An empirical study on the relation between interest
rate pass-through and banking competitiveness in
the Netherlands and Luxembourg
By: Eric Elderenbosch Faculty: Economics and Business Specilization: Economics and Finance
Thesis Supervisor: Ioana Neamtu Word count: 8155
January 2018
Abstract: One of the most important tools for the ECB to influence economic activity is
through interest rate-pass through. Using policy rates to achieve their goal of price and financial stability. Previous research has shown that there is heterogeneity in completeness and speed of adjustment within the eurozone. A proposed reason for this is banking
competitiveness. Therefore this paper will compare two countries with very different banking competitiveness. These countries are the Netherlands and Luxembourg. Results show that a more competitive banking scene results in quicker speed of adjustment to market interest rates and also a higher immediate pass-through rate. Results were obtained using a single equation correction model. Future research could look into the effect of banking competitiveness on final pass-through as my results are indecisive
Statement of originality
This document is written by Eric Elderenbosch, 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.
Table of contents
1. Introduction
1
2. Literature review
2
2.1 Interest rate pass-through
2
2.2 Previous Studies
3
2.3 Research Method
4
3. Research method
4
3.1 Research question
4
3.2 Herfindahl-Hirschman Index
4
3.3 Single equation error correction model (ECM)
5
3.4 Hypothesis
6
3.5 Augmented Dickey Fuller (ADF)
7
3.6 Engle-Granger Augmented Dickey-Fuller (EG-ADF)
8
4. Data and results
9
4.1 Herfindahl-Hirschman Index
9
4.2 Information on data used for the tests
10
4.3 Augmented Dickey Fuller (ADF)
14
4.4 Engle-Granger Augmented Dickey-Fuller (EG-ADF)
16
4.5 Single Equation Correction model
19
5. Conclusion
24
6. References
25
1
1. Introduction
The US banking crisis of 2007-2008 had huge economic repercussions for the whole world. As US banks collapsed, the government and central bank tried to save whatever financial institution could still be saved. The US banking crisis spread as the shock travelled all across the world and would later be described as the global financial crisis. Not only affecting financial institutions in the euro area through exchange rate and oversea savings, but also affecting the consumer side. As consumer confidence in the financial market was pushed to an all-time low.
The ECB had to intervene to soften the blow of decreased investing and consumption using one of their monetary tools to try and stop the economic down spiral. One of the tools they used was market interest rates. The ECB lowered its policy rates heavily in the hope to promote consumer spending and investing by reducing the revenue on savings and lowering the rate on loans. The way this worked would mean people could get cheaper loans to invest in projects that would yield more revenue than savings. However, the change in policy rates did not affect banks in the euro area as they might have predicted. Bank rates were only partly-passed through in the euro area (Hristov, et al, 2014).
After the crises Hristov et al (2014) find that this heterogeneity in the euro sector persisted. The heterogeneity meaning that the changes in the set market interest rate have varying speed and pass-through completeness rates depending on the European country. The speed in this case indicates how quick banking rates converge to the ECB’s rates both initially and in the long-term. Completeness then shows if the banking rates converge completely or if there is still a difference between ECB and banking rates after both rates have set.
The results from this thesis could give more insights on the effect of rates for certain countries depending on their banking competitiveness. Interest rates are an important instrumental tool for the ECB to try to achieve their goals (De Bondt, 2005). Therefore it is important to research the effects that influence the effectiveness of their rates, banking competitiveness possibly being one of them.
The reason banking competitiveness is looked at, is because much like other industries prices are closest to marginal costs the more competitors there are. Likewise in the banking industry if a bank would have a monopoly position there would be no reason to borrow from the central banks at the cheap set rate and to lend out at a similar low rate. They would instead borrow low and lend high. Therefor competitiveness could mean that retail rates would
2 In this thesis Luxembourg and the Netherlands will be looked at. As according to the ECB latest report on financial structures (2017) the respective market shares of the five largest credit institutions in total assets are for Luxembourg 28-34% and for the Netherlands 82-87%. Luxembourg is below the average of the euro area and the Netherlands is above it. Therefor it should be a good indication of the effect of banking competitiveness. These
countries also share similarities in culture both being Benelux countries, so therefor it is closer to a cetirus paribus. A big difference between the two countries however is that Luxembourg although very competitive; competition comes from foreign banks that have branches in Luxembourg, whereas in the Netherlands the market is controlled by Dutch Banks.
2. Literature review
To get a better understanding of interest rate pass-through, first the concept itself will be explained. After that previous studies will be discussed to shine more light on the subject and how it has changed over time.
2.1 Interest rate-pass through
The ECB has set a primary goal of price stability and financial stability. Firstly the price-stability, or monetary price-stability, which as stated by the ECB refers to a year-on-year increase in the Harmonised Index of Consumer Prices for the euro area of below 2 percent, which is to be maintained over the medium term. The ECB is aiming for an inflation rate of two percent. Secondly there is Financial-stability, which does not have a clear definition but a broad definition is given by Mishkin to what a financial crisis is (1991): “A financial crisis is a disruption to financial markets in which adverse selection and moral hazard problems become much worse, so that financial markets are unable to efficiently channel funds to those who have the most productive investment opportunities”. So financial stability can be defined as a state of financial markets that ensures an efficient allocation of financial resources (Issing, 2003)
To ensure that these goals are met it is important that the instruments the ECB can use are effective. The instruments include open market operations, standing facilities and
minimum reserve requirements for credit institutions. The effect of these instruments is described as the monetary transmission mechanism (MTM).
3 In this thesis the standing facilities are looked at. Standing facilities can be split into two facilities. A Marginal lending facility, for financial institutions to obtain overnight liquidity. A Deposit Facility where financial institutions can deposit money against an
overnight rate. The goal is that by setting these rates, retail banks will converge to the set rates with their own retail rates. This is the interest rate pass-through or as it is defined by Mishkin et al (2013): “the adjustment of retail bank interest rates in response to changes in the official policy rates set by the central bank”. Through these adjusted bank retail rates the ECB can influence economic activity. Because as banks are affected by the ECB rates they make choices to other rates that affect non-financial corporations and households. Now the issue starts when the interest rate pass-through is incomplete as it dictates whether there is an effect on the general public.
One of the reasons the pass-through could be incomplete is due to a lack of
competition among banks (Mishkin et al, 2013). High competition would encourage banks to adjust their rates quickly, as not doing so could result in market share loss. A paper by Mojon (2000) also comes to the same conclusions that competition among banks reduces the
“interest rate cycle asymmetry”, cycle asymmetry referring to the lag of banking rate adjustment.
2.2 Previous studies
Interest rate pass-through has been heavily researched both in the eurozone, country specifically or in a different part of the world. As I’ll be focusing on two countries from the eurozone I will focus this on previous studies on the eurozone. Before there was a European monetary union (EMU), countries that would join this union were looked at by Sander and Kleimeier (2004) and they concluded that country specific difference would still persists through the unions especially in countries where bank rates are a key determinant of the cost of capital and yield on savings, but over the years would become more homogenous over time.
This was then also found by Hristov et al. (2014) in that the pass-through was seen as complete but heterogeneous in speed.
This is partially contradicted by Sorensen and Werner (2006), who fined that before crisis interest rate pass-through is heterogeneous in both speed and completeness and hint in that especially the speed of adjustment could be explained by banking competitiveness. As
competitiveness could tend to lead to more risky borrowers to boost market shares and therefor retail rates would converge quicker.
4 Research by Aristei and Gallo (2014) indicate that during periods of financial distress bank lending rates to both households and non-financial corporation show a reduction of their degree of pass-through from the money market rate also indicating incompleteness, confirming that during the crisis interest pass-through is less complete.
In a similar kind of research, compared to this thesis, Leroy and Lucotte (2015) wrote about the effects of banking competitiveness in the period 2003-2010 and found increased competition would improve the effectiveness of monetary policy transmission through the interest rate channel. They did however use different tests for both as they used a Lerner index, which would normally be used to test a single corporations market share and not market concentration like a Hefindahl-Hirschman index. It will be interesting to see if I get similar results for the period 2008-2016.
2.3 Research method
The most important model used in this paper is the Single Equation correction model developed by Granger and Engle (1987). A model that is used to estimate the speed at which explained variables return to equilibrium after a change in their explanatory variables. This model is widely used for any research on interest rate pass-through as it can calculate speed and immediate and final pass-through. More on the methods used in the next chapter.
3. Research method
In this chapter used models and tests will be explained generally. This includes how the models are used and why they are chosen for this specific paper to answer the following research question.
3.1 Research question
What is the effect caused by banking competitiveness in relation to the speed and completeness of the interest rate pass-through effect from the ECB using Single Equation ECM with data ranging from 2008-2016 on Luxembourg and the Netherlands?
3.2 Herfindahl-Hirschman Index
To measure the competitiveness of the banking industry in a certain country the best metric to use is the Herfindahl-Hirschman Index. The Herfindahl-Hirschman index is an
5 indicator for market concentration, which could be compared to market competitiveness. This index squares all market shares of the respective firms and will be anywhere between zero and ten thousand. A high index number indicates that fewer firms own the bigger portions of the market and therefore the markets as a whole might be less competitive than those with a lower index. The formula looks as follows:
+ … + (1)
3.3 Single equation error correction model (ECM)
The ECM model is a time-series model used to show the interest rate pass-through effect and is the main econometric model used for research on the interest passed-through effect. With the use of the single equation error correction model I can find the speed and completeness on the effect of the changed market interest rates set by the ECB on the retail interest rates (Sørensen & Werner, 2006). The ECM models formula is as follows.
( ) (2)
First the variables BR and MR stand for bank retail rates and market rates, t stands for current period and t-1 previous period. The i stands for the specific observation. These rates will be matched, more on this later. The important coefficients for this model are as follows. In this equation shows the slope, and thereby the speed of adjustment or lag of the bank rates adjusting to the set market rate to the long-run equilibrium. As shown in the formula, it is the effect of the difference between retail rates and market rates in the previous period on the current difference on retail rates. A negative coefficient here means that the banking retail rate is moving towards the market equilibrium where a positive means it moves away. The size is then a measure of its speed. In short fulfills two rolls in both direction and speed. Indicates the immediate pass-through. It is the effect the immediate difference of the market rate has on the difference of the banking rate. Therefor this is the effect on the short run. If equals one it signals perfect Pass-through and at lower than one it might be an indication that the bank rates are “sticky” Above one would indicate that the immediate effect has an overshooting effect on retail rates. Overshooting might occur due to asymmetry
between banks and their borrowers reflected by credit risk factors.
For the long run reflects the final pass-through. As seen in the formula it’s the effect from last period’s difference in market rate on the currents period difference retail rate.
6 Which, similarly as the immediate Pass-through, if equals one it indicates perfect pass-through, whereas lower than one indicates incomplete pass-through and higher than one indicates overshooting.
The error term u is included to measure any omitted factors for a more complete statistical representation. These omitted factors could include for example risk aversion, difference in risk structures or term structures. Other omitted factors could have to do with the risk structure of banks and the countries issuing of rules on banking.
To use the model for Statistical research the following change has to be made:
Using the new variable to enable for linear regressions
(4)
Only one lag has to be tested as Sorensen and Werner (2006) concluded that adding parameters for a higher lag length had insignificant results and therefor need not be tested.
Single Equation EMC requires two assumptions to be fulfilled. The first one has to do with the interest rates being non-stationary. This means they will not return to their “rooted” value. The second assumption has to do with cointegration. This is a statistical property, which proves if time series variables move together or not.
3.4 Hypothesis
Now that the main model of this thesis is explained the hypothesis can be formulated in a way that makes sense. The hypothesis is as follows
The country with a more competitive banking scene will have interest rates converge quicker and more complete to the long-run equilibrium than those of less competitiveness. In this case the general Bank rate of Luxembourg would converge quicker and more complete than that of the Netherlands’ banks.
H0:
H1:
Speed of adjustment between absolute brackets, as more negative means faster speed of adjustment.
7
3.5 Augmented Dickey Fuller (ADF): test for unit root
The augmented Dickey Fuller test (Said & Dickey, 1984) will be used to determine whether our chosen variables have unit root or in different words are stationary. There are many different tests for unit roots on time series date, but the general consensus is that the tests get similar results. The augmented dickey fuller test is an alteration on the basic DF-test. The difference of these two tests resides in the amount of lags tested. The ADF-test can account for multiple lags and therefor removes the problem of possible serial correlation (Stock & Watson 2015). The standard Dickey-Fuller test (DF) can only test for one lag. The ADF formula is as follows:
∑ (5)
The addition of the lagged terms is implied in ∑ , is the error term. is a constant and is just like set to zero at the start as they correspond to modeling of a random walk. The null-hypothesis is of = 0 and the alternative hypothesis of . The test statistic is then . This is similar to a Student t-statistic. The ADF’s test null hypothesis is that the interest rate has a unit root or in other words is non-stationary and therefor the
alternative hypothesis is that there is no unit root and the variable is stationary. The preferred outcome is that this null hypothesis is not rejected as the ECM model requires the interest rates to have unit root. The reason for this is that if the null hypothesis is rejected it would mean the interest rates are stationary and therefor the researched interest rates would not move with their market proxy in a stochastic trend, but instead return to their own rooted value. Andries and Billon (2016) argue that interest rates are always non-stationary and can therefore be assumed while other research shows evidence that it cannot be assumed and thereby needs to be tested.
The amount of lags needed to be tested is researched by Schwert (1989). He concluded in his research that the estimated amount of lags ( ) needed can be calculated with the following formula.
( ( ) ) (6)
As I have 108 total observations per interest rate (T) the amount of lags used is twelve. This also means that the critical values of our ADF-statistic are -3.517, -2.894, and -2.582 for 1%, 5% and 10% significance level respectively
8
3.6 Engle-Granger Augmented Dickey-Fuller (EG-ADF): Test for cointegration
Besides the assumption of unit root there is one more assumption that needs to be tested. This is the assumption of cointegration. Cointegration shows that there is a stochastic trend in the long-run equilibrium between the tested bank retail interest rate and the chosen market proxy rate. Since we already used the Augmented Dickey-Fuller test for unit root, we can use a similar test in the Engle-Granger Augmented Dickey-Fuller test to check for cointegration. This is also the proposed test by Stock and Watson (2015).
The EG-ADF is a two-step test where the first step the cointegrated coefficient (µ) is estimated using the following OLS linear regression:
(7)
α is the constant, and when cointegration is present is the estimated constant difference between the interest rates for example, the overnight household interest rate could move the same direction as EONIA but will always be around one percent higher. Then if the two interest rates are cointegrated: - , will always result in a stationary variable as it
implies that both the error term is stationary and therefor constant on the long run and so is α.
The second step then is to test for unit root, using the Augmented Dickey-Fuller test. The null hypothesis is that the error term is non-stationary/unit root and therefor the alternative hypothesis indicated stationary/no unit root. Critical value for this test is found by Hamilton (1994, p.766 table B.9). At a significance level of 1%, 5% and 10%, the critical values are -3,39, -2.76. -2.45 respectively.
9
4. Data and results
In this chapter data and results will be displayed and discussed. First of all found results for the Herfindahl- Hirschman index will be shown. After that, information is given about the datasets used for the models and tests. Also motivations for using this data are explained. Thereafter results for the tests and models will be shown and discussed.
4.1 Herfindahl- Hirschman Index
As mentioned earlier the Hefindahl-Hirschman index is used to measure market concentration. The Hefindahl-Hirschman index for periods 2008-2016 Published by the ECB (2017,
October),
Table 1, HHI of countries and eurozone
2008 2009 2010 2011 2012 2013 2014 2015 2016 Mean Luxembourg 309 310 343 346 346 357 330 321 260 325 The Netherlands 2167 2034 2049 2067 2026 2105 2131 2104 2097 2087 Euro area 676 649 688 710 677 689 730 720 697 693
As seen here over time Luxembourg market concentration has gone up a little over time peaking in 2013. After 2013 it seems to get more concentrated again even going below the initial lower value from 2008 in 2016. It is safe to say that over this time period the HHI values for Luxembourg remained low. This indicates that Luxembourg indeed has a high banking market concentration or in other words a highly competitive banking sector. The biggest reason for this dense market concentration is that Luxembourg counts over a 100 different banks wherein the biggest five credit institutions own only between 28 and 31 percent of the total shares of assets for the time period 2008-2016. For instance, this is low compared to the biggest five credit institutions of the euro area, which own 44 to 48 percent of the total assets in the same time period.
The Herfindahl-Hirschman index for the Netherlands has ups and downs over time but seems to be fairly stable around the mean. The Netherlands is also in the opposite situation of Luxembourg in the fact that their market concentration is less concentrated than the euro area and therefor even less concentrated than Luxembourg’s. In contrast to Luxembourg’s biggest
10 five credit institutions the total shares in assets of the biggest five institutions from the
Netherlands own between 82 and 85 percent of total market assets.
In short Luxembourg’s market competitiveness is consistently above the Euro area and the Netherlands below both. There is also a significant difference between Luxembourg and the Netherlands. This significant difference confirms a useable choice in Luxembourg and the Netherlands to compare the interest pass-through rate based on bank market competitiveness.
4.2 Information on data used for the tests
The data used in this paper are harmonized monthly MFI interest rates from the period 2008-2016. This time period was chose in particular to see the interest pass-through rate after the financial crisis. The interest rates used are all collected from statistics from the Eurosystem of Central Banks (ESCB). Harmonized rates ensure that there are no matching issues that could skew the data. The models used in this paper work with time series data therefor
un-harmonized rates would mean the error terms could include time shifted data and therefor cause heterogeneity issues when compared to their market proxies and different sectors. To completely circumvent this problem no national retail interest rates are used but only MFI interest rates. The only exception to this is for the swap rates (both for 3 years maturity and 7 years), these rates are obtained daily but will only be looked at from a monthly perspective matching the other rates in date, using only the 15th of every month since that’s when the others are recorded. This gives us a total of 108 observations for each interest rate.
Several interest rates are used to check multiple sectors to see any inconsistencies between them. Previous research has indicated there are differences in convergence rates between different segments of the financial market (Sander & Kleimeier, 2004). To account for these different segments the interest rates used in this paper are chosen based on multiple differentiated factors for a more in depth picture on the pass-through rates. The differences are:
1. Differences between Short-term (maturities with less than one year) and Long- term (maturities one year and above) interest rates, to see how different maturities react to market rates matched with respective market rates of short and long maturities. More on the specific matching between retail and market rates later.
2. Differences between deposits and loans, to showcase a possible difference between interest rate Pass-through of borrowing- and savings- interest rates
11 3. Differences between rates given to households and non-financial corporations. This
difference gives more information about interest rate pass-through comparing different risk categories. An extra differentiation is made on the short-term non-financial
corporation loans of over and fewer than one million euros again to account for different risk structures.
There has been a lot of research of what bank retail rates are best compared to which interest rates. Belke, Beckmann and Verheyen (2012) conclude that short-term rates show higher cointegration with EONIA, but long-term rates less so. Therefor in this paper we’ll match rates based on ECB’s monthly bulletin (August, 2009, p.99) displayed in table 1 on the next page and graphically in graphs 1-5 on the page thereafter.
At first glance deposit rates seem to follow the same pattern of their corresponding market rate. A few exceptions already seem clear. First of all the NL – DH1 in respect with EONIA, the pattern is similar but NL- DH1 is about two percent above the EONIA line. Secondly NL – DH2 in respect with SWAP3 seems to not follow the pattern and is higher in general as well. Thirdly For Loans it seems EUR3 and SWAP7 are lower than the interest rate they are compared. The Engle-Granger Augmented Dickey-Fuller test especially should give a more descriptive description on these differences.
12
Table 2, Description of interest rates and their matched rates
Description interest rates Abbreviation, (NL - or LU - prefix depending on the country)
Matched market rate
Overnight Deposits households
DHO EONIA
Overnight Deposits Non-Financial Corporations
DNFCO EONIA
Short-term time Deposits households
DH1 Three-month
EURIBOR Short-term time Deposits
Non-Financial Corporations
DNFC1 Three-month
EURIBOR Short-term Loans for
household house purchases
LHHP1 Three-month
EURIBOR Short-term Loans to
non-financial corporations under one million euro
LNFCU1 Three-month
EURIBOR Short-term Loans to
non-financial corporations over one million euro
LNFCO1 Three-month
EURIBOR Long-term time Deposits
Households
DH2 Three-year swap rate
Long-term time Deposits Non-Financial Corporations
DNFC2 Three-year swap rate
Long-term Loans for household house purchase
LHHP2 Seven-year swap rate
Long-term Loans for household consumer credit
LHCC2 Seven-year swap rate
European overnight index average
EONIA Three-month Euro
Interbank offered rate
EURIBOR Three-year swap rate SWAP3 Seven-year swap rate SWAP7
13 -1,000 0,000 1,000 2,000 3,000 4,000 5,000 Jan -08 Au g-08 Ma r-09 Oct-09 Ma y-1 0 De c-1 0 Ju l-11 Fe b -12 Se p -12 Ap r-13 N o v-13 Ju n -14 Jan -15 Au g-15 Ma r-16 Oct-16 NL - DNFCO LU - DNFCO LU - DHO NL - DHO EONIA -1,000 0,000 1,000 2,000 3,000 4,000 5,000 6,000 7,000 Jan -08 Se p -08 Ma y-0 9 Jan -10 Se p -10 Ma y-1 1 Jan -12 Se p -12 Ma y-1 3 Jan -14 Se p -14 Ma y-1 5 Jan -16 Se p -16 NL - LHHP1 LU - LHHP1 NL - LNFCU1M1 LU - LNFCU1M1 LU - LNFCO1M1 NL - LNFCO1M1 EUR3 -1,000 0,000 1,000 2,000 3,000 4,000 5,000 6,000 Jan -08 Au g-08 Ma r-09 Oct-09 Ma y-1 0 De c-1 0 Ju l-11 Fe b -12 Se p -12 Ap r-13 N o v-13 Ju n -14 Jan -15 Au g-15 Ma r-16 Oct-16 LU - DH2 NL - DH2 LU - DNFC2 NL - DNFC2 SWAP3 -1,000 0,000 1,000 2,000 3,000 4,000 5,000 6,000 7,000 Jan -08 Au g-08 Ma r-09 Oct-09 Ma y-1 0 De c-1 0 Ju l-11 Fe b -12 Se p -12 Ap r-13 N o v-13 Ju n -14 Jan -15 Au g-15 Ma r-16 Oct-16 NL - LHHP2 LU - LHHP2 NL - LHCC2 LU - LHCC2 SWAP7 -1,000 0,000 1,000 2,000 3,000 4,000 5,000 6,000 Jan -08 Au g-08 Ma r-09 Oct-09 Ma y-1 0 De c-1 0 Ju l-11 Fe b -12 Se p -12 Ap r-13 N o v-13 Ju n -14 Jan -15 Au g-15 Ma r-16 Oct-16 NL - DH1 LU - DH1 LU - DNFC1 NL - DNFC1 EUR3
Graphs 1-5 of retail interest rates compared to their market proxy, separated by maturity and sort of rate
14
4.3 Augemented Dickey-Fuller (ADF): test for unit root
The augmented Dickey-Fuller test is the required test for the first assumption of the ECM to test for unit root. The null hypothesis states that there is unit root and the variable is non-stationary. The alternate hypothesis then means that there is no evidence of unit root and therefor the variable is stationary. The results are as followed:
Table 3, ADF test results
Variable ADF-statistic P- Value
NL - DHO -0.796 0.828 NL - DNFCO -0.523 0.888 NL - DH1 -3.139 0.024** NL - DNFC1 -2.913 0.044** NL - LNFCU1 -3.662 0.005* NL - LNFCO1 -2.566 0.101 NL - LHHP1 -0.543 0.883 NL - DH2 -0.133 0.410 NL – DNFC2 -2.496 0.116 NL – LHHP2 2.016 0.999 NL – LHCC2 -1.412 0.577 LU - DHO -0.959 0.768 LU - DNFCO -1.080 0.723 LU - DH1 -3.139 0.024** LU - DNFC1 -2.447 0.129 LU - LNFCU1 -1.775 0.393 LU - LNFCO1 -2.562 0.101 LU - LHHP1 -2.734 0.068*** LU - DH2 -2.626 0.088*** LU – DNFC2 -1.852 0.3351 LU - LHHP1 -2.734 0.068*** LU – LHHP2 -3.753 0.003* LU – LHCC2 -1.699 0.432
15
EONIA -1.741 0.410
EURIBOR3 -0.595 0.872
SWAP3 -1.313 0.623
SWAP7 -1.047 0.736
***significant at 10% level, **significant at 5% level, *significant at 1% level
Most variables indicate to have unit root and therefor are non-stationary. The exceptions for the Netherlands are both short-term time deposit rate and also the short-term loans for non-financial corporations fewer than one million euros. The short-term loans for non-non-financial corporations are just about not rejected at 10% significance level. In short all short-term rates that are not overnight for the Netherlands are close to stationary or are stationary depending on the significance level. If we use 1% significance level all, except short-term loans for non-financial corporations, are not rejected to have no unit root. On the long-term all variables are proven non-stationary for the Netherlands.
Luxembourg shares some similarities. The first similarity is that the short-term time deposit rate for households also is only stationary at a significance level of 1%. Secondly short-term deposits for non-financial corporations and loans for non-financial corporations over one million euros rates are close to being rejected at 10% significance level. The difference here is that for the Netherlands under one million euros had more significance to being stationary, whereas with Luxembourg over one million euros does. Short-term loans for households house purchase are non-stationary up till 5% significance level.
Luxembourg also has no unit root on the long-term rates. Between 5% and 10%
significance level is a rejection breakpoint for both long-term time deposits for households and loans for house purchases.
In respect to banking competitiveness it could mean that stationary rates are stationary because they are being kept so by the banking scene. The ADF test gives conflicting results on this. As long-term rates for the Netherlands are indicated to be non-stationary versus the stationary Luxembourg rates which do show statistic evidence for lack of unit root. On short term however the Netherlands has more statistically stationary rates than Luxembourg but only by one.
It is important to note, that as stated earlier some researchers like Andries and Billon (2016) and Mackinnon (1996) argue, that interest rate can never be stationary. This is because tests for unit-root can give a false indication of actual unit-root due to economic situations
16 where the interest rates are stable over long amounts of time. This is then not due to unit root, but due to the timing. While one of the assumptions for the single equation model is indeed that the interest rates have unit-root, it is important to take this criticism into consideration as it might mean the test results are in fact not a measurement of unit root. Therefore the interest rates could still be used for the ECM model as long as it is stated and discussed that certain rates seemingly lack unit root and the possible consequences of this.
4.4 Engle-Granger Augmented Dickey-Fuller (EG-ADF): test for cointegration
Now these are the results of the test for cointegration using EG-ADF. Both steps are included and combined in the results. First step α, µ cointegration coefficient, R2 and Adj. R2
from a linear regression and the t-statistic from the second step using the Augmented Dickey-Fuller test. All standard errors are indicated between round brackets, these are robust standard errors as we have no prove of homoscedasticity of the error term.
Table 4,EG- ADF test results
Bank retail rate Market rate α µ cointegration coefficient R2 Adj. R2 t-(EG-ADF) Statistic NL - DHO EONIA 0.387 (0.004) 0.0109 (0.006) 0.845 0.844 -1.405 NL – DNFCO EONIA 0.490 (0.021) 0.609 (0.016) 0.933 0.933 -1.680 NL – DH1 EURIB OR 1.894 (0.018) 0.551 (0.021) 0.864 0.863 -3.524* NL – DNFC1 EURIB OR -0.032 (0.018) 0.89 (0.010) 0.985 0.984 -4.197* NL – LNFCU1 EURIB OR 3.024 (0.019) 0.526 (0.012) 0.953 0.953 -4.035* NL - LNFCO1 EURIB OR 1.380 (0.023) 0.762 (0.010) 0.965 0.965 -5.045* NL – LHHP1 EURIB OR 2.859 (0.046) 0.632 (0.025) 0.861 0.860 -0.814 NL – DH2 SWAP3 3.55 (0.037) 0.061* (0.018) 0.113 0.105 -0.345
17
Bank retail rate Market rate α µ cointegration coefficient R2 Adj. R2 t-(EG-ADF) Statistic NL – DNFC2 SWAP3 0.093** (0.074) 0.930 (0.039) 0.845 0.843 -1.898 NL – LHHP2 SWAP7 4.069 (0.044) 0.232 (0.017) 0.714 0.712 0.320 NL – LHCC2 SWAP7 3.359 (0.025) 0.454 (0.015) 0.916 0.915 -2.516*** LU – DHO EONIA 0.388 (0.013) 0.583 (0.010) 0.968 0.968 -2.575*** LU – DNFCO EONIA 0.106 (0.011) 0.620 (0.015) 0.971 0.971 -4.682* LU - DH1 EURIB OR 0.204 (0.032) 0.754 (0.018) 0.941 0.940 -3.086** LU - DNFC1 EURIB OR -0.044 (0.017) 0.822 (0.018) 0.980 0.980 -4.632* LU - LNFCU1 EURIB OR 1.864 (0.016) 0.811 (0.009) 0.980 0.980 -3.019** LU - LNFCO1 EURIB OR 1.460 (0.020) 0.782 (0.013) 0.972 0.972 -6.673* LU – LHHP1 EURIB OR 1.694 (0.020) 0.654 (0.013) 0.954 0.954 -2.557*** LU – DH2 SWAP3 1.205 (0.026) 0.557 (0.028) 0.845 0.843 -3.017** LU – DNFC2 SWAP3 -0.153* (0.044) 0.834 (0.044) 0.847 0.846 -2.338 LU – LHHP2 SWAP7 1.513 (0.083) 0.521 (0.059) 0.583 0.579 -2.124 LU – LHCC2 SWAP7 1.563 (0.058) 0.736 (0.058) 0.758 0.755 -3.582*
18 First of all let us look at what each column means. α indicates the constant difference between the bank retail rate and market rate. For example NL – DH1 averages 1.894% above the EURIBOR rate. Looking at graph 3 this seems to be correct. Short-term rates on deposits are all within 0.5% of their respective mark rate, except for NL – DH1
The cointegration coefficient µ, is a measurement of how cointegrated the bank retail rate is to the corresponding market rate. This however is only the case when the t-statistic is rejected, as this implies that the error term has no unit root and is stationary. Otherwise the cointegration coefficient has no interpreted meaning as it is dependent with the error term.
R2 is how much the dependent variable, in this case the bank retail rate, is statistically explained by its models independent variables, in this case the market rate. This is purely statistical and requires theoretical background for interpretation. In terms of interest rates a high R2 could be an indication that the chosen proxy market rate indeed explains the movement of the bank retail rate and is therefore cointegrated.
The most important column is the t-statistic one indicating cointegration. All t-statistics of interest rates with a *, **, *** are cointegrated with their matched market rate, depending on significance level. For the Netherlands this means firstly that overnight banking rates are not cointegrated. Secondly neither are both short- and long-term loans for house purchases and thirdly long-term time deposits also are not cointegrated.
For Luxembourg only two rates are not cointegrated with their proxy, long-term time deposits for non-financial corporations and Long-term loans for household purchases.
The fact that Luxembourg has more cointegrated rates could be an indication that their banking scene follows the market rates more so. This aids my hypothesis that banking
competitiveness does lead to stricter following of the ECB’s set rates. A lack of cointegration for both countries on long-term loans for house purchases could indicate that the 7 year swap rate is an incorrect proxy. Similarly cross-country differences could require that different market proxies are used between countries. This however would make comparisons impossible, but could be used when looking at the countries individually.
19
4.5 Single equation correction model (ECM)
Results are divided by country and maturity for a better overview. When calculating averages, numbers that are not significant above 10% significance level are added as zero. All standard errors are between round brackets, and are robust as we have no prove of homoscedasticity of the error terms.
Table 5, short-term results for the Netherlands of the ECM model
Interest rate Speed of adjustment Immediate pass-through Final pass-through R2 Adj. R2 NL - DHO -0.064* (0.033) 0.05** (0.022) 0.142 0.238 0.200 NL - DNFCO -0.096*** (0.034) 0.443 (0.100) 0.712 0.602 0.582 NL - DH1 -0.264 (0.067) 0.187****. (0.242) 0.580 0.256 0.219 NL - DNFC1 -0.178** (0.081) 0.812 (0.l23) 0.835 0.700 0.682 NL - LNFCU1 -0.249* (0.071) 0.478* (0.1369) 0.524 0.544 0.521 NL - LNFCO1 -0.273 (0.073) 0.937 (0.118) 0.756 0.584 0.564 NL - LHHP1 -0.021**** (0.022) 0.274 (0.063) 1.09 0.596 0.576
****Not significant at >10% level, ***significant at 10% level, **significant at 5% level, *significant at 1% level
For the Netherlands short-term rates have an average speed of -0.154 which means it moves towards the long run equilibrium as it is negative and its size indicated that monthly it 15.4% of the divergence is adjusted for in the month after. The immediate pass-through is on average 0.428. This indicates that less than half the change is immediately passed-through.
20 Final pass through is on average 0.663, which indicate it is fairly incomplete as a little over 1/3rd of the market rate is not transferred to retail rates over time, it is however higher than the immediate pass through as speed did indicate. Looking at each rate specifically LHHP1 overshoots on final pass through even though speed would indicate it is sticky. LNFCO1 ends up with a lower final-pass through even though its speed is negative and adjusts rather
quickly.
Table 6,long-term results for the Netherlands of the ECM model
Interest rate Speed of adjustment Immediate pass-through Final pass-through R2 Adj. R2 NL - DH2 -0.002**** (0.016) 0.023*** (0.029) 6.478 0.172 0.130 NL - DNFC2 -0.067* (0.023) 0.052**** (0.043) 0.878 0.422 0.393 NL – LHHP2 0.016**** (0.012) 0.011**** (0.008) 0.401 0.295 0.260 NL – LHCC2 -0.085* (0.025) 0.059** (0.025) 1.053 0.314 0.279
****Not significant at >10% level, ***significant at 10% level, **significant at 5% level, *significant at 1% level
The speed of adjustment for the Netherlands on long-term interest rates is on average -0.038, showing that while moving towards the long-run equilibrium its slower than that of short-term rates, which is to be expected as long-term rates have a longer time till they reach maturity and there for a longer time to adjust. The Immediate pass through is on average 0.022, almost completely incomplete. However the final pass-through is on average 2.422. This is mostly because of the final pass through of DH2. As is calculated using the speed of adjustment. The average, without DH2 and LHHP2 as both are not significantly different from 0, Final pass through is 0.97, which is very complete.
21
Table 7, Short-term results for Luxembourg of the ECM model
Interest rate Speed of adjustment Immediate pass-through Final pass-through R2 Adj. R2 LU - DHO -0.103** (0.034) 0.226 0.575 0.716 0.702 LU - DNFCO -0.232** (0.090) 0.743 0.582 0.671 0.655 LU - DH1 -0.131** (0.066) 0.875 (0.086) 0.653 0.579 0.557 LU - DNFC1 -0.336 (0.093) 0.900 (0.117) 0.772 0.565 0.543 LU - LNFCU1 -0.153** (0.074) 0.858 (0.055) 0.776 0.739 0.726 LU - LNFCO1 -0.641 (0.122) 0.946 (0.102) 0.779 0.449 0.421 LU - LHHP1 -0.088** (0.036) 0.231 (0.063) 0.530 0.631 0.612
****Not significant at >10% level, ***significant at 10% level, **significant at 5% level, *significant at 1% level
For Luxembourg the speed of adjustment is on average -0.241, which is actually lower than that of the Netherlands, which is in line with the hypothesis that a more competitive banking scene would converge quicker. The average of immediate pass-through for short-term rates is 0.682 which is also what our hypothesis would have predicted that it is higher than the immediate pass through of the Netherlands. Final pass-through is on average 0.667, which while slightly higher than that of the Netherlands is not in line with the other results as the final pass-through actually went down for Luxembourg. As there is no way to calculate the standard errors of the Final pass-through as it is calculated after the regression it might just be that they are not significantly different from their immediate pass-through and therefor these weird results appear. The only rates that don’t have this result are DHO and LHHP1, which do show more complete in the long-term. In short, speed and immediate pass through shows evidence that it is indeed quicker and more complete for Luxembourg on short-term rates. Final
pass-22 through I cannot say that there is a definitive difference between the countries as standard errors might be high for Luxembourg.
Table 8, long-term results for Luxembourg of the ECM model
Interest rate Speed of adjustment Immediate pass-through Final pass-through R2 Adj. R2 LU - DH2 -0.181** (0.089) 0.169****. (0.103) 0.498 0.194 0.154 LU – DNFC2 -0.088*** (0.047) 0.317** (0.146) 0.640 0.297 0.262 LU – LHHP2 -0.054 **** (0.039) 0.080**** (0.116) 0.279 0.219 0.180 LU – LHCC2 -0.222** (0.099) -0.256**** (0.151) 0.684 0.222 0.183
****Not significant at >10% level, ***significant at 10% level, **significant at 5% level, *significant at 1% level
For Luxembourg the averages for the long-term rates are -0.136, 0.079 for speed and
immediate pass-through respectively, which again is quicker and more complete than those of the Netherlands. To compare Final pass through the average, not taken into account LHHP2 as its speed is not significantly different from 0, is 0.607. This contradicts the hypothesis that final pass-through is better for Luxembourg, at least for long-term rates. This might be due to the fact that the cointegration test for long-term rates did show that some were neither cointegrated ( NL- DH2, NL- DNFC2, NL – LHHP2, LU - DNFC2, LU - LHHP2) nor stationary (LU - DH2 ).
23
Averages of speed of adjustment , based on
Short Long Household Non-financial corporations Deposits Loans
NL -0.154 -0.039 -0.124 -0.173 -0.127 0.121
LU -0.241 -0.136 -0.121 -0.230 -0.184 -0.221
Maturity was already discussed, but from a risk point of view, speed of adjustment for
household is lower than for non-financial corporations and adjustment for household seems to be higher for the Netherlands over Luxembourg, indicating that banking competitiveness might have no effect on the speed of adjustment for rates for household but only for nfcs. A possible reason that the speed of adjustment is higher for non-financial corporations then households could be, because households are less likely to swap banks than nfcs where it is peoples jobs to swap to whichever bank has the more appealing rate, therefor banks do not risk losing their market share as much, when adjusting to household rates slower. The speed of adjustment for Deposits and loans does not take into account overnight deposits, as this could lead to a repeat of the results for maturities as there are no overnight loans to match overnight deposits. Loan rates adjust quicker than deposits for the Netherlands and opposite for Luxembourg.
Averages of immediate pass-through , based on
Short Long Household Non-financial corporations Deposits Loans
NL 0.428 0.022 0.138 0.541 0.209 0.350
LU 0.682 0.079 0.222 0.753 0.523 0.451
The immediate pass-through has similar results compared to speed of adjustment. In that short maturities are more complete and nfc are more complete than household, for probably similar reasons. Deposits and loans yet again are contradicting between the Netherlands and
Luxembourg, indicating that it might be a country specific difference.
I did not include the averages of final pass through for Luxembourg and the
Netherlands as it would require not taking into account several interest rates and different ones between the two countries.
24
5. Conclusion
The interest rate pass-through is still one of the most important instruments the ECB has to achieve their goal of price and financial stability. Evidence indicates that the interest rate pass-through is still incomplete for both EU countries as was in line with previous research (Hristov et al, 2014). Multiple interest rates were researched for a complete picture of interest rate pass-through. Some of these failed statistical prove of cointegration for their chosen market proxy, with cross-country differences. The Augmented Dickey-Fuller unit root tests were performed and here there was also a lack of statistical proof that interest rates are not stationary. However this might be purely statistical as it is argued that interest rates cannot lack unit root from a theoretical standpoint (Andries & Billon, 2016. MacKinnon, 1996).
The empirical results of the Netherlands and Luxembourg on interest rate pass-through confirm that there is indeed lag in pass-through and that the lag is higher for long-term rates than for short-term. This is logical from an economic point of view, as longer maturities do not need to adjust as quickly as shorter term maturities. Household speed of adjustment rates might be unaffected by banking competitiveness, due to bank loyalty being more prevalent for
households. The comparison between the Netherlands and Luxembourg also shows that the country with a more competitive banking scene has a more complete immediate pass-through and quicker speed of adjustment on both short- and long-term rates, similar to what Leroy and Lucotte (2015) found for the period of 2003-2010.
From these results there is an argument to be made for different policy rates per country. In which the ECB could “overshoot” their rates for countries with lower banking competitiveness to make up for the lower immediate pass-through and speed. This could however lead to arbitrage issues between countries as there is no exchange rate within the EU. Because banks have branches in different Eurozone countries, they could borrow cheap in one country and transfer it to a different country to loan out or vice versa. Therefor different policy rates between countries are not a possible solution for the heterogeneity of interest rate-pass through. Policy rates could however consider the average banking competitiveness of the eurozone. If policy rates are then adjusted to overshoot above the ECB’s planned rate
accordingly, the eurozone on average could have complete pass-through to the ECB’s planned rates. This would be in line with doing whatever is best for the region as a whole.
Future research could determine the effect of banking competitiveness on the final pass-through rate, either by researching a different model that could calculate the standard error of final-pass through or by perhaps choosing different proxies for certain rates that prove
25 cointegration for both countries. Different tests could also be used to check for different
cointegration results between tests e.g. the Phillips-Ouilaris test and Johansen test. Results from this thesis have indicated that the proposed policy rate change could work considering Luxembourg and the Netherlands. A broader research including all Eurozone countries could be done to investigate if the suggested change on policy rates can work on the Eurozone as a whole. All these recommendations together can help further the research on one of the ECB’s most important monetary transfer mechanics, the interest rate pass-through.
6. References
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. H., Beckmann, J., & Verheyen, F. (2012). Interest rate pass-through in the EMU–New evidence from nonlinear cointegration techniques for fully harmonized data.
De Bondt, G. J. (2005). Interest rate pass-through: empirical results for the Euro Area. German Economic Review, 6(1), 37-78.
Dickey, D., Fuller, W., 1979. Distribution of the estimators for autoregressive time series with a unit root. J. Am. Stat. Assoc. 74 (366), 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 (1999). “The stability-oriented monetary policy strategy of the
Eurosystem”, Monthly Bulletin, January, 39-50
Europiean Central Bank. (2017, October), Report on financial structures European Central Bank (2009). Recent developements in the retail bank
interest rate pass-through in the Euro area, Monthly Bulletin, August
Hamilton, J. D. (1994). Time series analysis (Vol. 2). Princeton: Princeton university press, 766
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,
26 Issing, O. (2003). Monetary and financial stability: is there a trade-off?. BIS Papers, 18,
16-23.
Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica: Journal of the Econometric Society, 1551-1580.
Kleimeier, S., & Sander, H. (2006). Expected versus unexpected monetary policy impulses and interest rate pass-through in euro-zone retail banking markets. Journal of Banking & Finance, 30(7), 1839-1870.
Leroy, A., & Lucotte, Y. (2015). Heterogeneous monetary transmission process in the
Eurozone: Does banking competition matter?. International Economics, 141, 115-134. MacKinnon, J. G. (1996). Numerical distribution functions for unit root and cointegration
tests. Journal of applied econometrics, 601-618.
Mishkin, F. S. (1992). Anatomy of a financial crisis. Journal of evolutionary Economics, 2(2), 115-130.
Mishkin, F.S., Matthews, K., & Giuliodori, M. (2013). The Economics of Money, Banking &
Financial Markets. Pearson Education.
Mojon, B. (2000). Financial structure and the interest rate channel of ECB monetary policy. 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.
Said, S. E., & Dickey, D. A. (1984). Testing for unit roots in autoregressive-moving average models of unknown order. Biometrika, 71(3), 599-607.
Schwert, G. W. (2002). Tests for unit roots: A Monte Carlo investigation. Journal of Business
& Economic Statistics, 20(1), 5-17.
Sørensen, C. K., & Werner, T. (2006). Bank interest rate pass-through in the euro area: a cross country comparison (No. 580). ECB working paper.
Stock, J. H., & Watson, M. W. (2016). Introduction to econometrics. Boston, Pearson/ Addison Wesley.
