Amsterdam Business School
MSc Business Economics: Finance Track
Master Thesis
The Impact of Price Leadership Bans on the Pricing Behavior in the Dutch
Mortgage Market
July 2016 Danny Schertz Thesis supervisor: Mr. M.A. Dijkstra MSc (University of Amsterdam)
ii
Statement of Originality
This document is written by student Danny Schertz who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are 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.
iii
ACKNOWLEDGEMENT
I would like to thank my thesis supervisor Mr. M.A. Dijkstra MSc of the Faculty of Economics and Business at the University of Amsterdam and my supervisor Dr. B. Overvest at the Netherlands Bureau for Economic Policy Analysis. Mr. Dijkstra helped me when I had questions about my research and frequently provided me with useful feedback. I would like to thank Mr. Overvest for the interesting input he provided me with and the many conversations we had about the research. His door was always open whenever I had questions about my thesis. Both Mr. Dijkstra and Mr. Overvest allowed this thesis to be my own work, but steered me in the right direction whenever they thought I needed it.
Furthermore, I want to thank the employees and other interns at the Netherlands Bureau for Economic Policy Analysis for helping me and making my internship enjoyable. Finally, I would like to thank the Netherlands Bureau for Economic Policy Analysis for providing me with the necessary data and offering me the opportunity to do this interesting internship over the last few months.
iv
Abstract
During the spring of 2009, the margins on mortgage rates in the Netherlands increased. This thesis investigates whether the price leadership bans that were imposed on ING, ABN Amro and AEGON by the European Commission influenced the pricing behavior of banks in the Dutch mortgage market. Building upon the theoretical and empirical literature about price leadership, Granger causality Wald tests, Error Correction Models, and Chow break tests are used to empirically analyze the impact of the price leadership bans on the pricing behavior of banks in the Dutch mortgage sector. The dataset allows controlling for characteristics of individual mortgages, making this thesis unique in its approach. Looking at newly issued mortgages with 10 year fixed rates and NHG coverage, the outcome of the Granger causality Wald test shows that Obvion is the price leader in the mortgage market and ING, SNS Bank, ABN Amro and Achmea are the price followers. Although it is not determined when the pricing behavior of the price followers exactly changed, the findings partly support the view that the price leadership bans disrupted competition in the Dutch mortgage market. Keywords: Dutch mortgage market, price leadership bans, European Commission, collusive price leadership, competition.
v
Table of contents
1 Introduction ... 1
2 Mortgage rates and price leadership bans ... 3
2.1 Mortgage rates in the Netherlands ... 3
2.2 The financial crisis, price leadership bans and the European Commission ... 4
2.3 ING ... 5
2.4 ABN Amro ... 5
2.5 AEGON... 6
2.6 Announcement effect ... 6
3 Price leadership bans and price setting behavior in the market ... 7
3.1 Theory: from a barometric to a collusive price leadership equilibrium ... 7
3.2 Theory applied to the Dutch mortgage market ... 8
3.3 Empirical studies on mortgage rates in the Netherlands ... 10
3.4 Granger causality and (Vector) Error Correction Models in previous empirical studies ... 12
4 Methodology ... 13
4.1 Model specification ... 13
4.2 Variable of interest and control variables ... 15
5 Data ... 16 5.1 Data sources ... 16 5.2 Dataset construction ... 18 5.3 Descriptive statistics ... 20 6 Empirical results ... 22 6.1 Results ... 22
6.1.1 Granger causality Wald test and correlation matrix ... 23
6.1.2 Time series analysis ... 25
6.1.3 Individual observations ... 27
6.1.4 Individual observations: announcement effect ... 29
6.1.5 Chow break test ... 31
6.2 Robustness checks ... 33
6.2.1 Interaction between cost variables and the price leadership ban dummy ... 33
6.2.2 Asymmetric short-run response to changes in mortgage rates of the price leader ... 34
7 Conclusion and discussion ... 36
8 References ... 39
1
1
Introduction
Since the start of the financial crisis in 2007, the European Commission has approved aid to twelve of the top 20 European banks (Adamczyk and Windisch, 2015). By March 2013, EU governments had spent around €1.5 trillion, equivalent to 13 per cent of total EU GDP on State aid in the banking sector (Franchoo, Baeten and Cranley, 2013). The bailouts were aimed at preventing contagion effects within the banking sector, as well as ensuring competition was not disrupted as a consequence of the State aid (De Kok, 2015). Subsequently, the European Commission set price leadership conditions, restricting banks that received State aid from offering their products at more favorable prices than their (three) best competitor(s) (Dijkstra, Randag and Schinkel, 2014).
Since the fall of 2008, ABN Amro, AEGON, ING and SNS REAAL received State aid in order to remain viable. Of these banks only SNS REAAL was not restricted by price leadership bans. Rabobank, the largest mortgage provider in the Netherlands, did not require State aid. In May 2009, the margins on Dutch mortgage rates increased significantly. The Association of Homeowners and the Dutch Consumer Organization noticed the increase in margins and suspected collusive behavior among the Dutch mortgage banks (Overvest and Tezel, 2014). The International Monetary Fund (IMF) also stated in 2010 that competition might have been weakened in the Netherlands as a result of the price leadership bans (IMF, 2010). The impact of the price leadership bans on the mortgage rates set by Dutch banks remains a topic of discussion and therefore the following research question is addressed:
What is the impact of the imposed price leadership bans by the European Commission on the pricing behavior of banks in the Dutch mortgage market?
De Haan and Sterken (2006) found that during the period preceding the price leadership bans, there was a price leader in the Dutch mortgage market and prices were set competitively. This is consistent with a form of barometric price leadership, where the price leader serves as a barometer for the rest of the industry (Markham, 1951; Cooper, 1996). Looking at the characteristics of the Dutch mortgage market, it can be assumed that Rabobank was the price leader in the market. Price leadership helps to facilitate collusion when firms have asymmetric information (Rotemberg and Saloner, 1990) and the imposed price leadership bans prohibited large banks to undercut their best priced competitors. Given the concentrated Dutch mortgage market, the price leadership bans effectively selected Rabobank as the price leader in the market (Dijkstra, Randag and Schinkel, 2014). Around May 2009, it became apparent that price leadership bans would be imposed on Dutch banks, which possibly resulted in a stable collusive price leadership from then onwards. In the collusive outcome, the mortgage rates set by the price followers are more dependent on the price set by the price leader when compared to the
2 competitive outcome. This study focuses on the impact of the price leadership bans on the pricing behavior of the price followers in the official time period in which the bans were put in place. The impact of the announcement effect of the price leadership bans, which is around May 2009, on the level of competition in the Dutch mortgage market is also investigated.
The price leader in the Dutch mortgage market is determined with the help of Granger causality Wald tests. To account for non-stationarity and cointegration in the data, an Error Correction Model (ECM) is used to investigate how the relationship between the mortgage rates set by the price leader and mortgage rates set by the price followers changed after the price leadership bans were imposed. The analyzed data in this study is obtained from the Dutch Securitisation Association1 and consists of Residential Mortgage Backed Securities (RMBS) from banks operating in the Dutch mortgage market. After combining data from multiple banks, the main dataset consists of 883,930 observations on mortgage loans in the time period between March 2006 and September 2014. By looking at the individual lender level and using a new data source, this thesis contributes to the discussion about the impact of both State aid and European regulation on the level of competition in the Dutch mortgage market. The total sample of 883,930 observations is used to indicate the price leader in the Dutch mortgage market with the help of Granger causality Wald tests.
A final subsample of 168,308 mortgage rates from Obvion, ING, SNS Bank, ABN Amro and Achmea is used for the ECM approach. Mortgage rates of Obvion, which is a subsidiary of Rabobank, serve as a proxy for the rates set by Rabobank. The availability of lender characteristics on individual mortgages makes this dataset more extensive than datasets used in previous research, wherein mortgage rates are studied using specific bank level data or overall banking sector data.2 Furthermore, to account for funding costs of mortgages, multiple cost variables are included in the cointegrating equation and ECM. The results show that Obvion is the price leader in the market and ING, SNS Bank, ABN Amro and Achmea are the price followers when newly issued mortgages that have National Mortgage Guarantee (NHG) coverage and 10 year fixed rates are considered. The findings of the ECM approach partly support the theory that the price leadership bans resulted in less competitive pricing behavior, thereby disrupting competition in the Dutch mortgage market.
This thesis is structured as follows. Section 2 discusses the development of mortgage rates in the Netherlands and focuses on the price leadership bans imposed by the European Commission. Section 3 discusses the theoretical impact of the price leadership bans on the price setting behavior in the Dutch mortgage market and offers an overview of previous empirical studies on pricing behavior. Section 4 describes the methodology. Section 5 contains the data sources, dataset construction and descriptive statistics. Section 6 presents the results and the robustness checks. Finally, the conclusion and discussion are presented in section 7.
1
Access is provided by the Netherlands Bureau for Economic Policy Analysis (CPB).
3
2
Mortgage rates and price leadership bans
This section focuses on the mortgage rates in the Netherlands and the price leadership restrictions. First, the development of mortgage rates in the Netherlands and surrounding countries is presented. Second, the price leadership bans that were imposed by the European Commission are described. After that, an overview of the restrictions that were imposed on Dutch banks is presented. Lastly, the potential announcement effect related to the price leadership bans is described.
2.1 Mortgage rates in the Netherlands
Before the spring of 2009, the average monthly mortgage rate in the Netherlands was closely linked to the average rates of surrounding countries like Belgium, France and Germany (Bijlsma et al., 2013). During the spring of 2009, the mortgage rates in the Netherlands remained relatively high, thereby not reflecting the lower swap rates and Euribor rates that the loosened ECB policy had triggered (Dijkstra, Randag and Schinkel, 2014). Figure I displays the average borrowing costs for house purchases in the Netherlands and surrounding countries. Starting in the spring of 2009, the lending rates for households in the Netherlands are relatively high compared to surrounding countries.
Figure I
Borrowing costs for house purchase in the Netherlands and surrounding countries
This figure shows the cost of borrowing (in %) for households for house purchase in the Netherlands, surrounding countries and the Euro area. Source: European Central Bank – Statistical Data Warehouse.
It remains unclear what exactly caused the mortgage rates in the Netherland to rise compared to the trend in neighboring countries in the spring of 2009. In the literature and the ongoing debate multiple
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 20 16-03 20 15-10 20 15-05 20 14-12 20 14-07 20 14-02 20 13 -09 20 13-04 20 12-11 20 12-06 20 12-01 20 11-08 20 11-03 20 10 -10 20 10-05 20 09-12 20 09-07 20 09-02 20 08-09 20 08-04 20 07-11 20 07-06 20 07-01 20 06-08 20 06-03 20 05-10 20 05-05 20 04-12 20 04-07 20 04-02
4 explanations are given, which can be categorized in two main classes: lowered competition and higher funding costs (Dijkstra, Randag and Schinkel, 2014). Appendix 1 provides an overview of the literature on the funding costs of mortgages in the Netherlands. Appendix 2 discusses the level of competition in the Dutch mortgage market and discusses the potential impact this could have on mortgage rates. An increase in funding costs, changes in regulation, higher levels of market concentration and entry barriers do not seem to be able to fully explain the relatively high Dutch mortgage rates. The price leadership bans that were imposed by the European Commission seem to offer a possible explanation for this increase, especially since they could justify the rapid margin increase in the spring of 2009. Therefore, the main focus of this thesis is to analyze the influence of the price leadership bans on the mortgage rates in the Dutch mortgage sector. The following section is about the price leadership bans set by the European Commission.
2.2 The financial crisis, price leadership bans and the European Commission
Between October 2008 and October 2014, the European Commission took over 450 State aid decisions and the European Commission approved State aid to 112 banks, which represent around one third of the European banking sector by assets (Franchoo, Baeten and Solek, 2015). During the financial crisis in 2008, there was no specific crisis management mechanism at the EU-level. The absence of such a mechanism resulted in State aid becoming the mechanism for bank restructuring and rescue. The European Commission established a set of exceptional rules, called the Crisis Communications, to specify how governments could use State aid in the financial sector (De Kok, 2015). The Crisis Communications framework outlines the support offered to the financial sector by governments, with the goal of protecting financial stability. Accordingly, the Crisis Communications framework tries to minimize the impact the rules have on competition across Member States and between banks (De Kok, 2015).
Banks that receive State aid have an advantage compared to better performing banks that do not receive State aid. This increases the moral hazard in the European banking sector. Moral hazard could lead to lower competitive incentives, since State aid prevents banks from driving competition out of the industry. This can potentially result in a distortion in the level of competition in the financial sector (De Kok, 2015).The State aid could also be used for predatory pricing, which can be seen as discounting prices in order to obtain market power at the cost of competitors (Bolton, Brodley and Riordan, 1999; NMa, 2011). Therefore, price leadership bans can be imposed on financial institutions that receive State aid, preventing them from offering their products at better terms than their competitors (Baudenbacher and Bremer, 2010).
A ban on aggressive pricing policies by banks that received State aid can result in the distortion of competition in the banking sector within a specific country. The price leadership bans seem especially incompetent for banks in countries (such as Greece) where most of their competitors
5 also received State aid, since this is likely to reduce effective competition. Therefore, other sufficient mechanisms could be imposed, as happened in the UK and Ireland where market entry was forced3 (Franchoo, Baeten and Solek, 2015).
The following paragraphs briefly highlight decisions made by the European Commission about State aid received by active banks in the Dutch mortgage market. Despite the fact that SNS Reaal did receive State aid,4 no price leadership ban was thought to be necessary.5 An overview of the decisions related to price leadership bans, the specific dates and some of the other restrictions that were imposed by the European Commission are presented in table I.
2.3 ING
After an agreement between the Dutch State and ING, the European Commission approved the restructuring plan of ING in November 2009. The plan included the prohibition of acquiring other firms, the carving out of Westland Utrecht Hypotheekbank (WUH) and a price leadership ban.6 The price leadership ban included the restriction that ING could not offer below its three best priced competitors (Overvest and Tezel, 2014). All these commitments were planned to stay in place for three years. On 16 November 2012,7 the European Commission decided that the price leadership ban for ING was not extended in the Netherlands for, amongst others, savings, mortgages and private banking.8 For ING in the Netherlands, it can therefore be concluded that the imposed price leadership ban was applicable from November 2009 until November 2012.
2.4 ABN Amro
In October 2008, Fortis Bank Nederland which included assets of ABN Amro Holdings, was acquired by the Dutch State. Following this acquisition, there was a lot of correspondence between the Dutch State and the European Commission.9 In March 2010, the Dutch State informed the European Commission that a price leadership ban would apply until the end of 2010.10 In July 2010, this price leadership ban commitment was extended by the Dutch State until June 201111 and in April 2011 the European Commission decided that ABN Amro could not be the price leader for mortgage products for a period of three years.12 The European Commission also restricted ABN Amro from acquiring
3
Costs of market entry were reduced in Ireland by providing new entrants access to the clearing system of Ireland’s Bank, thereby allowing them to postpone investments in their personal infrastructure.
4
See Commission (IP/10/82), Brussels, 28th January 2010. 5
See The President of the House of Representatives of the States-General, 23 August 2013. 6 Commission Decision No. 2010/608/EC (L 274/139) [84-85].
7
Commission Decision No. 2013/719/EU (L 337) Annex I: Catalogue of Commitments (d). 8
See The President of the House of Representatives of the States-General, 23 August 2013. 9
Commission Decision No. 2011/823/EU (L 333) [2-22]. 10 Commission Decision No. 2011/823/EU (L 333) [23-24]. 11
Commission Decision No. 2011/823/EU (L 333) [26]. 12 Commission Decision No. 2011/823/EU (L 333) [325-329].
6 control above a specific threshold. Multiple other behavioral restrictions were imposed, including the prohibition to pay dividends and the prohibition to advertise with the received State aid (De Kok, 2015). The price leadership ban on ABN Amro in the Netherlands expired on April 5th 2014.13 It can therefore be concluded that the imposed price leadership ban on ABN Amro was applicable from March 2010 to April 2014.
2.5 AEGON
After receiving State support in 2009, AEGON became one of the price leaders in the Dutch mortgage market. In order to avoid unfair competition, the European Commission decided in August 2010 that AEGON should be restricted by a price leadership ban.14 Under the price leadership ban, AEGON was not allowed to offer prices below its two or three (depending on specific mortgage characteristics) best priced competitors.15 Before the Dutch State was fully repaid, AEGON was also restricted to acquire stakes surpassing 20% in businesses. AEGON offered relatively low prices, thereby contributing to competition in the banking sector. The imposed price leadership ban on AEGON, at the same time that ING and ABN Amro had been restricted by price leadership bans, potentially had a negative impact on the level of competition in the Dutch banking sector (De Kok, 2015). AEGON repaid the last part of the State aid, a total of €3 Billion in June 201116 and the price leadership ban was subsequently lifted in July 2011. It can therefore be concluded that the imposed price leadership ban on AEGON was applicable from August 2010 until July 2011.
2.6 Announcement effect
The Netherlands Competition Authority (NMa, 2011) states that it is unlikely that the price leadership bans resulted in collusive behavior which led to the higher margins in the Dutch mortgage sector, since these margins increased before the first restriction by the European Commission was imposed in November 2009. In April 2009, European Commissioner Kroes stated that in order to provide ING with State support, ING should agree on a possible price leadership ban. The imposed price leadership ban on the German Commerzbank in May 2009 also showed the willingness of the European Commission to impose the bans. It is therefore possible that the imposed price leadership bans in the Dutch banking sector were anticipated by the banks from May 2009 onwards (Dijkstra, Randag and Schinkel, 2014). This announcement effect could potentially explain a shift towards a stable collusive outcome in May 2009, which is the subject of the next section. This thesis therefore also investigates
13
ABN AMRO N.V. – Registration Document, 30 May 2012. 14 Commission Decision No. 372/2009 [116-117].
15
Commission Decision No. 372/2009 [69-72].
7 whether this announcement effect had an impact on the pricing behavior in the Dutch mortgage market.
Table I
Overview of important dates related to price leadership bans in the Netherlands Bank Date price leadership ban
imposed
Expiration date price leadership ban
Announcement effect
ING November 2009 November 2012 May 2009
ABN Amro March 2010 April 2014 May 2009
AEGON August 2010 July 2011 May 2009
Other banks - - May 2009
This table presents the dates at which ING, ABN Amro and AEGON were restricted by the European Commission to be a price leader in the Dutch mortgage market. The expiration dates of the price leadership bans are also presented. The last column presents the date belonging to the date of the announcement effect of the price leadership bans, which is May 2009 for all banks.
3
Price leadership bans and price setting behavior in the market
This section focuses on the potential impact of the price leadership bans on the price setting behavior in the Dutch mortgage market. First, the impact of the price leadership bans on the pricing strategy of Dutch banks is described from a theoretical point of view. Second, an overview of empirical studies on mortgage rates in the Netherlands is presented. Finally, methodologies in previous studies to determine price leadership and pricing setting behavior are described.
3.1 Theory: from a barometric to a collusive price leadership equilibrium
Stigler (1947) describes barometric price leadership, referring to the case where one firm is the first to set its price. The price leader in the barometric leadership model has better knowledge of market conditions and therefore acts as a barometer for the other market participants (Markham, 1951). Lanzillotti (1957)17 describes a number of characteristics of the markets in which barometric price leadership could emerge and points out that, although the market can be highly concentrated with a small number of firms, it is important that there is a competitive fringe. Cooper (1996) addresses barometric price leadership in a duopolistic two stage model in which barometric price leadership can only exist if firms have asymmetric information, since uninformed firms are willing to wait for the informed firm to set its price. The uninformed players in the market can obtain information about the demand in the market without incurring costs, thereby free-riding on the informed firm (Cooper, 1996). In the case of barometric price leadership, the price followers undercut the price leader which eventually results in competitive prices (Markham, 1951; Rotemberg and Saloner, 1990). In
17
See Lanzilotti (1957) Competitive Price Leadership – A Critique of Price Leadership Models, for a full outline of the features of barometric price leadership.
8 oligopolistic markets with differentiated products, the leader has no power to coerce followers into setting their price uniformly and therefore a state of punishment is unlikely to occur (Rotemberg and Saloner, 1996).
Another kind of price leadership described in the literature is collusive price leadership. Rotemberg and Saloner (1990) use a duopolistic model in which they show that price leadership by the best-informed firm leads to collusion in the case of asymmetric information. In the model of Rotemberg and Saloner (1990) two firms compete in a super-game setting where the leader has better information than the follower. Rotemberg and Saloner (1990) show that the uninformed firm prefers the position of the follower. The price leader discloses its price, and the follower is supposed to set the same price. If the follower deviates from this price, a price war ensues in which the prices go back to the competitive equilibrium. Rotemberg and Saloner (1990) assume that the firms remain in the competitive equilibrium forever. Collusive price leadership can therefore only materialize with compliance of the follower, since deviation results in an extensive period of low pricing. Collusive price leadership results in higher prices and higher profits than when the competitive outcome is reached. Mouraviev and Rey (2011) also look into the role of price and quantity leadership in a super-game where firms move sequentially and show, in line with Rotemberg and Saloner, that leadership promotes collusion. Their model indicates that leadership is an extra powerful device for collusion when Bertrand competition is compared to Cournot competition. When the leader deviates, this is immediately noticed and penalized by the follower.
3.2 Theory applied to the Dutch mortgage market
The Dutch mortgage market is characterized by multiple large banks and a small competitive fringe (NMa, 2011). The findings of De Haan and Sterken (2006) that one bank behaves as a price leader and prices are set competitively is consistent with barometric price leadership. Rabobank has traditionally been a bank that consists of many different local cooperative banks (Groeneveld, 2016), which may have provided Rabobank with a better ability to obtain market information about demand changes than its competitors. Consequently, it is assumed that Rabobank was the better informed bank in the Dutch mortgage market and has been the barometric price leader in the Dutch mortgage market before the price leadership bans were imposed.18 Investigating the mortgage market between 2004 and 2011, the NMa (2011) reported that Rabobank stopped publishing their mortgage rates from June 2006 onwards. The fact that Rabobank did not publish its mortgage rates was also noticed by the European Commission.19 However, the rates set by Obvion, a subsidiary of Rabobank, were public20 and could have been used to serve as a proxy for the rates set by Rabobank. This facilitates Rabobank
18
See Randag and Schinkel (2012) for an extensive argument of price leadership in the Dutch mortgage market. 19
Commission Decision No. 372/2009 [70].
9 to act as a price leader for the rest of the Dutch mortgage market.
In 2004, Rabobank, ING and ABN Amro accounted for over 60 percent of the Dutch mortgage market (NMa, 2011). In November 2009, a price leadership ban was imposed on ING. In 2010, price leadership bans were also imposed on ABN Amro and AEGON.21 Two out of the top three largest players in the Dutch mortgage market were thereby restricted from undercutting mortgage rates set by their (three) best priced competitor(s). Given the concentrated Dutch mortgage market, the imposed price leadership bans thereby effectively selected Rabobank as the price leader in the market (Dijkstra, Randag and Schinkel, 2014). Price leadership can facilitate collusion (Rotemberg and Saloner, 1990; Mouraviev and Rey, 2011) and price leadership bans prohibited large banks to undercut their best priced competitors, which contributed to the stability of the collusive outcome after November 2009. Considering the low capacity of the small competitive fringe that was not restricted by the price leadership bans, it would have likely been more profitable to follow the prices set by Rabobank than to gain market share by undercutting Rabobank. This could explain why the non-restricted banks would adhere to prices set by Rabobank and the collusive outcome could become stable after November 2009.
As mentioned in section 3, there was also a potential announcement effect related to the imposed price leadership bans. Knowing that in the near future it would be impossible to undercut Rabobank, undercutting the price set by Rabobank decreased the expected future profits of deviation. This could have made the collusive outcome stable from May 2009 onwards.22 When it became apparent that price leadership bans would be imposed on Dutch banks (around May 2009), it is likely that Rabobank responded by increasing their mortgage rates which would result in higher prices and profits for both the leader and the followers (Rotemberg and Saloner, 1990). Higher entry barriers and high levels of concentration in the Dutch mortgage market could have also promoted a shift towards a collusive equilibrium (see appendix 2.1 and 2.2).
During the period preceding the price leadership bans, the prices were set competitively (De Haan and Sterken, 2006), meaning that the followers undercut the price set by the price leader and prices are mostly based on their funding costs (De Haan and Sterken, 2011). In the collusive equilibrium, prices and profits of the price leader and followers are higher than during competition and the prices set by the followers would be equal or above the price set by the price leader. In the collusive outcome, the mortgage rates set by the price followers are more dependent on the price set by the price leader when compared to the competitive outcome. When a shift from a barometric price leadership towards a collusive price leadership occurs it is expected that the mortgage rates set by the price followers become more dependent on the mortgage rate set by the price leader. The following
21
In 2010, AEGON had a top five position in newly registered mortgages with a market share of 8.3% (NMa 2011). 22
See Randag and Schinkel (2012) for an extensive argument on how the anticipation of the price leadership bans could have resulted in a shift towards a collusive outcome.
10 section describes the used methodology to investigate whether the pricing behavior of banks operating in the Dutch mortgage market changed during May and November 2009.
3.3 Empirical studies on mortgage rates in the Netherlands
Toolsema and Jacobs (2007) look at the average of the Dutch mortgage rates and investigate if mortgage rates are asymmetrically adjusted if the mortgage costs change. These costs are measured by the 10 year Dutch government bond and the 5 year Dutch swap rate. They find that positive cost shocks are reflected faster in the mortgages rates than negative cost shocks, which is disadvantageous to consumers. They conclude that the most likely explanations for this asymmetric response to capital market changes are tacit price coordination and search or switching costs.
De Haan and Sterken (2006 & 2011) look into the pricing behavior of the four largest banks in the Dutch mortgage market between 1997 and 2003 and examine Dutch mortgage rates for individual banks at a daily frequency. In their 2006 paper, they use a Vector Error Correct Model (VECM) and Granger causality tests to look into the pricing behavior of the four largest Dutch banks and find that the interest rate of one bank plays a significant role in the price setting process of other banks and behaves as a price leader. Using a structural conjectural variation model, they find that even though one bank behaves as a price leader, the followers set their prices competitively.
De Haan and Sterken (2011) use the most popular mortgages which are those with interest rate renewals after five and ten years. To capture both short-term and long-term effects, a Vector Error Correction Model is used. Their model assumes that in the long-run banks price mortgage rates at a surcharge for funding costs, which are measured by the 5 and 10 year Dutch government bond rates. They find that in the long-run, two out of the four largest Dutch banks have been less sensitive to cost changes than the others. The short-run results indicate that banks adjust their prices more strongly to funding cost decreases than increases, which is beneficial to consumers and indicates competitiveness. Therefore, they conclude that there is no reason to think that competition is constrained in the Dutch mortgage market (De Haan and Sterken, 2011).
Using monthly data between 2004 and 2010, Mulder and Lengton (2011) present the econometric analysis and results of the NMa (2011) report. Using time series analysis at the industry level, they look at the effect of cost, risk and market structure variables on the average mortgage rate in the Netherlands. The dependent variable in their model is the average monthly interest rate and swap rates, Euribor rates, deposit rates and interest rates on RMBS are used as cost variables. The market structure variables are the Herfindahl-Hirschmann Index (HHI), C3 and C4 levels. The standard deviation of the 6 month Euribor is used to reflect risks in the money market. Their findings indicate that mortgage risks and costs have a positive but insignificant effect on the interest rate and the lagged average monthly interest rate, which is included to account for autocorrelation, is the most essential explanatory variable. They also find that the variables related to the market structure do not
11 seem to influence the monthly interest rates (Mulder and Lengton, 2011).
Furthermore, Mulder and Lengton (2011) measure how often the banks that were subject to price leadership bans were offering the lowest mortgage rates in the market and a heavy decline can be seen in 2010. The results indicate that mortgage rates tend to increase when banks that are subject to price leadership bans, price less competitively. This indicates that the price leadership bans have led to higher mortgage rates (Mulder and Lengton, 2011; Overvest and Tezel, 2014). Another interesting finding of Mulder and Lengton (2011) is that mortgage rates increase between 0.10 and 0.20 percentage points if the market concentration, measured by C3, rises with one standard deviation. For robustness they also look at actual mortgage rates compared to window mortgage rates in the initial analysis, which results in the same conclusions (Mulder and Lengton, 2011).
Mulder (2014) extends the study of Mulder and Lengton (2011) and measures the level of competition with a bank-specific Lerner Index23 over the period 2005 - 2010. The marginal costs are measured using both a translog cost function and the derivative of the industry-wide funding costs, which are based on deposit rates, swap rates, Euribor rates, RMBS premiums and CDS spreads. Mulder finds that the increase in concentration in the Dutch mortgage market and the imposed price leadership bans had a negative impact on the level of competition, which is indicated by a higher Lerner index. Table II provides an overview of the studies about competition in the Dutch mortgage market and presents their main findings.
23
Measurement of market power which is calculated by subtracting the marginal costs of providing mortgages from the mortgage rate. This term is divided by the mortgage rate.
12 Table II
Overview of studies on pricing behavior in the Dutch mortgage market
Study Method and measure Sample Main results
De Haan and Sterken (2006)
Granger causality, VECM and discrete choice model to investigate price leadership and competitive pricing.
Daily advertised 5 and 10 year fixed mortgage rates between 1997 and 2003.
One bank behaves as the price leader and prices are set competitively.
Toolsema and Jacobs (2007)
Error Correction Model to see if Dutch mortgage rates respond asymmetrically to changes in costs.
Monthly averages 5 year fixed mortgage rates between 1978 and 2000.
Positive cost shocks are passed through faster than negative shocks. Tacit price coordination and search costs most likely explanations.
De Haan and Sterken (2011)
Error Correction Model to assess whether Dutch mortgage rates respond asymmetrically to changes in costs.
Daily advertised 5 and 10 year fixed mortgage rates between 1997 and 2003.
In the long-run prices are based on funding costs which is an indication of
competition. The short-run results indicate that banks adjust their prices more strongly to funding cost decreases than increases which is another indication of competitive pressure.
Mulder and Lengton (2011)
Time series analysis to look at the impact of the price leadership bans on mortgage rates.
Monthly and annual data between 2004 and 2010.
Banks subject to price leadership bans price less competitively and mortgage rates tend to increase. Indication that the price leadership bans have led to higher mortgage rates.
Mulder (2014)
Measures impact of concentration and price leadership bans on the competitiveness using a bank-specific Lerner Index.
Monthly averages of 5 year fixed mortgages with NHG between 2005 and 2010.
Indications that an increase in the industry concentration and price leadership bans had a negative impact on competition.
This table provides an overview of the reviewed literature on pricing behavior in the Dutch mortgage market. For each study the used methodology, sample of interest and main results are stated.
3.4 Granger causality and (Vector) Error Correction Models in previous empirical studies
De Haan and Sterken (2006) use Granger and instantaneous causality within their VECM and find that the mortgage rate set by one bank has a significant impact on the mortgage rates set by other banks. The results indicate that one bank shows price leadership behavior and does not follow mortgage rates set by other banks. Masih & Masih (2001) analyze dynamic causal relations between nine international stock market indexes using Granger causality tests within a VECM. The findings show that in both the short- and long-run the US and UK markets show leadership behavior. Peiers (1997) also uses Granger causality tests to identify a price leader on the foreign exchange market. Interventions by the central banks are used as triggers for the positioning of the price leader. The findings indicate that lagged quote changes of Deutsche Bank Granger-cause current quote changes of other banks, while current quote changes of Deutsche Bank are not Granger-caused by lagged quote changes of these banks.
As mentioned in the previous paragraph, De Haan and Sterken (2011) and Toolsema and Jacobs (2007) use ECMs to assess whether Dutch banks responded asymmetrically to cost shocks. Allen and McVanel (2009) also use an ECM to look into the price setting behavior of the six largest
13 banks in the Canadian mortgage market and find evidence that mortgage rates are more quickly adjusted upwards than downwards. One of the explanations they provide for this asymmetry in adjustments is that explicit or implicit collusive behaving banks are reluctant to lower their rates after a cost decrease, since it could signal that the bank is deviating from the collusive equilibrium (Allen and McVanel, 2009).
The banking sector is not the only industry were ECMs are used to test for asymmetrical price adjustments. There is also a substantial amount of literature that uses ECMs to look at asymmetric price adjustments in gasoline markets. Peltzman (2000) uses an ECM to look at asymmetric price adjustments in multiple industries like agriculture, textiles and crude oil. Using an ECM model, Borenstein, Cameron and Gilbert (1997) find that prices of retail gasoline respond quicker to increases in crude oil prices than to decreases in crude oil prices. If their sales remain above a certain level, firms with market power will try to keep their prices above the competitive level. Short-run market power could therefore be a possible explanation for their findings (Borenstein, Cameron and Gilbert, 1997).
Studies on (tacit) collusive price leadership often use ECMs to test whether cost shocks are asymmetrically passed through in the prices. Instead of focusing on cost shocks, this thesis investigates whether the pricing behavior of banks changed after price leadership bans were imposed on banks in the Dutch mortgage sector. To account for non-stationarity and cointegration in the mortgage rates an ECM is used to see whether the price setting behavior of the price followers on the Dutch mortgage market changed during the price leadership ban periods.
4
Methodology
This section describes the methodology used to test the hypothesis. First, the model is described. Second, the main variable of interest and the control variables are described.
4.1 Model specification
The previous section explained that the price leadership bans that were imposed by the European Commission seem to offer an explanation for the increase in mortgage rates. As explained in the theory, the price leadership bans could have resulted in a shift from a barometric price leadership towards a collusive price leadership, with Rabobank being the price leader. It is therefore expected that the mortgage rate set by the price leader is followed more closely by the price followers during the period of the price leadership bans. To test the hypothesis that the mortgage rates set by the price followers became more dependent on the mortgage rate set by the price leader as a result of the price leadership bans, multiple tests are performed.
14 and Peiers (1997) use Granger causality tests to indicate the price leader in the market. Granger causality in the mortgage market would indicate that past mortgage rates set by a specific bank contains information that is helpful to forecast changes in the mortgage rates set by other banks, beyond the information that is contained in the past mortgage rates of these banks (Stock and Watson, 2012). Accordingly, a Granger causality Wald test is used to test whether mortgage rates of one of the banks Granger-cause changes in the mortgage rates set by the other banks and therefore acts as a price leader.
The monthly mortgage rates set by banks can be non-stationary, as discussed by previous research on mortgage rates such as de Haan and Sterken (2006) and Mulder and Lengton (2011). It is important to test the data for non-stationarity, since non-stationarity would distort the analysis by making confidence intervals and hypothesis tests unreliable (Stock and Watson, 2012). Therefore, the monthly mortgage rates for all banks are tested on non-stationarity with the Augmented Dickey-Fuller test (ADF test).24 This test needs to be performed with the appropriate number of lags, which is determined using the Akaike Information Criterion (AIC) and the Schwarz Information Criterion (SIC) (Stock and Watson, 2012). To test for cointegration between the mortgage rates set by the price leader and the mortgage rate set by price followers, the Engle-Granger Augmented Dickey-Fuller test (EG-ADF test) for cointegration is executed (Stock and Watson, 2012). For this test the average monthly mortgage rates are used.
The theoretical background describes how Rabobank has traditionally been a price leader in the Dutch mortgage market. Mortgage rates set by mortgage provider Obvion, a subsidiary of Rabobank, are used as a proxy for Rabobank. To account for potential cointegration between the mortgage rates set by the price leader and the price followers an ECM is used (Stock and Watson, 2012). De Haan and Sterken (2011) and Toolsema and Jacobs (2007) use an ECM to test a long-run relationship between the mortgage rates set by banks and the funding costs. The model in this thesis assumes a long-run relationship between the mortgage rates set by the price leader and the mortgage rates set by the price followers, which results in the following long-run equation:
𝑟𝑖,𝑡 = α0+ 𝛽1𝑟𝑂𝑏,𝑡+ 𝛽2 𝑟𝑂𝑏,𝑡 𝑃𝐿𝐵 𝐷𝑢𝑚𝑚𝑦 + 𝛽3𝑃𝐿𝐵 𝐷𝑢𝑚𝑚𝑦 + 𝛽4𝐹𝐶𝑡 (1)
+ 𝛽5 𝑜𝑡ℎ𝑒𝑟 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝐹𝐸𝑡 + 𝜀𝑡
Where the dependent variable 𝑟𝑖,𝑡 is the average mortgage rate set by one of the price following banks
in a specific month t. 𝑟𝑂𝑏,𝑡 is the average interest rate from Obvion in the same month. To see whether the mortgage rates became more dependent on the rates set by Obvion, the variables 𝑟𝑂𝑏,𝑡 𝑃𝐿𝐵 𝐷𝑢𝑚𝑚𝑦 and 𝑃𝐿𝐵 𝐷𝑢𝑚𝑚𝑦 are included in the regression, which cover the time periods as depicted in table I. Subsequently, it is tested if the mortgage rates are stationary after first differencing
24
Recommended approach according to Engle and Granger (1987). De Haan and Sterken (2006) also use the Dickey-Fuller test.
15 (Stock and Watson, 2012). The variable of interest and the control variables are discussed in the following section. To account for unit root and to eliminate the stochastic trend, first differences are computed (Stock and Watson, 2012). This results in an ECM model that looks into the short-run dynamics between the mortgage rates set by different banks:
𝛥𝑟𝑖,𝑡 = 𝜃1𝛥𝑟𝑂𝑏,𝑡+ 𝜃2𝛥 𝑟𝑂𝑏,𝑡 𝑃𝐿𝐵 𝐷𝑢𝑚𝑚𝑦 + 𝜃3𝑃𝐿𝐵 𝐷𝑢𝑚𝑚𝑦 + 𝜃4𝛥𝐹𝐶𝑡 (2)
+ 𝜃5 𝛥𝑜𝑡ℎ𝑒𝑟 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝐹𝐸𝑡 + 𝜆𝜀𝑡−1+ 𝑉𝑡
The term 𝜆𝜀𝑡−1 is the error correction term from the cointegrating equation. Since a short-run shift
from the long-run relationship is supposed to be corrected, 𝜆 measures the speed of adjustment and should be negative and between 0 and -1 (Stock and Watson, 2012).
After testing the time series data for non-stationary and cointegration, this thesis also investigates the impact of the price leadership bans using the individual mortgage rates. This increases the number of observations and the usefulness of control variables regarding the specific mortgage characteristics. The assumption is made that the test results of the time series also apply to the individual mortgage data. In that case the dependent variable 𝑟𝑖,𝑡 in equation (1) is the individual mortgage rate of an individual mortgage set by one of the price following banks in a specific month t. Considering that the data is on the individual mortgage level and most mortgage rates are only renewed after multiple years, it is not possible to simply take the lag of the dependent variable (𝑟𝑖,𝑡−1) and the lag of the control variables. Therefore, the average mortgage rate and average values of the control variables in the previous month are taken.
4.2 Variable of interest and control variables
The hypothesis is that the price leadership bans resulted in Obvion, as a subsidiary and proxy for Rabobank, to shift from being a barometric price leader to a collusive price leader in the Dutch mortgage market. The coefficient of the interaction term is therefore the main variable of interest and expected to be significantly positive; indicating that the mortgage rate set by the price leader is more closely followed by the price followers after the price leadership bans were imposed. A significant positive value of 𝛽2 supports the hypothesis that the price leadership bans hindered competition in the Dutch mortgage market.
𝐹𝐶𝑡 in equations 1 and 2 includes control variables that are related to funding costs. As
described by Overvest and Tezel (2014), the 2011 NMa report accounts for the funding costs in the Dutch mortgage market by looking at the 10 year senior CDS spreads, 6 month Euribor rates, 10 year swap rates and the deposit rate (with a cancellation period of three months). Mulder and Lengton (2011) also include funding costs based on Euribor rates, swap rates, interest and deposit rates when looking at the effect of the price leadership bans on the level of competition in the Dutch banking
16 sector. The 10 year Dutch government bond rate is also included in the equation to control for funding costs, since Toolsema and Jacobs (2007) use this measure as a cost proxy for mortgages.25 The other control variables include control variables regarding the characteristics of the individual mortgage applicant. These control variables are the loan to value (LTV) and debt to income (DTI) in the time series analysis. When individual mortgage rate observations are used in the analysis, the age of the borrower and the region where the house is located26 are added as control variables. FEt and 𝜀𝑡 in equation 1 are respectively year fixed effects, and an error term.
Since there is a debate on the exact timing of the possible impact of the price leadership bans on the competitiveness and pricing behavior in the Dutch mortgage sector (Dijkstra, Randag and Schinkel, 2014; Overvest and Tezel, 2014; De Kok, 2015), the variable PLB Dummy is set with starting period May 2009, to investigate the announcement effect, and the official dates at which the price leadership bans were imposed (as reported in table I). Since SNS Bank and Achmea were not restricted by price leadership bans, their PLB Dummy covers the same time period as ING. This makes it possible to analyze whether these non-restricted banks also changed their pricing behavior on the mortgage market during the time period of the price leadership bans. The contribution of this paper is to look at the difference in price setting behavior in the Dutch mortgage market before and after the price leadership bans were imposed by the European Commission. The research is unique in that it uses individual mortgage, thereby controlling for specific mortgage applicant characteristics and funding costs.
5
Data
In this section the used data is discussed. First, the different data sources are described. Second, the construction of the dataset is described. Lastly, the descriptive statistics are presented.
5.1 Data sources
The dataset in this thesis differs from datasets used in previous research since it consists of monthly data obtained from the Dutch Securitisation Association on individual mortgages from specific banks. Previous research focuses on daily or monthly data for specific banks27 or the banking sector as a whole.28 No earlier research seems to make use of similar data in addressing the impact of the price leadership bans. Four data sources are used in this thesis: The Dutch Securitisation Association, De
25
The NMa (2011) and the associated study by Mulder and Lengton (2011) also use the RMBS spread to control for funding costs obtained from the Association for Financial Markets in Europe (AFME). Since this data source is not accessible, the RMBS spreads are not taken into account as funding costs in this thesis.
26 Based on the first two numbers of the postal code five regions are separated: <21, 21-40, 41-60, 61-80 and 81 – 99. 27
i.e. De Haan and Sterken (2011), Mulder and Lengton (2011) and Mulder (2014). 28 i.e. Toolsema and Jacobs (2007) and Mulder and Lengton (2011).
17 Nederlandsche Bank, Bloomberg and Thomson Datastream. Table III provides an overview of the data sources utilized in this study and what is obtained from them.
Table III
Overview of data sources
Variable Source Period Frequency
Mortgage rates & characteristics The Dutch Securitisation Association 2006 - 2014 Monthly 10 year senior CDS spreads Bloomberg & Thomson Datastream 2006 - 2014 Monthly 10 year Dutch gov. bond rate De Nederlandsche Bank 2006 - 2014 Monthly
Swap rates De Nederlandsche Bank 2006 - 2014 Monthly
Deposit rate De Nederlandsche Bank 2006 - 2014 Monthly
6 month Euribor rate De Nederlandsche Bank 2006 - 2014 Monthly
This table states this study’s key data sources and what is extracted from them. All variables cover the period 2006 – 2014 on a monthly basis.
10 year CDS spreads for ING, AEGON, SNS Bank and Rabobank are obtained from Bloomberg. 10 year CDS spreads for SNS Bank are only partly available over the time period 2006 - 2014. Therefore, the CDS spreads for SNS Bank from 2005 until May 2009 are obtained from Thomson Datastream and this data is augmented with data obtained from Bloomberg for the time period June 2009 - September 2014. For ABN Amro and Achmea it is not possible to obtain 10 year CDS spreads. Therefore, the average 10 year CDS spreads of ING, AEGON, SNS Bank and Rabobank are used as a proxy for the default risk of ABN Amro and Achmea. The 10 year Dutch government bond rate (table 1.2.1, series 3.1), 10 year swap rates (table 1.3.1), deposit rates with a cancellation period of three months (table 5.2.7, series 2.2.0.1.0) and 6 month Euribor rates (table 1.2.1, series 2.1.5) are obtained from De Nederlandsche Bank.
The thesis is written at the CPB Netherlands Bureau for Economics Policy Analysis, which provides access to The Dutch Securitisation Association. The Dutch Securitisation Association provides RMBS data from ABN Amro, Achmea, AEGON, DBV, Delta Lloyd, Ember, ING, Obvion and SNS. Obvion was established in 2002 as a joint venture between Stichting Pensioenfonds ABP (ABP) and Rabobank. Until 2012, Rabobank had a stake of 70% and ABP’s stake accounted for 30%. In March 2012, Rabobank announced to acquire the other 30% in Obvion and the transaction was completed in the mid-2012.29 Obvion is a subsidiary of Rabobank and serves as a proxy for Rabobank as explained in the theory. Furthermore, there is no data available for specific mortgages from Rabobank in The Dutch Securitisation Association RMBS database.
To account for a base cost price of mortgages, Euribor and swap rates are used. As in the NMa (2011) report, 10 year senior CDS spreads are used to account for the risk profile of the different banks. Toolsema and Jacobs (2007) look into the relationship between the mortgage rates and capital
18 market rate in the Netherlands and use the 10 year Dutch government bond rate for the capital market rate. The 10 year Dutch government bond rate is therefore also included to account for funding costs.
5.2 Dataset construction
Since not all initial observations have the necessary data available for the variables that are needed in the further analysis, several adjustments are made.30 All values are excluded for which no information on the date of original loan advance is available. Mortgage maturity is maximized at 30 years, thereby excluding observations that have a longer mortgage maturity.31 The observations that do not have information about the current interest rate are excluded from the sample since this is the dependent variable in the regressions. The assumption is made that negative interest rates are mistakes in the dataset and mortgage rates below 0.10% are also excluded. Observations with no data about the interval at which the interest rate is adjusted are also removed from the dataset, because this variable is needed to construct the date that belongs to the mortgage rate.
Furthermore, a NHG dummy is created for all observations that include a guarantee provider whose name is related to “Nationale Hypotheek Garantie”. Mortgages with NHG coverage are insured against default risk and therefore have relatively similar risks (Hassink and Leuvensteijn, 2003). Research from the NMa (2011) shows that due to switching costs mortgage providers apply lower rates to newly issued mortgages than to mortgages whose fixed rate period is ended. To account for this price differentiation, a dummy is generated to separate the two kinds of mortgages. Bank dummies are also created for each individual bank in the sample. The dataset includes the current mortgage rate (%), but does not include the date at which the mortgage rate was set. The date that belongs to the mortgage rates is calculated by subtracting the interval in months at which the interest rate is adjusted from the future date at which the next mortgage rate change is planned. This results in the date that belongs to the specific current mortgage rate (%) and this date is used in further analysis throughout this thesis.
Observations with birth years of the primary borrower are dropped if missing and all birth years before 1920 and after 2000 are excluded.32 Observations that have missing LTV data are dropped out of the dataset. To correct for outliers, LTVs below 7.5% or above 200% are left out of the dataset. Additionally, observations are dropped if the postal code is missing or zero. This leaves 883,930 observations in the total dataset.33 Finally, the datasets containing the funding costs are merged with the dataset containing the individual mortgages from the Dutch Securitisation
30 A full overview of the variables is available in the ECB Loan Level Template.
31 To account for small deviations and starting dates of mortgages, all mortgages with a maturity of 362 months (AR61 is measured in months) or less are included in the dataset.
32 The assumption is made that mortgage owners are at least 16 years old.
33 Since mortgages consist of multiple parts, it is important to keep in mind that the number of observations is not equal to the number of different mortgages. The reason that mortgages are sometimes split up in different parts is to account for differences in characteristics like maturity, mortgage type and interest type payments.
19 Association.34
Campbell and Dietrich (1983) expect a positive relationship between the LTV ratio and the default probability. A higher default probability is likely linked to higher mortgage rates and therefore a positive relationship between the LTV ratio and the mortgage rates is expected. A higher DTI ratio increases the future probability of default (Demyanyk and Hemert, 2011) on the mortgage and as a consequence a positive relationship between the DTI ratio and the mortgage rate is expected. To calculate the average mortgage costs, Mulder and Lengton (2011) use swap rates, Euribor rates, interest rates and rates on RMBS. To compensate for higher funding costs, banks ask higher mortgage rates and Mulder and Lengton (2011) expect that the funding costs are positively related to the mortgage rates. No specific relationship is expected between the age of the applicant or the location property and the mortgage rates. Table IV provides an overview, description, specification and the expected sign of the covariates that are used in the methodology.
Table IV Overview of covariates
Variable Description Specification
Expected sign 𝑟𝑂𝑏,𝑡 Average mortgage rate Obvion Continuous + 𝑟𝑂𝑏,𝑡× 𝑃𝐿𝐵 𝐷𝑢𝑚𝑚𝑦 Average mortgage rate Obvion × PLB Dummy Continuous +
LTV Loan to value ratio Continuous +
DTI Debt to income ratio Continuous +
Age Age of the mortgage applicant Continuous +/- Location property Location of the property Dummy +/- CDS spreads 10 year senior CDS spreads EUR Continuous + 10 year gov. bond rate Newest 10 year Dutch government bond rate Continuous + Base cost mortgages 10 year swap rates Continuous + Funding cost deposits 𝑟𝑑𝑒𝑝𝑜𝑠𝑖𝑡 Continuous + Euribor 6 month Euribor rates Continuous +
This table presents the main model’s covariates. The description of all covariates is provided in the second column. In the last column the expected sign is stated.
34
The datasets are merged on the monthly date corresponding with the costs and the monthly date corresponding with the mortgage rates.
20 5.3 Descriptive statistics
Table V provides an overview of the number of observations per bank and the number of observations per fixed rate period for the total sample.
Table V
Number of observations at the bank level and number of observations per fixed rate period
Bank N Percentage Cumulative Fixed rate period N Percentage Cumulative ABN Amro 503,464 57.0% 57.0% 0 - 1 year 2,893 0.3% 0.3% Obvion 193,783 21.9% 78.9% 1 - 5 years 144,709 16.4% 16.7% SNS Bank 68,290 7.7% 86.6% 5 - 10 years 44,087 5.0% 21.7% Achmea 47,946 5.4% 92.0% 10 years 413,369 46.7% 68.4% ING 41,984 4.7% 96.8% >10 years 278,872 31.6% 100.0% AEGON 28,463 3.2% 100% Total 883,930 100% 100% Total 883,930 100% 100%
The left side of the table reports the number of observations for each bank in the total sample. The right side of the table presents the number of observations for the different fixed rate periods. Percentages and cumulative percentages are also presented.
As shown in table V, 57% of the mortgage rates in the dataset are from ABN Amro, and Obvion accounts for approximately one fifth of the total sample. This seems to indicate that ABN Amro securitizes their mortgages more than the other banks in the sample. Even though ING and AEGON account respectively only for 4.9% and 3.0% of the dataset, their number of observations is still high. 46.7% of the mortgages in the total sample have a fixed rate period of 10 years. Only 0.3% of the mortgages have an interest reset interval between 0 and 12 months.Tables 1 and 2 in the appendix provide a more extensive overview of the total sample.
Based on this total sample, the monthly average mortgage rates are created to perform the Granger causality Wald tests. As described in the next section, the final subsample that is used to investigate the price setting relationship between the price leader and the price followers consists of 168,308 newly issued mortgages with a fixed rate period of 10 years and NHG coverage. Table VI provides the summary statistics of the main variables of this subsample. As shown in the table, the mean of the mortgage rate in the sample is 4.77% with a standard deviation of 0.44%. The average LTV is 93.44% with a standard deviation of 19.13%. The average DTI ratio is 4.26 with a maximum of 25.44. A possible explanation for this high maximum is that some mortgage applicants have a low income, but get a mortgage because they are wealthy. The cost variables are obtained on a monthly basis and therefore their number of observations differs from that of the other control variables.
21 Table VI
Descriptive statistics
Variable Mean Median S.D. Min Max N
Mortgage rate (%) 4.77 4.75 0.44 1.00 6.95 168,380 LTV (%) 93.44 100.08 19.13 7.60 199.66 168,380 DTI (ratio) 4.26 4.30 1.26 0.50 25.44 148,603 Age 41.78 40.00 9.84 22.00 96.00 168,380 Euribor (%) 1.96 1.35 1.47 0.20 5.22 103 Gov. bond rate (10y) 3.11 3.32 0.88 1.15 4.73 103 Deposit rate (%) 2.32 2.28 0.38 1.25 3.19 103 Swap rate (10y) 3.17 3.33 0.96 1.09 5.05 103
This table presents the mean, median, standard deviation, min, max and number of observations for the main variables that are used in the regressions. The cost variables are obtained on a monthly basis, which explains the different amount of observations compared to the other control variables. The dataset covers the time period between March 2006 and September 2014.
Table VII provides an overview of the number of observations of each bank in the subsample. ABN Amro has 116,891 mortgage rate observations, thereby accounting for approximately two third of the sample. Although ING mortgages only account for 1.8% of the sample, the number of observations is 3,102, which is enough to perform the analyses in the next section.
Table VII
Number of observations at the bank level
Bank N Percentage Cumulative
ABN Amro 116,891 69.5% 69.5% Obvion 30,659 18.2% 87.7% SNS Bank 13,673 8.1% 95.8% Achmea 3,983 2.4% 98.2% ING 3,102 1.8% 100% Total 168,308 100% 100%
This table provides the number of observations at the bank level for the subsample that is used in the long-run relationship and the ECM. Out of the 168,308 mortgages 69.5% belong to ABN Amro. Obvion accounts for 18.2% of the mortgages. AEGON is dropped out of this sample, since data was missing on multiple consecutive months.
Figure 2 displays the average monthly mortgage rates of Obvion, ING, SNS Bank, ABN Amro and Achmea for newly issued mortgages with a fixed rate period of 10 years and NHG coverage. The first price leadership ban in the Netherlands was imposed on ING in November 2009. Dialogues about the bans already started in the spring of 2009. The left vertical line therefore presents May 2009 and the right vertical line presents November 2009. From this graphical display, it is not clear to see whether the interest set by the price leader played a more important role in the interest rate set by the price followers after the price leadership bans were imposed. A thorough analysis executed in the next section provides an answer to the hypothesis.
22 Figure II
Average monthly mortgage rates and the Dutch government bond rate
This figure shows the average monthly mortgage rates of Obvion, ING, SNS Bank, ABN Amro and Achmea. The 10 year Dutch government bond rate is also included. The left vertical line corresponds to the date of the potential announcement effect of the price leadership bans, which is May 2009. The right vertical line corresponds to the date of the first price leadership ban that was imposed in the Netherlands (on ING in November 2009).
6
Empirical results
In this section, the results of the empirical analysis are presented. First, the main results are described and then the robustness checks are provided.
6.1 Results
First, the outcome of the Granger causality Wald test is discussed. Second, the time series results of the long-run relationship and the ECM are presented. Thereafter, individual observations are used to look at the impact of the price leadership bans on the pricing behavior of the price followers. Finally, the results of the Chow break test are provided.
