sets and was found reported that monetary policy also moves
to be more efficient in the analysis of the other interest rates like the treasury bills in
monetary transmission mechanism than a the same direction. Using a FAVAR, results
simple structural VAR. Indeed, a number of by Dungey and Fry (2010) revealed that in
recent studies like Rosoiu (2015), Jushan, Australia, a contractionary monetary policy
Lina and Kunpeng (2014) and Fry-McKibbin results in reduced government revenue and
and Zheng (2012) have used the FAVAR given is also associated with reduced debt-to-GDP
its properties. The FAVAR has been heralded ratios.
as one of the most plausible methodological
approach which is more inclusive in terms Haug, Jedrzejowisz and Sznajderska (2013)
of economic data to depict underlying combined the monetary and fiscal policy in
economic factors. A FAVAR model needs large a Structural Vector Auto Regression (SVAR)
data sets and is more appropriate in data and found that a positive shock on monetary
rich countries and hence this study could not policy (interest rate) had a significant
use it. The Structural Vector Autoregressive negative effect on GDP. This effect was
(SVAR) approach was used instead, which significant for more than 12 months. However,
remains one of the mostly used models in the shock did not have a significant effect
determining the effects of monetary policy on inflation, government expenditure and
using a few variables. revenue. The ordering of the variables by
Haug, Jedrzejowisz and Sznajderska (2013)
Fry-McKibbin and Zheng (2012) used the was different from that of Fry-McKibbin
Factor Augmented Vector Auto Regression and Zheng (2012) with the fiscal variables
ordered last whiles Fry-McKibbin and Zheng and the results revealed that, had Turkey
(2012) ordered them first. not undertaken counter cyclical and
discretionary interest rate cuts, the
Investigating the implications of monetary economy would have shrunk by -6.2 per
policy shocks on government debt cent instead of the -4.8 per cent that was
management in Barbados, Moore and Skeete realized. The Bank of the Republic of Turkey
(2010) reported that there is a positive reduced the policy rate by 1025 basis point
relationship between the bank rate and between November 2008 and November
treasury bills’ interest rate. An increase in 2009. During the financial crisis period,
the bank rate was found to be positively Turkey also adopted a floating exchange rate
correlated with rises in treasury interest system and the results revealed that if this
rate. Moore and Skeete (2010) also found exchange rate system was not adopted, the
a positive relationship between treasury economy would have shrunk by 8 per cent in
bills and the minimum savings rate which 2009. In the US, Mishkin (2009) argues that
is another monetary policy tool used by the the accommodative monetary policy stance
Bank and this relationship was even stronger adopted by the Federal Reserve Bank during
than that of the Bank rate. Using a SVAR, the financial crisis played an important role
Dlamini and Skosana (2016) also found a in ameliorating the effects of the recent
positive relationship between the discount financial crisis.
rate and the treasury bills interest rates in
Swaziland. Greenlaw et al. (2013) argue that monetary
policy can play a significant role in ensuring
Jannsen, Potjagailo and Wolters (2015) a sustainable outcome if the fiscal position
assert that monetary policy effects differ is moving in a positive direction from an
between crisis and non-crisis times as the unsustainable path to a sustainable path.
monetary transmission mechanisms might Greenlaw et al. (2013) further state that
be different due to credit constraints since fiscal consolidation is contractionary,
and additional effects of increases in the standard prescription for monetary
uncertainty. They found that in emerging policy is to ease during fiscal consolidation.
economies and Organisation for Economic Results by Hellebrandt, Posen and Tolle
Co-operation and Development (OECD) (2012) after examining historical records of
countries, expansionary monetary policy is fiscal consolidation in advanced economies
more effective on increasing GDP during a between the years 1978 and 2009 revealed
recession period of a financial crisis than that successful consolidations tend to be
in none-crisis times. Monetary policy was accompanied by higher monetary easing as
found to be having a negligible effect on measured by the interest rate. The authors
GDP during the recovery period. Findings further note that the easing has to start at
by Saliba (2013) revealed that monetary and the onset of the consolidation programme.
fiscal policy had a positive and significant
effect on growth of small economies post- Greenlaw et al. (2013) note that in a case
recession period. Saliba (2013) also found of fiscal dominance, a central bank has
monetary measures to be more important got no power to avoid the consequences
than fiscal measures in mitigating the effects of an unsustainable fiscal policy. Similarly,
of a recession. Hellebrandt, Posen and Tolle (2012) posit
that monetary policy cannot be effective
Alp and Elekdag (2011) investigated the when fiscal policies are unsustainable. They
impact of the monetary policy in Turkey further state that there is an argument that
during the financial crisis of 2007/2009 any monetary loosening should be preceded
established that when fiscal policy is 𝑌𝑌𝑡𝑡 = 𝐴𝐴−1 ∑between the reduced𝑒𝑒 𝐺𝐺form
𝑖𝑖=1 𝐵𝐵𝑖𝑖 𝑌𝑌𝑡𝑡−𝑖𝑖 + 𝑒𝑒𝑡𝑡 ………………...(2)
𝑡𝑡 VAR1 0residuals 0 𝑒𝑒 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷
0 𝑡𝑡 0 (=
follows; multiplying both sides by A -1 and it is is specified
𝑇𝑇𝑡𝑡 the discount 𝑎𝑎 as1 0rate 0 (Dr 0
determined before monetary policydisturbances. each In that regard, the uncorrelated,
relationship 𝑒𝑒 21 𝑒𝑒 𝐺𝐺𝐺𝐺𝐺𝐺 𝑡𝑡
where is a vector structural of serially shocks 𝑒𝑒(𝜀𝜀
𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡
) can but
𝑎𝑎31 be 𝑎𝑎32Where1
expresse
𝑒𝑒 0 the
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑒𝑒0𝐺𝐺𝑡𝑡
period, the monetary authority A𝑌𝑌 reducedshould
−1 ∑ form
𝑝𝑝
follows; of the model is obtained by 𝑡𝑡
structural VAR is specifie
𝑡𝑡 =between
𝐴𝐴 𝑖𝑖=1 𝐵𝐵where
the𝑖𝑖 𝑌𝑌𝑡𝑡−𝑖𝑖 + 𝑒𝑒𝑡𝑡 ………………...(2)
reduced is a form vectorVAR of seriallyresiduals uncorrelated,
𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷
𝑒𝑒( ) 𝑡𝑡 and = 𝑎𝑎41 𝑎𝑎42 𝑎𝑎43[ 𝑒𝑒1 𝑒𝑒]0𝑇𝑇𝑡𝑡 𝐷𝐷𝐷𝐷
focus exclusively on stabilizingmultiplying inflation.both notbutsides necessarily
notby A 𝑝𝑝follows:
necessarily
-1 and orthogonal,
it is specified orthogonal, reduced
as 𝑒𝑒 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃
𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡
reduced form 𝑎𝑎51 𝑎𝑎52 𝑎𝑎53 𝑎𝑎54𝑒𝑒 𝐺𝐺𝐺𝐺𝐺𝐺
a 𝑎𝑎vector 1
o
where is a vector of serially −1 uncorrelated,
∑ but
They argue that since too much inflation is disturbances. 𝑌𝑌 𝑡𝑡 = 𝐴𝐴 𝐵𝐵
𝑖𝑖=1 𝑖𝑖 𝑡𝑡−𝑖𝑖 𝑌𝑌 + 𝑒𝑒𝑡𝑡 ………………...(2) 𝑒𝑒 𝑎𝑎
𝑒𝑒 𝑡𝑡 61 162 063 64
𝐺𝐺 𝑎𝑎 𝑎𝑎 0𝑒𝑒 65 𝑎𝑎
𝐷𝐷𝐷𝐷𝐷𝐷
0
follows; structural shocks (𝜀𝜀In form disturbances. 𝑡𝑡 ) that can regard, Inbe expressed
that regard,
the [relationship
𝑒𝑒 𝐷𝐷𝐷𝐷 ]𝑒𝑒 𝑇𝑇as the
𝑡𝑡 [𝑎𝑎71 𝑎𝑎72 21the
𝑎𝑎173 𝑎𝑎74 0 𝑒𝑒 𝑎𝑎 0
𝐺𝐺𝐺𝐺𝐺𝐺
generated by lack of monetary notcommitment,
necessarilywhere orthogonal,
relationship is a reduced
vector
between of formreduced
serially
the uncorrelated, form 𝐺𝐺𝐺𝐺𝐺𝐺VAR but Where right 75
th
𝑎𝑎 𝑎𝑎 1 0
𝑃𝑃𝑃𝑃𝑃𝑃
an endogenous fiscal policy requires follows: an −1 ∑ between
𝑝𝑝 the reduced form VAR residuals 𝑒𝑒(𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 ) 𝑡𝑡and 31 32 𝑒𝑒 𝐷𝐷𝐷𝐷
𝑡𝑡
𝑌𝑌 𝑡𝑡 = 𝐴𝐴 𝐵𝐵
residuals 𝑌𝑌 + 𝑒𝑒 ………………...(2)
𝑖𝑖 𝑡𝑡−𝑖𝑖 ( 𝑡𝑡 ) and structural shocks 𝑒𝑒 𝑡𝑡 = 𝐴𝐴 𝜀𝜀 …………………….……
0( 𝑡𝑡 ) can 41coefficient
𝑒𝑒 = 𝑎𝑎 𝑎𝑎
a42vector 𝑎𝑎 43 𝑒𝑒 1
[
inflation conservative centraldisturbances. In that 𝑖𝑖=1 regard, the relationship
banker. not necessarily
be expressed
structural shocks as orthogonal,
follows:
(𝜀𝜀𝑡𝑡 ) can but be Wherereduced
expressed 𝑒𝑒 𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡
form
the left hand side 𝑎𝑎 𝑎𝑎 𝑎𝑎 53 of 𝑎𝑎5
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 as 𝑎𝑎
51 52
where is a vector
between the reduced form VAR residuals ( ) and of serially uncorrelated, 𝑒𝑒 the
𝑎𝑎 𝑎𝑎right 𝑎𝑎6
61
structural
62 63
disturbances.
follows: In that regard, the a vector relationship
[ 𝑒𝑒 𝐷𝐷𝐷𝐷of ]residuals [𝑎𝑎71 𝑎𝑎72 in𝑎𝑎73 the 𝑎𝑎7
3.0 METHODOLOGY not necessarily
structural shocks (𝜀𝜀𝑡𝑡 ) can𝑡𝑡 be 0expressed
𝑒𝑒
orthogonal, = 𝐴𝐴 𝜀𝜀
Based 𝑡𝑡 …………………….……..(3)
reduced on the
asform Cholesky decomposition coefficien
Wher o
between the reduced form VAR residuals the right( hand ) and side
The is the
ident sq
disturbances. In that regard,reduced the relationship formdecomposition
VAR, the studyofimposes structura
aconst vect
3.1 The Structural Vector follows: Based on the Cholesky Where
coefficients the left
associated hand withside
structural shocks 𝑒𝑒𝑡𝑡 = 𝐴𝐴 0 𝜀𝜀
) can
(𝜀𝜀𝑡𝑡VAR, 𝑡𝑡 …………………….……..(3)
bestudy expressed as this stud
Autoregression Model (SVAR) between the reduced
Based on the Cholesky the form
reduced VAR residuals
form
that decomposition (
define matrix A0 of )
the and as imposes
athe lower triangular rim
a vector of residualsThethe iden
Fry-McKibbin and Zheng (2012) and Naceur, follows: constraints that define matrix Astructural 0 as a lower shocks throughinth
structural shocks 𝐴𝐴0 𝜀𝜀(𝜀𝜀 𝑡𝑡 ) can be expressed
The identification as scheme follows Analyzing the or
Boughrara and Ghazouani (2009) assert reducedthat𝑒𝑒𝑡𝑡 = form
triangular …………………….……..(3)
𝑡𝑡VAR, the
matrix.
Based on the Cholesky decomposition study The imposesidentification constraints the schemeright
of the hand this
side coeffiisstud th
The identification of
cycles, the SVA
empirical analysis of the effects of follows:
monetary follows the original
that define reduced matrix form A0 VAR, aspaper a lower
the
paper
by
study Sims
triangular
by (1980),
imposes
Simscoefficients
matrix.
(1980),
constraints whereby Analyzing
associatedthe struct ChF w
policy have been largely investigated with whereby the𝑒𝑒 Choleski = 𝐴𝐴 𝜀𝜀 decomposition
…………………….……..(3) this study is is largely info
Based on the Cholesky decomposition Central
𝑡𝑡 of𝑡𝑡 the
0Bank of Swaziland © 2018 shocks
structural governme
cycles,throug
Vector Autoregressive (VAR) models. The As applied
identification
that definetomatrix theP acontemporaneous
scheme Ag0 easfollows a lower triangular theAnalyzing parameter
original matrix. the effects The
of fisc id
alluded to in the limitations reduced
of theform study VAR, the 𝑒𝑒
matrix = 𝐴𝐴 𝜀𝜀
studyA0imposes …………………….……..(3)
. constraints governm
𝑡𝑡 0 𝑡𝑡 that othe
and the literature review, the mostpaper ideal Based byTheSims identification
(1980), scheme wherebyfollows the cycles, theThe
Choleski original
identification
Fatas and this ofMiho the
that define matrix A0 as a on lower thetriangular Cholesky matrix. decomposition of the that othe
methodology for the study would have beenBank
Central paper
Estimating by Sims
of Swaziland the©(1980), impactwhereby
2018 of monetary thethis Choleski
policy
study is largely
The Based on
identification the scheme Choleskyfollows decomposition the original of the government spendingAnaly first
a FAVAR model as cited in Rosoiu (2015), Central P a g e reducedshockBank form (interest VAR,
of Swaziland therate), study imposes
© 2018Fry-McKibbin and constraints
reduced form(1980), VAR, a g the
P Zheng e study imposes constraints ordered thatAnalyzing other variables the effects likeofo
cycles
Jushan, Lina and Kunpeng (2014) paper and
by Sims Fry- whereby (2012) the recursively
Choleski the
that define matrix A0 as a lower triangular cycles, matrix. Fatas and M
thatBank
Central define matrix A©0 as
of Swaziland 2018 a lower triangular matrix. gover
The identification scheme follows the original spending
64
Page government
Central Bank Of Swaziland ©The
2018 identification scheme follows the original that
paper by Sims (1980), whereby the thatCholeski other variables lik
paper by Sims (1980), whereby the Choleski
ms (1980),
large set of data whereby is applied
decomposition the to the
Cholesky
contemporaneous
untry hence the parameter matrix A0.
osition is applied Cto the
ENT R A L contemporaneous
B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
Autoregressive
Estimating the impact of monetary policy shock
er matrix A. Thus, the order of the variables
(interest rate), Fry-McKibbin and Zheng (2012)
rcontains
to manyseven studies
variablesused
from infirstthe
recursively ordered context
the variables
to last
from of
first to last
as follows; total Fry-Mckibbin and Zheng (2012) ordered
rix form of the as follows; expenditure
total government (Gexpenditure (Gt),
advancedgovernment
ns follows; economies, including t), total taxation
Sims absorption (gross national expenditure) after
revenue (Tt), real
total taxation gross(Tnational
revenue t), real gross expenditure
national the fiscal variables in order to capture the
and is as follows;
…………..(1)
(GNE output
t), the
expenditure (GNEt(GDP),
ratio ),held inflation
by held
the ratio thebypublic to toGDP
the public contemporaneous effects of public spending
(Deb t), (Deb
GDP realt), GDP, inflation
real GDP, inflation(Inf(Inftt), augmented
), augmented and tax changes on output and absorption,
), money supply
ntaining the 7 factors
(M2), private sector credit
(C) and the interest rate (rt). This hence this study also orders absorption as
factors (C) and the interest rate (rt). This study
study adopts an ordering similar to that of the third variable. According to Fry-Mckibbin
nd the
quare policy
matrix of rate
adopts(DR). The
an ordering matrix
similar form
Fry-McKibbin and Zheng (2012) but without
to thatofof Fry-
and Zheng (2012), government spending and
is a vector of McKibbin and Zheng (2012)and but the without the
Rally model can the
be augmented
expressed as factors
follows: inflation taxation affects the level of public debt
orthogonal rate. This study
augmented factors also
and the adds private
inflation sector
rate. This which by extension affects GDP. The debt-
the number of credit
studyas a adds
also variable
privateafter real as
sector credit GDP and the
a variable to-GDP ratio is therefore ordered fourth
𝐺𝐺𝑡𝑡
last variable to be ordered
after real GDP and the last variable is the discount
to be ordered in the model whiles GDP is ordered fifth.
1 rate 0 (Dr).0 The 0matrix 0 form 𝜀𝜀 𝐺𝐺𝐺𝐺𝐺𝐺
of
𝑡𝑡
the structural The inclusion of the debt ratio is important
is the discount rate (Dr). The𝐶𝐶𝐶𝐶𝐶𝐶
𝐶𝐶𝑡𝑡 𝑎𝑎21 VAR 1 is 0
specified 0 𝜀𝜀 matrix
0 as follows; 𝑡𝑡 form of the
because it provides a feedback loop between
is obtained by
𝑡𝑡 = 𝑎𝑎31 𝑎𝑎32structural
1 VAR 0is specified
0 as𝜀𝜀follows;
𝑀𝑀𝑀𝑀𝑡𝑡
the fiscal variables and the public debt level
it is specified as
𝑃𝑃𝑡𝑡 𝑎𝑎41 𝑎𝑎42 𝑒𝑒 𝑎𝑎43 1 10 0 00 𝜀𝜀0 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅
𝐺𝐺𝑡𝑡 0 0 𝑡𝑡 𝜀𝜀 𝐺𝐺𝑡𝑡 (Fry-Mckibbin & Zheng, 2012) and improves
𝑡𝑡 ] [𝑎𝑎51 𝑎𝑎52 𝑒𝑒 𝑎𝑎53 𝑎𝑎 𝑎𝑎𝑎𝑎54 1 10] [0 𝜀𝜀 𝐷𝐷𝐷𝐷
𝑒𝑒 𝑎𝑎
𝑇𝑇𝑡𝑡 1
𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡
0 021
31
0 0 0
0 0
32
𝑡𝑡 ] 𝜀𝜀
𝜀𝜀 𝑇𝑇𝑡𝑡
𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡 the stability of the model (Fry & Pagan,
…………...(2) 𝑒𝑒 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑡𝑡 = 𝑎𝑎41 𝑎𝑎42 𝑎𝑎43 1 0 0 0 ∗ 𝜀𝜀 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝑡𝑡 …..(4) 2005). What happens in the economy in
𝑒𝑒 𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡 𝑎𝑎51 𝑎𝑎52 𝑎𝑎53 𝑎𝑎54 1 0 0 𝜀𝜀 𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡
ncorrelated, but 𝑒𝑒 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑎𝑎61 𝑎𝑎62 𝑎𝑎63 𝑎𝑎64 𝑎𝑎65 1 0 𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃
terms of GDP developments is more likely
[ 𝑒𝑒 𝐷𝐷𝐷𝐷 ] [𝑎𝑎71 𝑎𝑎72 𝑎𝑎73 𝑎𝑎74 𝑎𝑎75 𝑎𝑎76 1 ] [ 𝜀𝜀 𝐷𝐷𝐷𝐷 ] to impact on private sector credit hence
he left hand
reduced form side of the equation contains a
private sector credit is ordered sixth in the
he relationship
f residuals inWhere
the reduced
the
Where form,
lefthand
the left hand and
side
side of on thecontains
of the
the equation equation model. As in Fry-Mckibbin and Zheng (2012),
esiduals ( ) and the study orders the interest (discount rate)
contains
a vectoraofvector ofinresiduals
residuals the reducedinform,
the and
reduced
on
nd side is
expressed the squared
as form, andhand matrix
on side
the (A0) of is the last. This study runs three parsimonious
the right is right hand
the squared side
matrix (A0) of
squared matrix (A 0) of coefficients associated
models, and replaces the discount rate with
nts associatedcoefficients
with lagged
associatedvariables and and
with lagged variables
reserve requirement in the second model
with lagged variables and structural shocks
structural shocks through the column vector (ɛ ). and the liquidity requirement which are
al shocks through
………….……..(3) throughthe
thecolumn
column vector
vector (ɛ).
The identification of the SVAR model followed in other policy rates used by the Central Bank
nometric identification of monetary
Thethisidentification
study is largely of thepolicy
informed SVAR
by model
literature.
of Swaziland.
position of the
followed
is constraints
oses crucial any inmodel
to Analyzing this study is largely informed
specification,
the effects of fiscal policy on business
Impulse response and variance decomposition
by literature. Analyzing the effects of fiscal
cycles, Fatas and Mihov (2001) ordered analysis were carried out to determine
g SVARs. The
iangular matrix. policy
SVAR on identification
business cycles,exercise Fatas and Mihov
government spending first on
ws the original (2001) ordered government spending first
the assumption the effects of policy rate shocks in the
in this paper isthat explained as follows;
other variables like output, cannot affect estimated model. The impulse response
by the Choleski on the assumption that other variables like
output, cannot affect government spending function in a VAR analyses traces the effects
ordering of contemporaneously.
output and price Fry-McKibbin
level at theand 72 |
Zheng on the system when the model receives a
(2012) also ordered government expenditure shock of say one standard deviation. The
nning is because they react to an innovation variance decomposition reveals the amount
first on the same bases hence this study
e monetary policy rate with
also orders a lag due
government to their first in
expenditure of contribution of each variable to other
the model. Results by Nxumalo and Hlophe variables in the system. In essence, it shows
movement in nature.
(2016) revealed that the Swazi Government the amount of the forecast error variance of
follows a spend and tax approach hence each of the variables that can be explained
this study orders tax revenues second after by exogenous shocks to the other variables
government expenditure. in the system.
58 | P a gFry-McKibbin
e
and Zheng (2012) ordered tax second since
government expenditure is likely to affect The study uses monthly data spanning from
taxes. 2000M01 to 2015M12 to examine the impact of
Before uses
the study conducting the Johansen
the AIC and test, the 0.0 0.0
cts the appropriate model, -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 -1.5 -1.0
Inverse Roots of AR Characteristic Polynomial
-0.5 0.0 0.5 1.0 1.5
Prediction
stimation Errorthree
of the (FPE) suggest
most a lag of order 1.0
2. Given the different suggested lag orders,
threethe models indicated
study uses the
the AIC and the FPE, which
0.5
at model 3 isinvolves
model, which the the
most
estimation of the -1.5
three most relevant models. All three models -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
mum theEigenvalue
trace statistic results
results revealed that
4.4. Impulse Response Functions and
model 3 is the most appropriate with 4.4. at most Impulse Response Functions and the
the Variance Decomposition
e at most two cointegrating
five cointegrating vectors. The Maximum
Eigenvalue results indicate that there are at Variance Decomposition
4.4.1. The Impulse Response for a Shock
and 3.mostBased on the trace
two cointegrating vectors in model 2
on the Discount Rate
and 3. Based on the trace statistic 4.4.1. results, The Impulse Response for a Shock on the
l 3 was selected as theasbest The impulse response results presented
model 3 was selected the best model for
on Figure 3 indicate that a one standard
this study, and the variables were confirmed Discount Rate
y, and the variables were
to be cointegrated.
deviation shock to the discount rate does
The impulse response
not have results
a significant presented
effect on Figure
on all the
egrated.
3 indicate that a one standard deviation shock to
ests 66 Central Bank Of Swaziland © 2018
the discount rate does not have a significant effect
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
requirement and the discount rate shocks. The Response of SURPLUS_DEFICIT to DLDR
Response of DLT AX to DLDR
months’ period, including the fiscal variables effect on the fiscal variables, just like
Response of DLDR to DLDR
.03
.20
.04
.15
.03 .02
.10
shockreserve
also does not have a and
significant effect on
.02
(government budget deficit, tax revenue and the requirement the discount
.01
.05
.01
.00
.00 .00
-.05
-.01
-.01
the public domestic debt). rate shocks. The shock alsofordoes not have
-.10
-.02 2 4 6 8 10 12 14 16 18 20 22 24
2 4 6 8 10 12 14 16 18 20 22 24
credit,
Research Bulletin Volume 2which has a negative and significant
.010 .0075
.04
.0025
Research Volume 2
.00
.000
.0000
-.005 -.04
-.0025
Figure
Figure 5: Impulse Response
Responseforforaa Shock
.00
5: Impulse Shock
.00 .15 .00
.03 .02
-.05
.04 -.01
.10
Liquidity
Liquidity Requirement
Requirement
.00
.00 .00
.00
Response of DLGNE to DLDR -.05
-.01 Response of DLDEBT to DLDR Response of DLGDP to DLDR
-.02 .015 -.01
-.10 .0100
.08 2 4 6 8 10 12 14 16 18 20 22 24
-.02
2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12 14 16 18 20 22 24
Response to Cholesky One S.D. Innovations ± 2 S.E.
Source: Own calculations credit, the
whichother
has avariables
negative except for private sec
and significant
.010 2 4 6 8 10 12 14 16 18 20 22 24 .0075
.04
Response of DLLIQRQ to DLLIQRQ
.005 .0050
Response of SURPLUS_DEFICIT to DLLIQRQ Response of DLTAX to DLLIQRQ
.04
.00 .0025 .03
.000 Response of DLGNE to DLDR .15
Response of DLDEBT to DLDR .0000
Response of DLGDP to DLDR .03
.015
-.005 -.04
.08 .0100 .10 .02
-.0025 .02
response.
credit, which has a negative and signific
-.010
Requirement
2 4 6 8 10 12 14 16 18 20 22 24
response.
.06 2 4 6 8 10 12 14 16 18 20 22 24
-.010
.02
-.02
.06 2 4
Response of DLPSCR to DLDR
6 8 10 12 14 16 18 20 22 24
Liquidity Requirement Response
Response to Cholesky One of DLDEBT to DLLIQRQ
S.D. Innovations ± 2 S.E.
.012
Response of DLGDP to DLLIQRQ
Liquidity Requirement
.10 .02
.02 .025
.00 .005 .000
indicate that
Source: Own a shock on the reserve requirement
calculations
-.01
-.05 -.050
Source:
-.010
Own calculations
-.02 Response of DLLIQRQ to DLLIQRQ 2 4 6 8 10 12 14 16 18 20 22 24
2 4
2
6
4
8
6
10 12
8 10
14
12
16
14
18
16
20
18
22
20 22 24
24 -.10 Response of SURPLUS_DEFICIT to2DLLIQRQ
-.01 Response of DLTAX to DLLIQRQ
Requirement
4 6 8 10 12 14 16 18 20 22 24
.04 2 4 6 8 10 12 14 16 18 20 22 24
.03
.15
.03
.10 .02
Reserve Requirement
2 4 6 8 10 12 14 16 18 20 22 24
discountRequirement
.015 2 4 6 8 10 12 14 16 18 20 2
theanddiscount rate,credit.
the impulse response Own calculations Response of DLGNE to DLLIQRQ
Response of DLDEBT to DLLIQRQ
.012
results
discountindicate rate, thatthe aimpulse
shock onresponse
the reserve results4.4.4. Variance Decomposition for the Discount
.050 .004
.010
Response of DLPSCR to DLLIQRQ
.025
Rate
-.050
-.010 2 4 6 8 10 12 14 16 18 20 22 24
2 4 6 8 10 12 14 16 18 20 22 24
gross
the Reservenational expenditure, GDP and private 4.4.4.
Requirement
.06
sector credit.
Response to Cholesky One S.D. Innovations ± 2 S.E.
.02
Response of DLT AX to DLRSVRQ
Response of DLRSVRQ to DLRSVRQ Response of SURPLUS_DEFICIT to DLRSVRQ
Rate
.03
a minimal
shock to the effect
discount on the
has a fiscal
minimal variables.
effect on A
the
.08 .15
.00
.04 -.02
.05
.01 2 4 6 8 10 12 14 16 18 20 22 24
.02
shock Source:
on the own calculations
discount rate causes a 2.9 per
.00
Figure 4: Impulse
ImpulseResponse
Responsefor forthe
the Shock onon
.00
.015
Response of DLGNE to DLRSVRQ
causes a 2.9
Response of DLRSVRQ to DLRSVRQ
per cent, 1.4 per
on government budget deficit, tax revenues cent and 1.5 per the
centDiscou Response toResponse
Cholesky
.100
.075
One S.D.
of DLDEBT Innovations ± 2 S.E.
to DLRSVRQ
on Rate
.06 .050
.10 .02
period,
fiscal respectively.
variables. A shockOver on the thediscount
same period,
rate
-.05
-.02 -.004
-.010 -.050
-.01 2 4 6 8 10 12 14 16 18 20 22 24
2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12 14 16 18 20 22 24
-.04 -.10 2 4 6 8 10 12 14 16 18 20 22 24
months’
cent to period,
the 1gross respectively.
national a Over
thatexpenditure, the same0.4 deviat
.020 .100
credit
cent tothe hence
GDP and it 3.8
is an effective monetary
perper cent to private sector
.02
policy instrument.
-.02
Source:Liquidity Requirement
2 4 6 8 10 12 14 16 18 20 22 24 .012
.020 .100
own calculations
Source: Own calculations
.015 .075
.010 .050
per centvariation
to the gross national
on expenditure,
government budgetper deficit,
.004
.005 .025
standard Liquidity
deviation shock
Requirement to the liquidity revenues and domestic public debt over a
Liquidity Requirement an effective monetary policy instrument.
credit. These results indicate that the discount
Response of DLPSCR to DLRSVRQ
The
The results
requirementresultsdoes on not
on
.04
Figure
have55aindicate
Figure indicate
significant that a aone
effect
that one
on
.02
fiscal variables,
standard deviation
-.02
67
cent
Discount Rateto GDP and 3.8 per cent to private sec
Liquidity Requirement Central Bank Of Swaziland © 2018
Central Bank of Swaziland © 2018 credit. These results indicate 75that
| the disco
Page
The results on Figure 5 indicate that a one
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
Table 1: Variance Decomposition for the credit and 0.2 per cent variation on GDP
Discount Rate over the same period.
Period Discount Gov. Tax GNE Debt GDP PSCR
rate Deficit
Table 3: Variance Decomposition for the
1 100.0 0.000 0.0 0.00 0.00 0.00 0.00 Liquidity Requirement
4 93.02 0.637 0.9 0.28 1.02 0.23 3.87 Liquidity Gov. Tax GNE Debt GDP PSCR
8 90.92 1.673 1.36 0.35 1.53 0.26 3.90 Req. deficit
12 90.14 2.345 1.37 0.44 1.54 0.31 3.86 1 100.0 0.00 0.00 0.00 0.00 0.00 0.00
16 89.75 2.674 1.36 0.50 1.53 0.34 3.84 4 97.82 0.45 0.37 0.17 0.06 0.01 1.12
8 97.69 0.48 0.40 0.21 0.06 0.01 1.14
20 89.57 2.828 1.36 0.54 1.53 0.35 3.83
12 97.69 0.48 0.40 0.21 0.06 0.01 1.13
24 89.48 2.900 1.36 0.55 1.53 0.36 3.83
Source: own calculations 16 97.68 0.48 0.40 0.21 0.06 0.02 1.13
20 97.68 0.48 0.40 0.21 0.06 0.019 1.14
4.4.5. Variance Decomposition for the 24 97.67 0.49 0.40 0.21 0.06 0.02 1.14
Reserve Requirement Source: own calculations
but it also helps cash strapped consumers Prinsloo (2002) claims that household debt
to quicken payments for present needs ratios are an important analytical tool
than to prolong purchases into the future. which allows policymakers and economic
Nevertheless, increased borrowing that researchers to evaluate a household’s
escalates to unsustainable levels raises financial situation and to predict the
concerns about an individual’s ability to outcome of final consumption expenditure.
repay what they owe, especially in the In Swaziland, the household consumption
event of an unexpected change in the expenditure currently represents about
economic environment (Tudela & Young, 76 per cent of the gross domestic product
2005). Martinez- Carrascal and Del Rio (2004) (Central Statistics Office, 2015).
states that high debt levels imply a higher
debt service burden which can restrict the 1.1 Household Debt in Swaziland
household’s access to additional external Total household debt in Swaziland comprises
fund hence leaving a household defenseless of three components which include; credit
against unexpected shock to their income, extended to the housing sector, motor
assets or interest rates. vehicles, and other unsecured loans. As
shown in Figure 1, credit extended to the
To a number of countries household housing market contributes the largest share
consumption expenditure accounts for a of total household credit followed by loans
larger share of their gross domestic product extended for other unsecured loans such as
(GDP). According to the World Bank (2016), education and personal. Credit extended for
among high income countries final household the acquisition of motor vehicles remains
consumption expenditure contribution to the least in terms of contribution to the
gross domestic product between the year total household sector credit.
2000 and 2015, averaged an estimated
59.9 per cent while heavily indebted poor Figure 1: Household Credit Share
countries recorded an estimated 74.2 per
cent contribution. In the same period, the
Sub-Saharan African region’s consumption of
households grew slightly above that of high
income countries to eventually close at 66.3
per cent of GDP.
the substantial growth in household credit credit as shown by the household credit to
include a decline in prime lending rates from GDP ratio increasing from around 7% in 2006
a high of 9 per cent in 2006 to a low of 5.8 per peaking at 9.5% in the third quarter of 2011,
cent in 2015 combined with the competitive before declining to around 8.2% in 2013
credit products aggressively offered by before rising again to around 9.4% in 2015.
commercial banks to the household sector. This accommodative monetary policy stance
According to Khan et al (2016), recent favoured household credit as it generally
trends also show that the cost of living has accelerated during this period.
increased significantly especially due to
rising food prices which put upward pressure The expansion in lending to households
on household consumption especially among requires the Central Bank to respond
low and middle income population. Such appropriately through restrictive monetary
increases restrict household consumption policies to contain unwarranted increases
and may compel a household to acquire in credit extended to the household sector
credit in order to maintain or enhance which safeguards the sustainability of
consumption. However, increased household the country’s financial system. The Bank
borrowing relative to household income does this conscious of the fact that its
raises concerns about a household’s ability focus is not limited to the dynamics in
to service the loan, and has economic the household sector but the economy as
implications. a whole. The Central Bank of Swaziland is
ultimately tasked with the responsibility of
Figure 2: Share of Private Sector Credit ensuring price stability and sound financial
systems that ensures a long-term viable
growth path. However, notwithstanding the
Bank’s presence, household credit remains
elevated.
8.00
Source: Central Bank of Swaziland. 7.50
7.00
Figure 3 shows the ratio of credit extended 6.50
to households to GDP between the period 6.00
2006Q1
2006Q3
2007Q1
2007Q3
2008Q1
2008Q3
2009Q1
2009Q3
2010Q1
2010Q3
2011Q1
2011Q3
2012Q1
2012Q3
2013Q1
2013Q3
2014Q1
2014Q3
2015Q1
2015Q3
the model assumes that a household can A study in Australia which was conducted by
maximize utility over its lifetime subject to Meng et al (2011) explored the determinants
a budget constraint. of household debt by employing a Vector
Autoregressive (VAR) model. Findings of the
Another theory called the permanent income study revealed a positive and significant
hypothesis, developed by Milton Friedman impact of GDP, followed by house prices and
in 1957, also weighed in on the subject of the number of new dwellings on household
household credit and consumption. Friedman debt in Australia. Over and above that,
in this theory argued that consumption interest rates, unemployment rate and
should not depend on current disposable inflation were found to have a negative
income alone but on expectations relating effect on Australia’s household debt. From
to the total income which individuals expect the results, it was recommended that given
to earn during their lifetime. The model the large negative effect of uncontrolled
emphasized that consumers utilize the saving debt levels regulating and standardizing
and borrowing option to smoothen their the practice of mortgage financing could
consumption pattern in response to random limit the irresponsible behavior of financial
and temporary changes in their income institutions.
from year to year. From this perspective,
indebtedness is the result of a rational Rahman and Masih (2014) studied the
decision aimed at maximizing intertemporal relationship between household debt and
utility, now based on the wealth and income GDP, interest rate and house price in Malaysia
expected during the life cycle (Santos et al, using a Johansen test for cointegration.
2014). The results showed that there was a long
run positive relationship between house
Although rising household debt can be prices and household debt. In the short run,
attributed to the rationales advanced by changes in GDP, lending rates and house
the assertions highlighted above, there are prices according to the study findings could
many other important economic factors and not influence household debt.
reasons why household debt continues to
rise. One of the reasons relates to a drop in Khan et al. (2016) further explored the
real and nominal interest rates. The Bank of determinants of household debt in Malaysia
Canada (2012) claims that the upward trend by disaggregating the household sector
in household debt in recent years, especially into specific components of mortgage and
mortgage credit, is consistent with the growth consumer debt. By utilizing the bound
in population and ownership coupled with test for cointergration through the ARDL
improved mortgage affordability. Although modelling approach, the study revealed that
house prices may have risen, mortgage in the long run, a change in income level,
demand and financing has benefited from housing price and population had a positive
other factors such as income gains and lower impact on mortgage debt while a rise in
interest rates. interest rates and cost of living (consumer
price inflation) would exert a negative
2.2 Empirical Literature impact.
The discussion of the related literature is
provided in this section and mainly focuses Mutezo (2014) examined the relationship
on the main factors that influence household between household debt and consumption
debt among the developing economies spending in South Africa. Using the ARDL-
including the Southern African region and bounds testing approach the results revealed
the developed world. that there was a significant deterministic
relationship between household debt Correction Model to analyze the effect gross
and disposable income, net wealth and domestic product per capita, interest rates,
inflation. The results further indicated that inflation, household consumption and money
there was a long run relationship between supply had on household debt. Results from
household debt and income, interest rates this study showed that in the long run GDP
and inflation implying that low interest rates per capita, interest rates and money supply
and a general increase in household income determined changes in household debt in
during the period 2004-2011 supported high Botswana.
household indebtedness.
Panic (2010) in Sweden examined the
Cross country evidence from a study underlying factors that were responsible for
conducted by Coletta et al. (2014) of 32 the developments in the household debt to
countries which included 26 members of the disposable income ratio which had remained
European Union plus Japan, South Korea, relatively high for almost 30 years. Using the
Canada, Australia, New Zealand and the U.S. Engel-Granger two-step modelling approach
over the period 1995 to 2011 revealed that to cointegration, the results indicated that
household debt was highest in countries that household debt to disposable income ratio
had higher per capita GDP and household was determined by debt to assets, interest
wealth. The coefficient of GDP growth rate payments to disposable income and real
on the other hand was found to be negative estate price index in the long run. Inflation
and statistically significant, implying that was found to be a significant determinant
household debt is at its highest when only in the short run.
economic activity (growth rate) is at its Tu (2008) employed the multivariate
lowest. regression model to examine the lagged
effect of house prices and interest rates on
Mah et al. (2013) studied the household household debt to disposable income ratio in
debt and how it responded to shocks from New Zealand. Using data spanning the period
several macroeconomic variables over the from the first quarter of 1991 to first quarter
period 1985 quarter one to 2012 quarter 2008, the study observed that the 1980s
two. Using the Variance Decomposition and financial deregulation and the consequent
Generalized Impulse Response Function drops in interest rates and the increase in
analyses the study found that distortions in house prices were the main causes of the
the level of household debt in South Africa increased household debt in New Zealand.
were explained by the contributions in house
prices, household income, gross domestic A study by Hong (2011) explored the
product and the real prime rate. Based on determining factors of household
the study findings, the authors recommended indebtedness in the U.S. using quarterly data
that government should intervene by closely over the period of 1980-2010. The result
monitoring and properly managing the from a simple regression analysis showed
housing market to guard against excessive that the unemployment rate, interest rate,
credit to the housing sector. disposable personal income per capita,
share of retiring population and educational
Zimunya and Raboloko (2015) studied the attainment were negatively related to the
factors behind the growth of household debt household debt in the U.S. while housing
in Botswana. Using quarterly data from the prices, consumer confidence and share of
first quarter of 1994 to the second quarter working-age population were positively
2012, the study employed the Vector Error related to the household borrowing.
𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 Johans
Kilman (2016) investigated the effect of = 𝛽𝛽0 + 𝛽𝛽1 𝐿𝐿𝐿𝐿𝐿𝐿 𝐶𝐶𝐶𝐶𝐶𝐶 + 𝛽𝛽2 𝐿𝐿𝐿𝐿𝐿𝐿 𝑃𝑃𝑃𝑃 + 𝛽𝛽3 𝐿𝐿𝐿𝐿𝐿𝐿 𝐺𝐺𝐺𝐺𝐺𝐺
bound
household, corporate and government debt + 𝛽𝛽4 𝐿𝐿𝐿𝐿𝐿𝐿 𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽5 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿
on economic growth using panel data for 20 +µ (1) the m
advanced economies between 1980-2014. 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 integr
Through a dynamic panel data model, the = 𝛽𝛽0 + 𝛽𝛽1 𝐿𝐿𝐿𝐿𝐿𝐿 𝐶𝐶𝐶𝐶𝐶𝐶 + 𝛽𝛽2 𝐿𝐿𝐿𝐿𝐿𝐿 𝑃𝑃𝑃𝑃 + 𝛽𝛽3 𝐿𝐿𝐿𝐿𝐿𝐿 𝐻𝐻𝐻𝐻
+ 𝛽𝛽4 𝐿𝐿𝐿𝐿𝐿𝐿 𝑃𝑃𝑃𝑃𝑃𝑃 + 𝛽𝛽5 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 oppos
study concluded that there is a relationship
+µ (2) metho
between household debt and economic
growth in the long run. 3.2 Econometric Estimation Procedure regres
The study invokes three steps of estimation mutua
77
procedures which seek to examine the integr
Central Bank Of Swaziland © 2018
determinants of household debt in Swaziland and be a
its impact on economic growth. The first step explan
Thus in order to avoid such cases, stationarity
For long run relationships, the ARDL normally
tests of the variables are performed. For this
involves two steps of estimation. The first step is
purpose, the Augmented Dickey-Fuller
C E N T R A L B A N test
K O F(ADF)
S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
to examine the existence of a long-run
is employed.
Research Bulletin Volume 2 cointegration relationship among the variables in
After
3.2 the order ofEstimation
Econometric integration has been
Procedure the equations
For long runin question using the
relationships, thebounds
ARDLtest.
ave used Based on the Cholesky decomposition of the
The study
established andinvokes
it is foundthreethat steps all oftheestimation
variables normally
On condition involvesthat two there steps ofexists
estimation.
long run
basis, procedures
this which form
reduced seek VAR, to examinethe study theimposes
The first step is to examine the existence of
constraints
aredeterminants
integrated of order zero I(0)debt
of household or order one I(1),
in Swaziland acointegration
long-run cointegration among therelationship variables, the amonglong-run
uraltheVAR.
and
second thatisdefine
its impact
step ontoeconomic
check matrix
forgrowth.
theA0presence
asThe
a lowerof triangular
first the variables
coefficients matrix. in theshort-run
and equations coefficients
in question are
step considers examining the nature of the using the bounds test. On condition that
his long
study The identification
run cointegration using the schemeboundsfollowstest the originalestimated
thereafter
data or order of integration (stationarity). there exists long run cointegration among
paper via a parsimonious ARDL
atrixmethod
form through
If variablesby the areSimsAutoregressive
(1980), Distributed
non-stationary we may the
whereby thevariables,
model which
Cholesky the developed
is long-run using coefficients
Hendry’s
have spurious regression. Thus in order to and short-run coefficients are thereafter
Lag (ARDL) estimation procedure which was general-to-specific modelling approach. The
lectedavoid as such decomposition
cases, stationarity is appliedtests oftothe the contemporaneous
estimated via a parsimonious ARDL model
developed
variables by are
Pesaran et al.(2001).
performed. For this Thispurpose,
method selection
which of the parsimonious
is developed using Hendry’s model is done using
general-
parameter matrix A . Thus, the order of the variables
hasthe Augmented
several advantages Dickey-Fuller
over other testfrequently
(ADF) is to-specific
the built-in ARDL modelling
operator in approach.
EViews 9. The
employed. selection of the parsimonious model is done
is similar approaches
employed cointegration to many studies such as the used in the context of
using
To test thefor built-in
long run ARDL operator ininEViews
cointegration model9. 1 and
After the
Johansen order
andVARs
Juselius inofand integration
advanced
Engel-Granger has been
economies,
Two Toincluding
test for long Sims run cointegration in model
established and it is found that all the 12,and the 2,following the following Unrestricted Error Correction
Unrestricted Error
Stepvariables
method.(1992),
The Engel-Granger,
are integrated and is of as for
order example,
follows;
zero I(0) is
output (GDP),
or Correction inflation
Models (UECM), are constructed:
the five Models (UECM), are constructed:
ordertoone
limited I(1), the
bivariate second
testing of step
typical is to check
runs. It
for
rate (CPI), money supply (M2), private sector credit
the presence of long run cointegration
matrix of
identifies only a single cointegration relationship 𝛥𝛥𝛥𝛥𝛥𝛥𝛥𝛥 𝐻𝐻𝐻𝐻t =
using the(PSCR)
bounds and test the method through
policy rate (DR). The the
𝛽𝛽0 matrix form of 𝑡𝑡−𝑖𝑖 + 𝛴𝛴𝑝𝑝 𝛽𝛽2𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡−𝑖𝑖 +
𝑝𝑝
of serially + 𝛴𝛴𝑖𝑖=1 𝛽𝛽1𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐻𝐻𝐻𝐻
amongAutoregressive
what might have Distributed
been many Lag relations(ARDL)
and 𝑖𝑖=0
several
Central advantages
Bank of Swaziland © 2018 over other frequently 𝑝𝑝 𝑝𝑝
𝛴𝛴𝑖𝑖=0 𝛽𝛽5𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡−𝑖𝑖 + 𝛴𝛴𝑖𝑖=0 𝛽𝛽6𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐵𝐵𝐵𝐵𝐵𝐵𝑡𝑡−𝑖𝑖 + 88 | concluding tha
P a gemployed
e cointegration approaches such
𝑒𝑒 𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡 1 0 0 0 𝜀𝜀𝑡𝑡−1𝐺𝐺𝐺𝐺𝐺𝐺
0 𝜑𝜑1 𝑙𝑙𝑙𝑙𝑙𝑙𝐻𝐻𝐻𝐻 + 𝑡𝑡𝜑𝜑2 𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡−1 + 𝜑𝜑3 𝑙𝑙𝑙𝑙𝑙𝑙𝑃𝑃𝑃𝑃𝑡𝑡−1 + models share m
he above as the Johansen𝐶𝐶𝐶𝐶𝐶𝐶and Juselius and Engel- 𝐶𝐶𝐶𝐶𝐶𝐶
Granger Two Step
𝑒𝑒 𝑡𝑡 𝑎𝑎21 The 1 Engel- 0 0 4 0 𝜀𝜀𝑡𝑡−1 5
𝜑𝜑 𝑙𝑙𝑙𝑙𝑙𝑙𝐺𝐺𝐺𝐺𝐺𝐺 + 𝑡𝑡 𝜑𝜑 𝑙𝑙𝑙𝑙𝑙𝑙𝐵𝐵𝐵𝐵𝐵𝐵 𝑡𝑡−1 + However, if th
𝑀𝑀𝑀𝑀𝑡𝑡 method.
= 𝑎𝑎 𝑎𝑎 1 0 0 𝑀𝑀𝑀𝑀𝜀𝜀𝑡𝑡
of policy
Granger, for example,𝑒𝑒 is limited 32
31 to bivariate 𝜑𝜑 6 𝑙𝑙𝑙𝑙𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃 𝜀𝜀𝑡𝑡−1 𝑡𝑡 (3) lower bounds o
testing of typical
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡
𝑒𝑒 runs. It identifies 𝑎𝑎 41 𝑎𝑎 𝑎𝑎
42 only 43 a 1 0 𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅 𝑡𝑡
reject the nu
pretation [ 𝑒𝑒 𝐷𝐷𝐷𝐷𝑡𝑡 ] [ 𝑎𝑎 51 𝑎𝑎 52 𝑎𝑎 53 𝑎𝑎 𝛥𝛥𝛥𝛥𝛥𝛥𝛥𝛥 𝐺𝐺𝐺𝐺𝐺𝐺t =
]
54 1 𝑝𝑝 𝜀𝜀 𝐷𝐷𝐷𝐷𝑡𝑡 ]
[
single cointegration relationship among 𝑝𝑝
𝛽𝛽0 + 𝛴𝛴𝑖𝑖=1 𝛽𝛽1𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐻𝐻𝐻𝐻𝑡𝑡−𝑖𝑖 + 𝛴𝛴𝑖𝑖=0 𝛽𝛽2𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡−𝑖𝑖 + cointegration a
he model, what might have been many relations and it 𝑝𝑝 𝑝𝑝
𝛴𝛴𝑖𝑖=0 𝛽𝛽3𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝑃𝑃𝑃𝑃𝑡𝑡−𝑖𝑖 + 𝛴𝛴𝑖𝑖=0 𝛽𝛽4𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡−𝑖𝑖 + that the variab
is also limited to one regression. Similarly,
es by A-1 the JohansenWhere andthe Juselius
left hand approach, side of unlike
the equation
𝑝𝑝 𝑝𝑝
𝛴𝛴𝑖𝑖=0 𝛽𝛽5𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡−𝑖𝑖 + 𝛴𝛴𝑖𝑖=0 𝛽𝛽6𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐵𝐵𝐵𝐵𝐵𝐵𝑡𝑡−𝑖𝑖 +
contains a significant long
the bounds cointegration method, fails to 𝜑𝜑1 𝑙𝑙𝑙𝑙𝑙𝑙𝐻𝐻𝐻𝐻𝑡𝑡−1 + 𝜑𝜑2 𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡−1 + 𝜑𝜑3 𝑙𝑙𝑙𝑙𝑙𝑙𝑃𝑃𝑃𝑃𝑡𝑡−1 +
consider thevector
mixture of residuals
of regressors in the whichreduced
are form, and
𝜑𝜑4 𝑙𝑙𝑙𝑙𝑙𝑙𝐺𝐺𝐺𝐺𝐺𝐺 on the
𝑡𝑡−1 + 𝜑𝜑5 𝑙𝑙𝑙𝑙𝑙𝑙𝐵𝐵𝐵𝐵𝐵𝐵𝑡𝑡−1 + 4. Empirical Fi
not integrated of same order, that is, I (0) and 𝜑𝜑6 𝑙𝑙𝑙𝑙𝑙𝑙𝑃𝑃𝑃𝑃𝑃𝑃𝑡𝑡−1 𝜀𝜀𝑡𝑡 (4)
right hand
I (1). As opposed to the otherside multivariate
is the squared co- matrix (A0) of 4.1 Stationarit
integration methods the
coefficients ARDL is applicable
associated with lagged where
where HD
variablesHDtt,, GDP GDPtt refers
andrefers toto the dependent The stationari
the dependent
whether the regressors are not integrated variables corresponding
variables corresponding to equation
to equation (3) βand determine if ea
(3) and (4),
d, but of notsame structural
order or mutually shocks through co-integrated the column andvector
(4), βϕand (ɛ). coefficients
coefficients
are considered non-stationary
are considered as column
but as long as the order of integration is not as column vectors of the parameters, while
urbances. higher than The one,econometric
there can still identification
be a long run ofΔmonetary
vectors of thepolicy parameters, while Δ is the first
is the first difference operator and ε is the
integration. Th
relationship between the explanatory and error difference term. operator and ε is the error term. study’s validity
e reduced shocks is crucial to any model specification,
dependent variable provided that they are methodology;
Once the estimation of equation (3) and (4) is
co-integrated (Sultan,
including 2012). Fundamentally,
SVARs. The SVAR identification Once the estimation exercise of equation (3) and (4) is variables to be
ocks this approach is suitable for smaller sample completed,thethe completed, F-statistic
F-statistic will bewill
computed to test
be computed make the mod
observationsfollowed
of between in this30-80.paper is explainedto as follows;
test
whether whether
the independent the independent
variables have variables
a long-
this study, the
run relationship with household debt and
The ordering of output and price level at the and were foun
economic growth. This will be done by estimating
132 𝑎𝑎
031 01032
𝑒𝑒 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 01
𝜀𝜀 𝑀𝑀𝑀𝑀𝑡𝑡0𝑡𝑡 𝜀𝜀 𝑀𝑀𝑀𝑀 0𝑎𝑎0𝑡𝑡41 𝜀𝜀 𝑀𝑀𝑀𝑀 0𝑎𝑎𝑡𝑡42 𝜀𝜀 𝑀𝑀𝑀𝑀 𝑎𝑎𝐺𝐺𝐺𝐺𝐺𝐺
𝑡𝑡
43 1𝐺𝐺𝐺𝐺𝐺𝐺 0 𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅𝑡𝑡
𝐺𝐺𝐺𝐺𝑡𝑡 1 0 0 0 1 𝐺𝐺𝐺𝐺𝐺𝐺
0 0 𝐺𝐺𝐺𝐺𝐺𝐺
0 00 𝐺𝐺𝐺𝐺𝐺𝐺
0 0 0
𝑎𝑎141 43 𝑎𝑎 0[142 𝜀𝜀43
𝑎𝑎
0
𝑡𝑡 1
𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅
𝜀𝜀𝐷𝐷𝐷𝐷 ]𝐶𝐶𝐶𝐶𝐶𝐶 [𝜀𝜀𝑎𝑎151𝑡𝑡𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅
𝑡𝑡𝜀𝜀0𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅
𝑡𝑡
0 𝑡𝑡
0𝜀𝜀𝑎𝑎52𝑡𝑡𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅
𝑡𝑡 𝜀𝜀𝑎𝑎53 𝑡𝑡
𝑎𝑎𝜀𝜀 54 𝑡𝑡 1] [ 𝜀𝜀 𝐷𝐷𝐷𝐷𝑡𝑡 ]
42
𝐶𝐶𝐶𝐶𝑎𝑎
43 𝑒𝑒 𝐶𝐶𝐶𝐶𝐶𝐶 𝐶𝐶𝐶𝐶𝐶𝐶
𝑡𝑡
𝑎𝑎51 52
53
𝑡𝑡 0 𝑎𝑎 0
54 1 21 152 ] [𝑎𝑎1
53
54 021 𝜀𝜀 𝑎𝑎
1
𝐷𝐷𝐷𝐷 ]𝑡𝑡0[0]1𝜀𝜀𝑡𝑡 𝐷𝐷𝐷𝐷
𝜀𝜀53
54 𝜀𝜀54
𝑎𝑎
1 ]𝑡𝑡0[0]0𝜀𝜀𝑡𝑡 𝐷𝐷𝐷𝐷
1 𝜀𝜀]𝑡𝑡0[ ]0𝜀𝜀𝑡𝑡 𝐷𝐷𝐷𝐷𝜀𝜀 𝑡𝑡 0]𝑡𝑡 𝜀𝜀 𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡
𝐶𝐶𝐶𝐶𝐶𝐶
have a long-run 0 0𝑡𝑡relationship 0 𝑡𝑡 𝜀𝜀 𝑀𝑀𝑀𝑀with household integrated of order two I (2) hence the
𝑀𝑀𝑡𝑡𝑎𝑎132= 𝑎𝑎
031 0𝑎𝑎1032 31 𝜀𝜀 0 𝑎𝑎1
𝑀𝑀𝑀𝑀 032𝑡𝑡 𝜀𝜀 𝑀𝑀𝑀𝑀 001𝑡𝑡 𝜀𝜀 𝑀𝑀𝑀𝑀 𝑀𝑀𝑀𝑀
𝜀𝜀growth. 𝑡𝑡
debt and economic This will be done next step is to proceed with the three ARDL
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑎𝑎𝑡𝑡1 42
41
43 𝑎𝑎0𝑎𝑎142
4341 𝜀𝜀 𝑎𝑎143
𝑎𝑎
0
𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅 0
42𝑡𝑡𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅 𝑎𝑎1 43 0
𝑡𝑡𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅 1 0
𝑡𝑡𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅 𝑡𝑡 𝜀𝜀 𝑃𝑃𝑃𝑃𝑃𝑃𝑅𝑅𝑡𝑡
𝐷𝐷𝐷𝐷[𝑡𝑡𝑎𝑎51
Where [𝑎𝑎1𝑎𝑎52 theby
]51 left
𝑎𝑎53
𝑎𝑎 ]52 hand
estimating
𝑎𝑎𝑎𝑎54
]53 side theof
𝑎𝑎]54
Wald the ]𝑡𝑡 []equation
or F-test forcontains the joint a estimation procedures. The first step which
ide
quation
and ]
52
53
54
theofequation
side 53
54
thecontains [ 𝜀𝜀
ofequation 1
𝐷𝐷𝐷𝐷 54𝑡𝑡[ ]𝜀𝜀 𝐷𝐷𝐷𝐷
thecontains
significance equation 1
a contains𝑡𝑡[ ]𝜀𝜀 𝐷𝐷𝐷𝐷1
a contains
of the𝑡𝑡[ ]𝜀𝜀 𝐷𝐷𝐷𝐷
a 1
coefficients 𝜀𝜀
a
𝐷𝐷𝐷𝐷𝑡𝑡 ]
of the lagged involves the test for long run cointegration is
vector of residuals in
variables at their levels and comparing the reduced form, and on the
it presented in Table 1. In the ARDL model, the
educed
d
als the form,
inreduced
the form,
and reduced on form,
and the on form,
and the on and the on the
to the critical bounds values computed by existence of long run cointegration among
right hand side is the squared matrix (A0) of
d
eequared
the
tdhe ofequation
hand
side
isthe
left
matrix of
squaredequation
side
hand
thematrix the contains
squared
(A0) ofequation
side
Pesaran the
matrix contains
ofof
(A0) equation
etathe
matrix contains
of equation
al.
(A0) a (A0)
(2001). contains
of a of contains a a the variables for equations 4 and 5 are
coefficients associated with lagged variables and tested using the bounds F-test which checks
of
duals
ced
in
h residuals
ociated
ged
d reduced
the
withform,
lagged inreduced
variables the
with
laggedform,
and in
reduced
The
variables the
on form,
lagged
andand
null the
variablesreduced
on form,
and
hypothesis
and theon
variables andand
form,the
(Ho) on
and and
of the on the
non-existence of the joint significance of the coefficients of
structural long-run shocks through the column vector (ɛ).
she ide
and
sred
umn squared
ugh the
throughis
columnmatrix
side
the squared
vector the
column matrix
is(ɛ).
the
vectorsquared
(A0)
the
column matrix
vector ofeffect
squared
(ɛ).(A0) matrix
vector
(ɛ). of(nomatrix
(A0) cointegration):
(A0)
(ɛ).of (A0) ofH0: 1= the lagged variables at their initial levels.
of
2= 3= 4= 5 is tested against the Accordingly, the F-statistic tests the null
The econometric alternate identification
hypothesis (H1) ofthat monetaryexist policy
with
of ents
agged
ssociated
ted
ntification
ction with
lagged
associated
variables
identification
monetary
of monetary with
lagged variables
oflong-run
policy
monetary lagged
with
of and variables
policy
monetary lagged
andvariables
policy and variables
policy and there and a hypothesis that: the coefficients of the
shocks is crucialeffect to any (existence model of cointegration
specification, lagged level variables are jointly equal
hral
odel cks
rough
olumn
the
yoial modelshocks
anythrough
column
to the
vector
specification, through
model
any column
1:the
vector
(ɛ).
specification,
model column
1≠ vector
the(ɛ).
specification, column
2≠ vector
(ɛ).
specification, 3≠ vector
(ɛ). 4 ≠ (ɛ). 5. to zero (that is, there is no cointegration
including SVARs. The SVAR identification exercise or long-run relationship), which is tested
dentification
onometric
n tric
cation
ntification
R
s.SVAR of identification
monetary
identification
The of
SVAR monetary
identification
identification
exerciseof policy
monetary
identification
Should exercise of thepolicy
monetary
exercise ofpolicy
F-statistics monetary
exercise policy exceed policy the upper against the alternative hypothesis that;
followed inbound this paper of the iscritical
explained as follows;
model
any
ducial
plained
paperisis
toas crucial
model
any
isto
explained
follows;asspecification,
model
any
explained
follows; to
specification,
as model
any
follows; asspecification,
model
follows; specification, specification,then we there exists a long run relationship between
values, I(1),
may reject the null hypothesis of no long the variables.
The ordering of output and price level at the
ng ghe
output AR
Rs.
entification
price
and SVARs.
of SVAR
identification
The
output
price
level
and SVAR
identification
The runexercise
price
level
at
and SVAR
identification
the exercise
cointegration,
price
level
at identification
the exercise
level
at theat exercisethe exercise
concluding that the
beginning variables is because contained they react in thetomodels share Table 1: Bound Test for Cointegration
an innovation
explained
eyd ned
is
ct er in
because
seto paper
is this
as
react
they explained
follows;
paper
is
an innovation
to
react as
they explained
follows;
anmeaningful isas
innovation
to
react explained
follows;
an innovation
to an as follows;
long as
innovation follows;
run relationships. However, Bound Critical Values (unrestricted
in the monetary if the F-statistics policy rate are with inferior a lag to due
the to their
lower intercept with no trend) K=5
nd
put
fate
ary
thing
ordering
licyoutput
price
and
of
awith
policy
lag
ratedue outputprice
level
awithand
lagof
rate to output
due price
level
at
and
atheir
with
lagtodue the price
level
at
atheir and
lag the
todue level
price
at
their the
to their atlevel the at the
bounds of the critical values, I (0), we fail Models Estimated Critical Lower Upper
slow movement to reject in
the nature.
null hypothesis of no long F-Value Value Bound Bound
they
re. eact
nning
is
ause
ent because
nature. into
react
they
isanbecause
nature. innovation
to
react
they an innovation
to
react
they an innovation
to react an innovationto an innovation I(0) I(1)
run cointegration among the variables and
ywith
etary
policy
erate
monetary
awith
policy
lag
ratedue
awith
lag
policy
rate
todue
atheir
with
lag
concluderate
tothat
due
atheir
with
lagtodue
thetheir
avariables
lag
todue
theirto theirtested do
being Model (1) 10.235 1% 3.41 4.68
not share a significant long run relationship.
58 | P a g e 2.5% 2.96 4.18
wture.
ment
in
movement
nature.
in nature.
in nature.
58 | P a 58
g e| P a 58
g e| P a 58
g e| P a g e Model (2) 4.2380 5% 2.62 3.79
4.0 EMPIRICAL FINDINGS 10% 2.26 3.35
58
4.1| P aStationarity
58
g e| P a 58
g e| P aTest
58
g e| P a g58e| P a g e The bounds test results for long run
The stationarity test was done in order to cointegration are presented in Table 1. The
determine if each time series contains unit F-statistics in this case has been calculated
root or non-stationary and to determine the for two models in order to clearly explain
order of integration. This was necessary the relationship between household debt
to confirm the study’s validity in using and economic growth in Swaziland. Model 1
the ARDL Bounds testing methodology; in this regard is based on household debt as a
this methodology requires variables to be dependent variable while model 2 shows how
integrated of order I(0) or I(1) to make the household debt as an independent variable
model valid (Pesaran et el (2001). In this explains variations in economic growth.
study, the variables were first tested at levels
and were found to be none stationary. The Using Pesaran et al (2001) critical bounds
same test was conducted after differencing, value, the study finds that the estimated
and all the variables were found I(1). F-values for model 1 F(HD/GDP, CPI, BPA,
POP, PR) and model 2 F(GDP/HD,CPI, BPA,
4.2 Bounds Test POP, PR) is 10.24 and 4.24, respectively.
From the stationarity results it has been From the results we see that the estimated
established that none of the variables are F-values are greater than the upper bound
critical values at 1 and 5 per cent level The results further indicate that a 1%
of significance hence we reject the null increase in population is expected
hypothesis of no long run cointegration and to reduce household debt by 1.7 %.
conclude that there is long run cointegration Consequently, population growth was found
relationship between the variables specified to be contradicting theoretical expectations
in all the models and their dependent where it is expected to increase household
variables. debt. However, in this study, population
growth discourages growth in household
4.3 Long Run and Short Run Relationship debt and this could be due to the high
Analysis poverty and unemployment levels which
After generating a more parsimonious model hinders households from acquiring loans
which is traditionally found by gradually with commercial banks. Meng et al (2012)
deleting the insignificant variables. Table 2 purports that lower income or the absence
presents the long-run estimation results of of income due to high unemployment casts
the Model 1 based on the ARDL (1,3,0,2,1,1) doubt on future income. As a result, regular
lag length while Model 2 is based on ARDL unemployment of a household discourages
(1,1,2,0,1,2) lag length. The number of household debt because of concerns about
lags of both models were selected using the the ability to repay an existing loan hence
Akaike Information Criterion (AIC). From this discourages household demand for loans
the results of Model 1, the coefficients and constraints growth of household debt.
of Consumer Price Index, Building Plans
Approved and Population were found to be Even though the coefficients of GDP and prime
significant determinants of household debt. lending rates were found to be insignificant
The estimated coefficients (Model 1), determinants of household debt, they do
suggest that a 1 % increase in the Consumer bear the correct signs which are in line with
Price Index will result in a 0.70% increase theoretical expectations. The results as
in household debt in the long run implying presented in Table 2 imply that growth in
that as the cost of living rises they pose a GDP and lending rates discourage or retard
burden to household disposable income household borrowing. The signs of these
hence forcing households to acquire more parameters confirm the findings by Hong
loans. Khan et al (2016) further points out (2011), Khan et al (2011) and Cletta et al
that households may use debt to supplement (2014) who also found a negative relationship
their wages in order to meet their daily living between interest rates, GDP and household
expenses. debt in the respective countries.
On the other hand, a 1% increase in building The goodness of fit of Model 1 as shown
plans approved translates into a 0.23 % in by R-squared and Adjusted R-squared are
household debt. The positive coefficient of 0.75 and 0.61, respectively. These results
building plans approved could mean that imply that 61 per cent of total variation in
the desire for households to own urban household debt is explained by the specified
residential properties is facilitated through explanatory variables in the model, which is
the acquisition of mortgage which translates a fairly good fit.
into an increase in household debt.
Table 2: Estimated long run coefficients using Table 3: Error Correction Model
ARDL Model (1) Model (2)
Model (1) Selected ARDL Model (2) Selected ARDL Dependent Variable (LogHD) Dependent Variable
(1, 3, 0, 2, 1, 1) (1, 1, 2, 0, 1, 2) (LogGDP)
household debt by 0.02%. This result is found appropriate, several diagnostic tests were
to be consistent with the long run result conducted, these include, tests for serial
which found that there exists a negative but correlation and homoscedasticity. The
insignificant relationship between the two Ramsey RESET test for model specification
variables. The main reason for this is that was also conducted to check whether the
interest rate hikes in general discourage models non-linear combinations of the fitted
loan demand because it raises the cost of values help explain the dependent variable
borrowing, which discourages household or not.
from borrowing or at least reduce credit
demand. Table 4: Summary of Diagnostic Tests
Diagnostic Model 1(HD) Model 2(GDP)
On the other hand, a 1% increase in the Tests
consumer price index and building plans Test p-value Test p-value
Statistic Statistic
approved in the immediate short run results
Jarque-Bera 0.6045 0.7391 1.5311 0.4651
in household debt increasing by a 0.59 and (normality)
0.04%, respectively, indicating that these Breusch- 1.3316 0.5139 2.2391 0.3264
variables have a positive impact in short run Godfrey (serial
as well. However, this paper takes exception correlation)
that once a quarter elapses, building plan Breusch- 6.2471 0.9367 7.5185 0.8215
Pagan-
approved begin to generate a negative Godfrey
impact leading to a 0.08% reduction in (hetero-
household debt. scedasticity)
The error correction term also reflects the 4.5 Parameter Stability Test
required negative sign which is statistically In order to confirm the stability of the
significant. Thus, it can be concluded that estimated coefficients given that the model
any short run deviations are highly likely is said to be correctly specified, Pesaran
to converge towards long run equilibrium. (1997) proposed that the cumulative sum of
The coefficient of the error correction term recursive residuals (CUSUM) and the CUSUM
(ecm2) is -0.066 which implies that about of squares tests be applied to assess the
6.6% of disequilibrium between GDP and stability of the estimated parameters. As a
its determining factors is corrected within result, the plots of the cumulative sums are
one quarter. This shows a very slow speed expected to fall within the 5% critical lines
of adjustment towards long run equilibrium. of statistical significance where we shall
conclude that the coefficients are stable and
4.4 Diagnostic tests reliable.
To ensure that the results are robust and
Figure
Figure 4. 4. Stability
Stability test test results-
results-Model Model1 1 -0.4
-0.4
IV I II
Figure 4. Stability test results - Model 1 Figure 5: Stability test results - Model 2
(LogHD)
(LogHD)
(LogHD)
5% critical lines of statistical significance where
(LogGDP) Research Bulletin Volume 2
1.4 15
2009 20
1.2
1515
and reliable. 0.8 5
0.6
0.4
0 FiguF
1010 0.2 -5
0.0
-10
CUSU
C
-0.2
55
Figure 4. Stability test results- Model 1
from
-0.4 -15
(LogHD)
IV I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
2009 2010 2011 2012 2013 2014 2015
IV I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
2009 2010 2011 2012 2013 2014 2015 fr
0
0 CUSUM of Squares 5% Significance CUSUM 5% Significance
15
Figure 4 and 5 reflect the plots of CUSUM and stabs
-5 10
-5 Figure
CUSUMQ 4 and
tests5ofreflect
the twothe plotsThe
models. of CUSUM
results
5
and
fromCUSUMQ
the two tests
figuresofreflect
the two models. and
the reliability The resid
r
-10 0 results from the two figures reflect the
-10 -5 stability of all the estimated parameters as all the
reliability and stability of all the estimated
critic
-15-10
residuals were
parameters
tocritical
be falling
as allfound
within
thetoresiduals
be fallingwere
withinfound
the
c
-15-15 II III IV I II III IV I II III IV I II III IV I II III IV I II III IV bounds of 5% the
levelcritical bounds of 5%
of significance.
II2010 III IV I 2011 II III IV I 2012 II III IV I 2013
II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
II III IV I2014II III IV 2015
I II III IV level of significance.
2010 2011 2012 2013 2014 2015
CUSUM
2013 2014
5% Significance
2015
5.0 CONCLUSIONS AND 5. C
1.4
1.2
CUSUM 5% Significance
ThisRECOMMENDATIONS
paper investigates the relationship between 5
1.41.0 This paper debt
household investigates the relationship
and economic growth in
1.40.8
1.20.6
between household debt and
Swaziland by employing the Autoregressiveeconomic This
1.2
0.4
growth in Swaziland by employing the
Distributed Lag modelling approach using T
1.0 0.2
Autoregressive Distributed Lag modelling
quarterly using
approach data from 2006 todata
quarterly 2015.from
Using 2006
the
hous
h
0.0
1.0
0.8-0.2 tobounds
2015. test
Using the bounds
for long-run test for the
cointegration, long-run
study
0.8
0.6
-0.4
II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
cointegration, the study found evidence
found evidence of a long run relationship
of Swa
2010 2011 2012 2013 2014 2015 a long run relationship between household
between
andhousehold credit and economic growth. S
0.6
0.4 CUSUM of Squares 5% Significance
credit economic growth.
Distr
The results revealed that household credit had a
0.2Figure 5: Stability test results -Model 2
0.4
(LogGDP)
The results revealed that household credit
long-run negative and andsignificant impact on quar
D
0.0 had a long-run negative significant impact
0.2
oneconomic
economic growth,
growth, while thewhile the growth
economic economic had
-0.2
0.0 growth had an insignificant negative
an insignificant negative impact on household impact q
boun
on household credit in Swaziland. Such
-0.4 credit in Swaziland. Such findings imply that an
-0.2
II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
findings imply that an increase in household
increase
debt in long
in the household debt ineconomic
run deters the long rungrowth.
deters b
foun
-0.4 2010 2011 2012 2013 2014 2015 More importantly,
economic this importantly,
growth. More finding is in line
this with
finding
II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
CUSUM of Squares 5% Significance
the earlier assertion that excessive growth
is in line with the earlier assertion that excessive fo
betw
2010 2011 2012 2013 2014 2015 in household debt, if left unchecked, may
growth in household debt, if left unchecked, may
bring a negative effect on the performance
Central Bank ofCUSUM of Squares© 2018
5% Significance of the economy (Hammad et al, 2016). 94 | This
b
The
Swaziland
Figure 5: Stability test results -Model 2
Page calls for close monitoring of the household
credit market and the control of excessive The consumer price index was found to
households’ exposure to credit. This can lead to an increase in household debt.
be achieved by setting a household credit This finding of the study implies that an
threshold that promotes economic growth. increase in consumer prices encourages an
The study further explored the impact of increase in household debt in Swaziland. As
prime lending rate, population growth, prices go up, households are expected to
residential building plans approved and increase their credit in order to smoothen
consumer price index on household credit their consumption. Though the central bank
and economic growth. The study established expertly controls the interest rates when
that the consumer price index, population inflation increases, the households will have
growth and the residential building plan more loans to maintain and defaulting risks
approved had significant influence on will also increase. Dercon (2012) explains
household credit in the long run. that in most cases, households in developing
countries often lack risk management tools
Based on the aforementioned, the findings to meet adverse shocks. Thus encouraging
of this study indicated that an increase in financial literacy among consumers would
residential building plans approved leads to help households to manage their spending
an increase in household credit in the long pattern and help them make informed
run. This implies that the housing market decisions.
plays a significant role in the accumulation
of household credit in Swaziland which is a The ECMs for both models were found to be
cause for concern. In this regard, government statistically significant with the expected
needs to intervene and impose polices signs. In the case of the household debt
that require further scrutiny of the credit model, the estimated value was 0.83
extended towards residential development. implying that approximately 83 per cent
of disequilibria in the long run is corrected
Furthermore, the study revealed that an within one quarter while in the case of the
increase in population leads to a decrease GDP model the ECM implies that 6.6 per cent
in household debt in Swaziland in the long of disequilibrium is corrected within one
run. This finding implies it is possible for quarter.
household debt to fall as a result of an
increase in population. Aligned with this 5.1 Recommendations for further research
negative relationship could be the high For future research, we recommend that
level of unemployment and poverty which further studies be conducted to determine
shows that a number of households could be the household credit threshold that sustains
unbanked and highly incapable of acquiring economic growth in Swaziland. This would
loans with commercial banks. According to help policymakers to make informed policy
Zimunya and Raboloko (2015), it is possible decisions. Though the study found a long run
to reduce household debt by creating more relationship between household debt and
jobs. This study thus reiterates the need economic growth, the causal relationship
for more job creation opportunities in both has not been explained in this study hence
private and public sectors which would the granger causality is recommended for
increase productivity in the economy and further research.
enhance household income hence raising
households’ ability to acquire credit.
Prinsloo, J.W. (2002). Household debt, Tudela and Young (2005), The Determinants
wealth and saving. Quarterly Bulletin, of Household debt and Balance Sheets
SA Reserve Bank: 63-78. in the United Kingdom. Bank of
England wp 266.
Rahman S. and Masih M. (2014). Increasing
Household debt and its Relation TO Verner E., Sufi A. and Mian A. (2015).
gdp, Interest rate and house price: Household Debt and Business CCycles
Malaysia’s perspective. MPRA Paper Worldwide. wp 21581. http//:www.
No. 62365. nber.org/paper/w21581.
Santos C.S., Costa V., Teles N (2014). The Wildauer F.(2016). 7 Facts about U.S.
Political Economy of Consumption and Household Debt. https://www.
Household Debt: An Interdisciplinary boeckler.de/pdf/v_2016_10_22_
Contribution. RCCS Annual Review, wildauer.pdf.
Issue 6.
Weinberg J. A. (2006), Borrowing by U.S
Shim I., Mohnty M., Lombandi M. (2017), Households. Federal Reserve Bank of
The Real Effects of Household Debt Richmond, Economic Review, 92(3),
in the short and long-run. Bank of 177-194.
International Settlements wp no. 607.
World Bank (2016) http://databank.
Tu T. (2008), New Zealand Household Debt: worldbank.org/data/reports.
Is it too high. Aucklan University of
Technology. Zimunya F.M., Baboloko M. (2015).
Determinants of Household Debt
in Botswana: 1994-2012. Journal
of Economics and Public Finance,
1(1),14-36.
central banks define monetary policy in for Swaziland. As noted by Langa (2001),
various ways. In practice there are many Swaziland’s monetary policy framework is
instruments which can be used by the to a greater extent influenced (determined)
central banks in formulating their monetary by her membership to the quasi-currency
policies. They include controlling the level board, the Common Monetary Area (CMA),
of money supply, the reserve and liquidity where South Africa is a dominant member.
requirements, open market operations,
and the discount rate among others. The 1.1 Monetary Policy in Swaziland
discount or bank rate is the most popular The Central Bank of Swaziland has at
in most central banks. These banks target its disposal a variety of monetary policy
a certain level of inflation which guides instruments to achieve price stability.
them whether to increase or decrease However the country’s membership to a CMA
the discount rate. Central banks in most coupled with full economic integration and a
advanced economies now operate with some fixed exchange regime precludes discretional
form of either explicit or implicit inflation monetary policy or independence. Therefore
target. However, several emerging markets Swaziland has no formal role in the
and developing economies have adopted formulation of the monetary and exchange
inflation targeting since the 1990s, including rate policies. Therefore given the parity of
South Africa in 2000. the currency of Swaziland to the Rand and
the free mobility of capital, the scope of
Therefore, understanding the dynamics of monetary policies in Swaziland is limited and
inflation is very crucial in predicting the less independent to those of South Africa. The
future course of inflation in a precise manner exchange rate peg has to be fully supported
to achieve the goal of price stability. A number by well-coordinated monetary policy and
of efforts have been undertaken in different discipline fiscal policies. Expansionary fiscal
countries to identify suitable models for policy in Swaziland tends to undermine the
this analysis and has resulted into a large exchange rate peg by exerting pressure on
number of different models for inflation, external outflows reducing reserves and
which vary in terms of scale, and in terms threatening the peg, (Nxumalo, 2014).
of the variables assumed to drive Inflation.
According to Aron and Muellbauer (2011), In that regard, keeping the exchange rate
forecasting inflation in emerging markets is peg serves as an intermediate goal for
particularly challenging compared to the US monetary policy in Swaziland. To support this
and the Eurozone, due to the volatility in food exchange rate policy of the currency board,
prices, exchange rates, as well as structural Swaziland has to keep high levels of reserves
changes, hence a range of approaches are equivalent to the conventional international
used for forecasting inflation. These efforts standard of 3 months of imports cover.
have resulted in a large number of different
models for inflation, which vary in terms of Generally, Swaziland’s discount rate moves
scale, and in terms of the variables assumed in tandem to that of South Africa, however
to drive inflation (including the output gap, based on domestic economic fundamentals,
unemployment rate or unit labour costs). Swaziland sometimes deviate from the
This paper therefore seeks to identify a South African Reserve Bank (SARB) repo rate
suitable model for forecasting inflation usually by 50 basis points.
policy or 6
no formal 4 Figure 2: Swaziland and South Africa inflation con
2 Figure 2: Swaziland and South Africa inflation
rates
etary and 0 rates 200
22.0
2000
2002
2004
2006
2008
2010
2012
2014
2016
the parity 20.0 Exp
18.0
nd and the SD Discount Rate SA RepoRate 16.0 bre
14.0
per cent
monetary
Source: Central
Source: CentralBank of Swaziland
Bank of Swaziland 12.0 inte
10.0
and less 1.2 Theoretical Framework on Inflation 8.0 (cla
1.2 Theoretical Framework on Inflation
Dynamics 6.0
4.0
rica. The Dynamics 2.0 pur
Different schools of thought articulate to 0.0
ported by Different schools of thought articulate to the
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
dynamics ofof
dynamics inflation trends
inflation especially
trends in relationin
especially
discipline SD Inflation SA Inflation bas
relation
to economic togrowth.
economic
Centralgrowth.
to inflationCentral
dynamicsto
Source: Central Bank of Swaziland & STATS SA SA
policy in inflation dynamics is the issue of neutrality Source: Central Bank of Swaziland & STATS
is the issue of neutrality of money and labour
hange rate of money and labour markets adjustment
markets
to price adjustment
changes. The to traditional
price changes. route Theto Inflation trends for Swaziland and that of
l outflows her major trading partner, South Africa,
analyzing
traditional inflation is that ofinflation
route to analyzing isolatingisdemand
that of Central Bank of Swaziland © 2018
the peg, pull pressures and pressures
cost push Pin
a g Figure 2 depict that between 1989 and
and pressures.
e
isolating demand pull cost push 1994 the two countries’ inflation rates
According to the Keynesian model, demand
pressures.
pull According
inflation arises tofrom
the factors
Keynesian thatmodel,
lead averaged above 10 per cent, with South
to excess
demand demandarises
pull inflation in from
an factors
economy (i.e.
that lead Africa’s inflation rates higher than those
rate peg
increased in investment, in Swaziland. Between 1994 to 2000 South
monetary to excess demand in an economy rise in money
(i.e. increased in Africa’s inflation decreased at a faster pace
supply or government financing spending
hange rate investment,borrowing).
through rise in moneyCost supplypush
or government
inflation, than that of Swaziland; After adopting the
on the other
financing hand,through
spending links inflation
borrowing).to supply
Cost inflation targeting regime in 2001, South
nd has to
factors – predominantly rising production Africa’s inflation continued to slow down at
costs. Structuralists believe that structural 99 | a faster pace than that of Swaziland. The
bottlenecks such as distortionary government country’s inflation has remained higher than
policies, government budget constraint, the inflation rate of South Africa for the
inelastic supply of essential commodities better part of the last ten years.
and conflicts in income distribution are key
determinants towards inflation outcomes. 1.4 Swaziland Inflation Composition and
Weights
1.3 Swaziland and South Africa Inflation The composition of the consumer prices index
Trends (whose changes are used to measure inflation)
Inflation movements in Swaziland are often has changed over time predominately
linked with those of South Africa with the informed by household and income and
notion that there is high imported-inflation expenditure surveys (HIES) carried out in
pass-through effect from South Africa to the country. These surveys provide data for
Swaziland since Swaziland sources more than evaluation of the weighting structure of the
90 percent of its imports via South Africa. In items in the consumption basket. The latest
the long-run the relative purchasing power survey used is the 2001 Swaziland Household
Income and Expenditure surveys (SHIES Table 1: A Comparison of Old and the New CPI
2001). The breakdown of the subcomponents Weights
is based on the international classification Code COICOP Weight Weight Perc- %
Category Based Based entage change
standard COICOP (classification of Individual Classification on on point
consumption by purpose). Given the high SHIES SHIES change
2001 2010
poverty rates noted in the SHIES 2001 , food
dominates the consumption basket. Research Bulletin 01 Food &
Volume 2
non-alcoholic
37.73 28.31 -9.4 -25.0
beverages
Figure 3: Comparative Inflation Rate Trends
Figure 3: Comparative Inflation Rate Trends 02 Alcoholic 0.96 0.41 -0.6 -57.3
20.0 beverages &
18.0 tobacco
16.0
14.0 03 Clothing & 6.16 3.76 -2.4 -39.0
12.0
per cent
10.0 footwear
8.0
6.0 04 Housing & 14.33 29.24 14.9 104.1
4.0 Utilities
2.0
0.0 05 Furnishing 11.88 5.08 -6.8 -57.2
-2.0
Sep-15
Sep-10
Sep-11
Sep-12
Sep-13
Sep-14
Sep-16
Sep-17
Mar-10
Mar-11
Mar-12
Mar-13
Mar-14
Mar-15
Mar-16
Mar-17
90
11 Restaurants & 0.72 1.91 1.2 165.3
below shows the new weights that came into
Central Bank Of Swaziland © 2018 hotels
effect in 2013 and they are compared with the 12 Miscellaneous 4.67 4.78 0.1 2.4
eight in the consumption basket. Another undertaken worldwide. 2.1 Theoretical Literature
durable and semi-durable products. Components Phillips Curve (APC) where there is no long-run
gnificant shift was noted in the weightings of Research B
that saw their shares
C E N T R A L B A N K O F S WA Z I L A N D (weights) declining 2.1
| trade-off
Theoretical
R E S E A R C H between
by 2.1.1 Literature The
B U L L E T I N Vinflation
traditional
OLUME 2
Phillips
and unemployment a curve
urable and semi-durable products. Components curve may be subject to the Lucas critique.
esearch Bulletin notable Volume 2 margins are ‘furnishing and household
hat saw their shares (weights) declining by presented
2.1.1 The traditional
Estimated
below:
In 1958,
parameters Phillips
arecurve
A.W. Phillips
likely tointroduced change asthe the Ph
Research Bulletin equipment’
Volume
presents
Research 2 Bulletin
the (6.8Volume
methodology, percentage 2 section points), ‘clothing
4 presents theand Research Bulletin Volume 2
otable margins are ‘furnishing and household Curve, which postulates a trade-off bet
Researchpercentage
Bulletin Volume points), 2 ‘clothing and footwear’ In policy
1958, TheA.W.regime
traditional Phillipsvaries. PhillipsThiscurve
introduced led∗ to may
the thebe introduction
subject
𝑒𝑒Phillips
footwear’
models
presents
(2.4 the( methodology,
results,
percentage
2.4
goods percentage
forecasting
presents points)
points)
performance,
& section
the and methodology,
‘recreation4andpresents ‘recreation
as well
section
and theto4 the presents Lucas 𝜋𝜋 𝑡𝑡 =
the 𝛾𝛾(𝑈𝑈
presents
critique.
inflation and unemployment.Research𝑡𝑡 − 𝑈𝑈
Estimated the ) + 𝜋𝜋
methodology,
𝑡𝑡
parameters The Phillipssection
quipment’ (6.8 percentage points), ‘clothing and Bulletin
as culture’
and presents
culture’
forecast 3.5the
services
(3.5combinations,
percentage
(models methodology,
percentage points). points).
while section
Meanwhile
section Curve,
Meanwhile ofpresents
4well 5the which
are so-called
likelythe postulates
to Friedman-Phelps
change a as trade-off
the policy Accelerationist
between regime perfor
models results, forecasting results, performance,
forecasting asperformance, as well
representation models results,
states forecasting
that a lower
otwear’ ( 2.4items percentage that recorded points)
All Items and ‘recreation
notable increases
100 100 incurve
their
Where
inflation Phillips
mayvaries.
inflation,
and Curve
be This
unemployment.
subject led, to
𝜋𝜋
(APC) is the
negatively
where
to The
the introduction
there
Phillips
Lucas correlated
is no
curve
critique. theinflation
oflong-run withcu
0
asitems
concludes. models
that recordedresults, forecasting
notable increases performance, in whiletheir as well 𝑡𝑡
e methodology, forecast
section
weights are combinations,
4 presents
as forecast
‘communication’ the while
combinations,
(from section
1.43 per 5so-called section
will 5 as
Friedman-Phelps
tend to be forecast
associated combinations,
Accelerationist
with a higher whi
le
ndeCentral culture’ (
thereStatistics Office 3.5 percentage Source: points).
SHIES 2000/1 Meanwhile
and SHIES 2009/10, Central Statistics Office
cent
weights to as
3.17
are forecast
per cent)
‘communication’ combinations,
and ‘education’
(from 1.43 representation
Estimated
while
deviations
(from per trade-off
section
cent parameters
Phillips of states
between5 the
Curve that
inflation
are a
likely
unemployment
(APC)
concludes. lower
whereand
to inflation
unemployment
changethere rate rate
asis thefrom
no as th
it
tatistics Office concludes.
lts, forecasting performance, concludes. as well unemployment, and vice-versa. The econom
emsption
there that recorded
5.38
10 indicate that thereconcludes. notable
percent The to increases
results
8.77 from in
percent).The SHIES their 2010
different indicate that
long-run there trade-off between inflation and
to 3.17 per ofsection
cent) and will presented
policy tend
regime to be associated
below:
varies. This withtoathe
led higher level of
introduction ex
icatecombinations,
that therecomposition while Consumer 5 ‘education’
Price Indices(from natural
(CPI) 5.38 rate
unemployment U*,
specification and where
as presented of this thebelow:
relationship entire curve
can be wi
eights
.inThe
mption are ‘communication’
the consumption (from 1.43
has been a notable shift inunemployment, per cent the consumption
weighting
2.0percent
LiteraturetoReview structure 8.77 across countries
percent).The of different is one and
the so-called Friedman-Phelps∗ Accelerationist vice-versa. The econometric H
os. e 3.17
share consumption
The per
survey periods. The cent)
of the and often ‘education’
patterns cited (from
reasons
between the 5.38
why two inflation shifted
survey up
periods. or as:
The down𝜋𝜋 = one-for-one
𝛾𝛾(𝑈𝑈 − 𝑈𝑈 ) + 𝜋𝜋 with
𝑒𝑒 changes in
2.0composition
Literature of
Review Consumer Price Indices specification
(CPI) of this𝑡𝑡 relationship 2.0
𝑡𝑡
Literature can be 𝑡𝑡 written
Review
ey periods. The differential 2.0
may Literature
persist various across Reviewcountries. Phillips Curve (APC) where there is no long-run th
ercent
rages’
Sshare 2001 istoThis the 8.77 chapter
share major presents
percent).The changeReview indifferent
the SHIES theories 𝜋𝜋𝑡𝑡𝑒𝑒 and
2001
as: (the
is theexpected share rate of𝑡𝑡−1inflation). ∗However
2.0 Literature
weighting structure across countries istrade-off one of the
Where Where
between inflation,
inflation, inflation𝜋𝜋𝑡𝑡 ,𝜋𝜋is
𝑡𝑡 ,= is𝛼𝛼𝜋𝜋
negatively
negatively
and unemployment
+ 𝛽𝛽(𝑈𝑈 𝑡𝑡 − 𝑈𝑈 ) +with
correlated
correlated 𝜀𝜀
as 𝑡𝑡 ei
1 is the share
-alcoholic empirical
omposition
points
erages’ of
TheConsumer
This
beverages’ chapter
rest studies
of the
(weight)ThisPrice
presents
paper onof the
Indices
is
chapter choice
various
organised
‘food and(CPI)
presents as offollows;
theories
non-alcoholicinflation
various andwith
expectations beverages’
theories deviations
of and inflation
This
of thechapter unemployment
were
presentsrate
modelled
various
as som
re Review often section cited
2 is reasons
the review why of inflation
literature, differential
section presented 3 may
deviations below: of the unemployment ∗ rate from its
is ex
holic
hts points
.4 tobeverages’
eighting
percentage structure
forecasting
empirical
points
This
across studies
component chapter
countries
modelsempirical on is
forwhich presents
the onechoice
various
studies of the
fell various
economies.
by
on of9.4 the inflation
percentage In from
theories
choice
𝜋𝜋𝑡𝑡 =and
of
its
points
𝛼𝛼𝜋𝜋natural
where
inflation 𝑡𝑡−1 + rate
𝜋𝜋𝑡𝑡 𝛽𝛽(𝑈𝑈
denotes𝑡𝑡 − 𝑈𝑈 the
empirical ), studies
+and 𝜀𝜀𝑡𝑡 where
inflation onrate, the𝑈𝑈𝑡𝑡choic
the th
presents
persist across the methodology,
countries. section 4 presents function entire curve
ofof unemploymentpast is shifted
inflation up or down
(e.g. one-adaptiv
rcentage
ten cited points
spite
reasons
the
forecasting of
why
models empirical
the
inflation
results,
models abundant studies
differential
forecasting
for on performance,
literature
various may the
economies.choice
on natural
theofIn inflationrate U*, and where
forecasting
the
or 𝑒𝑒output,models
entire 𝑈𝑈 ∗foriscurve
the steady
various
is is
are
er hts of
presents
to
current weights to various theories
from 37.73
forecasting and percentmodels in the
for current
various
where 𝜋𝜋for-one weights
economies.
denotes 𝜋𝜋 =
withto In
𝛾𝛾(𝑈𝑈
the −
inflation
changes 𝑈𝑈 ∗
)
in + rate,𝜋𝜋 𝑈𝑈
(the is the
expected log
𝑡𝑡 𝑡𝑡 𝑡𝑡 𝑡𝑡 𝑡𝑡
ent
ersist weights
across
as
to
comparison
wellforecasting
countries.
The rest
as forecast
of
of forecasting
the papermodelscombinations,
is for
models
organised various
worldwide,
as expectations),
while shifted
economies.
follows; rate upof or
value
In at
inflation).down of its simplest
one-for-one
unemployment
However, just
with
or
expectations oneof lag
changes
output, and in 𝜀𝜀o𝑡𝑡
eased
tudies
are of on
weights. spite
The the share
section
of
choice
of5 28.31the ofspite
concludes.
abundant
inflation
percent of in literature
thetheabundant new weights. on the The share
literature
of unemployment onoforthe output, 𝑈𝑈 is thethe
spite ∗ of abundant
steady-state litera
𝑒𝑒 inflation were modelled
ht)s. The
of share
models
reased studies
for of that
section
comparison
various spite
2the of
specifically
isofcomparison
the
economies. theIn
review
forecasting
‘housing focus
and abundanton
ofofmodels
utilities’ Swaziland
literature,
forecasting
literature
Where
worldwide,inflation.
section
component 𝜋𝜋
are
models (the
3on
𝑡𝑡 inflation,
increased random
expected
the
Therefore
worldwide, 𝜋𝜋 𝑡𝑡 , is disturbance
in rate
negatively
practice
comparison ofas inflation).
some
term.
correlated
inflation
of
function
The traditional
forecasting However,
with of
depended mode Ph
component
he rest of the increased
paper is organised as follows; value ofpastunemployment
inflation (e.g. or output,
adaptive and 𝜀𝜀
expectations), 𝑡𝑡 is a
2.0 LITERATURE comparison REVIEW
of forecasting models ofworldwide,
expectations
onent
oholic
the
ght)
o a
increased
ofabundant
share scanty.
(weight) However,
studiesliteraturethat
of byspecifically
14.9
studies this
on paper
the
percentage
thatfocus will make
on Swaziland
points
specifically to deviations
use
on
arandom
focus share are
lagged
on at of
(weight)
Swaziland
disturbance
the
itsinflation ofofunemployment
simplest inflation
are
term. andjust the
studies were
one
The traditionalthat lag
output modelled
rate of gap,
specifically from
inflation.
Phillips
as its
hence someon
focus w
th
This chapter
ection 2 is the review of literature, section 3
Central Bank of presents
Swaziland various
© 2018 theories and Therefore in practice inflation depended on 1
hare
argest ofand (weight) other
forecasting of
scanty.
P a g
empiricale studies
empirical
However,
models studies
29.2 that
worldwide, this
percent, specifically
studies paper
on However,
the that
choice
replacing willthisfocus
have
ofmake inflation
‘food on
naturaluse
and Swaziland
beenfunction
ofrate
non-alcoholic areof
U*,inflationandofwherepast scanty. inflation
the However,entiregap, (e.g. curve
thishence adaptive
is
paper di
wil
od oholic non-alcoholic scanty. paper above will make
lagged
equation use becomes: and the output
dother
entral non-alcoholic
Bank of forecasting
Swaziland scanty. models
2018 However,
©worldwide. for various
this economies.
paper willwith make the use
above of equation becomes:
ent specifically
argest undertaken
other
focus empirical
on Swaziland
beverages’
other studies
are
as the
empirical that
component have
studies shifted expectations),
thatup
been theor have down
largest been at its
other
one-for-one simplest withjust
empirical changes 102one |
studies lag of
in thatan
a g e with theInlargest spite of the abundant literature on the
with thethis largest other empirical studies that have Another been
ngs on of
wever,
nother basket. undertaken
comparison
paper
Another
2.1studies
will
Theoretical
worldwide.
ofundertaken
make
weight forecasting
Literature
use
in the models
of worldwide.
consumption worldwide, 𝜋𝜋𝑡𝑡𝑒𝑒 basket.
inflation.
(the expected 𝜋𝜋𝑡𝑡 =rate
Therefore 𝛾𝛾(𝑈𝑈 undertaken
inof
𝑡𝑡 practice 𝑈𝑈 ∗ ) +
−inflation). worldwide.
𝜋𝜋𝑡𝑡−1
inflation , depended th
However,
asket. Another that
undertaken specifically worldwide. focus on Swaziland
nents
pirical
ngs
in the studies
of weightings are that
scanty. have
of significant
However, been
shift
thiswas paper noted will in theonweightings
expectations
make laggedofinflation of
inflation and
were themodelled
output gap, as some hence the ra
2.1 Theoretical2.1 Literature
Theoretical Literature where all the 2.1 Theoretical
variables are as Literature
previously
ng eworldwide.
weightings
by Components2.1.1 of The
use of traditional
other
durable empirical and Phillips
semi-durable curvethat have
studies where all the variables are as previousl
onents
roducts. 2.1 Theoretical Literature products. function above Components
equation
of
determined. past becomes: inflation (e.g. adaptive in
ts. Components been undertaken worldwide.
ehold 2.1.1 The traditional that2.1.1 sawThe Phillips
their traditional
sharescurve(weights)Phillips determined.curve
Research
decliningBulletin 2.1.1 The
Volume 2 just traditional Phillips cur
ng
eights) by declining
ical Literature In 1958,byA.W. Phillips introduced the expectations), Phillips 2.1.2 The atbyits simplest one lag of
2.1.1 TheLiterature traditional Phillips curve 𝜋𝜋 New = Keynesian
𝛾𝛾(𝑈𝑈 − 𝑈𝑈 ∗
) Phillips
+ 𝜋𝜋 , Curve 2.
g )
sehold declining
and by
2.1 Theoretical notable margins are ‘furnishing and household 𝑡𝑡 𝑡𝑡 𝑡𝑡−1
shing and household
curve
Curve,
In 1958, maywhich A.W. bepostulates
subject atotrade-off
the Lucas inflation.
between Different
critique.Therefore authors
current
in practice
In 1958, haveinflation
inflation augmented
A.W. Phillips depending
depended the
introduceon s
InPhillips1958, introduced A.W. Phillips theintroduced Phillips the Phillips
eation and
aditional householdPhillips curve 2.1.2 The
univariate New Keynesian
Phillips curve Phillips
relationship Curve with Th
ng and ‘clothing
points), 2.1.1and
inflation
Curve, TheInequipment’
and
which 1958,
traditional
unemployment.
postulates A.W.(6.8 percentage
Phillips
Phillips
a The curve
trade-offintroduced
Phillips
points),
on lagged
curve
between the
where ‘clothing
Phillips
all
inflation
and the and variables
the
Curve, output which aregap, as
hence
postulates previously
the a trad
Estimated parameters Curve, which are likely postulates to Phillips
change a trade-off as the between
inflation the output
expectations, thereby or avoiding unemployment the
s),
while ‘clothing and
eation
oints) and representation In 1958, A.W.
‘recreation footwear’ Phillips ( introduced
2.4 percentage thepoints) and ‘recreation im
inflation Curve,
and stateswhich
unemployment. that postulates
a lower
The a trade-off
inflation
Phillips above determined.
rate
curve between
implication
equation becomes: that inflation
there is a and long-run trade-off The
unemployment.
W. Phillips
and policy
‘recreation
introduced
Curve, regime whichvaries.the
inflation
postulatesPhillips and
This unemployment.
a trade-off
led to the between Different
introduction The
between
authors
Phillips curve
lower
have
expected augmented
unemployment value ofand the
inflation univariat
higher in(1
nwhile
ge their
points). Meanwhile inflation and
inflation
and culture’ and
unemployment. ( 3.5 percentage
unemployment. The The points).
Phillips Phillips Meanwhile
curve
ch postulates will tend
representation
a trade-offto be representation
associated
states
between thatwith a lower a higher
states thatlevel
inflation a lower of inflation
rate inflation (Norman representation
raterelationship and Richards,states 2010).that Galí a lowe
nts). of
Meanwhile the
curve so-called
representation Friedman-Phelps states that a Phillips
Accelerationist
lower 2.1.2 The curveNew Hence
Keynesian ∗ thePhillips resulting with
Curve inflation
inflation foe
rlecent their
increases in tend items that recorded
theirrepresentation
unemployment, and states The
vice-versa. thatnotablea lower
econometric
increases
inflationandin𝜋𝜋Gertlerrate = 𝛾𝛾(𝑈𝑈(1999)
𝑡𝑡their 𝑡𝑡 − 𝑈𝑈 ) were + 𝜋𝜋𝑡𝑡−1among , authors
d unemployment. will
inflation Theto rate bewill
Phillipsassociated
will curve
tend
tend to with
to a higher
be associated
be associated level
with of
with a higher level of will tend to be associated with a
reases
cent in their Curve
Phillips (APC) where there is (from expectations,
no long-run whoper introduced thereby the
thenunemployment,
makes New Keynesian
avoiding the inflation the Phillips implicationrate,va π
er ’ 5.38
(from 1.43 per
a
specificationcentwill
higher weights
level
oftend of to
this
are be ‘communication’
unemployment,
associated
relationship can and
with
be vice-
a
written higher 1.43 level cent
of and vice-versa.
unemployment,
on states that a lower inflation and
unemployment, vice-versa.
rate The econometric
and vice-versa. where Different
The Curve
all authors
econometric
the (NKPC), variables have
whichaugmented had current
are as the univariate Th
inflation
previously
m 1.43 per cent
erent versa. Thetoeconometric 3.17inflation per cent) specificationandunemployment of this (from 5.38
‘education’
m
ducation’ 5.38
be associated
(from
trade-off
as: 5.38unemployment,
withbetween
specification
relationship a higher of this level
specification
can be ofandofas:
relationship
written and
vice-versa.
thiscan
that
The
bedetermined.
relationship written there
depending
can as
econometric is written
be curve
a long-run
on eithersomespecification trade-off
measure
the output of this
of between
the
gapoutput or unemp
relationship lowe fu
Phillips or unemployment relationship
gap and the with expected inflation
on’(CPI)
fferent (from 5.38 percent to 8.77 percent).The different
cent).The differentspecification of this relationship can be
unemployment written and higher inflation (Norman
HenceEt πand ou
ent, and presented as:
vice-versa. below:
The econometric
as: ∗ value
expectations, of inflation expected
thereby as:in the future
avoidingnext inflation,
period.
the implication t+1
The
of the different 𝜋𝜋 = 𝛼𝛼𝜋𝜋𝑡𝑡−1 + 𝛽𝛽(𝑈𝑈 of𝑡𝑡 − 𝑈𝑈 ) + 𝜀𝜀𝑡𝑡 Price Indices
(CPI)
Price Indices (CPI) 𝑡𝑡composition Consumer 2.1.2 The the New (CPI)
resultingKeynesian inflation Phillipsequation Curve (NKPC) then
n of this relationshipas: can be written ∗ Richards, that there 2010). is a is Galí
presented
long-run and trade-off Gertler
as follows:
between (1999) lower wer re
∗
Indices
lountries
may
of the iswhere (CPI)where 𝜋𝜋 = 𝛼𝛼𝜋𝜋
𝑡𝑡 denotes
weighting + 𝛽𝛽(𝑈𝑈 −
the𝜋𝜋𝑡𝑡inflation
structure 𝑡𝑡= 𝑈𝑈
𝛼𝛼𝜋𝜋
∗across ) +
rate,𝜀𝜀
+𝑒𝑒countries
𝑡𝑡𝛽𝛽(𝑈𝑈𝑡𝑡 − is 𝑈𝑈 is ∗ makes
)+ one the
𝜀𝜀𝑡𝑡 of the inflation rate, 𝜋𝜋 ,= to𝛼𝛼𝜋𝜋 depend + on
𝛽𝛽(𝑈𝑈 − 𝑈𝑈 )
one of𝜋𝜋the denotes 𝜋𝜋 𝑡𝑡 = 𝑡𝑡−1
the 𝛾𝛾(𝑈𝑈
inflation
𝑡𝑡 − 𝑈𝑈 )
rate, +
𝑡𝑡−1 𝜋𝜋
𝑈𝑈 𝑡𝑡 is the log
𝑡𝑡 𝑡𝑡−1 𝑡𝑡
hy
iesmay is one of the
the log 𝑡𝑡 of unemployment or output,
𝜋𝜋𝑡𝑡 = 𝛼𝛼𝜋𝜋𝑡𝑡−1 + 𝛽𝛽(𝑈𝑈𝑡𝑡𝑡𝑡 −Different ∗ is
𝑈𝑈among)+ 𝜀𝜀𝑡𝑡 either
unemployment authors
authors the have output andwho gap or inflation
higher
augmented unemployment
introduced the univariate (Norman gap and
the New
al
ation differential the may often cited reasons why inflation differential may
ofwhere steady-state
unemployment
𝜋𝜋𝑡𝑡 denotes where or value
output,
the 𝜋𝜋inflation of 𝑈𝑈 ∗
unemployment
is the
𝑡𝑡 denotes the inflation
steady-state
rate, 𝑈𝑈𝑡𝑡 is the log or rate, and𝑈𝑈expected
𝑡𝑡 is the log futurewhere 𝜋𝜋𝜋𝜋𝑡𝑡 𝑡𝑡 =denotes
inflation, 𝛼𝛼𝐸𝐸 𝜋𝜋 the.+ 𝛽𝛽(𝑈𝑈𝑡𝑡 −rat
inflation
Thus tr𝑈𝑈
= 𝛼𝛼𝜋𝜋𝑡𝑡−1Where
differential + may
𝛽𝛽(𝑈𝑈 inflation,
𝑡𝑡 −
output, ∗
𝑈𝑈where
)persist
and + 𝜀𝜀𝑡𝑡 𝜋𝜋is𝜋𝜋𝑡𝑡across
,random
is negatively
𝑡𝑡adenotes disturbance
countries. Phillips
correlated Keynesian
term. Richards,the curve
with NKPC 2010).
Phillips is Galí
relationship
presented Curve and as with 𝑡𝑡 𝑡𝑡+1
Gertler
(NKPC),
follows: (1999)
inflationwhich were had
lows; value of unemployment
of unemployment or output,
of unemployment or the 𝑈𝑈 ∗ isinflation
output, or and
theoutput, rate,
steady-state𝜀𝜀𝑡𝑡𝑈𝑈is ∗ 𝑈𝑈a𝑡𝑡 is the log
is the steady-state of unemployment or output, 𝑈𝑈 is t
∗
deviations of of the
unemployment unemployment oroutput,
output, and 𝑈𝑈rate∗expectations,
is theis
among
from
steady-state itsauthors
thereby where who U introduced
avoiding thetheimplication
isunemployment unemploymentthe New
random disturbance term. The traditional Phillips
91
enotes
ion 3 the value
inflation of unemployment
rate, 𝑈𝑈valueis the of log or
unemployment or 𝜀𝜀 output, a and 𝜀𝜀 is a value of or outpu
organised as follows; The𝑡𝑡 rest
ollows; the paper is organised as follows; 𝑡𝑡
Central Bank of
𝑡𝑡
Swaziland © 2018
value of and unemployment that Keynesian
there
and is aPhillips
aislong-run
Central Bank Of Curve (NKPC),
trade-off
Swaziland © betweenwhich
2018 lower had
sed
ment
tion as
3 or natural
follows;
random
output, 𝑈𝑈 ∗rate
disturbance
is the U*,
sectionsteady-state
random 2 term.
is where The
disturbance
the review theor
traditionalterm.
of
output,
entire Phillips
The
literature,
P a g curve
traditional
e
𝜀𝜀𝑡𝑡 is
section Phillips3 differencerandom disturbance between the term.formulaThe trad W
literature, section 3
unemployment and higher inflation (Norman and
102 |
is presented as follows: less volatile correlated. This model was in reference to the
than it wa
(1998) ha
either
thethe
een where Uoutput
is the gap or new
unemployment
unemployment rate. The gap
onlyand lower-case letters designate
formulation of the inflation in thewhere
onsome eras. In rate
inflation lightπof this dilemma,
conveys inertia.supply
of normalized D is an first henc
shock variable
three basic determinants t−1 of the inflation t rate:
difference between the formulation ∗ ) of the new mean squared error
correcti for
expected 𝑡𝑡future𝑡𝑡 inflation,
nal Phillips 𝜋𝜋 =
curve 𝛼𝛼𝐸𝐸 𝜋𝜋
lies𝑡𝑡+1 +
in 𝛽𝛽 (
the 𝑈𝑈 𝑡𝑡 E−
factt π
𝑈𝑈
Gordon
CENTRAL BANK O 𝑡𝑡
t+1 .
+ Thus
𝜀𝜀(1982)
𝑡𝑡
index the
F S W Aof
NKPC
differences
developed
Z I Lnormalized
AND | the of logarithms,
empirical
R E Sexcess H demand,
E A R C serially triangle
BULLETIN
upper-case
VzOt LisUThe
uncorrelated aEvector
M 2
letters
error term. forec
inertia, demand, and supply. forecasts general has declined
relies and the
on expectation traditional Phillips curve
of the future,model. That wasdesignatelies in the fact logarithms of levels, and L is a(2008),
univ
is presented as follows:
where U is the unemployment rate. The of normalized after
onlyspecification the US
supply experiencedshock Watson the
variables, first (2008) and ϵt presented is a the
of this model can be written as:
hard to forecast Se
due
that theto
st inflation former determine relies on expectation of the future,
current conditio
base
difference between the formulation of oil shock between
serially
the polynomial 1973-1975
new uncorrelated error in the lag
which operator.
caused
term. DistributedStockThe dependent
andLag (ADL)
regressive percentage terms inP
rather𝜋𝜋𝑡𝑡than = 𝛼𝛼𝐸𝐸 past inflation
𝑡𝑡 𝜋𝜋𝑡𝑡+1 + 𝛽𝛽(𝑈𝑈toinflation
𝑡𝑡 − 𝑈𝑈 ∗ 𝑡𝑡 ) and
determine + 𝜀𝜀current
𝑡𝑡 unemployment 𝜋𝜋 = 𝑎𝑎(𝐿𝐿)𝜋𝜋 + 𝑏𝑏(𝐿𝐿)𝐷𝐷 + 𝑐𝑐(𝐿𝐿)𝑧𝑧 + 𝜖𝜖 𝑡𝑡 of lagsof intere the w
and the traditional Phillips curve lies in the Watson fact variable (2008)
𝑡𝑡π is the
t to inflation
presented be positively
𝑡𝑡−1 rate
the
𝑡𝑡 and a series
The
𝑡𝑡
Auto-
inflation. Phillips curve as the
forecastingversion of
modelsthe Go
ov
where U is the unemployment rate. The where lower-case letters designate first gap have
in (1t
idthat the correlated. This onmodel
the was
inflation in reference
rate πt−1 to
conveys the inertia. Dtfirst
is an
wherePhillips
only Uformer
curve
is the
difference relies on
unemployment
between expectation of rate.
the formulation theregressive
future, Theof whereonly
differences lower-case
Distributed ofLag letters
(ADL)
triangle
logarithms, designate
Phillips
model without
henceCurve
upper-case itthe
hassupplybecome sh
cop
the
2.1.3 new
The and
Hybrid the
Phillipstraditional threePhillips
curve basic determinants
curve of the inflation rate: VECM forec
ryrather
power than ofpast theinflation NKPC towas determine current
Phillips index of
letters
differences normalized
designate
of excess
logarithms
logarithms, demand, of levels,
upper-case z is a
and vector
letters
on curve as the version andofpresented
the Gordon (1982)
t
itforecasters
as:
difference lies inbetween the fact that the formulation
the former relies of the new to provid
The
inflation.
tically whenexplanatory Gali and
expectation of the future, rather than powerGertler inertia,
of the demand,
NKPC past was
of
L is asupply.
and
normalized
designate
polynomial
supply
logarithms
The in the
shock
of
general lag operator.
variables,
levels, and and
L
The
ϵ
is is
a aabout(2 fo
triangle dependent
model without variablethe supply is the shock inflation rate and model. Al
variables
univariate t
and improved
the traditional Phillips curve lies in the fact co
inflationfeatures
ed additional to determine
drastically in thewhen current
pure Galiinflation.
specification and of Gertler
this model
seriallya series
polynomial
canofbelags
uncorrelated
in
written
the on the
lag
as:
error 𝜋𝜋term.
inflation
operator. = rate
𝑡𝑡 The 𝑎𝑎(𝐿𝐿)𝜋𝜋
Stock
dependent𝑡𝑡−1 + and withSeve
𝑏𝑏(𝐿𝐿)𝐷𝐷 +
𝑡𝑡rel
2.1.3 The Hybrid Phillips curve and presented it as:
conveys inertia. Dt is an index of normalized based on the US data,
that the
(1999)
g NKPC. 2.1.3
former
introduced
TheyThe added
relies the
Hybridadditional
on expectation
Phillips pastcurve features in of the thepure future, in off
The The explanatory power of ofthetheNKPC 𝜋𝜋 = 𝑎𝑎(𝐿𝐿)𝜋𝜋
was variable
Watson + π
𝑏𝑏(𝐿𝐿)𝐷𝐷t is
(2008) +the inflation
𝑐𝑐(𝐿𝐿)𝑧𝑧 presented
+ 𝜖𝜖 rate and the a seriesThe
the world. of lags
Auto- It Alturki
is for th
explanatory power 𝑡𝑡
NKPCthe was 𝑡𝑡−1 excess demand, 𝑡𝑡 where all theofvariables
𝑡𝑡 is a 𝑡𝑡vector normalized are as explain
(199
rather
on an than
asforward explanatory past variable,
looking inflation
NKPC. and to determine
They added current
pastsupply
on the 𝜋𝜋 = 𝑎𝑎(𝐿𝐿)𝜋𝜋
shock
𝑡𝑡 inflation +
variables,
rate πt−1Lag
𝑡𝑡−1 𝑏𝑏(𝐿𝐿)𝐷𝐷 and
conveys +
𝑡𝑡(ADL) 𝜖𝜖 , is
𝑡𝑡inertia.a serially
Dtpersistently
is an
ga
improvedimproved drasticallydrastically when whenGali Galiand and Gertler Gertler regressive Distributed Phillips
have Curve tr
values of inflation as an explanatorywhere variable,
lower-case and uncorrelated
letters error
designate Central
term.Bank Stock
first of Swaziland
and Watson © 2018 andcorre
long
VE
inflation.
al unit labor
(1999) cost
introduced other than
additional the features in the index ofcurve
normalized Pexcess
a g e demand, zt is a vector
(1999) pure introduced
forward additional
looking NKPC. featuresThey inaddedthe
where pure
the Phillips
all (2008)
the as the
presented
variables are version
as the The
explained of Auto-regressive
thebefore. Gordon
forecasting (1982) inflation, d
further
proxy for used marginal real unit cost.labor cost other than
Thedifferences theDistributed
of logarithms, upper-case letters Curve Tajikista
(200ab
pastlooking values of inflation of normalized Lag (ADL)
supply shock Phillipsvariables, Phillips
andvariables
ϵt is a
forward NKPC. Theyasadded an explanatory
the past triangle model without the supply shock
2.1.3
ulation The
output was gap
variable, Hybrid
referred andto proxy Phillips
to asfor
further curve
marginal
real unitcost.
thedesignate
used CentrallaborThe
logarithms Bank curveof as
Swaziland the
of levels, and L is a version
© 2018 of the Gordon (1982) cond
P a g e serially
triangle uncorrelated
model without error the term. supply Stock shock and wi
valuescost of inflation
other than as an the explanatory
output gap variable,
to proxy and
forand presented it as: Several authors have
The resulting
explanatory
s it displays formulation
features ofpowerboth was referred
of
thepolynomial the toNKPC inasthethelag was
variablesoperator. and The presented dependent it as: of in
further marginal
used real cost. unit The resulting
labor costformulation
other than was theWatson (2008) presented the in Theforecasting Auto- inflat
Ali
he newhybrid
improved referred
PhillipsNKPC, to
curve
drastically asthe
as it
of displays
hybrid
the form:
when features
NKPC,
variable
Galias itπof both inflation
displays
t is theGertler
and the rate and a series+ of𝑏𝑏(𝐿𝐿)𝐷𝐷
lags + 𝜖𝜖 , gap
outputfeatures gap toof both proxythefor marginaland cost. The regressive 𝜋𝜋𝑡𝑡 = 𝑎𝑎(𝐿𝐿)𝜋𝜋
Distributed 𝑡𝑡−1 Lag (ADL) 𝑡𝑡 Phillips 𝑡𝑡 Curve
traditional the new (1998) improved an th
traditional and the new Phillips on curve
the of the form:
inflation rate π conveys inertia. D is an VECM
(1999)
resulting 𝑈𝑈introduced
+ 𝛽𝛽 (Phillips 𝑈𝑈 ∗curve
𝑡𝑡 −formulation
ofadditional
𝑡𝑡 ) + 𝜇𝜇𝜋𝜋𝑡𝑡−1was
the + form:
𝜀𝜀𝑡𝑡 referred features to asin the thePhillips
pure t−1
where curveall as thethevariables
version t are as explained
of the Gordon correction (1982)
+1
index of normalized where before.
excess all the
demand,variables z isarea as explained
vector before. modelabou (VE
Ta
∗ t
forward
hybrid NKPC, = 𝛼𝛼𝐸𝐸
𝜋𝜋𝑡𝑡 looking + 𝛽𝛽 (𝑈𝑈𝑡𝑡 features
as𝑡𝑡 𝜋𝜋it𝑡𝑡+1displays
NKPC. −They𝑈𝑈 𝑡𝑡 ) +added𝜇𝜇𝜋𝜋both
of 𝑡𝑡−1 + the 𝜀𝜀𝑡𝑡 triangle
the past model without the supply shock variables
(2008), by imposin
ables are as explained before. Theof normalized supply Centralshock Bank of Swaziland
variables, and© 2018ϵt is a with
traditional and the new Phillips curve of the form: P and
a g
2.2
e
Empirical
presented it as:Literature
values
ce inWhere of inflation
Where
this the the
hybrid
variables
variables as an
NKPC are isexplanatory
are as
asthe explained
explained variable,
before.
before. Theand Stock and Watson (2006) could conditions
not have put to better an
serially uncorrelated
The major difference in this hybrid NKPC it any better when they stated that inflation error term. Stock and
of, interest rates, the Alture
furthermajor
term 𝜋𝜋 𝜇𝜇𝜋𝜋
is =used
the
𝑡𝑡 𝑡𝑡−1 ,difference
𝛼𝛼𝐸𝐸 real+ unit
𝑡𝑡 𝜋𝜋𝑡𝑡+1 are
which
addition (in
𝛽𝛽of labor
this
𝑈𝑈𝑡𝑡the
the − 𝑈𝑈term
lagged 𝑡𝑡 )
cost
∗ hybrid
+
Watson other
NKPC
𝜇𝜇𝜋𝜋𝑡𝑡−1 +,(2008)
𝜀𝜀which
𝑡𝑡
than
is the
thepresented 𝜋𝜋 𝑡𝑡 = 𝑎𝑎(𝐿𝐿)𝜋𝜋
the
has become both harder and easier to The𝑡𝑡−1 + 𝑏𝑏(𝐿𝐿)𝐷𝐷
Auto- 𝑡𝑡 + 𝜖𝜖 𝑡𝑡
are theof lagged
addition the proxy term values 𝜇𝜇𝜋𝜋𝑡𝑡−1 of, which
current areinflation.
thecost. laggedThe forecast. They state that it’s easiergap in the model.and
because The
output
ent inflation. gap
Furthermore
to
Furthermore
there must
for
there
be
marginal
regressive
the Distributed
restriction Lag (ADL) Phillips Curve
Where the variables are as explained before. Thewhere inflationall thehas variablesbecome areless as explained
volatileVECM before.
than it Tajik
values
triction that
that of𝛼𝛼 + current
𝜇𝜇 = 0 to
toinflation.
rule
rule out
out Furthermore
a long run trade- there provide consid
resulting
majoroffdifference formulation in past this was hybrid referred
PhillipsNKPC curve istoas theas the in the
the version
was of the1970’s
Central Bank of Swaziland © 2018
Gordonand the (1982) root mean squared
between
must be the restriction 103 the and
that expected inflation error (RMSE) of naïve inflation about
forecasts forecasting the
| 𝛼𝛼 + 𝜇𝜇 = 0 to rule Pout
triangle model without a g e the supply shock variables
hybrid
additionNKPC, of the as
variables. term it 𝜇𝜇𝜋𝜋
displays
𝑡𝑡−1 , which features
are the lagged of both has the declined. They further state that it is
103 | with relatively small e
and presented it as: hard to forecast due to less improvement in
values
traditional of
2.1.4 and current
The triangle inflation.
the newModel Furthermore
Phillips curve of the form: there
percentage terms of standard multivariate
According to the
must be the restriction that 𝛼𝛼 + 𝜇𝜇 = 0 to rule new Keynesian theories, out forecasting models over Alturki and Vtyurina
univariate
one of the drawbacks of∗ the standard benchmarks, hence it has become much 𝜋𝜋 𝑡𝑡 = 𝑎𝑎(𝐿𝐿)𝜋𝜋 𝑡𝑡−1 + 𝑏𝑏(𝐿𝐿)𝐷𝐷 𝑡𝑡 + 𝜖𝜖 𝑡𝑡 ,
= 𝛼𝛼𝐸𝐸𝑡𝑡curve
𝜋𝜋𝑡𝑡Phillips 𝜋𝜋𝑡𝑡+1 + 𝛽𝛽 (𝑈𝑈𝑡𝑡 − 𝑈𝑈 𝑡𝑡 )as+the
was identified 103 |+ 𝜀𝜀𝑡𝑡
𝜇𝜇𝜋𝜋failure
𝑡𝑡−1 and long term dynami
difficult for inflation forecasters to provide
to explain the behavior ofwhere inflation all the in some variables value are as addedexplained beyond before.a univariateTajikistan model. using the
eras. In light
Where the variables are as explained of this dilemma, Gordon (1982)
Central Bank before. of Swaziland The
Although
© 2018 this assertion was based on the
developed the empiricalP atriangle ge
model. US data, it is quite true to the rest of the
major That difference
was after in the this hybrid NKPC
US experienced the firstis world. the It is for that reason that forecasters
oil shock between 1973-1975 which caused have persistently tried different models in
addition inflationof the andterm 𝜇𝜇𝜋𝜋𝑡𝑡−1 , which
unemployment to be are positively the lagged forecasting inflation, deriving mixed results.
correlated. This model was in reference to
valuestheofthree current inflation. Furthermore
basic determinants of the inflation there Several authors have tried different
rate: inertia, demand, and supply. The approaches in forecasting inflation. Engert
must be the restriction that 𝛼𝛼 + 𝜇𝜇 = 0 to rule out
general specification of this model can be and Hendry (1998) improved the M1-based
written as: 103 vector
| error-correction model (VECM)
developed by Hendry (2008), by imposing a 95) to last quarter of 2007-08 (Q4: 2007-
set of equilibrium conditions to better anchor 08), a VAR was developed and subsequently
the long run behavior of interest rates, the using Minnesota prior or Litterman’s prior
exchange rate and the output gap in the proposed by Litterman in 1980, a BVAR model
model. They found that this extended-VECM was developed. Based on the comparison of
provide considerable leading information forecasting performance of a VAR and BVAR
about forecasting the eight-quarter inflation model, measured in terms of out-of-sample
rate with relatively small errors. percentage root mean square error (RMSE),
they found that BVAR model performed
Alturki and Vtyurina (2010) explained the better than VAR model in the case of inflation
short and long term dynamics and inflation as well as IIP growth forecast.
forecasts in Tajikistan using the VECM and
autoregressive moving average (ARMA). Other authors who assessed the performance
Their results show that the ARMA model out- of the BVAR against other models includes
of-sample forecast clearly outperforms the Kenny et al. (2008), Giannone et al.
forecasts of the VECM, possibly due to the (2010), Cuaresma et al. (2009) and Huang
limited sample used in the VECM. Engert and (2012). Kenny et al. (2008), focused on the
Hendry (1998) improved the M1-based vector development of multiple time series models
error-correction model (VECM) developed for forecasting Irish Inflation. They employed
by Hendry (2008), by imposing a set of the BVAR, which allows the estimated
equilibrium conditions to better anchor the models to combine the evidence in the data
long run behavior of interest rates, the with any prior information which may also
exchange rate and the output gap in the be available. A large selection of inflation
model. They found that this extended-VECM indicators were assessed as potential
provide considerable leading information candidates for inclusion in a BVAR. The
about forecasting the eight-quarter inflation results confirm the significant improvement
rate with relatively small errors. Alturki and in forecasting performance which can be
Vtyurina (2010) explained the short and long obtained by the use of Bayesian techniques
term dynamics and inflation forecasts in over the other forecasting models.
Tajikistan using the VECM and autoregressive
moving average (ARMA). Their results show Giannone et al. (2010) constructed a large
that the ARMA model out-of-sample forecast Bayesian Vector Autoregressive model (BVAR)
clearly outperforms the forecasts of the for the Euro Area that captures the complex
VECM, possibly due to the limited sample dynamic inter-relationships between the
used in the VECM. main components of the Harmonized
Index of Consumer Price (HICP) and their
Biswas et al. (2010) developed a forecasting determinants. They evaluated the model in
model for inflation as well as Index of Industrial real time and found that it produces accurate
Production (IIP) growth in a multivariate forecasts than other multivariate models.
time series Bayesian framework, known
as Bayesian Vector Autoregressive (BVAR) Akdogan et al. (2012) produced short-term
Model. They argued that the main advantage forecasts for inflation in Turkey, using a large
of using this model is the incorporation of number of econometric models such as the
prior information which may boost the univariate ARIMA models, decomposition
forecasting performance of the model. Using based models, a Phillips curve motivated
the quarterly data on wholesale price index time varying parameter model, a suit of
(WPI), narrow money ( M1) and IIP during the VAR and Bayesian VAR models and dynamic
period of first quarter of 1994-95 (Q1: 1994- factor models. Their result suggests that the
the variables which requires pretesting of on the parameters of the VAR. This ideology The BVAR
parameters is
of in
the the
VAR. Thisfamily
ideology ofof the
imposing VAR models;
The
available ARDL
in Bayesianmodel 3.2.5
can
Statistics be
by T
t
restrictions
of imposing in econometric estimations is analyse the long-run relati
the stationarity property of the time series. however it differsrestrictions
by imposingin econometric
restrictions information
on the on parameters
(ARDLof inte
94
some known distribution called prior test, we determine the orde
Central Bank Of Swaziland © 2018 idea behind the BVAR is to impose restrictions ARDL
In this study a BVAR model for Sw
distributions. variables using the unit root
developed which will allow fo
(shrinkage) by the way of prior information
between the on
to ensure develo
that macroeconomi
major the variables
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
on coefficients of the VAR. Therefore in The ARDL Phillips curve model used in this
BVAR model, the coefficients are considered study will follow that in Stock and Watson
as variables with some known distribution (2008) and be presented as follows:
called prior distributions. Research Bulletin Volume 2
∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡 = 𝑎𝑎0 +
∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡 = 𝑎𝑎0 + 𝑏𝑏1 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡−1 + 𝑏𝑏2 ∆𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡 Thus, an e
In this study a BVAR model for Swaziland will 𝑝𝑝 +
as follows
be developed which will allow for interactions + ∑ 𝑐𝑐𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡−𝑖𝑖
between the major macroeconomic variables 𝑖𝑖=1 Research Bulletin V
and further produce inflation forecasts 𝑝𝑝 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶
+
which will be compared with results from + ∑ 𝑑𝑑𝑖𝑖 ∆𝑂𝑂𝑂𝑂𝑂𝑂
other forecasting models. ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡 = 𝑎𝑎0 + 𝑏𝑏1 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 +𝑡𝑡−𝑖𝑖𝑏𝑏+ 𝜖𝜖∆𝑂𝑂𝑂𝑂𝑂𝑂
𝑡𝑡−1
𝑡𝑡
2 𝑡𝑡 Thu
𝑖𝑖=1
𝑝𝑝 Where ∆ is the diffe
Despite the abundance of the priories inWhere ∆ is the difference operator, 𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡 islogarithm the
as f
of Swaz
literature, many authors found that thelogarithm + ∑of𝑐𝑐𝑖𝑖Swaziland ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 𝑡𝑡−𝑖𝑖
CPI and 𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡 islogarithm
Research Bullet
the ResearchofBulletin
Litterman priori outperforms the other Where Where outp
𝑖𝑖=1 is the
∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 difference
𝑡𝑡 = 𝑎𝑎0 + 𝑏𝑏1 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 operator, + 𝑏𝑏2 ∆𝑂𝑂𝑂𝑂𝑂𝑂
𝑡𝑡−1 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 𝑡𝑡 = 𝑡𝑡is𝑎𝑎0 + 𝑏𝑏1ci ∆T
logarithm
priories. For example Huang (2012) found the logarithm of output minus the logarithm of
potential output, a
∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 = 𝑎𝑎 +
𝑡𝑡 of 0Swaziland 𝑏𝑏 𝑝𝑝∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 CPI𝑡𝑡−1 and2+ 𝑏𝑏 ∆𝑂𝑂𝑂𝑂𝑂𝑂 𝑡𝑡 coefficien
is 𝑝𝑝 Th
that the Minnesota prior of Litterman (1986)potential
𝑝𝑝 1
output, and 𝜖𝜖𝑡𝑡 is 𝑝𝑝the error term. first The step in the ∆𝑙𝑙aA
the logarithm of output + ∑ 𝑐𝑐𝑖𝑖minus ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 the 𝑡𝑡−𝑖𝑖 logarithm + ∑as𝑐𝑐
equilibriu
is the best among all priors consideredfirst+ of ∑in𝑑𝑑the
step
potential ∆𝑂𝑂𝑂𝑂𝑂𝑂
ARDL
output, + bounds
∑ and +𝑐𝑐 𝜖𝜖 approach
is
∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 the is
error to term.
𝑖𝑖 𝑡𝑡−𝑖𝑖 𝑖𝑖 𝑡𝑡
𝑖𝑖=1 𝑡𝑡−𝑖𝑖 estimate the abov
Once 𝑝𝑝the
𝑖𝑖=1
when forecasting Chinese inflation and The first step in the𝑖𝑖=1 ARDL by bounds ordinaryapproach is
𝑝𝑝
estimate the above equation least
output. Furthermore the Minnesota prior to 𝑖𝑖=1 estimate the above equation
squares (OLS). cumulativ This
∆
𝑡𝑡−𝑖𝑖 by 𝑡𝑡ordinary + ∑ 𝑑𝑑
+ 𝑝𝑝∑ 𝑑𝑑 ∆𝑂𝑂𝑂𝑂𝑂𝑂 + 𝜖𝜖
𝑖𝑖
squares
is fine-tuned to most macroeconomic data least squares (OLS). (OLS). This would test for the existence a of
long-run relation
+∑ 𝑖𝑖=1𝑑𝑑𝑖𝑖 ∆𝑂𝑂𝑂𝑂𝑂𝑂
This would 𝑡𝑡−𝑖𝑖 + test𝜖𝜖𝑡𝑡 for the and the 𝑖𝑖=1 C
in particular somethingWhere that shows whyadifference
∆ is the long-run operator,
relationship
existence of a long-run relationship among
𝑖𝑖=1 𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶
the variables is the
by among
𝑡𝑡 conducting an F te
applied to
Bayesian VARS can provide more freedom the Where variables ∆ is the by difference
conducting operator, Where
an F𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶
∆test 𝑡𝑡 is
is the for the
difference
conducting an F test for the joint significance of
the coefficients
to modelers by allowing logarithm
subjective as ofwellSwaziland
the Where
joint
logarithm ∆ isCPI the
significance and
of difference
Swaziland 𝑂𝑂𝑂𝑂𝑂𝑂
operator,
ofCPI the andis the
𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶
coefficients
𝑂𝑂𝑂𝑂𝑂𝑂 𝑡𝑡 is
isthe ettheal., 199o
the coefficients of the lagged levels 𝑡𝑡logarithm
of the 𝑡𝑡 of Swaziland
as objective fine tuning (Litterman, 1986). of logarithm the lagged levels
ofofSwaziland ofminus the and
CPI variables. variables.
𝑂𝑂𝑂𝑂𝑂𝑂 𝑡𝑡 is The the The Wh nullW
In that regard, this studylogarithm
will use onlyofthevariables. logarithm output the
logarithm logarithm of of
output m
outputnull Theminus null hypothesis
hypothesis the
to be to logarithm
be tested isb H=0:of
tested is b =
3.2.6 The
0 against
Wh c
Litterman or Minnesota prior in estimatingb = b logarithm potential of output andminus the logarithm of and 𝜖𝜖
1 2
2 = 0 against
against the the
output, 𝜖𝜖𝑡𝑡 is the
alternative H1: ci = diTwo
alternative
error
potential term.
= 0..sets
output,
Two
The
The coe
fourth co 𝑡𝑡
1
of criticale
the BVAR. potential output, sets andpotential
first 𝜖𝜖step
𝑡𝑡
ofcritical
critical is
output, the
in values and error
the ARDL abounds 𝜖𝜖 𝑡𝑡 is term.
the error
firstapproach The
term.
step in is The
the toARDL
Two sets of values for for a given given
significance significance models eq
wO
first
level step
can be
estimate in the
the determinedARDL
above equation bounds (Pesaranby approach level can
ordinary
estimate the above et is
al., to
least
equ
be deter
equ
3.2.5 The Autoregressive first step in
Distributed Lagthe
level ARDL
can be determined
2001).
estimate The bounds
the firstabove level approach
(Pesaran
equationis
et al., 2001).
calculated
by
is
Theon
ordinary
tofirst the
least
the Trian
level is cc
On
(ARDL) Phillips Curve The assumption
squares
first level is that
(OLS).
calculated
This
allwould
would test
on the assumption
variables
for
squares
included
the existence
(OLS).
in the
of
This
developedOnc
woul cua
that all variables in
The ARDL model can be used estimate the above
to empirically equation
squares
a long-run (OLS). This by
relationship ordinary test for
amonga the the
long-runleast
existence
variables of
relationshipby basi
that ARDL model
all variables are
included integrated
in the ARDL of model order are zero, three
cum ana
analyse the long-run relationships and a conducting
long-run relationship among integratedbyof order
the significance
variables
while the second an F test onefor is calculated
the joint
conducting on an the F oftest for
dynamic interactions among squares (OLS). integrated
the variables This would
assumption
of order
conducting
test zero,for
that
an F the
while
test
the theexistence
variables
for
second one
thelaggedjoint are integrated
significance
isof
calculatedofon the ea
inertia, ap
de
the coefficients
of interest in a model. The model can becalculated on the assumption that the variables
of the order one. The
of the
nullvariables hypothesis
the levels
coefficients
of
of
nowith and
the of thethe
et
a long-run relationship among
coefficients of the the lagged levels are by integrated
of the of o
estimated by using the ARDL bounds test variables. The null hypothesis variables. to be tested TheisincludesH0: hypo3
null
cointegration
are integrated of order is rejected
one. The null when theofvalue
hypothesis of
no cointegration app
cointegration procedure, developed by variables.
1 = b2
btest The
= 0 againstnull hypothesis
the alternative to be tested is H =0:0. the Ta
3.2
the statistic exceeds the 1 =H
bupper b12: =ccritical
i0= against
di triangle
conducting
Pesaran et al. (2001). The ARDL bounds test an F
of test
no
b1Two
for
cointegration
= b2sets
the is
= 0ofagainst
joint
rejected
the
significance
when
alternative
the value
H1sets the of of
test statistic
: csignificance exc
Th
i =ifdi the = 0. value
bounds
is based on the assumption that the variablesthe test statistic
value, while
critical valuesit
exceeds the upper critical bounds
isforaccepted
aTwo given of critical et
variables. am
F-statistic is lower than the lower bounds
are I(0) or I(1). So, beforethe coefficients
this test,value,of the canoflagged levels of et the
Two sets critical values for a given value,
significance while it is acc
mo
applying level be determined (Pesaran level can al.,be2001).determined t
while itOtherwise,
value. is accepted ifthe the F-statistic
cointegration is lower test specificati
is
we determine the order of integration of level can be determined (Pesaran et thanal., the
2001). lower b
thed
all variables using the unitvariables. The
root tests. Thethan the
The first level is calculated on
nullinconclusive.
hypothesis
lower boundstovalue. be tested Otherwise,is
Thethefirst assumption
theH0level
cointegration
: used 3.2
is calcul
in th
test is
The det
thatfirst level is calculated
all variables included inon the the
that ARDL allassumption
model are
variables include
objective is to ensure that the variables arecointegration test is inconclusive.
b1 = results.
b2 = 0Inagainst the
Whenever
that allalternative
integrated
the
variables of order
existence
included H1in:while
zero, cthe
iof= ARDL
the da isecond
integrated =model 0.ofone
long-run areis The
order
thr
zeroin
not I(2) so as to avoid spurious the Whenever the exist
relationship
integrated
is
ofon
found,
order
the short-run dynamic ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡
ine
presence of variables integrated of orderWhenever the existence of azero, long-run whilerelationship
the second onethe is w
Two sets of critical calculated
values
parameters for
can a the
be assumption
given
obtained significance that
bycalculated
estimatingthefound,
is variables
on an mod
the assum
short
two, we cannot interpret the values of Fis found, calculated
the short-runon theof assumption
dynamic parametersthat the can variables wiin
error are correction
integrated model
order associatedone. The arenull with
integrated
be hypothesis
obtained the of order
by es
statistics provided by Pesaran et al. (2001). the
level can be determined long-run
be obtainedareof integrated
noby (Pesaran
estimates.
estimating
cointegration of order Thus, et
one.
anrejected
is erroranal.,
Theerrorwhen
of 2001).
nocorrection
null
correction hypothesis
the value
cointegration
model associatedof is ret
inc
model of no cointegration
associated with theis long-run
rejected when the value of
estimates. tri
the test statistic
The first level is calculated onexceeds
the the upper critical
assumption
the test bounds
statistic
Central
Page
devv
Bankexceeds
of Swazil
95
the test
Central Bank statistic ©
of Swaziland exceeds
2018 the upper critical bounds vas
Page
value, while it is accepted if the value,
F-statistic
whileisitlower
is accepted
that all variables included
value, thein
than while the
it is
lower ARDL
accepted
bounds model
Central Bank Of Swaziland © 2018
if the are
F-statistic
value. islower
Otherwise,
than the lower thre
sp
theboundu
than the lower bounds value. Otherwise, the us
steady inflation when the output gap and the A
as follows: Where the lagged rate supply the lags
of inflation shock of variable.
oil
𝐶𝐶𝐶𝐶𝐶𝐶 prices, This
conveys which is ARIMA
specification a proxy predictsfor(2001)
model, thewhe
𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡−𝑖𝑖Research Bulletin Volume 2 supply shock terms are which is the and
𝑡𝑡−𝑖𝑖 zero,
all proxy hencefor excess it is demand, 𝑂𝑂 T
inertia, 𝑂𝑂𝑂𝑂𝑂𝑂 are steady
the supplylags shock
inflation
of the variable.
when
logarithm the This of specification
output For gap the and predictsof the f
the
Swaziland
BtheU L Llags ofV Ooil L U Mprices, which is a prox(
correction 𝑝𝑝
𝑏𝑏2 ∆𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡 modelThus, willan beerror specified 𝑡𝑡
correction
C E N T R A Lmodel B A N Kwill O F be A Z I L Aalways
S Wspecified ND | estimatedR E S E A R C H without E T I Na constant E 2 term
∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡 = 𝑎𝑎0 + ∑output minus
𝑐𝑐𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 the logarithm steady
supply shock inflation terms
of potentialsupply
when
are output,
shock
the
all zero, output
variable. and
inflation. gap and
hence
This it rate
isthe of
specification
as follows: (Gordon,
𝑡𝑡−𝑖𝑖 2009). This model will be estimated by o
𝑂𝑂𝑡𝑡−𝑖𝑖 + 𝜖𝜖𝑡𝑡 always supply
excessestimated shock terms without are all zero, and hence The
it isAtk
𝑡𝑡−𝑖𝑖 are a when constant term
𝑖𝑖=1which is the proxy for demand, 𝑂𝑂𝑂𝑂𝑂𝑂
ordinary least squares (OLS) steady inflation the output gap r
𝑝𝑝 𝑝𝑝 3.3.2 The random-w
model will be specified as follows: the lags of oil 𝑝𝑝
inflation
prices, always
(Gordon,when
whichthe estimated
2009).
is output
asupply This gap
proxy model
forwithoutand the will
the abe
supplyconstant by
estimated term
shock termsAtkeson-Ohanian are all zero, and h(T
0 + ∑ 𝑐𝑐𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 𝑡𝑡−𝑖𝑖 𝑡𝑡 = 𝑎𝑎0 + ∑ 𝑐𝑐𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 + ∑𝑡𝑡−𝑖𝑖 𝑑𝑑𝑖𝑖 ∆𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡−𝑖𝑖 −3.3 𝛼𝛼𝐸𝐸𝐸𝐸𝐸𝐸
shock The terms
naïve
𝑡𝑡−1 + 𝜖𝜖𝑡𝑡benchmark
(Gordon, are all
2009). zero, models
This and
model hence will beit estimated
is by
tor, 𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 is the supply shock variable. ordinary least squares
This specification always (OLS)
predicts
estimated without
𝜖𝜖𝑡𝑡 𝑖𝑖=1 𝑡𝑡 𝑖𝑖=1 𝑖𝑖=1 always estimated without a constant The term seconda bench constam
Where
𝑝𝑝 𝑝𝑝 steady inflation 3.3.1
(Gordon, Theordinary
when the output
Autoregressive
2009). least squares
This gapMoving
model
(Gordon, andwill (OLS)
2009). the
be Average
estimated
This model will inflatio be esti
and 𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡 is the 3.3 The naïve benchmark models (2001) random walk
+ ∑ 𝑑𝑑𝑖𝑖 ∆𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡−𝑖𝑖 − 𝛼𝛼𝐸𝐸𝐸𝐸𝐸𝐸Where +
𝑡𝑡−1 + ∑ 𝜖𝜖 ci and di
𝑡𝑡 𝑑𝑑𝑖𝑖 ∆𝑂𝑂𝑂𝑂𝑂𝑂𝑡𝑡−𝑖𝑖 − are the
𝛼𝛼𝐸𝐸𝐸𝐸𝐸𝐸𝑡𝑡−1
supply short
shock by
+ 𝜖𝜖(ARMA) run
ordinary dynamic
𝑡𝑡 terms are all zero, and
least squares (OLS).
henceleast it issquares (OLS) W
the𝑖𝑖=1 logarithm of 3.3 The naïve ordinarybenchmark models of the four-quarter r
𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶 is the 3.3.1 The
𝑡𝑡
coefficients 𝑖𝑖=1
of thealways model’sestimated convergence
The
3.3 firstThe
without
benchmark to Autoregressive
naive a benchmark
constant
model toterm
Moving Average 3.4 Psei
be used in the
models
e error
𝑂𝑂𝑂𝑂𝑂𝑂 𝑡𝑡 is the term. The (ARMA) 3.3.1 The Autoregressive 3.3 The naïve Moving benchmark rateAverage of models
inflation method ove
di are the short Where runciequilibrium
and di are and
dynamic the short𝜶𝜶 is(Gordon,
the
run speed 2009).
dynamic of This
study adjustment.
is model will be estimated
the Autoregressive Movingby Average
3
ds
ogarithmapproach Whereof iscitoand di are the short run dynamic 3.3.1 The The (ARMA) Autoregressive
first benchmark 3.3.1 The Moving
model Autoregressive beThe
toAverage used Atkeson-Ohania
in the Aver
Moving
the model’s coefficients convergence Once of the to ECM
the model’s model ordinary
convergence least
has been(ARMA). squares
toestimated, (OLS) This
the model combines both
coefficients (ARMA) 𝜋𝜋𝑡𝑡+m
orbyterm. ordinary The least of the model’s convergence study The is firstthebenchmark Autoregressive
(ARMA)
model Moving to be used Average inAll thefore
d 𝜶𝜶 is to the equilibrium
speed equilibrium
of adjustment. and 𝜶𝜶 is issum
and
cumulative thethe speed speed
of recursive of adjustment. ofresiduals The first
autoregressive (CUSUM) benchmark (AR) andmodel moving to averagebe used(MA) in
proach is to 3.3 The naïve the benchmark (ARMA).
study study ismodels
is the the
This Autoregressive
Autoregressive model combinesMoving
Moving Average
both
or the existence adjustment.
model has been estimated,
of Once the ECM model has been terms. It basically states
Once the ECM model has been
theCUSUM of square (CUSUMSQ) estimated, the The first
that benchmark
the time Where model
series 𝜋𝜋 to out-of-s
𝑡𝑡+4 be
is theuse
and
estimated, the cumulative sum of 3.3.1 the recursive Average tests (ARMA).
(ARMA). are This This model model combines both
combines both A
g dinary
the least
variables by The Autoregressive autoregressive Movingstudy (AR) and
Average is the moving average (MA)
Autoregressive
inflation. forecas
Moving
m of recursive residuals cumulative
residuals
(CUSUM) sum
andofto
(CUSUM) the recursive
CUSUM residuals
of square (CUSUM) behavior
autoregressive of a variable𝑦𝑦
(AR) and 𝑡𝑡 is largely
moving determined
average (MA) by
e existence of applied assess the parameter stabilityterms. (Pesaran
autoregressive It basically (AR) states andthat moving average (MA) o
oint significance
(CUSUMSQ) of tests are applied to (ARMA)
assess the terms.
its own Itvalue
basically in the states
(ARMA).
preceding that the This
periodtimethe series
andmodel time
the
series selectio
combine
M of square (CUSUMSQ) and the CUSUM of square (CUSUMSQ) tests are
variablesparameter by stability ettests are
al.,(Pesaran
1997). et al.,The 1997). first benchmark behavior behavior of a variable
terms.
model ofIttoabasically
variable𝑦𝑦
be is largely
used states
𝑡𝑡inis the
determined
largely 3.4
thePseudo
that determined time series out-of-s
by
availab
f
edthelevels of the value of the immediate autoregressive
past error. (AR)
The and
general moving avera
ss parameter applied
stability to (Pesaran
assess the parameter stability (Pesaran by its own value in the preceding period and
gnificance of study is thethe its
Autoregressive behavior
own value of ainvariable𝑦𝑦
Moving the preceding
Average 𝑡𝑡 is largely
methodology.
period determined and the by s
section
3.2.6 isThe form value
of the of
ARMA the model terms.
immediate as It
presented basically
past error.
in states
The
Asterou that the tim
to be tested :Triangle
H0al.,
et 1997). 3.2.6Model The Triangle Phillips Model Curve Phillips Curve
a
general value its
form own of thevalue
of thecombines ARMA
immediate in the
model preceding
as presented period and the
vels of The the fourth and last model under the (ARMA).theory This and Hall
model
(2007) is an behavior
ARMA (p, ofpast
both
q)
error. The
a variable𝑦𝑦
model of
general
is largely
𝑡𝑡 the
t are
deter ba
tive H1:based ci = di models = 0. The fourth
whose forecasting and lastability model under will the theoryvalue
in Asterou basedandofHall (2007) is an past ARMA (p, forecasts
All q) will bes
tested is H
ngle Model Phillips Curve 0 : 3.2.6 The Triangle Model Phillips Curve
autoregressive (AR) form and the the
ofmoving ARMA its
immediate
average model
own as presented
(MA)
value in
error.
the inThe
preceding Asterougeneral
period,
period
be tested is the models
given significance Trianglewhose modelforecasting
Phillips curve ability form; model
will be of tested the is
form
form;
of the ARMA model as presented out-of-sample in Asterou forecat
: ci =model
1last diwhich
= 0.under The
wasthefourth
developed
theory andbased
last model by Gordon under the terms. theory
(1982), It basedbasically and states Hallthat (2007) the value istime anofARMA series
the immediate (p, q) model pastoferror. the
date Th t, w
aran etwhich al., 2001). ismodels
basedwhose the
on three Triangle basicmodel
Research determinants Phillips
Bulletin
behavior Volume curve2 which
of aisvariable𝑦𝑦 form; andis was Hall
𝑝𝑝
largely (2007) determined is an ARMA
𝑞𝑞
by (p,
forecast
q) model made offor at dap
the parin
significanceability
forecasting will be tested forecasting
is ability will be tested 𝑡𝑡 form of𝜖𝜖 the ARMA model as presented
of inflation rate, namely; inertia, demand, 𝑦𝑦 𝑡𝑡 = 𝜇𝜇 + ∑ 𝛾𝛾 𝑦𝑦
𝑖𝑖 𝑡𝑡−1 + 𝑡𝑡 + ∑ 𝜃𝜃 𝜖𝜖
𝑖𝑖 𝑡𝑡−1 selection, etc, will bd
etnd al.,
the2001). assumption developed by Gordon its(1982),
own valuewhich in isthe based form;
preceding on
model ratePhillipsand of supply. the Triangle
inflation
curve The 𝐶𝐶𝐶𝐶𝐶𝐶
which major Research
model
was Phillips
conveys
difference Bulletincurvewith Volume
which
ARIMAthe was 2 model, where 𝑖𝑖=1 period the
and 𝑝𝑝 Hall I and denotes (2007) the isintegration.
𝑖𝑖=1 𝑞𝑞an ARMA (p, will q) mod be
𝑡𝑡−𝑖𝑖
three basic determinants available through df
eassumption
Gordon
ARDLADRL model
(1982),
are
Phillips
developed
which
curve
is by isGordon
based
that the
on (1982),latter which includes
value isofof inflation
based theon Where
rate, namely;
immediate µ is past =error.
a𝑦𝑦𝑡𝑡constant,𝜇𝜇 +form; ∑ The 𝛾𝛾 general
𝑝𝑝 𝑖𝑖 𝑦𝑦is
𝑡𝑡−1the + 𝜖𝜖𝑡𝑡coefficient
+ ∑ 𝜃𝜃 𝑞𝑞 𝑖𝑖 𝜖𝜖𝑡𝑡−1 quarter
are
ate of the lags
inflation
only the demand𝑡𝑡−𝑖𝑖 of the
𝐶𝐶𝐶𝐶𝐶𝐶 logarithm
conveys
shock, whereas the triangle of ARIMA For the
model,
Where Swazilandμwhereis a model,
the
constant, I 𝛾𝛾 denotes
is 𝑦𝑦
the refers
coefficient to
integration. for the the CPI
will be recurw
𝑡𝑡
+section
𝑖𝑖
DL themodel second are onethree
is basic inertia,
determinants demand, and form
of inflation supply.
rate, TheARMA
of namely;
the major
for the difference
model lagged 𝑦𝑦𝑡𝑡 =autoregressive
as presented 𝜇𝜇 𝑖𝑖=1
+∑ in 𝛾𝛾Asterou
𝑖𝑖 𝑦𝑦𝑡𝑡−1 + term, 𝜖𝜖𝑡𝑡 𝑖𝑖=1 ∑ 𝜃𝜃𝑖𝑖 𝜖𝜖𝑡𝑡−1
erminants of inflation rate, namely;
model also includes supply shock variables. lagged autoregressive term, and 𝜃𝜃𝑝𝑝𝑖𝑖moving 3.5𝑞𝑞 For
he the logarithm
lags of of
the potential
logarithm output,
of For inflation.
the Swaziland
andis is the model,
coefficient 𝑦𝑦 𝑖𝑖=1refers
of theoflagged to t is
the arethe
𝑖𝑖=1 CPIbased on all aq
that
econd the Following
one variables
is Gordon
inertia, with
demand, (1997) the
and ADRL
the
supply. Phillips
specification
The and
major Hall
curve of (2007)
difference is that an
the
Where ARMA
latter μ is (p, a q)
constant, model 𝑡𝑡 𝛾𝛾 is the the coefficient for the
d, and supply. The major difference average term.
coefficient of theWhenever lagged moving 𝑦𝑦
the 𝑡𝑡 =𝑖𝑖 𝜇𝜇
data + ∑
average 𝛾𝛾
forperiod,𝑦𝑦
𝑖𝑖 term.
𝑡𝑡−1 +
is which 𝜖𝜖 + ∑
𝑡𝑡 Pseudo 𝜃𝜃𝑖𝑖 𝜖𝜖𝑡𝑡
the triangle model which will be used in this is 19
Lhe
xy
ogarithm
he forhypothesis
variables
null
Phillips
excess
curveis is
ofwith demand,
potential
the includes 𝑂𝑂𝑂𝑂𝑂𝑂
ADRL output,
the latter only
Phillips are
𝑡𝑡−𝑖𝑖 curve
the demand inflation.
isform;
that theshock, latter non-stationary
whereas lagged Where μand is aintegrated
theautoregressive constant, term, 𝛾𝛾of𝑖𝑖 is order
the
𝑖𝑖=1andcoefficient
one, 𝜃𝜃𝑖𝑖 is for the the 3
𝑖𝑖=1
paper as that
follows: 3.3.2 The Whenever random-walk the data for model 𝑦𝑦𝑡𝑡 is non-stationary proposed date and
by
t, which capture
is 2013
the model lagged becomes an ARIMA model,
autoregressive term, where and 𝜃𝜃𝑖𝑖 term.is the P
ll hypothesis
prices, thewhich includes
ofis whereas only
a proxy the demand shock, whereas the coefficient of the lagged moving average
for
when excess value demand, 𝑂𝑂𝑂𝑂𝑂𝑂the
triangle 𝑡𝑡−𝑖𝑖for arethe
model also includes supply shock
𝑝𝑝 𝑞𝑞
Where μmodel is the a constant, 𝛾𝛾𝑖𝑖 is
the demand shock, integrated
the I denotes of order integration. one, theFor becomes
Swaziland for anthe coefficie
parameter instabil
estim
3.3.2 Atkeson-Ohanian
The
𝑦𝑦𝑡𝑡 shock
= 𝜇𝜇model, random-walk
+ (1997)
∑ 𝛾𝛾 coefficient
𝑖𝑖 𝑦𝑦𝑡𝑡−1
(2001)
+ 𝜖𝜖𝑡𝑡 the + model
∑of the
𝜃𝜃𝑖𝑖 CPI
𝜖𝜖𝑡𝑡−1 proposed
lagged moving by average term.
per the critical
riable.
ces,
l also
value ofbounds
which This
includes
triangle variables.
isspecification
asupply
model also Following
proxyshock forpredicts the
includes supply Gordon Whenever refersthe to data
the
lagged forinflation.𝑦𝑦𝑡𝑡 is non-stationary
autoregressive term, andand 𝜃𝜃c
𝑖𝑖=1 𝑖𝑖=1 will be 2014:Q1 t
itical bounds variables.specification Following ofGordon The
Atkeson-Ohanian
(1997) second
theCentral Bank
benchWhenever
integrated of Swaziland
(2001) mark of the order ©data
is 2018
the
one, forthe 𝑦𝑦𝑡𝑡 is
Atkeson-Ohanian
model non-stationary
becomes anand i
F-statisticGordon
llowing is lower (1997) the the triangle model a which
P3.3.2 g e The will be
random-walk coefficient model proposed of the lagged moving
by ahead. avera
ble. when This the output
specification gap predicts and the quarters
istic is lower specification of the triangle model Wherewhich μ will
is a be constant, integrated
𝛾𝛾 is the of
coefficient order one,
for the the model becomes an
e. Otherwise, the
the triangle model which used will in bethis paper is asThe (2001)
follows: second Atkeson-Ohanian
random
bench walk
𝑖𝑖
mark
Central Bank of Swaziland © 2018
(2001)
model,
is the
Whenever in which
Atkeson-Ohanian
the data the for forecast
𝑦𝑦 𝑡𝑡 is non-statio
rms are
hen the the all output zero, in gap and andhence is asthe it is lagged autoregressive The second bench mark is the Atkeson-
herwise,
per
used this paper follows: P a g e term,
Central Bank of
and integrated
Swaziland
𝜃𝜃𝑖𝑖 is ©ofthe order 3.5
2018 one,Forecast the model Evaluabec
e. is as follows: Research𝑝𝑝Bulletin(2001) Volume of the 2 random four-quarter
Ohanian
𝑝𝑝 walk (2001) rate
model,
random
P a g emoving average term. of in inflation
walkwhich model, theis inthe
forecast
which average
Pseudo out-of-sam
sedarewithout all zero, aandconstant hence𝑝𝑝 it is term coefficient 𝑝𝑝
ofthe the forecast
lagged of the four-quarter rate of
ed rate Where
of inflation the𝑝𝑝 lagged𝐶𝐶𝐶𝐶𝐶𝐶∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶
rate
conveys 𝑡𝑡of= inflation
∑ 𝑏𝑏𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶
ARIMAof𝑐𝑐the
Whenever rate model,
𝑡𝑡−𝑖𝑖 +of∑
four-quarter 𝑐𝑐𝑖𝑖 𝑙𝑙𝑙𝑙𝑙𝑙𝑂𝑂𝑂𝑂𝑂𝑂
inflation
where
inflation foristhe 𝑦𝑦theover
rate ofthe
𝑡𝑡−𝑖𝑖I non-stationary
denotes
𝑡𝑡 is average
inflation
P a g rate
previous
Central Bank of Swaziland © 2018
integration. of andisinflationfouraverage
the quarters.
over
captures model spe
ng-run relationship ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡 =𝑡𝑡−𝑖𝑖∑ 𝑏𝑏𝑖𝑖 ∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝑡−𝑖𝑖 + ∑ 𝑙𝑙𝑙𝑙𝑙𝑙𝑂𝑂𝑂𝑂𝑂𝑂the data e
This
without
∑relationship
𝑏𝑏
are
model
∆𝑙𝑙𝑙𝑙𝑙𝑙𝐶𝐶𝐶𝐶𝐶𝐶
𝑖𝑖 the conveyslags
a
+
𝑡𝑡−𝑖𝑖 of
will
∑ the𝑐𝑐
inertia, be
constant estimated
𝑙𝑙𝑙𝑙𝑙𝑙𝑂𝑂𝑂𝑂𝑂𝑂
𝑖𝑖 logarithm
term
are 𝑖𝑖=1 by
of the lags For
𝑖𝑖
of
the the
𝑡𝑡−𝑖𝑖
𝑖𝑖=1
the previous four quarters. The Atkeson-
1ic parameters logarithm can of
𝑖𝑖=1 output
𝑖𝑖=1 𝑡𝑡−𝑖𝑖
minus +
the
integrated
rate
𝑖𝑖=1
∆𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙
logarithm The +ofSwaziland
𝜖𝜖
of order model,
Atkeson-Ohanian
inflationOhanian
one,
over
model
thethe 𝑦𝑦model
thus
refers
𝑡𝑡 modelprevious
is
becomes to
presented thus the four anisCPI
as:presented instability,as:and e
quarters.
ameters
is model
quares can
(OLS) willofbe estimated + ∆𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 by + 𝜖𝜖𝑡𝑡 inflation. 𝑡𝑡 𝑡𝑡
the logarithm
ofcorrection
potential potential
output, output, which is𝑡𝑡 the proxy for
n+ error
∆𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 𝑡𝑡 + 𝜖𝜖𝑡𝑡 The Atkeson-Ohanian
Central Bank of Swaziland © 2018 𝜋𝜋𝑡𝑡+4 model = 𝜋𝜋𝑡𝑡thus + 𝜀𝜀𝑡𝑡+4 is presented as:
r correction
oxy
res for
(OLS) excess
excess demand,
demand, 𝑂𝑂𝑂𝑂𝑂𝑂 𝑡𝑡−𝑖𝑖 are
are the lags
P a g ofe oil
nng-run
enchmark
estimates. estimates.
prices, models
which is a proxy for the supply 3.3.2shock The random-walk
Where 𝜋𝜋𝑡𝑡+4 =
model
is the 𝜋𝜋𝑡𝑡 proposed
+
four 𝜀𝜀𝑡𝑡+4 quarter ahead rate
by
prices, which is a proxy for the
variable. This specification predicts Where
steady
Atkeson-Ohanian 𝜋𝜋𝑡𝑡+4(2001)
108 | of inflation.
is108the | four quarter ahead rate of
chmark
ariable.
regressive Moving AverageThismodels specification predicts
108 |
The Where inflation.
second 𝑡𝑡+4 ismark
𝜋𝜋bench the isfour quarter ahead rate of
the Atkeson-Ohanian
n when the output gap and the
ressive Moving Average (2001) random walk model, in which the forecast
markwithout
96
erms are all zero, Central
model ato constant
and hence Bank Of Swaziland
be usedterm
it is
in the
© 2018
inflation.
of the3.4 Pseudo rate
four-quarter out-of-sample of inflation is forecast the average
ated
3.2 pecification
The random-walk predicts
demand, 𝑂𝑂𝑂𝑂𝑂𝑂 𝑡𝑡−𝑖𝑖 are model proposed by
3.3.2second
The bench mark model
The random-walk is the Atkeson-Ohanian
proposed by
output
hkeson-Ohanian gap
is a proxy forC(2001)and the
the (2001) random walk
E N T R A L B A N K O F S WA Z I L A N D
Atkeson-Ohanian
|
model, in which the forecast
RESEARCH BULLETIN VOLUME 2
(2001)
zero, and hence
epecification
second bench
predicts itmark
is is the Atkeson-Ohanian
of thesecond
The four-quarter
bench mark rate ofisinflation is the average
the Atkeson-Ohanian
001)
output a constant
random
gap3.4 walk
and termthemodel, in whichforecast the forecast
Pseudo rate of random
out-of-sample
(2001) inflationwalk overmodel,
theunderlying
previous
in which four
this testquarters.
the is that if a single forecast
forecast
will
the be estimated
four-quarter
zero, and hence it is methodology.
rateby of inflation is the average contains all information contained in the
All forecasts willof Thebe theAtkeson-Ohanian
computed
four-quarter themodel
using rate ofother thus isispresented
individual
inflation theforecasts.
average as:The forecasts
e aof inflation
constant over
pseudo termthe previous four quarters.
out-of-sample forecast methodology, will be combined in pairs to produce three
that is, for a forecast rate of madeinflation 𝜋𝜋
at dateover = 𝜋𝜋 +
, alltheforecast
𝑡𝑡+4 𝜀𝜀
𝑡𝑡previous four quarters.
𝑡𝑡+4combinations. The forecasts from
e Atkeson-Ohanian
will be estimated estimation, bymodel
lag length thusselection,
is presentedetc, will as:
be the VECM will be combined with those of
models The Atkeson-Ohanian
available throughmodel thus
BVAR,isand presented
performed using only
𝜋𝜋 = 𝜋𝜋 + Where
𝜀𝜀
data
𝜋𝜋𝑡𝑡+4 is the fourthe quarter ahead rate as:
the benchmarks
of combined
date 𝑡𝑡+4 . The forecasts
𝑡𝑡 𝑡𝑡+4in this section will be together.
oving Average recursive, so thatinflation. forecasts at date𝜋𝜋𝑡𝑡+4 = 𝜋𝜋𝑡𝑡 + 𝜀𝜀𝑡𝑡+4
are
all available data from the initial 3.8 Data
here 𝜋𝜋𝑡𝑡+4 based
models is theonfour quarter ahead rate of
period, which is 1990:Q1 in this study through All the data to be used in this study are
Where
3.4 Pseudo is the four quarter
𝜋𝜋𝑡𝑡+4out-of-sample forecastahead rate of
lation. date t, which is 2013:Q4. This period will be secondary data which will be sourced
to beAverage
oving used in the
used for parameter estimation. The forecast from the Central Bank of Swaziland (CBS)
inflation.
methodology.
ve Moving Average horizon will be 2014:Q1 to 2015:Q4, which is and the Central Statistical Office (CSO).
4 Pseudo out-of-sample eight quarters ahead. forecast Data transformation will be carried out
3.4 Pseudo out-of-sample forecast
tocombines
ethodology. be used inboth the
3.5 Forecast Evaluation
where necessary. Since the study will use
quarterly data, annual data like GDP will be
All forecasts will be computed using the pseudo
methodology.
oving
ve Moving average Pseudo
Average(MA) out-of-sample forecast evaluation disaggregated to quarterly while monthly
captures model out-of-sample specification uncertainty, data like inflation
forecast methodology, that is, willfor bea aggregated to
hatcombinesthe time series
modelboth instability, and estimation uncertainty, quarterly.
forecasts will be computed using the pseudo
in addition to the forecast
All usual made
forecasts at date
uncertainty
will be of t, all estimation,
computed using the lagpseudo
length
rgely determined
oving average future (MA) by (Stock
events and Watson, 2008).
t-of-sample forecast methodology, that is, for a 4.0 FINDINGS AND DISCUSSIONS
Therefore the performance selection, etc, will
evaluation be
out-of-sample forecast methodology, ofperformed using that only data
is, for a
ding
hat period
the time and seriesthe
ecast made at date t, all estimation, lag length
the competing models is to determine which 4.1 Unit Root Tests
of general
them are more available
forecast
precise madeandthrough at date
reliable date t.According
for t, all The forecasts
estimation, lag in this
length
trgelyerror. The
determined by to Granger (1969), stationarity
ection, etc,forecastingwill be performed headline inflation usingover only thedata
eight tests are the pre-tests for avoiding spurious
section
selection, will be
etc, recursive,
will so that forecasts
beofperformed using at date
presented
ding period in and
Asterou
quarter forecast
the horizon. The quality regressions. Theyonly are datathe starting point
ailable through date t.forecasts
the obtained The forecasts
will be in this
tested using
t are based
available on all available
through
in any
date t.regression
The
cointegration
dataforecasts
from theininitial analysis
this
as well as
(p,
t error. q) model four of the
classical statistical loss functions: Mean
ction willThe general so that forecasts at date
beAbsolute
recursive, analysis. A series is said to be
Error (MAE), period, will
section
Mean which
Absolute is 1990:Q1
Percent
be recursive, sointhat thisforecasts
nonstationary study through
if it contains
at date a unit root, hence
presented in Asterou
Error (MAPE),
re based on all available data from the initialthe Root Mean Squared Error testing for stationarity is simple testing for
(RMSE), and the Thiel’s date t, which
Coefficient. is
t are based on all available 2013:Q4. This
the data period
presence from will
of the be
a unit used
root in a series. In
initial
(p, q) model of the
riod, which is 1990:Q1 in this study through
𝑞𝑞 nonstationary series, the order of integration
3.6 Forecast Combination for parameter estimation.
period, which is 1990:Q1is determined The forecast
in this study by the horizon
through
number of times it has
𝜖𝜖te𝑡𝑡 +t, ∑
which 𝜃𝜃𝑖𝑖 𝜖𝜖𝑡𝑡−1
is 2013:Q4.
Bates and Granger This(1969)period will bethe
introduced usedidea to be differenced to attain stationarity. One
𝑖𝑖=1 that a combination will
date t,bewhich
of forecasts 2014:Q1
outperforms to 2015:Q4,
is 2013:Q4. Thisofperiod
way
which
testing will
is
be eight
used or otherwise
for stationarity
parameter
𝑞𝑞 anyestimation.
individual forecast, The forecast as different horizon
models is through graphical inspection of the series
quarters
for parameter ahead.estimation. The forecast horizon
he coefficient havefor their
the own merits. Acknowledging the to be used in the estimations. The graphical
𝜖𝜖ll𝑡𝑡 +be∑2014:Q1𝜃𝜃𝑖𝑖 𝜖𝜖𝑡𝑡−1
existence to of2015:Q4,
different approaches which is eight
to combine analysis serves as a benchmark for the formal
𝑖𝑖=1 forecasts, in this will
3.5 Forecast Evaluation measure ofwhich
study be
the 2014:Q1
Combination to
Test 2015:Q4, is eight
m, arters andahead. 𝜃𝜃𝑖𝑖 is the unit root. The figure below shows
of Chong and Hendry (1986) and refined by the graphs of the variables to be used in this
quarters
Pseudo ahead.
out-of-sample forecast evaluation
oving average
he coefficient for the term.
Timmermann (2006) will be used. The idea paper.
5 Forecast Evaluation
m,non-stationary
and 𝜃𝜃𝑖𝑖 is and the captures
3.5 Forecast model specification uncertainty, model
Evaluation
eudo out-of-sample forecast evaluation
oving model becomes
average term. an instability,out-of-sample
Pseudo and estimation forecast uncertainty,
evaluation in
ptures model specification uncertainty, model
8
non-stationary and captures model specification uncertainty, model 109 |
stability, and estimation uncertainty, in
model becomes an instability, and estimation uncertainty, in
109 |
Central Bank Of Swaziland © 2018
97
8 109 |
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
Swaziland CPI affects the direction of local the same variables which were used in the
CPI after 2 lags, whereas RSA CPI affects it VECM. Furthermore since cointegration
after 1 lag in the short run. Furthermore the was found to exist among the variables, we
speed of adjustment towards equilibrium is proceed to estimate the BVAR, by applying
30 per cent as shown by the coefficient of the the Litterman/Minnesota prior as explained
ECM. That means about 30 per cent of any in the methodology that they are the most
disturbance from equilibrium is corrected common in literature. A key issue in the
every quarter. estimation of Bayesian VARs is the choice of
hyper-parameters which determine overall
Table 2: ECM Model Results tightness, the tightness on the prior mean of
Variable Coefficient Std. Error t-Statistic zero on cross lags in each equation, and the
D(LSDCPI(-1)) 0.038855 0.10020 0.38776 decay parameter. A complete search over all
possible hyper-parameters is not justifiable
D(LSDCPI(-2)) 0.281508 0.09856 2.85615
as it merely transfers the problem of over-
D(LSACPI(-1)) 0.57484 0.14720 3.90512
parameterization to one of two many hyper
D(LSACPI(-2)) -0.137704 0.15622 -0.88147 parameters to estimate. For the purpose
Variable Coefficient Std. Error t-Statistic of this study, the decay parameter was
D(LPSCR(-1)) 0.004702 0.01920 -0.40073 therefore set equal cross-variable weight
D(LPSCR(-2)) -0.047151 0.01941 -2.42861
to 0.99; which are commensurate with the
Minnesota prior. The results are in the table
D(LM2(-1)) -0.010839 0.02705 -0.40073
below.
D(LM2(-2)) 0.041884 0.02671 1.56791
ECM (-1) -0.303950 0.00616 -4.93338 Table 4: BVAR Model results
Source: Model outcomes Variable Coefficient Std. t-Statistic
Error
The results in the table below shows that D(LSDCPI(-1)) 0.619348 0.05644 10.9744
the VAR characterizes the data generating
D(LSDCPI(-2)) 0.100586 0.04251 2.36605
process fairly well, as the model passes the
D(LSACPI(-1)) 0.319830 0.07675 4.16703
tests of autocorrelation, normality, functional
D(LSACPI(-2)) -0.034866 0.06319 -0.55172
specification and heteroskedasticity,
D(LPSCR(-1)) 0.022807 0.01082 2.10754
whereby the probabilities are grater than
0.05, which is the rule of thumb. Furthermore D(LPSCR(-2)) -0.000331 0.00881 -0.03754
the model is correctly specified as reported D(LM2(-1)) -0.007342 0.01352 -0.54297
by the F-test values, hence it can be used D(LM2(-2)) 0.000390 0.01218 0.03204
for forecasting.
Just like the VECM, the estimated model
Table 3: Results of diagnostic tests passes all the diagnostic tests. The analysis
Significance X2 statistic Probability of the results are almost similar to those of
Breusch–Godfrey serial 17.34 0.36 the VECM. From the results, save for money
correlation LM test supply, all the variables are significant in
White 578 0.12 explaining the movements in the domestic
Heteroskedasticity test CPI, with the exception of lag 2 for RSA CPI
Jarque–Bera test 5.55 0.70 and credit extention.
Ramsey RESET test (log 0.14 0.72
likelihood ratio) 4.4 The Autoregressive Distributed Lag
(ARDL) Phillips Curve
4.3 The Bayesian VAR The ARDL Phillips curve model can be
The BVAR used in this paper arise from estimated by using the ARDL bounds test
2
generally t
The lags two and three were found not 92 94 96 98 00 02 04 06 08 10 12
the actuals
to be statistically significant in the AO ACTUAL
PHILIPS
VECM
TRIANGLE
BVAR
ARIMA
period, ho
model, however we proceeded to use it for AOM
Table 13: In-sample Evaluation statistics is the best model in forecasting inflation
Forecast RMSE MAE MAPE Theil eight quarters ahead.
VECM 1.5932 1.217162 16.45786 0.093329
BVAR 1.7069 1.288774 16.55504 0.100007 Table 14: Out-sample Evaluation statistics
PHILIPS 1.3424 0.996753 13.06176 0.078916 Forecast RMSE MAE MAPE Theil
TRIANGLE 1.2337 0.914317 12.36371 0.072342 VECM 1.392668 1.307597 25.42017 0.123055
ARIMA 1.4492 1.081336 14.81903 0.084847 BVAR 0.809370 0.680443 12.64060 0.074615
AOM 1.3543 1.009655 13.31094 0.079596 PHILIPS 1.391884 1.123842 20.38211 0.138399
Mean 1.3318 0.995632 12.99070 0.078217 TRIANGLE 1.525228 1.087103 19.32947 0.153205
square
error ARIMA 0.931882 0.815626 15.12816 0.087513
9
variable from which
benchmarks to choose.
in the eight quartersThese models
ahead forecasts,
and forecasts may differ in the underlying
albeit with only the RMSE measure. These results
8
6
assumptions, or may employ different
5 show that theTraditionally
information. BVAR, when correctly specified, can
the forecasting
4
103
model. Based on majority, we conclude that the “best” out of the individual forecasts available. To
Central Bank Of Swaziland © 2018
AO model is the best model in forecasting test whether an average, or combination, of the
C E N T R A L B A N K O FResearch
S W A Z I LBulletin
A N D Volume
| RE2S E A R C H B U L L E T I N V O L U M E 2
curve with those from the triangle model Phillips lowest values of all the measures, followed by the
curve, and the benchmarks combined together. combination of the VECM and BVAR.
be just as good as a combination of all of the Table 15: Forecast Combination Evaluation
The figure below shows the comparison of the
forecasts. Table
Statistics 15: Forecast Combination Evaluation
forecasts combinations with the actual data in the Forecast
Statistics RMSE MAE MAPE Theil
combined in pairs to produce three forecast TRIANGLEPHILIPS 1.445551 1.106913 19.897 0.144445
that the forecasts
combinations. Thewere tracking
forecasts fromthe
theactuals,
VECM Forecast
AOARIMA RMSE
0.716811 MAE
0.621490 MAPE0.065879
11.887 Theil
were combined with those of the BVAR, those
except for the graph of the triangle and Phillips
Mean square error NA NA NA NA
found to be the best, with lowest values Bank of Swaziland should consider using the
of all the measures. Four quarters ahead BVAR, over and above the naïve benchmarks
forecasts evaluation selected the AO naïve in forecasting inflation. This conclusion is
benchmark as the best model. Results for similar to that of Doan, Litterman, and Sims
the eight quarter ahead forecasts were (1984), who found that BVAR performs better
mixed, with the RMSE selecting the BVAR than the unrestricted vector autoregression
while all the other measures selected the AO models.
naïve benchmark. Generally the best model
in this category is the AO naïve benchmark,
followed by the BVAR. REFERENCES
Forecasts combinations were also performed. Akdogan, K., Beser, S., Chadwick, M.G.,
The forecasts from the VECM were combined Ertug, D., Hulagu, T., Kosem, S.,
with those of the BVAR, and the benchmarks Ogunc, F., Ozmen, M.U. and Tekalti,
combined together. Results showed that the N. (2012). Short-term Inflation
best combination in tracking the actuals four Forecasting Models for Turkey and a
quarters ahead were the benchmarks. The Forecast Combination Analysis, Turkiye
results further show that the combination Cumhuriyet Merkez Bankasi Working
of the VECM and BVAR performs better than Paper No 12/09.
each individual model and the AO performs Alturki, F. and Vtyurina, S. (2010) Inflation
better than its combination with the ARIMA in Tajikistan: Forecasting Analysis
to forecast inflation four quarters ahead. and Monetary Policy Challenges,
IMF Working Paper, WP/10/17.
Secondly, the model combinations forecasting
ability was evaluated for eight quarters Aron, J. and Muellbauer, J. (2008). Some
ahead. The benchmark combination was Issues in Modeling and Forecasting
the best model in forecasting inflation eight Inflation in South Africa. CSAE
quarters ahead, followed by the combination Working Paper WPS/2009-01,
of the VECM and BVAR. Furthermore, the Department of Economics, Oxford
BVAR was found to perform better than University.
its combination with the VECM to forecast
inflation eight quarters ahead, while the Asteriou, D. and Hall, S.G. (2007). Applied
AO model was found to perform better Econometrics. Revised edition.
than its combination with the ARIMA only A Modern Approach, Macmillan,
with the MAE and MAPE statistics. These Palgrave.
results therefore suggest that the VECM and
BVAR models are not better than the naïve Atkeson, A. and Ohanian, L.E. (2001). Are
models in forecasting inflation in Swaziland, Phillips Curves Useful for Forecasting
however with the BVAR correctly specified, Inflation? Federal Reserve Bank
it can be a suitable model for forecasting of Minneapolis, Quarterly Review
inflation in Swaziland. 25(1):2-11.
These results contradicts those by Akdogan Bates, J.M. and Granger, C.W.J. (1969). The
et al. (2012) which suggested that the Combination of Forecasts, Operations
models which incorporate more economic Research Quarterly, 20, 451-68.
information outperformed the random walk
model at least up to two quarters ahead. This Biswas, D., Singh S. and Sinha A. (2010),
paper therefore suggests that the Central
Kenny, G., Meylar, A. and Quinn, T. (1998). Pesaran, M. H., Shin, Y. and Smith, R. J.
Bayesian VAR Models for Forecasting (1997), Testing for the Existence of
Irish inflation, Working Paper, a Longrun Relationship. Cambridge
Technical Paper 4/RT/98 Ireland: Working Papers in Economics 9622,
Central Bank of Ireland. Faculty of Economics, University of
Cambridge.
Koop, G. and Korobilis, D.(2010),
Forecasting Inflation Using Dynamic Pesaran, M.H., Shin, Y. and Smith, R.J.
Model Averaging, University of (2001). Bounds testing approaches
Strathclyde. to the analysis of level relationship.
Journal of Applied Economics 16: 289-
Langa, T. (2001). Monetary Policy in 326.
Swaziland, Paper presented at a
conference on Monetary Policy Phillips, P.C.B. and Perron, P. (1988).
Frameworks in Africa. Pretoria SA. Testing for a Unit Root in Time Series
Regression, Biometrika, 75, 335–346.
Litterman, R.B. (1986) Forecasting with
Bayesian Vector Autoregressions - Sims, C. (1980). Macroeconomics and
Five Years of Experience, Journal of Reality, Econometrica, 48, 1-48.
Business and Economic Statistics,
January 1986, Vol. 4, No. 1, pp. 25-38. Stock, J.H. and Watson, M. (2001). A
Comparison of Linear and Nonlinear
Norman, D. and Richards, A. (2010), Univariate Models for Forecasting
Modelling Inflation in Australia, RBA Macroeconomic Time Series, in
Research Discussion Paper No 2010-03. R.F. Engle and H. White (eds.),
Cointegration, Causality, and
Nxumalo, W. (2014). Understanding Forecasting: a Festschrift in Honour
the Conduct of Monetary Policy in of Clive Granger, Oxford: Oxford
Swaziland: An Internal Perspective, University Press.
Central Bank Of Swaziland Draft
Working Paper. (Unpublished). Stock, J.H. and Watson, M.W. (2002). Has
the Business Cycle Changed and Why?
NBER Macroeconomics Annual 2002,
159–218.
112
cond
and important
(pg. sub-components
83). The GECM
aspect to
sub-components ascertain
understanding sources
contribution
contribution totothe inflation
inflation
to the overall
overall CPI
3). sub-components
The GECM for each
contribution country.the
Central Bank Of Swaziland © 2018
To study
overall these
CPI effects, it is first
GECM
CM
by: for for each
each country. country.
To To study
study these these
effects, effects,
127 || it is first
ECM forfor each
eachcountry.
country.ToTostudy
studythese
importantthese effects,
toeffects,it itisitisfirst
ascertain 127 is first
sources of inflation
first
ntial between Swaziland
important to and South
ascertain Africa
sources of inflation
model
is partly because the comparability of the data
with a similar base year of December 2012).CPI. The
Notes: 𝑦𝑦 and 𝑥𝑥 represent logg
*, ** and *** represents sta
𝑡𝑡 𝑡𝑡
between the is
prices
the estimated
partly because
ofthe(such
the
coefficients astwo
are
the use
comparability
much countries
bigger.classification
of similar
of the
That
data k 1 =− γ
COICOP
τγ−γSpecification (1) runs the estimated model for the
2
is partly because the comparability of the data Notes: Notes: 𝑦𝑦 and 𝑥𝑥 represent
period logged
January Swaziland
1993CPI and
–CPI logged South2016.
December African The
thereby supporting the relative
is partly because the
and comparability
the review PPP toofhold.
of weights the
and data
rebasing of Notes:
the
CPI. *, **
Notes:
yand
𝑦𝑦CPI
*, **𝑦𝑦CPI.
and
and
and***
𝑡𝑡
xrepresent
𝑡𝑡 t represent
𝑥𝑥 represents
𝑥𝑥 represents
loggedlogged
𝑡𝑡
𝑡𝑡
represent logged
t
statistical
Swaziland Swaziland
significance
Swaziland CPI and
and logged
at 10%,5%
logged
CPI
and 1%
South
and African
South logged South
African
has improved with newer methodologies adopted African CPI.
respectively.and ****,
CPI. *, ** and ***model
𝑡𝑡
** and statistical
***
𝑡𝑡 significance
represents
is statistical
estimated with an
at 10%,5%
statistical
additional
and 1%
significance
and 1%control
at
has improved
The long-run with
multiplier
has improved newer methodologies
is a derived
with
with newer similar base year
methodologies adopted
toofadopted
be 10%,5%
December respectively.
2012). The
Specification
represents significance
(1) runs the estimated model for the
at 10%,5%
and 1% respectively.
respectively.
(such as the use of similar classification COICOP example Specification (1)
variable runs
related theto estimated
inflation targeting model for for
using the a
1.14. (such
This as means
the asuse
(such that
the ofuse any
similar
of permanent
contemporaneous
classification
similar classification
effect one
for
COICOPCOICOP
Specification
after the
period January dummy
(1) runs
1993
variable
the estimated
–thatDecember
takes value
model
2016.
of zero
the
The
between
and the review of weights
percentage
and theand point
review increase
of
the review weights and
of weights inand
rebasing and
Southrebasing
rebasing
and
reweighting
rebasing
of the
Africa’s
of theof the
of CPI
the CPI period
in
model
2013 January
period
CPICPI Specification January (1)
is estimated
1993-2000 and
1993 –– December
1993 runs
with onean
December2016.
the
for additionalestimated
2001-2016
2016.
control
which
TheThe
makes model
with a similar base year of December 2012). The
shows that the contemporaneous effect became
priceswith
would
a with
lead
similar
to
a similar
a 1.14
base as
base year year
percentage
of December
of December
point
2012). The formodel
model
the is
is estimated
period estimated with
with an an additional
additional
it comparable to the other specifications.2016.
January 1993 – December control
control
0.672012). The variable
with a related to inflation targeting using a
permanentcontemporaneous
increase in effecthigh
the forasexample
consumer
percentage
after
price thepoints variable
per The model is that estimated value with
variable related
related
Specification to inflation
to takes and (3)targeting
inflation
(2) ran onan
istargeting using
shorter additional
using a a
time
contemporaneous effect
contemporaneous for for
convergence example
rate of after
aboutafter thethe
22.3 dummy
cent. Thus, variable of zero between
level for rebasing and effect
Swaziland. reweighting example
of the CPI in 2013
rebasing and reweighting any inflation of differential
the CPI in
control
2013 atdummy
is corrected a faster variable
dummy
1993-2000
frame
variable
variableand one
to
that
that related
capture
takes
fortakes
most
value
2001-2016
to inflation
value
recent
of zero
of
which
dynamics.
between
zeromakes targeting
between
All
rebasingshows andthatreweighting
the contemporaneousof the CPIeffectin became
2013
pace and given the high pass-throughusing a dummy and onevariable thatwhich takes value of
specifications have no serial correlation.
1993-2000
effect, the for 2001-2016 makes
shows that the contemporaneous effect became 1993-2000 it comparable and one for 2001-2016
to specification
the other which
specifications. makes
Specification
showsasthat (2)
high and
theascontemporaneous(3) which
0.67 long-run
percentage effect
points shows
became
with
multiplier is slightly lower a zero at
it 1.11
However,
between 1993-2000
comparable to the other and
(2) and (3) reflects
one for
specifications.
that
2001-
as high as 0.67 percentage points with a itSpecification comparable the(2) and
estimatedto (3) theis ran
coefficientsother onareshorter time
specifications.
not stable. Hence
recentas trends support
convergence rate the
of percentage
about
high as 0.67 percentage points with a notion points.
22.3 per that
cent. in
Thus, 2016 which
Specification makes
(2) and (1) it comparable
(3)is is rantoon shorter to timeandother
the
convergence rate of about 22.3 per cent. Thus, Specification frame to capture specification most taken
recent be more
dynamics. reliable
the post 2000’s the
any inflation
convergence
estimated
rate differential
of about 22.3
coefficients
is corrected
per cent. at aThus,faster specifications.
frame to inference
(2)
capture
and (3)
Specification
most based
is largely
is
recent
ran on(2) andAll
shorter
on it.dynamics. All
(3)time is ran
are much bigger.
any That
inflation is partly
differential is because
corrected
pace and given the high pass-through effect, the on
at athefaster specifications
shorter
frame to capturetimehave have frame mostno toserial
capture
recent correlation.
dynamics. most recent All
any inflation differential is corrected at aeffect,
fasterthe specifications The estimated no serial correlation.
comparability paceof and the
given data has
the high improved
pass-through with However, specification
dynamics. All (2)results
specifications and (3) can reflects
also
have
be thatshown
no serial
long-run multiplier is slightly lower at 1.11 specifications
However, graphically specificationhaveas shown no and
(2) inserial 1. correlation.
(3) reflects
figure As noted that in
newerpace and given multiplier
methodologies
long-run the high
percentage points.
adoptedpass-through
Table
is 1: (such
Estimated
slightly effect,
lower as
results the
at the
of the GECMthe
1.11 model
correlation. estimated However,
coefficients are specification
not stable. Hence (2) and
∆yt = α0 + γ(yt−1 − xt−1 )However, specification (2)areand (3) reflects that
+the
τ1 ∆xestimated
figurecoefficients
1, inflation differential notdoes tend
stable. to deviate
Hence
use oflong-run
similar classification
multiplier
percentage points. is EstimatedCOICOP
slightly lower and at the 1.11 (3) reflects
t
specification (1) thatis takenthe to estimated
be more reliable coefficients
and
model + τ2 xt−1 specification (1) is taken to be more reliable and1.1
from the long-run boundary (of about -1.1 and
reviewpercentage
of weights the estimated coefficients are not stable. Hence
points.and rebasing Specification
of the
(1)
CPI (2)
are not isstable.
inference
(3) aslargely
per the based Hence on it. inspecification
estimation Specification 1). For(1) is
with a similar base year of December
Sample
2012).
1993M1 – 2001M01 taken
inference is largely
specification
2012M12 to be
(1)
example, moreis based
taken
a common
on to
reliable it.be more reliable and
shockand of ainferencerise in is
The estimated results can also be shown
The contemporaneous effect for2016M12 example – largely
–inference based
is largely
internationalon
The estimated results can also be shown it.
based
pricesonparticularly
it. oil and food
2016M12 graphically prices
2016M12 as shown in selected
for some figure 1. As noted
months in 2007inand
Table 1: Estimated results of the GECM model
graphically as shown in figure 1. As noted in
Table 1: EstimatedConstant
results(αof
0 ) the GECM model
-0.053*** -0.086*** The
figureestimated
-0.114** 1, inflation results
differential
2008 resulted can also to be
does tend
in higher inflation shown
deviate
differential
Estimated ∆yt = α0 + γ(yt−1 − xt−1 ) + τ1 ∆xt
figure 1, inflation differential does tend to deviate
113
Estimated ∆ytCentral
= α0 + γ(yoft−1
Bank − x ©) 2018
Swaziland + τ1 ∆xt from the long-run
graphically boundary
as shown in (of about1.-1.1
figure Asand 1.1
noted
129 |
in
Table model
1: Estimated results P a g eof the+GECMτ2 xt−1t−1model
model + τ2 xt−1 from the long-run
Central Bankboundary (of about
Of Swaziland -1.1 and 1.1
© 2018
Specification (1) (2) (3) as per1,the
figure estimation
inflation in Specification
differential does tend1).
to For
deviate
Estimated ∆yt = α(1)
Specification 0 + γ(yt−1 −(2) xt−1 ) + τ1 ∆x
(3)
t as per the estimation in Specification 1). For
Sample 1993M1 – 2001M01 2012M12 example, a common shock of a rise in
from the long-run boundary (of about -1.1 and 1.1
between Swaziland and South Africa. Towards the Source: Author’s calculations using data from CSO and STATSSA
end of 2009 the differential fell below the lower 4.2. Analysis of CPI Sub-components trends
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
bound mainly due to a collapse in commodity This sub section presents graphical analysis of the
prices which was mainly influenced by the second CPI components. Special focus goes into
roundThe effects of theresults
estimated global can
financial
also becrisis. 4.2. Analysis
shown computation of of CPI Sub-components
inflation differential between
graphically
Throughout the yearas shown in figuredifferential
2012 inflation 1. As noted trends
tradables and non-tradables. Due to data
in figure 1, inflation differential does tend This sub section presents graphical analysis
was elevated
to deviate abovefrom3 percentage
the long-run points largely (of limitations,
boundary of the CPI this analysis is only
components. done for
Special the goes
focus year
due toabout
domestic-1.1taxandreforms
1.1 asthat
per had
the inflationary
estimation in 2014 into –computation of inflation
2016 focusing on recent inflation differential
Specification 1). For example, a17common Research Bulletin Volume 2 tradables and non-tradables. Due
between
pressures on Swaziland’s consumer prices . This differential and this using
is sufficient
shock Swaziland
between of a rise and in international
South Africa. Towards prices
the to data
Source: limitations,
Author’s calculations this CSOfor
analysis
data from and this
is study
only
STATSSA done as
studyparticularly
takes more interestoil andonfood the sources
prices for of thesome the forfocus
the year 2014 – 2016
is mainly on the focusing
recentoninflation
recent
end of 2009 the differential fell below the lower 4.2. Analysis of CPI Sub-components trends
recentselected
inflation months in 2007
differential andin2008
observed 2016.resulted
The inflation differential and this is sufficient
in higher
bound mainly inflation differential
due to a collapse between differential
in commodity for this
This sub
observed
study
section
as in 2016.
the
presents
focus Theis CPI
graphical
mainlyfor goods
analysis
is
onof the
the
unpacking of theand
Swaziland sources
SouthisAfrica.
done inTowards
detail inthe theend recent inflation differential observed in
prices which was mainly influenced by the second used CPI
to proxy
components.
tradables while the CPI
Special focus
for services
goes into
of 2009
subsequent the differential fell below the lower 2016. The CPI for goods is used to proxy
section.
round
boundeffects
mainlyofduethe to aglobal
collapsefinancial crisis. iscomputation
in commodity
used to proxy non-tradables. As noted in
tradables while the CPI for
of inflation servicesbetween
differential is used
Figureprices which
1: Swaziland
Throughout was2012
and
the year mainly
South Africa influenced
inflation equation
differentialby tradables
Inflation 11, the interest is to study
to proxy non-tradables. As noted in equation the sources of
and non-tradables. Due to data
the second
Differential Dynamics round effects of the global 11, thedifferential interest iswhether to study theythe sources of
was elevated above 3 percentage points largely inflation limitations, this analysis is only done
emanate
for the
from
year
financial crisis. Throughout the year 2012 inflation differential whether they emanate
5.0
due to domestic
inflation tax reforms
differential wasthat
elevated above 3 the
had inflationary from
2014
tradable
the
– 2016
differential
tradable between
differential
focusing on recent betweenthe two the
inflation
Inflation differential (percentage points)
percentage
4.0 points largely due to domestic
pressures on Swaziland’s consumer prices . This countries
17 two countries
or fromthis or from
respective respective internal
internal differential
tax reforms that had inflationary pressures differential differentialand between is sufficient for this study as
the tradable and non-
study
3.0 takes more interest on the sources the between thesector.
tradable andon non-tradable
the recentsector.
on Swaziland’s consumer prices17. Thisofstudy tradable
the focus is mainly inflation
takes
recent more interest
2.0 inflation differentialonobserved
the sources
in 2016. ofThe
the differential observed in 2016. The CPI for goods is
Figure 2: Sources of Inflation Differential
recent inflation differential Figure 2: Sources of Inflation Differential
unpacking
1.0 of the sources is done in observed
detail in thein between
used Swaziland
to proxy tradables andwhile
Souththe Africa
CPI for services
2016. The unpacking of the sources is done between Swaziland and South Africa
subsequent
in0.0detail insection.
the subsequent section. is used 8.00 to proxy non-tradables. As noted in
inflation differential (percentage
6.00
Jan-07
Jan-09
Jan-11
May-12
Jan-13
Jan-15
May-08
May-10
May-14
May-16
Sep-07
Sep-09
Sep-11
Sep-13
Sep-15
Figure equation4.00
11, the interest is to study the sources of
Figure1:1:
Swaziland and South Africa Inflation
-1.0
Swaziland and South Africa Inflation
Differential Dynamics inflation2.00
differential whether they emanate from
Differential
-2.0 Dynamics
5.0 0.00
the tradable differential between the two
points)
Jan-14
Jan-15
Oct-15
Jan-16
Oct-14
Oct-16
Apr-14
Apr-15
Apr-16
Jul-14
Jul-15
Jul-16
-3.0 -2.00
Inflation differential (percentage points)
Author’s6.00
Jan-07
Jan-09
Jan-11
May-12
Jan-13
Jan-15
May-08
May-10
May-14
May-16
Sep-07
Sep-09
Sep-11
Sep-13
Sep-15
Source:
STATSSA calculations using data from CSO and STATSSA
-1.0 4.00
According 2.00 to figure 2 the main drivers of the
-2.0 According 0.00
to figure 2 the main drivers of
observed
the observedinflation differential
inflation between Swaziland
differential between
points)
Jan-14
Jan-15
Oct-15
Jan-16
Oct-14
Oct-16
Apr-14
Apr-15
Apr-16
Jul-14
Jul-15
Jul-16
-3.0 -2.00
17 Swaziland
and South -4.00 andinSouth
Africa 2016 was Africa in 2016
largely due towas (i)
The tax reformsCPI_diff
included an Amendment
Upper boundof the Sales Taxbound
lower in
December 2011 and the introduction of VAT in April 2012. largely due to (i) the differential between
Source: Author’s calculations using data from CSO and the differential
tradables forbetween
the twotradables
Overall inflation diff for the two
countries and (ii)
EffectsSTATSSA
of VAT on Swaziland’s prices are well documented by
Goods inflation diff
Nxumalo and Mabuza (2013) the internal differential between the
Goods-Service diff_SA
Central Bank of Swaziland © 2018 tradable and non-tradables in Swaziland.
130 |
Page Goods-Services diff_SW
The tax reforms included an Amendment of the
17 The differential between tradables of the
Sales Tax in December 2011 and the introduction of two countries reflects a violation of the
Source: Author’s calculations using data from CSO and STATSSA
VAT in April 2012. Effects of VAT on Swaziland’s prices ‘Law of One Price’ (LOP) and this paper
are well documented by Nxumalo and Mabuza (2013) would explore
According to figure some 2 theof mainthe drivers
factorsofthatthe
observed inflation differential between Swaziland
17 Centralan
The tax reforms included Bank
114
Of Swaziland
Amendment © 2018
of the Sales Tax in
December 2011 and the introduction of VAT in April 2012.
and South Africa in 2016 was largely due to (i)
the differential between tradables for the two
Effects of VAT on Swaziland’s prices are well documented by
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
might have led to this in 2016. The second Notes: SA_inflation_Weight_adj series represents the
aspect of the differential between the local SA inflation calculated using Swaziland’s CPI weights.
The difference between the bars and SA inflation
tradables and non-tradables can be inferred represent the portion of inflation differential that
to provide some information on what can be is explained by weights-differential (40 %)and the
termed the internal version of the Balassa- difference between the bars and Swaziland inflation
Samuelson effect which according to Dedu represent the portion of inflation differential
and Dumitrescu (2010) can be interpreted arising from other factors other than the weights-
differential.
as the Baumol-Bowen effect. To unpack this
effect further information on productivity
4.2.2. Supply Side Inflation Pressures
in the tradable and non-tradable sectors is
The next step is to investigate some of the
required. However, Swaziland does not have
reasons apart from weights-differential that
good productivity data to support further
explains the 60 per cent differential observed
analysis for this.
from figure 3. The first stop is looking at the
important sub-component which is “Food
4.2.1. The Role of CPI Weights
and non-Alcoholic Beverages” (FNAB). This
The different composition of Consumer Price
component accounts for 29.22 per cent in
Indices (CPI) weighting structure across
the consumption basket. Food prices are
countries is one of the often cited reasons
highly linked to food productions and shocks
why inflation differential may persist
thereof and thus represent a good proxy to
across countries. To test this in the case of
understanding supply side factors in the CPI.
Swaziland the paper uses a crude method of
recalculating South Africa’s inflation using
Figure 4: Food Inflation Trends
Swaziland’s CPI weights composition and then
compares the CPI-weight adjusted series to
the Actual for Swaziland. The results of these
calculations are shown in figure 3. As noted
from figure 3, the CPI weight differential does
explain some of the differential observed in
2016. The weights only explain an average
Research Bulletinof
Volume
about2 0.6 percentage points of the total
erential
1.5 percentage points observed. That is; the
differential partially explains 40 per cent of the
weights differential partially explains 40 per
bles in observed
cent of differential
the observedin 2016.
differential in 2016.
Source: CSO and STATSSA
ables of
Figure
Figure3:3:
Inflation
InflationDifferential
Differentialdue
duetoto
he ‘Law Differences in CPI Weights
Differences in CPI Weights
explore 9.0%
CPI year-on-year % growth
7.0%
o this in
5.0%
erential 3.0%
Jan-14
Jan-15
Jan-16
Apr-14
Oct-14
Apr-15
Oct-15
Jul-14
Apr-16
Oct-16
Jul-15
Jul-16
adables
ation on
SA_inflation_Weight adj
n of the SA_inflation_Actual
ding to SW_inflation
is effect
Notes: SA_inflation_Weight_adj series represents the SA
in the
inflation calculated using Swaziland’s CPI weights. The
equired. difference between the bars and SA inflation represent the Central Bank Of Swaziland © 2018
115
e good portion of inflation differential that is explained by weights-
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
Bread and Cereals Price Index Sugar and Sugar Products Price Index
30 35
25 30
25
y-o-y % growth
y-o-y % growth
20
20
15
15
10
10
5 5
0 0
Jul-15
Jul-16
Jul-15
Jul-16
Nov-16
Jan-15
May-15
Nov-15
Jan-16
May-16
Nov-16
Jan-17
Jan-15
May-15
Nov-15
Jan-16
May-16
Jan-17
Mar-15
Sep-15
Mar-16
Sep-16
Mar-15
Sep-15
Mar-16
Sep-16
SA_Bread n Cereal SW_bread n cereals SA_sugar prod SW_sugar prod
6 8
5
4 6
3 4
2
2
1
0 0
Jul-15
Jul-16
Jul-15
Jul-16
May-15
Nov-15
May-16
Sep-16
Nov-16
May-15
Nov-15
May-16
Nov-16
Jan-15
Mar-15
Sep-15
Jan-16
Mar-16
Jan-17
Jan-15
Mar-15
Sep-15
Jan-16
Mar-16
Sep-16
Jan-17
SA_meat SW_meat SA_eggs n dairy SW_eggs n dairy
y-o-y % growth
20.00 10
15.00
5
10.00
5.00 0
Jul-15
Jul-16
Jan-15
May-15
Nov-15
Jan-16
May-16
Nov-16
Jan-17
Mar-15
Sep-15
Mar-16
Sep-16
-
-5
Jul-15
Jul-16
May-15
Mar-16
May-16
Jan-15
Mar-15
Jan-17
Sep-15
Nov-15
Jan-16
Sep-16
Nov-16
(5.00)
-10
Source: Bank
Central CSO and
ofStatsSA
Swaziland © 2018 133 |
Page
As shown in figure 4 Swaziland and South The other notable differential is on ‘sugar
Africa’s food inflation tends to track and sugar related’ products. Like prices of
each other though there were significant ‘bread and cereals’ the trend is similar but
differentials in 2007-2008; 2012 and 2016. growth rates for Swaziland are significantly
The differential observed in 2007-2008 higher than those of South Africa. This is
and 2016 were driven by common shocks. surprising from two grounds: first Swaziland
In 2007-08 there was a significant surge in is a net exporter of sugar; secondly sugar
international oil and food prices but the price setting is harmonized between
observed food inflation was notably higher Swaziland and South Africa amid on a one-
in Swaziland than in South Africa. A similar month lag basis. In 2016, sugar prices rose
story was observed in 2016, a common shock twice; by 12.5 per cent in February and 15
from an El’nino induced drought resulted per cent in August. However, in the second
in asymmetric results in the food inflation half of the year after the second sugar price
outcomes for the two countries while the increase (in both countries) Swaziland sugar
2012 differentials were mainly informed prices are 10 percentage points higher than
by domestic factors that were unique to those of South Africa. This reflect that the
Swaziland at that time. cumulative 27.5 per cent increase went
through smoothly whilst for South Africa it
To analyse the asymmetric results in food might have not been fully implementable
inflation especially the most recent ones arguably because of an increased influx of
arising from the negative effects of the sugar imports (to the SA market) that might
drought, food inflation can be unpacked have put pressure on prices not to increase by
further to its sub-components. Box 1 shows the full 15 per cent in August 2016. This also
the comparison of major sub-components reflects differences in demand elasticities
within food inflation. It can be observed for sugar prices in the two countries.
that significant asymmetric trends or levels
of price changes come from the price indices Another worrying divergence within the food
for ‘bread and cereals’, ‘sugar and sugar components is that of prices of vegetables.
products’ and ‘vegetables’. For further Between November 2015 and March 2016
analysis, it is important to investigate the there was a significant acceleration in prices
sources of differences in the growth rates for vegetables in both Swaziland and South
of these price indices. As expected the Africa. However, for most part of the second
impact of drought on maize weighed heavily half of 2016 South Africa’s vegetable’s prices
on cereals in what could be termed the have been on a downward trend closing the
first round effects of the drought. Prices of year 2016 at below 5 percent. An opposite
‘bread and cereals’ products particularly trend was observed in the case of Swaziland
mealie meal accelerated significantly early with vegetable prices reaching 35 per cent
in the year reflecting food supply shortages year-on-year increase in the last two months
in both Swaziland and South Africa. of 2016.
However, the increase was too pronounced Additional pressures on the supply side can
in Swaziland cereal prices with a differential also be studied from administered prices.
of about 10 percentage points. One of the Administered prices broadly relates to the
developments that might have contributed prices that are regulated by some governing
to this differential is the bottlenecks in body. While administered prices vary from
supply mainly the ban on the importation of country to country in terms of coverage, the
mealie meal implemented through National common ones for both Swaziland and South
Agricultural Marketing Board (NAMBOARD). Africa are electricity, water, fuel and public
transport. Figure 5 shows the trends for the for these components tend to be higher
two countries in administered prices. with an average of 1.7 percentage points
reflecting mark-up in prices that incorporate
Though there is no clear relationship transportation and other retailing costs. The
Research Bulletin Vo
between the administered prices for these inflation differential for these components Research Bulletin V
two countries (as this reflect domestic pressures in the in
rose significantly economy particularly
the second in the
half of 2016 upwa
policies on regulated pricing), it can be pressures in the
signalling that economy
there particularly
were demand in the
pressures upw
second half of the year. mean
observed from figure 5 that administered in the economy particularly in the second
second half of the year. mea
prices in Swaziland were relatively higher half of 6:
the year. and footwear and 15 pe
Figure Clothing
throughout the year 2016. Administered furnishing and household equipment 15 p
Figure 6: Clothing and footwear and put
price increases included among others; a 25 Figure 6: Clothing and footwear and
furnishing and household equipment put
per cent increase in public transport fares furnishing and household
Clothing equipment
and Footwear index empl
and an 11.7 per cent increase in electricity. Clothing and Footwear index emp
evide
rch Bulletin Volume 2 12
evid
and
% GROWTH
Figure 5: Administered Prices 10
12
od 14 % GROWTH 8
10 and
incre
es. 12 68
10 46
incr
2016
Y-ON-Y
re 8 24 201
y-on-y % change
Y-ON-Y
or 6 02 4
4
May-16
Jan-13
May-13
Jan-14
May-14
Jan-15
May-15
Jan-16
Sep-13
Sep-14
Sep-15
Sep-16
a. 0 4
2 As h
May-16
Jan-13
May-13
Jan-14
May-14
Jan-15
May-15
Jan-16
Sep-13
Sep-14
Sep-15
Sep-16
16 0
As h
stron
Jan-14
Jan-15
Jan-16
Apr-14
Oct-14
Apr-15
Oct-15
Apr-16
Oct-16
Jul-14
Jul-15
Jul-16
-2
a -4 Clothing and Footwear SA
instro
a
5 -6 Clothing
Clothingand
andFootwear
FootwearSW
SA
in a
holds
he Administered_SA Administered_SW
Clothing and Footwear SW
hold
provi
ng Source:
Source: CSO CSO
Furnishing and Household
wo In South Africa administered price movements prov
differ
Equipment
Furnishing index
and Household
In South Africa administered price movements
were merely a reflection of petrol price diffe
mean
were merely a reflection of petrol price Equipment index
10
movements
movementsin in lineline
withwith
international oil prices
international oil mea
viabi
Y-ON-Y-%GROWTH
so 8
prices and exchange rate developments. 10
and exchange rate developments. On the 6
viab
CMA)
Y-ON-Y-%GROWTH
Jan-14Jan-14
May-14
Jan-15Jan-15
May-15
Jan-16Jan-16
May-16
Sep-14
Sep-13
Sep-15
Sep-16
feed intodo
Swaziland inflation differential
also feed in Swaziland.
into inflation differential 0
ry nom
May-13
May-14
May-15
May-16
Sep-14
Sep-13
Sep-15
Sep-16
in Swaziland. As sh
th 4.2.3. Demand Side Inflation Pressures FURNISHING, HOUSEHOLD EQUIPMENT SA
To study
4.2.3. the demand
Demand side, Pressures
Side Inflation the paper differ
As s
er, FURNISHING,and
HOUSEHOLD EQUIPMENT SA
studies inflation differential for selected FURNISHING HOUSEHOLD EQUIPMENT
SW track
diffe
he components
To that covers
study the demand side, semi-durable
the paper studiesand
FURNISHING and HOUSEHOLD EQUIPMENT
ed durable products. These are ‘clothing and Source:
Source:CSO andand
STATSSA wher
trac
inflation differential for selected components that CSO SW STATSSA
footwear’ and ‘furnishing and household Swaz
covers semi-durable and 6durable products. These The demand
Source: pressures
CSO and STATSSA mainly came from the whe
equipment’. Figure depicts the growth The demand pressures mainly came from
rates
are for the
‘clothing price
and indices
footwear’ for‘furnishing
and ‘clothing and
and fiscal
The side. According
the demand
fiscal side. to the
According
pressures medium
to the
mainly term
camemediumbudget
from the Africa
Swa
he
footwear’ and ‘furnishing and household term budget
review reviewdelivered
for According
2016/17 for 2016/17
by thedelivered
minister of 50 ba
as household equipment’. Figure 6 depicts the fiscal side. to the medium term budget Afric
equipment’ indices. Swaziland’s prices by the minister of Finance, Government
ed growth rates for the price indices for ‘clothing Finance,forGovernment
review implemented
2016/17 delivered a salaryof
by the minister arbitr
50 b
and footwear’ and ‘furnishing and household
118
at review forGovernment
Finance, the civil service in July 2016.
implemented This
a salary arbi
Figur
Central Bank Of Swaziland © 2018
equipment’ indices. Swaziland’s prices for these resultedforin the
a civil
26 percent and S
ly review service inincrease in This
July 2016. the
Figu
components tend to be higher with an average of
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
implemented a salary review for the civil Figure 7: Nominal Convergence – Swaziland
Research Bulletin V
service in July 2016. This resulted in a 26 and South Africa Nominal Policy Rates
percent increase in the government wage- 14.00 6.
bill for 2016/17 financial year and stretched 12.00 4.
the fiscal deficit for the year revised upwards 2.
10.00
from 13.7 per cent to 16.1 per cent. This 0.
meant that public sector wages rose by 8.00
-2.
more than 15 per cent in real terms. Such 6.00 -4.
adjustments also put pressure on private
4.00 -6.
sector wages as employees demanded similar
adjustments. This evidently put pressure on 2.00
prices of semi-durable and durable products 0.00 Sou
and hence the observed increases in these STAT
Jan-07
Jan-09
Jan-11
Jan-13
Jan-15
May-16
May-08
May-10
May-12
May-14
Sep-07
Sep-09
Sep-11
Sep-13
Sep-15
-2.00
prices in the second half of 2016. A c
-4.00
Sou
4.3. Monetary Policy Considerations infl_diff SW_nom_int SA_nom_int
As highlighted earlier inflation differential thr
Source:
Source: Authors
Authors calculations
calculations using
using data from data from
the SARB, theCSO
CBS, SARB,
and
has strong implications for monetary policy STATSSA. mo
CBS, CSO and STATSSA.
especially in a currency union. The fact that Of important note though; is the implied policy out
relative PPP holds between South Africa and
Of important
Swaziland provides relief that even though movements in note though;
real terms is the
which implied
takes into ran
policy movements in real
inflation differentials emerge from time to account inflation movements. This gives a better terms which takes
pol
time, they are mean reverting thereby they into account inflation movements. This gives
view of whether monetary policy is expansionary
do not threaten the viability and relevancy a better view of whether monetary policy is
rela
of the currency union (i.e. CMA). However, or expansionary
contractionary. or contractionary.
Policy rates in real Policy
termsrates
are per
in real
the inflation differentials impose some shown in figure 8. terms are shown in figure 8.
the
complications in terms of trade-off between
nominal convergence and real convergence. As As itit can
canbebeseenseeninin figure
figure 8; 8; whenever
whenever there there
are per
are significant inflation differentials
As shown in figure 7; regardless of the significant inflation differentials pursuing similar pursuing
On
inflation differential nominal policy rates in similar nominal rates to those of South Africa
Swaziland track closely those of South Africa widens the differential in real policy rates
nominal rates to those of South Africa widens the Cen
and translates
and in cases where there is a differential differential in realtopolicy
different
rates monetary
and translates policy
to pol
stances.
in policy rates, Swaziland policy rates fall different monetary policy stances. con
below those of South Africa amid at narrow
Figure 8: Real Convergence – Swaziland and the
margins of not more than 50 basis Research
points.Bulletin Volume 2
South Africa Policy Rates in Real Terms
This is expected given the capital arbitrage
14.00 6.00 Con
that comes with the CMA membership.
12.00 4.00
The
2.00
10.00
0.00 imp
8.00
Jul-08
Jul-14
Oct-07
Apr-09
Jul-11
Oct-10
Apr-12
Jan-13
Oct-13
Apr-15
Oct-16
Jan-07
Jan-10
Jan-16
-2.00 cur
6.00 -4.00
mo
4.00
Figure
-6.00
8: Real Convergence – Swaziland and
South Africa Policy Rates in Real Terms the
2.00 inf_diff SW_real_int SA_real_int
diff
0.00 Source: Authors
Source: Authors calculations
calculations using datausing data
from the from
SARB, CBS,the
CSOSARB,
and
STATSSA . and
Jan-07
Jan-09
Jan-11
Jan-13
Jan-15
May-16
May-08
May-10
May-12
May-14
Sep-07
Sep-09
Sep-11
Sep-13
Sep-15
119
throughout the year reflecting a restrictive
Source: Authors calculations using data from the SARB, CBS, CSO and Central Bank Of Swaziland © 2018
STATSSA. monetary policy as headline inflation stayed
Of important note though; is the implied policy
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
A case in point is what was noted in 2016, the In 2016, the common shock was the El’nino
South African real policy rate was positive induced drought and it affected both
throughout the year reflecting a restrictive countries but for Swaziland it raised food
monetary policy as headline inflation stayed prices by more than 5 percentage points
outside the South African Reserve Bank target from those of South Africa arguably because
range of 3-6 per cent. However, Swaziland’s of supply bottlenecks in Swaziland. From
real policy rate was consistently negative the demand side, the paper notes that in the
reflecting a relatively looser monetary second half of the year; the expansionary
policy stance even in a period of elevated fiscal policy particularly adjustment in public
inflation. This can also explain the inflation wages following the implementation of the
differentials observed during this period. salary review created inflationary pressures
which may have fed through to the widening
On the basis of this analysis it important of inflation differentials between Swaziland
that the Central Bank factors in inflation and South Africa. This was further worsened
differential in its policy reaction function by a relatively looser monetary policy when
in order to ensure real convergence and considering real policy rates movements.
harmonize policy stances within the currency The traditional culprit in explaining inflation
union. differential - differences in CPI weights -
only explained 40 percent of the recent
CONCLUSION differential. The paper recommends that
The analysis of inflation differential remains inflation differential be taken into account
an important subject of interest for countries in the monetary policy reaction function so
in a currency union so as to ensure harmony as to correct for real convergence in policy
of monetary policy stances and hence the rates and thereby harmonize monetary
viability of the union itself. This paper policy stances within the union.
studied inflation differentials between
Swaziland and South Africa and implications REFERENCES
for monetary policy in Swaziland. The paper Braun, W., 2010. Inflation Differentials,
studies long-term convergence and short- Price Differentials and Convergence in
run dynamics using the PPP model. The the Eurozone.
results showed that relative PPP holds for [Online] Available at: https://www.
Swaziland and South African inflation rates amherst.edu/media/view/329608/
and that there is high contemporaneous pass- original/Braun-InflaDiffer,
through effect from South African prices to PrizeDiffer,ConverEurozone.pdf.
Swaziland’s prices. These results support Carolina, M. G., 2006. PPP Theory in a
that the currency union remains viable and Fixed Exchange Rate System, Antillen:
relevant for Swaziland. BNA Working Paper/06/02.
Disaggregating inflation differentials Coakley, J. & Snaith, S., 2004. Testing for
between tradables and non-tradables reflect Longrun Relative Purchasing Power
that recent inflation differentials (in 2016) Parity in Europe, Colchester, UK:
mainly came from the tradeable sector University of Essex.
which reflected the violation of the law of
one price between the two countries. From Coudert, V., 2004. Measuring the Balassa-
the supply side, the paper noted that; even Samuelson Effect for Countries
though Swaziland and South Africa faced of Central and Eastern Europe.
similar shocks, the results and adjustment Banque de France Bulletin Digest,
mechanism to the shocks were asymmetric. February(122), pp. 23-43.
De Boef, S., 2000. Modelling Equilibrium Rabanal, P., 2008. Inflation Differential in
Relationship: Error Correction Models a Currency Union: A DSGE Perspective.
with Strong Autoregressive Data. Tarragona, Centre for Economic Policy
Political Analysis, 9(1), pp. 78-94. Research (CEPR), pp. 1-45.
Dedu, V. & Dumitrescu, B. A., 2010. The Ridhwan, M. M., 2015. Inflation
Balassa-Samuelson Effect in Romania. Differentials, Determinants and
Romanian Journal of Economic Convergence: Evidence from Indonesia
Forecasting, Volume 4, pp. 44-53. Sub-National Data. Kuala Lumpur, The
Journal of Developing Areas, pp. 108-
Haan, J. d., 2010. Inflation Differentials 130.
in the Euro Area: A Survey. In: The
European Central Bank at Ten. Berlin: Taylor, A. M. & Taylor, M. P., 2004. The
Springer-Verlag, pp. 11-32. Purchasing Power Parity Debate,
London: Centre for Economic Policy
Mihaljek, D., 2002. The Balassa-Samuelson Research.
Effect in Central Europe: A
Disaggregated Analysis. Budapest, Wang, J.-Y., Masha, I., Kazuko, S. &
Bank for International Settlements. Leighton, H., 2007. The Common
Monetary Area in Southern Africa:
Miletic, M., 2012. Estimating the Impact Shocks, Adjustment and Policy
of the Balassa-Samuelson effect Challenges, African Department: IMF
in Central and Eastern European Working Paper 158.
Counries: A Revised Analysis of
Panel Data Cointegration Tests. Wickremasinghe, G., 2004. Purchasing
PANOECONOMICUS, Volume 4, pp. 475- Power Parity Hypothesis in Developing
499. Economies: Evidence from Sri Lanka,
Victoria: Monash University Caulfield.
QUARTZ Africa (2017) reveals that the The rapid increase in the consumer price
collusion was a global effort as banks across inflation in the first half of 2008 was driven by
four continents worked together to distort most notably a rise in oil prices, international
the currency market. The banks are alleged commodity prices and domestic food prices.
to have used the Reuters currency trading The weakening of the Lilangeni/Rand
platform, Bloomberg instant messaging exchange rate aggravated by rising inflation
services and telephone calls to organize in South Africa also contributed to further
appropriate trading times, prices and taking increases in the domestic consumer price
turns to transact, hold or pull bills. As a inflation as imported inflation, given that
result, these colluded trades would affect the country imports over 90 per cent of its
anyone buying Rands or using US Dollars to goods from South Africa.
buy Rand (QUARTZ Africa, 2017).
Conversely, the downturn in inflation over
In essence, the act of collusion by the banks the subsequent periods can be attributed
can be transmitted through various channels to reductions in international oil prices and
of economic activity which include inflation, the strengthening of the Lilangeni/Rand
foreign debt, trade and gross reserves exchange rate. The overall inflation is seen
resulting in a misdirected policy stance to have moderated in line with a moderation
by the various policy institutions of the in the volatility in the Lilangeni/Rand
member countries. As a member of the CMA, exchange rate between fourth quarter 2009
Swaziland is not spared from the envisaged and first quarter 2011.
brunt of these currency distortions. This As oil prices increases began to emerge
paper therefore provides a brief review coupled with a depreciation of the Lilangeni/
of the relationship between exchange Rand exchange rate against the U.S dollar,
rate movements and the major economic petrol prices also rose sharply. This saw
indicators in Swazilandmindful of the date transport inflation increasing to almost
in which the Rand-Dollar manipulation took 15 per cent in the first quarter of 2012.
place. Throughout the remainder of the review
period, overall inflation remained relatively
1.1 Inflation and Exchange Rate low and such developments were largely
By the dawn of the 2007/2008 financial crisis, attributed to lower food and transport
Swaziland’s overall rate of inflation had prices which minimised the effects of the
skyrocketed to double digits and maintained continued weakening of the local currency.
an upward trajectory for the first three
quarters of 2008. Having averaged a high of Exchange rate depreciation(appreciation)
13.9 per cent in the third quarter of 2008, by nature may quicken or gradually lead
the consumer price inflation thereafter to a rise (fall) in consumer price inflation.
rapidly decelerated in the subsequent years The analysis presented by Figure 1 shows
but still averaged around 5 and 9 per cent stronger historic cases of the relationship
over the period 2009 and 2014. Up until the between inflation and the exchange rate.
fourth quarter of 2014, Swaziland’s overall For instance, between the first quarter of
consumer price inflation had remained at 2009 and second quarter 2011 the consumer
single digits showing that the monetary price inflation is found to be decreasing
policy has been effective in curtailing consistent with an appreciation of the local
undesirable levels of inflation. Accounting currency. On the other hand, between the
for these developments (in inflation) were fourth quarter of 2013 and fourth quarter
a combination of domestic and external 2014 a depreciation coincides with a rise in
factors. the rate of inflation.
Pu
2008 2009 2010 2011 2012 2013 2014
quarter of 2013 and fourth quarter 2014 a Public External Debt Exchange rate
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
depreciation coincides with a rise in the rate of Source: Central Bank of Swaziland
inflation.
By the end of the fourth quarter of 2010,
Figure 1:
1: Exchange
ExchangeRate
Rate
andand Inflation
Inflation in in Swaziland’s grossappreciation
driven by the external outstanding debt had
in the exchange
Swaziland value of athe
followed Lilangeni
declining exchange
trend rate.at E2.5
which closed
40.0 12.0
billion. The downward movement in foreign debt
Exchange Rate(E/US$)
Inflation Rate(%)
30.0 10.0
The downward trajectory (appreciation) in
8.0 as
20.0
6.0
theshown in Figure
exchange was2 thus
was mainly drivenfor
a benefit by debt
the
10.0
4.0 servicing obligations
appreciation given that
in the exchange value external
of the
0.0 2.0 debt for central
Lilangeni government
exchange constitutes
rate. The downwardthe
Mar
Mar
Mar
Mar
Mar
Mar
Mar
Sept
Sept
Sept
Sept
Sept
Sept
Sept
-10.0 0.0 largest share of total loans portfolio. On the
trajectory
other hand,(appreciation) in thedepreciation
the substantial exchange wasof
2008 2009 2010 2011 2012 2013 2014 Research Bulletin Volume 2
Transport Inflation
Food Inflation the Lilangeni
thus a benefit forfrom June 2011
debt servicing to December
obligations given
2014 external
coupled debtwith increases the reduct
for centralin government
drawdowns
Overall Inflation
Exchange rate that
on Trade
1.3 existing
andprojects
Exchangeresulted
Rates in the drastic import bi
Source:
Source: Central
Central Bank Bank of Swaziland
of Swaziland constitutes
increases the largest
in external share of total loans
debt.
Under normal circumstances, a weaker domestic quarter-on
portfolio. On the other hand, the substantial
1.2
1.2 Foreign
ForeignDebt
Debtand
andExchange
ExchangeRate
Rate currency boosts growth in export earnings while substantia
1.3 Trade
depreciation and Exchange
of the LilangeniRates
from June 2011 to
Though expressed
Though expressedin inEmalangeni
Emalangeni terms,
terms, the Under normal
discouraging circumstances,
imports. In the case of aSwaziland,
weaker imports g
the country’s foreign debt position is a December 2014 coupled with increases in
rch Bulletin Volume
country’s 2 foreign debt position is a function of domestic currency boosts growth in export
function of exchange rate developments and trade activity is almost entirely catered for by the Over the
ed denominated
exchange rateexternal liabilities,
developments for instance,
andexternal
drawdowns a
on earnings while
drawdowns discouraging
on existing projectsimports. In the
resulted in the
drawdowns on the existing loans. neighbouring South Africa to which the Rand and weakened
case of Swaziland, trade
drastic increases in external debt.activity is almost
re For dollar
depreciate
the ofdenominated
existing the external external
local unit against
loans. the liabilities,
ForUS dollar
dollar entirely are
Lilangeni catered
fixed onfora one-on
by theone neighbouring
basis. As a 2014 to E1
for instance, a depreciate of the local
ort would
Central lead
Bank to a rise in ©
of Swaziland outstanding
2018 foreign debt, South Africa to which the Rand and Lilangeni 142 |
unit against the US dollar would lead to a result, more than 80 per cent of the country’s the dollar.
he
Page
and the reverse is true. As seen in the figure are fixed on a one-on one basis. As a result,
rise in outstanding foreign debt, and the imports come 80from
more than perSouth
centAfrica.
of theOn the other
country’s
reverse is true. As seen in the figure
below, outstanding external debt over the review below,
imports
hand, morecome
than from
half ofSouth Africa. exports
the country’s On the 1.4 Gross
outstanding external debt over the review
period has been mimicking the movements in the other hand, more than half of the country’s
by period has been mimicking the movements (approximately 60%) are destined for the South Rates
exchange rate. exports (approximately 60%) are destined for
in the exchange rate. African market with market
the remainder accounted for The curre
se the South African with the remainder
sis
accounted
by forthe
the rest of by world.
the rest of therefore
This the world. This
means reserves is
Figure 2:
Figure 2: Exchange
ExchangeRate
Rateand
andPublic
PublicExternal
External
Debt
DebtininSwaziland
Swaziland therefore means that a percentage of the
ric that a percentage of the developments in the USD and
4,000 12.0 developments in the country’s trade arena
country’s
would betrade arena would be susceptible to other fore
Exchange Rate (E/US$)
3,000
rst 2,500 8.0 tradingincurrencies.
changes the major trading currencies. expected t
2,000 6.0
he 1,500 4.0
Figure
Figure 3:3:Exchange
Exchange RateTrade
Rate and and Trade gross reser
1,000 Developments in Swaziland
Developments in Swaziland
ng 500 2.0 be affected
7,000.0 12.0
- 0.0
Exchange Rate(E/US$)
Exports & Imports (E'Million)
Mar
Mar
Mar
Mar
Mar
Sept
Sept
Sept
Sept
Sept
Sept
Sept
5,000.0
th 8.0 reflects on
2008 2009 2010 2011 2012 2013 2014 4,000.0
6.0
a Public External Debt Exchange rate 3,000.0 Between t
4.0
2,000.0
of Source:
Source: Central
Central Bank
Bank of of Swaziland
Swaziland 1,000.0 2.0 quarter 20
- 0.0 high of E7
By
By the
the end
end of
of the
the fourth
fourth quarter
quarterofof 2010,
2010,
Mar
Mar
Mar
Mar
Mar
Mar
Mar
Sept
Sept
Sept
Sept
Sept
Sept
Sept
Swaziland’sgross
gross external outstanding debt respectivel
in Swaziland’s external outstanding debt had 2008 2009 2010 2011 2012 2013 2014
had followed a declining trend which closed Imports Exports Exchange rate partly due
followed a declining trend which closed at E2.5
0 at E2.5 billion. The downward movement in Source: Central Bank of Swaziland of the Rand
billion.
foreignThe downward movement in2foreign debt
Exchange Rate(E/US$)
Research Bulletin V
Reserves (E'Million)
7,000
review. Accounting, in part, for the positive 10.0 resp
6,000
development in the trade account was a 5,000 8.0 also
depreciation of the local currency when 4,000 6.0
3,000 stab
compared to major trading partners and 2,000
4.0
March
March
March
March
March
March
March
Sept
Sept
Sept
Sept
Sept
Sept
Sept
2014, quarter-on-quarter exports improved
mac
substantially to an estimated E5.87 billion,
2008 2009 2010 2011 2012 2013 2014 use
while imports grew considerably above
Gross Official Reserves Exchange Rate(EP)
E4.5 billion. Over the same period, the con
Lilangeni had weakened from E10.87 in the Source: Central Bank of Swaziland
Source: Central Bank of Swaziland
the
first quarter of 2014 to E11.21 in the fourth
Consistentwith
Consistent with a sharp
a sharp declinedecline of the
of the Lilangeni Ulti
quarter of 2014 to the dollar.
Lilangeni against the dollar in the period
against the dollar in the period between 2012 to infla
between 2012 to 2014, reserves accelerated
1.4 Gross Official Reserves and Exchange 2014, reserves
from a low of accelerated
E3.8 billionfrom a low 2012
in March of E3.8to of i
Rates
E8.0 billion
billion in December
in March 2012 to E8.0ofbillion
2014.inDespite
Decemberthe and
The currency composition of the country’s
positive outcome on the country’s reserves,
reserves is kept at approximately 60% of 2014. Despite the positive outcome on the be i
it is important to note that in 2014 the
ZAR, 25% USD and the rest comprises of
country’s
Lilangenireserves, it is important
had performed badly toin note that in
comparison rate
the numerous other foreign currencies.
to thethe
2014 major currencies
Lilangeni specifically
had performed the in
badly US
Consequently, it is expected that a sizeable
dollar. This
share of the country’s gross reserves comparison to the major currencies specifically
especially those held in dollars will be rela
the
2.0USPolicy
dollar. Implication: The Inflation
affected by either an upward or downward the
movement of the value of exchange which Model
2.0 Policy Implication: The Inflation Model pre
also reflects on the country’s level of import Swaziland together with other members
cover. Between the first quarter of 2009 and Swaziland
of the CMA together
the LNSwithcountries
other members
adopted of the
the stud
second quarter 2011 gross official reserves fixed the
CMA exchange rate targeting
LNS countries adopted system
the fixed to infla
fell from a high of E7.5 billion and a low of monetary policy implementation. Under this
exchange rate targeting system to monetary Dlam
E3.4 billion respectively. Contributing to the system, government and the Central Bank
decline was partly due to an appreciation policy
have tied implementation. Under this
the official exchange ratesystem,
at par the
in the external value of the Rand/Lilangeni with the South African Rand. In this
government and the Central Bank have tied the regard, doll
exchange rate to the major currencies most the intermediate objective of the monetary
official exchange rate at par with the South and
importantly, due to the fact that most of the policy in Swaziland is to maintain the fixed
fluctuations in the reserves emanates from exchange
African rate
Rand. In between
this regard,the
theLilangeni and
intermediate infla
government budgetary obligations especially the Rand.
objective Thismonetary
of the requirespolicy
thatinthe country’s
Swaziland is Janu
because of the revenue from the Southern currency be backed by the international
to maintain
reserve the fixed exchange
requirement rate between
standard the
of a three
African Customs Union (SACU) which add to Fro
reserves yet they are then allocated towards months import
Lilangeni and thecover.
Rand. This requires that the
Ord
the national fiscus. country’s currency be backed by the international
con
Central Bank of Swaziland © 2018
Page
Over and above the fixed exchange rate Instead of using a dummy variable which
responsibility, the Central Bank of Swaziland would have helped to capture the difference
is also tasked with the responsibility of in time events before and after the alleged
maintaining stable and low prices (inflation) collusion, the paper specifically focuses on
and a sound financial system that will ensure the period within which the crime is alleged
sustainable growth in the economy. Informed to have been committed. Once the statistical
by a number macroeconomic variables the significance of the variable of the Rand/
Central Bank thus uses the discount rate Lilangeni exchange rate is determined we
to influence monetary conditions in the shall analyse the regression outcome under
country, which almost mimics the South the pretence that the manipulation had an
African Reserve Bank’s repo rate. Ultimately, impact on the consumer price inflation in
the Bank monitors developments in inflation Swaziland but still subject to the enquiry’s
and guards against any undesirable rates of outcome in South Africa.
inflation by responding with an appropriate
and effective monetary policy action that Table 1: Long-run regression results
would be in the form of a manipulation of Variable Coefficient Std. t-Statistic Prob.
the discount rate. Error
LOGCPI_SA 0.87243* 0.09826 8.8784 0.000
This section therefore seeks to establish the
relationship and impact of the exchange LOGEXR_ 0.620057* 0.04225 14.673 0.000
rate over the period in which the said USD
collusions are presumed to have occurred. LOGOIL_ 0.172665* 0.02431 7.1014 0.000
PRIC
From previous studies that have investigated
the determinants of inflation in Swaziland C -1.50135* 0.45176 -3.3276 0.001
(Ndzinisa, 2008: Dlamini & Dlamini, 2001), R-squared 0.800614 S.E
the magnitude of the impact of the Rand/ 0.058032
Lilangeni exchange rate to the US dollar, Adjusted R-squared 0.793137
South Africa’s consumer price inflation and *=1%,**=5%,=***=10%
oil prices on the consumer price inflation(CPI_
SD) in Swaziland over the period January The positive coefficient of the exchange
2008 to December 2014. rate as depicted by Table 1 is found to be
significant a 1 percent level signifying that a
From the preceding variables, the following depreciation in the Rand/Lilangeni exchange
Ordinary Least Squares (OLS) model is rate against the US dollar over the period in
constructed: question was inflationary. If there is concrete
Research Bulletin Volume 2
evidence that these banks infiltrated the
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆𝑡𝑡 = 𝛽𝛽0 + 𝛽𝛽1 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆𝑡𝑡 currency market by altering the forces of
0.058032
Research Bulletin Volume 2 supply and demand, then a depreciation
+ 𝛽𝛽2 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑈𝑈𝑈𝑈𝑈𝑈𝑡𝑡 in the RandR-squared
Adjusted which is transferrable 0.793137 to the
_𝑆𝑆𝑆𝑆𝑡𝑡 + 𝛽𝛽3 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 peripheral
0.058032 economies
*=1%,**=5%,=***=10% of the CMA is a cause
Research 𝑡𝑡 + 𝜇𝜇𝑡𝑡 Volume
Bulletin Research2 Bulletin Volume 2
Research Bulletin for concern
Volume
The 2 about the future stability and
𝑈𝑈𝑈𝑈𝑡𝑡 Research Bulletin Volume 2 positive coefficient of the exchange rate as
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆
where 𝑡𝑡𝛽𝛽=denotes
𝛽𝛽 + 𝛽𝛽Adjusted
0denotes 1 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆 the elasticity
=R-squared
𝛽𝛽0 + coefficients,
𝑡𝑡 𝛽𝛽coefficients, 0.793137 credibility of the monetary 0.058032 policy stances0.058032
where 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆 = the
𝛽𝛽 elasticity
𝑡𝑡+ 𝛽𝛽 1 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆𝑡𝑡𝜇𝜇𝑡𝑡 the
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆 0.058032
𝑃𝑃𝑃𝑃𝑃𝑃𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆 *=1%,**=5%,=***=10% that depicted
have been by Table
taken 1 is
and/or found
will to pursued
be be significant a 1
𝑡𝑡 + 𝜇𝜇𝑡𝑡 the error 𝑡𝑡 = 𝛽𝛽term, 0 + 𝛽𝛽1 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑆𝑆𝑆𝑆𝑡𝑡 is South Africa
𝑡𝑡 0 1 𝑡𝑡 0.058032
+ 𝛽𝛽2 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑈𝑈𝑈𝑈𝑈𝑈 + 𝛽𝛽
𝑡𝑡 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑈𝑈𝑈𝑈𝑈𝑈 byR-squared
the countries in the union. Hence,
error term, price The
𝐶𝐶𝐶𝐶𝐶𝐶_𝑆𝑆𝑆𝑆 𝛽𝛽𝑡𝑡positive
is South 2 coefficient
Africa consumer of𝑡𝑡 the exchange
Adjusted rate as R-squared
0.793137
consumer + 𝛽𝛽2+ index, 2 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑈𝑈𝑈𝑈𝑈𝑈𝑡𝑡 is the local
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑈𝑈𝑈𝑈𝑈𝑈 𝑡𝑡
percent
Adjusted
imposing
*=1%,**=5%,=***=10%
level
Adjusted
R-squared
a set
signifying
of restraints
that a 0.793137
0.793137 depreciation
against such
in the
+ 𝛽𝛽 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 + 𝜇𝜇 Adjusted R-squared
*=1%,**=5%,=***=10% 0.793137
ficients,
priceunit the 𝐸𝐸𝐸𝐸𝐸𝐸_𝑈𝑈𝑈𝑈𝑈𝑈
𝜇𝜇index,
𝑡𝑡 against 3 the depicted USD, is+andby𝛽𝛽3Table
the
+𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃
𝑡𝑡local
𝛽𝛽3 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃
𝑡𝑡unit
1 𝑡𝑡is+against
found
𝜇𝜇𝑡𝑡 𝑡𝑡 The
+to 𝜇𝜇𝑡𝑡be significant
isthe a 1exploitexchange
*=1%,**=5%,=***=10%
Rand/Lilangeni rate against
the international + 𝛽𝛽3 price of oil. 𝑡𝑡For estimation + 𝜇𝜇 𝑡𝑡
unscrupulous
positive coefficient
*=1%,**=5%,=***=10%
The positive becomes
of the exchange
coefficient aofrelevant
rate as the US
the exchange rate a
rica USD, consumerand percent
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 level
is the signifying
international that a depreciation
price The The positive
necessity.
positivein the coefficient
coefficient of theof the exchange
exchange rate rate
as as
𝛽𝛽 denotes purposes,
where the elasticity
𝛽𝛽thedenotes datacoefficients,
series
the is transformed
𝑡𝑡 elasticity the
𝜇𝜇𝑡𝑡 coefficients, into
depicted dollar
𝜇𝜇𝑡𝑡 theby Table over
1 is foundthe toperiod
be in question was
where 𝛽𝛽 denotes the elasticity coefficients, 𝜇𝜇 the depicted by Table 1 significant
is found to abe 1 significant a
unit
hereof against
𝛽𝛽 oil.
denotes
natural Fortheestimation
the elasticity
logarithms. Rand/Lilangeni coefficients,
purposes, exchange
the data𝜇𝜇𝑡𝑡 the 𝑡𝑡
series rate is depicted
against
depicted the
by by
US
Table Table
1 is 1 is found
found to betosignificant
be significant
a 1 a1
term, 𝐶𝐶𝐶𝐶𝐶𝐶_𝑆𝑆𝑆𝑆error 𝑡𝑡 term, is South 𝐶𝐶𝐶𝐶𝐶𝐶_𝑆𝑆𝑆𝑆 Africa𝑡𝑡 is South
consumer Africa percent consumer inflationary.
level signifying Ifthat
there is concreteinevidence
a depreciation the that
error term, 𝐶𝐶𝐶𝐶𝐶𝐶_𝑆𝑆𝑆𝑆 is South Africa consumer percent level signifying that a depreciation in th
rnational
ror transformed price
term, 𝐶𝐶𝐶𝐶𝐶𝐶_𝑆𝑆𝑆𝑆 into𝑡𝑡 is
dollar
𝑡𝑡
naturalSouth over Africa
logarithms. the consumerperiod in percent
question
percent level level
was signifying
signifying that that
a a depreciation
depreciation in in
the the
ndex, 𝐸𝐸𝐸𝐸𝐸𝐸_𝑈𝑈𝑈𝑈𝑈𝑈 is the𝐸𝐸𝐸𝐸𝐸𝐸_𝑈𝑈𝑈𝑈𝑈𝑈
local unit isagainst theunit against the theseRand/Lilangeni
banks infiltrated the currency market by
126
price index, the local Rand/Lilangeni exchange rateexchange
against the
rate US against the U
price index, 𝐸𝐸𝐸𝐸𝐸𝐸_𝑈𝑈𝑈𝑈𝑈𝑈 is the local unit against the Rand/Lilangeni exchange rate against
eicedata index,series is
𝐸𝐸𝐸𝐸𝐸𝐸_𝑈𝑈𝑈𝑈𝑈𝑈 is the local unit againstisthe the the US
Central Bank Of Swaziland © 2018
nd 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃
USD, and 𝑡𝑡 isinflationary.
the international
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃
If there
is the price concrete
international dollar
evidence
Rand/Lilangeni
price altering
over the
that
theexchange
forces ofinrate
period supply against
and demand,
question was
USthen a
USD, Instead
and of using a dummy
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 is the variable
𝑡𝑡 international which price would dollar over the period in question was wa
dollar over the period in question
s.
D, and 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙_𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 is the international price 𝑡𝑡
C E N T R A L B A N K O F S WA Z I L A N D | RESEARCH BULLETIN VOLUME 2
The results as shown in the Table 1 also 3.0 CONCLUSION AND POLICY
indicate that there exists a statistically RECOMMENDATION
significant positive link between the The purpose of this paper was to examine
consumer price inflation of South Africa, oil the link and/or the relationship between the
prices and the consumer price inflation of exchange rate and the different economic
Swaziland. The magnitude of the coefficient indicators in Swaziland with special focus
implies that a percentage point increase in on the time period in which the Rand-
South Africa’s consumer price inflation over Dollar manipulations occurred. It further
the period under review resulted in a 0.87 demonstrated through a simple regression
per cent increase in the dependent variable analysis, that prices in South Africa which
(inflation). In recognition of this economic are informed by the exchange rate and
fundamental, Swaziland’s inflation is to the actual exchange rate itself do become
a large extent influenced by the price transferred to the domestic economy.
developments in South Africa, to which a
significant component of the goods procured Key towards this discussion, therefore,
by the domestic economy comprises is that the impact of the manipulations,
imported inflation. the outcome and the mitigating action to
restore order as it will be instituted by the
Table 2: Long-run Cointegration commission, should not be viewed in isolation
intercept intercept none of the South African economy but should
& trend
take cognisant of the fact that the CMA
ADF test statistic -2.401 -2.984 -2.419***
links South Africa with the LNS countries to
ADF test critical -2.586 -3.159 -1.614 which their currencies are pegged at parity
values
***-represents statistical significance at 10%.
with the Rand. Hence, given that these
peripheral economies are extremely open to
While the long run regression results in trade with South Africa, the transferability
revealed statistical significance of the of the impact of the distortion is inevitable.
variable under review, the robustness of the Thus pending the outcome of the enquiry
findings is thus confirmed by the significance this paper will assume the following policy
of the residuals through the long run recommendations:
cointegration.
zz Other CMA Member countries should
2.1 The Status of the Investigations conduct studies to examine the extent to
The Competition Commission of South Africa which such collusions might have affected
has since referred the matter to the Tribunal the conduct and implementation of their
where the commission seeks an order macroeconomic policies.
declaring that the seventeen banks indeed
contravened South Africa’s Competition zz That the proceeds from penalty or fine
Act. Upon establishing beyond reasonable imposed by the commission on the banks
doubt that the banks had conspired to be shared among CMA members using a
influence market forces, the commission specified formulae of revenue sharing
will thereafter impose a penalty of 10 per as imposed by the Southern African
cent on their annual turnover. Customs Union. The funds may also be
incorporated as additional payment when
seigniorage compensation payments are
made to LNS countries.
zz The CMA should also come up with policies LNS Economies. Journal of Economic
tailored to safeguard the credibility and Integration 25(2), 324-352.
relevance of the institution and where
there are loop-holes the union should Ndzinisa P. (2008). Efficacy of Monetary
be ready to respond appropriately Policy on Economic Growth in
especially if such a case were to present Swaziland, Central Bank of Swaziland,
itself again. Mbabane.
130
E-mail: research@centralbank.org.sz
Website: www.centralbank.org.sz
Central Bank Of Swaziland © 2018