How do euro deposits in CESEE react to exchange rate shocks? (OeNB Bulletin Q1/25)

Nico Petz, Thomas Scheiber, Julia Wörz ¹

JEL classification: C32, E41, F31, F41

Keywords: deposit euroization, exchange rate shocks, vector autoregression, time-varying parameter regression, Bayesian estimation

The use of foreign currencies as secondary currencies and safe-haven assets has a long history in Central, Eastern and Southeastern Europe (CESEE). High inflation and hyperinflation episodes as well as banking sector turbulences in the form of currency and banking crises in the early stages of transition have made foreign currencies attractive to individuals and firms. ² Despite substantial macroeconomic stabilization since the early turbulent days of catching-up, foreign currencies – and in particular the euro – have remained attractive in the region, leading to a considerable degree of euroization in many CESEE countries.

The long-standing literature on euroization initially started by focusing on the use of the US dollar, i.e. dollarization. In general, both dollarization and euroization refer to “the use by the residents of one country of assets (or liabilities) denominated in another country’s currency” (Kokenyne et al., 2010, p. 4.). While the terms are used interchangeably in the literature, in this paper, we will use “euroization” as our research focuses on the CESEE region, where the euro has been superseding the respective local currencies. While often unavoidable, (excessive) euroization is widely considered an unfavorable phenomenon as it has a number of undesirable consequences: A high share of foreign currency loans increases vulnerabilities in financial stability by adding an element of (clustered) exchange rate risk to credit risk (de Nicolo et al., 2005; Rancière et al., 2010) and a high share of foreign currency in monetary aggregates weakens monetary policy transmission (Levy Yeyati, 2006).

The literature distinguishes between three types of euroization: currency euroization (with the euro being used as a means of payment in domestic transactions), real euroization (with domestic transactions being indexed to the exchange rate of the euro) and financial euroization (with assets and liabilities being held in euro instead of the local currency) (see e.g. de Nicolo et al, 2003; Manjani, 2015). In this paper, we focus on deposit euroization, i.e. the asset side of financial euroization, that is the use of euro deposits ³ as a safe-haven asset.

The determinants of deposit euroization are manifold. Khvedchuk et al. (2019) identify three categories of euroization determinants. First, macroeconomic factors play a role in the extensive usage of a foreign currency: It is expectations of high inflation and exchange rate volatility that predict higher levels of euroization (rather than current exchange rates or persistent inflation; see, originally, Ize and Yeyati, 2003). Similarly, experiences of past high inflation are likely to result in extensive euroization. The literature on euroization further emphasizes the role of minimum variance portfolio (MVP) motives for euroization in the long run. This means that in the long run, relative volatilities of inflation and nominal exchange rates influence decisions on the currency portfolio. In the short run, the interest rate differential between local and foreign currency deposits (i.e. the real interest rate spread) acts as a driver of euroization dynamics (see, among others, Rajković and Urošević, 2017).

Second, structural factors – especially low-quality governance and low macroeconomic policy credibility – are harbingers of euroization (Honig, 2009; de Nicolo et al., 2005). In the same vein, Brown and Stix (2015) argue that residents’ distrust in the stability of the local currency (as a result of a country’s policy quality) is an important motive for deposit euroization. Behavioral economic factors and path dependency also play a role: Euroization has become a habit (in the CESEE region) that improving governance quality will not necessarily be able to overwrite soon, as hysteresis effects and memories of past crisis episodes can be sticky.

Third, prudential regulations influence euroization trends, albeit controversies exist in the literature. Khvedchuk et al. (2019) argue that favoring the local currency vis-à-vis the euro is a functional means of promoting de-euroization. In contrast, Brown and Stix (2015) state that deposit euroization is by and large demand driven, which means that supply-side interventions (such as bank regulations favoring the local currency) may prove to be inefficient. ⁴

Overall, the drivers behind rather high and persistent levels of euroization have remained somewhat inconclusive so far. Further, households and nonfinancial corporations (NFCs) are likely to react to different signals. In addition, their sensitivity to changing signals may differ structurally and over time. While MVP motives may affect households more strongly, NFCs – especially exporters – may pay more attention to exchange rate movements. In this respect, special attention should be paid to the effects of unanticipated exchange rate movements.

In a recent article, Scheiber and Wörz (2024) reviewed the prevalence and drivers of euro deposit holdings in ten CESEE economies over the last 25 years. The authors confirmed notable differences between countries and sectors but also across different time periods. In particular, the impact of the interest rate differential on the currency composition of household deposits seems to vary across episodes, highlighting a rather rapid adjustment of household portfolios to different macroeconomic circumstances in line with the literature. Fluctuations in NFC deposits appear to be less affected by interest rate spreads and portfolio considerations but to correlate more often with exchange rate movements, which likely reflects valuation effects.

In this paper, we specifically zoom in on unanticipated exchange rate movements and their impact on households’ and NFCs’ demand for euro deposits. Moreover, we examine whether the sensitivity of changes in euro deposits to exogenous exchange rate shocks varied over the last 20 years. Hence, we restrict our attention to CESEE countries with flexible exchange rates. Our results vary significantly across countries, sectors and time periods. In general, our findings indicate that changes in NFCs’ euro deposits show stronger sensitivity to exchange rate shocks than changes in households’ euro deposits. Moreover, the sensitivity of NFC deposits shows greater time variation than that of households’ euro deposits. In line with expectations, this suggests that currency choices by NFCs are more responsive to unanticipated exchange rate movements than those made by households.

This paper is structured as follows: Section 1 motivates the research questions. In section 2, we estimate a VAR model, identify structural shocks using sign restrictions and compute impulse response functions to an exogenous exchange rate shock. In section 3, we use a time-varying parameter (TVP) model in order to obtain the time-varying sensitivity of euro deposits to exogenous movements in the nominal exchange rate. We do this for each country and separately for households and NFCs, given assumed differences in the sensitivity to exchange rate shocks. Section 4 concludes.

1 Motivation of research questions

The main objective of this paper is to shed light on the following two research questions: Do unanticipated exchange rate movements explain swings in euro deposits of households and NFCs in CESEE? And if they do, did the sensitivity of euro deposits to exogenous exchange rate shocks vary over the last 20 years?

In our analysis, we focus on six CESEE countries with floating exchange rates, namely Czechia, Croatia, Hungary, Poland, Romania and Serbia. In general, deposit euroization in CESEE has declined since the early 2000s, especially in Croatia, Poland and Hungary (see chart 1). Yet, households in Croatia continued to hold around 60% of their total deposits in euro on the eve of euro adoption. In Serbia, euroization continues to be high. We observe a peak in the share of euro deposits at around 90% of total household deposits between the mid-2000s and mid-2010s, before the deposit euroization level receded somewhat to 70%. The share of euro deposits in total NFC deposits is smaller but remained at 40% in 2023. Romania also exhibits a rather high level of deposit euroization, reaching a relatively stable share of 40% for household deposits and recording a decline to less than 30% for NFC deposits. While in Croatia, Romania and Serbia, deposit euroization has been notably higher in the household sector, Czechia, Hungary and Poland show higher deposit euroization levels in the NFC sector. Overall, the share of euro deposits in total deposits is lower in these three countries, with Hungary posting the highest levels of between 35% to 40% for NFC deposits.

Many CESEE countries were building on macroeconomic stabilization policies, i.e. the return to sustainable fiscal policies and a sound banking sector. This should not only reduce exchange rate volatility and inflation but also support de-euroization as it reduces the need for a safe-haven asset to protect the value of savings. On the other hand, the associated lower interest rate differentials reduce the insurance premium (or opportunity cost) of holding euro deposits as compared to that of holding local currency deposits.

But euroization turned out to be persistent despite successful macroeconomic stabilization. Micro evidence (Brown and Stix, 2015) from CESEE individuals shows that inflation and exchange rate expectations are still important drivers of the demand for euro deposits. Such expectations are, in turn, partly driven by people’s past experiences and their current assessment of economic prospects and trust in public sector governance.

The question we examine is whether – and if so, to what extent – the observed dynamics of euro deposits are still influenced by exchange rate fluctuations. Note that the countries under observation had different exchange rate regimes in place during our observation period. Poland allowed its currency to float freely throughout the period and Hungary also maintained a floating exchange rate. Czechia interrupted its period of a free floating exchange rate by applying a temporary exchange rate floor between 2013 and 2016. The removal of this floor was very smooth and – in contrast to expectations – did not lead to strong currency reactions. Romania and Serbia, by contrast, allowed their currencies to float until 2017 and 2016, respectively, and then moved to a stabilized arrangement. Croatia followed a tightly managed float throughout the period, with a crawl-like arrangement until 2015 followed by a stabilized arrangement. ⁵ Given these differences between countries, we focus specifically on unexpected exchange rate movements (due to looming crisis events), which have the potential to induce substantial portfolio shifts. ⁶

Chart 1

In general, fluctuations in deposit euroization reflect both valuation effects due to (small) exchange rate movements ⁷ and actual portfolio rebalancing. For households, portfolio rebalancing may be driven by various factors, e.g. changes in disposable income and the savings rate, changes in risk perception and loss aversion, changes in inflation and exchange rate expectations as well as interest rate differentials, i.e. the insurance premium (or opportunity cost) of holding euro deposits. For NFCs, demand for euro liquidity may result from their integration in European value chains and their need to settle payments (related to trade, revenues, installments, wages) in euro. In addition, NFCs at times demand a safe-haven asset from a risk and liquidity management perspective. Therefore, the size of their euro deposits might fluctuate with seasonal swings, the business cycle, external demand, changes in risk perception, inflation and exchange rate expectations as well as with interest rate differentials. Based on the considerations and stylized facts presented so far, we will analyze households and NFCs separately.

2 VAR analysis

To analyze the economic impact of unanticipated movements in the nominal exchange rate, i.e. exogenous exchange rate shocks, we set up a vector autoregressive (VAR) model that captures the complex interactions among a set of economic and financial variables and that uses sign restrictions to identify structural shocks.

Our model uses monthly data from the early 2000s through June 2024 ⁸ and incorporates six endogenous variables: the year-on-year growth rate of industrial production, the year-on-year changes in the Harmonised Index of Consumer Prices (HICP) in percent, the month-on-month growth rate of the nominal exchange rate to the euro, the interest rate differential (IRD) between the domestic three-month interbank offered rate (IBOR) and the three-month EURIBOR ⁹ , a measure of relative economic activity (domestic industrial production relative to euro area industrial production) and the year-on-year growth rate of euro deposits ¹⁰ . We estimate the model separately for each country in our sample, and then twice: once including euro deposits of households and once including euro deposits of NFCs.

The reduced form of the VAR model is given by

$$\mathbf{y}_{t} = \mathbf{A}_{1}\mathbf{y}_{t - 1} + \ldots + \mathbf{A}_{p}\mathbf{y}_{t - p} + \mathbf{u}_{t},\quad\mathbf{u}_{t}\ \sim\ N\left( 0,\mathbf{\Sigma} \right),$$

where $\mathbf{y}_{t}$ is an m-dimensional vector containing economic and financial variables, $\mathbf{A}_{n}(n = 1,\ldots,p)$ is a set of $m \times m$ coefficient matrices subject to the number of lags, $p$. The model features a zero mean error term with variance covariance matrix, $\mathbf{\Sigma}$. We can rewrite the model in a more compact fashion with $\mathbf{x}_{t} = \left( \mathbf{y}_{t - 1}',\ldots,\mathbf{y}_{t - p}' \right)$ and $\mathbf{A} = \left( \mathbf{A}_{1},\ldots,\mathbf{A}_{p} \right)$,

$$\mathbf{y}_{t} = \mathbf{A}\mathbf{x}_{t} + {\mathbf{B}\mathbf{\epsilon}}_{t}.$$

The reduced form residuals, $\mathbf{u}_{t}$, are linked to the structural errors, $\mathbf{\epsilon}_{t}$, via $\mathbf{u}_{t} = \mathbf{B}\mathbf{\epsilon}_{t}$, where $\mathbf{\epsilon}_{t}\sim N(0,\mathbf{I})$ and $\mathbf{B}$ constitute the contemporaneous impact matrix. Since our main interest lies in the economic interpretation of structural shocks, we need to formulate adequate restrictions on the $\mathbf{B}$ matrix to achieve identification. Following Uhlig (2005) and Rubio-Ramírez et al. (2010), the identification procedure imposes sign restrictions on the impacts of the multiplier matrix, thus constraining the contemporaneous responses of the variables within the system. We formulate the restrictions in accordance with Leiva-Leon et al. (2022) and present them in table 1.

We identify a domestic aggregate supply shock $\epsilon_{t}^{AS}$, a domestic aggregate demand shock $\epsilon_{t}^{AD}$, an exogenous exchange rate shock $\epsilon_{t}^{ER}$, a monetary policy shock $\epsilon_{t}^{MP}$ and an external demand shock $\epsilon_{t}^{Ext}$. We leave responses of a shock to the change in the euro deposit $\epsilon_{t}^{DE}$ unrestricted. We anticipate that in response to a positive aggregate supply shock, domestic industrial production and relative economic activity both increase. Simultaneously, the HICP, the nominal exchange rate and the IRD are expected to decline on impact of the positive aggregate supply shock. A positive domestic aggregate demand shock is expected to lead to increases in industrial production, the HICP, the exchange rate, the IRD and relative economic activity, while the response of euro deposits is unclear. Following the literature (see, e.g., Rajković and Urošević, 2017, and Ize and Yeyati, 2003), we assume that an exogenous exchange rate shock (domestic currency appreciation) causes declines in the HICP, the IRD and euro deposits. We identify a conventional monetary policy shock (monetary tightening) by assuming that it triggers negative responses in industrial production, relative economic activity and the HICP. A positive external demand shock is expected to increase domestic production and the HICP. However, since euro area demand is anticipated to rise more significantly, relative economic activity will decrease. Consequently, as the euro area is expected to raise interest rates following the demand shock, the IRD will decline.

We estimate the VAR model using Bayesian techniques and include 12 lags, which is a lag length commonly selected when working with monthly data. The model features a hierarchical version of the Minnesota prior developed by Giannone et al. (2015). The Minnesota prior follows the common economic belief that a variable’s own lags are more important for predicting its future values than the lags of other variables in the system. Additionally, more distant lags are given less weight than recent lags. Giannone et al. (2015) propose a conjugate prior setup, where the hyperparameters are treated as additional parameters to be estimated. ¹¹ For our estimation, we save 10,000 draws from the posterior distribution, after discarding the first 10,000 draws as burn-in.

Posterior quantiles (16th, 50th, 84th) of the impulse response functions to a 1% appreciation in the domestic currency relative to the euro are presented in chart 2. For clarity, we focus on the response of three key variables: the HICP, the IRD and euro deposits (DEUR). We present results from the model estimation including euro deposits of households (red) and from that including euro deposits of NFCs (blue). We choose to show NFC results only for DEUR, since NFC results for the HICP and IRD are qualitatively very similar ¹² . Note that we calculated growth rates of euro deposits denominated in euro ¹³ , hence valuation effects on the growth rates of euro deposits are zero by construction.

Since we focus on understanding the reaction of euro deposits to unanticipated exchange rate movements, we restrict the discussion of the impulse response functions to the exogenous exchange rate shock. In response to a positive exogenous exchange rate shock, implying a 1% appreciation of the domestic currency, year-on-year growth in the HICP declines by 0.2 to 0.5 percentage points on impact. This effect deepens within the first year following the shock and levels out after approximately 20 to 30 months. Responses of the IRD show initial declines ranging from 0.1 to 0.7 percentage points. Again, for most countries we find a further strengthening of the effect within the first 12 months. The lowest persistence is to be observed in Croatia, the highest in Serbia.

Euro deposits fall on impact, in line with expectations. We find that, across the countries in our sample, the reactions of household deposits (ranging from 0.5 to 3.5 percentage points) are weaker than those of NFC deposits (ranging from 1.5 to 7.5 percentage points). Our results imply that NFCs are more responsive to unanticipated exchange rate movements than households. This favors the notion that NFCs might be more active in managing their liquid reserves and adjusting their portfolios, while households tend to exhibit an inattentiveness bias and a certain degree of inertia. Moreover, households are more concerned to avoid losses than to hunt for higher returns, which implies some asymmetric behavior with regard to currency appreciations and depreciations; yet the VAR model cannot capture these nonlinearities. The difference in the strength of effects between households and NFCs is particularly elevated in Croatia and Serbia.

Interestingly, effects for households show higher persistence in Czechia, Hungary and Romania. In the latter two countries, the effect even increases after impact and peaks only after five months. Once households react to unanticipated exchange rate movements, they adjust their savings behavior more persistently.

To better understand the significance of exogenous exchange rate shocks, we calculate the relative importance of these shocks in total exchange rate fluctuations. On average, these shocks account for 5% to 17% of exchange rate movements across our sample countries. Country-specific results for both model specifications can be found in table A3.

Chart 2

3 Sensitivity analysis

This section explores the time-varying sensitivity of euro deposits to exogenous variations in the exchange rate. To this end, we draw on the exogenous exchange rate shock identified in section 2 and estimate a time-varying parameter (TVP) regression using euro deposits as our dependent variable. We perform this exercise twice for each country, once for household euro deposits and once for NFC euro deposits.

Let us start by setting up a standard TVP model given by

$$y_{t} = \mathbf{x}_{t}\mathbf{\beta}_{t} + \nu_{t},\quad\nu_{t}\sim N\left( 0,\sigma^{2} \right),$$

where $y_{t}$ is a scalar time series and $\mathbf{x}_{t}$ contains a set of $K$ predictors. Since states in $\mathbf{\beta}_{t}$ are allowed to change over time, a law of motion must be defined. We assume that they evolve according to a random walk process given by

$$\mathbf{\beta}_{t} = \mathbf{\beta}_{t - 1} + \mathbf{\eta}_{t},\quad\mathbf{\eta}_{t}\sim N\left( 0,\mathbf{\Omega} \right),$$

with $\mathbf{\Omega} = diag\left( \omega_{1},\ldots,\omega_{K} \right)$. Since $\mathbf{\Omega}$ is a diagonal matrix, the state innovations are conditionally independent. To estimate the model in an efficient manner, we make use of the noncentered specification of a state space model. We follow Frühwirth-Schnatter and Wagner (2010) and rewrite the measurement equation as

$$y_{t} = \mathbf{x}_{t}\mathbf{\beta}_{0} + \mathbf{x}_{t}\widetilde{\mathbf{\Omega}}\widetilde{\mathbf{\beta}_{t}} + \nu_{t},$$

where $\widetilde{\mathbf{\Omega}} = diag\left( \sqrt{\omega_{1}},\ldots,\sqrt{\omega_{K}} \right)$ and with the state equation now reading

$$\widetilde{\beta_{t}} = \widetilde{\beta_{t - 1}} + \mathbf{\eta}_{t},\quad\mathbf{\eta}_{t}\sim N\left( 0,\mathbf{I} \right).$$

We can now intuitively interpret the regression coefficients as a time-varying part that oscillates around a constant component. Additionally, we can also treat the square root of the state innovation variances as additional parameters in our regression. Hence, if our estimation finds that $\omega_{j} = 0$, we observe no time variation in the coefficient of the $j$th variable. Conversely, if we find that $\omega_{j} \neq 0$, we do observe parameter change over time.

In a Bayesian fashion, we estimate the model using a set of prior assumptions. We choose an uninformative normal prior on the constant part of the model, $\beta_{j0}\sim N(0,1)$ with $j\ = \ 1,\ldots,K$. To control the amount of time variation in the model, we place a normal prior on the state innovation variances given by

$$\sqrt{\omega_{j}}\ |\ \lambda\ \sim\ N(0,\ \lambda).$$

To inform our prior beliefs from the data, we set $\lambda = \sigma_{OLS}^{2}$, where $\sigma_{OLS}^{2}$ is the ordinary least squares (OLS) estimate of the variance obtained from fitting a first-order autoregressive (AR(1)) model to the year-on-year change in euro deposits. We use a Gibbs sampler to save 10,000 draws of the posterior distribution, after discarding the first 10,000 draws as burn-in.

For our estimation, we set the year-on-year change in euro deposits as our dependent variable and construct a set of predictors to further control for domestic and global economic developments. This set includes:

• the country-specific exogenous exchange rate shock, $\epsilon_{t}^{ER}$, which is obtained from the identified VAR model in section 2;

• the Global Economic Conditions indicator (GECON, Baumeister et al., 2022), which controls for global economic trends. It considers a wide range of activity, financial and uncertainty measures;

• country-specific five-year sovereign credit default swaps, to control for domestic risk and market expectations;

• the IRD between the domestic three-month IBOR and the three-month EURIBOR;

• the inflation differential between domestic and euro area HICP, to also consider price pressures in the euro area;

• domestic industrial production in year-on-year growth rates.

The sample consists of monthly data and starts in the 2000s, continuing through June 2024. ¹⁴ Our particular interest lies in the time-varying coefficient related to $\epsilon_{t}^{ER}$, which is the measure of sensitivity of euro deposits to unanticipated exchange rate fluctuations. We present the time-varying coefficient estimates for households (red) and NFCs (blue) in chart 3.

3.1 Results for households

The time-varying coefficients exhibit the expected sign according to the literature, i.e. an appreciation of the local currency vis-à-vis the euro is associated with a decline in euro deposits and vice versa. The sensitivity of euro deposits with respect to exogenous exchange rate shocks varies gradually in Czechia, Hungary and Poland and does not show strong changes over time in Croatia, Romania and Serbia. In Czechia, we find a significant reaction to exogenous exchange rate shocks between 2006 and 2010, i.e. around the time of the global financial crisis (GFC). The reaction becomes significant again after the euro area debt crisis and peaks in the years of negative interest rates in the euro area. Croatia exhibits a stable and significant relationship with almost no variation over time, which points to the important role of hysteresis and habits in explaining the persistence of euroization. This finding is in line with micro evidence on the determinants of deposit euroization (Brown and Stix, 2015). Both Hungary and Poland show a similar pattern to Czechia. The results become significant after the GFC in 2010, peaking in the years of negative euro area interest rates and staying marginally significant until end-2023 in Hungary and end-2022 in Poland, respectively. In Romania and Serbia, we detect a significantly negative response throughout the observation period; however, time variation is limited.

3.2 Results for NFCs

As discussed above, the euro deposits of NFCs fluctuate with more volatility than those of households (see chart 1). Similarly, the sensitivity of NFC euro deposits to exogenous exchange rate shocks shows a pattern of higher volatility, lower persistency and stronger reactions than that of household euro deposits in the respective country. NFCs seem to be more vigilant and responsive to exchange rate changes across the six countries. In Czechia, we detect episodes of heightened sensitivity after the GFC-induced recession and during the negative interest rate period in the euro area. The sensitivity of NFCs’ euro deposits during the negative interest period in the euro area was elevated and substantially stronger than the sensitivity of households’ euro deposits. The sensitivity of NFC euro deposits in Croatia fluctuated strongly and fast between 2002 and 2013, when it suddenly converged to zero and remained insignificant until 2023. Croatia operated a managed float when it joined the EU in 2013, simultaneously entered the Exchange Rate Mechanism II and the Single Supervisory Mechanism in 2020 and finally adopted the euro in 2023. These institutional changes, which permanently reduced the risk of large depreciations, may have permanently weakened the responsiveness of NFC euro deposits in the country after 2013.

In Hungary we observe a heightened sensitivity of euro deposits prior to the GFC and after the euro area debt crisis. However, we observe a somewhat gradual overall decline over time, which turns insignificant in 2019. For Poland, we find stronger sensitivity after the euro area debt crisis and during the negative interest rate period in the euro area. In recent years, the strength of the effect moderated. Romanian NFCs turn out to be different: Their sensitivity is found to be weak and insignificant prior to 2016 and gradually turns negative thereafter. Results for Serbia show more time-varying features for NFCs than for households. The coefficient turns negative during the euro area debt crisis and stays negative after 2016, covering the negative interest rate period and the COVID-19 pandemic-induced recession. These results may also reflect the change in the management of the exchange rate by the National Bank of Serbia around 2016.

Chart 3

4 Conclusions

We investigate the effects of unanticipated exchange rate movements on euro deposits in selected Central, Eastern and Southeastern European (CESEE) economies characterized by flexible exchange rate regimes and considerable deposit euroization. In our analysis, we separate the deposits of households from those of nonfinancial corporations (NFCs) as these different economic agents are likely to react differently to the same signals. Our empirical approach involves a two-step process in which we first identify structural shocks and then compute impulse responses to an unanticipated exchange rate movement in the form of a 1% exogenous domestic currency appreciation.

In line with expectations, we find that euro deposits decrease on impact. Across the six countries in our sample, the response of household deposits (ranging from 0.5 to 3.5 percentage points) is weaker than that of NFC deposits (ranging from 1.5 to 7.5 percentage points). The effect is most pronounced in Croatia and Czechia, where households’ euro deposits decrease on impact by about twice as much as in all other countries. Also, the effects on households are more persistent in Croatia, Czechia, Hungary and Romania than in the other countries. The difference in the strength of the effects between households and NFCs is particularly pronounced in Croatia and Serbia. Our results support the notion that NFCs are more sensitive to unanticipated exchange rate movements whereas households pay more attention to other factors – such as preserving the value of their assets rather than maximizing returns, which implies an asymmetric behavior – or even remain inattentive to exchange rate shocks within a certain range. This supports the idea that NFCs may be more active in managing their liquid reserves and adjusting their portfolios, whereas households tend to exhibit an inattention bias and a certain degree of inertia. While euro deposit levels correlate overall with differences in exchange rate regimes, we find no such systematic differences for the sensitivity of euro deposits to exogenous exchange rate shocks. The shock reactions are highest in Croatia, which has a tightly managed float (and thus is the country with the lowest exchange rate volatility in our sample), and Czechia, which had a floating exchange rate over most of the period under observation.

The second step in our process allows us to examine how the sensitivity of changes in euro deposits changes over time in response to unanticipated exchange rate movements. To this end, we estimate a time-varying parameter regression for each of the six countries in our sample. As expected from the literature, we find that an appreciation of the local currency against the euro is associated with a decline in euro deposits and vice versa. The sensitivity of household euro deposits to exogenous exchange rate shocks varies gradually over time in Czechia, Hungary and Poland but shows little time variation in Croatia, Romania and Serbia. Hence, with respect to response over time, we find some difference between those countries with a more freely floating exchange rate and those with a more managed float, at least when we look at households. It should be kept in mind that, in general, households react less strongly to unanticipated exchange rate movements than NFCs. However, once they react, they will also adjust their savings behavior more persistently.

In contrast, NFCs show greater volatility in their responsiveness to exchange rate shocks over time. In all countries observed, the sensitivity of NFCs’ euro deposits to exogenous exchange rate shocks is more frequent, faster and sometimes larger than that of households in the respective country. The time-varying effects confirm that NFCs seem to be more vigilant and responsive to exchange rate changes. Croatia is an interesting case, displaying strong fluctuations in NFCs’ euro deposits until 2013 and almost zero reaction after the country’s accession to the EU and the adoption of a stabilized arrangement.

Following the classification of Khvedchuk et al. (2019), a discussion on what factors may drive a shift in the sensitivity of demand for euro deposits, we distinguish three different factors: macroeconomic factors, structural factors and changes in regulatory policy. In our interpretation, the episodes of heightened sensitivity identified in the six countries observed are mainly related to macroeconomic and country-specific structural factors. The correlation between euro deposits and unanticipated exchange rate movements increases substantially for both households and NFCs in Czechia, Hungary and Poland during the episode of very low inflation and negative interest rates in the euro area. In more strongly euroized economies, like Croatia and Serbia, in contrast, structural factors may have dominated. Croatia operated a managed float when it joined the EU in 2013, entered both the Exchange Rate Mechanism II and the Single Supervisory Mechanism in 2020 and adopted the euro in 2023. These institutional changes, which permanently reduced the risk of large currency depreciations, may have permanently weakened the responsiveness of NFC euro deposits in Croatia after 2013. In contrast, the sensitivity of households’ euro deposits in Croatia remained significant and stable throughout the observation period, pointing to the important role of hysteresis and habits in explaining the persistence of euroization (Brown and Stix, 2015). Finally, the results for Serbia suggest that policy changes, both in the exchange rate management of the National Bank of Serbia and in the dinarization strategy of the Serbian authorities, may have played an important role in the sensitivity of euro deposits to exogenous exchange rate shocks.

5 References

Baumeister, C., D. Korobilis and T. K. Lee. 2022. Energy markets and global economic conditions. In: Review of Economics and Statistics 104(4). 828–844.

Brown, M., R. De Haas, C. Tille and L. Halpern. 2012. Foreign banks and foreign currency lending in emerging Europe. In: Economic Policy 27(69). 57–98.

Brown, M. and H. Stix. 2015. The euroization of bank deposits in Eastern Europe. In: Economic Policy 30(81). 95–139.

de Nicolo, G., P. Honohan and A. Ize. 2005. Dollarization of bank deposits: Causes and consequences. In: Journal of Banking & Finance 29(7). 1697–1727.

Frühwirth-Schnatter, S. and H. Wagner. 2010. Stochastic model specification search for Gaussian and partial non-Gaussian state space models. In: Journal of Econometrics 154(1). 85–100.

Giannone, D., M. Lenza and G. E. Primiceri. 2015. Prior selection for vector autoregressions. In: Review of Economics and Statistics 97(2). 436–451.

Harkmann, K. and K. Staehr. 2021. Current account drivers and exchange rate regimes in Central and Eastern Europe. In: Journal of International Money and Finance 110. 102286.

Honig, A. 2009. Dollarization, exchange rate regimes and government quality. In: Journal of International Money and Finance 28(2). 198–214.

Ize, A. and E. L. Levy Yeyati. 2003. Financial dollarization. In: Journal of International Economics 59. 323–347.

Khvedchuk, K., V. Sinichenko and B. Topf. 2019. Estimating a natural level of financial dollarization in Ukraine. In: Visnyk of the National Bank of Ukraine No. 247. 38–44.

Kokenyne, A., J. Ley and R. Veyrune. 2010. Dedollarization. IMF Working Paper 10(188). 1–50.

Leiva-Leon, D., J. Martínez-Martín and E. Ortega. 2022. Exchange rate shocks and inflation comovement in the euro area. In: International Journal of Central Banking 18(1). 239–275.

Levy Yeyati, E. L. 2006. Financial dollarization: evaluating the consequences. In: Economic Policy 21(45). 62–118.

Manjani, O. 2015. Estimating the determinants of financial Euroization in Albania. Graduate Institute of International and Development Studies Working Paper No. HEIDWP07-2015. 1–28.

Rancière, R., A. Tornell and A. Vamvakidis. 2010. Currency mismatch and systemic risk in emerging Europe. In: Economic Policy 25(64). 597–658.

Rubio-Ramírez, J. F., D. F. Waggoner and T. Zha. 2010. Structural Vector Autoregressions: Theory of Identification and Algorithms for Inference. In: Review of Economic Studies 77(2). 665–696.

Scheiber, T. and J. Wörz. 2024. Exporting stability to the European neighborhood – the role of deposit euroization in CESEE revisited after 25 years of EMU. In: Monetary Policy & the Economy Q4/23, 25 years of EMU in Austria. Oesterreichische Nationalbank. 61–78.

Slavov, M. S. T. 2017. Exchange Rate Regimes in Central, Eastern and Southeastern Europe: A Euro Bloc and a Dollar Bloc?. IMF Working Paper 2017(83).

Uhlig, H. 2005. What are the effects of monetary policy on output? Results from an agnostic identification procedure. In: Journal of Monetary Economics 52(2). 381–419.

Urošević, B. and I. Rajković. 2017. Dollarization of deposits in the short and long run: Evidence from CESE countries. In: Panoeconomicus 64(1). 31–44.

6 Annex

The annex can be found in the pdf version of this study.