Dollarization of deposits in the short and long run: evidence from CESE countries

Transcription

1 Dollarizaion of deposis in he shor and long run: evidence from CESE counries Ivana Rajković, NBS Branko Urošević, Faculy of Economics, Universiy of Belgrade and NBS Ranko Jelić, Universiy of Birmingham Skopje, April 2014

2 Agenda Inroducion Background on dollarizaion A model of deposi dollarizaion Empirical resuls Robusness checks Concluding remarks

3 Objecive & Resuls (1) People of Cenral, Easern and Souheas European (CESE) as well as many oher emerging markes ofen save in foreign currencies. For counries wih inf. argeing regime, financial dollarizaion renders radiional moneary policy less effecive. We invesigae deposi dollarizaion in CESE counries in shor and long run. Consruc a version of he Minimal Variance Porfolio model ha helps disinguish beween facors ha deermine shor run and long run dynamics of deposi dollarizaion. Tes i on a balanced panel of six CESE counries wih flexible exchange raes, wih monhly daa from January 2008 o December Error-correcion model in a dynamic panel conex in order o es if MVP drives he long-run dynamics of deposi dollarizaion; dynamic panel daa regression wih counry specific fixed effecs for modeling shor-run dynamics.

4 Objecives & Resuls (2) There exiss a coinegraion relaionship beween permanen componen of deposi dollarizaion and MVP share : In he long run deposi dollarizaion is increasing in inflaion volailiy and exchange rae pass-hrough and decreasing in volailiy of depreciaion. Transiory componen of deposi dollarizaion displays high persisence in he shor-run: In conras o long-run, inflaion and is volailiy seem o have lile effec on deposi dollarizaion in he shor-run. Insead, depreciaion and is volailiy seem o be he main drivers (hey work in opposiion o each oher).

5 Conribuion o he exising lieraure We conribue o he exising lieraure in he following ways: We find a simple way o model disincion beween he shor and long run dynamics of deposi dollarizaion. We apply a novel economeric echnique o esing he deerminans of dollarizaion based on panel coinegraion and esablish coinegraed relaionship beween he level of dollarizaion and MVP. Demonsrae ha exchange rae volailiy works in opposiion of he level of FX depreciaion in he shor run. In conras o mos of he exising lieraure, we excluded from consideraion ransacional deposis ha do no reflec opimizing behavior of agens.

6 Definiion of dollarizaion Alongside he local currency FX used as a means of paymens, sore of value or uni of accoun. Differen forms of dollarizaion: 1. Official (de iure) 2. Unofficial (de faco) (De Nicolo, Honohan & Ize (2005)) Financial dollarizaion (asse subsiuion) - use of FX as a sore of value Real dollarizaion - prices and wages indexed o FX, FX as a uni of accoun Paymens dollarizaion (currency subsiuion) - use of FX as a means of paymen Financial dollarizaion Real dollarizaion Paymens dollarizaion The paper is concerned wih financial dollarizaion, or, more precisely, deposi dollarizaion.

7 Sylized facs abou financial dollarizaion in CESE (1) Financial dollarizaion common for counries suffering from hisory of macroeconomic insabiliy (high and persisen inflaion, episodes of currency crises). Reflecs lack of confidence in domesic currency. Driven also by impor of foreign savings Counries wih underdeveloped financial markes borrow in foreign currency, since borrowing in local currency is no possible abroad (Original sin). Economies frequenly hi by sudden crises end o be more dollarized (Safe-haven effec). Posiive sides of dollarizaion: Increases financial deph Moivaes savings inside he banking secor Negaive sides of dollarizaion: Weakens moneary ransmission mechanism Balance shee effecs (currency mismach and loan defaul risk) Fear of floaing (highly dollarized counries are more prone o inervene on FX marke in order o preven sharp depreciaions)

8 Loans Sylized facs abou financial dollarizaion in CESEC Banks in he region mosly mach currency composiion of asses and liabiliies. Exposed o he currency risk mosly indirecly hrough defaul risk HUN ROM ALB CRO SRB Figure1: The share of FX-denominaed and FX-indexed ineres-baring deposis and loans of households and enerprises in CESE counries, end of April 2013 (in %) CZE POL RUS Source: BoA, CNB, CNB, MNB, NBP, NBR, CBR, NBS, auhors calculaions Deposis Posiive correlaion beween deposi and loan dollarizaion for banks in he region.

9 Relaionship beween deposi dollarizaion and depreciaion and is volailiy he case of Serbia (1) Figure 2: Deposi dollarizaion raio and exchange rae changes in Serbia from January 2004 o December 2013 (in %) 8% 6% 4% 2% 0% -2% -4% 92% 90% 88% 86% 84% 82% 80% 78% 76% 74% Source: NBS, auhors calculaions Depreciaion raes(lhs) Deposi dollarizaion (rhs)

10 Relaionship beween deposi dollarizaion and depreciaion and is volailiy he case of Serbia (1) Figure 2: Deposi dollarizaion raio and exchange rae changes in Serbia from January 2004 o December 2013 (in %) 8% 92% 6% 90% 88% 4% 86% 2% 84% 82% 0% -2% -4% Tigh managed fx rae regime Floaing Fx rae regime Managed floaing fx rae regime, frequen inervenions 80% 78% 76% 74% Source: NBS, auhors calculaions Depreciaion raes(lhs) Deposi dollarizaion (rhs)

11 Relaionship beween deposi dollarizaion and depreciaion and is volailiy he case of Serbia (2) From 2004 o igh managed exchange rae regime, low exchange rae volailiy (uncondiional volailiy of exchange rae of 0.01) and high bu relaively sable level of deposi dollarizaion (beween 89% and 91%). From 2006 o he beginning of he crisis in lae more flexible exchange rae regime (uncondiional volailiy increased o he level of 0.21), deposi dollarizaion raio sared o decrease and achieved is minimum value of 81% in December A decline in dollarizaion in 2006 and 2007 may also be explained wih exchange rae appreciaion ha was caused by speculaive acions on FX marke and huge foreign capial inflow in he previous period. Lae 2008 and he beginning of exchange rae depreciaions followed by he reurn o igh managed exchange rae regime ha resuled in a less volaile exchange rae depreciaions and increasing level of deposi dollarizaion.

12 Why do we focus on deposi dollarizaion? Some empirical findings ha increase in deposi dollarizaion likely drives loan dollarizaion (bu no necessarily he opposie) Given hisory of macroeconomic insabiliies in many CESE counies savings in foreign currency more aracive o people Ausrian banks asked by OeNB o primarily use deposis in hos counries as sources of funding loans. Thus, srucure of deposis largely deermines FX srucure of loans.

13 Relaed lieraure (1) Ize & Yeyai, Journal of Inernaional Economics, 2003 This paper uses minimum variance porfolio approach o explain long erm dollarizaion of deposis and loans. They find ha main drivers in seady sae is relaionship beween volailiy of inflaion and volailiy of real depreciaion. The model assumes ha here exiss only financial dollarizaion (no real dollarizaion as measured by he pass-hrough coefficien of exchange rae changes on prices). UIP holds

14 Relaed lieraure (2) Winkelried and Casillo, Journal of Inernaional Money and Finance, 2010 Dynamic model of deposi dollarizaion. Try o explain high dollarizaion persisence in counries ha achieved a saisfacory level of macroeconomic sabiliy (Peru and Poland). Assumpions: Agens opimize Markowiz uiliy funcion Fixed real ineres raes on HCD and variable real ineres raes on FCD ( UIP does no hold) Agens differ in heir abiliy o process available informaion The share of FCD in agens porfolio depends on he expeced excess reurn on FCD relaive o HCD and on volailiy of excess reurn. A core mechanism for dollarizaion persisence is heerogeneiy among agens, i.e. he difference in how hey exrac informaion

15 Relaed lieraure (3) Neanidis & Savva, (WP, 2009) Empirical sudy of deerminans of deposi and loan dollarizaion in he shor run for 11 ransiion economies from February 1993 o July Shor-run loan dollarizaion is mainly driven by banks maching of domesic loans and deposis (posiive correlaion beween deposi and loan dollarizaion). Depreciaion posiively affecs deposi dollarizaion, while he effec is absen in he case of loan dollarizaion. De Nicolo, Honohan & Ize, JBF 2004 Panel regression on a large sample of abou one hundred economies, yearly daa from 1990 o Average deposi dollarizaion (raio of onshore FCD o oal bank deposis) serves as dependen variable. Do no sigle ou effecs on ineres-bearing deposis. Focus on long-run drivers, include boh MVP and facors ha ener ino i such as inflaion.

16 Our model assumpions Borrows several key assumpions from I & Y: Agens maximize quadraic uiliy funcion Shor selling no allowed Agens do no hold cash Nominal ineres raes fixed during he life of he conrac In conras o I and Y: Agens can save eiher in domesic or foreign currency bu only in domesic banks UIP holds in he long, bu no in he shor run

17 The model (1)

18 Minimum Variance Porfolio Model (2) ) ( 2 ) ( 1 1 r Var c r E U ) ( ) ( H F F H r r E x r r E ), ( 2 ) ( ) ( ) ( ) ( H F H F H F F H r r r Cov x r r Var x r Var r Var Agens maximize quadraic uiliy funcion: Expeced real reurn on he deposi porfolio based on informaion available up o period Variance of porfolio reurns The share of foreign currency deposis in he porfolio: x F

19 Minimum Variance Porfolio Model (3) Opimal share of FCD: x E ( r r ) F H F* 1 1 e1 1 2 ce 1 e 1 UIP does hold in he shor run (carry rade (Menkhoff e al., JF, 2012)): E F H ( r r 1) 1 0 In he long run, UIP is expeced o hold: x MVP F* e1 1 e1

20 Daa descripion (1) Table 1: Counry coverage and deposi dollarizaion daa availabiliy Counry Daa availabiliy Number of observaions Albania 2007: :12 73 Czech Republic 1997:3 2013: Hungary 2001:5 2013: Poland 1996: : Romania 2007:1 2013:12 84 Serbia 2004:1 2013: Baseline: balanced monhly panel of 6 CESE counries for he period from February 2008 o December Robusness checks conduced based on differen samples of counries and ime periods and esimaes of exchange rae passhrough

21 Daa descripion (2) Table 2 Summary saisics of mos imporan variables from February 2008 o December 2013 ounry Deposi dollarizaion Monhly inflaion raes Monhly home currency depreciaion raes Exchange rae pass-hrough* in % in % in % in % Mean Min Max Mean Min Max Mean Min Max Mean Min Max lbania zech Republic ungary omania land rbia *Esimaed using ADL mehodology

22 Table 3: Definiion of variables Daa Descripion (3) DOL DOL_PERM DOL_TRANS DEP INF VOL_INF VOL_DEP PASS MVP Share of fx-indexed and fx-denominaed ineres-bearing deposis in oal ineres bearing deposis for households and enerprises Permanen componen of deposi dollarizaion obained using Beveridge Nelson-mehodology (log values) Transiory componen of deposi dollarizaion obained using Beveridge-Nelson mehodology (log values) Nominal depreciaion rae (differenced logarihm of nominal exchange raes) Monhly inflaion rae (differenced logarihm of CPI) Volailiy of inflaion calculaed using GARCH and EGARCH mehodology Volailiy of nominal depreciaion calculaed using GARCH and EGARCH mehodology Exchange rae pass-hrough calculaed using auoregressive disribued lag (ADL) mehodology

23 Pass-hrough esimaion Mehodology (3) In our basic model, exchange rae pass-hrough esimaed by Auoregressive Disribued Lag (ADL) mehodology: k e i i i1 Where π sands for inflaion rae, e for nominal depreciaion rae, and β is esimaed shor-erm exchange rae pass-hrough coefficien. In order o perform robusness checks we esimaed exchange rae passhrough using Kalman Filer mehodology:

24 Empirical resuls: panel uni roo ess Several panel uni roo ess are applied o es he saionariy of panels : Levin-Lin-Chu es assumes homogeneiy of all individual panels and heerogeneiy of deerminisic componen. The hypohesis of nonsaionariy of all panels is esed agains he alernaive of saionariy of all panels. Im-Pesaran-Shin es is appropriae for dynamic heerogeneous panels and is based on he average of ADF saisics calculaed for each crosssecion in he panel. I es ess he null of a uni roo in he enire panel agains he alernaive ha some panels are saionary. Fisher ype ess (combined panel uni roo es) es he hypohesis of nonsaionariy of all panels agains alernaive ha a leas one panel is saionary. Uni-roo ess for each panel conduced individually, and hen p-values from hese ess are combined o produce an overall es. Uses he inverse chi-squared, inverse normal, inverse logi ransformaions and modified version of he inverse chisquared ransformaion proposed by Choi.

25 Empirical resuls Prior o esimaion of long-run and shor-run dynamics of deposi dollarizaion, we es for he presence of uni roos in he panel daa se According o every applied crieria, DOL_PERM and MVP conain panel uni roo, while DOL_TRANS, VOL_DEP, INF, VOL_INF and DEP are saionary. Table 6: Resuls of Panel uni roo ess: DOL_PERM DOL_TRANS MVP VOL_INF VOL_DEP INF DEP Levin, Lin & Chu (0.28) -3.40*** (0.00) (0.29) -1.32* (0.09) ***(0.01) ***(0.00) -0.02*** (0.00) Im, Pesaran and Shin (0.18) ***(0.00) (0.12) -1.26* (0.10) -3.31*** (0.00) -9.95*** (0.00) ***(0.00) Fisher ype P (0.22) 36.93*** (0.00) (0.25) 69.91*** (0.00) ***(0.00) *** (0.00) ***(0.00) Z (0.19) -3.95*** (0.00) (0.13) -5.92*** (0.00) ***(0.00) *** (0.00) ***(0.00) L (0.19) -4.11*** (0.00) (0.14) -7.82*** (0.00) ***(0.00) *** (0.00) ***(0.00) Pm 0.70 (0.24) -5.10*** (0.00) 0.59 (0.28) ***(0.00) 4.79*** (0.00) 38.45*** (0.00) ***(0.00) Sample: Albania, Czech Republic, Hungary, Poland, Romania and Serbia Sample period: April 2008 April 2013 Noe: ***, **, * indicae saisical significance of 1%, 5% and 10% respecivelly

26 Esimaed shor-run dynamics of deposi dollarizaion (1) In he shor run we esed several hypohesis based on expression: wih included dynamics x E ( r r ) F H F* 1 1 e1 1 2 ce 1 e 1 H1: Transiory dollarizaion exhibis persisence, i.e. agens decisions on dollarizaion rely on pas dollarizaion raio. H2: Deposi dollarizaion is increasing in ineres rae spread beween foreign and local-currency deposis and MVP in he shor-run, i.e, ransiory dollarizaion is increasing in real ineres rae and MVP share and decreasing in volailiy of nominal depreciaion. Table 7: CESE Panel regression for shor-run dynamics of deposi dollarizaion Dependen variable: DOL_TRANS (1) (2) (3) Mehod DPD DPD DPD Variable Coefficien p-value Coefficien p-value Coefficien p-value CONST DOL_TRANS *** *** *** 0.00 DEP ** ** ** 0.04 VOL_DEP * ** ** 0.05 VOL_INF INF Wald chi 2 (3) =156.9 Wald chi 2 (3) =139.8 Wald chi 2 (3) =141.2 p-value =0.00 p-value =0.00 p-value =0.00

27 Esimaed shor-run dynamics of deposi dollarizaion - resuls Transiory componen of deposi dollarizaion exhibis high degree of persisence (i.e. high and significan coefficien of he lagged dependen variable). Inflaion and is volailiy do no have saisically significan impac on deposi dollarizaion in he shor run (equaion 1 and equaion 2). Depreciaion and is volailiy saisically significanly deermine deposi dollarizaion in he shor run. Sign of volailiy negaive: o he bes of our knowledge, firs o esablish ha volailiy of nominal exchange rae depreciaion works in opposiion wih he level of depreciaion in he shor run. Findings robus wih respec o differen specificaions of he model.

28 Long run: Panel coinegraion Since variables DOL_PERM and MVP conain uni roo, we esimae panel coinegraion relaionship for long run dynamics. Weserlund ess are applied o es for he presence of coinegraion by deermining wheher here exiss error correcion for individual panel members or for he panel as a whole. I sars from he error-correcion model where all variables in level are assumed o be I(1) and where a i is an esimae of he speed of error correcion owards he long run equilibrium: Δy i =ci+a i1 Δy i-1 +a i1 Δy i b i0 Δx i + b i1 Δx i-1 + +b ip 0Δx i-p +a i (y i-1 -b i x i-1 )+ε i Weserlund es comprises of four ess: Ga and G saisics es if a i = 0 for all i, versus H 1 a i < 0 for a leas one i. Pa and P saisics es if a i = 0 for all i versus a i < 0 for all i, which means ha here is evidence of coinegraion for he panel as a whole.

29 Esimaion of long run dynamics of dollarizaion, Panel coinegraion ess Table 8: Summary of panel coinegraion ess Tes Saisics p-value Weserlund ECM panel coinegraion ess G -2.99** 0.03 Ga ** 0.02 P -8.30*** 0.00 Pa *** 0.00 Noe: ***, **, * indicae saisical significance of 1%, 5% and 10% respecivelly Conclusion: According o all four Weserlund ess, we can rejec null hypohesis of no coinegraion beween permanen componen of dollarizaion and esimaed MVP share. New resul

30 Esimaion of panel coinegraion coefficiens (1) In order o esimae long-run coefficiens, we assume he following long-run deposi dollarizaion funcion: DOL_PERM i =c 0i +c 1i LOG(MVP) i +u i, i=1,..,6, =1,,63 Depending wheher lag-differenced explanaory variable is included in he model, wo differen error-correcion equaions are esimaed : Eq1: ΔDOL_PERM i =ϕi(dol_perm i-1 -c 0i -c 1i LOG(MVP) i )+b 11i ΔLOG(MVP) i-1 +ε i and Eq2: ΔDOL_PERM=ϕi(DOL_PERM i-1 -c 0i -c 1i LOG(MVP) i )+ ε i Coefficiens are esimaed by applying PMG and MG mehod: PMG: long run coefficiens are considered o be equal across all panels (c 1 ), while he shor run coefficiens and error variances are allowed o differ across panels. MG: coefficiens of he model are calculaed from he unweighed average of he unconsrained, fully heerogeneous model (long-run coefficiens are heerogeneous as well). Consisency of esimaors is hen esed wih he Hausman es.

31 Esimaion of panel coinegraion coefficiens (2) Firs es he model ha includes lag differenced log(mvp), ie. Eq1: ΔDOL_PERM i =ϕi(dol_perm i-1 -c 0i -c 1i LOG(MVP) i )+b 11i ΔLOG(MVP) i-1 +ε i Coefficien of lag of differenced Log(MVP) alhough posiive is no saisically significan Table 9: MG and PMG esimaion of he long-run coefficiens of he deerminans of deposi dollarizaion when lagged difference of log(mvp) is included (2008:2 2013:12) Dependen variable: DOL_PERM Mehod MG PMG Variable Coefficien p-value Coefficien p-value Log(MVP) 0. 08* ** 0.04 Error-correcion erm -0.11*** *** 0.00 d.log(mvp) Consan -0.13*** *** 0.00 Noe: ***, **, * indicae saisical significance of 1%, 5% and 10% respecivelly We esablish posiive coinegraing relaionship beween permanen componen of deposi dollarizaion and MVP ha ses I&Y claim on firmer ground

32 Esimaion of panel coinegraion coefficiens(3) Afer excluding log difference of MVP, we esimaed Eq 2: ΔDOL_PERM=ϕi(DOL_PERM i-1 -c 0i -c 1i LOG(MVP) i )+ ε i Table 10: MG and PMG esimaion of he long-run coefficiens of he deerminans of deposi dollarizaion (2008:2 2013:12) Dependen variable: DOL_PERM Mehod MG PMG Variable Coefficien p-value Coefficien p-value Log(MVP) 0.07* * 0.08 Error-correcion erm -0.10*** *** 0.00 Consan -0.12** *** 0.00 Noe: ***, **, * indicae saisical significance of 1%, 5% and 10% respecivelly Esimaed coefficiens remained of expeced sign and significance.

33 Esimaion of panel coinegraion coefficiens(4) Hausman es The consisency of esimaors is esed by applying Hausman es ha assumes ha difference in coefficiens is no sysemaic. Obained Hausman es saisics of 0.34 (p-value=0.56) suggess ha PMG esimaors are preferred o MG, since hey are consisen and efficien under he null hypohesis. Thus, here is no reason o assume ha esimaed coefficiens significanly differ across counries (same forces a play across he region).

34 Esimaion of long run dynamics of dollarizaion main resuls There exiss a posiive and saisically significan relaionship beween permanen componen of deposi dollarizaion and MVP share, which means ha volailiy of inflaion and pass-hrough posiively affec deposi dollarizaion, while volailiy of depreciaion negaively affecs i in he long run. The coefficien on he error-correcion erm across all empirical specificaions is saisically significan a he 1% level suggesing ha he seleced variables in he model show a reurn o a long-run equilibrium.

35 Concluding remarks Our resuls indicae ha policymakers ineresed in conaining and, perhaps, reversing dollarizaion need o disinguish beween shor and long run effecs. In he shor run people seem o care relaively less abou inflaion han abou exchange rae moves since UIP does no hold and: inflaion does no ener ino real ineres rae spread beween FCD and HCD real ineres rae spread is increasing in nominal depreciaion. Appreciaion of he domesic currency (resuling, also, in increased volailiy) may, a leas emporarily, reduce he level of dollarizaion. However, according o our findings, due o srong long-erm relaion beween he level of dollarizaion and MPV, in he long run inflaion argeing combined wih freely floaing exchange rae seems more desirable moneary policy if one is o conain dollarizaion Favoring local-currency deposis by offering higher ineres raes and subsequen decline in ineres rae spread may resul in lower dollarizaion of deposis only in he shor run. In he long run, when real ineres raes are equaled due o UIP condiion, only credible inflaion argeing policy combined wih floaing exchange rae will resul in lower dollarizaion raios.

36 Robusness check To es if he resuls are robus o he applied esimaion echnique in exchange rae pass-hrough, we esimae pass-hrough using Kalman Filer mehodology Table 13: MG and PMG esimaion of he long-run coefficiens of he deerminans of deposi dollarizaion (2008:2 2013:12) Dependen variable: DOL_PERM Mehod MG PMG Variable Coefficien p-value Coefficien p-value Log(MVP) 0. 05* * 0.09 Error-correcion erm *** *** 0.00 Consan -0.12** ** 0.03 Noe: ***, **, * indicae saisical significance of 1%, 5% and 10% respecivelly Hausman es of 3.01 (p-value=0.10) sugges an 10% significance ha PMG esimaors are preferred

37 Appendix Beveridge-Nelson decomposiion