Monetary policy pass-through with negative policy rates
An empirical analysis on retail bank deposits
Abstract
This thesis explores the interest rate pass-through from monetary policy to retail deposit rates using a panel data approach covering data of 17 European countries. This study builds on the discussion about the pass-through effectiveness which was especially vivid at the beginning of the century, but has revived since central banks introduced negative policy rates in the aftermath of the global financial crisis. Most studies focus primarily on bank lending rates and bank profitability with low interest rather than deposit rates. Estimating the pass-through to deposit rates directly while discriminating between the period with positive and negative policy rates, I find evidence that the pass-through to deposit rates has significantly weakened since central banks introduced negative policy rates. In addition, the analysis shows that the degree to which banks are funded by deposits is a significant factor in the pass-through. Furthermore, this study finds an unexpected largely homogeneous pass-through across European countries. These results are robust for various model assumptions and specifications and reveal that the degree of pass-through is especially dependent on the distance of the zero bound for deposits.
Keywords: Monetary policy, Retail deposit rates, Interest pass-through, Negative interest rate policy
University of Groningen, Faculty of Economics & Business Master Thesis
MSc Finance 11 January, 2021
Author: Maurits Pieter Roorda Student number: S3485544
Content
1. Introduction ... 3
2. Brief background theory of monetary transmission ... 5
3. Literature review ... 7
3.1 The interest rate channel and deposit rates ... 7
3.1.1 ‘Conventional times’ ... 7
3.1.2 Facing the ZLB ... 8
3.3 Literature gap and hypotheses ... 10
3.3.1. Pass-through literature gap ... 10
4. Empirical strategy ... 12
5. Data ... 17
5.1. Interest data ... 17
5.1.1. Retail deposit rates ... 17
5.1.2. Monetary policy and funding cost rates ... 19
5.2. Third factor data ... 20
5.3. Concise data statistics ... 22
6. Pass-through analysis results ... 23
6.1 Exploring breakpoints... 23
6.2. Full sample pass-through results ... 24
6.3. Pass-through results for individual countries ... 27
6.4. Robustness of results ... 33
6.4.1. Fee income and POLS assumptions ... 33
6.4.2. Time deposits with fixed maturity ... 33
6.4.3. Marginal cost rate stability ... 37
7. Conclusions, limitations and further research ... 39
3
1. Introduction
The decade 2010-2020 was characterized by the aftermath and revival from the global financial crisis of 2008. In order to combat declining inflation and economic growth, many central banks reduced their policy interest rates multiple times. Ever since, inflation has been unstable and lagged behind the objectives of the central bank mandates. In June 2014, the European Central Bank (ECB), lowered their policy rates for the European Monetary Union (EMU) countries below zero percent for the first time, following the example of the Danish central bank in July 2012. Shortly after, also the Swedish and Swiss central bank followed. Leading up to these decisions, many central bankers and economists believed that the arrival at the zero bound would lead to a problem referred to as the Zero Lower Bound (ZLB). At the zero nominal policy interest bound, central banks would face a liquidity trap and lose their main tool to stimulate the economy by lowering policy rates. In a monetary policy report of 2015, the Swedish central bank wrote that policy rate cuts in negative territory would not have a different effect on the transmission channels of monetary policy, provided that the cuts are not very large. Figure 1 below suggests however that the response of overnight retail deposit rates to policy rate cuts stop when their zero bound is reached, indicating a reluctance of banks to pass on negative cuts in the deposit facility rate (DFR) of the ECB to their customers.
Figure 1: EMU policy rates versus aggregated retail deposit rates. Source: ECB Statistical Data Warehouse MFI interest rate statistics.
4 degree of market competition and deposit funding, volatility of the money market and other macroeconomic variables seem to influence the transmission as well (Gigineishvili, 2011). In order to add to the knowledge about the pass-through to retail deposits with low interest, I focus primarily on how the pass-through to retail deposit rates has changed after the introduction of negative interest rate policy (NIRP). In addition, I examine the roles of competition and deposit funding in particular and explore whether there exists heterogeneity in the pass-through of countries. The data for this research is from 14 EMU countries and the countries Sweden, Switzerland and Denmark (see appendix A), covering the recent period from April 2008 until July 2020. These countries have in common that their central banks pursue negative policy rates for a while. The pass-through effectiveness is estimated using according to two approaches which differ in the main explanatory variable of interest; the monetary policy approach using short-term interbank rates to reflect policy rates, and the cost of fund approach using a more mature market rate as proxy for the relevant cost of fund faced by banks. The innovation of this paper relative to the existing literature is that I directly examine whether the pass-through to deposit rates has been impaired after the introduction of NIRP, using recent data from 17 European countries.
5
2. Brief background theory of monetary transmission
This section provides a brief theoretical background of monetary policy and its different channels. Its mechanisms are quite technical by nature and this short introduction helps to understand the remainder of this study. The full theoretical background is presented in appendix B.
Monetary policy transmission is built on four main channels that direct policy rate changes to the real economy through financial markets. Each of them acts as a catalyst for the pass-through to affect different pillars of the financial system in the desired way (Bernanke and Gertler, 1995). The interest rate channel is nowadays perceived as the most important by many central banks. It passes changes in policy rates on to long-term interest rates and retail interest rates set by for instance commercial banks. Thereby, it indirectly influences the amount of savings and investment via retail credit institutions that receive deposits and grant credit. Central banks act as ‘banker of the banks’. Banks store their surpluses at central bank deposit accounts and can borrow from the central bank. In addition, banks can also use an interbank market to exchange liquidity. Central banks set the interest rates that apply to either surplus deposits stored at the central bank or short-term credit granted to banks. The overnight rate used in the interbank market is derived from rates set by the central bank, referred to as standing facilities. This rate is generally regarded as the policy rate. Since policy rates became negative in the EMU, the lower limit of the standing facilities referred to as the DFR, approximately equals the overnight interbank rate and became a leading reference for the monetary policy stance of the ECB. The standing facilities framework directly influences short-term money market interest rates and indirectly also the interest rates set by banks for retail loans and deposits. A simple but adequate model to explain the pass-through of policy rates to retail rates is the marginal cost price model, which is often used as a basic framework in pass-through studies (Rousseas, 1985; De Bondt, 2005; Kleimeier & Sander, 2002) because it reflects the price setting of banks quite accurately. The theory behind the model states that short-term money market rates controlled by central banks, are ultimately transmitted to various retail rates through marginal funding costs faced by banks (Illes, Lombardi & Mizen, 2019). The market rates reflect marginal costs of funds for banks because they rely on them for short-term borrowing. They also represent opportunity costs for retail deposits.
This relationship can approximately be captured by the following simple model
Dr = α+ γMr (1)
7
3. Literature review
This section provides an overview of the key literature on the topic of interest rate pass-through and assesses it critically. It is written to present the main findings related to the transmission of interest rates from former research. I start by focusing on the interest rate channel in a positive interest environment. Because this thesis builds on the line of research on pass-through in a low interest rate environment, it is important to explore whether the mechanisms of transmission holds once policy interest rates become negative. I draw conclusions from the analyzed literature and identify its gap. I conclude by describing research questions, each accompanied by a formal hypothesis.
3.1 The interest rate channel and deposit rates
3.1.1 ‘Conventional times’
Literature on monetary policy transmission is abundantly present but the vast majority explores the interest rate transmission to retail lending rates. This is not surprising if one considers that bank deposits are a funding source for cash, with the ultimate objective to lend it out for profitable projects. The volume of credit granted for these projects is an important economic indicator and target of central banks. Given that deposit rates have predominantly been positive, this might clarify why researchers have been particularly interested in retail lending rates. Early empirical studies like Cottarelli & Kourelis (1994) and Borio and Fritz (1995) focus on the pass-through from money market rates on short-term corporate loan rates and compare estimates in a cross-country setting. Both find a short-term pass-through well below the policy rate change for most of the Euro area countries. In addition, the long-term pass-through on equivalent loan rates is significantly higher for almost all sample countries, but there exists clear cross-country differences in the degree of pass-through for European countries before realization of the EMU in 1999.
8 In a positive interest rate environment with the standing facilities well above the ZLB, several studies document differences in the pass-through over time, depending on the state of the interest rate cycle (Heinemann & Schüller, 2002; Mester & Saunders, 1995). The transmission on retail rates seems to be faster when the change is advantageous for banks. Hofmann and Mizen (2004) and Kopecky & Van Hoose (2012) concluded that the speed of adjustment of retail deposit rates to policy rates depends on expectations of future rates and whether the gap between policy rates and retail rates is growing or closing. Deposit rates adjusted faster to a cut in policy rates than to an equivalent increase. Volatility of the money market rate slows down the pass-through on deposit rates (Mojon, 2000; Gigineishvili, 2011). The European integration project entailed a common monetary policy stance for the EMU countries and reduced volatility in the money market. Although heterogeneity remained, it resulted in a more uniform pass-through for the respective countries.
Examining the ‘stickiness’ of deposit rates, many studies find evidence about the importance of competition in the banking market with respect to the pass-through (De Bondt, 2005; Neumark & Sharpe, 1992; Kleimeier & Sander, 2002; Hannan, 1997; Mojon, 2000). The effect of competition on the pass-through is found to be uniformly positive. It requires some degree of market power by banks to be able to adjust only partially to policy rate changes and banks which exhibit some market power can afford to some extent to pay their own price for deposits because they are not directly trumped by competitors. Each of these studies find evidence that competition and contestability in the retail bank market segments affect pricing behavior of banks regarding interest rates when banks are no price-takers1.
The pass-through is found to be heterogeneous across countries (Bernhofer & Van Treeck, 2013; Verheyen, 2013; Marotta, 2009; Sørensen and Werner, 2006). For the EMU countries, after the introduction of the common euro currency, continuing differences in policy rate transmission are reported. The ongoing heterogeneity is mainly attributed to the relatively large differences in competition between national banking markets. Furthermore, the pass-through is more complete in countries whose banks report a higher share of provisions, typical for competitive markets. Research on bank risk taking confirms this suggestion (Jiménez, Lopez & Saurina, 2007; Boyd & De Nicolo, 2005).
3.1.2 Facing the ZLB
With interest rates dropping dramatically since the global financial crisis, a growing body of literature paid attention to the effectiveness of economic stimulus through low policy rates2. Hiroshi (2018) compares Japanese banks that were temporary subject to negative policy rates,
1 As assumed in perfect competitive markets, where profit-maximizing agents face a market price equal to its marginal cost price (P=MC) and average profit equals zero.
9 to banks that were not3. He focuses on the impact on lending rates and finds that the former group responded with lower rates after a policy rate cut while the latter group did significantly less. Yoshino, Taghizadeh-Hesary & Miyamoto (2017) find a limited transmission to lending rates after entering NIRP. Both papers ignore how deposit rates responded which makes it difficult to derive its role in the findings.
Jackson (2015) and Brunnermeier & Koby (2018) discuss the limits of the effective lower bound (ELB), which is the point beyond which further expansionary monetary policy becomes counterproductive. They document a limited transmission of negative policy rates on aggregate deposit rates leading up to that point. Also profitability and funding sources can negatively be affected with NIRP. Altavilla, Boucinha & Peydr (2017) show that net interest income of banks is significantly harmed after the introduction of NIRP, suggesting that banks are reluctant to lower their deposit rates because of the potential loss of a cheap funding source. Jobst & Lin (2016) do not find an impaired pass-through to deposit rates. Their data covers only the first mild initiative to go beyond the ZLB and suggest that substantial policy rate cuts in negative territory will alter the observed effect through the fear of losing household deposits.
The degree of deposit funding is found to be another factor influencing the pass-through of negative policy rates (Jobst & Lin, 2016; Heider, Saidi & Schepens, 2016; Eggertson, Juelsrud, & Summers, 2019). Banks relying heavily on deposit funding are moving to more risky lending with higher potential returns. Negative policy rates induce a wedge between deposit funding and market-based funding. As high deposit banks are reluctant to pass-on negative rates, the lower policy rates make market-based funding cheaper relative to deposit funding and this lowers the net worth of high deposit banks relative to those that depend less on deposit funding. In addition to increased risk-taking, these banks also borrow less as alternative source of funding. The findings suggest that banks are unwilling to lower their interest rates once the deposit rate tends to become zero, as these banks might have more to suffer from deposit withdrawals. It follows that banks with a higher deposit share have more to lose from lowering deposit rates below zero and are likely to be more conservative in doing so. On the contrary, high deposit banks pay relatively more interest to depositors and have therefore an incentive to lower deposit rates. The fear of losing clients versus decreasing profitability seem to be important opposing forces, making the effect of deposit funding ambiguous.
Research that puts more emphasis on the magnitude of the direct pass-through of policy rates to deposit rates in a low interest environment is scarce. A number of studies on the pass-through in a low interest environment do not document an impaired pass-through (Jensen & Spange, 2015; Hristov, Hulsewig & Wollmerhouser, 2014; Bech & Malkhozov, 2016; Horvath, Kotlebova & Siranova, 2018). All these papers also provide a reason that questions their
10 findings, like the use of data covering only mildly negative policy rates which could be just tolerable for banks. Another mentions stimulus of complementary unconventional monetary policy measures such as quantitative easing (QE), introducing a way for banks to attract liquidity other than deposits and fostering a stronger pass-through with expansionary stimulus. In contrast, Eisenschmidt & Smets (2018) does find a zero lower bound on household deposits for the EMU, as they stopped responding to policy rate cuts in negative territory. They do not observe the same for corporate deposits and suggest that it is much easier for households to substitute deposits with cash because of the relatively small volumes, yielding limited storage costs. Basten & Mariathasan (2018) explore several aspects of the banking business and find an incomplete pass-through to deposit rates. They document reluctance of banks to introduce negative deposit rates to protect their reputation and prevent a loss of current and future customers. Interestingly, they find that banks increase their fee income which is likely to compensate for the losses of keeping the deposit rate above its intended equilibrium level. 3.3 Literature gap and hypotheses
3.3.1. Pass-through literature gap
11 Either way, the results of this study will contribute to the knowledge about the effectiveness of conventional monetary policy in unconventional times.
3.3.2. Research questions and hypotheses
I aim to conduct empirical research to find out whether the pass-through to retail deposit rates has remained equally strong with NIRP. In order to do so, the following research questions need to be answered and are translated into concrete hypotheses.
Research question 1 (Q1): Does the monetary policy transmission from policy rates to
commercial bank deposit rates remain stable after central banks introduced negative interest rate policy?
Hypothesis 1: The monetary policy transmission from central bank policy rates to retail deposit rates becomes weaker once policy rates are below zero percent.
(Q1a): Does the transmission of negative policy rates to retail deposit rates depends on the degree to which banks are funded by deposits?
Hypothesis 1a: The transmission of negative policy rates to retail deposit rates weakens more when commercial banks hold a large share of liabilities as retail deposits, relative to banks that hold fewer deposits.
(Q1b): Does the transmission of negative policy rates to retail deposit rates depend on the degree of competition in the banking sector?
Hypothesis 1b: The transmission of negative policy rates to retail deposit rates weakens more in less concentrated banking markets, relative to more concentrated banking markets.
(Q1c): Is there heterogeneity in the transmission of policy rates to retail deposit rates between countries after central banks introduced negative interest rate policy?
12
4. Empirical strategy
This section describes the empirical strategy I follow to test the hypotheses from section 3.3.2. The first step in the analysis is to explore the relationship between the monetary policy rate and the retail deposit rate over time. More specific, whether there is a systematic breakdown in the relationship between the two somewhere in the sample period. According to the theory, a structural break could be expected at the introduction of NIRP. A structural break could not be interpreted as evidence for an impaired pass-through relationship but it does indicate a sudden disturbance in the relationship.
Subsequently, I use panel data to estimate the pass-through relationship. Panel data provides rich information and facilitates efficient estimation. Acknowledging heterogeneity across countries is important because the sample period covers data of events that are likely to have had different impact on national financial systems, such as the European sovereign debt crisis. Another point of attention is that interest rate models are often estimated in first difference to avoid spurious regression problems4 (Granger & Newbold, 1974). Converting levels into first difference has the property of removing unit roots, making the data first difference stationary. Another advantage of the first difference transformation is that it eliminates unobserved individual effects causing heterogeneity factors other than the change in the policy rate. Hence, I choose to difference all continuous data once. An alternative solution could be to model the pass-through by applying a method suitable for estimating autoregressive models like the Panel Vector Autoregression Model, but this method is less suitable for identifying differences between periods. Combining these considerations, estimating differenced data using a linear estimator like by pooled ordinary least squares (POLS) is the best choice. POLS is the standard estimation technique for panel data, which is adequate to study the time series of multiple countries without controlling for unobserved individual effects which are eliminated by the first difference transformation.
In order to isolate the pass-through effectiveness from the data, I perform regressions of retail deposit rates on the policy interest rate and other market rates. The general model to estimate the pass-through of monetary policy in this study is as follows.
∆Dri,t = α + β∆Mri,t + Ʃ𝑗=1𝑛∗ γj∆Mri,t-j + δBi,t postbound + η∆Mri,t*Bi,t postbound + Ʃ𝑗=1𝑛∗ θj∆Mri,t-j*Bi,t postbound + ιComp
i + κ∆Mri,t*Compi*Bi,t postbound + Ʃ𝑗=1𝑛∗ λj∆Mri,t-j*Compi*Bi,t postbound +πDepfi,t +
ρ∆Mri,t*Depfi,t*Bi,t postbound + Ʃ𝑗=1𝑛∗ τj∆Mri,t-j*Depfi,t*Bi,t postbound + υ∆Appi,t +χ∆Voli,t + ψ∆Infli,t +
ԑi,t (2)
13 The model is estimated in first difference, which is taken after constructing the model in levels first. Variables Dr and Mr are the retail deposit rate and the market rate respectively. For the market rate, lags are required because deposit rates are not immediately reacting to policy rate changes (Gambacorta & Iannotti, 2007; Sørensen & Werner, 2006). This version of a finite
distributed lag model is often used to model the impact of monetary or fiscal policy. The
optimal lag length for the market rate is indicated by n* and further motivated in the results section. Most pass-through studies use the monetary policy interest rate as key explanatory variable while others also advocate the use of longer term money market rates to capture the marginal funding cost that banks face. The former is referred to as the monetary policy
approach using the monetary policy rate and the latter as the cost of fund approach using the marginal cost rate. The latter approach is followed in industrial organization based studies
which view the pass-through strength as an indication for imperfections in the banking sector. Kleimeier and Sander (2002) stress that both methods should be considered complementary because they could offer different insights when the results are compared. I follow their recommendation and follow both approaches, which basically only differs in the explanatory variable used to reflect Mr.
Variable B is a dichotomous indicator for the NIRP bound and also interacts with the market rate and its lags to measure the post-bound pass-through with negative policy rates. This bound discriminates between the periods with positive interest rates and negative interest rates and will be set such that it equals the point where the monetary policy rate first entered negative territory. At this point, the constant markdown α in model (1) could theoretically be zero such that the nominal bank deposit rate equals a fraction γ of the market rate in (1). The marginal cost rates for the cost of fund approach also become negative within only a few months from this point.
Third factors influencing the pass-through significantly are scarce in theoretical models and empirical research confirms that factors that do play a role are limited (Gigineishvili, 2011; Mojon, 2000). Despite, the degree of competition or market concentration as well as the degree of deposit funding seem to be important factors, as discussed in sections 2 and 3. I use indicator variable Comp to indicate whether the concentration of a country its banking sector is above or below the sample median. A similar approach5 will be employed for the degree of deposit
funding in each country, using indicator Depf. Both indicators interact with the market rate as well as the indicator for the post-bound period to capture their effects on the post-bound pass-through.
Variable App measures the effect of the Asset Purchasing Programme (APP) initiated by the ECB at the end of 2014 to bring back inflation to the desired level while interest rates are at the
14 ZLB. With this program, better known as QE, the ECB buys a range of assets in the financial sector to stimulate the economy. The purchases drive up asset prices, reduce the yield curve and signals trust to financial markets (Gambetti & Musso, 2017). In addition, there are spillover effects of the APP program to other countries, among others in Sweden, Denmark (Korus, 2019) and Switzerland (Bernhard & Ebner, 2017). Because one of the channels of the APP works through a reduction in lending rates, bank profitability would suffer unless deposit rates are sufficiently reduced. Therefore, variable App is expected to have a negative effect on deposit rates which could overlap the post-bound pass-through and needs to be controlled for.
Volatility of the money market rate is interpreted as uncertainty in the underlying market by banks. They respond to changes in the money market depending on the reliability of information carried by interest rates (Gigineishvili, 2011). As a consequence, banks are more reluctant to act on the signal when volatility is high and wait for more reliable information. Volatility of the money market rate is therefore expected to have a direct disturbing effect on change in the deposit rate. Money market volatility is represented by variable Vol. High inflation could lead to higher uncertainty of future inflation, which in turn can lead to higher interest rates and further uncertainty regarding other economic variables6. Gigineishvili (2011) argues that the
risk following from uncertainty is transmitted from banks to borrowers in order to share that risk and protect returns from unfavorable unforeseen outcomes. For deposit rates, this would mean that in economies that pursue contractionary monetary policy as a natural response to high inflation, the pass-through would likely be stronger in case banks raise deposit rates proportionally with lending rates. Monetary policy itself does in general not affect the interest margin much and the interest spread is typically held constant while adjusting bank rates (Pétursson, 2001). On the other hand, in current times of exceptional low interest, banks profitability might struggle and maintaining a constant interest spread might no longer be preferred. While banks have an incentive to increase lending rates quickly when the policy rate goes up, the pass-through to deposit rates might react more sluggish to widen the interest margin for additional protection of profitability. On the contrary with expansionary monetary policy which is more evident with lower inflation, lending rates might react more sluggish as it could be beneficial for profit. At the same time, banks have an incentive to decrease deposit rates stronger to widen the interest margin and protect profitability. This contrasts with the theory that states implicitly that with lower inflation, risk from uncertainty is transmitted to a lesser extent (Friedman, 1977; Ball, 1992; Gigineishvili, 2011). As the sample period is dominated by relatively low inflation and even deflation, one would expect a weak effect of inflation on the deposit rate according to this theory, but as the counterarguments denote, the effect of inflation is highly ambiguous in the deposit rate setting. Inflation is represented by variable Infl.
The variables Vol and Infl are not interacted with the post-bound dummy because their effects
15 on the pass-through are expected to be continuous and not dependent on the sign of the change of the policy interest rate. In addition, interacting with the post-bound dummy is purely intended to identify effects over two different periods. It does not add to the reflection of reality to interact the variables Vol and Infl with the post-bound indicator and moreover, exploring their effects between two periods is beyond the scope of this study. As the motivation to include these variables makes clear, their effects are directly considered in the price setting process and are therefore only expected to have a contemporaneous effect. Finally, ԑ is the error term in model (2).
Model (2) includes the indicator variable for the post-bound and its multiplicative term with the market rate and its lags. The pre-bound period is referred to as the base group and the post-bound indicator measures the relative difference of the post-post-bound period with the pre-post-bound period. Hence, the market rate interaction with the post-bound indicator and the lagged versions of that interaction measure only the difference in the pass-through with respect to the base group, which are the market rate variable and its lags that do not interact with the post-bound indicator.
Wooldridge (2001) argues that binary indicator variables could be differenced if necessary. Following Cochrane (2018) however, one has to be careful with differencing data and pursue to model relationships in the most realistic way. Because differencing the binary variables leads to either one or zero datapoints in each panel, the post-bound indicator as well as those for competition and deposit funding stay in the model (2) in levels. The intercept α reflects a time trend in a differenced model. The three-way interactions measure how the post-bound pass-through varies across the levels of the third variables Comp and Depf, as motivated in Rafiq and Chin (2019) and Brambor, Clark & Golder (2006). The coefficients of the three-way multiplications are reflecting the difference with the base groups which are the two-way interactions for the post-bound pass-through without the dummies Comp and Depf.
16 Additionally, I can easily measure the effect of competition and deposit funding on the pass-through over the full sample period to compare it with the post-bound effects. A slight modification of model (2) omits the post-bound indicator as well as its interactions, resulting in two multiplicative terms that measure the effects of competition and deposit funding on the pass-through over the full sample period rather than the post-bound period only. The multiplicative terms with the post-bound indicator in model (2) act only as a bound to discriminate between a variable its effect between two period and leaving it out results in model (3) below.
∆Dri,t = α + β∆Mri,t + Ʃ𝑗=1𝑛∗ γj∆Mri,t-j + δBi,t postbound + ηCompi + θ∆Mri,t*Compi + Ʃ𝑗=1𝑛∗ κj∆Mr i,t-j*Compi +λDepfi,t + π∆Mri,t*Depfi,t + Ʃ𝑗=1𝑛∗ ρj∆Mri,t-j*Depfi,t + τ∆Appi,t + χ∆Voli,t + ψ∆Infli,t +
ԑi,t (3)
In addition to estimating the interest transmission for all countries together, I estimate it for individual countries to explore the heterogeneity of the pass-through within the EMU as well as the difference between all sample countries. Model (4) below is basically specified as model (2) but includes the three-way interaction of indicator Country with the market rate and the post-bound dummy. The three-way interactions with the competition and deposit funding dummies are replaced with the standalone indicators of competition and deposit funding. Note that model (4) does not aim to capture their effects in the post-bound period, which is done by model (2). Nonetheless their effects need to be controlled for and therefore they appear as standalone dummy controls in the model. Model (4) will be estimated multiple times for each separate country, meaning that only the country dummy will be replaced each time to represent a new country.
∆Dri,t = α + β∆Mri,t + Ʃ𝑗=1𝑛∗ γj∆Mri,t-j + δBi,t postbound + η∆Mri,t*Bi,t postbound + Ʃ𝑗=1𝑛∗ θj∆Mri,t-j*Bi,t postbound + κ∆Mr
i,t*Countryi*Bi,t postbound + Ʃ𝑗=1𝑛∗ λj∆Mri,t*Countryi*Bi,t postbound + πCompi + ρDepfi,t
+ τ∆Appi,t + χ∆Voli,t + ψ∆Infli,t + ԑi,t (4)
17
5. Data
This section discusses the data used for the variables presented in the previous section. Section 5.1 explains how the retail deposit rates are measured and what considerations have been taken into account to find a suitable monetary policy rate for the monetary policy approach and marginal cost rate for the cost of fund approach. The remaining explanatory variables are discussed n section 5.2 and section 5.3 provides a brief overview of the data statistics. Detailed information regarding the sources for all data used in this study is presented in appendix C. 5.1. Interest data
5.1.1. Retail deposit rates
The data for this study is from 14 EMU countries and the countries Sweden, Switzerland and Denmark. These countries have in common that their central banks pursue low interest rates for the last couple of years to adhere to their mandate. To achieve that, they have adopted negative policy rates and subsequently cut them further into negative area. Figure 1 in the introduction already suggested a disturbance in the pass-through for the EMU countries from that point on. Figure 2 below presents a clear and similar decoupling for the policy rates of the other sample countries (dotted lines) and their overnight retail deposit rates (solid lines). The choice for retail deposit rates is among other, determined by data availability. Some pass-through studies use retail deposit rate data at the bank level. Ideally, I would have used bank level data as well, however this requires data collection through individual banks or with the help of central bank data which is not publicly available. As these methods are highly time consuming and not providing clarity about a useful dataset on forehand, I choose to use retail deposit rate data aggregated at the country level expressed in basis points change. The drawback is that information is lost when data is aggregated within a country, increasing the probability of inaccurate estimates (Holderness, 2016).
A distinction can be made between different groups of retail deposit rates, based on the maturity and size of the deposits and the client group (i.e. households, corporations or other non-financial firms etc.). I use deposit rates faced by households and non-profit institutions serving households and private clients on annualized agreed rate. Regarding the maturity, the deposit classes of overnight or sight deposits and time deposits with fixed maturities up to one year two years are used. For Denmark and due to unavailability of data, deposits redeemable at notice with maturity up to three months are chosen instead of overnight deposits7. Also because of
18 unavailability, data for the maturity up to two years misses for Switzerland and data for the maturity up to one year misses for Denmark.
Figure 2: policy rates versus retail deposit rates of European countries. Source: ECB Statistical Data Warehouse MFI interest rate statistics.
The reasons for focusing on households and private clients and the chosen maturities are twofold. First, the sluggishness of the retail deposit rate reaction after a policy rate cut is expected to be more expressed in rates for private clients and households, as also suggested by Eisenschmidt & Smets (2018)8. This is logically the first group of deposits that show a decoupling from the monetary policy and market rates because the threshold to withdraw deposits is lower through the limited storage costs of relatively small volumes of cash. Second, the availability of comparable aggregated data across countries or central bank policies is relatively limited and by using the chosen retail rate classes, the best possible comparability between countries is achieved. This gives three different variables to capture the change in household retail deposit rates; overnight deposits (DrOn), time deposits up to one year (Dr1Y)
and time deposits up to two years (Dr2Y). These will each individually be used in the regressions.
The overnight households household deposits is used first for testing the hypotheses and the other maturities are used for a robustness check.
19
5.1.2. Monetary policy and funding cost rates
Variable Mr reflects two different market rates to measure the pass-through according to the monetary policy approach and the cost of fund approach separately. Following the former, it is common in pass-through studies to use a close ‘tracker’ of the official policy rate due to its infrequent rate of change, which would bring about issues due to insufficient variation in the data. A good proxy for central bank policy rates are overnight money market rates. The overnight interest rate is targeted by central banks using the standing facilities and open market operation (Pérez Quirós & Rodríguez Mendizábal, 2001) and previous studies show close correlations between central bank policy rates and overnight interest rates9. Since the global financial crisis of 2008, the Euro Overnight Index Average (EONIA) closely tracks the Main Refinancing Operations (MRO) rate of the EMU. Altavilla, Canova & Ciccarelli (2020) note that EONIA closely tracks the MRO rate in periods of normal liquidity, the DFR when liquidity is abundant and does not have a floor at zero. They conclude that EONIA is a good indicator of the degree of monetary accommodation the ECB tries to achieve, both in normal times and when credit easing policies are used. I follow these authors in using a proxy for the official policy rates.
Because the correlations between overnight market rates and policy rates may differ across regions, I collected data of the main overnight market rate and official policy rates steered by the central banks of the EMU, Sweden, Switzerland and Denmark to conduct correlation analyses confirming the usefulness of my proxies (appendix D). Although these proxies are not perfect, the correlation analyses over the sample period in appendix D show that the overnight rates are still closely correlated with the policy rates. For the monetary policy approach, I adopt the EONIA10 rate, Stockholm Interbank Offered Rate (STIBOR), Swiss Average Rate Overnight (SARON) and Tomorrow/Next (T/N) overnight rate for the monetary policy proxies of the EMU, Sweden, Switzerland and Denmark respectively. These rates form the variable which reflects the change in Mr for the monetary policy approach expressed in basis points. Following among others De Bondt, Mojon & Valla (2005), De Bondt (2005) and Kleimeier & Sander (2002) for the cost of fund approach, money market rates with varying maturities are examined on their correlations with retail deposit rates (Appendix D). Because market rates accurately reflect the marginal funding cost faced by banks, the idea is that they are most appropriate in reflecting the marginal cost price. In addition, the increasing competition between bank-based and market-based products may have induced banks to increasingly pay
9 See for instance the correlation analyses performed by De Bondt (2005).
20 attention to market rates with higher maturity when setting their own retail rates. A natural starting point for finding the appropriate market rate per deposit category is to look for the closest matching maturities (Sørensen & Werner, 2006) to capture the best proxy for the marginal cost rate, which I consider the opportunity cost in this deposit rate setting. Because the overnight market rate is already used for the monetary policy approach, for the cost of fund approach, the Euro Interbank Offered Rate (Euribor) one week, STIBOR one week, London Interbank Offered Rate Confoederatio Helvetica Franc (LIBOR CHF) one month and Copenhagen Interbank Offered Rate (CIBOR) three months are chosen for the EMU countries, Sweden, Switzerland and Denmark respectively. These rates form the change in Mr for the cost of fund approach expressed in basis points. For the time deposits with maturities of one year and two years, the longest maturity market rate is chosen, in all cases the 12 month market rate. Some studies select the cost of fund by picking the market rate with the closest correlation11. One might notice that choosing the highest correlated rate is biasing the results as it can be considered as ‘searching for significance’. I assessed the correlations of all market rate maturities (See appendix D) as well. The market rates based on matching maturities do not yield the highest correlations, but seem still closely related to retail deposit rates.
5.2. Third factor data
The degree of competition in banking or market concentration represented by variable Comp is measured by taking the combined annual asset value of the three largest banks or the five largest banks12, divided by the total annual asset value of all commercial banks for each country. The annual median value of each country is calculated and the countries are categorized as above or below the median. Below median values are indicated as relatively competitive markets with the value one. Above median value countries are the base group with a value zero. Due to unavailable monthly data I use annual data. Each year of each country is categorized above or below median value. This yields that there is hardly change in the country time series. Unfortunately there is no data available for the years 2019 and 2020. Because there is hardly change over time in the value above or below the median within countries, I make the assumption that the competitiveness of a country in the years 2019 and 2020 remains the same as in the previous years. Ultimately, a county is either indicated above or below median value with respect to competition. Despite monthly data being ideal, yearly data gives still a quite accurate reflection of the competitive position per country as relative competitiveness is not expected to change much within years.
The degree of deposit funding is measured as the share of all household deposits relative to total liabilities at the monthly level and reflects the degree of deposit funding relative to other
11 See De Bondt (2005) and Kleimer & Sander (2004) for instance.
21 funding sources. Again, the median value of all countries is calculated for each month and indicated as above or below median13. The banks of the countries in the former group attract relatively more deposits for funding and are indicated with a one, the other group forms again the base group. Deposit funding is represented by variable Depf.
Figures 3: pass-through correlations versus market concentration of sample countries.
Figure 4: pass-through correlations versus deposit funding of sample countries
The correlations between retail deposit rates and the short-term market rate for each of the 17 sample countries are plotted against the average degree of market concentration and degree of deposit funding, in figure 3 and 4 above. In figure 3, the correlations weaken with decreasing market concentration, which is in line with the theory stating that competition leads to price
22 taking rather than price setting. Figure 4 illustrates that the pass-through correlation decreases with the degree of deposit funding, as demonstrated in prior research. The choice for using indicators for competition an deposit funding instead of continuous data is among others because there is only annual data available for competition and because a three-way interaction with continuous data is hardly interpretable and not meaningful.
Variable Country is a dichotomous variable which has either a value one or zero for the whole range of months of a given sample country. It has variable one for the country whose pass-through is measured in relation to the whole group as in model (4).
Finally, I use three variables to control for their potential disturbing effects. The APP (App) is measured as the monthly change in purchases of the whole program in billion euros, consisting of the corporate sector purchase programme (CSPP), public sector purchase programme (PSPP), asset-backed securities purchase programme (ABSPP) and the third covered bond purchase programme (CBPP3). The money market volatility (Vol) is measured as the logarithmic14 change of the standard deviation of monthly short-term overnight interbank rates over a rolling period of the previous 12 months. Inflation (Infl) is measured as monthly growth rate of the price level according to the consumer price index excluding energy and food prices standard.
5.3. Concise data statistics
Table 1 below summarizes the data statistics of the variables described in this section. The full sample is an unbalanced panel due to the missing observations for the two different deposit rate maturities of Switzerland and Denmark and dropping the first observations of each country its time series as a result of differencing. The number of observations for this study is relatively large which is favorable when splitting the sample using the post-bound indicator, as the separate periods will still have sufficient observations. The interest rate covariates show broad ranges between the minimum and maximum values, illustrating quite some difference in the movement of the variables over time. For instance, during the sample period the monthly changes in the monetary policy market rate and the cost of fund market rate are showing significant decreases of far over a hundred basis points. This is mainly due to rapidly collapsing market rates during the 2008 financial crisis and the sovereign debt crisis in Europe. This is accompanied by severe money market volatility and a drop in inflation as can be seen from the relatively high standard deviations. What stands out is that standard deviation and the minimum of the monthly change in the overnight deposits are lower compared to those of the other maturity deposit rates. It could be that banks are more careful when setting the overnight deposit rate resulting in weaker changes, especially when retail deposit rates become low. There remain
23 some outliers in the data which are mainly the result of the volatile period at the beginning of the global financial crisis in 2008 and 2009.
Table 1 Descriptive statistics of variables
Variable Obs Mean Std.Dev. Min Max
∆Overnight deposits (DrOn) 2499 -.009 .054 -.63 .45
∆Time deposits up to one year (Dr1Y) 2349 -.027 .169 -1.49 .75
∆Time deposits up to two years (Dr2Y) 2351 -.022 .095 -1.6 .51
∆Monetary policy market rate (MrMP) 2499 -.03 .135 -1.2 .53
∆Cost of fund market rate (MrCF)
∆Asset purchasing program (App)
2499 2499 -.034 .114 .15 8.009 -1.804 -24.66 .618 49.03
∆Money market volatility (Vol) ∆Inflation (Infl) 2499 2499 -.008 -.029 .359 .394 -2.54 -2 2.37 1.65
Competition indicator (Comp) 2516 .529 .499 0 1
Deposit funding indicator (Depf) 2516 .444 .497 0 1
6. Pass-through analysis results
This section starts with a short analysis to explore whether there is a structural break in the relationship between the monetary policy rate and the overnight deposit rate. The results of estimating the empirical models described in section 4 will be presented and analyzed in the following section. In order to test the consistency of the results, several robustness checks are conducted to evaluate the results for different specifications and assumptions of the models. 6.1 Exploring breakpoints
To differentiate between two interest rate environments, I explore the relationship between the monetary policy rates and overnight deposits for structural breaks in time series of individual countries. These are points in time where time series abruptly change which could be caused by a significant change in the parameters of the process that produces the series. As discussed in section 4, the bound that discriminates between the periods with positive interest rates and negative interest rates is set such that it equals the point where the monetary policy rate first entered negative territory, the introduction of NIRP. I use this analysis as preliminary check to validate whether the points of entering NIRP correspond with a structural break.
24
1993; Quandt, 1960). Besides those already mentioned, there are also other significant events which occurred during the the sample period15 and the time series reveal multiple breakpoints.
The first breakpoints correspond with the emergence of the Greek government debt crisis in late 2009 and the subsequent European sovereign debt crisis. This breakpoint is rather caused by an incidental spike in interbank market risk than a persistent decoupling of retail deposit rates from the monetary policy rates. Other breakpoints are found around the point where the overnight short-term market rate becomes negative and this holds for all sample countries. At these points, the retail deposit rates are still just above zero percent (See appendix E for the estimated break range for each country). This finding provides an indication for the broken relationship between the monetary policy rates and retail deposit rates.
Using this first clue, the next section explores the actual pass-through relationship while controlling for the effects of other covariates.
6.2. Full sample pass-through results
Table 2 below reports the results from estimating model (2) which captures the full panel pass-through for both the monetary policy approach and the cost of fund approach. First of all, the number of observations reported at the bottom of the table deviate from those in table 1. This is the result of taking lags of the market rate variables. The lag order for market rates is typically one or two months (De Bondt, 2005). He performs a lag structure test based on Akaike information criterion (AIC), Hannan–Quin and Schwartz criteria and finds an optimal lag order of two months. Because finite distributed lag models are sensitive for the number of lags included, I follow this approach to find the optimal lag length n* that minimizes AIC and the number of lags (Hill, Griffiths & Judge, 2001). This results in taking two lags of variable Mr which holds for both the monetary policy approach and cost of fund approach. The coefficients of the contemporaneous term and the lags are jointly representing the pass-through. I will account for heteroscedasticity by using robust standard errors while estimating models (2), (3) and (4).
The model F-statistics for both regressions are significant, indicating that the covariates provide an improvement to the fit of the data relative to the intercept-only model. In addition, the coefficient of determination (R2)values are relatively low which points towards a moderate
explanation of the variation in the retail deposit rates. The remainder of the variation is absorbed by the error term which is not reported. The relatively low R2 values could imply that relevant variables are omitted, however, taking into account that the theory and empirical literature learns that deposit rates are not heavily affected by other factors than the monetary policy, it is
25 reasonable to state that the low values are for an important part just the result of differencing the data. The downside of differencing data is that it reduces the variation in the data with a significant lower R2 and larger standard errorsas a result. Estimating level data on the other hand, caused serious autocorrelation in the residuals of preliminary estimates. Cochrane (2018) describes this trade-off as ‘throwing the baby out with the bathwater’ versus efficient estimation. As the residual autocorrelation is severe with level data, differencing is the preferred option.
For the monetary policy approach in column 1, a significant contemporaneous pass-through coefficient of 0.074 as well as its second lag of 0.095, indicate that a one basis point change in the monetary policy rate causes a combined 0.169 basis points change in the retail deposit rate during the pre-bound period. The coefficient of the first lag is insignificant and not reliable for drawing inference. Despite that the deposit rate moves in the same direction as the monetary policy rate, the magnitude of the transmission is already weak during the period prior to NIRP. Looking at the coefficients of the multiplicative terms of the monetary policy rate with the post-bound indicators in the same column, the contemporaneous coefficient and the first lag are insignificant. However, a significant second lag coefficient of -0.119 means that the post-bound pass-through is 0.119 basis points weaker compared to that of the pre-bound pass-through. This yields that the post-bound pass-through is only 0.05 basis points after a basis point change in the monetary policy rate and significantly weakened. Moving on to the cost of fund approach in column 3, the results vary in the magnitude of the pass-through coefficients. The contemporaneous pre-bound pass-through coefficient of 0.085 as well as its second lag coefficient of 0.088 are statistically significant and add up to a combined pre-bound pass-through of 0.173 basis points change in the deposit rate after a basis point change in the marginal cost rate. This pre-bound pass-through is comparable to the one measured according to the monetary policy approach. The sign of the coefficients is positive as expected, but the strength of the transmission is again quite limited. The contemporaneous post-bound multiplicative term in the same column as well as its first lag are insignificant but the second lag reports a significant coefficient of -0.154. Comparing this coefficient to that of the base group, the post-bound pass-through is thus 0.154 basis points weaker during the post-bound period compared to the pre-bound pass-through.
The deposit funding indicator also reports a significant coefficient of 0.005, indicating that the average change in the deposit rate is 0.005 basis points higher for high deposit banks relative to low deposit banks. The intercept shows an expected significant downward trend.
26 Table 2 Full panel pass-through to overnight deposits by model (2) and (3)
Monetary policy approach Cost of fund approach Column Model (1) (2) (2) (3) (3) (2) (4) (3) VARIABLESa ∆Overnight deposits ∆Overnight deposits ∆Overnight deposits ∆Overnight deposits
∆Market ratei,t 0.074** 0.098* 0.085*** 0.111***
∆Market ratei,t-1 0.007 0.045 -0.006 0.009
∆Market ratei,t -2 0.095*** 0.082* 0.088*** 0.084**
Post-bound dummyi,t 0.003 0.002 0.003 0.003
∆Market rate*boundi,t -0.196 -0.267
∆Market rate*boundi,t-1 -0.194 0.053
∆Market rate*boundi,t-2 -0.119** -0.154**
Compi
∆Market rate*compi,t
∆Market rate*compi,t-1
∆Market rate*compi,t-2
∆Market rate*comp*boundi,t
-0.000 -0.259 0.002 -0.005 0.027 0.021 -0.000 -0.350 0.002 -0.003 0.048 -0.004
∆Market rate*comp*boundi,t-1
∆Market rate*comp*boundi,t-1
-0.179 0.024
-0.026 0.057
Deposit funding dummyi,t
∆Market rate*deposit fundi,t
∆Market rate*deposit fundi,t-1
∆Market rate*deposit fundi,t-2
0.005** -0.000 -0.051 -0.102* -0.014 0.005** -0.000 -0.073 -0.086 0.018
∆Market rate*deposit fund*Boundi,t -0.049 -0.063
∆Market rate*deposit fund*Boundi,t-1 -0.055 0.038
∆Market rate*deposit fund*Boundi,t-2 0.089* 0.089*
∆Appi,t
∆Money market volatilityi,t
∆Inflationi,t Constant -0.000 0.004 -0.000 -0.007** -0.000 0.004 -0.001 -0.006** -0.000 0.004* -0.001 -0.007** -0.000 0.003 -0.001 -0.006** Observations 2465 2465 2465 2465 R2 Number of countries Model F-statistic 0.150 17 3.88*** 0.174 17 4.84*** 0.160 17 4.42*** 0.187 17 5.84*** *** p<0.01, ** p<0.05, * p<0.1
27 While the weak pass-through to overnight deposits is in line with most literature, the difference found here indicates that after entering NIRP, the pass-through has weakened significantly further. The severe weakening in the pass-through after the introduction of NIRP, suggests that complete unresponsiveness of the deposit rate is expected to be found at the point where the deposit rates are zero. Testing this is not possible because it requires to measure the post-bound at the very end of the sample period, resulting in too limited observation for the post-bound period and unreliable inference.
There is no indication that competition affects the pass-through after the market rates have become zero and entered negative territory, as the indicator and multiplicative terms are insignificant in columns 1 and 3. The same conclusion can be drawn for the influence of deposit funding on the post-bound pass-through.
The effects of competition and deposit funding over the full sample period are estimated by model (3) and reported in columns 2 and 4 of table 2. Estimating them directly by model (2) would require to add two-way interactions of Mr with Comp and Mr with Depf. Preliminary estimates led to very high collinearity and inflated standard errors, regardless of the way the variables are measured. The panel structure improves the efficiency of estimates due to the relatively high number of observations, reducing collinearity among explanatory variables. On the other hand, including multiple lags, multiplicative terms and its constitutive elements in the same equation often raises the probability of high collinearity (Brambor, Clark & Golder, 2006; Hill, Griffiths & Judge, 2001). Measuring competition and deposit funding using continuous data could unfortunately not solve the collinearity problems. A formal test (See appendix F) points out that issues following from high collinearity can safely be ignored. While the pass-through coefficients are only weakly significant in column 2, column 4 reports a significant pass-through of 0.195 basis points after a basis point change in the marginal cost rate over the full sample period. The coefficients of the multiplicative terms with competition and deposit funding which were of primary interest in estimating model (3) are not found to be significant.
6.3. Pass-through results for individual countries
Table 3 below displays the outcomes of estimating model (4) that includes the three-way country dummy interaction with the monetary policy rate and the post-bound indicator. The model F-statistics indicate a proper fit of the data and the R2 values are comparable to the full sample R2 values from table 2. Again, severe collinearity issues arise when each country dummy is represented in one regression. To overcome this problem, model (4) is estimated 17 times, each time including the country dummy of one country.
28 indicator including all lags, there is no difference across countries, which is straightforward because it is measured as the panel pass-through which is the same way for each country. The three-way interaction with the country dummy measures the deviation in the post-bound pass-through of a specific country with respect to the panel post-bound pass-pass-through. For Latvia, the post-bound through is -0.177 basis points weaker compared to the panel post-bound pass-through. Hence, the sluggishness in the pass-through after NIRP is stronger in Latvia. For Slovakia, a significant post-bound coefficient indicates that the post-bound pass-through is 0.189 basis points stronger compared to the full panel post-bound pass-through.
Except for these countries, there is no significant difference in the pass-through found for individual countries. For the EMU countries which do not have a very well-integrated banking sector, this largely homogeneous pass-through is surprising. These findings are not in line with previous work such as Sørensen & Werner (2006) who find clear differences in the transmission to deposit rates for individual EMU countries at the beginning of the century. In addition, more recently Holton & Rodriguez d’Acri (2018) find large differences in the pass-through to lending rates across European banks after the Eurozone crisis. The difference with the former finding might be explained by the gap in the timeline and the fact that the European banking integration was in an earlier stage and far before the establishment of for instance the banking union16. The
latter authors do focus on a recent period but still report significant heterogeneity between European banks.
In addition to the pass-through variables, the deposit funding indicator coefficient is again found to be significant with a coefficient of 0.05.
Following the cost of fund approach and replacing the monetary policy rate with the marginal cost rate, the results as presented in table 4 below remain stable. The pre-bound pass-through coefficients are slightly higher compared to those estimated by the monetary policy approach, as was also the case when estimating model (2). Also the post-bound pass-through is still weaker for the full panel of countries. Significant deviations in the post-bound pass-through are again only found for Latvia and Slovakia and are comparable to those found following the monetary policy approach. The largely homogeneous pass-through leads to the rejection of hypothesis 1c that the pass-through of negative policy rates is heterogeneous across countries.
29 Table 3 Individual country pass-through to overnight deposits following the monetary policy approach
Model (3) Column (1) (2) (3) (4) (5) (6) (7) (8) (9) Country VARIABLES AUS ∆Overnight deposits CYP ∆Overnight deposits GER ∆Overnight deposits SPA ∆Overnight deposits FIN ∆Overnight deposits FRA ∆Overnight deposits GRE ∆Overnight deposits LAT ∆Overnight deposits LIT ∆Overnight deposits ∆Market ratei,t
∆Market ratei,t-1
∆Market ratei,t -2
Post-bound dummyi,t
0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003*
∆Market rate*boundi,t 0.063 0.063 0.063 0.063 0.063 0.063 0.063 0.063 0.063
∆Market rate*boundi,t-1 0.105 0.105 0.105 0.105 0.105 0.105 0.105 0.105 0.105
∆Market rate*boundi,t-2 -0.095*** -0.095*** -0.095*** -0.095*** -0.095*** -0.095*** -0.095*** -0.095*** -0.095***
∆Market rate*country*boundi,t
∆Market rate*country*boundi,t-1
∆Market rate*country*boundi,t-2
Compi
Deposit funding dummyi,t
-0.008 -0.047 0.049 0.001 0.005** -0.008 -0.047 0.049 0.001 0.005** -0.088 -0.159* 0.049 0.001 0.005** -0.009 -0.049 0.002 0.001 0.005** -0.054 -0.045 0.083 0.001 0.005** -0.201 -0.123 -0.073* 0.001 0.005** -0.045 -0.029 0.118 0.001 0.005** -0.129 -0.177** -0.077 0.001 0.005** -0.107 -0113 0.024 0.001 0.005** ∆Appi,t
∆Money market volatilityi,t
30 Table 3 (continued) Column (10) (11) (12) (13) (14) (15) (16) (17) Country VARIABLES LUX ∆Overnight deposits NED ∆Overnight deposits POR ∆Overnight deposits SLN ∆Overnight deposits SLK ∆Overnight deposits SWE ∆Overnight deposits SWI ∆Overnight deposits DEN ∆Overnight deposits ∆Market ratei,t
∆Market ratei,t-1
∆Market ratei,t -2
Post-bound dummyi,t
0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003* 0.075*** 0.007 0.095**** 0.003*
∆Market rate*boundi,t 0.063 0.063 0.063 0.063 0.063 0.063 0.063 0.063
∆Market rate*boundi,t-1 0.105 0.105 0.105 0.105 0.105 0.105 0.105 0.105
∆Market rate*boundi,t-2 -0.095*** -0.095*** -0.095*** -0.095*** -0.095*** -0.095*** -0.095*** -0.095***
∆Market rate*country*boundi,t
∆Market rate*country*boundi,t-1
∆Market rate*country*boundi,t-2
Compi
Deposit funding dummyi,t
∆Appi,t -0.101 -0.002 -0.021 0.001 0.005** -0.000 -0.127 0.139 -0.019 0.001 0.005** -0.000 -0.182 -0.056 -0.029 0.001 0.005** -0.000 -0.125 -0.079 0.074 0.001 0.005** -0.000 -0.076 -0.006 0.189** 0.001 0.005** -0.000 -0.129 -0.112 0.013 0.001 0.005** -0.000 -0.226 -0.154 0.012 0.001 0.005** -0.000 0.429 0.333 0.072 0.001 0.005** -0.000
∆Money market volatilityi,t
31 Table 4 Individual country pass-through to overnight deposits following the cost of fund approach
Model (3) Column (1) (2) (3) (4) (5) (6) (7) (8) (9) Country VARIABLES AUS ∆Overnight deposits CYP ∆Overnight deposits GER ∆Overnight deposits SPA ∆Overnight deposits FIN ∆Overnigh t deposits FRA ∆Overnight deposits GRE ∆Overnight deposits LAT ∆Overnight deposits LIT ∆Overnight deposits ∆Market ratei,t
∆Market ratei,t-1
∆Market ratei,t -2
Post-bound dummyi,t
0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003
∆Market rate*boundi,t 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071
∆Market rate*boundi,t-1 0.049 0.049 0.049 0.049 0.049 0.049 0.049 0.049 0.049
∆Market rate*boundi,t-2 -0.107*** -0.107*** -0.107*** -0.107*** -0.107*** -0.107*** -0.107*** -0.107*** -0.107***
∆Market rate*country*boundi,t
∆Market rate*country*boundi,t-1
∆Market rate*country*boundi,t-2
Compi
Deposit funding dummyi,t
-0.019 0.025 0.029 0.001 0.004** 0.178 0.177 -0.028 0.001 0.004** -0.116 -0.018 0.019 0.001 0.004** 0.013 0.052 -0.024 0.001 0.004** -0.084 0.014 -0.075 0.001 0.004** -0.226* -0.079 -0.034 0.001 0.004** -0.052 -0.032 0.175 0.001 0.004** -0.153 -0.147** -0.049 0.001 0.004** -0.123 -0.061 0.016 0.001 0.004** ∆Appi,t
∆Money market volatilityi,t
32 Table 4 (continued) Column (10) (11) (12) (13) (14) (15) (16) (17) Country VARIABLES LUX ∆Overnight deposits NED ∆Overnight deposits POR ∆Overnight deposits SLN ∆Overnight deposits SLK ∆Overnight deposits SWE ∆Overnigh t deposits SWI ∆Overnight deposits DEN ∆Overnight deposits ∆Market ratei,t
∆Market ratei,t-1
∆Market ratei,t -2
Post-bound dummyi,t
0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003 0.085*** -0.007 0.089**** 0.003
∆Market rate*boundi,t 0.071 0.071 0.071 0.071 0.071 0.071 0.071 0.071
∆Market rate*boundi,t-1 0.049 0.049 0.049 0.049 0.049 0.049 0.049 0.049
∆Market rate*boundi,t-2 -0.107*** -0.107*** -0.107*** -0.107*** -0.107*** -0.107*** -0.107*** -0.107***
∆Market rate*country*boundi,t
∆Market rate*country*boundi,t-1
∆Market rate*country*boundi,t-2
Compi
Deposit funding dummyi,t
∆Appi,t -0.079 0.026 0.025 0.001 0.005** -0.000 -0.277* 0.099 0.031 0.001 0.005** -0.000 -0.158 -0.030 0.003 0.001 0.005** -0.000 -0.136 -0.049 0.089 0.001 0.005** -0.000 -0.126 0.037 0.240** 0.001 0.005** -0.000 -0.144 -0.029 0.040 0.001 0.005** -0.000 -0.312 -0.023 0.038 0.000 0.005** -0.000 0.563 -0.001 0.031 0.000 0.005** -0.000
∆Money market volatilityi,t
