dynamics of fiscal and current account deficits in thailand: an empirical investigation

Dynamics of fiscal and currentaccount deficits in Thailand:an empirical investigation

Ahmad Zubaidi BaharumshahDepartment of Economics, Faculty of Economics and Management,

Universiti Putra Malaysia (UPM), Selangor Darul Ehsan, Malaysia, and

Evan LauFaculty of Economics and Business, Universiti Malaysia Sarawak (UNIMAS),

Sarawak, Malaysia

Abstract

Purpose – The purpose of this paper is to contribute further on the twin deficits debate in adeveloping economy.

Design/methodology/approach – The data for Thailand over three decades are used as a casestudy.

Findings – The major findings are: first, a stable, long-run equilibrium relationship between fiscaldeficit, interest rate, exchange rate, and current account was found. Second, the causal relationshipbetween the two deficits runs from fiscal deficit to current account deficit. This evidence is supportiveof the twin deficits hypothesis. Further econometric analysis reveals that the two financial variables(interest rate and exchange rate) act as intermediating variables – that is an increased fiscal deficitcauses interest rate to rise, and this in turn puts pressure on the exchange rate. The appreciation of thedomestic currency causes a current account deficit.

Originality/value – The paper is of value by showing both direct and indirect channels to uncoverthe twin deficits phenomena. Based on a persistent profile response, it was found that the adjustmentprocess may take as long as a year to complete.

Keywords Fiscal measures, Exchange rates, Interest rates, Thailand

Paper type Case study

1. IntroductionThailand’s economy emerged as one of the fastest growing economies in the Asia-Pacificregion in 1990 and it has been hailed as the “Fifth Tiger” (Warr, 1999). The spectaculargrowth performance of the economy prior to the 1997 financial crisis has been mainlyattributed to the regime switch from import substitution industrialization and toexport-oriented industrialization strategy in the late 1970s[1]. Thailand was thehardest-hit economy during the East Asian financial crisis. To cope with the crisis, it wasforced to embark on the IMF-mandated program by committing to float the exchangerate and tighten both monetary and fiscal policies. It has been more than nine years sincethe onset of the East Asian financial crisis and Thailand has returned to a position of

The current issue and full text archive of this journal is available at

www.emeraldinsight.com/0144-3585.htm

This paper has benefited from the valuable comments of the anonymous referee and the Editor ofthis journal. This research was partly funded by UNIMAS Fundamental Research (Grant No.03(72)/546/05(45), which is gratefully acknowledged. All remaining errors are the responsibilityof the authors.

JES34,6

454

Journal of Economic StudiesVol. 34 No. 6, 2007pp. 454-475q Emerald Group Publishing Limited0144-3585DOI 10.1108/01443580710830943

relative financial stability and a recorded modest growth rate. Overall, inflation is low,current account is in the surplus and exchange rates are less volatile. However, thelong-term impact of the crisis is still unclear, especially in terms of its fiscal stance.

When the crisis occurred, the government absorbed substantial losses from theprivate sector. Additionally, the government followed an expansionary fiscal policyaimed at stimulating the economy immediately after the crisis. Together, these measurescreated a large increase in public debt reaching a peak at 58 percent of the GDP in thefirst quarter of 2001. The overall fiscal deficit, however, has been contained at below 3percent of GDP every year since the crisis, except in 1999 when in reached 3.3 percent.Nonetheless, new government stimulus programs (e.g., educational reforms, healthinsurance, and village funds), the economic slowdown in the USA, and possibly theAsian Tsunami have the potential to generate additional demand and stress onThailand’s future fiscal budget. Many observers are also concerned about the threats tofiscal soundness in the wake of the economic crisis resulting from: structuralvulnerability due to incomplete financial sector reforms that could require substantialbank bailout programs; and payout of contingent liabilities, particularly for bankguarantees and pension benefits (see Rosengard, 2004). Together they may add to thegovernment’s fiscal burden. Hence, the issue of sustainability as well as its impact on theperformance of the economy becomes important. This study focuses on the second issuesince previous studies have examined the earlier issue (Cuddington, 1996; Bank forInternational Settlements, 2003). In particular, the question of concern is: will the fiscalstimulus package introduced by Thailand lead to deterioration in external balance?

A prolonged and unsustainable budget deficit (BD hereafter) can pose a serious threatto the recovery path. Policymakers in the region are concerned with the large BD as theyhave the potential to derail the economic recovery process. The policy concern is basedon a well-known theoretical argument that suggests a higher BD could put upwardpressure on interest rate, in turn drive up the value of domestic currency internationally,since higher interest rates attract capital flows from the rest of the world, ultimatelyleading to a widening in current account deficit (CAD hereafter) widen[2]. This logic fitsthe US economy in the late 1980s, early 1990s and more recently during the post-Iraq warwhere large BDs incarnate in the imbalance of current accounts – dubbed as the twindeficit hypothesis (TDH) Bahmani-Oskooee (1995), Alse and Bahmani-Oskooee (1992)and Darrat (1988), to name a few, have demonstrated the issue via empirical analysis.

Fiscal imbalances are frequently associated with economic disruptions in theemerging markets as they are more vulnerable to real or financial shocks, and withtheir governments more susceptible to financial constraint. The purpose of this studyis to contribute further to the TDH. To shed further insight on the twin deficit issue inThailand, this paper utilized quarterly frequency data from 1976:Q1 to 2001:Q4 toaddress the following related questions:

. Does BD contribute to CAD (twin deficit) or are they twin divergent – when BDworsens, the CAD improves?

. What are the roles of interest rates (IRs) and exchange rates (EXCs) in theBD-CAD nexus?

Specifically, does empirical evidence support Abell’s (1990 a, b) findings that reveal IRsand EXCs are the primary transmission mechanisms that connect the two deficitsclosely together.

Fiscal andcurrent account

deficits

455

Earlier studies on TDH were based mainly on data from the major industrializedcountries, particularly the USA (Abell, 1990 a, b; Hatemi and Shukur, 2002). It is notunreasonable to expect the empirical evidence from these studies to not fit those of thedeveloping economies due to the different economic and financial structures[3]. Toexamine the TDH, which is the main objective of this study, we relied on severaltime-series econometric methods. More rigorous systematic statistical tests ofintegration, cointegration and causality tests are offered in the present work. In thismanner, we were able to ascertain the robustness of our empirical findings in relationto the link between the two deficits.

Recently, there has been a revival of interest in the twin-deficit phenomenon into theforefront of the policy debate especially for the US economy (see for example, Frankel,2006; Bartolini and Lahiri, 2006; Coughlin et al., 2006). Besides that a series of papers inthe special issue of Journal of Policy Modeling (Vol. 28 No. 6, pp. 603-712, 2006) arededicated to the debate on “Twin deficits, growth and stability of the US economy”.This interest is due to the recent declines in the US current account and fiscal balancesand the impact to the world economic instability.

The rest of the paper is organized as follows. Section 2 presents the theoreticalparadigms and related literature on the connection between the two deficits. In Section3, we briefly discuss the methodological issues and provide an account of the data usedin the analysis. Section 4 reports the empirical findings and Section 5 contains someconcluding remarks.

2. Theory and related literatureThe theoretical explanation for the TDH is based on the well-known Mundell-Flemingframework. According to this model, an increase in BD induces upward pressure oninterest rates that in turn trigger capital inflows and appreciation of the exchange rate.Ultimately, the appreciation of the domestic currency will lead to an increase in CAD.Private saving remains the same as the public perceived the government bond issue tofinance the deficit as increasing their wealth. The response of domestic investment andCAD to a large extent depends on capital mobility. In the case when capital is highlymobile, domestic interest rate is unresponsive (inelastic) to fiscal shock. Hence, there isno crowding-out effect on domestic investment as foreign capital will quickly offset thefall in domestic investment. Capital inflow in turn puts upward pressure on EXCthrough either a rising nominal exchange rate in the case of a flexible exchange ratesystem or rising prices under a fixed exchange rate system. Therefore, theconventional Mundell-Fleming model predicts a positive relationship between the twodeficits.

Beside the Mundell-Fleming framework, there is the Keynesian absorption theorythat links the two deficits. According to the absorption theory, an increase in BD wouldincrease domestic absorption and hence imports, and the expansion of imports leads tothe worsening of the CAD. Hence, like the Mundell-Fleming model, the Keynesianmodel suggests that the causal relationship between the two variable runs from BD toCAD and not the other way round.

At the other end of the spectrum of the twin deficit debate is the RicardianEquivalence Hypothesis (REH). The REH proposed by Barro (1974) suggests that thepublic anticipate future increase in taxes. Knowing that their future disposable incomewill be reduced because of the impending increase in taxes, households reduce their

JES34,6

456

consumption spending and raise savings to smooth out the expected reduction inincome. Thus, there is no effect on national savings, investment, and CAD following aBD. Put differently, REH states that shifts between taxes and BD do not matter for realinterest rate, the quantity of investment, or the current account balance.

Turning to the empirics, the evidence so far does not provide a clear consensus onthe debate. Researchers like Bachman (1992), Rosensweig and Tallman (1993),Bahmani-Oskooee (1995), Vamvoukas (1999), Piersanti (2000) and more recentlyAkbostanci and Tunc (2001) and Leachman and Francis (2002) found support for theconventional view that a worsening BD stimulates an increase in CAD[4]. In contrast,the empirical evidence in Miller and Russek (1989), Enders and Lee (1990), Rahman andMishra (1992), Evans and Hasan (1994), Evans (1987), Wheeler (1999) and Kaufmannet al. (2002) offer support for REH, that is, BD does not affect trade deficits[5].

To sum up, the controversy reflects that two opposing views on fiscal policy prevailin the literature. One is based on the conventional view that BD has importantconsequences on the economy. The other based on the REH view is that BD has noeffect at all (see Piersanti, 2000). However, the impact of increasing BD and CAD is onlyone aspect of TDH. Another important concern is how innovation in the stance of fiscalpolicy impinge on interest rates and other macroeconomic variables such as exchangerate, money supply, inflation (see, e.g. Baffoe-Bonnie, 2004; Apergis, 1998:Bahmani-Oskooee, 1995)[6]. In particular, several authors have shown that higherbudget deficits have led to higher interest rates and, therefore to an appreciation of thenational currency.

3. Methodological issues3.1 Integrational testsIn carrying out cointegration analysis, the first step is to implement unit root tests. Inthis study, we deployed Said and Dickey (1984) (ADF), the modified ADF by Elliottet al. (1996) (DFGLS) that tests the null of nonstationary. In contrast, we also adoptedthe Kwiatkowski et al. (1992) (KPSS) procedure for level (hm) or trend stationarity (ht)against the alternative of a unit root. As such, the KPSS principles involve differentmaintained hypothesis from the ADF and DFGLS unit root tests[7].

3.2 Cointegration procedureA number of tests for cointegration have been developed in the literature. In this study,we adopted the popular Johansen and Juselius (1990) method. For detailed discussionon the procedure, the reader may refer to Baffoe-Bonnie, 2004; Apergis, 1998, amongothers. The system-based cointegration procedure was used to test the absence (orpresence) of long-run equilibria of the TDH. The test utilizes two likelihood ratio (LR)test statistics for the number of cointegrating vectors: namely the trace test and themaximum eigenvalue test. The importance of applying a degree-of-freedom correctionfor the Johansen-Juselius framework is necessary to reduce the excessive tendency ofthe test to falsely reject the null hypothesis of no cointegration. In this study, we reliedon the correction factor suggested by Reinsel and Ahn (1992) that multiplies the teststatistic by (T 2 nk)/T to obtain the adjusted test statistics where T is total number ofthe observations, n is number of series in the system and k is the lag length order ofVAR system.


deficits

457

3.3 Granger causality testsIf cointegration is detected, then the Granger causality must be conducted in vectorerror correction model (VECM) to avoid problems of misspecification (see Granger,1988). Otherwise, the analyses may be conducted as a standard first difference vectorautoregressive (VAR) model. VECM is a special case of VAR that imposescointegration on its variables. The relevant error correction term (ECTs) must beincluded in the VAR to avoid misspecification and omission of the importantconstraints.

However, the workhorses for testing non-causality of ECM and VECM when thevariables are cointegrated are cumbersome and sensitive to the values of nuisanceparameters in finite samples (Rambaldi and Doran, 1996, p. 3). One way to circumventthis problem is to posit a VAR in which the variables appear purely in their level form.Motivated by this development, Toda and Yamamoto (1995) had proposed themodified WALD (MWALD) for testing Granger non-causality linkages that allowcausal inference to be conducted in the level VARs that may contain integrated and(non-) cointegrated processes whether the individual variables are I(0), I(1) or I(2)process. More importantly, the procedure overcomes the pre-test biases thatpractitioners may be confronted with the VECM and other modeling formulationinvolving unit root and cointegration tests.

Toda and Yamamoto (1995) proved that in the integrated and (non-) cointegratedsystem, the MWALD test for restrictions on the parameters of a VAR(k) has anasymptotic x2 distribution when a VAR (p ¼ kþ dmax) is estimated, where dmax is themaximum order of integration suspected to occur in the system and k is the lag lengthselected for the estimation. Furthermore, this procedure imposes (non-) linearrestrictions on the parameters of VARs models without pretest for unit root andcointegrating rank and the MWALD test statistics could be easily computed using theSeemingly Unrelated Regression (SUR) method technique. These restrictions implylong run causal inference since unlike ordinary first difference VAR, this formulationinvolves only variables appearing in their levels. With these caveats in mind, we reportthe results from both procedures for comparison.

3.4 Dynamic analysis: GVDCs and GIRFsIn order to gauge the relative strength of the variables and the transmissionmechanism responses, we shocked the system and partitioned the forecast errorvariance decomposition (FEVD) for each of the variables in the system. However, it iswell established that the results of FEVD based on Choleski’s decomposition aregenerally sensitive to the ordering of the variables and the lag length (see Lutkepohl,1991). To overcome this shortcoming, the generalized variance decomposition (GVDCs)suggested by Lee et al. (1992) is applied here. The innovation of the GVDCs will berepresented in percentage form and strength of each variable to their own shocks andothers is measure by the a value up to 100 percent. The GVDCs are executed using timehorizons of one to 24 quarters. Similarly, we conducted the generalized impulseresponse functions (GIRFs), based on the work by Pesaran and Shin (1998) in thisstudy. Intuitively, the GIRFs trace the response over time of a variable say x due to aunit shock given to another variable, say y. The mapping of the GIRFs is representedby a graphical illustrative. Both are obtained from the moving average (MA)representation of the original VAR model.

JES34,6

458

3.5 Data sourcesQuarterly seasonal nominal data from 1976:1 to 2001:4 was utilized in the analysis. Thedata were gathered from several International Financial Statistics issues, published bythe International Monetary Fund (IMF). The variables employed in the study were thecurrent account (CAD), fiscal position (BD), discount rate representing the short-termnominal interest rate (IR) and nominal exchange rate (EXC) denominated in US dollar.Both the CAD and BD are expressed as ratio of the nominal Gross Domestic Product(GDP)[8] in order to account for the economy’s growth. All the variables are expressedThai baht, except for interest rate and exchange rate.

Thailand’s current account (CAD) and fiscal positions (BD) as a share of GDP over1976-2001 period are plotted in Figure 1 to provide a visual perspective on the twindeficit hypothesis. The plots illustrate that fiscal deficits are accompanied by widecurrent account deficits for the 1976Q1-1987Q1 period. In contrast, we observed thatthe two variables appear to move in the opposite direction for the remaining period.Fiscal surplus averaged around 2.6 percent of the GDP while current account was indeficit, averaging 25.8 percent between 1987Q2 and 1997Q2. The investment boomcontributed to the enlargement of current account deficits during this period[9]. Thecurrent account, however, is noticed to swing from deficit to surplus following thesharp devaluation of the Thai baht in July 1997. The improvement in Thailand’sexternal balance is primarily due to the sharp depreciation in the value of the Thai bahtagainst the US dollar and, to a lesser extent, the yen. Meanwhile, the rise in BD duringthe post-crisis era (averaging 22.9 percent) is largely due to the fiscal stimuluspackage introduced by the Thai’s authority in the aftermath of the currency crisis.Notice that the rise in current account surplus during the post-crisis period is unrelatedto the fiscal position after the financial crisis. Indeed, Higgins (2005) pointed out thecurrent account surplus recorded in the post-crisis period is not due to excessivesavings but due fall in investment (investment drought). Meanwhile, an examination ofthe contemporaneous correlation between BD and CAD suggests that a systematicrelationship exists over the entire sample. The correlation coefficient turns out to be0.89 – meaning fiscal deficits are associated with current account deficits.

Figure 1.Percentage of current

account and budgetaryposition to GDP


deficits

459

While the evidence presented in Figure 1 can be informative, it provides only a weakbasis for assessing the validity of the TDH. The simple correlation between the twovariables is suggestive because it is influenced by many other variables that are notaccounted for in the analysis. It is possible that the Thai current account surplus mighthave turned into large deficit during the post-crisis period had economic growth notcollapsed (Bartolini and Lahiri, 2006). Therefore, further analysis is needed to obtain amore reliable conclusion on TDH. We pursue this task in the next section.

4. Empirical results4.1 Integrational tests resultsSince the data-dependent methods are sensitive in lag selection criteria (k), we followedthe recursive t-statistic procedure suggested by Ng and Perron (1995) with an upperbound of kmax on k. We set kmax to be 12 to overcome this shortcoming. If the lastincluded lag was significant, we would choose k ¼ kmax. If not, we would reduce k byone until the last lag became significant. If no lags were significant then k was set tozero (k ¼ 0). The 5 percent value of the asymptotic normal distribution, 1.96 was usedto assess the significance of the last lag. The procedure adopted here falls into thecategory of the general to specific sequential procedure[10]. Table I summarizes theoutcome of the ADF, DFGLS and KPSS testing in level and first differences performedon all series. Overwhelmingly, all the testing procedures suggest the existence of unitroot or nonstationarity in level or I(1) for all the variables. This is also in line with theliterature that suggest macroeconomic variables are I(1) process (e.g. Apergis, 1998;Daly and Kearney, 1998; Bahmani-Oskooee, 1995). The findings that all the variableshave the same order of integration allowed us to proceed with the Johansenmultivariate cointegration analysis[11].

4.2 Cointegration and hypothesis testing resultsBefore testing for the existence of any cointegrating relationship between thefour-dimensional variables using Johansen procedure, it is necessary to determine thedynamic specification of the VAR model. It is widely known that the lag orders (k) canaffect the number of cointegrating vectors as well as the shape of the impulse function.For this purpose, multivariate generalization of Akaike Information Criteria (AIC)proposed by Gonzalo and Pitarakis (2002) were used to determine the optimal laglength for the vector autoregressive (VAR)[12]. Additionally, we relied on multivariatediagnostic tests for autocorrelation, constant variance and normal distribution tofinally arrive at the optimal lag length of the VAR model. The results tabulated inPanel B, Table I, indicate that VAR(5) is most appropriate for Thailand. The diagnostictest for the model with five lags showed no evidence of autocorrelation and both theconstant variance and normality tests were satisfied[13].

Results of the cointegration procedure (with and without the adjustment factor) arepresented in Panel A, Table II. The null hypothesis of no cointegrating vector (r ¼ 0) infavor of at least one cointegrating vector is rejected at 5 percent significance level. Wenoted that both the trace and the maximum eigenvalue test led to the same conclusion– the presence of one cointegrating vector. Rejecting the null hypothesis of nocointegration implies that the four variables do not drift apart and share at least acommon stochastic trend in the long run.

JES34,6

460

To determine if all the four variables in the system of TDH belong to the cointegratingspace, we applied the log-likelihood ratio (LR) test for the exclusion of each of theindividual variable (i.e. imposing zero restrictions on respective coefficients) asdiscussed in Johansen and Juselius (1990, p. 195). Panel B, Table II provides the testresults of the exclusion restriction on CAD, BD, IR and EXC. Test results reveal thatthe null of restricting the coefficients of CAD, BD, IR and EXC to zero can be easilyrejected at the 5 percent significance level. Clearly, all the variables belong to thecointegrating space and cannot be ruled out from the analysis. Overall, the resultsindicate that the variables share a long-run co-movement that is bounded by theirlong-run equilibrium relationship[14]. To provide further evidence of the TDH, we also

Test statisticstm tt tm tt hm ht

Panel A: unit root testsA: levelCAD 21.527(2) 21.699(2) 21.166(3) 21.659(3) 0.636(3) * 0.236(3) *

BD 22.508(2) 22.438(2) 21.577(3) 21.550(3) 0.676(4) * 0.273(4) *

IR 22.001(2) 22.711(2) 21.424(1) 21.859(1) 1.128(1) * 0.210(1) *

EXC 20.197(2) 21.725(2) 20.533(2) 21.905(2) 1.248(5) * 0.165(5) *

B: first differencesDCAD 26.533(2) * 26.614(2) * 24.098(3) * 24.175(3) * 0.108(3) 0.043(3)DBD 214.70(2) * 214.77(2) * 24.075(3) * 25.668(3) * 0.062(4) 0.047(4)DIR 24.721(2) * 24.870(2) * 25.381(1) * 25.492(1) * 0.188(1) 0.072(1)DEXC 25.828(2) * 25.930(2) * 25.607(2) * 25.776(2) * 0.156(5) 0.065(5)

Panel B: lag selection based on multivariate AICLag AIC1 373.8202 387.6893 393.8364 426.6675 439.944 * *

6 430.7417 427.2428 428.4709 432.559

10 429.15211 421.84812 427.203

Notes: The t, t, and h statistics are for ADF, DFGLS and KPSS respectively. The subscript m in themodel allows a drift term while t allows for a drift and deterministic trend. Refer to the main text forthe notations. Asterisks ( *) indicate statistically significant at 5 percent level. Figures in parenthesesare the lag lengths. The asymptotic and finite sample critical values for ADF are obtained fromMacKinnon (1996) while the KPSS test critical values are obtained from Kwiatkowski et al. (1992,Table 1, p. 166). The DFGLS for the drift term (m) follows the MacKinnon (1996) critical values whilethe asymptotic distributions for the drift and deterministic trend (t) are obtained from Elliott et al.(1996, Table 1, p. 825). Both the ADF and DFGLS test examine the null hypothesis of a unit rootagainst the stationary alternative. KPSS tests the null hypothesis that the series is stationary againstthe alternative hypothesis of a unit root. D denotes first difference operator. ( * *) indicates the optimallag selected for the VAR estimation

Table I.Unit root tests and lag

selection for VAR


deficits

461

k¼

5r¼

1l

max

Tra

ceN

ull

Alt

ern

ativ

eU

nad

just

edA

dju

sted

95p

erce

nt

CV

Un

adju

sted

Ad

just

ed95

per

cen

tC

V

PanelA:Johansenmultivariate

test

r¼

0r¼

142

.993

*34

.395

*23

.920

59.0

35*

47.2

28*

39.8

10r,¼

1r¼

213

.089

10.4

7117

.680

16.0

4212

.833

24.0

50r,¼

2r¼

32.

434

1.94

711

.030

2.95

322.

363

12.3

60r,¼

3r¼

40.

519

0.41

54.

160

0.51

90.

415

4.16

0

PanelB:testof

exclusion

restrictionsbasedon

Johansenprocedure

Thailand

Variables

x2-statistics

(p-value)

CA

D25

.331

(0.0

00)*

BD

18.6

18(0

.000

)*

IR12

.914

(0.0

00)*

EX

C8.

127(

0.00

4)*

PanelC:Normalizingthecointegratingvectors

Variables

CAD

BD

IREXC

21.

0000

0.43

987

20.

0149

890.

0331

98

Notes:

*S

tati

stic

ally

sig

nifi

can

tat

5p

erce

nt

lev

el.T

hek

isth

ela

gle

ng

than

dr

isth

eco

inte

gra

tin

gv

ecto

r(s)

.Ch

osen

r:n

um

ber

ofco

inte

gra

tin

gv

ecto

rsth

atar

esi

gn

ifica

nt

un

der

bot

hte

sts.

Th

eu

nad

just

edan

dth

ead

just

edst

atis

tics

are

the

stan

dar

dJo

han

sen

stat

isti

csan

dth

est

atis

tics

adju

sted

for

smal

lsa

mp

leco

rrec

tion

fact

orac

cord

ing

toR

ein

sela

nd

Ah

n(1

992)

met

hod

olog

y.T

he

test

stat

isti

cof

the

excl

usi

onte

stis

giv

enb

yT

[ln

(12

l* )

–ln

(12

l)]

,w

her

eT

isth

en

um

ber

ofob

serv

atio

ns.

Ith

asx

2d

istr

ibu

tion

wit

hr

(k2

s)d

egre

eof

free

dom

.Her

er

den

otes

the

nu

mb

erof

coin

teg

rati

ng

vec

tors

,kis

the

dim

ensi

onof

the

rest

rict

edco

inte

gra

tin

gsp

ace

ands

isth

en

um

ber

ofv

aria

ble

sin

the

syst

em.T

he

esti

mat

edco

effi

cien

tsw

ere

obta

ined

by

nor

mal

izin

gth

ecu

rren

tac

cou

nt

var

iab

le

Table II.Cointegration test andhypothesis testing

JES34,6

462

tested for simultaneous exclusion of both BD and CAD in the cointegrating vector. TheLR test indicates the test is rejected at high confidence level (x2

ð1Þ ¼ 12:79( p-value ¼ 0.000)). Hence, the statistical evidence indicates that REH is rejected infavor of TDH since both BD and CAD lie in the cointegrating vector.

Next, we proceeded with the estimation of the long-run parameters of the model bynormalizing the current account variable. There was only one significant vectordetected and thus, we did not have the problem of having to identify the equation thatrepresents the current account. Results of normalizing with respect to current accountvariable are presented in Panel C, Table II.

In general, the sign of the estimated parameter normalizing on CAD is consistent witha priori expectation. Importantly, the results suggest that BD does have a positive effecton current account. An increase in the BD would trigger correspondingly an upward hikein CAD imbalances. This supports the conventional view that there exists strongcorrelation between CAD and BD. It is observed in Table II (Panel C) that a 1 baht increasein BD leads to a corresponding increase of 0.44 baht in the CAD. Similarly, the statisticallysignificant positive sign on the EXC variable suggests that devaluation leads to animprovement in the CAD in the long run. From a policy perspective, this finding suggeststhat EXC could be used as a policy target to manage the external imbalances. This is inline with Bahmani-Oskooee and Payesteh (1993) that showed higher budget deficits tendto be followed by an appreciation of the US dollar. Meanwhile, the article by Apergis(1998) reported mixed evidence for a group of OECD countries.

Interestingly, we also found that the IR has a negative impact on CAD. One possibleexplanation could be that higher IR reflects greater returns on investment due to say,higher productivity. This makes domestic goods cheaper and hence, expansion inexports and imports to fall, causing a decrease in CAD. Another possibility is thathigher IR reflects higher investment risk, which may deter foreign capital flows intothe economy leading to a fall in CAD.

4.3 ECM and diagnostic testingAfter identifying the long-run equilibrium estimates of the model, we proceeded withestimating the short-run dynamics of the twin deficits model for Thailand[15]. Themethodology utilized to find the error correction representation is the general tospecific paradigm (see Hendry, 1987). Given that VAR(5) is most appropriate forThailand, we eliminated sequentially the insignificant lags and the most parsimoniouserror correction mechanism for changes in current account along with the singleequation statistics which are summarized in the following equation:

DCAD ¼ 2 0:004 2 0:132ECTt21 2 0:451DCADt21 2 0:376DCADt22

ð22:111Þ ð28:433Þ* ð25:873Þ* ð24:995Þ*

2 0:364DCADt23 2 0:194DCADt25 þ 0:243DBDt þ 0:205DBDt21

ð25:012Þ* ð22:559Þ* ð3:994Þ* ð3:253Þ*

2 0:119DIRt25 þ 0:103DEXCt þ 0:118DEXCt21

ð2:052Þ* ð2:935Þ ð3:264Þ*

Asterisks denote significance at 5 percent level. Figures in parentheses represent thet-statistics.


deficits

463

As reported in the above equation, the estimated ECT in the cointegratingregression appears to be negative and statistically significant, a condition necessaryfor the stability of the estimated model. Importantly, the t-statistics on the laggedresidual is statistically significant supporting the Johansen results reported earlier.The coefficient of the ECT measuring the speed of the temporal adjustment to the longrun equilibrium in the system is denoted by the cointegration relationship. From theabove equation, we found that the speed of adjustment to long run equilibriumfollowing a disturbance is rather slow.

Results also show that short-run movements in BD, IR and EXC all have asignificant impact on current account. The coefficient of BD are positively signed andthey are larger compared to IR and EXC. This result suggests that BD has a positiveimpact on CAD in Thailand. This evidence is consistent with the long-run equilibriumestimates. For the remaining variables, EXC does have positive impact while IRinversely relates to CAD. Overall, the empirical results by the error correction modelappear to support those obtained by the Johansen method.

The diagnostic statistics are tabulated in Table III. The estimated model seems to berobust to various departures from the standard regression assumptions in terms ofserial correlation of the residuals (AR), autoregressive conditional heteroskedasticity(ARCH), misspecification of functional form (RESET), non-normality (J-B), orheteroskedasticity of residuals (White). Furthermore, the plot of the cumulative sumsquares (CUSUM squares) in Figure 2 reveals that the null hypothesis of parameterstability cannot be rejected at the 5 percent level of significance[16]. Therefore, thetemporal stability of the relationship appears to be unaffected by two major events that

Figure 2.Plot of CUSUM squarestest

AR(4) ARCH(4) RESET(1) J 2 B(2) White3.551 2.510 1.137 0.339 0.069(0.470) (0.643) (0.286) (0.844) (0.792)

Notes: The distributional properties of diagnostics are: LM (4) is a test of 4th order serial correlation.ARCH (m) is an m-order test for autoregressive conditional heteroskedasticity. Ramsey’s RESET(Regression Specification Test) test uses the square of the fitted values. J-B (Jarque-Bera) is the test ofthe normality of the residuals. The White general heteroscedasticity test is based on the regression ofsquared residuals on squared fitted values. Figures in parentheses refer to the p-values

Table III.Diagnostic testing

JES34,6

464

occurred during the sample period – the economic recession of the mid-1980s and the1997 Asian financial crisis.

4.4 Granger causality results (VECM and MWALD)In this paper the main emphasis is on the channel through which BD influences CAD.Given the presence of a unique and single cointegrating vector for the four-dimensionalVARs following the cointegration test, this provided us with one error correction term(ECT) to construct the vector error correction model VECM. The purpose is to know themechanism through which BD affects CAD. The CAD equation is the only one in thesystem where the ECT is statistically significant. This suggests that CAD solely bearsthe brunt of short run adjustment to bring about the long run equilibrium in Thailand.In other words, the CAD acts as the initial receptor of any exogenous shocks thatdisturb the equilibrium system. This also highlight that BD, IR and EXC are weaklyexogenous in the system (see Urbain, 1993).

It is evident from Panel A, Table IV that the null hypothesis of BD does not cause (inGranger-sense) CAD and is easily rejected at 5 percent significance level (BD ! CAD).However, we did not find any significant causality running from CAD ! BD. Thisfinding appears to support the conventional view (i.e. the TDH) that emphasizes thatthe causal relationship runs from BD to CAD and not vice versa. We also estimated thebivariate model and test to see if the correlation between the two deficits is consistentwith the other competing hypothesis. The results from the analysis (not shown here)appear to support TDH.

Next, we examined the role of the causing variables (IR and EXC) in the TDH. Wesought for the causal chain that runs from BD to IR, to capital flows, to EXC and finallyto the CAD (BD ! IR ! EX ! CAD) (see Volcker, 1984; Abell, 1990a, b). As presentedin Panel A, Table IV, it is observed that BD causes CAD and it operates through IR andEXC (BD ! IR ! EX ! CAD). It is also observed in Panel A, Table IV there is a thirdchannel in the BD-CAD nexus (i.e. BD ! IR ! CAD). This finding is noteworthy as it

DCAD DBD DIR DEXC ECTDependent variable x2-statistics

Panel A: short run causality in VECM environment (k ¼ 5)DCAD – 31.433(0.000) * 5.834(0.016) * 13.294(0.000) * 13.793(0.000) *

DBD 1.711(0.191) – 0.999(0.318) 0.015(0.902) 0.033(0.856)DIR 1.353(0.245) 4.988(0.026) * – 2.742(0.098) 0.852(0.356)DEXC 0.249(0.617) 1.944(0.163) 9.031(0.003) * – 0.380(0.537)

Panel B: long run Granger non-causality using MWALD (k ¼ 5 d ¼ 1)CAD BD IR EXC

Dependent variable MWALD (x2-statistics)CAD – 12.140(0.032) * 13.615(0.018) * 28.779(0.000) *

BD 7.823(0.166) – 3.776(0.582) 4.033(0.545)IR 4.904(0.427) 12.045(0.034) * – 5.948(0.311)EXC 6.482(0.262) 8.729(0.120) 20.769(0.000) * –

Notes: The x2-statistic tests the joint significance of the lagged values of the independent variables,and the significance of the error correction term(s); D is the first different operator; k ¼ optimum lagand d ¼ maximal order of integration; *statistically significant at 5 percent level

Table IV.Granger causality results


deficits

465

identifies an additional channel for the Thai economy through which CAD may berelated with BD[17].

The causal inference implies that BD in Thailand has a significant effect on IR. Onthe other hand there is no compelling evidence to suggest that the relationship runsfrom IR to BD as reported in Darrat (2002) for the Greek economy[18]. Accordingly, ourfinding is consistent with the Keynesian crowding-out hypothesis that suggests BDraises IR. However, it refutes the Ricardian contention that suggests BD does notimpact IR but instead changes in IR change the stance of fiscal policy (e.g. Ibrahim andKumah, 1996; Darrat, 2002, among others).

It is worth noting that the empirical evidence presented above depends crucially onthe integration as well as the cointegration properties of all the variables in the system(see Toda and Yamamoto, 1995; Masih and Masih, 1999). To complement the VECMprocedure, we applied the Toda-Yamamoto test that has been shown to be valid underless restrictive assumptions. The results of the tests of restrictions from a VARestimated by the procedure prescribed by Toda and Yamamoto (1995, MWALD) arereported in Panel B, Table IV. For the application of the MWALD approach, (kþ 1)order of the VAR was estimated with restrictions performed on lagged terms up to thekth lag. There are no short run causality flows in this case since all the variables in theVAR system appear in level form. In adopting the methodology as a complementarytool, the model employing a VAR in level form will add an extra dimension to theanalysis in addition to providing us with another facet to assess the general robustnessof the results generated from the VECM.

Among others, the MWALD test tallies with the results obtained from the VECMresults discussed earlier. First, we found that BD ! CAD and not vice versa. Second,as portrayed in Panel B, the interplay of the transmission mechanisms work from BDto CAD operating through IR and EXC. Third, it is clear that CAD is the initialreceptors of the exogenous shocks from the other variables in the system of the twindeficits model. All in all, we have demonstrated the important role of IR and EXC inexplaining the twin deficit nexus.

To sum up, the statistical evidence based on two procedures strongly suggests thatBD and CAD are highly correlated and at least three channels exist between the twodeficits. The Thai experience over the past three decades is consistent with the TDH.IR and EXC are important policy variables and both appear to be weakly exogenousand hence could be potentially employed to manage the current account. From a policyperspective, the results suggest that an improvement in BD is necessary to improve theCAD.

4.5 Dynamic analysis: GVDCs and GIRFsIn order to strengthen the empirical evidence from causality analysis, the dynamicanalysis of the system are examined. We relied on GVDCs to gauge the strength of thecausal relationship among CAD, BD, IR and EXC. All in all, the results strengthen thefindings from the causality tests presented earlier. Table V provides the decompositionof the forecast error variances of each variable up to 24-quarter horizon. The threemajor results from this table may be summarized as follows:

(1) CAD seems to be the most interactive variable in the system. The GVDCs showthat almost 47 percent of the forecast error variance can be explained by BD (36percent), IR (1 percent) and EXC (10 percent) at the end of the 24-quarter

JES34,6

466

horizon. This provides for strong direct causality originating from BD to CAD.Furthermore, it is proven that CAD is the most endogenous variable as itemerged as the recipient of shocks originating from the other macroeconomicvariables in the system.

(2) EXC is the most exogenous variable in the system with only about 15 percent ofits forecast variance being explained by the remaining variables in the entireforecast horizon.

(3) Changes in IR are largely due to the movement in BD. The effect of BD on IRappears to be become stronger as the horizon increases. The picture thatemerged from the GVDCs supports the uni-direction causality relationship fromBD ! IR.

Given a system of four-dimensional variables, we may construct up to 12 possiblescenarios of GIRFs for each of the variable taken separately (ignoring their ownshocks). Visual illustrations of the GIRFs up to 50 quarters are presented in Figure 3.Most of the graphs visually exhibit a delayed response to the shock. They are rathersluggish but able to settle after 25 quarters interval. Meanwhile, CAD and BDresponded negatively to the shock in IR implying the existence of a negativerelationship. A striking illustration is detected for the response of BD to the shock inCAD. This picture is quite in contrast with the conventional view of the positive impactfrom BD to CAD. We also detected that CA reverts rapidly back to the equilibrium,

Horizon Due to innovation in:Percentage of variations in (quarters) DCAD DBD DIR DEXC DCU

Quarters relative variance in: DCAD1 68.804 25.486 0.198 5.513 31.1964 54.312 39.506 0.883 5.299 45.6888 51.941 40.077 0.663 7.320 48.059

24 53.185 36.672 0.510 9.632 46.815

Quarters relative variance in: DBD1 6.028 87.255 4.533 2.184 12.7454 12.138 74.348 7.465 6.049 25.6528 13.749 72.779 6.714 6.758 27.221

24 12.299 78.704 4.630 4.367 21.296

Quarters relative variance in: DIR1 0.455 3.116 95.916 0.513 4.0844 1.259 5.552 92.164 1.025 7.8368 2.612 11.580 84.702 1.107 15.298

24 3.760 15.982 79.099 1.159 20.901

Quarters relative variance in: DEXC1 7.728 0.381 7.896 83.996 16.0044 6.732 1.770 8.612 82.886 17.1148 6.437 1.834 8.492 83.237 16.763

24 5.705 1.833 7.197 85.264 14.736

Notes: The last column provides the percentage of forecast error variances of each variable explainedcollectively by the other variables. The columns in italic represent the impact of their own shock

Table V.Generalized variance

decomposition


deficits

467

Figure 3.Generalized impulseresponse function (IRFs)

JES34,6

468

after the shock from EXC, implying that devaluation is taking place in the subsequentquarters. This of course does not follow the J-curve pattern.

Finally, we generated the persistent profile of a system shock following theinformation obtained from the cointegrating vector. Persistence profile of asystem-wide shock is an alternative complementary application of GIRFs. It is aunique measure of the effect by the shock in estimating one-step ahead forecast errorfor the whole system. The persistent profile system-wide shock provides the speed ofconvergence to equilibrium for a cointegrating system. One interesting feature of thiskind of experiment is that it shows how long it takes the system to adjust to thelong-run equilibrium after a real disturbance, or shock occurs. Figure 4 graphs thepersistent profile of a system-wide shock to cointegrating vector (CV). Thecointegrating vector has a tendency to converge to the equilibrium (around 7 to 12quarters). Since the figure does not drift upwards at any time, there is no overshootingimpact for Thailand. Thus, the persistent profile shock shows that thefour-dimensional system in Thailand has the tendency to revert to long-runequilibrium.

5. Summary and conclusionsThe aim of this study is to investigate the causal relationship between BD and CAD inThailand. To discern the causal chain link among the I(1) macroeconomic variablestypically used to test the twin deficit hypothesis, we relied on several econometricprocedures that permit both short-run and long-run linkages. The purpose is tocircumvent some of the problem associated with an individual technique and to assessthe robustness of our empirical results. The following conclusions can be derived formthe empirical evidence.

First, the study finds that the data are consistent with the twin deficit hypothesis –that is BD Granger causes CAD and not the reverse. It turns out, however, that there islack of evidence to support the hypothesis that CAD causes BD or current accounttargeting. The statistical evidence suggests that the fiscal deficit if left unchecked willadversely affect the current account. In other words, a prolonged fiscal deficit willcause a change in current account.

Second, it worth mentioning that the connection between BD and CAD is detectedthrough three channels: directly from BD to CAD; indirectly from BD to higher IR

Figure 4.Persistence profile shockfor cointegrating vector


deficits

469

which in turn appreciates the domestic currency and eventually leads to CAD; andfrom BD to higher IR and leads to CAD. While the first two results have appeared inthe literature, the last finding is uncommon in the TDH literature. Third, the dynamicanalysis based on the generalized variance decomposition (GVDCs) reveals that EXC isthe most exogenous variable among the four variables in the system and IR is largelyinfluenced by BD – that is BD Granger causes IR.

From a policy perspective, a clear conclusion emerged from the empirical work. Agrowing BD will adversely affect the external balance and if the Thai authorities are tocorrect the imbalance in the current account, they cannot ignore the growing size of theBD. In addition, our results suggest that increasing BD will crowd private investmentsthrough its effect on IR. We note that not all previous findings support this conclusion.Baffoe-Bonnie (2004), for example, found that fiscal policy has minimal effect on GNPand exports, and failed to reduce inflation in Ghana.

Finally, further research in this area can be undertaken by including moremacroeconomic variables (money supply, inflation, private investment, income etc.) todepict the twin deficit phenomena. Primarily due to data limitation, we have restrictedour analysis to a four-dimensional system to provide the tentative evidence on the twindeficit hypothesis in Thailand. Therefore, while the results are appealing, some cautionis necessary because earlier studies have shown that there are other variables thatconnect the two deficits.

Notes

1. Besides that, several other factors such as rising labor cost in Japan and the other newlyindustrialized countries (NICs: South Korea, Taiwan, Hong Kong and Singapore); theopening up of the capital markets and increasingly becoming more integrated with theglobal financial markets; the rising yen-dollar rate in the later half of 1980s and theassociated higher Japanese foreign direct investment and stable macroeconomicenvironments have also contributed to the economic progress of the Thai economy.

2. We note that divergent results have emerged in the literature on relationship between realbudget deficits and exchange rates. Apergis (1998), for example, empirically examined therelationship between BD, exchange rate, prices and money supply for eight OEDC countries.The results revealed that higher budget deficits lead to currency appreciation in Germany,the UK, Switzerland, Canada and the Netherlands. They are also in line with Alse andBahmani-Oskooee (1992) for the USA. For the remaining OEDC countries (Italy, Belgium andFrance), Apergis (1998) found that higher deficits lead to the appreciation of domesticcurrency.

3. In developing economies, lack of domestic capital markets, inefficiency of the system oftaxation and huge public spending imply that a substantial portion of the BD needs to relyon international capital markets. A recent study by Kouassi et al. (2004) using data from1975 to 1998 found evidence in favor of feedback between BD and CAD for Thailand. Thetest results for Malaysia and Singapore, however, suggest rising BD does not cause a surgein CAD and vice versa. We note that their analysis was based on a simple bivariate model.

4. Some earlier works by Hutchison and Pigott (1984) and Zietz and Pemberton (1990) alsoidentify a causal relationship running from BD to CAD.

5. Literature on the twin deficits issue has mainly centered on two major theoretical paradigms.However, as pointed out by Darrat (1988) these are not the two possible outcomes betweenthe two deficits. A high correlation between the two deficits is also consistent with two othercompeting hypotheses: namely (1) two variables are mutually dependent, and (2) the

JES34,6

470

causality runs from CAD to BD. Interested readers may refer to the paper by Darrat (1988)and the literature quoted therein.

6. Stabilization aspect of fiscal and exchange rate policy have been investigated in developingeconomies (see Little et al., 1994; Baffoe-Bonnie, 2004). An empirical investigation on Ghanaby Baffoe-Bonnie (2004) revealed that fiscal policy is less effective in influencing inflation,output and exports. The impact of changes in government expenditure on these variables, ifany, is very small.

7. We note that the simulated test by Kwiatkowski et al. (1992) has shown that the KPSS teststatistics perform well in a fairly small sample size (30) and the test yields asymptoticallyvalid results with desirable properties.

8. Quarterly observations of GDP were extrapolated from the annual series employing theGandolfo (1981) quadratic interpolation approach and it is also outlined in Bergstrom (1990).Authors like Arize et al. (2000, 2005) and Lau and Baharumshah (2004) also adopted thisinterpolation technique in extracting the annual GDP series to obtain quarterly observationsfor their study.

9. Thailand was among the top ten recipients of net private capital flows (together withMalaysia and Indonesia) during this period. The boom period coincided with relaxation ofcapital inflows. The Bangkok International Banking Facility (BIBF) was established in early1993. Financial institutions under BIBF were authorized to accept deposits and loans fromabroad in foreign currency and extend loans to both overseas and local markets, as well asengage in cross-currency foreign currency trading and loan syndication (Rajan, 2002).

10. The liberal sequential testing strategy leads to a better size-power trade-off in thesubsequent inferences in the univariate framework (Ng and Perron, 1995). See also Gonzaloand Pitarakis (2002) for detailed discussion.

11. Throughout this paper, the computations were done with E-Views and Microfit computerpackages.

12. The multivariate generalization of AIC was chosen due to its superiority as the bestperforming criterion in lag selection techniques when the system dimension increases(Gonzalo and Pitarakis, 2002).

13. To conserve space, we do not report these results but they are available from the authors onrequest.

14. In the estimation process, we also included an exogenous dummy (DUM) variable in themodel to account for the financial crisis (DUM ¼ 1 for the 1997-1998 and zero elsewhere) andused the LR test to see if the variable was significant. As variable appeared to beinsignificant, we did not include it the final estimation.

15. The Granger representation theorem proves that in the presence of cointegration, therealways exists a corresponding error correction representation (Engle and Granger, 1987).This implies that changes in the dependent variable are a function of the level ofdisequilibrium in the cointegrating relationship (captured by the error correction term) aswell as changes in other explanatory variable(s). The error correction terms (ECT) can beconsistently obtained from the corresponding lagged residuals of the single equationcointegration regression (King et al., 1991).

16. Briefly, if the plot of the CUSUM squares sample path moves outside the critical region (atthe 5 percent significance level), the null hypothesis of stability overtime of the intercept andslope parameters is rejected.

17. This finding may simply reflect the artifact of the exchange rate arrangement systemadopted by the Thai authorities. Thailand follows the de facto peg exchange rate system inmost of the time period in the sample (Reinhart and Rogoff, 2002; Parsley and Popper, 2006).


deficits

471

Despite this setback, this additional channel provides a better understanding of the twindeficits phenomenon. We thank an anonymous referee for pointing this out.

18. Darrat (2002) argued that a government in a small open economy like Greece reacts to largebudget deficits by raising IR. In other words, IR plays an important role in shaping thebudgetary process.

References

Abell, J.D. (1990a), “Twin deficits during the 1980s: an empirical investigation”, Journal ofMacroeconomics, Vol. 12, pp. 81-96.

Abell, J.D. (1990b), “The role of budget deficits during the rise of the dollar exchange rate from1979-1985”, Southern Economics Journal, Vol. 57, pp. 66-74.

Akbostanci, E. and Tunc, G.I. (2001), “Turkish twin deficits: an error correction model of tradebalance”, Working Papers in Economics, No. 6, Economic Research Center (ERC), MiddleEast Technical University, Ankara, Turkey.

Alse, J. and Bahmani-Oskooee, M. (1992), “Are twin deficits really related? A comment”,Contemporary Policy Issues, Vol. 10, pp. 108-11.

Apergis, N. (1998), “Budget deficits and exchange rates; further evidence from cointegration andcausality tests”, Journal of Economic Studies, Vol. 25 No. 3, pp. 161-78.

Arize, A.C., Malindretos, J. and Grivoyannis, E.C. (2005), “Inflation-rate volatility and moneydemand: evidence from less developed countries”, International Review of Economics andFinance, Vol. 14, pp. 57-80.

Arize, A.C., Osang, T. and Slottje, D.J. (2000), “Exchange-rate volatility and foreign trade:evidence from thirteen LDCs”, Journal of Business and Economic Statistics, Vol. 18,pp. 10-17.

Bachman, D.D. (1992), “Why is the US current account deficit so large? Evidence from vectorautoregressions”, Southern Economic Journal, Vol. 59, pp. 232-40.

Baffoe-Bonnie, J. (2004), “Dynamic modelling of fiscal and exchange rates policy effects in adeveloping country; a non-structural approach”, Journal of Economic Studies, Vol. 31 No. 1,pp. 57-75.

Bahmani-Oskooee, M. (1995), “The long-run determinants of the US trade balance revisited”,Journal of Post Keynesian Economics, Vol. 17 No. 3, pp. 457-65.

Bahmani-Oskooee, M. and Payesteh, S. (1993), “Budget deficit and the value of the dollar: anapplication of cointegration and error-correction modeling”, Journal of Macroeconomic,Vol. 15 No. 4, pp. 661-77.

Bank for International Settlements (2003), Fiscal Issues and Central Banking in EmergingEconomies, BIS Papers No. 20, Bank for International Settlements, Basel.

Barro, R.J. (1974), “Are government bonds net wealth?”, Journal of Political Economy, Vol. 82,pp. 1095-117.

Bartolini, L. and Lahiri, A. (2006), “Twin deficits, 20 years later”, Current Issues in Economicsand Finance: Federal Reserve Bank of New York, Vol. 12 No. 7.

Bergstrom, A.R. (1990), Continuous Time Econometric Modelling, Oxford University Press,Oxford.

Coughlin, C.C., Pakko, M.R. and Poole, W. (2006), “How dangerous is the US current accountdeficit?”, The Regional Economist, April, pp. 5-9.

Cuddington, J.T. (1996), “Analysing the sustainability of fiscal deficits in developing countries”,mimeo, Georgetown University, Washington, DC.

JES34,6

472

Daly, K.J. and Kearney, C. (1998), “Fiscal financing decisions and exchange rate variability”,Journal of Economic Studies, Vol. 25 No. 4, pp. 309-27.

Darrat, A.F. (1988), “Have large budget deficits caused rising trade deficits?”, Southern EconomicJournal, Vol. 54, pp. 879-86.

Darrat, A.F. (2002), “On budget deficits and interest rates: another look at the evidence”,International Economic Journal, Vol. 16, pp. 19-29.

Elliott, G., Rothenberg, T.J. and Stock, J.H. (1996), “Efficient tests for an autoregressive unit root”,Econometrica, Vol. 64, pp. 813-36.

Enders, W. and Lee, B.S. (1990), “Current account and budget deficits: twins or distant cousins?”,The Review of Economics and Statistics, Vol. 72, pp. 373-81.

Engle, R.F. and Granger, C.W.J. (1987), “Cointegration and error correction: representation,estimation and testing”, Econometrica, Vol. 55, pp. 251-76.

Evans, P. (1987), “Interest and expected future budget deficits in the United States”, Journal ofPolitical Economy, Vol. 95 No. 1, pp. 34-58.

Evans, P. and Hasan, I. (1994), “Are consumers Ricardian? Evidence for Canada”, QuarterlyReview of Economics and Finance, Vol. 34, pp. 25-40.

Frankel, J.A. (2006), “Twin deficits and twin decades”, in Kopcke, R., Tootell, G.M.B. andTriest, R.K. (Eds), The Macroeconomics of Fiscal Policy, MIT Press, Cambridge, MA,pp. 321-36.

Gandolfo, G. (1981), Quantitative Analysis and Econometric Estimation of Continuous TimeDynamic, North-Holland Publishing Co., Amsterdam.

Gonzalo, J. and Pitarakis, J.Y. (2002), “Lag length estimation in large dimensional systems”,Journal of Time Series Analysis, Vol. 23, pp. 401-23.

Granger, C.W.J. (1988), “Some recent development in a concept of causality”, Journal ofEconometrics, Vol. 38, pp. 199-211.

Hatemi, A. and Shukur, G. (2002), “Multivariate-based causality tests of twin deficits in the US”,Journal of Applied Statistics, Vol. 29, pp. 817-24.

Hendry, D.F. (1987), “Econometrics methodology: a personal perspective”, in Bewley, T. (Ed.),Advances in Econometrics, Vol. 2, Cambridge University Press, Cambridge, pp. 29-48.

Higgins, M. (2005), “The global investment slump and the US current account deficit”, mimeo,Federal Reserve Bank of New York, New York, NY.

Hutchison, M.M. and Pigott, C. (1984), “Budget deficits, exchange rates and current account:theory and US evidence”, Federal Reserve Bank of San Francisco Economic Review, Vol. 4,pp. 5-25.

Ibrahim, S.B. and Kumah, F.Y. (1996), “Comovements in budget deficits, money, interest rate,exchange rate and the current account balance: some empirical evidence”, AppliedEconomics, Vol. 28, pp. 117-30.

Johansen, S. and Juselius, K. (1990), “Maximum likelihood estimation and inference oncointegration with application to the demand for money”, Oxford Bulletin of Economicsand Statistics, Vol. 52, pp. 169-210.

Kaufmann, S., Scharler, J. and Winckler, G. (2002), “The Austrian current account deficit: drivenby twin deficits or by intertemporal expenditure allocation?”, Empirical Economics, Vol. 27,pp. 529-42.

King, R.G., Plosser, C.I., Stock, J.H. and Watson, M.W. (1991), “Stochastic trends and economicfluctuations”, American Economic Review, Vol. 81, pp. 819-40.


deficits

473

Kouassi, E., Mougoue, M. and Kymn, K.O. (2004), “Causality tests of the relationship between thetwin deficits”, Empirical Economics, Vol. 29, pp. 503-25.

Kwiatkowski, D., Phillips, P.C.B., Schmidt, P. and Shin, Y. (1992), “Testing the null hypothesis ofstationarity against the alternative of a unit root: how sure are we that economic timeseries have a unit root?”, Journal of Econometrics, Vol. 54, pp. 159-78.

Lau, E. and Baharumshah, A.Z. (2004), “On the twin deficits hypothesis: is Malaysia different?”,Pertanika Journal of Social Science and Humanities, Vol. 12, pp. 87-100.

Leachman, L.L. and Francis, B. (2002), “Twin deficits: apparition or reality?”, Applied Economics,Vol. 34, pp. 1121-32.

Lee, K.C., Pesaran, M.H. and Pierse, R.G. (1992), “Persistence of shocks and its sources inmultisectoral model of UK output growth”, Economic Journal, Vol. 102, pp. 342-56.

Little, I.M.D., Cooper, R.N., Corden, W.M. and Rajapatirana, S. (1994), Boom, Crisis, andAdjustment: The Macroeconomic Experience of Developing Countries, Oxford UniversityPress, New York, NY.

Lutkepohl, H. (1991), Introduction to Multiple Time Series Analysis, Springer-Verlag, Berlin.

MacKinnon, J.G. (1996), “Numerical distribution functions for unit root and cointegration tests”,Journal of Applied Econometrics, Vol. 11 No. 6, pp. 601-18.

Masih, A.M.M. and Masih, R. (1999), “Are Asian stock market fluctuations due mainly tointra-regional contagion effects? Evidence based on Asian emerging stock markets”,Pacific-Basin Finance Journal, Vol. 7, pp. 251-82.

Miller, S.M. and Russek, F.S. (1989), “Are the twin deficits really related?”, Contemporary PolicyIssues, Vol. 7, pp. 91-115.

Ng, S. and Perron, P. (1995), “Unit root tests in ARMA models with data dependent methods forthe selection of the truncation lag”, Journal of the American Statistical Association, Vol. 90No. 429, pp. 268-81.

Parsley, D.C. and Popper, H.A. (2006), “Exchange rate pegs and foreign exchange exposure inEast and South East Asia”, Journal of International Money and Finance, Vol. 25,pp. 992-1009.

Pesaran, M. and Shin, Y. (1998), “Generalized impulse response analysis in linear multivariatemodels”, Economic Letters, Vol. 58 No. 1, pp. 17-29.

Piersanti, G. (2000), “Current account dynamics and expected future budget deficits: someinternational evidence”, Journal of International Money and Finance, Vol. 19 No. 2,pp. 255-71.

Rahman, M. and Mishra, B. (1992), “Cointegration of US budget and current account deficits:twin or strangers?”, Journal of Economics and Finance, Vol. 16, pp. 119-27.

Rajan, R. (2002), “Exchange rate policy options for post-crisis Southeast Asia: is there a case forcurrency basket?”, The World Economy, Vol. 25, pp. 137-63.

Rambaldi, A.N. and Doran, H. (1996), “Testing for Granger non-causality in cointegrated systemsmade easy”, Working Paper in Econometrics and Applied Statistics No. 88, Department ofEconometrics, University of New England, Armidale.

Reinhart, C.M. and Rogoff, K.S. (2002), “A modern history of exchange rate arrangements: areinterpretation”, Working Paper No. 8963, National Bureau of Economic Research.Cambridge, MA.

Reinsel, G.C. and Ahn, S.K. (1992), “Vector autoregressive models with unit roots and reducedrank structure: estimation, likelihood ratio test and forecasting”, Journal of Time SeriesAnalysis, Vol. 13, pp. 353-75.

JES34,6

474

Rosengard, J.K. (2004), “Will bank bailouts bust budgets? Fiscalisation of the East Asianfinancial crisis”, Asia-Pacific Economic Literature, Vol. 18, pp. 19-29.

Rosensweig, J.A. and Tallman, E.W. (1993), “Fiscal policy and trade adjustment: are the deficitsreally twins?”, Economic Inquiry, Vol. 31, pp. 580-94.

Said, E.S. and Dickey, D.A. (1984), “Testing for unit roots in autoregressive-moving averagemodels of unknown order”, Biometrika, Vol. 71, pp. 599-607.

Toda, H.Y. and Yamamoto, T. (1995), “Statistical inference in vector autoregressive with possiblyintegrated processes”, Journal of Econometrics, Vol. 66, pp. 225-50.

Urbain, J. (1993), Exogeneity in Error Correction Models, Springer-Verlag, Berlin.

Vamvoukas, G.A. (1999), “The twin deficits phenomenon: evidence from Greece”, AppliedEconomics, Vol. 31, pp. 1093-100.

Volcker, P.A. (1984), “Facing up to the twin deficits”, Challenge, March/April, pp. 4-9.

Warr, P.G. (1999), “What happen to Thailand?”, Working Paper in Trade and Development No. 3,The Australian National University, Canberra.

Wheeler, M. (1999), “The macroeconomic impacts of government debt: an empirical analysis ofthe 1980s and 1990s”, Atlantic Economic Journal, Vol. 27, pp. 273-84.

Zietz, J. and Pemberton, D.K. (1990), “The US budget and trade deficits: a simultaneous equationmodel”, Southern Economic Journal, Vol. 57, pp. 23-34.

Corresponding authorAhmad Zubaidi Baharumshah can be contacted at: [email protected]


deficits

475

To purchase reprints of this article please e-mail: [email protected] visit our web site for further details: www.emeraldinsight.com/reprints

dynamics of fiscal and current account deficits in thailand: an empirical investigation

Documents