asymmetric pass-through and risk of interest rate: an empirical exploration of taiwan and hong kong
TRANSCRIPT
This article was downloaded by: [Moskow State Univ Bibliote]On: 20 February 2014, At: 01:07Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK
Applied EconomicsPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/raec20
Asymmetric pass-through and risk of interest rate: anempirical exploration of Taiwan and Hong KongKuan-Min Wang a & Thanh-Binh Nguyen Thi ba Department of Finance , Overseas Chinese Institute of Technology , Taichung 407, Taiwanb College of Business, Feng Chia University , Taichung 407, TaiwanPublished online: 03 Jan 2008.
To cite this article: Kuan-Min Wang & Thanh-Binh Nguyen Thi (2010) Asymmetric pass-through and risk of interest rate: anempirical exploration of Taiwan and Hong Kong, Applied Economics, 42:5, 659-670, DOI: 10.1080/00036840701704444
To link to this article: http://dx.doi.org/10.1080/00036840701704444
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of theContent. Any opinions and views expressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon andshould be independently verified with primary sources of information. Taylor and Francis shall not be liable forany losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use ofthe Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
Applied Economics, 2010, 42, 659–670
Asymmetric pass-through and
risk of interest rate: an empirical
exploration of Taiwan and
Hong Kong
Kuan-Min Wanga,* and Thanh-Binh Nguyen Thib
aDepartment of Finance, Overseas Chinese Institute of Technology,
Taichung 407, TaiwanbCollege of Business, Feng Chia University, Taichung 407, Taiwan
This study employs the asymmetric threshold cointegration test suggested
by Enders and Siklos (2001) and creates asymmetric EC-EGARCH(1, 1)-
M model to investigate the pass-through of money-market rate to banking
retail rates in Taiwan and Hong Kong. It further explores the impact of
interest volatility on interest rates. Over the period of February 1988 to
December 2004, we find that the interest pass-through mechanism of these
two markets is noncomplete. In addition, based on the asymmetric
threshold cointegration test, we discover the existence of asymmetric
cointegration relationship between retail rates and market rate in both
markets. In particular, while employing asymmetric EC-EGARCH (1, 1)-
M model to test for the influence of money-market rate adjustment and
volatility on retail rates in short-run, we find robust evidence that there
exist the upward rigidity in deposit rate and the downward rigidity in
lending rate in both Taiwan and Hong Kong. This finding supports the
hypothesis of the collusive pricing arrangements. Furthermore, interest
volatility should cause a smaller margin of variation for Taiwan’s deposit/
lending rates and wider margin for Hong Kong’s lending rate.
I. Introduction
The pass-through process of money-market rate to
retail interest rates is an important link. The Central
Bank of one country could control the money-market
interest rates through its short-term policy interest
rate. Afterward, the financial institutions and market
behaviour determine the performance of this mone-
tary policy. The interest rate transmission mechanism
first controls the short-term interest rate, and then
influences the long-term interest rates or other
retail rates. De Bondt (2005) indicates that the
deposit/lending rates determined by commercial
banks should have great effect on the investments
and expenditures of depositors/borrowers, further, on
economic growth.The pass-through process of money-market rate to
retail rates differs among countries based on their
economy policy and control degree. Moreover, the
different retail rates among banks should be caused
by a variety of pass-through processes, speeds,
markup/markdown rates. As banks have their own
characteristics, depositors, maturity structures and
interest rates. Beside, disequilibrium market caused
*Corresponding author. E-mail: [email protected]
Applied Economics ISSN 0003–6846 print/ISSN 1466–4283 online � 2010 Taylor & Francis 659http://www.informaworld.com
DOI: 10.1080/00036840701704444
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
by business cycle should make retail rates reactasymmetrically to money-market rate. Accordingly,there exists an asymmetric pass-through process.
What is interest rate ‘pass-through’? Governmentadjusts policy rates according to business-cyclevolatility, which is followed by correction of money-market rate among banks (interbank-interest rate).Banks should transfer the change of money-marketrate to retail rates (including deposit and lendingrates). This process is so-called interest rate pass-through. If these cost adjustments are transferredcompletely to retail rates, which is called a completepass-through. However, banks are unable to transferthis cost immediately to retail rates because of theircontract maturities, financial structure, or operatingsystem. In general, one part of cost is borne bycustomers; the other part is passed through markup/markdown on fixed rates. Hence, the pass-through ofmoney-market rate to retail rates is not in 1:1proportion, which is called as noncomplete pass-through. And it is an over pass-through if thisproportion is >1. However, this pass-though isusually accompanied with markup/markdown onfixed rates.
The relationship of market rate and banking retailrates of either complete or noncomplete pass-throughmust be stable in long-run. In other words, there mustexist long-run cointegration relationship. One mone-tary policy considered to be effective or that does notdepend on the existence of interest pass-throughmechanism. Further, adjustment speed is a criticalinfluential factor during the adjustment of short-rundisequilibrium. This speed depends on asymmetricinformation costs that induce problems of adverseselection and moral hazard (Stiglitz and Weiss, 1981).Accordingly, the asymmetric adjustment of interestpass-through is resulted. Adjustment speed ondeposit and lending is identical if asymmetricinformation costs do not influence speed of interest-rate adjustment. Then it becomes what is calledsymmetric rate pass-through mechanism. Moreover,the markup (markdown) of interest and pass-throughparameter, because of different speeds, should resultin different degrees of pass-through.
Borio (1997) considers that Central Banks ofseveral countries implement monetary policy bymeans of control over short-term money-marketrate. Manna et al. (2001) also suggests that alterationof short-term money-market rate is an extremelyimportant step in order to execute monetary policythrough interest pass-though process. Moreover,Bredin et al. (2001) points out that the maininstrument of monetary policy is both the degree ofand speed of pass-through of short-term money-market rate to retail rates.
Focusing on reason of why the policy interest ratesare noncomplete pass-through, Cottarelli andKourelis (1994), Mojon (2000), Ehrmann et al.(2003) and Horvath et al. (2004) argue that thecompetition among banks, or among banks andnonfinancial agencies, the capitalization degree ofbank systems in each country, and the volatility ofinterest rate and so on induce the different marginand speed during pass-through process. Alternatively,Heffernan and Fuertes (2006) point to the impact ofsunk costs, borrowers/depositors’ habit and implicitcollusion among financial institutions on retail ratedetermination. Banks could select price discrimina-tion for their interest rates, making retail rates beunable to reflect the change of policy rate in shortrun.
Most of the traditional studies analyse the interestpass-through with symmetric (or linear) assumption.Thus, Central Banks are unable to effectively utilizethe policy interest rate to implement monetary policywithout the existence of this mechanism.Nevertheless, we consider that the pass-throughmechanism could actually exist but the bank retailrate adjustment, because of influence of asymmetricinformation cost, results in several forms, such asnonadjustment, slow adjustment or immediateadjustment.
There are some studies examining the existence ofmonetary mechanism and the interrelationshipbetween money and other variables with asymmetricviewpoint, such as the study of Altavilla andLandolfo (2005) that attempts to exploit whethermonetary authorities have a different behaviourduring recession and expansion. The study stronglysuggests that central banks cannot neglect the regimewhere the monetary action takes place. It follows thatthe phase of business cycle is an important matter inmonetary policy decision process. Florio (2005)discovered that the effects of positive and negativeshocks on output are statistically different from zero,and the null of symmetry between the two is rejectedin favour of negative shocks having a greater impacton real output growth, confirming an asymmetriceffect of monetary policy even for Italy. Afterward,Maki and Kitasaka (2006) use the threshold-coin-tegration test to investigate the long-run equilibriumrelationship among money, income, prices andinterest rates in Japan and provide clear evidence ofthe cointegration relationship characterized by asym-metric adjustment. The semi-structural modellingapproach that is developed by Clausen and Hayo(2006) is applied to study asymmetric monetarytransmission in Europe and find asymmetries on thedemand side in the strength of interest-rate transmis-sion and on the supply side in the effects of the output
660 K.-M. Wang and T.-B. N. Thi
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
gap on inflation. It is inferred that the financialmarket of every country has its own interest pass-through mechanism, such as the rigidity in interestadjustment, the noncomplete or complete pass-through.
The relatively dense theoretical research on rela-tionship between long-run and short-run pass-through and on interest-rate adjustment focusesmostly on analysing the long run cointegration offirst moment, adjustment of short-run error correc-tion, the traditional symmetric cointegration test.They neglect the function of short-run bias accom-panied with asymmetric adjustment. Therefore, theempirical results lead to conclusion that there is nocointegration relationship between retail rates andmarket rate. Moreover, they also ignore the long-runpass-through accompanied with asymmetry, whichcauses them to be unable effectively explain theadjustment of interest with the error-correctionmodel in short run as the adjustment of interestappears the heteroskedasticity phenomenon.
Therefore, in order to overcome the above-mentioned problems while examining the interestpass-through process, we employ the asymmetricthreshold cointegration test suggested by Enders andSiklos (2001), TAR (threshold autoregression) andMTAR (momentum-threshold autoregression) testfor whether there is an asymmetric threshold coin-tegration relationship between retail rates andmoney-market rate. To overcome the appearance ofheteroscedasticity along with interest change, withwhich the precise error correction model is not able tobe fitted, we first add the error correction term ofasymmetric adjustment to the conditional meanequation. This method allows us not only toinvestigate the short-run asymmetric adjustment,interest risk, the response of retail rates to changesin money-market rate, but also to explain the firstmoment relation of retail rates. Afterward, the secondmoment relation within EGARCH-M–model frame-work is utilized to capture the retail rate volatility andthe influence of asymmetric interest shock onvolatility, moreover, to test if model has asymmetricvolatility. Finally, we name this model asymmetricEC-EGARCH-M.
This study investigates the pass-through of money-market rate to banking-retail rates in Taiwan andHong Kong. Why the topic of interest rate pass-through is interesting? Hong Kong, at present, servesas a service platform for many Taiwanese companiesventuring into the mainland market. According to theTaipei Trade Centre in Hong Kong, there are over3000 active Taiwanese companies in the territory,with the majority having investments or otherbusiness activities in the mainland. In addition,
there are over 10 000 inactive ‘paper companies’.Among the surveyed Taiwanese companies, 80.9%indicate that they use Hong Kong banks for fundtransfers among Taiwan, the mainland and HongKong. The monetary policy, therefore, has a greatimpact on capital flows between Hong Kong andTaiwan. One of the major factors to whichGovernments, Banks and Investors play close atten-tion is policy interest rate (or official interest rate).This study aims to provide evidences of interest-ratepass-through and interest risk in these two markets,which help to evaluate and determine the effective-ness of one interest policy and help to make precisedecision on investment along with a new interestpolicy.
This article chiefly examines the interest pass-through mechanism, the short-run and long-runasymmetric interest adjustment in Taiwan andHong Kong, with main purpose of (1) investigatingif the cointegration relationship between money-market rate and bank retail rates exists? (2) consider-ing the impact of long-run asymmetric adjustment,and exploring the short-run interest adjustment withthe second-moment-based method, (3) analysing theeffect of the adjustment speed of asymmetric interestin short run, the pricing behaviour in commercialbanks’ retail rates, and the impact of asymmetricinformation on interest adjustment, (4) checking ifthe interest risk influences the adjustment of retailrates.
The article is organized as follows. Section Iintroduces the motivations and purposes of thisstudy. Section II focuses on the explanatory powerof error correction term and the asymmetric volatilityof interest. Section III describes the methodology thatis employed and presents the data and empiricalmodels. Section IV reports the summary statistics,discusses the empirical results and its economicalimplication. Finally, the article is concluded inSection V.
II. The Explanatory Power of Error-Correction Term and the AsymmetricVolatility of Interest Rate
The phenomenon of price-volatility cluster, foundmost popular in studies about price of financial assets,plays an important role in financial analysis because itimplies the impact of risk on price. Earlier studies byEngle et al. (1990) shows that the informationtransmission should influence the volatility of condi-tional variance. Nelson (1991) finds that the degree ofinformation influence on conditional variance is
Asymmetric pass-through and risk of interest rate 661
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
different if information is different. Hence, he estab-lishes the EGARCH model to explain the asymmetricphenomenon. The EGARCHmodel, therefore, is usedmost in describing the volatility-cluster phenomenonof stock price and exchange rate. To our knowledge,this is the first study to apply this model for the pass-through of money-market rate to bank retail rates.Moreover, we employ EGARCH-M to investigate theimpact of interest risk on interest rate.
Based on the above mentioned theories andempirical studies on interest, the relation betweenretail rates and money-market rate should be coin-tegration, which ensures the effectiveness of monetarypolicy. Consequently, if market price temporarilydeviates from long run equilibrium it will correctitself in the next period. Engle and Yoo (1987) andLee (1994) conclude that the error-correction termhas a great explanatory power for the conditionalmean of cointegration series. In the meanwhile, Lee(1994) applies the error-correction term in GARCHmodel. This study investigates the extension of thismodel with the asymmetric error-correction term,asymmetric volatility and the impact of interest riskon interest, which could be called EC-EGARCH-Mmodel.
III. Methodology
The data
Our analysis is based on monthly data to investi-gate the interest pass-through of money-market rateto bank retail rates in Taiwan and Hong Kong.The empirical work analyses data collected byTaiwan Economic Journal (TEJ) and InternationalFinancial Statistics (IFS), which includes informa-tion on deposit rates, lending rates and money-market rate. The description of variablesis presented in Table 1. TEJ data from February1988 to December 2004 are used for Taiwanesemarket and IFS data from January1994 to December 2004 are used for Hong Kongmarket.
Figure 1 displays the time trend of the money-
market rates, the deposit rates and the lending rates
of Taiwan and Hong Kong during study period.The interest behaviours evidently differ from
each other. It is observed that lending rates, relative
to market rates and deposit rates, are almosthighest. Money-market rates that quickly
respond to capital elasticity have higher volatilityand sometimes are even higher than the lending
rates. Taiwan’s market rates are mostly close to
lending rate while Hong Kong’s market rates,inversely, are close to deposit rate, which presents
the different behaviours of these two money
markets until 2001 when Taiwan’s market ratesbegan to be close to deposit rates. During study
period, Taiwan’s retail rates are more stable than
Hong Kong’s.
Asymmetric threshold cointegration test
While analysing time series data, determining
whether a series is stationary or not is very important,for the stationary or otherwise of a series can strongly
influence its behaviour and properties. When model-ling several unit root nonstationary time series
jointly, one may encounter the case of cointegration.
A co-integrating relationship can be seen as a long-run or equilibrium phenomenon, since it is possible
that co-integrating variables may deviate from their
relationship in the short run, but their associationwould return in the long run. Therefore, TAR and
MTAR model, suggested by Enders and Siklos
(2001), could be used for testing for the long runasymmetric cointegration.
To assume that all co-integrating ranks of variables
{y1t, . . ., ynt} are I (1). According to the cointegration
test introduced by Engle and Granger (1987), thelong-run equilibrium relationship between variables is
as follows:
y1t ¼ �_
0 þ �_
1y2t þ . . .þ �_
nynt þ et ð1Þ
In Equation (1), �i are the estimated parameters and
et is an error term. When long-run relation exists, etbecomes a stationary time series. In order to test for
Table 1. Description of variables
Country Hong Kong :HK Taiwan : TWN
Deposit rate Deposit rate :DI_HK Deposit rate: DI_TWNLending rate Lending rate: LI_HK Lending rate: LI_TWNMoney-Market rate Money-market rate: MI_HK Overnight call loan rate between banks: MI_TWN
Note: Variable DI_i denotes the deposit rate of country i, LI_i denotes the lending rate of country i, and MI_idenotes money-market rate of country i.
662 K.-M. Wang and T.-B. N. Thi
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
the cointegration between variables, we use Equation
(2) to test for unit root.
�et ¼ �et�1 þ "t ð2Þ
In Equation (2), "t is considered as a white-noise
process. There exists a long-run relationship of
symmetric stationarity if �2< �<0. In a symmetric
model, no matter what the value of et�1 is – positive
or negative– the change of et value is always �multiplied by et�1 value. Hence, the symmetric
assumption should lead to a wrong model if long-
run relationship is asymmetric stationary. To over-
come this problem, Enders and Granger (1998) and
Enders and Siklos (2001) assumed that the equili-
brium error being negative or positive is the
signal source of asymmetric adjustment. Then, the
TAR is generated to test for the existence of
asymmetric cointegration equilibrium.
�et ¼ It�1et�1 þ ð1� ItÞ�2et�1 þ "t ð3Þ
It of Equation (3) is an indicator variable,
It ¼1 if et�1 � �0 if et�1 < �
�ð4Þ
Equation (4) shows that if et�1 value is greater (or
equal) than the threshold value �, the adjustment
coefficient is �1 and adjustment margin is �1et�1 And
if et�1 value is smaller than the threshold value �, theadjustment coefficient is �2 and adjustment margin is
�2et�1 Since the actual characteristics of nonlinear
model is unknown, Enders and Siklos (2001) assume
further that �et�1, which is the first differential of
et�1, represents the momentum of interest adjust-
ment. It is also possible that an indicator variable,
which causes the interest asymmetric adjustment
under asymmetric structure. Thus, the asymmetric
TAR model could be called MTAR model. MTAR is
generated as below:
�et ¼Mt�1et�1 þ ð1�MtÞ�2et�1 þ "t ð5Þ
0
4
8
12
16
20
88 90 92 94 96 98 00 02 04
DI MI LI
TAIWAN
0
4
8
12
16
20
88 90 92 94 96 98 00 02 04
DI MI LI
HONG KONG
Fig. 1. The time trend of deposit rate (DI), money-market rate (MI), lending rate (LI)
Asymmetric pass-through and risk of interest rate 663
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
The indicator variable Mt, at this moment, is:
Mt ¼1 if �et�1 � �0 if �et�1 < �
�ð6Þ
Equation (6) demonstrates that if �et�1 value is
greater (or equal) than the threshold value �, the
adjustment coefficient is �1 and adjustment margin is
�1et�1 and if �et�1 value is smaller than the thresholdvalue �, the adjustment coefficient is �2 and adjust-
ment margin is �2et�1.In addition, if the residuals are correlated in
Equation (3) and (5), TAR and MTAR model
might be corrected as:
�et ¼ It�1et�1 þ ð1� ItÞ�2et�1 þXp
j¼1�j�et�j þ "t
ð7Þ
�et ¼Mt�1et�1 þ ð1�MtÞ�2et�1 þXp
j¼1�j�et�j þ "t
ð8Þ
In order to get {et} series of both Equations (7) and
(8) to be stationary, the abundant condition,
�2< (�1, �2)<0, must be satisfied. When {et} is
stationary and the threshold value is known, the
asymptotic distribution of parameter �1 and �2converges to a multivariate normal. Enders and
Siklos (2001) use the statistic � to test for theasymmetric threshold cointegration. And statistic �
test for the null hypothesis �1¼ �2¼ 0 with F
distribution. If the null hypothesis is rejected, then
there exists cointegration. Thus, F test could be
further utilized to investigate if the null hypothesis
(�1¼ �2) is true. If the null hypothesis of symmetric
adjustment is accepted, then the long-run relationship
between variables, resulted by Engle–Granger coin-tegration test, is symmetric. If the null hypothesis
�1¼ �2 is rejected, there, consequently, exists an
asymmetric cointegration. Hence, Engle–Granger’s
symmetric cointegration test is a special case of
Enders and Siklos (2001)’s threshold cointegration
testBeside, in order to estimate a suitable threshold
value, we employ the method of Chan (1993) toestimate the threshold value � of TAR and MTAR
model. Based on this �, we establish the indicator
variables of TAR or MTAR to proceed with
asymmetric cointegration test. Relating to threshold
critical value, see the simulation results of Wane et al.
(2004) for reference.When there exist the cointegration between
interest rates, analysing the adjustment of short-
run deviation should use the error correctionmodel. However, when the conditional heteroske-
dasticity appears in interest rates, using the error
correction model to estimate the adjustment ofshort-run deviation could result in bias.Interestingly, EC-EGARCH-M could overcome allproblems stated above.
Asymmetric EC-EGARCH(1, 1)-M model
This study constructs an univariate asymmetric EC-EGARCH(1, 1)-M model to take the conditionalmean equation of one-period lag asymmetric error-correction term into account. Firstly, the long-runnoncomplete pass-through of market rate to bankretail rates is built:
Rt ¼ d0 þ d1MIt þ et ð9Þ
Rt of Equation (9) represents bank deposit rate orlending rate. MIt is money-market rate, et is long-runerror term. d0 is defined as the fixed markup(markdown) of retail rates. It is a partial pass-through if the pass-through parameter, d1 is smallerthan 1; a complete pass-through if d1 is ¼1; and overpass-through if d1 is >1.
According to the interest pass-through explanationof Borio (1997), Bredin et al. (2001), Manna et al.(2001) and De Bondt (2005), the central bank of onecountry could execute the monetary policy withtransmission mechanism. In other words, the adjust-ment in policy interest rate affects the retail ratesthrough the market rate. The interest pass-throughmeans that the change of money-market (interbank)rate is followed by the adjustment in policy interestrate, and then banks try to shift this cost to the retailrates (deposit and lending rate). This process is chieflyconcerned with the relationship between two vari-ables, money-market rate and retail rates. We,therefore, examine the relationship of each coupleof variables–money-market rate and deposit rate,money-market rate and lending rate. We can notsimultaneously examine the relationship of threevariables because of biased result possibly caused bythe displacement or offset of exogenous effect ofmoney-market rate, or because the pass-throughprocess is from deposit rates to lending rate or viceversa.
Based on Equation (9), when there is cointegra-tion and the adjustment of asymmetric errorcorrection model is not able to get the residual oferror correction model to be a white noise, wecreate the asymmetric EC-EGARCH(1, 1)-Mmodel:
�Rt ¼ a0 þXp
i¼1ai�Rt�i þ
Xq
j¼1bj��t�j
þm�MIt þ s�t þ �1Mtet�1
þ �2ð1�MtÞet�1 þ �t�tj�t�1 � Nð0, �2t Þ ð10Þ
664 K.-M. Wang and T.-B. N. Thi
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
1: logð�2t Þ ¼ !þ �t�1�t�1
��������þ � �t�1�t�1
þ � logð�2t�1Þ ð11Þ
Equation (10) is referred to as conditional meanequation of asymmetric EC-EGARCH (1, 1)-Mmodel. The impact of interest risk (the conditionalSD �t) on interest rates is taken into consideration.The interest risk causes an increase of interestvolatility margin if s value is significant positive;and a decrease, on the contrary. �1 and �2 representthe adjustment speed of the positive and negativeerror correction term. e
_
t�1 is the error correction termof previous period long-run equilibrium. The volati-lity of retail rate adjustment, following the long-runequilibrium error, presents a positive adjustment ifthe adjustment speed coefficient is positive; and anegative adjustment if it is negative. In addition, weadd the lag autoregressive AR(p) and the moving-average MA(q) to rectify the remaining autocorrela-tion. The maximum-likelihood method is applied toestimate the asymmetric EC-EGARCH(1, 1)-Mmodel. Taking the logarithm of the maximum like-lihood function is as follows:
logL ¼ �ðT=2Þ logð2Þ � ð1=2ÞXT
t¼1logð�2t Þ
� ð1=2ÞXT
t¼1ðv2t =�
2t Þ ð12Þ
The formation of Equation (10) could be used toexamine whether the rigidity in retail rates exists.�e
_
it�1 � � implies that the adjustment of retail ratesis higher than the change of equilibrium error afterthe money-market rate changed. Therefore, theadjustment margin of retail rates must be downward.�e
_
it�1 < �, in contrast, suggests an upward adjust-ment. The variation margin of retail rates can beadjusted upward or downward via the error correc-tion term Mite
_
it�1 and ð1�MitÞe_
it�1. When �1 and �2are not equal, the adjustment of rigidity appears inretail rates. |�1|>|�2| indicates an upward rigidity inretail rates. Naturally, |�1|<|�2| indicates a down-ward rigidity in retail rates.
Equation (11) is called the conditional variableequation. The conditional variable of a logarithmensures a positive value. When � 6¼ 0, the condition
variable possesses an asymmetric effect. When �<0,the condition variable possesses a leverage effect. Themodel of this study is different from others in: (1)considering the existence of asymmetric interest pass-
through, (2) speculating on interest short-run adjust-ment that possesses heteroskedasticity phenomenon,(3) analysing the impact of interest risk on interestrates, (4) first creating asymmetric EC-EGARCH(1,1)-M and examining the adjustment of short-runinterest asymmetry.
Diagnostic checking on asymmetric EC-EGARCH(1,1)-M model
With regard to how adequate the asymmetric EC-EGARCH(1, 1)-M model is fitted, we adopt the signbias test, negative size bias test, positive size bias testetc. introduced by Engle and Ng (1993) to check if the
standard residual is an asymmetry. The diagnosticmodel is as follows:
�t�t
� �2
¼ �þ c1S�t þ c2S
�t �t�1 þ c3S
þt �t�1 þ �t ð13Þ
Where ð�t=�tÞdenotes the standard shock. �t repre-sents the error term. S�t ðS
þt Þ is the dummy variable,
which is 1 if �t�1 is smaller or greater than 0, and is 0if otherwise. The adequate fitted model must satisfythe null hypothesis H0:ci¼ 0 and the joint testH0:ci¼ c2¼ c3¼ 0.
IV. Empirical Evidence and Analysis
Unit root test and cointegration test
Table 2 reports the variable levels and the firstdifferential of Augmented Dickey Fuller(ADF) unitroot test. The lag selection is determined by the
minimum Akaike’s information criterion (AIC). Allinterest variables are I(1) series under 5% significantlevel.
The long-run parameter estimation of retail ratesis revealed in Table 3. It is discovered that, with
Table 2. ADF unit root test
Country HK TWN HK TWN
Variables Level First differentialDI_i �0.620 (0) 0.407 (2) �10.48***(0) �6.878***(1)LI_i �1.415 (3) �2.367 (1) �9.606***(0) �5.640***(3)MI_i �1.332 (2) �0.660 (8) �12.63***(1) �8.582***(7)
Notes: ADF equation includes constant, (.) is the adequate lag determined by the minimum AIC, the maximum lag is 12period.*** represent 1% significant level. Concerning the critical value, see MacKinnon (1996).
Asymmetric pass-through and risk of interest rate 665
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
parameter d0, a raise of interest in both markets is
significant with lending model; but not with deposit
model. The estimation of pass-through parameter
d1 indicates the noncomplete pass-through of both
deposit and lending model. When the pass-through
degree is below 1, the increase of bank’s financing
costs are not complete passed through to its retail
rates. The test results of TAR and MTAR
presented in Table 4 exhibit the asymmetric
cointegration of interest models in both Taiwan
and Hong Kong.
Summary statistics of interest variables
Table 5 focuses on the summary statistics of first
differential interest variables. The means of deposit
and lending rates are negative except Taiwan’s
lending rate. The SD shows that Hong Kong’s
interest risk is greater than Taiwan’s. Hong Kong’s
deposit rate risk, opposite to Taiwan’s, is greater
than its lending rate. Beside, all skewness, kurtosis
and Jarque–Bera statistics reject the normality
hypothesis.If one series is an independent and identically
distributed (i.i.d), it is still an i.i.d. after transform-
ing through any linear or nonlinear. We infer that
reason of rejecting the normality hypothesis could
be interest change accompanied with correlation.
Therefore, the interest change level and square
Ljung–Box Q-statistic(LB) test, stated in Table 5,
display the existence of autocorrelation in both
change level and square of interest variables. The
Table 5. The summary statistic of first differential of interest variable
Variable DI_HK DI_TWN LI_HK LI_TWN
Mean �0.017 �0.008 �0.030 0.002SD 0.447 0.047 0.243 0.181Skewness �0.463 �3.042 �0.031 6.179Kurtosis 9.074 24.33 8.680 62.591J-B 206.0*** 4142.9*** 225.8 31173.3LB(12) 29.47*** 47.23*** 45.17*** 56.47***LB2(12) 62.23*** 24.10** 22.20** 33.24***
Notes: Variable DI_i denotes the deposit rate of country i, LI_i denotes the lending rate of country i, and MI_i denotes themoney-market rate of country i.J-B is the statistic Jarque-Bera normal distribution test. LB(12) is Ljung-Box statistic of 12month lags of return. LB2(12) is Ljung-Box statistic of 12 month lags of square return.*** and ** respectively 1% and 5% significant level.
Table 4. TAR and MTAR cointegration test
TAR test MTAR testCountry Interest model lags � F � lags � F �
HK DI_HK 1 13.31*** 5.742(0.018)** �0.889 1 15.40*** 9.360(0.002)*** �0.410LI_HK 1 15.05*** 5.822(0.017)** �0.756 1 17.29*** 9.602(0.002)*** �0.313
TWN DI_TWN 6 6.691* 4.432(0.036)** 0.317 8 19.07*** 33.78(0.000)*** 0.081LI_TWN 5 12.63*** 15.47(0.000)*** �0.601 3 12.82*** 16.68(0.000)*** �0.127
Notes: We take the maximum lag as 12 periods. The principle of lag determination is ensuring residuals being white noise.The differential lag results via the interest model TAR and MTAR test is represented above. The values in (.) are p-value.***, ** and * respectively 1%, 5% and 10% significant level. The critical value of test statistic � is stated in table 2 of Waneet al. (2004).
Table 3. The Long-term parameter estimation
Country HK TWN HK TWN
Parameter Deposit rate model Lending rate modeld0 �0.113 (0.579) 0.510 (0.130) 4.921 (0.000) 6.712 (0.000)d1 0.855 (0.000) 0.163 (0.004) 0.627 (0.000) 0.250 (0.000)
Note: The long-term relationship equation of deposit (lending) rate model is Rt¼ d0þ d1MItþ et, (.) denotes p-value.
666 K.-M. Wang and T.-B. N. Thi
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
autocorrelation of square means the interdepen-
dence of nonlinear, which could be caused by the
conditional heteroscedasticity.
Asymmetric EC-EGARCH(1, 1)-M model
The estimations of asymmetric EC-EGARCH(1, 1)-
M are reported in Table 6. The diagnostic checking
model Q12(uh�1/2) and Q12(u
2h�1) are employed to
test for the correlation of retail rate models in each
country. The results of sign bias test, negative size
bias test, positive size bias test and joint test,
introduced by Engle and Ng (1993) show that the
standard residuals do not remain any asymmetry
effect. It proves an equate model fitted for interest in
both markets.Due to estimation of parameter m, the influence of
change in Hong Kong current money-market rates on
retail rates is positive. Moreover, the impact on the
deposit rate (0.054) is greater than on the lending rate
(0.049). The situation is quite different from Taiwan,
the impact on deposit rate is negative (�0.0002), and
on lending rate is positive (0.038). Analysing the
influence of interest volatility (risk), based on
estimation of parameter s, documents that the
impact on Hong Kong lending rate is significantly
positive and the impact on Taiwanese lending/deposit
rate is significantly negative.
Analysing the effect of the asymmetric error
correction term on retail rates could be progressed
through �1 and �2 parameter. Based on the test result
of H0:�1¼ �1, the deposit/lending model of both
markets rejects H0, which indicates that the adjust-
ment of interest in short-run disequilibrium is
asymmetry. Comparing the adjustment speed via
absolute value of �1 to �2, we discover, in both
regions, the upward rigidity adjustment in deposit
rate (|�1|>|�2|) and the downward rigidity adjust-
ment in lending rate (|�1|<|�2|). This finding supportsthe hypothesis of the collusive pricing arrangements.
Furthermore, testing for the asymmetry of condi-
tional variance through � parameter, it is found that
the asymmetry is significant in both regions.
However, � of Taiwanese lending rate model is
negative, which implies a leverage effect. The
empirical results are categorized in Table 7.
Empirical evidence and its economical implication
Why does the monetary market rate pass-through
noncompletely to retail rates? According to classical
theory, with perfect competition and sufficient
information, the price is equal marginal cost.
Therefore, the ratio of price change to marginal
cost change is ¼1. In other words, the interest pass-
through mechanism is symmetric and complete.
Table 6. Estimation of EC-EGARCH(1,1)-M model
Country HK TWN
DI_HK LI_HK DI_TWN LI_ TWN
Interest modelEstimatedvalue p-value
Estimatedvalue p-value
Estimatedvalue p-value
Estimatedvalue p-value
a1 0.229 0.013 0.494 0.000 0.102 0.000m 0.054 0.000 0.049 0.000 �0.0002 0.000 0.038 0.000s �0.009 0.747 1.214 0.000 �0.329 0.000 �0.121 0.055�1 0.044 0.005 �0.1226 0.000 �0.004 0.000 �0.010 0.023�2 0.009 0.247 0.1227 0.000 �0.002 0.000 �0.088 0.000! �0.906 0.000 �6.268 0.000 �5.699 0.000 �2.793 0.000 1.150 0.000 1.349 0.000 6.380 0.000 1.193 0.000� 0.481 0.000 0.363 0.000 5.128 0.000 �0.137 0.063� 0.979 0.000 �0.459 0.000 0.697 0.000 0.575 0.000H0: �1¼ �2 4.978 0.025 59.32 0.000 10.06 0.001 63.70 0.000Q12(uh
�1/2) 9.561 0.654 12.55 0.323 7.284 0.776 8.736 0.725Q12(u
2h�1) 5.808 0.886 15.91 0.144 1.175 1.000 3.636 0.989SB 0.643 0.279 0.928 0.129NSB 0.898 0.829 1.000 0.105PSB 0.343 0.372 0.941 0.914Joint 0.814 0.746 0.996 0.351Log L 20.16 49.70 692.2 203.3
Note: Q12(uh�1/2) and Q12(u
2h�1) represent the standardized residual and its squared Ljung–Box statistic of 12 month lags.SB,NSB and PSB denote the sign bias test, negative size bias test, and positive size bias test suggested by Engle and Ng (1993), ofwhich p-value of t-test is presented. And p-value of Chi-square test belongs to Joint test.Log L is the value of maximumlikelihood function.
Asymmetric pass-through and risk of interest rate 667
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
Table
7.Empiricalresultsandcomparison
Country
Interest
model
Degreeof
Interest
markup
(markdown)(d
0)
Degreeof
interest
pass-through(d
1)
Interest
pass-through
mechanism
(MTAR
test)
Impact
of
interest
volatility
(s)
Asymmetric
effect
of
conditional
variable
(�)
Interest
rigidity
adjustment(�
1,�
2)
Conform
ing
with
theory
HK
DI_HK
—Noncomplete
Asymmetric
—Positive
Upward
rigidity
Collusivepricing
arrangem
ents
LI_HK
markup
Noncomplete
Asymmetric
Positive
Positive
Downward
rigidity
TWN
DI_TWN
—Noncomplete
Asymmetric
Negative
Positive
Upward
rigidity
Collusivepricing
arrangem
ents
LI_TWN
markup
Noncomplete
Asymmetric
Negative
Negative(leverageeffect)
Downward
rigidity
Note:Symbol‘—
’showsthenonsignificanteffect.
668 K.-M. Wang and T.-B. N. Thi
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
Nevertheless, when market is distant from the perfectcompetition and near to monopoly, the ratio of pricechange to marginal cost change is 6¼1. There arenumber of theories proving these outcomes, such asadverse selection, switching cost, consumer irration-ality and risk sharing etc. (Lowe and Rohling, 1992).Under certain circumstances and assumptions, asshown in Horvath et al. (2004), the bank rates mighteven overreact or forego changes in the policy rate.According to one argument, bank rates can changeprior to changes in marginal funding. Banks may, forinstance, anticipate a rise in funding costs andincrease loan rates in advance. This argument hashigher relevance if banks finance longer-term loanswith shorter-term deposits, which is usually the case.
Interest volatility is defined as interest risk. Thus,interest change could get risk of one asset vary. Ourempirical evidence points out that the effect ofinterest risk on interest rate is not consistent in bothmarkets. In other words, the levels of evaluatingbank’s interest risk by Financial SupervisoryAgencies and the risk degrees reflected in bank’sinterest pricing are different. Beside, we discover theleverage effect of Taiwanese lending rate whilemeasuring the asymmetry of volatility, which meansthat the negative shock of interest adjustment isgreater than positive shock. This discovery couldpartly explain the margin of lending rate changedecreases as interest risk increases. Why does interestrates of both markets exist the asymmetric adjust-ment? According to bank behaviour, collusive pricingarrangements and reverse customer reaction, asym-metric information costs should cause the rigidityadjustment of deposit and lending rates. Evidently,Taiwan and Hong Kong conform to the theory ofcollusive pricing arrangements.
Literature comparison
Sander and Kleimeier (2002) discover, with imperfectcompetitive market and cost adjustment framework,the upward rigidity in deposit rate and the downwardrigidity in lending rate in European countries, whichis consistent with the evidence of this study. Thesefindings supports hypothesis that interest rigidity iscaused by bank’s collusive pricing arrangements.Based on data of Malaysia, Singapore, the empiricalresults of Scholnick (1996) are not able to reject thehypothesis of bank’s collusive pricing arrangements.
Additionally, Lim (2001) finds the upward rigidityadjustment in both lending and deposit rate withAustralia. Iregui et al. (2002) discover the downwardrigidity adjustment in both lending and deposit ratewith Colombia and Mexico. These two findingssupport the hypothesis of bank’s collusive pricing
arrangements. The empirical result of Iregui et al.(2002) might relate with assumption of completepass-through of Interbank Call Loan Rates to retailrates. Utilizing Johansen cointegration test, Hofmann(2002) concludes that the adjusting time of lendingrate, in France, Germany, Italy and Spain is ratherslow, which possess the characteristic of rigidityadjustment. This study uses the asymmetric cointe-gration model to examine the asymmetry of interestadjustment, which is different from Hofmann (2002)who use Johansen cointegration model with sym-metric assumption.
Recently, De Graeve et al. (2007) analyse the pass-through from market interest rates to retail bankinterest rates in Belgium. The study advocates aheterogeneous approach and applies it to the Belgianbanking market. The results find that the long-termpass-through is typically less than one-for-one,rejecting the completeness hypothesis. The situationof Belgian banking market is consistent with those ofTaiwan and Hong Kong. They also discover someevidence that deposit rates adjust faster downwardthan upward.
V. Conclusion
This study utilizes the asymmetric threshold cointe-gration test and creates asymmetric EC-EGARCH(1,1)-M model to examine the pass-through of money-market rate to retail rates and the relationshipbetween interest volatility and interest rate inTaiwan and Hong Kong. We find that there is anoncomplete pass-through in both markets. We alsodiscover the asymmetry cointegration in deposit andlending-interest model. The results estimated byasymmetric EC-EGARCH(1, 1)-M for both marketsconform with the hypothesis of collusive pricingarrangements. Moreover, according to results ofconditional variance, there exists a leverage effect inTaiwanese lending rate model.
The study makes several contributions. First, weprovide more evidences for understanding the beha-viour of interest pricing in financial market viaanalysing the interest pass-through of money-market rates to bank retail rates in Taiwan andHong Kong. Then, the capital cost and income sourceof lending/borrowing agencies or of individual areclearer. Second, the asymmetric information mightcause the asymmetric adjustment during interest pass-through process, from which derive the differentadjustment speed. Therefore, in order to timely adjustthe policy rate, the Central Bank must be deeplyconcerned about the influence of information
Asymmetric pass-through and risk of interest rate 669
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014
variation on bank’s interest adjustment and pricingmethod. Third, according to this study’s evidence,banks might have the interest risk in hand by meansof interest rigidity adjustment so as to control theircosts.
Together with economic integration day by daybetween two markets, the empirical evidence of thisstudy really provides various evaluations and deter-mination on interest policy for investors, banks andgovernments. From an academic perspective, thisarticle creates and applies the methods and modelsthat have not been used before on the same subject.Accordingly, several new evidences, that are usefulfor real world, are discovered.
References
Altavilla, C. and Landolfo, L. (2005) Do central banks actasymmetrically? Empirical evidence from the ECB andthe Bank of England, Applied Economics, 37, 507–19.
Borio, C.V. (1997) The implementation of monetary policyin industrial countries: a survey, BIS EconomicsPapers, No.47.
Bredin, D., Fitzpatrick, T. and O’Reilly, G.. (2001) Retailinterest rate pass-though: the Irish experience, CentralBank of Ireland Technical Paper, No.6.
Chan, K. S. (1993) Consistency and limiting distribution ofthe least squares estimator of a threshold autoregres-sive model, The Annals of Statistics, 21, 520–33.
Clausen, V. and Hayo, B. (2006) Asymmetric monetarypolicy effects in EMU, Applied Economics, 38,1123–134.
Cottarelli, C. and Kourelis, A. (1994) Financial structure,bank lending rates, and the transmission mechanism ofmonetary policy, IMF Staff Papers, 41, 587–623.
De Bondt, G. (2005) Interest rate pass-through: empiricalresults for the euro area, German Economic Review, 6,37–78.
De Graeve, F., De Jonghe, O. and Vander Vennet, R.(2007) Competition, transmission and bank pricingpolicies: evidence from Belgium loan and depositMarkets, Journal of Banking and Finance, 31, 259–78.
Ehrmann, M., Gambacorta, L., Martinez-Pages, J.,Sevestre, P. and Worms, A. (2003) The effects ofmonetary policy in the euro area, Oxford Review ofEconomic Policy, 19, 58–72.
Enders, W. and Granger, C. W. J. (1998) Unit-root test andasymmetric with an example using the structure ofinterest rates, Journal of Business and EconomicStatistics, 16, 304–11.
Enders, W. and Siklos, P. (2001) Cointegration andthreshold adjustment, Journal of Business andEconomic Statistics, 19, 166–76.
Engle, R. F. and Granger, C. W. J. (1987) Cointegrationand error correction: representation, estimation, andtesting, Econometrica, 55, 251–76.
Engle, R. F., Ito, T. and Lin, W. L. (1990) Meteor showersor heat waves? heteroscedastic intra-daily volatility inthe foreign exchange market, Econometrica, 58,525–42.
Engle, R. F. and Ng, V. (1993) Measuring and testing of theimpact of news on volatility, Journal of Finance, 48,1749–778.
Engle, R. F. and Yoo, B. S. (1987) Forecasting and testingin co-integrated systems, Journal of Econometrics, 35,143–59.
Florio, A. (2005) Asymmetric monetary policy: empiricalevidence for Italy, Applied Economics, 37, 751–64.
Heffernan, S. A. and Fuertes, A. (2006) Bank heterogene-ities in the interest rate transmission mechanism,Working Paper, 2006 Finance Faculty WorkingPaper Series.
Hofmann, B. (2002) The pass-through of money-marketrates to business loan rates in the euro area countries,mimeo, Center for European Integration Studies (ZEI),University of Bonn, Bonn.
Horvath, C., Kreko, J. and Naszodi, A. (2004) Interest ratepass-through in Hungary, MNB working paper.
Iregui, A. M., Milas, C. and Otero, J. (2002) On thedynamics of lending and deposit interest rates inemerging markets: a non-linear approach, Studies inNonlinear Dynamics and Econometrics, 6.
Lee, T. (1994) Spread and volatility in spot and forwardexchange rates, Journal of International Money andFinance, 13, 375–83.
Lim, G. C. (2001) Bank interest rates adjustment: are theyasymmetric?, Economic Record, 77, 135–47.
Lowe, P. and Rohling, T. (1992) Loan rate stickiness:theory and evidence, Research Discussion Paper,Reserve Bank of Australia.
Maki, D. and Kitasaka, S. (2006) The equilibrium relation-ship among money, income, prices, and interest rates:evidence from a threshold cointegration test, AppliedEconomics, 38, 1585–592.
Manna, M., Pill, H. and Quiros, G. (2001) The eurosystemsoperational framework in the context of the ECB’smonetary policy strategy, European Central Bank,mimeo.
Mojon, B. (2000) Financial structure and the interestchannel of the ECB monetary policy, ECB WorkingPaper, No. 40.
Nelson, D. B. (1991) Conditional heteroskedasticity inasset pricing: a new approach, Econometrica, 59,347–70.
Sander, H. and Kleimeier, S. (2002) Asymmetric adjust-ment of commercial bank interest rates in the euroarea: an empirical investigation into interest rate pass-through, Kredit und Kapital, 35, 161–92.
Scholnick, B. (1996) Asymmetric adjustment of commercialbank interest rates: evidence from Malaysia andSingapore, Journal of International Money andFinance, 15, 485–96.
Stiglitz, J. E. and Weiss, A. (1981) Credit rationing inmarkets with imperfect information, AmericanEconomic Review, 71, 917–26.
Wane, A., Gilbert, S. and Dibooglu, S. (2004) Criticalvalues of the empirical F-distribution for threshold,Calendar Year 2004 Discussion Papers for theDepartment of Economics Southern IllinoisUniversity at Carbondale, Illinois.
Winker, P. (1999) Sluggish adjustment if interest rates andcredit rationing: an application of unit root testing anderror correction modelling, Applied Economics, 31,267–77.
670 K.-M. Wang and T.-B. N. Thi
Dow
nloa
ded
by [
Mos
kow
Sta
te U
niv
Bib
liote
] at
01:
07 2
0 Fe
brua
ry 2
014