The effect of the 2008 financial crisis on the interest pass-through rate : the case of the Netherlands

The effect of the 2008 financial crisis on the interest pass-through rate:

the case of the Netherlands.

Marieke van Brussel (10663827) University of Amsterdam

Economics and Business Bachelor Thesis Specialization: Economics and Finance

Supervisor: D.H.J. Chen June 2016

Abstract In response to the 2008 financial crisis, the European Central Bank lowered its policy rate in order to stimulate the economy. However, the efficiency of this monetary policy is dependent on the functioning of the interest rate pass-through model. Though previous studies on the Euro area concluded the crisis to have impaired the interest rate pass-through, these results cannot be assumed to hold for individual Euro countries. This paper therefore examines the effect of the 2008 financial crisis on the interest pass-through rate in the Netherlands and consequently compares the results with previous empirical literature on the Euro area. To this end, data on the retail and policy rate from 2003 to 2014 was collected. Consequently, an Error Correction model was used to estimate the short- and long-run pass-through dynamics for the pre- and post-crisis period. The results

indicate that the 2008 financial crisis impaired the interest rate pass-through model in the Netherlands, resulting in a less efficient monetary policy. However, a comparison with results of previous empirical research indicated that it was to a lesser extent affected than the Euro area.

Statement of originality

This document is written by Marieke van Brussel who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

1. Introduction

In September 2008 the collapse of the Lehman Brothers bank took place and, according to Sathye, this indicated the beginning of the 2008 financial crisis (2013). In response to this the European Central Bank (ECB) started decreasing its policy rate, bringing it to an all-time low of zero percent in March 20161. This was done in a continuous effort to stimulate the economy that had significantly weakened as a result of the financial crisis (Paries et al., 2014). However, the Central Bank cannot directly affect the output of the economy, but does so through its monetary policy channels. By setting their official rate they influence the retail rate, asset prices, exchange rate, expectations, and consequently, the demand in the economy (Bank of England [BoE], 1999). The transmission from policy to retail rate is considered to be the most important one of these channels, especially in the Euro area where banks are the main source of finance for non-financial corporations (Aristei & Gallo, 2013). By lowering its policy rate, the Central Bank lowers the cost at which commercial banks can lend and, through this, the rates charged by commercial banks. This boosts the spending behaviour of consumers and firms, as the time value of money decreases, and consequently increases total demand in the economy (Aristei & Gallo, 2013). The effectiveness of the Central Bank’s policy in boosting demand is thus dependent on the extent to which changes in the policy rate are passed-through to the retail rate. This is commonly known as the interest pass-passed-through rate.

A simultaneous graphing of the policy and retail rate shows that after the ECB started lowering its policy rate in response to the financial crisis, the spread between the rates increased from being approximately 0.75 percentage points to varying around 1.75 percentage points. The increased spread indicates that the banks were less willing or able to lower their rates (Hristov et al., 2014). This raises the question whether the pass-through has significantly decreased during the crisis period, which would imply a less efficient and effectively weaker monetary policy of lowering interest rates during a time at which the ECB relies heavily on precisely such policy measures to re-establish financial stability. Several studies have therefore analysed the effect of the 2008 financial crisis on the interest pass-through rate in the Eurozone (Aristei & Gallo, 2013; Paries et al., 2014; Hristov et al., 2014; Bank of International Settlements [BIS], 2013). The overall results show that the pass-through decreased and that monetary policy consequently became less efficient. In response to this, the ECB adopted additional unconventional monetary policy tools to further stimulate the economy (ECB, 2009).

1

Figure 1; retail and policy rate2

However, it has also been shown that there exists heterogeneity across Euro area countries when it comes to the interest rate pass-through and that this heterogeneity increases during times of financial distress (Aristei & Gallo, 2013). Results on the Euro area therefore do not necessarily hold for individual Euro countries. Consequently, it is of relevance to individual countries to examine the effect of the crisis on the interest rate pass-through. Since the Eurozone is characterised by a common monetary policy, a comparison of these findings with results for the entire Euro area can assist the ECB in determining which additional measures to take and where these are most needed. Such empirical research has been done for Italy, Spain and Germany (BIS, 2014; de Sola Perera & van Nieuwenhuyze, 2014), all being in the top five European countries with the highest GDP3. However, though the Netherlands displays the fifth highest GDP of all members of the European Monetary Union, the effect of the crisis on its interest pass-through rate has to this date not been addressed. This paper will therefore extent previous research by looking at the Netherlands in specific. The primary research question of this papers is as follows;

Has the 2008 financial crisis significantly affected the interest pass-through rate in the Netherland?

2 _{A detailed explanation of the data used in this graph is provided in section 3} 3

Based on data obtained from

http://ec.europa.eu/eurostat/tgm/refreshTableAction.do?tab=table&plugin=1&pcode=tec00001&language=en 0,00 0,50 1,00 1,50 2,00 2,50 3,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00 ja n u ar i-03 au gu stu s-03 m aa rt-04 o kt o b er-04 m ei-05 d ec em b e r-0 5 ju li-06 fe b ru ari -07 se p tem b er-07 ap ril-08 n o ve m b e r-08 ju n i-09 ja n u ar i-10 au gu stu s-10 m aa rt-11 o kt o b er-11 m ei-12 d ec em b e r-1 2 ju li-13 fe b ru ari -14 se p tem b er-14 Perc en ta ge p o in ts Perc en t

In addition to this, it is relevant to compare these findings with the effect the 2008 financial crisis had on the pass-through of the Eurozone. Since literature on the effect of the crisis on the pass-through mechanism in the Euro area is widely available, this paper will furthermore address how the results found for the Netherlands compare with previous findings on the Euro Area. To this end, an Error Correction Model (Engle & Granger, 1987) is used to first estimate the interest rate pass-through model for the pre-crisis period, providing a detailed explanation of its functioning in the Netherlands and how this contrasted to the Euro area. This contributes to a better understanding of any

differences in reaction to the crisis that might occur. Next, the crisis period is incorporated into the model allowing for an analysis of its effect on the pass-through rate. The results are consequently compared with the findings of previous empirical research on the Euro area.

The remainder of this paper consist of six sections. In section 2 an overview of previous literature on the interest rate pass-through and the effect of the financial crisis will be provided. Next, section 3 will describe the data used in this empirical research. Section 4 illustrates and explains the method used. Consequently, the results obtained will be presented and discussed in section 5. Finally, section 6 will conclude this paper.

2. Literature review

This section will first provide an overview of the factors influencing the pass-through and consequently discusses the results found by previous empirical research.

The main objective of the European Central Bank is achieving a stable inflation of

approximately two percent (ECB, 2009). As mentioned in the introduction, the interest rate pass-through is one of the monetary policy transmission mechanisms (MPTM) pass-through which the ECB can influence inflation. Moreover, as the Euro area has a bank-based financial system, it is considered to be the most important transmission mechanism. Literature generally distinguishes between the immediate and final pass-through4, thereby allowing for an analysis of the short- and long-run efficiency of the MPTM. In a perfectly competitive market the pass-through will be complete, resulting in an efficient MPTM. However, due to the existence of market imperfections, this is often not the case. The main causes of an ineffective MPTM are the existence of switching costs (Ahmad et al., 2013) and information asymmetries (Sander & Kleimeier, 2004), both leading to banks having a certain degree of market power.

4 _{The immediate and final pass-through are also referred to as the short- and long-run pass-through}

Moreover, previous literature has shown the pass-through to be dependent on several additional factors. Aristei and Gallo (2013) prove that the pass-through is dependent on both the type and size of the loan. Their research indicates that loans to corporations display a higher pass-through than loans to consumers as information on firms is better accessible, thereby reducing asymmetric information. In addition to this, the pass-through also increases as the size of the loan grows. The two main reasons for this are a decline in asymmetric information, due to the increased benefit of collecting information, and a higher degree of competition between banks. Sander and Kleimeier (2004) conclude that the characteristics of the national financial market as well as the growth within a country affects the degree of pass-through. Likewise, countries’ legal and cultural differences also attribute to variation in the pass-through rate. This implies the existence of heterogeneity in the pass-through rate across the Euro area.

Empirical analysis of the interest pass-through rate has been performed for individual countries such as Ireland (Bredin et al., 2002), the United Kingdom (Ahmad et al., 2013), Australia (Sathye, 2013), the Netherlands (Bernhofer & van Treeck, 2013; Sorensen & Werner, 2006),

Germany, Italy and Spain (BIS, 2014; de Sola Perera & van Nieuwenhuyze, 2014). In addition to this, several papers have applied analyses to the Euro area as a whole (Aristei & Gallo, 2013; De Bondt, 2005; Paries et al., 2014; Hristov et al., 2014). Furthermore, empirical research differs in the retail rate used when examining the interest rate pass-through. These can broadly be classified into papers using rates on loans to consumers and those using rates on loans to non-financial corporations (NFC).

There are two main approaches to modelling the interest rate pass-through. First, the Error Correction Model introduced by Engle and Granger (1987), which allows for inclusion of both the short- and long-run characteristics. Second, the Vector Error Correction Model (Johansen, 1991), extending the ECM by allowing for interdependence between the variables. Despite the

heterogeneity of the target countries, retail rate, and the approaches used to model the pass-through, existing literature commonly shows an incomplete short-run pass-through (Ahmad et al., 2013; Hristov et al., 2014; Belke et al., 2013;).

The long-run through is consistently found to be higher than the immediate pass-through. This is for several reasons. First, the benefit of collecting information increases when a longer period is taken into consideration therefore decreasing the amount of asymmetric

information. Second, corporations often have alternative sources of finance available and though the benefit of shifting to this source does not compensate for the switching costs in the short-run, it does in the long-run when a decrease in the policy rate is consistently not passed-through. Third, entry into the market is not possible in the short-run whilst in the long-run it is, thereby diminishing the degree of market power. Lastly, in the short-run it is often unsure whether the changes in policy rate are permanent or temporary. Due to the existence of menu costs, banks make predictions on the

future value of the policy rate. This forward-looking behaviour can result in an incomplete pass-through in the short-run, whereas it does not affect the final pass-pass-through (Cottarelli and Kourelis, 1994 ).

Even though the long-run through is found to be higher than the immediate pass-through, there is no consensus on whether it is complete or incomplete. De Bondt (2005) and Cordemans and de Sola Perera (2011) used the rate on loans to NFC as a proxy for the retail rate and found a complete long-run pass-through. The same goes for Belke et al. (2013), who used the rate on loans to NFC of at least one million euros. On the contrary, an incomplete final pass-through was found by Ahmad et al. (2013) and Bredin et al. (2002), who made use of the rate on loans to consumers as a representation of the retail rate. This is consistent with the findings of Aristei and Gallo (2013), who showed the pass-through to depend on the type as well as the size of the loan. A comparison of the interest rate pass-through across different Euro area countries is provided by Sander and Kleimeier (2004), and confirms the existence of heterogeneity.

In addition to research on the interest-rate pass-through, several papers have examined the effect of the 2008 financial crisis (Hristov et al., 2014; Ahmad et al., 2013; Sathye, 2013). Ahmad did so by forecasting the change in retail rates using the pass-through model determined in the pre-crisis period and comparing the findings with the actual occurred changes. Sathye tested for structural breaks in the model to identify a significant change, whereas Hristov examined the change in the spread to show that the retail rate’s deviation from the policy rate significantly increased in the post-crisis period. The results show that the post-crisis has significantly impaired the pass-through mechanism, as it caused an increase in credit risk and information asymmetries. Furthermore, Belke et al.

concluded that the heterogeneity across countries also increased in response to the crisis.

However, the methods used do not allow for an analysis of the effect on different aspects of the mechanism. Aristei and Gallo (2013) therefore extended previous literature by comparing the pass-through model under normal conditions with the model under high volatility. Nevertheless, they only incorporated the short-run characteristics into the model and, though they found the immediate pass-through to have significantly decreased, no result for the long-run pass-through was provided. De Sola Perera and van Nieuwenhuyze (2014) and the Bank for International Settlement (2014) did include the long-run characteristics, whereas the short-run characteristics were not addressed. They both concluded the 2008 crisis to have impaired the long-run pass-through by significantly increasing credit risk.

All in all, the interest rate pass-through and the effect of the 2008 financial crisis have been widely discussed in literature, where the general consensus is that it led to an impairment of the monetary policy transmission mechanism. As there exist heterogeneity in the Euro area, which has increased as a result of the crisis, it is of relevance to study the pass-through rate of individual

countries. This paper will therefore contribute to existing literature by analysing the effect of the 2008 financial crisis on the pass-through rate in the Netherlands and addressing both the short- and long-run characteristics.

3. Data

The dataset used consists of monthly data on the retail and policy rates from January 2003 to

December 2014. This period was chosen as it contains the 2008 financial crisis, including enough data of the pre- and post-crisis period. A total of 143 observations was collected, of which 68 are pre-crisis and 76 post-crisis5.

The key policy rate used by the ECB is the rate on main refinancing operations (MRO), however, this rate changes only infrequently. As the model of the interest rate pass-through uses first differences6, a different proxy for the policy rate is needed when performing the empirical analysis (De Bondt, 2005). Aristei and Gallo (2013) proposed the use of the EURIBOR, which is determined by the average interest rate a panel of the biggest banks in the Euro zone charge each other. Thus whereas the rate on MRO determines the cost at which banks can borrow from the ECB, the Euribor determines at which cost they can borrow from each other. Figure 2 shows a

simultaneous graphing of the Euribor and the rate on Main Refinancing Operation. Up until September 2008 both rates are approximately equal with an average disparity of only 0.05

percentage points. However, after this, the spread between them increases reaching a maximum of 0.684 percentage points in March 2012. The reason for this is that the ECB reacted to the crisis by not only lowering their policy rate, it also implemented unconventional measures to further lower the cost of financing for banks, thereby decreasing the Euribor even more (Hristov et al., 2014). In this empirical research, the benefit of not using an infrequently changing rate outweighs the downside of a non-perfect correlation with the rate on Main Refinancing Operations. In addition to this, it is appropriate to use short-term rates in order to avoid complications that can arise due to term structures (Sander & Kleimeier, 2004). This paper therefore uses the 1 week Euribor obtained from Datastream as a representation of the policy rate.

5 _{The starting point of the 2008 financial crisis is addressed in section 3.1} 6

Figure 2; Euribor MRO comparison

The interest pass-through rate is considered to be especially relevant in the Euro area as commercial banks play a leading role in providing loans to non-financial (European Central Bank [ECB], 2009). This paper therefore uses the rate on new loans to NFC of at least 1 million euros as representation of the retail rate7. The data was obtain from statistics published by “De Nederlandse Bank”, where the rates are calculated by taking the average rate weighted by the size of the new contract.

The data of both rates has been plotted in figure 1. Additionally, table 1 in the appendix provides a summary of the statistics.

3.1 Structural break

As the financial crisis period is included in the dataset, the possibility of a structural break exists. This would denote a shift in the data caused by an exogenous factor (Ahmad et al., 2013). It is of

importance to identify the occurrence of a structural change as it indicates when the model shifted and thus the start of the crisis took place. Referring back to figure 1, the significant increase in the spread after September 2008, gives the impression that a structural break took place. To formally test if this is indeed the case, the Wald test for structural change (Andrews & Fair, 1988) has been used. The results are summarized in table 2 and indeed confirm the occurrence of a structural break on September 2008. This paper therefore uses this date as the beginning of the 2008 financial crisis.

7 _{When using the rate on loans to Non-Financial Cooperation, only data on loans over 1 million Euros was}

available 0 1 2 3 4 5 6 1- 6-2000 1- 3-2001 1- 12-200 1 1- 9-2002 1- 6-2003 1- 3-2004 1- 12-200 4 1- 9-2005 1- 6-2006 1- 3-2007 1-12 -20 0 7 1- 9-2008 1- 6-2009 1- 3-2010 1- 12-201 0 1- 9-2011 1- 6-2012 1- 3-2013 1- 12-201 3 1- 9-2014

P

erc

en

t

Table 2, Wald test for structural change

Sample Break date H0 Chi-squared statistic

2003m2 - 2014m12 2008m9 No structural break 33.0187***

Note [a]: *, **, ***, denote a rejection of the null hypothesis at a 10, 5, and 1 percent significance level respectively

3.2 Stationarity

A dataset of interest rates consists of time series data and are in literature commonly identified as non-stationary (Bredin et al., 2002; Ahmad et al., 2013; Aristei & Gallo, 2013). This means that the interest rates fluctuates around a long-term movement over time, referred to as the stochastic trend, that is dependent on the state of the economy. If two variables that are non-stationary both follow an increasing, or decreasing, stochastic trend at a certain point in time, a regression can indicate that they are related when this is not necessarily the case. The regression is than a spurious one, meaning that the variables have a unit root. To make the regression non-spurious, the series should be transformed into stationary data. This can be done by taking the first differences of the variables, as this eliminates the stochastic trend (Stock & Watson, 2015). When a variable is non-stationary whilst its first difference is, it is said to be integrated of order one.

Figure 3; First differences

note [a]: dRR and dPR refer to the change in retail and policy rate respectively -1,20 -1,00 -0,80 -0,60 -0,40 -0,20 0,00 0,20 0,40 0,60 0,80 ja n u ar i-03 se p tem b er-03 m ei-04 ja n u ar i-05 se p tem b er-05 m ei-06 ja n u ar i-07 se p tem b er-07 m ei-08 ja n u ar i-09 se p tem b er-09 m ei-10 ja n u ar i-11 se p tem b er-11 m ei-12 ja n u ar i-13 se p tem b er-13 m ei-14 Per ce n tage p o in ts dRR dPR

Figure 1 indicates that both the policy and retail rate indeed have a stochastic trend, whilst a graphing of their first differences in figure 3 shows that they do not. In order to formally test this, the Augmented Dickey Fuller test for a Unit Root (Dickey & Fuller, 1979) was performed. The results are reported in table 3. It can be concluded that both the policy and the retail rate are integrated of order one and that their first differences should therefore be used when running a regression.

Table 3, Augmented Dickey Fuller test for Unit Root

Effect H0 Test statistic Stationary

Retail rate Unit root -0.998 no

Policy rate Unit root -0.588 no

Delta retail rate Unit root -12.210*** yes

Delta policy rate Unit root -8.766*** yes

Note [a]: *, **, ***, denote a rejection of the null hypothesis at a 10, 5, and 1 percent significance level respectively

3.3 Cointegration

As discussed, figure 1 displays that both the policy and retail rate have a stochastic trend. This requires the use of their first differences when running a regressing and therefore only provides information on their short-run relationship. However, it also shows that both rates move closely together, suggesting that they have a common stochastic trend. If this is the case, the series are said to be cointegrated, meaning that there exists a long-run equilibrium between both rates to which they revert. This makes it possible to also include the long-run relationship into the regression analysis.

Though figure 1 shows a cointegrated relationship, Andries and Billon (2016) stress the importance of formally testing it. To do so, the Engle-Granger Augmented Dickey Fuller test for cointegration (Stock & Watson, 2015) was used. If the retail and policy rate have a common stochastic trend, a linear combination of them should result in a stationary variable,

𝑅𝑅 − 𝐶 ∗ 𝑃𝑅 (1) 8

where 𝐶 is chosen to eliminate the common trend from their differences and is referred to as the cointegrating coefficient. Though the approximately constant spread suggests that this cointegrating

8 _{RR and PR refer to the retail rate and policy rate respectively. In the remainder of this paper, the shortened}

coefficient is equal to one, it is not certain and should therefore first be estimated by use of a regression.

𝑅𝑅𝑡= 𝐴 + 𝐶 ∗ 𝑃𝑅𝑡+ 𝜀𝑡 (2)

If the retail and policy rate are cointegrated, this equation correspondingly displays the long-run equilibrium to which the retail rate reverts (Engle & Granger, 1987). A is a constant and therefore stationary, whilst 𝑅𝑅 − 𝐶 ∗ 𝑃𝑅 will only result in a stationary variable when the two rates are cointegrated. If this is the case the error term will be stationary as well, since a linear combination of stationary variables results in a stationary variable. After running regression 2, the Augmented Dickey Fuller test (Dickey & Fuller, 1979) was performed to test for stationarity of the error term. The results of the regression and the Augmented Dickey fuller test are summarized in table 4 and 5 respectively. As the results show, the retail and policy rate are cointegrated with a cointegrating factor of 0.723. In addition to this, the constant of 1.789 implies that the retail rate remains approximately 1.8

percentage points higher than the policy rate, which is confirmed by the spread displayed in figure 1. Table 4, Results regression (2)

Variable Coefficient t statistic P > |𝒕| Period Observations R2

𝐴 1.789 (.034) 46.13 0.000 2003M1 – 2014M12 144 0.9374 PR .723 (.016) 52.26 0.001

Table 5, Augmented Dickey Fuller test for Unit Root

Variable Test statistic Stationary

Error term -5.104*** yes

Note [a]: *, **, ***, denote a rejection of the null hypothesis at a 10, 5, and 1 percent significance level respectively

4. Method

The following section will address the method used to answer the research question. First, subsection 4.1 will explain the general interest rate pass-through model and provide the

expectations for the pre-crisis period values. Next, subsection 4.2 will discuss how the financial crisis was incorporated into the model and what the expected effect of the crisis on the interest rate pass-through model is.

4.1 The interest rate pass-through model

In order to capture the effect of a change in the policy rate on the retail rate, this paper uses the Error Correction Model as introduced by Angle and Granger (1987). This model is based on the condition that the variables are cointegrated and stationary in their first differences, which has been proven to hold for the retail and policy rate. It allows for inclusion of the short- and long-run

dynamics and, though it does not take the interdependence of the variables into account, it provides an economically attractive interpretation of the parameters (De Bondt, 2005). The basic form of the Error Correction Model is displayed in equation (3),

∆𝑅𝑅t= 𝜏 + 𝛾𝐸𝐶𝑇t-1+ 𝛽∆𝑃𝑅t+ 𝜀𝑡 (3)

where ECT stands for the Error Correction term. This term is based on the assumption of

cointegration, which confirms that the retail rate reverts to the long-run equilibrium represented by equation (2). The error correction term displays by how much the retail rate deviated from its long-run equilibrium in the previous period, as this influences the change in retail rate in the current period.

𝐸𝐶𝑇𝑡 = 𝑅𝑅t− (𝐴 + 𝐶 ∗ 𝑃𝑅t) (4)

Combining equation (3) and (4) gives;

∆𝑅𝑅t= 𝜏 + 𝛾(𝑅𝑅t-1− (𝐴 + 𝐶 ∗ 𝑃𝑅t-1) + 𝛽∆𝑃𝑅t+ 𝜀_𝑡 (5)

Expanding this formula provides the following equation, which represent the final form Error Correction Model commonly used when studying the interest pass-through rate (De Bondt, 2005).

∆𝑅𝑅t= 𝛼 + 𝛾𝑅𝑅t-1+ 𝛿𝑃𝑅t-1 + 𝛽∆𝑃𝑅t+ 𝜀𝑡 (6)

Equation (6) contains four important aspects of the transmission mechanism that will be discussed in this paper. These are the final pass-through and the mark-up, representing the long-run

characteristics, and the immediate pass-through and speed of adjustment, representing the short-run characteristics.

4.1.1. Long-run Characteristics and pre-crisis expectations

The final pass-through is represented by C and measures to what extent adjustments in the policy rate are passed onto the retail rate’s long-run equilibrium (Ahmad et al., 2013). As equation (6) is formed by expanding equation (5), C can be calculated as follows;

𝐶 = −𝛿_𝛾 (7)

When the long-run pass through is found to be complete, a one percentage point change in the policy rate leads to a one percentage point change in the retail rate’s long-run equilibrium. A complete final pass-through would result in C being equal to one, indicating a perfectly competitive market. The null and alternative hypothesis for this two sided t-test are;

H0: 𝐶 = 1 H1: 𝐶 ≠ 1 (T1)

As this paper investigates the pass-through of the policy rate to the rate on loans to NFC of at least one million euros, the long-run pass-through is expected to be complete and the null-hypothesis should therefore not be rejected.

The second factor determining the long-run relationship is the mark-up, denoted by A. This indicates by how much the retail rate is marked above, or below, the policy rate in the long run. Similar to the long-run pass-through, A is a non-linear combination of two of the regression coefficients found in equation (6);

𝐴 = −𝛼

𝛾 (8)

Whilst the mark-up represents a cost when it concerns deposit rates, it denotes revenue for lending rates and is therefore expected to be bigger than zero. To test this, a one-sided t-test will be

performed, where the alternative hypothesis is expected to hold.

H0: 𝐴 = 0 H1: 𝐴 > 0 (T2)

4.1.2. Short-run characteristics and pre-crisis expectations

The short-run pass-through is represented by 𝛽 and indicates how much of the change in the policy rate is passed through within the same month. When 𝛽 is significantly different from one the pass-through is found to be sticky, indicating a slow response of the retail rate to changes in the policy rate. When this is not the case, the short-run pass-through is said to be complete (Bredin et al., 2002). An incomplete short-run pass-through is, just as the long-run pass-through, caused by the

existence of market imperfection (Ahmad et al., 2013). To test for a complete immediate pass-through, a two sided t-test with the following null and alternative hypothesis will be performed;

H0: 𝛽 = 1 H1: 𝛽 ≠ 1 (T3)

Previous research has shown the short-run pass-through to be lower than the final pass-through, indicating that the existence of market imperfections diminishes in the long-run (De Bondt, 2005). It is therefore expected that the short-run pass-through, contrary to the final pass-through, is

incomplete. This should result in a rejection of the null hypothesis.

When the short- and long-run pass-through are equal to each other and the retail rate is currently at its long-run equilibrium, an adjustment in the policy rate changes the long-run

equilibrium by the exact same amount as the immediate change in the retail rate. The retail rate will therefore remain at its long-run equilibrium and the error correction term will equal zero. However, when this is not the case and either the short- and long-run pass-through display different values or the retail rate is currently not at its long-run equilibrium, the adjustment speed becomes relevant.

The speed of adjustment is represented by the absolute value of 𝛾 and specifies how much of the retail rate’s deviation from its long-run equilibrium in the previous period is corrected in the current period (Ahmad et al., 2013). As the long-run pass-through is expected to be higher than the immediate pass-through, a decrease in the policy rate should result in a larger decrease. Assuming the retail rate was at its long-run equilibrium, this leads to it being above its long-run equilibrium it the current period, resulting in a positive ECT in the next period. For the retail rate to revert to its long-run equilibrium, the ECT should thus be negatively related with the change in retail rate. Consequently, 𝛾 is expected to be negative, making it mean reverting. To test this, a one sided t-test will be used where the null hypothesis is presumed to be rejected;

H0: 𝛾 = 0 H1: 𝛾 < 0 (T4)

4.2 The 2008 financial crisis incorporated into the model

To see how the financial crisis affected the interest rate pass-through model, regression (6) was ran twice. First using the pre-crisis data and consequently using the post-crisis data. These will be referred to as regression (6a) and (6b) respectively. Though the results of these regression allow for an analysis of the pass-through characteristics in both periods, they cannot be used for comparison. Both regressions use a different sample and have therefore different statistical characteristic. Hence, the results cannot be used to test whether the change in coefficients is significant.

In order to be able to formally check if the crisis had a significant effect, a dummy variable is added to equation (6) (Clogg et al., 1995).

∆𝑅𝑅t= 𝛼₃+ 𝛾₃𝑅𝑅t-1+ 𝛿₃𝑃𝑅t-1 + 𝛽₃∆𝑃𝑅t+ 𝐷(𝜃 + 𝜌𝑅𝑅t-1+ 𝜔𝑃𝑅t-1 + 𝜑∆𝑃𝑅_𝑡) + 𝜀_𝑡 (9)

The dummy value equals zero during the pre-crisis period and one after. The found values for a3, y3, d3 and b3 will equal the coefficients found in regression (6a). Additionally, the values of 𝜃, 𝜌, 𝜔 𝑎𝑛𝑑 𝜑 indicate the change of the coefficient in the post-crisis period with respect to the pre-crisis period. If these coefficients are simultaneously equal to zero, the 2008 financial crisis did not have a significant effect on the interest pass-through model. To formally test this an F-test will be performed, as this allows for simultaneous testing of coefficients (Stock & Watson, 2015).

H0: 𝜃 = 𝜌 = 𝜔 = 𝜑 = 0 H1: 𝜃 𝑜𝑟 𝜌 𝑜𝑟 𝜔 𝑜𝑟 𝜑 ≠ 0 (T5) As it has been shown that a structural break occurred at the beginning of the crisis period, H0 is expected to be rejected. Subsequently, the effect on the four aspects of the transmission mechanism should be examined in order to determine in what way the mechanism is affected.

4.2.1 Expectations post-crisis long-run characteristics

As formula (7) and (8) show, the long-run pass-through and the mark-up are a non-linear

combination of coefficients. In order to test for significance, it should therefore be tested if these coefficients are simultaneously equal to zero. If this is the case, their non-linear combination can be concluded to equal zero as well. As a result, the F-test will again be used. This gives the following null and alternative hypothesis when testing for a change in the final pass-through;

H0: 𝜔 = 𝜌 = 0 H1: 𝜔 𝑜𝑟 𝜌 ≠ 0 (T6)

Due to the rise in credit risk and asymmetric information caused by the crisis, the long-run pass-through is expected to decreases. The null hypothesis is therefore expected to be rejected. The hypotheses for testing for a change in the mark-up can be displayed as follows;

H0: 𝜃 = 𝜌 = 0 H1: 𝜃 𝑜𝑟 𝜌 ≠ 0 (T7)

Referring back to figure 1, the spread between the retail and policy rate increased with

approximately 1 percentage point. This one time permanent increase in the spread indicates that the mark-up has significantly increased as a result of the crisis. It is therefore expected that the null hypothesis will be rejected and the alternative hypothesis can be assumed to hold.

4.2.2 Expectations post-crisis short-run characteristics

The change in short-run pass-through is represented by 𝜑. Similar to the final pass-through, the rise in credit risk and asymmetric information caused by the crisis is expected to negatively affect the immediate pass-through. This will be tested with a one sided t-test, where the null hypothesis is expected to be rejected;

H0: 𝜑 = 0 H1: 𝜑 < 0 (T8)

As for the adjustment speed, the increase in asymmetric information in the post-crisis period leads to a slower speed of adjustment (De Bondt, 2005). However, the crisis has also led to a clearer signalling of the Central Bank’s monetary policy stance, thereby improving the banks’ ability to anticipate policy rate changes. This increases their responsiveness, allowing for a faster adjustment towards its long-run equilibrium. These two effects could be offsetting, but previous literature (Aristei & Gallo, 2013) has shown the adjustment speed to increase and this results is expected to hold for the Netherlands. As the adjustment speed is represented by the absolute value of 𝛾3, who’s

sign is expected to be negative, an increase in the speed of adjustment would be represented by a negative 𝜌. By use of a one-sided t-test we will test the following hypotheses, where the alternative hypothesis is expected to hold.

H0: 𝜌 = 0 H1: 𝜌 < 0 (T9)

5. Results and result analysis

The results of regression (6a), (6b) and (9) are displayed in table 6, 7 and 8 respectively, which can be found in the appendix. In addition to this, an overview of all the hypothesis tests performed is listed in table 9 in the appendix. Table 10 displays an overview of the most relevant results.

Table 10, summary of pass-through characteristics

Pre-crisis Post-crisis

Effect Variable Value Complete Value Complete Change Significant

short-run pass-through 𝛽 .690*** No .410*** No -.280 No long-run pass-through − 𝛿 𝛾⁄ .974*** yes .769*** No -.204*** Yes mark-up _{− 𝛼 𝛾}⁄ .969*** 1.824*** .855*** Yes Adjustment speed |𝛾| .688*** .662*** .025 No

Note [a]: *, **, ***, denote significance at the 10, 5, and 1 percent level respectively

5.1 Pre-crisis result analysis

This section will discuss the results of the pre-crisis interest rate pass-through model for the Netherlands and how these compare with previous found results on the Euro area.

5.1.1 Long-run characteristics

In the pre-crisis period the final pass-through is found to be complete with a value of 0.974, meaning that a change in the policy rate is eventually fully passed-through to the retail rate. The market can therefore be considered as highly competitive and without market imperfections in the long-run (Ahmad et al., 2013). The result of a complete final pass-through is consistent with the expectations, as well as with previous literature on the Netherlands (Sorensen & Werner, 2006) and the Euro zone (Sander & Kleimeier, 2004), when compared with similar loan rate types.

Subsequently, the mark-up is concluded to be significant with a value of 0.969, coinciding with the expectations. Together with the final pass-through, the mark-up displays the long-run relationship between the retail and the policy rate. This relationship shows that even though changes in the policy rate are eventually fully passed-through, the retail rate remains significantly higher. This finding is supported by the simultaneous graphing of the policy and retail rate displayed in figure 1. Moreover, it is in line with findings on the Euro area, as the Bank for International Settlements (2014) found an average mark-up of 0.950. It can therefore be concluded that the interest rate

5.1.2 Short-run characteristics

The immediate pass-through in the pre-crisis is found be 0.690, meaning that a one percentage point change in the policy rate results in a 0.690 percentage point change in the retail rate within the same month. This is incomplete and indicates the existence of market imperfection in the short-run. A comparison with the final pass-through confirms the disappearance of short-run market

imperfections in the long-run, as it is found to be complete.

The result of an incomplete short-run pass-through is in line with both Bernhofer’s findings on the Netherlands (2013) and the expectations, as well as the results on the Euro area. However, it is 3.6 times as high when compared with the pass-through of 0.19 found by the Bondt for the Euro area (2004). This confirms the existence of heterogeneity within the Euro area and implies a higher degree of competition, and therefore less asymmetric information, within the Dutch banking system.

Table 10 shows an adjustment speed of 0.668 in the pre-crisis period, meaning that 68.8% of the retail rate’s divergence from its long-run equilibrium is correcting in the following month. Assuming that the rate is equal to its equilibrium at the beginning of the month, a one percentage point increase causes an immediate increase of 0.690 percentage points. As its equilibrium increase by one percentage point due to the complete long-run pass-through, the rate will deviate with 0.310 percentage points from this equilibrium. The next month, in addition to the immediate pass-through of any policy rate change that might take place, the 68.8% adjustment speed causes the retail rate to rise with 0.207 percentage points towards its long-run equilibrium.

The found adjustment speed is again in line with previous studies on the Netherlands as Bernhofer concluded it to be 0.526 (2004). However, results on the adjustment speed in the Euro area range from 0.052 (Aristei & Gallo, 2013) to 0.11 (De Bondt, 2005), being significantly lower than in the Netherlands. Since asymmetric information is one of the main factor influencing the speed of adjustment, it can be concluded that the Dutch banking system has a lower degree of asymmetric information (De Bondt, 2005). This is in line with the results of a higher immediate pass-through.

5.2 Effect of the 2008 financial crisis

The results of hypothesis test (T5) displayed in table 9 shows that, as expected, the 2008 financial crisis significantly affected the interest rate pass-through model. This section will therefore provide a detailed analysis of how the model was affected. In addition to this, the results will be compared with the findings of previous literature on Euro area.

5.2.1 Long-run characteristics

Table 10 shows that both the long-run pass-through and the mark-up have changed significantly with a decrease of 0.204 and an increase of 0.855 percentage points respectively, thereby confirming the expectations. The foremost reason for this change is the upsurge in bad debt and credit risk, which have led to an increase in information asymmetries and consequently moral hazard and adverse selection. This caused banks to conduct stricter lending criteria regardless of the declining policy rate (Ahmad et al., 2013). Moreover, the failure of several banks during the crisis period led to the remaining banks wanting to improve their financial stability (Aristei & Gallo, 2013). Consequently, banks became more reluctant to lower the lending rate and increased their mark-up, as this denotes revenue.

The results of a decrease in the long-run pass-through is consistent with the findings of De Sola Perara (2014) and the Bank for International Settlements (2014), who found an average change of 0.445 and 0.348 percentage points respectively. Moreover, the BIS also included the mark-up in its model and the results show that it has significantly increased for all the countries included in the empirical analysis, with an overall average of 1.174 percentage points. This indicates that the long-run characteristics of the interest rate pass-through model in the Netherlands has been impaired in response to the 2008 financial crisis, resulting in a less efficient monetary policy. However,

comparison of the results with previous literature on the Euro area shows a larger impairment in the Euro area as both the long-run pass-through and mark-up displayed a greater change.

5.2.2 Short-run characteristics

In contrast to the long-run characteristics, the short-run features were not affected by the crisis. Though the immediate pass-through displayed a change of 0.280 percentage points, it was found to be insignificant as it at the same time displayed a relatively high standard error of 0.211. This is in line with neither the expectations nor the findings of Aristei and Gallo (2013) on the Euro area, who concluded the short-run pass-through to have significantly decreased. The immediate pass-through was expected to be negatively affected as the crisis caused a rise in credit risk thereby making banks less willing to lower their lending rate. However, the previously discussed results of the post-crisis period indicated a higher degree of competition within the Netherlands when compared with the Eurozone. This results in a higher pressure on banks to lower their lending rates in order to not lose clients, providing a possible explanation for the insignificant change in the immediate pass-through.

The speed of adjustment decreased with an insignificant amount 0.025 percentage points, indicating that the faster adjustment because of a clearer policy stance was most likely offset by an increase in asymmetric information. The results contradict the expectations and findings of Aristei

and Gallo(2013), who found the adjustment speed to increase from 0.05 to 0.59 in response to the crisis. However, their obtained post-crisis value remains lower than the post-crisis adjustment speed of 0.662 found for the Netherlands. This indicates that the contradicting result was most likely found because the Netherlands already reacted fast to deviation from the long-run equilibrium, leaving less room for improvement.

6. Conclusion

This paper examined the effect of the 2008 financial crisis on the interest rate pass-through in the Netherlands and consequently compared the results with the findings of previous empirical research on the Euro area. To do so, the Euribor and the rate on loans to NFC of at least one million euros were used as a proxy for the policy and retail rate respectively. Consequently, monthly data from January 2003 to December 2014 was used where September 2008 was selected as the beginning of the crisis. Three regressions were performed, one for the pre-crisis period, one for the post-crisis period, and one for the entire period incorporating a dummy variable to account for any change in coefficients caused by the crisis. The results of these regressions led to several findings.

First, the Netherlands exhibits an incomplete immediate pass-through in the pre-crisis period, whilst the final pass-through is complete. This implies an inefficient function of monetary policy that becomes efficient in the run. However, as the speed of adjustment towards its long-run equilibrium is found to be high, the transition from inefficient to efficient monetary policy is quick. It can therefore be concluded that the functioning of the monetary policy transmission mechanism in the Netherlands was effective in the pre-crisis period.

Second, a comparison with previous literature the Euro area shows that, though the long-run characteristics are similar, the Netherlands displays a higher efficiency of monetary policy in the short-run. Not only is the immediate pass-through higher, the speed of adjustment is also shown to be faster, resulting in a quicker convergence of the retail rate to its long-run equilibrium. These findings confirm the existence of heterogeneity in the Euro area.

Third, the results show that the 2008 financial crisis significantly impaired the functioning of the monetary policy transmission mechanism in the Netherlands. Whist the effect on the short-run characteristics is insignificant, the long-run efficiency significantly decreased due to a diminished final pass-through and an increase in the mark-up.

Lastly, a comparison of the effect on the different aspects of the interest rate pass-through model with previous findings on the Eurozone indicate dissimilarities in the extend and way of impairment. While the long-run efficiency of monetary policy in both the Netherlands and the Eurozone decreased in response to the crisis, this was to a lesser extent in the Netherlands as both the final pass-through and the mark-up changed less than previous papers on the Eurozone

concluded. Moreover, previous empirical research on the Euro area indicated a significant decrease in the immediate pass-through whereas this paper finds no significant change for the Netherlands.

These findings lead to the overall conclusion that the 2008 financial crisis significantly impaired the monetary policy transmission mechanism in the Netherlands. This indicates a less efficient monetary policy of lowering interest rates during a time at which the ECB relies heavily on precisely this policy, and the use of additional measure to stimulate the economy is therefore warranted. However, as the efficiency of conventional monetary policy remain high in the

Netherlands when compared with the Euro area, these finding do not indicate the need for the ECB to increase its focus on the Netherlands when implementing unconventional tools of monetary policy.

Nevertheless, this paper only examined the pass-through to the rate on loans to NFC whilst it has been shown that the pass-through is, amongst others, dependent on the type of rate. A

suggestion for further research is therefore an analysis of the effect the 2008 financial crisis had on a wider range of retail rates. Moreover, the existence of heterogeneity in the interest pass-through model across countries is confirmed by this paper. This emphasises the importance of continuing with researching the interest rate pass-through for individual countries, in order to assist the monetary authorities in deciding on their monetary policy.

Bibliography

Ahmad, A. H., Aziz, N., & Rummun, S. (2013). Interest rate pass-through in the UK: has the

transmission mechanism changed during the financial crisis?. Economics Issues, 18(1), 17-38 Andrews, D. W., & Fair, R. C. (1988). Inference in nonlinear econometric models with structural

change. The Review of Economic Studies, 55(4), 615-640.

Andries, N., & Billon, S. (2016). Retail bank interest rate pass-through in the euro area: An empirical survey. Economic Systems, 40(1), 170-194.

Aristei, D., & Gallo, M. (2014). Interest rate pass-through in the Euro area during the financial crisis: A multivariate regime-switching approach. Journal of Policy Modeling, 36(2), 273-295.

Belke, A., Beckmann, J., & Verheyen, F. (2013). Interest rate pass-through in the EMU–New evidence from nonlinear cointegration techniques for fully harmonized data. Journal of International money and finance, 37, 1-24.

Bernhofer, D., & van Treeck, T. (2013). New evidence of heterogeneous bank interest rate pass-through in the euro area. Economic Modelling, 35, 418-429.

Bredin, D., Fitzpatrick, T., & Reilly, G. O. (2002). Retail interest rate pass-through: the Irish experience. Economic and Social Review, 33(2), 223-246.

Clogg, C. C., Petkova, E., & Haritou, A. (1995). Statistical methods for comparing regression coefficients between models. American Journal of Sociology, 1261-1293.

Cordemans, N., & de Sola Perea, M. (2011). Central bank rates, market rates and retail bank rates in the euro area in the context of the recent crisis.National Bank of Belgium Economic Review, 1, 27-52.

Cottarelli, C., & Kourelis, A. (1994). Financial structure, bank lending rates, and the transmission mechanism of monetary policy. Staff Papers, 41(4), 587-623.

Darracq Paries, M., Moccero, D., Krylova, E., & Marchini, C. (2014). The retail bank interest rate pass-through: The case of the euro area during the financial and sovereign debt crisis. ECB Occasional Paper, (155).

De Bondt, G. J. (2005). Interest Rate Pass‐Through: Empirical Results for the Euro Area. German Economic Review, 6(1), 37-78.

de Sola Perea, M., & Van Nieuwenhuyze, C. (2014). Financial integration and fragmentation in the euro area. Economic Review, (i), 99-125.

Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of the American statistical association, 74(366a), 427-431.

Engle, R. F., & Granger, C. W. (1987). Co-integration and error correction: representation, estimation, and testing. Econometrica: journal of the Econometric Society, 251-276.

European Central Bank. (2009). Recent developments in the retail bank interest rate pass-through in the Euro area. ECB Monthly Bullitin

Gambacorta, L., Illes, A., & Lombardi, M. J. (2014). Has the transmission of policy rates to lending rates been impaired by the Global Financial Crisis?.

Hristov, N., Hülsewig, O., & Wollmershäuser, T. (2014). The interest rate pass-through in the Euro area during the global financial crisis. Journal of Banking & Finance, 48, 104-119.

Illes, A., & Lombardi, M. J. (2013). Interest rate pass-through since the financial crisis. BIS Quarterly Review, September.

Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica: Journal of the Econometric Society, 1551-1580. Monetary Policy Committee. (1999). The transmission mechanism of monetary policy. Bank of

England.

Sander, H., & Kleimeier, S. (2004). Convergence in euro-zone retail banking? What interest rate pass-through tells us about monetary policy transmission, competition and integration. Journal of International Money and Finance,23(3), 461-492.

Sander, H., & Kleimeier, S. (2006). Convergence of interest rate pass-through in a wider Euro zone?. Economic Systems, 30(4), 405-423.

Sathye, M. (2013). Financial Crisis and Interest Rate Pass-Through in Australia. Review of Pacific Basin Financial Markets and Policies, 16(04), 1350023.

Sørensen, C. K., & Werner, T. (2006). Bank interest rate pass-through in the euro area: a cross country comparison. Frankfurt am Main: European Central Bank.

Stock, J. H., & Watson, M. W. (2015). Introduction to econometrics. Boston: Pearson/Addison Wesley.

APPENDIX

Table 1, Summary of statistics

Entire period Pre-crisis Post-crisis

PR RR PR RR PR RR Observations 144 144 68 68 76 76 Mean 1.669 2.995 2.867 3.754 .596 2.315 Std. Dev. 1.416 1.057 .867 .849 .827 .704 Minimum -.013 1.419 2.049 2.829 -.013 1.419 Maximum 4.839 5.438 4.839 5.438 3.857 5.276

Note [a]: PR and RR are used to denote the policy rate and retail rate respectively

Table 6, Regression pre-crisis period (6a)

Coefficient Value t statistic P > |𝒕| Period Observations R2

𝛼 .666 (.138) 4.82 0.000*** 2003M1 – 2008M8 67 0.3839 𝛽 .690 (.142) 4.86 0.000*** 𝛾 -.687 (.128) -5.36 0.000*** 𝜹 .669 (.125) 5.35 0.000*** − 𝛿 𝛾⁄ .974 (.028) 34.47 0.000*** − 𝜶 𝜸⁄ .969 (.083) 11.62 0.000***

Note [a]: *, **, ***, denote significance at the 10, 5, and 1 percent respectively Note [b]: The standard errors are displayed between brackets

Table 7, Regression post-crisis period (6b)

Coefficient Value t statistic P > |𝒕| Period Observations R2

𝛼 1.208 (.195) 6.20 0.000*** 2008M9 – 2014M12 76 0.4853 𝛽 .410 (.121) 3.38 0.001*** 𝛾 -.662 (.104) -6.38 0.000*** 𝜹 .509 (.084) 6.06 0.000*** − 𝛿 𝛾⁄ .769 (.045) 17.12 0.000*** − 𝜶 𝜸⁄ 1.824 (.043) 42.29 0.000***

Note [a]: *, **, ***, denote significance at the 10, 5, and 1 percent respectively Note [b]: The standard errors are displayed between brackets

Note [c]: The standard error of − 𝛿 𝛾⁄ and − 𝛼 𝛾⁄ were calculated based on the delta method

Table 8, Regression with Dummy variable (9)

Coefficient Value t statistic P > |𝒕| Period Observations R2

𝛼2 .666 (.178) 3.74 0.000*** 2003M1 – 2014M12 143 0.4717 𝛽2 .690 (.183) 3.77 0.001*** 𝛾2 -.687 (.165) -4.16 0.000*** 𝛿2 .669 (.161) 4.15 0.000*** 𝜃 .542 (.245) 2.22 0.028** 𝜌 .025 (.188) 0.13 0.895 𝜔 -.1600 (.1771) -0.90 0.368 𝝋 -.280 (.210) -1.33 0.187

Note [a]: *, **, ***, denote significance at the 10, 5, and 1 percent respectively Note [b]: The standard errors are displayed between brackets

Table 9, Results hypothesis tests

Test H0 H1 t statistic F statistic p-value

T1 𝐶 = 1 𝐶 ≠ 1 -.94 - 0.351 T2 𝐴 = 0 𝐴 > 0 11.62 - 0.000*** T3 𝛽 = 1 𝛽 ≠ 1 -2.18 - 0.033** T4 𝛾 = 0 𝛾 < 0 -5.36 - 0.000*** T5 𝜃 = 𝜌 = 𝜔 = 𝜑 = 0 𝜃 𝑜𝑟 𝜌 𝑜𝑟 𝜔 𝑜𝑟 𝜑 ≠ 0 - 8.25 0.000*** T6 𝜔 = 𝜌 = 0 𝜔 𝑜𝑟 𝜌 ≠ 0 - 6.07 0.003*** T7 𝜃 = 𝜌 = 0 𝜃 𝑜𝑟 𝜌 ≠ 0 - 15.14 0.000*** T8 𝜑 = 0 𝜑 < 0 -1.33 - 0.094* T9 𝜌 = 0 𝜌 < 0 0.13 - 0.448