Monetary and financial policies for 'de-euroization'--a case study of recent Croatian experience.

Galac, Tomislav ; Kraft, Evan

INTRODUCTION

Episodes of significant currency depreciation during the financial crisis have exposed some of the risks associated with the widespread practice of foreign currency lending in Central and Eastern Europe. The debt burden and debt-servicing burden increased steeply and suddenly in many countries (EBRD, 2010). As a consequence, the non-performing loan ratios in banks' portfolios soared (Rainer and Haiss, 2010, p. 4). To alleviate balance-sheet effects, monetary authorities in countries with higher levels of credit euroization (1) were often forced to respond to depreciation pressures with monetary measures to support the domestic currency. But many of these measures were pro-cyclical, and thus it is unclear what their total effects on the quality of banks' assets were.

Monetary policymakers face a difficult dilemma in small open highly euroized countries when simultaneously faced with exchange rate pressures and economic downturn, hardly independent events. Attempting to preserve a stable nominal exchange rate in these circumstances can be pro-cyclical; allowing the exchange rate to adjust increases balance-sheet risks, and can also turn out to be pro-cyclical. Clearly, the more rigid the exchange rate regime, the greater the risks associated with letting the nominal exchange rate adjust. Regardless of the regime type, the expected cost of exchange rate adjustment is likely to be proportional to the level of credit euroization at the time of the shock.

Regulatory limits on the direct foreign currency exposure of financial institutions that require 'covering' foreign currency denominated lending by foreign currency denominated liabilities create another policy dilemma in non-crisis times. Many highly euroized countries have suffered from high and persistent deposit euroization, typically due to a history of high inflation or other forms of political or financial instability. Policymakers in such countries often attempted to develop their credit markets by allowing or even stimulating the development of foreign currency denominated lending. So, sometimes, a choice must be made between slower financial development and a higher level of euroization. Thus, the causation between credit and deposit euroization might run both ways, possibly creating a vicious circle of total financial euroization.

On these grounds, a sensible policy recommendation for highly euroized countries would be to increase nominal exchange rate flexibility and to take steps to reduce both deposit and credit euroization levels during good times, to make the financial system more resilient in bad times. This is precisely the advice found in most recent literature on credit euroization, such as Zettelmeyer et al. (2010), Rainer and Haiss (2010), and Kokenyne et al. (2010). This literature analyzes the causes and implications of credit euroization in theory and practice, as well as policies deemed successful at de-euroization (or de-dollarization in the case of Latin American economies).

In this paper, we test the robustness of their findings on the interesting case of Croatia, one of the most euroized countries in Europe in terms of both financial sector assets and liabilities. Policymakers in Croatia have spent considerable efforts trying to prevent the adverse effects of rapid credit growth fueled by abundant capital inflows. One of the presumed adverse effects of capital inflows is financial euroization. To our knowledge, our analysis is the first in-depth country case study, based on very detailed in-house central bank data, to assess the impact of specific regulatory and macroprudential policies on financial euroization. Our analysis indicates that the monetary and financial policies implemented by the Croatian central bank in 2004-2010 had a profound impact on the evolution of both deposit and credit euroization in Croatia. However, due to the large number of intertwined central bank measures, it is difficult to assess the effect of specific measures. Unexpectedly, feedback between deposit and credit euroization, at least in the period under review, runs only from credit euroization to deposit euroization. We find strong statistical evidence that central bank measures to curb foreign borrowing and domestic credit growth by domestic banks have contributed, if not actually caused, partial financial de-euroization in 2004-2007, while crisis fighting central bank measures in 2008-2009 might have contributed to partial re-euroization.

The remainder of this paper is organized as follows. In the next section, we review the literature, focusing mostly on the recent literature on the causes of and policies to combat financial euroization in small and open emerging market economies. In the third section, we present a descriptive analysis of the history of financial euroization in Croatia. The fourth section is devoted to monetary and financial (prudential) policies implemented in Croatia in 2004-2010 to reduce the negative consequences of rapid capital inflows, and how they might have affected financial euroization. In the fifth section, we present an econometric analysis of determinants of financial euroization in Croatia, with emphasis on central bank policies, and the last section discusses how these results could be generalized to other countries.

LITERATURE REVIEW

The causes and consequences of unofficial (or spontaneous) euroization are surveyed in several recent articles, including Ize and Levy-Yeyati (2003), and Levy-Yeyati and Rey (2006). The main consequence of high credit euroization of relevance to policymakers is Foreign Currency Induced Credit Risk (FCICR)--a systematic risk of default by foreign currency borrowers without corresponding foreign currency income streams or assets adequate to hedge the exchange rate risk. This risk materializes when the domestic currency depreciates against the borrowing currency. This often happens during cyclical downturns, and measures to support the domestic currency in these circumstances are necessarily pro-cyclical. Moreover, depreciation often affects the economy asymmetrically as households and small and medium service-oriented enterprises (and others not involved in export activities) are naturally unhedged against currency risk. These empirical facts clearly make credit euroization an undesirable phenomenon from the policymakers' viewpoint.

The main adverse consequence of a high level of deposit euroization is also related to credit euroization. If banks are allowed (or required) to extend (mostly) domestic currency loans when the majority of deposits are denominated in foreign currency, the whole banking system may become exposed to direct exchange rate in the event of nominal depreciation of the domestic currency. Two solutions to this problem are often employed in practice: (1) fixing the nominal exchange rate or/and (2) imposing restrictions on banks' net-open (short) foreign currency positions. The first solution reduces the degrees of freedom of monetary policy, while the second solution either increases credit euroization or leads to underbanking of potential borrowers who are not naturally hedged against foreign currency risk.

Regarding the causes of euroization, much more has been written on deposit euroization. (2) High levels of deposit euroization are typically found in countries with history of high and volatile inflation or sharp nominal exchange rate depreciations. Scheiber and Stix (2009) provide cross-country econometric evidence that this history does not matter once the current quality of institutions is adequately accounted for in the analysis. However, their analysis confirms the positive effect of inflation volatility on deposit euroization through a more complex channel, the so called 'minimum variance portfolio dollarization ratio' (following Ize and Levy-Yeyati, 2003) stipulating that deposit euroization increases as the ratio of inflation volatility to real exchange rate volatility grows.

On credit euroization, Zettelmeyer et al. (2010) survey the existing literature and provide their own econometric evidence. They find the following main causes and catalysts of credit euroization: (1) institutional weaknesses, especially if regular cyclical fluctuations often resulted in economic crises in the past, (2) history of volatile inflation, (3) lower cost of foreign currency loans, especially when combined with a perceived state guarantee for the case of major exchange rate disturbance, (4) de facto rigid exchange rate regime, regardless of the de lure setup, (5) high accessibility of banking services, (6) cheap and accessible foreign funding for domestic banks, and (7) expensive or inaccessible instruments for hedging foreign currency risk, especially when banks' net foreign currency exposures are subject to regulatory limits and when deposit euroization is also high.

Rainer and Haiss (2010) undertake their own cross-country econometric investigation of drivers of credit euroization, and identify factors very similar to Zettelmeyer et al. (2010). They then drill down to find that (1) cheap and accessible foreign funding of domestic banks drives credit euroization, regardless of the share of foreign ownership in the banking sector, and that (2) extension of foreign currency loans to households is well-correlated with higher household deposit euroization, while the supply of foreign funding appears more important for higher levels of corporate credit euroization. Like the other authors above, they conclude that policies should be designed to curb both deposits and loans in foreign currency to reduce the level of credit euroization.

Regarding policies aimed at reducing credit euroization, Zettelmeyer et al. recommend four categories of measures: (1) reforming macroeconomic policies and institutions, primarily to build a more flexible exchange rate regime in a low inflation environment; (2) developing markets for local currency debt instruments, with the sovereign leading; (3) increasing the price of foreign currency borrowing and decreasing the scope for legal uses of foreign currency through monetary and prudential regulation; and (4) building foreign currency reserve buffers for periods of exchange rate pressures, including arranging contingency lines with the ECB and IMF.

Kokenyne et al. (2010) also address specific measures to reduce financial (both credit and deposit) euroization. Like Zettelmeyer et al., they are also strong advocates of a flexible exchange rate regime as a first step in this direction. They, however, argue that low inflation and flexible exchange rate are not sufficient of themselves, but that monetary and prudential measures aimed at making foreign currency borrowing more expensive may need to be employed initially and temporarily when financial euroization is persistent. They also explain why forcible de-euroization attempts have failed in the past, and like other authors argue strongly against them.

In addition, for countries that decide to fight credit euroization, but at the same time choose to retain their rigid exchange rate regimes, they stress the importance of building symmetrical and credible exchange rate targets. Specifically, they argue that the monetary authorities in such countries should expend equal effort on preventing trend-appreciation (which could drive credit euroization) and trend-depreciation (which could drive deposit euroization).

There are also some recent studies that focus on how central bank policies affect foreign currency lending in particular. Brzoza-Brzezina et al. (2010) estimate a panel VAR with data on four large CEE countries, and reach the conclusion that monetary policy tightening in those countries (increasing the domestic interest rate) slows down domestic currency bank credit but at the same time speeds up foreign currency borrowing from banks, that is, it contributes to currency substitution of bank credit. This study is expanded by Rosenberg and Tirpak (2008) who add additional policy variables and analyze a 10-country panel consisting of nine new EU member states and Croatia. They construct an index of central bank policies to discourage foreign currency borrowing based on a spectrum of five degrees of central bank intervention. They conclude that, in addition to the standard determinants found by others, primarily interest rate differentials and financial development, central bank policies also affect the currency composition of bank lending. However, they point out that their panel model does not capture the evolution of credit euroization in some countries in the sample very well, including Croatia.

HISTORY OF EUROIZATION IN CROATIA

In this paper, we examine credit euroization in Croatia from 2000 to 2010. To this end, we use three data sets, each with its own merits. The data source for all figures and tables in this paper is CNB in-house databases, and authors' calculations. First, the quarterly data set is based on banks' statistical/supervisory reports to the Croatian national bank and runs from 1999Q3 to 2010Q3. It provides a detailed breakdown of banks' assets and liabilities into their Croatian kuna (HRK), FX, and FX-indexed components, but contains only 45 quarterly observations. Measures of deposit and credit euroization based on these data are shown as dotted lines in Figure 1.

Second, there is a monthly data set from the same source, which runs from January 2004. The 81 monthly observations include the interesting period of central bank policy initiatives. However, this data set provides only the aggregate currency breakdown, that is, HRK, FX, and FX-indexed totals which include transaction deposits (M1), government deposits, foreign borrowings, and lending to those sectors, which should ideally be excluded from the analysis or analyzed separately. The third data set consists of monthly data from the monetary survey, starting in December 1993. It can only be used for an initial descriptive analysis of deposit euroization because this data set treats HRK loans and deposits indexed to foreign currency as pure HRK loans and deposits. This type of aggregation should not affect the measure of deposit euroization much (thick solid line in Figure 1), with the exception of the 2004-2006 period of rapid growth of HRK deposits indexed to foreign currency. However, it makes no sense at all to measure credit euroization while bunching together foreign currency-linked loans and pure HRK loans (thin solid line in Figure 1).

[FIGURE 1 OMITTED]

Euroization in Croatia in earlier periods has been studied by Sonje and Vujcic (1999), Feige et al. (2002), Kraft (2003), and Kraft and Sosic (2006) among others. Those authors argued that the prolonged period of high deposit euroization after the successful disinflation program in 1993 and the end of the war in the late 1995 was due to the deep trauma of these experiences, resulting in 'persistent' deposit euroization. This conclusion is reinforced by the findings of analyses of other financial institutions. For instance, the introduction of 'housing savings banks' in 1998 met a lukewarm welcome by households until they were allowed to link their savings to foreign currency in the second half of 1999. After that, this form of savings started growing rapidly (Tepus, 2006). Similarly, standard life-insurance contracts hardly existed in Croatia until both premiums and payouts were allowed to be linked to the euro (Stipic et al., 2009). Finally, heavy government borrowing in or linked to foreign currency throughout 1995-2010 directly contributed to credit euroization, but it probably also contributed indirectly by making foreign currency and indexed borrowing acceptable to all parties, and by failing to establish a yield curve for purely domestic currency borrowing.

Sonje and Vujcic (1999) argued that the only way to develop the domestic loan market under these circumstances was to allow foreign currency (including foreign currency-indexed) lending. More precisely, they argued that this was the only way to develop the loan market without fully exposing the entire financial sector to foreign currency risk that would automatically have occurred had loans been extended in domestic currency without indexation. And, that is precisely what the monetary authorities did: foreign currency and indexed lending, both to households and enterprises, were allowed, while banks' net-open foreign currency position has been limited by regulation since 1995. The result was a rapid financial deepening in the entire period after 1995, driven initially by FX-indexed bank lending with only a very gradual increase of pure Croatian kuna bank loans' share in the total (Figure 1).

Banks fulfilled this regulatory requirement by extending domestic currency (HRK) loans indexed to foreign currency (mostly euros, but an important pre-crisis episode of heavy Swiss Franc-indexed lending (3) was also observed). On the liability side, they mostly attracted foreign currency savings (euros, and some US dollars mostly in the Southern parts of the country) and foreign funding (primarily loans and deposits from their EU-based owners). Foreign currency loans on the asset side and foreign currency-linked deposits on the liability side have played a much smaller part, with the former usually extended to corporate clients to pay for imports, and the latter being solicited from the public as part of a regulatory arbitrage scheme which ran from 2004 to late 2006 when this regulatory gap was closed. An obvious consequence is that (when treating foreign currency-linked instruments as part of the foreign currency class) deposit and credit euroization in Croatia moves together (two dotted lines in Figure 2).

[FIGURE 2 OMITTED]

The stylized features of the evolution of financial euroization in Croatia can be read directly from Figure 1. In short, deposit euroization was extremely high in the entire post-war period until about December 2001, regardless of the measure used (between 85 % and 88%, using monetary survey data). The quarterly measure of credit euroization (treating indexed loans as foreign currency loans) shows that it too was very high in this period, hardly varying at all. The credit euroization measure based on monthly data, which treats indexed loans as pure HRK loans, shows a steeply declining trend in the period prior to 2000, likely reflecting exchange rate movements and the rise of household lending, which we suspect was almost entirely FX-indexed in this period. Unfortunately, this cannot be verified in the monthly data set prior to 2004. Household lending rose from virtually zero in the war-years to about two fifths of total banking assets at the beginning of 2000. (4)

Sometime after the peak of the euro cash conversion process in January 2002 (for more details on euro-conversion see Kraft and Sosic, 2006), a clear trend of 'de-euroization' develops, with the end-dates depending on the measure. For deposits, 'de-euroization' lasts until the third quarter of 2006 by the monthly measure and the end of 2007 by the quarterly measure. The difference can probably be explained by a change of regulation in the last quarter of 2006 equating indexed deposits with foreign currency deposits for the purpose of some important monetary measures to be described later.

For loans, de-euroization runs in two waves, also until the end of 2007. A temporary and mild reversal in the period from the second quarter of 2004 until the end of 2005 should probably be attributed to the rise of long-term loans to households which were at the time almost all denominated in foreign currency (first euros and later Swiss francs). The stock of these loans grew at a very rapid pace during this period. The rapid reduction of the credit euroization ratio that followed in 2006-2007 could probably be attributed to the governments' strategy to reduce foreign currency denominated borrowing (including that of large public enterprises), implemented rather forcefully from the start of 2006 (see Government of Croatia, 2010 and EU Commission, 2005). However, it could have also been a consequence of the introduction of longer-term pure domestic currency loans to households in that period as a response of banks to central bank measures intended to make foreign currency and indexed lending more expensive.

The last period is one of 're-euroization', when the share of foreign currency and indexed deposits rose from the low of 67% at end-2007 to 78% by the end of third quarter of 2010, and the analogous share for loans rose from 61% to 72 %. Notably, re-euroization started in early 2008, but it intensified after the Lehman Brothers failure. While a spike in exchange rate volatility in the late 2008 and early 2009 is certainly the first choice for explaining this process, it is conceivable that other factors were also important.

Most importantly, the central bank acted forcefully to release foreign currency liquidity and sterilize I-IRK liquidity in this period, attempting to stem heavy exchange rate pressure. The measures employed (for details, see Galac, 2010) effectively reduced the regulatory cost of holding foreign currency liabilities relative to HRK denominated liabilities, and these measures have not been reversed since. Thus, they could have had a direct positive effect on deposit euroization. Moreover, 2008 and 2009 were marked by massive outflows of hot money from the Croatian stock market. A point in case is open equity investment funds, which had about HRK 16 billion under management in October 2007, and only HRK 2.5 billion in February 2009 (see HANFA, 2007, 2009). While it is difficult to decompose this change into its valuation and flow components, it is likely that a non-negligible amount of this flow found its way back from equities into foreign currency savings deposits, or ended up going abroad.

Finally, the first three quarters of 2010 saw stable deposit and loan euroization measures, at levels last seen in early 2006, before their great, but temporary reduction. These may be some early signs that by that time the re-euroization process of 2008-2009 had come to its natural end.

MONETARY AND PRUDENTIAL MEASURES AND FINANCIAL EUROIZATION IN CROATIA

From the previous section, it appears probable that financial euroization in Croatia was at least in part shaped by central bank macroprudential measures. During 2002-2007, the central bank was in the business of 'leaning against the wind', trying to reduce the pace of rapid credit growth fueled by abundant foreign funding (for more details see Kraft and Galac, 2011), often in the form of FDIs flowing into the almost wholly foreign-owned Croatian banking sector. As a side effect, these measures increased the cost for banks of holding foreign currency deposits and extending foreign currency loans relative to their HRK denominated counterparts during that period, while the opposite is true for 2008-2010. Thus, they should have also affected the evolution of financial euroization in the country, from the credit supply/deposit demand side of the equation. Given the large number of macroprudential measures implemented, many of which could be considered nonstandard central bank tools, an overview of these measures is given in Table 1 (see also Figure 2), while a more detailed description is provided below.

To make the discussion easier to follow, we will group the measures according to their main purposes: measures to restrict funding and/or immobilize assets, measures to create higher liquidity buffers, measures to limit exchange rate risk on banks' balance sheets, and measures to create higher capital buffers.

We start with measures to restrict funding and/or immobilize assets. Concerned about strong deposit growth fueling a credit boom, the Croatian National Bank imposed rather high reserve requirements on deposits. During the 1990s, the rate was higher on foreign exchange deposits, but in 2000 a uniform reserve requirement (RR) for domestic and foreign currency deposits was introduced. Although the reserve requirement was a key sterilization measure, the central bank was aware that the high reserve requirement constituted a significant tax on deposits, and sought to reduce the rate when the opportunity presented itself. The rate was decreased to facilitate recovery from the 1999 recession and banking crisis in 2000-2001, and was later lowered in December 2004 and February 2006. Finally, as part of the countercyclical package during the global crisis, the central bank reduced the RR rate again in December 2008 and February 2010.

Even while it lowered the overall reserve requirement rate, the central bank found a way to use the reserve requirement to drain the domestic currency (HRK) component of liquidity from the banking system. To this end, it required that a certain portion of the reserve requirement on foreign exchange deposits be held at the central bank in the form of an HRK denominated deposit. By increasing this proportion, the central bank could drain HRK liquidity. This kuna holding requirement (KHR) was raised in several steps from only 25 % in 2003 to 75 % in early 2009, when the central bank was waging an all-out defense of the kuna in the face of strong depreciation pressures. KHR changes could have easily had an indirect negative impact on the level of deposit euroization, since KHR effectively applied HRK required reserve rules completely under the control of the central bank to a large portion of foreign currency deposits.

Another of the measures aimed to restrict funding sources was the Marginal Reserve Requirement (MRR), introduced in July 2004 as a way to slow down capital inflows. In particular, the MRR aimed at banks' foreign funding sources. The MRR initially applied only to the increase in banks' foreign liabilities, with a reserve of 24% required. The measure was extended to cover deposits or other assets of leasing companies, in an effort to close one of the significant loopholes in the credit growth reserve measures of 2003. The MRR rate was increased in numerous steps, reaching 55 % in 2006. At this rate, it seemed that foreign borrowing would provide very little profit for banks.

Foreign borrowing nonetheless continued. However, the large jump in the MRR from 40% to 55 % in early 2006, in conjunction with higher risk weights on foreign currency-linked loans to unhedged borrowers since mid-2006, to be described below, created strong incentives for banks to increase their capital rather than continue foreign borrowing. These balance-sheet adjustments should have had a negative effect on the dynamics of the widely defined deposit euroization measure (which includes capital items), at least in 2006 and 2007 when the largest increases in banks' capital were observed.

In the immediate aftermath of the failure of Lehman Brothers in September 2008, several large foreign-owned banks experienced substantial deposit withdrawals. To allow the parent banks to support their subsidiaries, the central bank removed the MRR completely in October. The parent banks did support their subsidiaries, initially with deposits and short-term loans, so that bank deposit withdrawals of October and November of 2008 were largely reversed.

Turning now to liquidity management, the foreign currency liquidity requirement (FCLR) required banks to hold higher foreign currency liquidity as a form of self-insurance for the contingency of runs on foreign currency deposits. It required the holding of liquid foreign assets with a maturity of no more than 3 months to cover the reserve requirement. In this sense, the asset side of the regulation was less onerous than an ordinary reserve requirement. In addition, the FCLR was lowered in numerous steps, and was never increased.

There were attempts to evade the FCLR. After its enactment in 2003, banks began to offer local currency deposits indexed to an exchange rate, usually the euro-kuna rate. Until the central bank altered its regulation in September 2006, these deposits, which display the same currency risk as 'pure' foreign currency deposits, were not covered by FCLR. This is the likely explanation for why the monthly measure of deposit euroization had been falling in the entire period from 2003 to 2007, while the quarterly measure, treating indexed deposits as foreign currency deposits, only fell after the third quarter of 2006. Overall, one would expect a negative effect of a higher FCLR requirement on deposit euroization.

The central bank also directly targeted credit growth. The credit growth reserve (CGR) was the first purely macroprudential measure, introduced at the beginning of 2003. Specifically, the measure defined a set of items on banks' balance sheets and certain off-balance sheet items. If the sum of these items grew more than 4% in a given quarter, the bank would be required to purchase special CNB bills paying only 0.5 % interest. The amount of bills to be purchased was twice the excess of credit growth over the 4% maximum.

The measure remained in force throughout 2003, but was withdrawn as of the beginning of 2004. It was reintroduced in 2007, with some modifications. The amount of central bank bills required to be purchased was now only 50% of the overrun. This was raised to 75% in December 2007. The credit growth reserve appeared quite successful on the surface as very few banks exceeded the credit growth limits, but it was accompanied by heavy disintermediation, with leasing portfolios and direct foreign borrowing, dominated by corporate borrowing, growing rapidly whenever CGR was in place. Thus, this measure could have had an indirect impact on credit euroization in Croatia by reducing the FX portion of the new loans granted during the periods when the measure was in effect.

The credit growth reserve was formulated in nominal terms, so by keeping the CGR in place, the central bank gave commercial banks a strong motivation to avoid currency depreciation. This may have contributed to the success of the central bank's defense of the exchange rate in the first quarter of 2009. The CGR was finally withdrawn in November 2009, when depreciation pressures had substantially receded. When the credit growth reserve initially expired at the end of 2003, it was replaced by a capital requirement for rapidly growing banks.

The capital requirement for rapidly growing banks was the first of a number of measures aimed at increasing capital buffers at banks whose credit portfolio was growing rapidly. It compelled banks whose assets and off-balance sheet items grew by more than 20 % to retain a portion of dividends, unless the bank's capital adequacy ratio exceeded a high level. In addition, the measure required banks to form general provisions for unidentified losses. Because the limit for asset growth was set at the relatively high level of 20% in 2004, and because the measure did not take effect if capital adequacy levels were high, the dividend retention provisions rarely kicked in during 2004 and 2005. However, lowering the growth limit to 15 % in 2006 made the dividend-retention measure relevant.

The general provision has a capital-like function. Thus, this measure, like other measures which induced capital-raising by banks, could have affected deposit euroization negatively, since capital is local currency denominated by definition. It also could have affected credit euroization negatively by inducing slower bank credit growth, with the effect on credit euroization akin to that of the CGR. Afterwards, as the global crisis set in, bank credit growth rates began declining, so there was a significant drop in new general reserves formed in both 2007 and 2008. Beginning in 2009 the measure was abandoned altogether, but an even stricter replacement measure was already in its place: since beginning of 2008 banks growing faster than 12 % per annum had to maintain a higher than usual minimum capital adequacy ratio. However, lower bank credit growth and capital-raising efforts before and during the crisis have made this measure non-binding for most banks.

Matching assets and liabilities by currency is a basic principle of risk management. The Croatian National Bank has imposed regulatory limits on banks' net-open foreign exchange position (NOP) since the early 1990s. By raising or lowering these limits, the CNB was able to affect credit conditions. The limit was lowered to 20% of the regulatory capital during the boom, and also there were efforts to better capture embedded options in some kinds of foreign-currency indexed loans. The limit was raised to 30 % of the regulatory capital during the crisis to give banks more flexibility to accept foreign currency deposits or loans without necessarily having to extend indexed loans.

Currency matching may lower the direct exchange rate risk faced by banks, but the practice of extending mostly foreign currency-indexed loans exposes banks' clients who are not hedged to that same risk. It is just a transfer of risk, not risk-reduction. In particular, in the case of permanent depreciation of domestic currency, not followed by a proportionate increase in incomes, this reduction of direct foreign exchange risk could easily be compensated for by an increase in the credit risk of unhedged borrowers (foreign currency-induced credit risk, or FCICR).

In Croatia, however, it became apparent that banks were not pricing the FCICR, as bank loans to unhedged borrowers and hedged borrowers bore the same interest rates. In response, the central bank adopted the prudential measure of increased risk weights for the calculation of the capital adequacy regulatory ratio for loans in or indexed to foreign currency to unhedged borrowers (FCICR risk weight add-ons). Banks were required to document whether borrowers had natural hedges, and if not, they were required to apply 25 percentage points higher risk weights for their loans. It turned out that very few borrowers were hedged, and banks were required to raise substantial amounts of capital. The central bank was able to increase these riskweights by another 25 percentage points when it felt that risks were building up in the banking system as a whole at the beginning of 2008. However, the introduction of Basel II forced the Croatian National Bank to abandon this approach to FCICR in April 2010.

The FCICR risk-weight add-ons were certainly a prudential measure intended to better align capital requirements with risk profiles at the level of individual banks. In practice, they might have had a negative effect on both deposit and credit euroization. Regarding the former, they had an impact by spurring capital raising by banks, and regarding the latter they just might have been the main driver of a sudden rise of pure HRK-denominated long-term loans in 2007. However, their possible effect on de-euroization in Croatia must have been short-lived, as it was first interrupted by the sudden re-euroization during the escalation of the global financial crisis in 2008, and then put out of effect with the change in bank risk management standards in 2010.

Altogether, the CNB's monetary and financial policies should have affected the evolution of financial euroization in Croatia through the loan supply/deposit demand equation. A glance at Figure 2 reveals that the most likely candidates for contributing to de-euroization are the MRR, after its rate was set to its maximum level of 55 %, the introduction of FCICR risk weights, before they were increased to 50 percentage points, and the inclusion of foreign currency-indexed deposits into the calculation of the FCLR requirement. If measures contributed to re-euroization in the crisis period, then the most likely candidates for this role would be the two reductions of the FCLR rate, but also increasing KHR to 75 % may have played a role. Removing the FCICR risk weights could have contributed to the halt in the re-euroization trend in 2010.

DETERMINANTS OF EUROIZATION IN CROATIA - AN ECONOMETRIC ANALYSIS

We undertake an econometric analysis of the impact of the CNB measures on financial euroization in Croatia using monthly data for the period from January 2004 until September 2010 (the second data set from the third section of this paper, providing 80 observations after differencing the data). The dependent variables of interest are overall deposit and credit euroization, dep_eur and cred_eur. They are shares of foreign currency (including indexed) deposits and loans in total bank deposits and loans, respectively. The independent variables of immediate interest are CNB instruments described in the previous section of the paper. We code these as dummy variables, since we have no ex ante reason to assume any functional form between various policy parameters and changes in financial euroization. Additional (not policy related) independent variables are motivated by the influence of uncovered interest parity (UIP) condition and the role of inflation expectations. We assume that banks set interest rates for both deposits and loans so that the short-run UIP condition can be interpreted as stipulating that any change in the euroization level would require that banks (temporarily) misprice risk premium for holding HRK denominated deposits or taking out HRK denominated loans. Put simply, a change in the euroization level would require that the relevant interest rate spread under- or over-compensates for the nominal exchange rate risk perceived by the non-banking sector.

On top of the UIP condition, in the case of a rigid exchange rate regime, inflation expectations may have a role in determining the credibility of the exchange rate target or band, and thus they should also affect the level of deposit euroization from the supply side. In the context of credit euroization, inflation expectations may also play a role if banks do not offer certain types of domestic currency loans at all (there is anecdotal evidence of such behavior in Croatia in the past, especially for important categories of long-term household loans, such as car and home loans). In this case, the (non-) existence of expected wage-to-inflation indexation may (decrease) increase the relative demand for foreign currency (including indexed) loans.

On the deposit demand (loan supply) side, provided that their exchange rate and interest rate asset-liability gaps are not too large, banks rarely have to worry about (moderate) inflation since they operate on the margin. Nevertheless, an increase in expected inflation should put upward pressure on banks' domestic currency deposit rates, and consequently on their domestic currency loan rates, other things equal. If, on the other hand, relatively high inflation is expected, especially relative to inflation in the euro area, that should also affect bank behavior through its impact on exchange rate expectations, or the assessment of the credibility of the exchange rate target in the case of a rigid exchange rate regime. A higher risk premium in the form of a spread between the interest rate on HRK deposits and non-HRK deposits (dep_spr) would then be required for holding domestic currency deposits, and a lower one in the form of a spread between the interest rate on HRK loans and non-HRK loans (cred_spr) for taking out foreign currency loans.

Given the above discussion, explanatory variables considered for modeling financial euroization in Croatia include (1) interest rate spreads (dep_spr, cred_spr); (2) proxies for financial development and foreign capital inflows in the form of the bank loans to deposits ratio (itd) and banks' foreign liabilities to deposits ratio (fltd), respectively; (5) the kuna-euro nominal exchange rate (er) and CPI inflation (infl), and their 12-month volatilities (er_uol and infl_vol); (4) dummy variables representing central bank policy measures (5) (mrr, cgr, [clr, fcicr, khr, and fxi_fclr assuming the value of 1 after September 2006 when FCLR base includes FX-indexed deposits) which alter the relative regulatory costs for banks of holding non-HRK liabilities or extending non-HRK loans against holding HRK liabilities and extending HRK loans; (5) period dummy variables for extreme observations (identified as studentized residuals greater than 2 in unreported preliminary regressions of deposit and credit euroization measures on central bank measures) on dependent variables which may or may not have a natural explanation; (6) the crisis dummy variable (crisis) taking the value of one after September 2008, and (7) two auxiliary variables--Croatia's EMBI spread (embispr) to measure external expectations about exchange rate movements, and the share of blocked in total deposits (ipo) to measure the effect of several large stock IPOs on bank deposit composition.

Combining these explanatory variables with the presumed co-movement of the deposit euroization and credit euroization discussed earlier, we begin modeling deposit euroization and credit euroization separately using single equation regression techniques, and assuming exogeneity where necessary.

Estimates of alternative models of deposit euroization are shown in Table 2. Same as before, the dependent variable (dep_eur) represents the share of foreign currency (including indexed) deposits in total bank deposits. Models D1 and D2 provide alternative accounts of how the CNB measures might have impacted deposit euroization dynamics. As shown in Table 2, these models explain much of the variation in dep_eur, and pass several standard diagnostic checks. Model D3 shows the best fit of the class of models which involve lags 0 to 3 of the 'traditional' explanatory variables (related to inflation, exchange rate, interest rates, financial development and external funding), and do not contain any of the CNB measures on the right-hand side. This latter model has a poorer in-sample fit than the previous two models, and it also fails some residual diagnostic tests. Thus, together, estimation of models D1-3 strongly suggests that the CNB measures might have been the main if not the sole drivers of the observed deposit de-euroization, and that they did not work through the standard interest rate/exchange rate channels.

In particular, Model D1 suggests that setting the MRR rate to its maximum level of 55 % could have been the trigger for de-euroization, while including the foreign currency-indexed deposits into the FCLR calculation added steam to it. However, once the CGR rate was increased to 75%, which happened simultaneously with increasing the FCICR risk weights to 50 percentage points, the re-euroization started, which then intensified after the MRR was abolished, that is, one month after the Lehman Brothers event. At the same time, while Model D2 also suggests that setting the de-euroization in motion could be credited to hiking the MRR rate to its maximum, this model credits the reductions in the FCLR rate for driving the subsequent re-euroization. Thus, the analysis conducted here confirms the casual observation made earlier that while the CNB measures appear to have had an impact on the dynamics of deposit euroization in Croatia, it is difficult to attribute this impact to particular measures.

In a similar fashion, several competing single-equation models of credit euroization are shown in Table 3. Same as before, the dependent variable (cred_eur) is calculated as the share of foreign currency (including indexed) loans in total bank loans. The models using policy dummy variables (C1 and C2) provide two slightly different and equally plausible accounts of how the CNB measures might have impacted credit euroization during the observed period. However, in contrast to the case of deposit euroization, they explain less of the variation of the dependent variable than Model C3 based on traditional variables, and do not pass the test of serial autocorrelation of the residuals. However, Model C3 may suffer from a serious form of misspecification, as it decisively fails two diagnostic tests.

Altogether, these estimates suggest that, at least in the single-equation framework, modeling credit euroization in Croatia is more problematic than modeling deposit euroization. Regarding the impact of CNB measures on credit euroization, both dummy variable models suggest that it is the introduction of the FCICR weights that triggered credit de-euroization. Also, both indicate that this impact doubled after the introduction of the CGR in 2007.

However, the proximity of the former event to the earlier setting of MRR rate to 55 %, and the latter event to the earlier introduction of foreign currency indexed deposits into the FCLR calculation leaves open the possibility that coefficients in these models reflect the impact of deposit de-euroization on credit de-euroization instead. Regarding re-euroization in the later period, the models suggest that it was triggered by either the Lehman event or by abolishing the MRR, two chronologically almost indistinguishable events, and it subsided after the CGR was abolished in late 2009.

Having estimated some reasonable single-equation models of financial euroization in Croatia during 2004-2010, we examine their robustness against relaxation of the key underlying assumption of no feedback between the deposit and credit euroization. First, we explicitly allow for the potential and expectedly positive impact of the changes in deposit euroization level on the changes in credit euroization level. We also model a potential reverse impact. One could expect the latter to be positive as well, given regulatory limits on net open foreign currency positions of banks.

We approach this robustness exercise by formulating a two-variable system of stochastic equations formed to nest the single equation models analyzed so far. The system also explicitly models the feedback between (a few lags of) dep_eur and cred_eur. We estimate the model coefficients using a 3SLS optimization routine, using lagged dependent and contemporaneous and lagged independent variables as instruments in this estimation, in order to properly assess the direction of contemporaneous impact between deposit and credit euroization changes. The first model to be examined embeds earlier Models D2, D3, C1 and C3, and its coefficient estimates are presented in Table 4 as Model DC1. Its parsimonious version, derived by successively removing from Model DC1 variables whose coefficients are associated with p-values greater than 0.25, is presented as Model DC2 in the same table.

Models DC1 and DC2 seem to tell quite a different and simpler, although perhaps counter-intuitive story about financial euroization in the period under review, than their single-equation OLS counterparts. First of all, they suggest that the 2006-2007 financial de-euroization was driven by credit de-euroization and not by deposit de-euroization. In particular, the introduction of FCICR risk weights apparently had a direct effect of stimulating relatively more pure HRK bank lending, and then this impulse was transmitted with a short lag (one month) to the deposit side of the banks' balance sheets via the asset-liability matching mechanism. As in the single equation case, in the re-euroization period 2008-2010, the reduction of the FCLR rate seems to have contributed to the speed of this process.

Thus, unlike single equation models, models DC1 and DC2 do not find any use for the changes made by the CNB to its MRR, CGR, and FCLR facilities in modeling financial euroization in Croatia. The only exception is around the months where these changes took place, for which the models confirm statistically significant single-period effects. So, including the foreign currency indexed deposits into the FCLR calculation had a large and expecredly negative impact on the deposit euroization in the first two months of the change (it also had a significant and negative impact on credit euroization in those two months, most likely due to the asset-liability matching mechanism). Significant single period effects in the expected direction are also found around the hike of the CGR rate to 75 %, and the reduction of the FCLR rate to 20%.

Regarding the independent impact of 'traditional' variables in models DC1 and DC2, the signs on the coefficients related to lagged inflation changes are all negative and mostly statistically significant, likely reflecting the fact that our data are constructed from nominal and not real money balances. The impact of nominal exchange rate changes is statistically significant only in the credit euroization equation where it is positive, thus reflecting supply factors. The cumulative impact on credit euroization of two lags of inflation volatility changes seems positive and that of three lags of nominal exchange rate volatility also appears positive, both again reflecting supply factors. On the other hand, the significant and positive effects of credit spread changes on credit euroization probably reflect demand factors.

There are two other variables initially considered for modeling financial euroization. First, financial development, measured by the ratio of loans to deposits, seems to have a significant and unexpectedly negative impact on credit euroization, and possibly also on deposit euroization. This unusual finding is probably just a reflection of statistical properties of the sample at hand, since the loan-to-deposit ratio displays a steady trend-increase throughout the estimation period, while measures of both deposit and credit euroization eventually decrease in the period as shown in Figures 1 and 2 (although this trend-decrease is blurred by a sharp decrease in 2006-2007, and almost equally sharp reversal in 2008-2009, as already analyzed). Second, dependence on external funding, measured by the proportion of external in total bank liabilities, seems to have an expected net positive impact but only on credit euroization (6).

In particular, the period of de-euroization seems to be related to the introduction of FCICR weights which directly stimulated credit de-euroization and indirectly contributed to deposit de-euroization through the asset-liability matching mechanism. Inclusion of foreign currency indexed deposits into the FCLR calculation appears highly significant only in the first two months of implementation when it had a strong negative impact on both deposit and credit euroization changes. The asset-liability matching impulse seems to run only from credit euroization to deposit euroization and not in reverse.

Summing up, single equation models of financial euroization in Croatia indicate that the CNB measures are likely to have contributed if not outright caused deposit and credit de-euroization of 2006-2007, although it is not completely clear which of the many measures implemented or altered during that short period were instrumental in producing this effect. It appears most likely that deposit de-euroization started with setting the MRR rate to 55% in January 2006, and intensified after the foreign currency indexed deposits were included in the FCLR calculation. The process reverted into reeuroization when CGR rate was increased to 75 % and FCICR weight add-ons were increased to 50%, although it is not clear why that happened, given that the initial introduction of FCICR weights is correlated with a strong negative impulse to deposit euroization.

A more plausible and statistically equally possible explanation provided by a different model is that deposit re-euroization started with lowering of the FCLR rate which effectively reduced the regulatory costs of holding non-HRK liabilities. Yet another possibility is that deposit de-euroization and subsequent re-euroization is mostly if not entirely due to asset-liability matching, and that the primary effect of the CNB measures was on credit euroization. The single equation models indicate that credit de-euroization started with the introduction of FCICR weights and intensified after CGR was introduced in early 2007. The beginning of credit re-euroization could most likely be attributed to the Lehman Brothers event.

Multiple equation models which explicitly model the asset-liability matching mechanism due to the regulatory requirement that banks keep their open foreign currency positions limited seem to favor the latter alternative. In particular, these models strongly suggest that the period of de-euroization seems to be related to the introduction of FCICR weights which directly stimulated credit de-euroization which in turn contributed to deposit de-euroization through the asset-liability matching mechanism. Inclusion of foreign currency indexed deposits into the FCLR calculation appears highly significant only in the first two months of implementation when it had a strong negative impact on both deposit and credit euroization changes. The asset-liability matching impulse seems to run only in the direction from credit euroization to deposit euroization.

CONCLUSION

Financial euroization in Croatia appears to have been a by-product of a complex political and economic environment during the last 20 years. In the first 10 years, it was most likely driven by the success of policies to develop the credit market in the presence of high currency substitution following the war years, which resulted in increasing deposit and credit euroization, and decreasing currency substitution. In the last 10 years, it appears to be strongly influenced by the monetary and financial policies to curb the negative side effects of strong capital inflows, in conjunction with banks' efforts to minimize the costs of holding liabilities through regulatory arbitrage. In this context, one should distinguish the period of high and persistent financial euroization prior to 2002, the period of partial de-euroization 2002-2007 (especially pronounced in 2006-2007), and finally the global crisis period 2008-2010 which featured partial re-euroization.

During 2004-2010, central bank measures, which had strong effects on regulatory costs, appear to have played a very important role in explaining the evolution of financial euroization in Croatia. This finding comes with the usual disclaimer related to difficulties in explicitly modeling exchange rate and inflation expectations. That notwithstanding, the marginal reserve requirement and foreign currency liquidity requirement, appear at first glance to be the key drivers of partial de-euroization in 2006-2007 (as also argued by Kokenyne et al., 2010), while the financial crisis, especially after the Lehman Brothers event, probably is to be credited with partial re-euroization in 2008-2010, with a possible contribution from the removal of the marginal reserve requirement and the lowering of the foreign currency liquidity requirement.

However, explicitly modeling the asset-liability matching mechanism due to the regulatory requirement that banks keep their open foreign currency positions limited provides an interesting alternative explanation for the dynamics of financial euroization in Croatia in 2004-2010. This model strongly suggests that the 2006-2007 de-euroization happened entirely due to the introduction of risk weight add-ons for foreign currency induced credit risk in the calculation of regulatory capital of banks. This unorthodox measure then caused relatively faster domestic currency lending, which necessitated adjustment on the liability side of banks' balance sheets to keep their asset-liability currency gaps in check. Interestingly, Rosenberg and Tirpak's (2008) model for Croatia fails precisely in the 2006-2007 period, although its index of central bank policies accounts for risk weight add-ons for foreign currency induced credit risk in the calculation of banks' regulatory capital. Perhaps the failure of this model can be attributed to the omission of the asset-liability currency matching regulation in their analysis (see Rosenberg and Tirpak, 2008, p. 11). In this alternative model, re-euroization is not clearly attributed to particular CNB measures, so it is likely driven by other, more 'fundamental', variables such as inflation, exchange rate, and interest rate spreads.

Overall, our findings should be interpreted as indicating that even within a rigid exchange rate arrangement, a wide array of vigorously implemented monetary and prudential measures can lead to a reduction in financial euroization during normal times, as also suggested by Kokenyne et al. (2010). It also appears that the policies aimed at preserving exchange rate stability in such a regime may contribute to increases in financial euroization levels during periods of strong exchange rate pressures, although they ultimately pay off when judged against a wider set of criteria (as shown by Galac, 2010).

However, it should also be noted that, while policy measures seem to have had some effect in Croatia, they were far from creating full de-euroization in the sense of the literature: euroization level, measured as foreign currency deposits in broad money, reduced by 20 percentage points and below 20% without major macroeconomic costs (see Erasmus et al., 2009, p. 9, fn 13). Likewise, the euroization level measured in terms of loans did drop by approximately 20 percentage points, but remained elevated at a minimum of 61%, suggesting that credit euroization was far from eliminated during the de-euroization episode.

Moreover, most of the measures implemented were not explicitly aimed at de-euroization, but rather they intended to meet other central bank targets, so that their signaling effect could not have been accounted for in our analysis. Thus, it remains an open question, whether central bank measures whose side-effect in the particular time-period analyzed here was to reduce the level of financial euroization in Croatia would be as effective if used to directly target de-euroization in different circumstances.

It should also be noted that financial euroization in Croatia may be qualitatively different compared to other highly euroized European nations (eg, Hungary). This could also have implications for generalizing our findings. First, historically, Croatia has a higher dependence on tourism and a high fraction of the population who have worked abroad for significant periods of time. As these mainly result in Eurozone tourist receipts and remittances, a higher proportion of households and small and medium sized enterprises may have a better natural hedge against borrowing in foreign currency in Croatia than in other highly euroized countries. At the same time, these inflows also may have had an impact on the propensity to euroize, at least historically. Also, the prevalence of foreign currency borrowing by the government itself is not generally applicable to other highly euroized countries.

REFERENCES

Brzoza-Brzezina, M, Chmielewski, T and Niedzwiedzinska, J. 2010: Substitution between domestic and foreign currency loans in Central Europe. Do central banks matter? ECB Working paper no 1187, European Central Bank, Frankfurt am Mein, April.

EBRD. 2010: Cross-border banking in the transition region: Lessons from the crisis and beyond. Speech by EBRD President, Mr. Thomas Mirow at the Joint IIF-EBRD Conference on Financial Systems in Emerging Europe, London, May.

Erasmus, L, Leichter, J and Menkulasi, J. 2009: Dedollarization in Liberia--Lessons from cross-country experience. IMF Working paper WP/09/37, International Monetary Fund, Washington DC, March.

European Commission. 2005: Croatia 2005 progress report. European Commission, Brussels, 9 November 2005, SEC (2005) 1424 {COM (2005) 561 final}.

Feige, EL, Faulend, M, Sonje, V and Sosic, V. 2002: Currency substitution, unofficial dollarization and estimates of foreign currency held abroad: The case of Croatia. In: Blejer, M and Skreb, M (eds). Financial Policies in Emerging Markets. MIT Press: Chicago, pp. 217-250.

Galac, T. 2010: The central bank as crisis manager in Croatia--A counterfactual analysis. Croatian National Bank Working paper W-27, Croatian National Bank, Zagreb, December.

Government of Croatia. 2010:2009 Pre-Accession Economic Programme. January, http://www.mfin .hr/adminmax/docs/2009%20-%20Pre-Accession % 20Economic%20Programme.pdf.

HANFA. 2007: October, Monthly Report, at http://www.ripe.hanfa.hr/repository_files/file/568/.

HANFA. 2009: February, Monthly Report, at http://www.ripe.hanfa.hr/repository_files/file/552/.

Ize, A and Levy-Yeyati, E. 2003: Financial dollarization. Journal of International Economics 59(2): 323-347.

Kokenyne, A, Ley, J and Veyrune, R. 2010: Dedollarization. IMF Working paper WP/10/188, International Monetary Fund, Washington DC, July.

Kraft, E. 2003: Monetary policy under dollarization: The case of Croatia. Comparative Economic Studies 45(3): 256-277.

Kraft, E and Galac, T. 2011: Macroprudential regulation of credit booms and busts--The case of Croatia. World Bank Policy Research Working paper no. 5772, World Bank, Washington DC, August.

Kraft, E and Sosic, V. 2006: Floating with a large life jacket: Monetary and exchange rate policies in Croatia under dollarization. Contemporary Economic Policy 24(4): 492-506.

Levy-Yeyati, E and Rey, H. 2006: Financial dollarization: Evaluating the consequences. Economic Policy 21(45): 63-118.

Rainer, W and Haiss, P. 2010: Credit euroization in central, eastern and southeastern Europe. Conference paper presented at the Annual Meeting of the Austrian Economic Association (NOeG), Vienna.

Rosenberg, CB and Tirpak, M. 2008: Determinants of foreign currency borrowing in the new member states of the EU. IMF Working paper WP/08/173, International Monetary Fund, Washington DC, July.

Scheiber, Tand Stix, H. 2009: Euroization in central, eastern and southeastern Europe--New evidence on its extent and some evidence on its causes. OeNB Working paper no. 159, Austrian National Bank, Vienna, November.

Sonje, V and Vujcic, B. 1999: Croatia in the second stage of transition 1994-1999. Croatian National Bank Working paper W-1, Croatian National Bank, Zagreb, December.

Stipic, M, Sustic, I and Krolo, Crvelin A. 2009: New product development in Croatian insurance industry. Paper presented at the International Conference on Entrepreneurship and Innovation PODIM, Maribor, March.

Tepus, MM. 2006: Analiza poslovanja stambenih sedionica: Rezultati drugoga HNB-ova projekta anketiranja stambenih stedionica. Croatian National Bank Survey P-23, Croatian National Bank.

Zettelmeyer, J, Nagy, P and Jeffrey, S. 2010: Addressing private sector currency mismatches in emerging Europe. EBRD Working paper no. 115, European Bank for Reconstruction and Development, London, June.

TOMISLAV GALAC & EVAN KRAFT

Croatian National Bank, Trg hrvatskih velikana 3, Zagreb 10000, Croatia. E-mail: [email protected]

(1) We use the term 'credit euroization' to denote the notion of dollarization of financial system assets in countries where the reference currency is euro. Analogously, we use the term "deposit euroization' to denote dollarization of deposit institutions' liabilities in those countries. Furthermore, we use the term 'financial euroization' as an umbrella term for both credit and deposit euroization, while we use the usual 'currency substitution' term to denote the situation in which euro notes and coins are used for carrying out daily transactions, rather than the domestic currency. More on terminology can be found in, for example, Ize and Levy-Yeyati (2003).

(2) Deposit euroization should be distinguished from currency substitution, a term usually reserved for describing the use of foreign currency (cash) as usual means of payment and a store of value. Currency substitution is usually found in countries with history of banking crises, especially when they resulted in forced conversion of foreign currency deposits into domestic currency or in some other form of restrictions being placed on the disposal of foreign currency deposits. Often, after a successful stabilization of the banking system such countries faced increased deposit euroization (with the proportionate decrease in currency substitution), as foreign currency cash and foreign deposits flowed back into the domestic banking system. For a description of this phenomenon in Croatia in the 1990s, see Feige et al. (2002).

(3) According to private CNB data available to us, the share of CHF denominated loans in total FX denominated loans increased from almost zero at end-2003 to peak at almost 30% in late 2007, as the CHF became the leading carry-trade currency internationally. This proportion then gradually declined thereafter after the practice of extending CHF denominated loans was completely abolished in many countries including Croatia, for well-known reasons. However, international developments in exchange rates and interest rates had much less impact on the deposit side, where the share of euro denominated deposits in total FX denominated deposits remained stable and high throughout the 2000-2010 period, in the range between 80 and 90%, with CHF in the distant second place peaking at end--2007 at 9 %. While the currency composition of FX denominated loans or deposits is not the focus of this paper, the potential for substitution between different foreign currencies should be kept in mind when interpreting the results of our study, and especially when generalizing the results for Croatia in the future or for other seemingly similar countries: Poland and Latvia come to mind as examples of EU countries where USD was an important currency in the past.

(4) The large drop in December 1999 reflects a methodological change to account for the privatization of the second largest bank in the country.

(5) Actually, these dummies are set to specific levels of CNB instruments with the same name described in Table 1. For instance 'mrr = 55' is a 0/1 dummy assuming the value of 1 whenever the MRR rate equals 55%. Note that in Figure 2 this same notation is used somewhat differently - the "MRR = 55' there marks the vertical line passing through the month in which the MRR rate was set to 55% from some previous level.

(6) We assessed the robustness of Model DC1 by replacing, one at a time, the embedded single equation models D2 and C1 with their alternatives D1 and C2, and observing the impact on the overall picture of financial f painted by the resulting three new models. Overall, the main features of Model DC1 persist over all three alternative specifications. Table 1: CNB macroprudential measures, 2002-2010 Symbol Official name of the Meaning CNB measure rr Required reserve Standard fractional reserve requirement levied on deposits khr Portion of required Sterilization tool for unwanted reserve assessed on the inflows of external bank funds basis of FX liabilities but allocated in kuna mrr Marginal reserve Additional fractional reserve requirement levied on banks' external funds srr Special reserve Additional fractional reserve requirement levied on banks' wholesale funds fclr Minimum required amount Liquidity buffer for foreign of foreign currency currency liquidity risk claims nop Limit on foreign Limit on foreign currency currency exposure asset-liability gap cgr Compulsory CNB bills Fractional reserve on excess subscription credit growth fcicr Capital adequacy risk Capital buffer for foreign weight add-ons for bank currency induced credit risk loans to borrowers with unhedged foreign currency position car Capital adequacy ratio Standard prudential requirement General banking risk Additional prudential requirement reserves for fast growing banks Dividend retention Additional prudential requirement requirement for fast growing banks Symbol Official name of the Expected impact Main purpose CNB measure on euroization rr Required reserve Restrict funding khr Portion of required - Drain liquidity, reserve assessed on the restrict funding basis of FX liabilities but allocated in kuna mrr Marginal reserve - Restrict funding requirement srr Special reserve - Restrict funding requirement fclr Minimum required amount - Liquidity buffer of foreign currency claims nop Limit on foreign Limit on-balance currency exposure sheet currency risk cgr Compulsory CNB bills - Limit credit subscription growth fcicr Capital adequacy risk - Capital buffers weight add-ons for bank loans to borrowers with unhedged foreign currency position car Capital adequacy ratio Capital buffers General banking risk - Capital buffers reserves Dividend retention - Capital buffers requirement Table 2: CNB policy measures and deposit euroization Variable Model #Dl (OLS) Coefficient p-value const. 0.001 0.578 jan. 05 -0.016 0.036 mar. 05 0.003 0.697 apr. 08 -0.025 0.002 fclr dec. to 20 0.028 0.001 fxi dep. incl. into [fclr.sub.t] -0.014 0.061 fxi dep. incl. into [fclr.sub.t-1] -0.034 0.000 cgr inc. to 75 0.020 0.014 cgr inc. to [75.sub.t-1] 0.007 0.412 [ipo.sub.t] 0.603 0.000 MRR = 55 & CGR <75 -0.012 0.000 MRR = 55 & FXI FCLR = O 0.008 0.015 MRR = 55 & CGR = 75 0.004 0.240 CGR = 75 & MRR = O 0.006 0.023 CGR = O & KHR = 75 0.003 0.402 CGR = O & KHR = 75 & FCICR = O -0.004 0.360 MRR = 55 & FCLR &gt; 28.5 MRR = 55 & FCLR &gt; 28.5 & FXI FCLR = 1 FCLR <28.5 & FCICR = 50 [infl.sub.t-1] [infl.sub.t-2] [infl.sub.t-3] [er.sub.t] [embispr.sub.t-1] [ltd.sub.t] [ltd.sub.t-3] [fltd.sub.t] Adjusted [R.sup.2] 0.671 F-statistic 11.763 Schwarz Information Criterion -6.347 Jarque-Bera Statistic (p-value) 0.276 LM serial correlation test (p-value) 0.252 Whites heteroscedasticity test 0.560 (p-value) RESET test (p-value) 0.747 Variable Model #D2 (OLS) Coefficient p-value const. 0.002 0.188 jan. 05 -0.017 0.021 mar. 05 0.002 0.751 apr. 08 -0.009 0.195 fclr dec. to 20 0.027 0.000 fxi dep. incl. into [fclr.sub.t] -0.015 0.047 fxi dep. incl. into [fclr.sub.t-1] -0.034 0.000 cgr inc. to 75 0.035 0.000 cgr inc. to [75.sub.t-1] 0.022 0.003 [ipo.sub.t] 0.601 0.000 MRR = 55 & CGR <75 MRR = 55 & FXI FCLR = O MRR = 55 & CGR = 75 CGR = 75 & MRR = O CGR = O & KHR = 75 CGR = O & KHR = 75 & FCICR = O MRR = 55 & FCLR &gt; 28.5 -0.005 0.074 MRR = 55 & FCLR &gt; 28.5 & -0.008 0.011 FXI FCLR = 1 FCLR <28.5 & FCICR = 50 0.006 0.006 [infl.sub.t-1] [infl.sub.t-2] [infl.sub.t-3] [er.sub.t] [embispr.sub.t-1] [ltd.sub.t] [ltd.sub.t-3] [fltd.sub.t] Adjusted [R.sup.2] 0.700 F-statistic 16.395 Schwarz Information Criterion -6.558 Jarque-Bera Statistic (p-value) 0.119 LM serial correlation test (p-value) 0.511 Whites heteroscedasticity test 0.549 (p-value) RESET test (p-value) 0.722 Variable Model #D3 (STEPLS) Coefficient p-value const. 0.001 0.469 jan. 05 mar. 05 apr. 08 -0.018 0.053 fclr dec. to 20 0.049 0.000 fxi dep. incl. into [fclr.sub.t] -0.032 0.001 fxi dep. incl. into [fclr.sub.t-1] -0.045 0.000 cgr inc. to 75 0.019 0.067 cgr inc. to [75.sub.t-1] [ipo.sub.t] 0.819 0.000 MRR = 55 & CGR <75 MRR = 55 & FXI FCLR = O MRR = 55 & CGR = 75 CGR = 75 & MRR = O CGR = O & KHR = 75 CGR = O & KHR = 75 & FCICR = O MRR = 55 & FCLR &gt; 28.5 MRR = 55 & FCLR &gt; 28.5 & FXI FCLR = 1 FCLR <28.5 & FCICR = 50 [infl.sub.t-1] -0.796 0.002 [infl.sub.t-2] -0.435 0.071 [infl.sub.t-3] -0.424 0.046 [er.sub.t] 0.192 0.173 [embispr.sub.t-1] 0.000 0.123 [ltd.sub.t] -0.162 0.093 [ltd.sub.t-3] -0.093 0.227 [fltd.sub.t] 0.055 0.078 Adjusted [R.sup.2] 0.566 F-statistic 8.077 Schwarz Information Criterion -6.057 Jarque-Bera Statistic (p-value) 0.825 LM serial correlation test (p-value) 0.008 Whites heteroscedasticity test 0.735 (p-value) RESET test (p-value) 0.018 Note: The dependent variable is logarithmic change in dep_eur The sample runs from February 2004 until September 2010 for the total of 80 observations. Table 3: CNB policy measures and credit euroization Variable Model #C1 (OLS) Coefficient p-value const. -0.001 0.709 jan. 05 -0.014 0.117 mar. 05 0.046 0.000 apr. 08 0.003 0.746 fclr dec. to 20 0.010 0.291 fxi dep. incl. into [fclr.sub.t] 0.009 0.320 fxi dep. incl. into [fclr.sub.t-1] -0.004 0.691 cgrinc. to 75 0.004 0.637 cgr inc. to [75.sub.t-1] -0.005 0.581 [ipo.sub.t] -0.334 0.004 MRR < 30 & CRISIS = O 0.005 0.144 FCICR = 25 & CGR < 75 -0.012 0.000 FCICR = 50 & MRR = 55 0.002 0.557 MRR = O & CRISIS = 1 0.010 0.003 MRR = O & CGR = O & CRISIS = 1 -0.008 0.042 FCICR = O & FXI_FCLR = O FCICR = 25 & CGR = O FCICR = 25 & CGR = 50 FCICR = 50 & CRISIS = O FCICR = 50 & CRISIS = 1 [infl.sub.t-1] [er.sub.t-1] [infl_vol.sub.t] [infl_vol.sub.t-1] [infl_vol.sub.t-2] [infl_vol.sub.t-3] [er_vol.sub.t-1] [er_vol.sub.t-2] [er_vol.sub.t-3] [ltd.sub.t] [ltd.sub.t-1] [fltd.sub.t] [fltd.sub.t-1] [fltd.sub.t-2] [Cred_Spr.sub.t-2] [Cred_Spr.sub.t-3] Adjusted [R.sup.2] 0.525 F-statistic 7.235 Schwarz Information Criterion -6.037 Jarque-Bera statistic (p-value) 0.368 LM serial correlation test (p-value) 0.012 Whites heteroscedasticity test 0.315 (p-value) RESET test (p-value) 0.411 Variable Model #C2 (OLS) Coefficient p-value const. 0.000 0.990 jan. 05 -0.011 0.209 mar. 05 0.043 0.000 apr. 08 0.003 0.751 fclr dec. to 20 0.011 0.242 fxi dep. incl. into [fclr.sub.t] 0.008 0.392 fxi dep. incl. into [fclr.sub.t-1] -0.004 0.666 cgrinc. to 75 0.004 0.643 cgr inc. to [75.sub.t-1] -0.005 0.587 [ipo.sub.t] -0.331 0.005 MRR < 30 & CRISIS = O FCICR = 25 & CGR < 75 FCICR = 50 & MRR = 55 MRR = O & CRISIS = 1 MRR = O & CGR = O & CRISIS = 1 FCICR = O & FXI_FCLR = O 0.001 0.751 FCICR = 25 & CGR = O -0.012 0.028 FCICR = 25 & CGR = 50 -0.013 0.005 FCICR = 50 & CRISIS = O 0.002 0.756 FCICR = 50 & CRISIS = 1 0.008 0.055 [infl.sub.t-1] [er.sub.t-1] [infl_vol.sub.t] [infl_vol.sub.t-1] [infl_vol.sub.t-2] [infl_vol.sub.t-3] [er_vol.sub.t-1] [er_vol.sub.t-2] [er_vol.sub.t-3] [ltd.sub.t] [ltd.sub.t-1] [fltd.sub.t] [fltd.sub.t-1] [fltd.sub.t-2] [Cred_Spr.sub.t-2] [Cred_Spr.sub.t-3] Adjusted [R.sup.2] 0.507 F-statistic 6.796 Schwarz Information Criterion -5.999 Jarque-Bera statistic (p-value) 0.188 LM serial correlation test (p-value) 0.020 Whites heteroscedasticity test 0.454 (p-value) RESET test (p-value) 0.508 Variable Model #C3 (STEPLS) Coefficient p-value const. 0.001 0.328 jan. 05 mar. 05 0.043 0.000 apr. 08 fclr dec. to 20 0.037 0.001 fxi dep. incl. into [fclr.sub.t] fxi dep. incl. into [fclr.sub.t-1] -0.024 0.005 cgrinc. to 75 cgr inc. to [75.sub.t-1] [ipo.sub.t] 0.217 0.198 MRR < 30 & CRISIS = O FCICR = 25 & CGR < 75 FCICR = 50 & MRR = 55 MRR = O & CRISIS = 1 MRR = O & CGR = O & CRISIS = 1 FCICR = O & FXI_FCLR = O FCICR = 25 & CGR = O FCICR = 25 & CGR = 50 FCICR = 50 & CRISIS = O FCICR = 50 & CRISIS = 1 [infl.sub.t-1] -0.372 0.079 [er.sub.t-1] 0.210 0.089 [infl_vol.sub.t] -0.031 0.013 [infl_vol.sub.t-1] -0.045 0.001 [infl_vol.sub.t-2] 0.026 0.028 [infl_vol.sub.t-3] -0.036 0.002 [er_vol.sub.t-1] 0.018 0.062 [er_vol.sub.t-2] -0.023 0.036 [er_vol.sub.t-3] 0.026 0.008 [ltd.sub.t] -0.434 0.000 [ltd.sub.t-1] 0.118 0.073 [fltd.sub.t] 0.118 0.000 [fltd.sub.t-1] -0.052 0.112 [fltd.sub.t-2] 0.057 0.040 [Cred_Spr.sub.t-2] 0.006 0.033 [Cred_Spr.sub.t-3] 0.011 0.000 Adjusted [R.sup.2] 0.617 F-statistic 7.204 Schwarz Information Criterion -6.004 Jarque-Bera statistic (p-value) 0.216 LM serial correlation test (p-value) 0.000 Whites heteroscedasticity test 0.893 (p-value) RESET test (p-value) 0.022 Note: The dependent variable is logarithmic change in cred_eur The sample runs from February 2004 until September 2010 for the total of 80 observations. Table 4: Joint models of cnb policy measures and financial euroization in croatia Variable Model DC1 dep_eur Coefficient p-value const. -0.001 0.677 jan. 05 -0.011 0.183 mar. 05 apr. 08 -0.024 0.023 fclr dec. to 20 0.019 0.176 fxi dep. incl. into [fclr.sub.t] -0.018 0.036 fxi dep. incl. into [fclr.sub.t-1] -0.026 0.006 cgr inc. to 75 0.025 0.007 cgr inc. to [75.sub.t-1] 0.017 0.036 [ipo.sub.t] 0.553 0.001 MRR = 55 & FCLR &gt; 28.5 -0.001 0.688 MRR = 55 & FCLR &gt; 28.5 & 0.000 0.925 FXI_FCLR = 1 FCLR < 28.5 & FCICR = 50 0.004 0.157 MRR < 30 & CRISIS = O FCICR = 25 & CGR < 75 MRR = O & CRISIS = 1 MRR = O & CGR = O & CRISIS = 1 [cred_eur.sub.t] 0.559 0.138 [cred_eur.sub.t-1] 0.159 0.101 [dep_eur.sub.t] [dep_eur.sub.t-1] [infl.sub.t-1] -0.323 0.152 [infl.sub.t-2] -0.332 0.101 [infl.sub.t-3] -0.341 0.057 [er.sub.t] 0.071 0.585 [er.sub.t-1] [embispr.sub.t-1] 0.000 0.550 [[infl_vol.sub.t] [infl_vol.sub.t-1] [infl_vol.sub.t-2] [infl_vol.sub.t-3] [er_vol.sub.t-1] [er_vol.sub.t-2] [er_vol.sub.t-3] [ltd.sub.t] 0.151 0.375 [ltd.sub.t-1] [ltd.sub.t-3] -0.065 0.392 [fltd.sub.t] -0.064 0.180 [fltd.sub.t-1] [fltd.sub.t-2] [cred_spr.sub.t-2] [cred_spr.sub.t-3] Adjusted [R.sup.2] 0.601 Durbin-Watson Statistic 2.384 Portmanteau residual AC test (min. p-value for lags 1-12) Lutkepohl residual Normality test (p-value) Variable Model DC1 cred_eur Coefficient p-value const. 0.003 0.026 jan. 05 0.002 0.796 mar. 05 0.041 0.000 apr. 08 fclr dec. to 20 0.027 0.039 fxi dep. incl. into [fclr.sub.t] fxi dep. incl. into [fclr.sub.t-1] cgr inc. to 75 cgr inc. to [75.sub.t-1] [ipo.sub.t] 0.149 0.470 MRR = 55 & FCLR &gt; 28.5 MRR = 55 & FCLR &gt; 28.5 & FXI_FCLR = 1 FCLR < 28.5 & FCICR = 50 MRR < 30 & CRISIS = O 0.000 0.970 FCICR = 25 & CGR < 75 -0.014 0.026 MRR = O & CRISIS = 1 0.004 0.289 MRR = O & CGR = O & CRISIS = 1 -0.004 0.300 [cred_eur.sub.t] [cred_eur.sub.t-1] [dep_eur.sub.t] -0.060 0.857 [dep_eur.sub.t-1] 0.029 0.721 [infl.sub.t-1] -0.179 0.309 [infl.sub.t-2] [infl.sub.t-3] [er.sub.t] [er.sub.t-1] 0.174 0.042 [embispr.sub.t-1] [[infl_vol.sub.t] -0.012 0.203 [infl_vol.sub.t-1] -0.018 0.167 [infl_vol.sub.t-2] 0.036 0.000 [infl_vol.sub.t-3] -0.022 0.006 [er_vol.sub.t-1] 0.012 0.092 [er_vol.sub.t-2] -0.021 0.006 [er_vol.sub.t-3] 0.011 0.181 [ltd.sub.t] -0.412 0.000 [ltd.sub.t-1] 0.071 0.173 [ltd.sub.t-3] [fltd.sub.t] 0.058 0.016 [fltd.sub.t-1] -0.067 0.047 [fltd.sub.t-2] 0.034 0.090 [cred_spr.sub.t-2] 0.005 0.008 [cred_spr.sub.t-3] 0.010 0.000 Adjusted [R.sup.2] 0.741 Durbin-Watson Statistic 2.022 Portmanteau residual AC test 0.220 (min. p-value for lags 1-12) Lutkepohl residual Normality 0.894 test (p-value) Variable Model DC2 dep_eur Coefficient p-value const. -0.002 0.109 jan. 05 -0.008 0.250 mar. 05 apr. 08 -0.020 0.004 fclr dec. to 20 0.037 0.000 fxi dep. incl. into [fclr.sub.t] -0.026 0.001 fxi dep. incl. into [fclr.sub.t-1] -0.039 0.000 cgr inc. to 75 0.031 0.000 cgr inc. to [75.sub.t-1] 0.014 0.052 [ipo.sub.t] 0.634 0.000 MRR = 55 & FCLR &gt; 28.5 MRR = 55 & FCLR &gt; 28.5 & FXI_FCLR = 1 FCLR < 28.5 & FCICR = 50 0.006 0.005 MRR < 30 & CRISIS = O FCICR = 25 & CGR < 75 MRR = O & CRISIS = 1 MRR = O & CGR = O & CRISIS = 1 [cred_eur.sub.t] [cred_eur.sub.t-1] 0.243 0.001 [dep_eur.sub.t] [dep_eur.sub.t-1] [infl.sub.t-1] -0.351 0.044 [infl.sub.t-2] -0.365 0.039 [infl.sub.t-3] -0.443 0.009 [er.sub.t] [er.sub.t-1] [embispr.sub.t-1] [[infl_vol.sub.t] [infl_vol.sub.t-1] [infl_vol.sub.t-2] [infl_vol.sub.t-3] [er_vol.sub.t-1] [er_vol.sub.t-2] [er_vol.sub.t-3] [ltd.sub.t] -0.116 0.128 [ltd.sub.t-1] [ltd.sub.t-3] [fltd.sub.t] [fltd.sub.t-1] [fltd.sub.t-2] [cred_spr.sub.t-2] [cred_spr.sub.t-3] Adjusted [R.sup.2] 0.663 Durbin-Watson Statistic 1.841 Portmanteau residual AC test (min. p-value for lags 1-12) Lutkepohl residual Normality test (p-value) Variable Model DC2 cred_eur Coefficient p-value const. 0.005 0.000 jan. 05 mar. 05 0.037 0.000 apr. 08 fclr dec. to 20 0.023 0.000 fxi dep. incl. into [fclr.sub.t] fxi dep. incl. into [fclr.sub.t-1] cgr inc. to 75 cgr inc. to [75.sub.t-1] [ipo.sub.t] MRR = 55 & FCLR &gt; 28.5 MRR = 55 & FCLR &gt; 28.5 & FXI_FCLR = 1 FCLR < 28.5 & FCICR = 50 MRR < 30 & CRISIS = O FCICR = 25 & CGR < 75 -0.016 0.000 MRR = O & CRISIS = 1 MRR = O & CGR = O & CRISIS = 1 [cred_eur.sub.t] [cred_eur.sub.t-1] [dep_eur.sub.t] [dep_eur.sub.t-1] [infl.sub.t-1] [infl.sub.t-2] [infl.sub.t-3] [er.sub.t] [er.sub.t-1] 0.162 0.051 [embispr.sub.t-1] [[infl_vol.sub.t] [infl_vol.sub.t-1] [infl_vol.sub.t-2] 0.039 0.000 [infl_vol.sub.t-3] -0.018 0.020 [er_vol.sub.t-1] 0.011 0.082 [er_vol.sub.t-2] -0.019 0.007 [er_vol.sub.t-3] 0.009 0.145 [ltd.sub.t] -0.390 0.000 [ltd.sub.t-1] [ltd.sub.t-3] [fltd.sub.t] 0.051 0.012 [fltd.sub.t-1] -0.052 0.011 [fltd.sub.t-2] 0.038 0.041 [cred_spr.sub.t-2] 0.006 0.007 [cred_spr.sub.t-3] 0.010 0.000 Adjusted [R.sup.2] 0.757 Durbin-Watson Statistic 1.978 Portmanteau residual AC test 0.551 (min. p-value for lags 1-12) Lutkepohl residual Normality 0.999 test (p-value) Note: The dependent variables are Logarithmic changes to dep_eur and cred_eur. The sample runs from April 2004 until September 2010 for the total of 78 observations.