1
Hold-Up Problem in Commercial Real Estate Finance
Are the fruits of information accumulation shared equally between
lender and borrower?
Master thesis Jasper Whitmore 0415812
Supervisor: Dr F.J. de Graaf
MSc Business Studies, specialisation: Governance & Finance University of Amsterdam, Amsterdam Business School Faculty of Economics and Business
2
Abstract
This thesis adds to the existing literature regarding relationship lending by investigating the extent and effect of information accumulation and the presence of the Hold-Up Problem within a loan database from a leading commercial real estate lender in the Netherlands. The contribution to the existing theory is largely generated by the character of the data, as it features 100% collateralised loans. Furthermore, the lender puts a lot of effort into learning about the borrower and the collateral. Besides data on the collateral and borrower, this thesis investigates price and non-price aspects of loan terms. Surprisingly, no significant relationship is established between duration and information accumulation, contrary to a key assumption that is prevalent throughout the existing body of literature. The effects of the Hold-Up Problem are also found to be insignificant, which is consistent with the previous finding as the Hold-Up Problem theory is itself based on the effects of information accumulation.
Acknowledgements
This thesis marks the final chapter of the part-time master course in Business Studies, with a specialisation in Governance & Finance. It turned out to be a longer story than planned, with ebbs and flows that followed the movements of an extracurricular body – my life. Before embarking on the final piece of work, I would like to take the opportunity thank a number of people. I would like to thank Dr Frank Jan de Graaf for his continuous support and understanding: much additional patience was required and provided. I would like to thank both my current and previous employer in supporting me in my endeavour to become a Master of Science. A special thanks to Charlotte Spanjaard, whose help has been immeasurable. I would like to thank my family and friends, who had incorruptible faith in a good outcome. I hope to provide the same in return. This study has brought me friends, led me on international adventures and has tickled my mind in many inquisitive ways. My life has been enriched and I am proud to be a University of Amsterdam alumnus.
3
Inhoudsopgave
1. INTRODUCTION 4 CONTEXT 5 PROBLEM STATEMENT 6 2. THEORETICAL FRAMEWORK 7 2.1RELATIONSHIP LENDING 72.2DOWNSIDES OF RELATIONSHIP LENDING 9
2.3HOLD-UP PROBLEM 10
3. DATA, VARIABLE DESCRIPTION AND METHODOLOGY 13
3.1DATA 16
3.2 VARIABLE DESCRIPTION AND SOURCES 17
3.2.1INDEPENDENT 17
3.2.2DEPENDENT VARIABLES 18
3.2.3CONTROL VARIABLES 19
3.3METHODOLOGY 19
4. RESULTS 22
4.1COLLATERAL VALUE VOLATILITY 23
4.2LOAN TERMS 24 4.2.1MONOPOLY RENTS 25 4.2.2AVAILABILITY OF CREDIT 27 5. LIMITATIONS 31 6. CONCLUSIONS 32 REFERENCES 36 APPENDIX 38
SECTION 1 – VARIABLE DESCRIPTIONS AND DESCRIPTIVE STATISTICS 38
SECTION 2 - REGRESSION ANALYSES OUTCOMES 39
4
1. Introduction
The past seven years have been marred by multiple financial crises. Recovery has taken place in many countries but even today there are many remnants of the severity of the sub-prime mortgage crisis of 2007, the collapse of Lehman Brothers in 2008, the Greek debt crisis of 2010, and the Euro-banking crisis of 2011. Many nationalised banks have yet to be brought back into private hands. Interest rates remain at historically low levels. Many developed economies have not surpassed their pre-crisis size. Unemployment remains stubbornly high in the Eurozone. That said, many improvements have been made to regulations of the financial system. Especially banks have seen regulatory demands increased, with higher capital requirements, to ensure sufficient capacity to endure downturns, and information duties, to provide regulators oversight into banking activities.
The cause of the turbulent period has not been pinpointed to a single factor. It is safe to assume that many aspects contributed. This thesis does not attempt to solve that riddle but may contribute to understanding one aspect better, namely information asymmetry and how soft information is treated in the banking system. Berger & Udell (2006) note that this information oftentimes is proprietary in nature, and that this soft information is difficult to transfer through hierarchies and organisation. In contrast, hard information consists of easy-to-quantify data and is presented through financial ratios. Due to the homogenous nature of this type of data, it transfers more easliy through hierarchies.
Borrowers that rely on businesses processes that generate mainly soft information can overcome this by engaging in relationship lending, whereby the lender acquires information over time and multiple interactions. Diamond (1991) presents this as a choice between an arrangement with monitoring, as is the case in relationship lending, or borrowing without monitoring, as is the case when acquiring debt through transactional lending. Through the extensive efforts of the borrower and lender to transfer soft information, the lender accumulates more information about the borrower than competitor lenders have. In his widely cited literature review, Boot (2000) lists the ability to closely monitor collateral in asset-backed lending as one of the major advantages of relationship lending. One of the main investment classes that is financed through asset-backed lending is commercial real estate. This is relevant to this thesis as the study is done on a real estate financier.
5
The reduced information asymmetry leads to a reduced risk profile. However, switching costs arise if the borrower decides to switch to another lender. Which dynamic takes the upper hand depends on many factors. The relevant literature has focused on the Hold-Up Problem, whereby the lender uses the switching costs to extract monopoly rents from the borrower, in the form of increased loan rates. Empirical research has come to opposite conclusions. For example, Degryse & Van Cayseele (2000) find a positive relationship between loan rates increase and relationship duration. Petersen & Rajan (1994) find that loan rates do not change with increased duration of the lending relationship. And finally, Berger & Udell (1995) find that loan rates decrease in relation to increasing duration of the lending relationship. Boot (2000) states that the Hold-Up Problem can be present without affecting loan terms, as there might be other factors that dominate the Hold-Up Problem.
Context
The Dutch banking system is characterised by a high amount of concentration, caused by two decades of consolidation and reorganisations in the 1980’s and 90’s (Van Leuvensteijn et al, 2011), explaining the below-average number of average bank relationships. Rajan & Petersen (1995) find that in markets where a limited number of banks provide credit, relationship lending provides a larger share of the total amount of debt. A prime example is the Dutch commercial real estate financing market, which is largely a relationship-based lending system and is characterised by a high concentration of lenders. The three largest lenders supply over 80% of the lending volume1.
The commercial real estate market is one of the sectors most affected by the recent turmoil on the financial markets. The Dutch commercial real estate market was not an exception. Over the period of 2005 – 2008, the average investment volume was €9.4 billion2. With the financial crisis taking hold in all severity, over the period 2009 – 2012 the average volume was just over half of that, averaging €4,9 billion3. Furthermore, asset prices had seen a positive development over the period 2005 – 2008 of 4-7% per annum. This turned into a
1 Sectorstudie Vastgoedfinanciering, 2011, SEO Economic Research: M. Kerste, J. Poort, P. Risseeuw, N. Rosenboom, page 29
2 Dutch Capital Markets Bulletin 2013, 2013, Jones Lang Lasalle: D. van Leeuwen, M. Hesp, S. Bertens, T. van den Noort, page 7
6
negative development over the subsequent period (2009-2012), ranging from -1% and -8% per annum.4
The financial turmoil affected the European banking system, with governments in virtually all EU countries bailing out or nationalising banks. Whilst banks do exert a level of inter-bank monitoring (Furfine, 2001), observed in differentiated inter-bank lending, the differences in margins do not represent the differences in risk profiles. This leads to volatile swings in risk perception, with a too mild perception in a good economic environment and too harsh in bad economic environment.
This dynamic contributed to a shortage of liquidity in the Dutch banking system, which was priced into interest margins charged to the borrowers through the introduction of a liquidity spread5. This is an additional component of the interest margin, aimed to compensate the lender for the heightened costs of funding. The liquidity spread is based on the development of inter-bank lending availability. The liquidity spread was introduced in 2008 at a limited 10bps but increased to a maximum of 185bps in 2010, and averaging approximately 120bps over the period 2009-20126.
Problem Statement
This thesis aims to contribute to the existing body of literature on relationship lending and a sub-theory of that field, namely the Hold-Up Problem. Data is mined from a loan database from one of the leading commercial real estate financiers of the Netherlands. By examining the theory of relationship lending on the specific characteristics of real estate financing, hopefully new insights will be uncovered. The database offers an accurate insight into the development of margins, collateral values, fundamental cash flows, relationship duration, and other variables. By investigating the extent of information accumulation in the first step, the assumption that duration equals information accumulation is tested. Secondly, the relationship between duration and price and non-price loan terms is tested against the findings of the existing literature.
Therefore, the research question is:
4 Overview of Financial Stability, 2012, no. 16, De Nederlandsche Bank, Page 23 5 Idem, page 10
7
Does increased relationship duration lead to a continuous information accumulation and affect the loan terms?
To assist in answering this research question, the following sub-questions are posed:
(i) Does a longer relationship-duration lead to a lower volatility in change of collateral value? (ii) Are monopoly rents extracted by the lender from borrowers with longer relationship duration?
(iii) Do relationship lenders provide a higher availability of credit to borrowers with longer relationship duration?
The remainder of this thesis is organised as follows. Section 2 will provide an account of the main aspects of relationship lending, and discuss the body of literature on the Hold-Up Problem and the research gap that exists. Throughout the section, hypotheses are articulated. Section 3 describes the data, the variables and methodology. The results of the study are presented in section 4, including a discussion on the findings. The limitations of this study are discussed in section 5. And finally in section 6 the conclusions will be discussed.
2. Theoretical Framework
2.1 Relationship Lending
Berger & Udell (2006) state that relationship lending technologies are “based on soft qualitative information gathered through contact over time”. Boot & Thakor (2000) describe how engaging in relationship lending can add value to the borrower, with banks needing to develop costly expertise to be able to provide this type of service. The authors call this expertise “sector specialization”. Boot (2000) defines relationship banking as: “(i) investing in obtaining customer-specific information, often proprietary in nature; and (ii) evaluating the profitability of these investments through multiple interactions with the same customer over time and/or across products” - similar to Berger & Udell’s definition. Berger (1999) offers three conditions that test whether relationship lending is present:
(i) the bank manager undertakes efforts to acquire information that is not readily and/or publicly available;
8
(iii) the information is treated confidentially.
Berger & Udell (2006) give a number of examples of what might encompass qualitative information, including “the character and reliability of the SME’s owner, based on direct contact over time by the institution’s bank manager; the payment and receipt history of the SME gathered from the past provision of loans, deposits, or other services to the SME by the institution”. With the reference to small and medium-sized enterprises (SME), the authors give a hint towards the influence of size of organisation. According to Sufi (2007), firm size and governance type are also of influence on the amount of information asymmetry. Banks endeavour to acquire this information in order to reduce the risk of moral hazard and the amount of asymmetric information through Delegated monitoring, as described by D’Auria et al (1999).
Measuring information asymmetry is difficult. In asset-backed lending all loans are collateralised by assets. Boot (2000) names the enhanced ability to monitor collateral assets as one of the advantages of relationship lending. Monitoring leads to adjustment of the perception of the assets, which might change through enhanced knowledge or market developments. In each case, the value of these assets is not static but develops over time.
In database used in this thesis, all real estate assets are physically revalued by the lender once every three years, incidentally in case of material changes and yearly based on portfolio analysis. When an asset has undergone the valuation process multiple times, the lender will have accumulated more information on the asset. With an increased duration, the lender will also have acquired more information on the borrower. The expectation is that through reduced information asymmetry, the lender can make a better prediction of the development of the value of the collateral. If information is accumulated, the volatility of changes in value should decrease. Therefore, if relationship duration equals information accumulation, there should be a negative relationship between relationship duration and volatility of collateral value.
H1: With an increased relationship duration, the volatility of value of the collateral decreases.
9
Degryse & Van Cayseele (2000) state that with an increase in duration of the relationship, a more effective process to overcome information asymmetry is established. This scope covers Boot & Thakor’s (2000) choice of relationship bank financing versus capital market financing and it is reasonable to assume that being better-informed encompasses Boot’s (2000) definition of relationship lending (obtaining customer-specific information, evaluating profitability based on information acquired through multiple interactions). Hadlock & James (2002) argue that borrowers better served by bank debt are those that rely on information that is more opaque due to the nature of the underlying projects – projects that can only be fully described through soft information. In their article, the authors draw on the financial distress literature and suggest that bank debt is provided by better-informed lenders than debt financed through public markets.
Cole (1998) researches the likelihood of credit line extensions and finds borrowers with a pre-existing relationship are more likely to be granted extension than those without. The author establishes that the information accumulation is effective and enables lenders to make a more accurate assessment of the risks. Berger & Udell (1995) find that increased duration of relationships result in lower collateral requirements. However, both articles miss one consideration, concerning the selection bias. Bad borrowers will have been forced to exit the relationship with the lender, as a good business practice of the lender. The positive relation between extension and/or collateral requirements might therefore, in part or wholly, be a result of the ejection of bad borrowers. By testing the volatility of the collateral value, an objective measure of information accumulation is researched.
2.2 Downsides of Relationship Lending
Whilst relationship lending has a positive effect on reducing information asymmetries between borrowers and lenders, Rajan & Petersen (1995) argue that it also causes illiquidity. The authors argue that due to the effort that goes into acquiring proprietary information, the financial asset becomes difficult to refinance by a challenging bank. The illiquidity becomes a risk and becomes a downward pressure on the value of the financial asset. It is the expectation that real estate, which is one of the most illiquid of investment classes, suffers from this dynamic.
As a result of the illiquidity, the relationship lender sees its bargaining power increase and “the financier attempts to secure her return on investment by retaining some kind of
10
monopoly over the firm she finances” (Rajan & Zingales, 2001). However, the relationship lender is also forced to continue to provide finance even when the borrower is in financial distress. Rajan & Zingales (2001) describe this as the practice of charging “above-market rates in normal circumstances in return for an implicit agreement to provide below-market financing when their borrowers get into trouble”. However, underwriting standards change throughout the credit cycle and, according to DELL’ARICCIA et al (2012), this is the result from both the credit supply (bank) and demand (borrower) side. The demand/supply balance at the time of initial loan disbursement does not have to be the same as at the moment of an extension. Any change in demand/supply equilibrium will have a material impact on loan terms.
The potential downsides of relationship lending are recognised as being that of the Hold-Up Problem and the Soft Budget Constraint (Boot A. W., 2000). The HDP is a theory that arose from literature focussing on optimal contracts vs incomplete contracts (Schmidt & Noldeke, 1995). The premise of that literature is the non-verifiability of every eventuality in a contract. This leads to inevitability of renegotiations and the accompanying assumption of both contracting parties of the possibility to renegotiate. Rajan (1992) articulates this dynamic eloquently by stating that “while informed banks make flexible financial decisions which prevent a firm’s projects from going awry, the cost of this credit is that banks have bargaining power over firm’s profits, once projects have begun”.
2.3 Hold-Up Problem
Through all the efforts of the lender and borrower to learn from each other a closer relationship emerges, enabling both parties to provide service to each other. However, as a result of information accumulation, a borrower looking to change to another relationship lender will have to exert effort into transferring the information to the new relationship lender (Barone, Felicic, & Pagnini, 2011) and will face relatively high switching costs. Elsas (2005) makes the connection between private information and switching costs as well, stating that with “private information accumulation over time, the lock-in of a borrower should increase with duration. Duration then reflects switching costs, the severity of the HDP, and relationship intensity in general”. According to Shy (2002), switching costs arise from the physical and human capital that a consumer invests upon purchasing. In
11
relationship lending, this translates into the financial capital and pledged collateral as physical capital and in-depth knowledge of the collateral as human capital.
Ongena & Smith (2002), Rajan (1992) and Sharpe (1990) argue that the increased switching costs, as a result of the accumulated information, enable banks to extract monopoly rents through modification of the loan terms. This could occur at renewal of existing loans or when disbursing new loans. The body of literature on HDP has largely focused on these monopoly rents by empirically researching the development of loan rates. If HDP exists, loan rates should increase over time.
The counter-argument here is that the accumulation of (proprietary) information leads to reduced information asymmetry. A more accurate assessment of the borrower’s risk profile leads to lower risk profile exists and the lender can pass on (part of) the monopoly rents to the borrower (Berger & Udell, 1995).
The findings of these studies have largely focused on establishing a relationship between duration of the lending relationship and monopoly rents. The findings have been contradictory, which Boot (2000) expects is because the HDP can be dominated by other factors. Boot outlines four factors that might come into play, three internal to the relationship and one external: (i) availability of credit, which sees a positive correlation to relationship duration in empirical research (Peteresen & Rajan, 1994; Berger & Udell, 1995). (ii) Lenders engage in marketing tactics towards young firms, offering generous loan terms to assist these firms in their infancy and with the aim of recouping lost revenues at a later stage after accumulating additional information (Petersen & Rajan, 1994; Petersen & Rajan, 1995). (iii) Boot (2000) takes from a working paper of Kracaw and Zenner (1998) which researches the effects of bankers having directorship in the lender’s organisation. In their working paper, the authors find that the presence of these directorships increase the HDP. (iv) Finally, the amount of competition between sources of credit affects the degree in which switching costs develop and thereby influences the severity of the HDP (Degryse and Cayseele, 2000).
This thesis hypothesizes that monopoly rents are passed on to the borrower. H2a follows the existing literature and investigates the margin. The quality of the available data allows for an accurate measurement of net margin, as opposed to earlier research that did not have access to similar data and had to make corrections to arrive at a proxy for net margin. The
12
net margin is determined at the discretion of the bank manager, whose considerations are largely in line with the factors that are outlined by Boot (2000).
H2a: With an increased relationship duration, the net margin decreases.
H2b investigates the share the lender extracts from the free cash flow generated by the collateralised assets. This is a more nuanced perspective on monopoly rents as it considers the available opportunity to extract rents. If monopoly rents are passed on to the borrower, the lender will take a smaller share of free cash flow. This has not been researched in previous literature but can be measured due to the nature of the database.
The real estate assets generate income through rental contracts, which offer predictable cash flow. Part of the lender’s valuation process is to establish a fundamental cost structure for each real estate asset. After deduction of those costs and interest payments, cash flow is available for (a) regular repayments and (b) dividend payments. The regular repayment as share of the available free cash flow then excludes effects of the interest margin and the amount of credit vis-à-vis collateral value. It is assumed that this is an accurate variable to illustrate the ability and desire of the lender to claim monopoly rents.
H2b: With an increased relationship duration, a lower share of the cash flow after operating costs and interest costs is paid as repayment.
Degryse & Van Cayseele (2000) include the availability of credit in their research, using the repayment duration as a proxy. This concerns the duration needed to repay the entire loan based on the repayment schedule. Oftentimes, this duration exceeds the contractual loan term and thereby hints at an implicit extension agreement. Whereas the regular repayment as share of the available free cash flow illustrates dynamics of monopoly rents, the repayment duration researches the availability of credit. This thesis follows Degryse & Van Cayseele (2000) in investigating availability of credit and uses this convincing variable.
H3a: With an increased relationship duration, the loan features a longer repayment duration.
13
Using a loan database that is fully collateralised against real estate offers the opportunity to quantify the availability of credit against the value of the collateral. The so-called Loan-to-Value ratio (‘LtV’) is a key risk indicator of commercial real estate financing and collateral features prominently in determining the other loan terms. This hard information ratio allows observation of differences in availability of credit. With greater insight into the risks of the borrower and the collateral, this thesis hypothesizes that long-standing relationships offer greater availability to credit than short-standing relationships. This is consistent with the findings of Petersen & Rajan (1994), Berger & Udell (1995) and Degryse & Van Cayseele (2000). The assumption is made that there is no difference in desired amount of leverage between clients with long-standing and short-standing relationships. Based on this assumption and the findings of Petersen & Rajan (1994), Berger & Udell (1995) and Degryse & Van Cayseele (2000), this thesis hypothesises that the average LtV is higher in long-standing relationships than in short-long-standing relationships.
H3b: With an increased relationship duration, the loan features a higher Loan-to-Value ratio.
3. Data, Variable Description and Methodology
Petersen & Rajan (1994), Burger & Udell (1995) and Degryse & Van Cayseele (2000) refer to two strands of empirical research into relationship lending, based on direct and indirect measures of the strength of the relationship. The authors offer an example of the indirect method, namely what the effect of a banking relationship is on the share price of the borrower. In Boot’s (2000) review of the literature on Relationship lending, he states that although this indirect method shows that the existence of the relationship has a positive effect on the value of the company, the sources for this added value are not uncovered by this method.
The aforementioned authors opt for the direct measure, with the independent variable being the duration of the relationship. Boot (2000) also confirms that this is the typical continuous measure used in the body of literature. This thesis follows the direct method. Data is mined from a loan database from a commercial real estate financing arm of a Dutch bank, which deploys a relationship-based lending business model. The Dutch commercial
14
real estate financing market is characterised by its high concentration of providers of credit (banks).
This thesis follows the research methodology of Degryse & Van Cayseele (2000). In their research, the authors focus on very small firms operating in a financial system in which banks are the main source of debt. 81% of the dataset consists of single-person businesses and 17% of small businesses. Besides the character of the data, Degryse & Van Cayseele (2000) also investigate the association between relationship lending and collateral requirements. Finally, the authors investigate whether borrowers that make use of additional information sensitive products are subjected to the same price development as those that only make use of a loan facility.
Before outlining the differences in methodology used in this thesis vis-à-vis that of Degryse & Van Cayseele (2000), an assessment of the dataset used in their research follows. This is important, as there is a large homogeneity in their dataset that might make the data analysis susceptible to the dominating factors that were mentioned by Boot (2000). Specifically, firms of such small nature rely on bank managers for strategic advice (Ian Burke & Jarratt, 2004), which puts the bank manager in a similar position as a director in the company. As Kracaw & Zenner (1998) note, the presence of a bank on the board of directors increases the intensity of the HDP. As for the other three factors that might dominate the HDP as mentioned by Boot (2000): Availability of credit; Marketing to young firms; and Competition between lenders – the fundamental characteristics of the dataset do not offer any grounds to make additional assessments.
One other element of the dataset that requires highlighting is the proxy used for margin. Degryse & Van Cayseele (2000) do not have access to the net margin. The dataset is based on the total interest paid by the borrower, including the cost of funding the loan and other elements that are indiscriminate to the borrower lending-profile. The authors correct for this by comparing the loan duration with a government bond with the same maturity. The reasoning behind this correction is to extract the risk-free rate (government bond) and isolate the cost of capital that is fundamental to the borrower. However, there might be other elements incorporated in the spread that are not linked to the borrower profile. An example thereof is the earlier mentioned liquidity spread.
The dataset used in this thesis includes the net margin. That is, the component of the interest charge that is determined by the bank manager on the basis of the risk and
15
commercial considerations. In other words, the net margin should be a representation of the risk/return proposition. It excludes the cost of funding and any other component of the interest costs. This allows for an analysis with fewer corrections than that of Degryse & Van Cayseele (2000), which should contribute to the accuracy of the analysis.
As a consequence of the nature of the underlying business of the database used in this thesis, 100% of the loans feature collateral. This offers potential for new research avenues. The so-called Loan-to-Value (‘LtV’) is a key measure in determining the amount of credit that a borrower can take on. With a fully collateralised loan portfolio, this clear measure of availability of credit emerges can be analysed on a continuous scale as opposed to a binary scale (collateralised/not collateralised) used by Degryse & Van Cayseele (2000). Petersen & Rajan (1994) include a debt-to-asset ratio in their research, a comparable variable that might be less robust as the assets include non-real assets (for example goodwill). Degryse & Van Cayseele (2000) use a different variable for availability of credit, being the duration of repayment scheme. This thesis also includes this analysis.
A second additional avenue to the research method is a new variable to measure the amount of monopoly rents that the lender might extract, as a consequence of the HDP. The collateralised real estate assets produce a stable cash flow arising from rental contracts. The database includes fundamental cost-structures of each real estate asset. After subtracting from the rental income the cost of operating and the total interest costs, an amount of cash flow is left for repayment of the bank loan and dividend payment to investor. If the lender aims to extract monopoly rents, the share of the repayments will be higher in borrowers with longer relationship duration.
Finally, this thesis does not include a variable that makes a distinction between borrowers that make use of additional information sensitive products and those who do not. Whereas the dataset of Degryse & Van Cayseele (2000) is based on a loan database of small businesses, the data used in this thesis is mined from a loan database from a homogeneous business-lending platform. The real estate financing business unit collects specialist information on borrowers, which is reflected, for example, in the fundamental cost-structure of each real estate asset.
16
3.1 Data
The data for this study is mined from the loan database of a leading Dutch commercial real estate financier. This lender operates in two markets, the institutional/international environment and non-institutional/domestic market. In the non-institutional/domestic market, a relationship-based business model is deployed. Data is taken exclusively from this market.
Two datasets are constructed. One aims to follow the research methodology of Degryse & Van Cayseele (2000). Data is extracted from the database as of 30th of September 2014. The second dataset follows the development of collateral value. Data is extracted on the 30th of
September 2010 and measured again on September 30, 2014. Please refer to Table I in the appendix for an overview of descriptive statistics and definitions.
The first dataset consists of a total of 3,362 loans and 1,204 borrowers. All loans started after July 23, 1993. Borrowers are defined as Legal Ultimate Parent (‘LUP’). These entities are not necessarily the Main Loan Partner (‘MLP’) counterparty for the lender but are a controlling owner (>50% of voting shares) of the MLP. It is the assumption that the effects of duration of relationship are based on the ultimate controlling party. Additionally, the annual credit revision is executed at LUP level, as is the three-yearly revaluation cycle. A common feature in the client database is the ‘combi’, a joint-venture based on 50/50 split ownership. As there is no controlling owner, this entity is administered as a separate LUP, which does not reflect the duration of relationship. These are therefore excluded from the database.
A second correction is done on the lower boundary of the size of LUP. LUPs with a smaller outstanding amount than €1,000,000 have been excluded. The commercial real estate lender from which the data has been taken has a specialised portfolio for owner-occupiers, which features different loan terms to those of owners who seek returns from their real estate investments.
37% are BV’s (limited liability partnerships), 15% are foundations (‘stichting’), 3% concern private individuals or legal entity with personal liability and 8% are CV’s (non-recourse financing). A catch all group covers the rest and includes not assigned data points and less commonly used forms (mainly foreign entities). Approximately 65% of LUPs have an outstanding loan amount between €1 mln and €5mln, 22% have an outstanding loan amount between €5mln and €15mln, 11% have an outstanding loan amount between €15mln and €50mln, and 2% have an outstanding loan amount above €50mln.
17
For the second data set on collateral values, a total of 978 assets are included. Only assets in the office and retail asset classes are included as these concern the core business of the lender. Secondly, only assets that were present from September 30, 2010 to September 30, 2014 are included. All collateral assets had a base value on September 30, 2010. Between the start date September 30, 2014 each asset underwent at least one physical revaluation and at least one other valuation data point. The lender adheres to a three-yearly valuation cycle, which is central to the valuation policy of the lender. Another part of the valuation cycle is an incidental valuation as response to a material change to the asset. Material changes can be negative, for example a bankruptcy of a tenant, or positive, for example a large capital expenditure done by the borrower. All assets were also subjected to revaluation based on portfolio analysis, in which general market trends are applied. This type of revaluation excludes soft information but affects all values equally.
3.2 Variable Description and Sources 3.2.1 Independent
Relationship duration is the term of the existence of the relationship since the first loan contract until yearend 2014. The mean value for relationship duration is 7.2 years, comparable to the 7.8 years value reported in Degryse & Van Cayseele’s (2000) research. A standard deviation of approx. 4.5 is calculated, offering a wide range in the dataset, which is conducive to researching the relative impact of relationship duration on dependent variables.
The relationship duration is also included in the second data set, which is constructed for the analysis of the effect of information accumulation on the volatility of the market value. A difference in mean value (9.7 years) and standard deviation (3.88 years) is recorded in this dataset, hinting at a selection bias towards long-standing borrowers as a result of the criteria for selection of collateral assets. As the main criteria are based on the asset being present both on September 30, 2010 and September 30, 2014, an effect on the duration does not come as a surprise.
To combat part of this selection bias, the object duration is included. This concerns the term from the first time the lender appraised the real estate object until yearend 2014. An advantage of the dataset is that objects have been sold from one borrower to another and
18
that the object duration can be longer than the relationship duration. Whilst the relationship duration should equal the information accumulation, it stands to reason that the specific knowledge of the real estate object forms a large part of the information accumulated. The mean value for object duration is 7.4 years with a standard deviation of 2.13 years.
3.2.2 Dependent variables
The first dependent variable used to examine the dynamics of monopoly rents is NET_MARGIN. The net margin excludes any component of the total interest costs that is not determined at the discretion of the bank manager. What is left is the margin that the bank manager deems acceptable vis-a-vis the underlying risk profile combined with the commercial importance of the borrower. A mean value of approx. 2.11% is calculated for the net margin, with a standard deviation of 0.51%.
The second variable used for monopoly rents is Repay_share_FCF. Total interest costs are calculated by multiplying the total interest percentage with the total outstanding amount. Operating costs are determined as part of the valuation process, and are included in the database. This variable is determined by the share that the repayment schedule takes in the cash flow that remains after subtracting operating costs and total interest costs from the rental income. With a mean value of 47.63% and a standard deviation of 71.54%, a minimum of minus 598.7% and a maximum of 686.9%, this variable shows a high degree of variance.
To research the availability of credit, this thesis follows Degryse & Van Cayseele (2000) in using Repay_DUR. This is the expression of the repayment schedule in unit of time (years). It is calculated by calculating the amount of time in which the current total outstanding credit amount will be repaid if following the repayment scheme. In the data set used in this thesis, the mean value is 43.55 years, which equates to an average repayment of 2.29% of total outstanding each year. A second variable is used, following Petersen and Rajan (1994) in adapted form. Petersen and Rajan (1994) use asset-to-debt ratio, which includes intangible assets such as goodwill and difficult to value receivables, often as determined by the owner. This thesis uses the Loan-to-Value ratio (LTV), which is determined on real assets that are valued regularly by the lender. A mean value of 74.27% is recorded, with standard deviation of 32.79%.
COLLAT_VAL_VOL. The change in value of the collateral from data point 30 September, 2010 and 30 September, 2014. All real estate assets are physically valuated the start of the
19
loan. As part of a prudent loan management, the collateral is revaluated at least every three years to ensure that the value is up to date. The mean value for this variable is 9.37% with a standard deviation of 30.67%.
3.2.3 Control Variables
SIZE. The commercial real estate financier services small and medium sized investors in the domestic market, although a number of clients have achieved institutional size (>€100mln). LUPs are divided into four brackets: €1mln - €5mln; €5mln - €15mln; €15mln - €50mln; >€50mln based on total amount of outstanding credit. Degryse & Van Cayseele (2000) use classifications extracted from the chamber of commerce. This thesis has elected to use the current outstanding credit amount, as it offers the chance to analyse on a continuous scale. Furthermore, Degryse’s & Van Cayseele’s (2000) determination of size led to a sensitivity as 81% of data points fell into one category (single person business) and a further 17% fell into a second category (small business).
GOV. Governance characteristics are followed on the categories in Degryse & Van Cayseele (2000), but adapted to the Dutch legal system based on the same amount of liability that the type of legal entity carries. In the dataset of this thesis, a number of legal entities are paired: BV & Stichting (foundation) are paired due to fact that foundations are commonly used as directors in BV’s. A total of 617 of the data points are represented in this grouping. Secondly, single person businesses and general partnerships are combined as they offer the same level of liability. A total of 32 of the data points are represented in this grouping. Thirdly, CV is grouped together with Maatschap (partnership) as these have the same non-recourse characteristic. A total of 71 of the data points are represented in this grouping. Finally, a catch-all group is made to include other legal forms and undetermined data points. A total of 484 of the data points are represented in this grouping.
3.3 Methodology
The statistical analyses focuses on two domains of relationship lending, namely the effect of duration on information accumulation and the effect of duration on the effects of the Hold-Up Problem (‘HDP’). The following conceptual model is used:
20
The first hypothesis aims to investigate the effects of relationship duration on information accumulation. Elsas (2005) equates relationship duration with information accumulation. A regression analysis based on the development of collateral value over time, with relationship duration as independent variable, should offer insight into the effect of relationship duration on information accumulation. It is assumed that the collateral values incorporate soft qualitative information that is learned about the borrower. The hypothesis is formulated as follows:
H1: With an increased relationship duration, the volatility of value of the collateral decreases. The following four hypotheses aim to investigate the relationship between duration and HDP. As Boot (2000) points out, there can be numerous factors that dominate the HDP, leading to counterintuitive results. This thesis attempts to overcome other factors by incorporating two conceptual elements of the HDP (Monopoly rents & Availability of credit) and two dependent variables for each conceptual element. Relationship duration is used as independent variable, with control variables in the form of borrower size and legal entity. For the conceptual element Monopoly rents, the following two hypotheses are formulated:
Relationship lending - Duration
Legal entity
Client selection / risk reduction Increased switching costs
>> HDP Information
accumulation
Monopoly rents ---H2a Net margin (-) ---H2b Amo % of FCF (-)
Availability of Credit ---H3a Repayment DUR (+) ---H3b LTV (+) H1 H3 H2 Size LTV
21
H2a: With an increased relationship duration, the net margin decreases.
This follows the methodology that is present throughout the literature on relationship lending. The net margin is the element of the interest costs that is at the discretion of the bank manager and which would encapsulate the risk/return assessment of the borrower. If the borrower shows a reduced risk profile, the bank manager should reward the borrower with a lower margin. If information accumulation is achieved, the provision for unknown data is reduced and a reduced risk profile emerges.
H2b: With an increased relationship duration, a lower share of the cash flow after operating costs and interest costs is paid as repayment.
This variable is a contribution to the literature. Lenders extract cash flow from borrowers in the form of interest and repayments. Repayment durations have been investigated (see H3a) as variable for availability of credit. However, the share of repayments of the available cash flow (after fundamental operational costs and interest costs) has not been researched as variable for monopoly rents, whilst it directly affects the amount of cash flow the borrower can extract as dividend from the investment portfolio. If information accumulation takes place, a reduced amount of repayment as share of remaining cash flow would be expected to be acceptable for the lender. If the HDP is present, a higher share is expected to emerge.
For the conceptual element Availability of credit, the following two hypotheses are formulated:
H3a: With an increased relationship duration, the loan features a longer repayment duration. This thesis follows the existing method, present in the body of literature on relationship lending and HDP. This features the duration in which the total outstanding credit will be repaid if the current repayment schedule is maintained. This is a measure of availability of credit, measured in time. It is expected that the lender would prefer to have a longer exposure to borrowers with whom information accumulation is achieved, above borrowers with whom no information accumulation is achieved. If the HDP is present, the lender will attempt to extract increased repayments and reduce the repayment duration.
22
H3b: With an increased relationship duration, the loan features a higher Loan-to-Value ratio. An advantage of researching a fully collateralised loan database, with elaborate collateral valuation policies, is that a clear measure of availability of credit emerges: Loan-to-Value. This follows previous literature that features a similar ratio, debt-to-asset, in determining availability of credit. The advantage of the Loan-to-Value ratio is that is valued by the lender, whereas the debt-to-asset also relies on the borrower’s valuation of assets (such as goodwill). The assumption is that demand for amount of credit is equal between new and established borrowers and that due to information accumulation the lender is willing to bear higher Loan-to-Value ratios with borrowers that have a longer relationship duration.
Four hierarchical multiple regression analyses are executed aiming to determine whether the HDP influences the amount of monopoly rents extracted by the lender and the amount of credit available to the borrower. In the first step, the control variable Size and the dummy variables for governance characteristics (Eenm_VOF, CV, Overige) are incorporated. The dummy variable for BV_Stich is used as reference category in relation to the variable governance characteristics, as this is the most common group of legal entities.
In the second step, the research variables are added to the regression analyses. For H2a, H2b & H3a these concern the variables Relationship duration and Loan-to-Value. For H3b, Loan-to-Value is the dependent variable in the regression analysis and therefore the only addition in the second step is Relationship duration.
4. Results
This section analyzes the empirical results of the determinants of (i) information accumulation, (ii) monopoly rents, and (iii) availability of credit. It is assumed that duration has an effect on all three, with monopoly rents and availability of credit being dependent on information accumulation. This is consistent with the basic premises of the theory on relationship lending.
Some remarks are needed on the assumptions made for the regression models. The first assumption is the normality of the residuals. Inspection of the histograms (for each regression analysis included in the appendix section 3) of the standardised residuals reveals that the residuals do not show prominent deviations from normality. The second assumption
23
is the linearity of the regression model. The scatterplots (for each regression analysis included in the appendix section 3) of the standardised residuals on the standardised predicted values show that the assumption with regards to linearity also is satisfied as the results show a random distribution. With regards to the third assumption, regarding the homoscedasticity of the residuals, the scatterplot and the random distribution of its outcome satisfy this assumption
4.1 Collateral value volatility
Hypothesis 1, as described in sections 2.1 and 3.3, is made under the assumption that information accumulation is a result of the extensive efforts the lender exerts to learn about the collateral objects. The change in value will be less volatile when information is achieved as the lender will have less unknown information to account for. The variables are described in section 3 and the definitions and sources are detailed in table I.
To test hypothesis 1, this thesis runs a regression analysis with the relative change in collateral as dependent variable and relationship duration and object duration as independent variables. The following functional form is executed:
Collateral value volatility
= β0 + β1 Relationship duration + β2 Object duration
The results of the multicollinearity analysis are shown in Table II and show significant relationships between the start value, end value and change in values. Intuitively this would be the expectation as the variables are based on the same underlying asset. This is also supported by the fact that all VIF scores are smaller than 5.
An overview of the results is presented below in Figure 1. The full results of this regression analysis are detailed in tables IIIa, b & c in the appendix.
24
The regression model as a whole does not produce a significant outcome as R Square=0.001. Furthermore, the partial effects of Relationship duration (beta=0.001) and Object duration (beta=0.027) are neither significant. Without a significant relationship, hypothesis 1 will therefore be rejected.
4.2 Loan terms
The second set of hypotheses is made under the assumption that increased relationship duration equals information accumulation. This follows Elsas (2005), who explicitly makes this assumption, and which is prevalent throughout the relationship lending literature. The hypotheses address two components of loan terms: monopoly rents and availability of credit. Control variables for size and governance characteristics are included. Please refer to section 3 and table I for definitions and descriptions of the variables. The assumptions regarding normality, linearity, homoscedasticity and random distribution are verified in the same manner as was done for the regression analysis for H1. Please refer to Appendix section 3 for the histograms and scatterplots.
An interclass correlation analysis is executed, of which the results are presented in table IV. A number of significant relationships are found, albeit with generally limited effects. Predictably, ‘Repayment duration’ is negatively related with ‘Repayment as % of FCF’, which intuitively makes sense as both are derived from the repayment schedule. ‘Loan-to-Value’ shows a positive effect on ‘Net margin’, which makes sense as this would mean that the lender prices higher risk propositions with higher returns. Making less sense, a significant positive relationship is found between ‘Loan-to-Value’ and ‘Repayment duration’. This is counterintuitive as the lender would likely endeavour to reduce the higher LTV more quickly than a low LTV, which would lead to an opposite relationship. Perhaps the most telling and consistent finding in this correlation analysis is the positive relationships found between ‘Size’ and ‘Repayment duration’, ‘Loan-to-Value’ and ‘Relationship Duration’. This might offer insights when discussing the results later in this section.
No significant relationships are found on the independent variable Relationship duration, which does not bode well for the outcomes of the following regression analyses.
25
4.2.1 Monopoly rents
Hypotheses 2a and 2b, as described in sections 2.1 and 3.3, aim to investigate the effect of duration on monopoly rents. The results of the hierarchical regression analysis are presented in the appendix in tables V & VI. Hypothesis 2a follows the approach used by Degryse & Van Cayseele (2000), but without the need to incorporate corrections to come to a component of the interest costs that reflect the bank manager’s assessment of risk/return profile of the borrower. In the loan database that is used, the net margin is included. This is the component that is at the bank manager’s discretion and therefore should adjust discriminately on the bank manager’s perception of the borrower. The following form function is used in the analysis for H2a:
Net margin =
β0 + β1 Relationship duration + β2 Size + β3 Governance characteristics + β4 Loan-to-Value.
An overview of the results is presented below in Figure 2. The full results of this regression analysis are detailed in tables Va&b in the appendix.
26
In the first step, the model is not significant as R Square = 0.006. The second step adds a significant amount of variance to the analysis but this is caused by the control variable LTV (beta=0.109) and not the dependent variable Net margin (beta=0.015). This outcome was predictable as a significant relationship between LTV and Net margin was established in the interclass correlation analysis. H2a is therefore rejected.
Hypothesis 2b is based on a new variable Repayment as % of FCF’, which is used as a proxy for monopoly rents. Although this is without empirical evidence, it intuitively makes sense as valuable additional variable for monopoly rents as it represents a method of extracting cash flow from the borrower and is derived from the same data as the empirically researched variable Repayment duration. The same form function is used as for H2a, with only the dependent variable replaced:
Repayment as % of FCF=
β0 + β1 Relationship duration + β2 Size + β3 Governance characteristics + β4 Loan-to-Value.
27
An overview of the results is presented below in Figure 3. The full results of this regression analysis are detailed in tables VI-a & b in the appendix.
As in the regression for Net margin, the model is not significant in the first step as R Square = 0.005. The second step again adds a significant amount of variance to the analysis (R Square Change = 0.010). Unsurprisingly, the control variable LTV (beta=0.102) is the cause for the additional variance and not the dependent variable Net margin (beta=-0.006), as detailed in table VIb.
These outcomes are consistent with those of the regression analysis for H2a. The fact that LTV shows a significant relationship with both variables for monopoly rents indicates that higher risks call for increased capturing of economic rents. Whether these are monopoly rents is not determined in this analysis. With the outcomes being consistent, this does indicate that Repayment as % of FCF can be a viable variable. However, hypothesis 2b is rejected as no significant relationship is found.
4.2.2 Availability of credit
Hypotheses 3a and 3b, as described in sections 2.1 and 3.3, aim to investigate the effect of duration on availability of credit. The results of the hierarchical regression analysis are presented in the appendix in tables VII & VIII. H3a follows the hypothesis of Degryse & Van
28
Cayseele (2000) by investigating the availability of credit in terms of time, using the duration of the repayment scheme as variable. The form function repeats the previous two regression analyses, with the dependent variable being replaced.
Repayment duration =
β0 + β1 Relationship duration + β2 Size + β3 Governance characteristics + β4 Loan-to-Value.
An overview of the results is presented below in Figure 4. The full results of this regression analysis are detailed in tables VII-a & b in the appendix.
As opposed to the two regression analyses for H2a and H2b, the model is significant as R Square = 0.012. In this model, the control variable Size shows a significant relationship with beta = 0.101. The second step does not add a significant amount of variance (R Square Change = 0.002). This is not an expected outcome, as Degryse & Van Cayseele (2000) find a significant relationship between their dummy variable for collateral (in their dataset 27% of loans are collateralised) and repayment duration. Intuitively, this would lead to the expectation that the variable used for collateral in this thesis (LTV) would show a significant relationship with repayment duration. However, this is not the case. Hypothesis 3a is rejected, as no significant relationship is found between repayment duration and relationship duration.
29
Hypothesis 3b is based on the variable debt-to-asset ratio used by Petersen & Rajan (1994), adapted to the market standard use in commercial real estate finance – being Loan-to-Value. In the regression analyses for both monopoly rent variables, LTV shows a significant relationship. This is consistent with the finding of Degryse & Van Cayseele (2000) who establish a significant relationship between collateral pledging and net margin. Following upon these indications, it is the expectation that higher LTV’s are only acceptable when the lender is willing to accept an increased risk profile. Based on the assumption of selection of good clients and ejection of bad clients over time, it stands to reason that a positive relationship exists between relationship duration and LTV’s. The form function is therefore as follows:
Loan-to-Value =
β0 + β1 Relationship duration + β2 Size + β3 Governance characteristics.
An overview of the results is presented below in Figure 5. The full results of this regression analysis are detailed in tables VIII-a & b in the appendix.
This regression model is not significant as R Square = 0.007. However, the control variable Size shows a significant positive relationship within the model (beta = 0.085, t(1196)=2.938, p=0.003). Although the model in itself is not significant, the fact that Size again is a
30
significant factor within the model, as it was in the regression analysis for Repayment duration, makes for consistent findings. Step 2 of this model does not add a significant amount of variance (R Square Change = 0.000, F-change(1, 1195)=0.560, p=0.454). H3b is rejected as no significant relationship is found.
Although the findings fail to establish significant relationships, in the interclass correlation analysis (Table IV) a correlation between Size and Relationship duration, LTV and Repayment duration is found to be significant. This hints at a relationship between size and quality of borrower. Intuitively, it makes sense that borrowers of higher quality are enabled to grow more and would be consistent with the selection of high quality borrowers and deselection of low quality borrowers. That selection would be done on the basis of accumulated information, which the five presented regression analyses have failed to establish a relationship with their respective dependent variables.
The findings are not in line with Degryse & Van Cayseele (2000), who find that loan rates have a positive relationship with relationship duration. The findings are also not in line with Berger & Udell (1995), who find a negative relationship between loan rates and relationship duration. The findings are in line with Petersen & Rajan (1995) who do not find any significant relationship between relationship duration and loan rates.
Interestingly, the article of Petersen & Rajan (1995) includes a variable on the amount of credit in relation to value of the asset of the borrower (Debt-to-Asset ratio), which is similar to variable Loan-to-Value used in this thesis. However, this thesis does not find a significant relationship between relationship duration and availability of credit, as Petersen & Rajan (1995), Berger & Udell (1995) and Degryse & Van Cayseele (2000).
Perhaps most surprising is the insignificant relationship that this thesis finds between relationship duration (as well as object duration) and the volatility of the collateral value. The effort the lender puts into accumulating information regarding the collateralised real estate assets leads to an expectation that the lender would become better at assessing the value of the real estate assets. This, in turn, would lead to smaller changes in value. This hypothesis was made under the assumption that collateral values include soft qualitative information, as described by Berger & Udell (2006). Specifically “the character and reliability of the SME’s owner” (Berger & Udell, 2006) should impact the collateral value under this
31
assumption as it proves to the lender the capability of the borrower to generate cash flow with the collateral real estate asset.
This assumption might be false as the appraisers have an assignment to produce a value that is a viable market price. This is largely dependent on external factors. Especially during an economic downturn it is possible that the accumulated information might not weigh up against external factors that influence pricing of the collateralised real estate assets. However, the efforts of learning about the real estate assets should lead to reduced information asymmetry and lead to differences between borrowers’ higher and lower degrees of information asymmetry.
5. Limitations
This thesis suffers from a number of limitations. The most apparent is the period that has been researched. The Dutch commercial real estate financing market suffered severely during the financial crisis from 2008 – 2014. As neither the lender nor the borrower were operating under normal business practice, the decision making process of both will have been affected.
That said, the database is substantial in size and offers transparency in the form of the net margin. The control variables had limited significant effects: dOverig showed a significant relationship with Net Margin and Repayment as share of FCF, and Size showed significant relationships with Repayment duration and LTV. The dummy variable dOverig acts as a catch-all group, and thereby creates a limitation by not clearly showing its underlying characteristics. Further research could determine the sources of the significant relationships of dOverig.
Another limitation that arises from the dataset is the variable used for size. Whereas Degryse & Van Cayseele (2000) use a classification on firm size, this thesis uses the total outstanding credit amount as proxy for size. In case a borrower has a small outstanding credit amount but is a large competitor in the market, this proxy will not encapsulate the dynamic of size. However, with the database skewed towards smaller end of the size, with 65% of borrowers falling in the €1mln - €5mln bracket, this limitation is not assumed to make a big impact. Secondly, an underlying portfolio of real estate asset(s) serves as
32
collateral, and their heterogeneous characteristics will negate the impact of Size in the cases where the borrower has a significantly larger size than the credit amount indicates.
With regards to the changes of collateral value, the financial crisis not only had a direct effect on the price of real estate assets but also caused a prolonged economic downturn in the Netherlands. It also strongly affected the income generating capacity of the lenders, in other words the ability to attract rental income. The change in collateral value will be influenced by this and its effect may have been of such an extent that it was dominant, causing the effect of the accumulated information to be insignificant.
Finally, the nature of the business in which this research has been conducted may be subject to an above-average amount of dominating factors, as described by Boot (2000). Firstly, the real estate investment asset class offers low-obstacle entry to new sources of capital, both debt and equity, and the sector is not a knowledge-intensive industry that produces proprietary information such as patents. Secondly, it stands to reason that information regarding the real estate asset, rather than information regarding the borrower, should determine the extent of the Hold-Up Problem. This would be caused by the fact that the main business determinants take place within the real estate assets and not at the owner level.
This thesis, however, did not face a limitation in this regard, as the database included the duration of which the lender had knowledge of the individual objects. A potential avenue for further research would be to investigate whether information might not be accumulated linearly but may be achieved for the largest part in one specific timeframe (for example during the first year).
6. Conclusions
This thesis has focused on the effects that duration of relationship between lender and borrower has on the interaction between the two actors. Existing literature has made assumptions regarding the effect of duration in the first step on information accumulation and in the second step the loan terms. It is widely accepted that increased relationship duration leads to increased information accumulation. However, measuring information accumulation is challenging and measuring its predictive power even more so. That is what this thesis has attempted to undertake, by researching the relationship between duration (both relationship and object duration) and volatility of collateral value.
33
The findings do not support the hypothesis that increased relationship duration leads to increased information accumulation. Neither do the findings support the assumption that object duration has a significant relationship with information accumulation. However, the variable used in this analysis, Collateral value volatility, is an untested variable that would benefit from further verification of its viability. It is difficult to imagine that the extent of effort that the lender goes through to learn about the collateral does not lead to information accumulation. Then, the question is whether it manifests in other variables or whether external factors remain of such an impact that the accumulated information only serves for defensive uses.
Secondly, existing literature features two opposing concepts with regards to the effects of information accumulation (measured in relationship duration) and loan terms. Literature based on the Hold-Up problem expects that increased switching costs will put the lender in a position to extract monopoly rents. The other field in the literature expects that the increased information accumulation leads to a lower risk profile of the borrower, which should allow the lender to pass on lower pricing to reflect the decrease in risk. This thesis supports neither, as no significant relationship is found between relationship duration and monopoly rents (measured in the form of (i) net margin and (ii) repayments as share of cash flow after operating costs and total interest costs).
Finally, the availability of credit also does not show a significant relationship with relationship duration. A positive significant relationship is generally established in existing literature. By researching a loan database that is fully collateralised, from a lender who features an explicit and extensive valuation policy (with the aim of accumulating information), it was expected to follow the findings of similar research, following Degryse & Van Cayseele (2000) in repayment duration and Petersen & Rajan (1995) for a comparable variable for debt to company ratio. However, this thesis does not find a significant relationship between relationship duration and either Repayment duration or Loan-to-Value.
Reaching back to the finding that there is no significant relationship between relationship (as well as object) duration and volatility of collateral value, this indicates that the link between relationship duration and information accumulation is not as strong as presented in earlier research. The findings regarding Monopoly rents and Availability of credit could then
34
be considered to be consistent with the literature. If there is no change in information accumulation, there is no change in switching costs and risk profile.
Academic relevance
This thesis contributes to the body of literature regarding relationship lending, specifically on the subjects of information accumulation, monopoly rents and availability of credit. The methodology is based on commonly used research methods and contributions are made in the form of new variables on the basis of the specific character of the dataset. Against a backdrop of a body of literature that does not produce consistent findings, adding to the body of work strengthens the academic foundation. By including an analysis into the relationship between relationship (and object) duration and information accumulation, the analyses into relationships between relationship duration and variables for monopoly rents and availability of credit were based on a finding and not an assumption. It stands to reason that with less information accumulation, the effects of relationship lending, both positive and negative, are reduced.
Further research
Further research could focus on establishing the degree of information accumulation and the amount of proprietary information prior to researching the effects of relationship duration on monopoly rents and availability of credit. This might reduce the amount of inconsistent findings in the body of literature. As “relationship duration = information accumulation” is one of the main assumptions on which most of the Hold-Up Problem research is executed, this thesis proves that is worthwhile to first establish that assumption before researching the effects of the Hold-Up Problem.
Additional avenues for research are also present in making the variables more accurate. Different business sectors will likely produce varying degrees of knowledge that can be considered as proprietary information. Future research could incorporate a clear frame on the expected relative impact of the respective business sectors so as to ensure that dominating factors, as outlined by Boot (2000), are considered in formulating the variables. For example, one could investigate whether information accumulation in the respective business sectors progresses at a stable rate or if there is a concentration at one time period. This would enable the researcher to build a database on an accurate time frame.
