An Empirical Study of Subordination Levels in Commercial Mortgage Backed Securities by Romina Padhi Bachelor of Arts in Economics, University of Mumbai, 1997 Masters in Business Administration, University of Mumbai, 2000 Submitted to the Department of Architecture in Partial Fulfillment of the Requirement for the Degree of Master of Science in Real Estate Development at the Massachusetts Institute of Technology September 2005 © 2005 Romina Padhi. All Rights Reserved. The author hereby grants MIT permission to reproduce and to distribute publicly paper and electronic copies of this thesis document in whole or in part. Signature of Author ……………………………………………………………………… Department of Architecture August 5, 2005 Certified by ……………………………………………………………………… Brian Anthony Ciochetti Professor of the Practice of Real Estate, Thesis Supervisor Accepted by ……………………………………………………………………… David Geltner Chairman, Interdepartmental Degree Program in Real Estate Development An Empirical Study of Subordination Levels in Commercial Mortgage Backed Securities by Romina Padhi Submitted to the Department of Architecture on August 5, 2005 in Partial Fulfillment of the Requirements for the Degree of Master of Science in Real Estate Development ABSTRACT The CMBS market has been in existence since the mid 1980s; however, it was during the mid 1990s that the market began to grow. A combination of favorable interest rate environment, entry of new players in the market and the amount of demand for commercial real estate assets, led to a record US CMBS issuance in 2004, with the 2005 outlook being even better. However, the subordination or credit enhancement level of these securities has been on a downward trend since 1995. The thesis attempts to analyze the risk factors such as loan to value ratio, debt service coverage ratio, floating versus fixed rate, large and conduit deal types, as well as diversification factors (property type and geographic location), and their impact on subordination levels. Finally, market forces such as spreads on CMBS are also analyzed for their influence on subordination levels. For the analysis, data were collected on 430 commercial mortgage backed securities issued from 1995 through mid 2005. The information was obtained from Trepp, which tracks all the commercial mortgage backed securities issued in the market. The trend in subordination levels of each of the tranches or bond classes was analyzed over the period of study and a quantitative regression analysis was performed to analyze the influence of the above mentioned factors on the subordination levels. The results indicate that the loan to value ratio, interest rate type (fixed versus floating) and the deal type (conduit and large loans) have a significant impact on the subordination levels. Also, certain other market factors, including the spread differential between CMBS and corporate bonds, strong property market performance, increased liquidity and increased number of investors may also have influence the subordination levels of these securities. Thesis Supervisor: Brian Anthony Ciochetti Title: Professor of the Practice of Real Estate 2 Acknowledgement I would like to thank Don Belanger of CSFB for his help in compiling the data used for the regression analysis. His insights and valuable comments are appreciated. I would like to thank my advisor, Tony Ciochetti, who has been a guiding factor for this thesis. His wisdom and inputs on various aspects of the thesis were truly inspirational. 3 TABLE OF CONTENTS ABSTARCT ……………………………………………………………………… 2 ACKNOWLEDGEMENT ……………………………………………………….. 3 Chapter 1 – INTRODUCTION 1.1 Scope of Research ………………………………….……………………… 5 1.2 History and Background of CMBS …………….………………………….. 6 1.3 CMBS and Subordination Levels ………………….……………………… 8 1.4 CMBS Rating Methodology ………………………………………………. 12 1.5 Summary …………………………………………………………………... 16 Chapter 2 – LITERATURE REVIEW 2.1 Mortgage Default and Loss Severity Research ……………………………. 17 2.2 Real Estate Diversification Research ……………………………………... 25 2.3 Summary …………………………………………………………………… 28 Chapter 3 – DATA AND METHODOLOGY 3.1 The Data …………………………………………………………………… 30 3.2 Methodology ……………………………………………………………….. 33 Chapter 4 – DATA ANALYSIS 4.1 Subordination Levels ………………………………………………………. 36 4.2 Discussion of Regression Results ………………………………………….. 38 4.3 Other Factors affecting Subordination Levels ……………………………... 42 4.4 Summary …………………………………………………………………… 45 Chapter 5 – CONCLUSION ……………………………………………………….. 48 BIBLIOGRAPHY ………………………………………………………………….. 50 APPENDIX ………………………………………………………………………… 53 4 Chapter 1 INTRODUCTION 1.1 Scope of Research While the US commercial mortgage backed securities (CMBS) market has been in existence since the mid 1980s, it has emerged as an important sector since 1996 and has grown tremendously since then; with US issuance growing from $26 billion in 1996 to $93 billion in 2004. The CMBS sector has developed a broad investor base and a liquid secondary market. These securities have provided an attractive higher yield alternative to traditional corporate bonds, asset backed securities, and residential mortgage backed securities. From the investor’s perspective, CMBS are desirable as a security because risk is spread among different investors and tranches, and these assets are more liquid than owning a whole loan. The popularity of CMBS is also a function of the unique risk / return characteristics that these complex securities offer, particularly the lower rated tranches that exhibit both debt and equity like characteristics. With the development of the market, the structure of CMBS transactions has become increasingly important. The primary structure of CMBS involves a senior – subordinate structure that gives cash flow priority to the senior classes and shifts the default risk down the structure to the lower classes. These classes satisfy the demand of various classes of investors. The lower bond classes provide credit support to the higher rated tranches; this is referred to as the subordination or the credit enhancement level, which is determined for each bond class. A lot of research has been done on the pricing of these securities, how prepayments affects these securities and on the default of such securities. However, there exists minimal literature on the subordination levels in CMBS, which is a structural aspect of the transaction and plays an important role in determining how defaults and losses affect each of the bond classes. Subordination levels have fallen dramatically from the levels of 1995 / 1996 when they used to be in the range of 30% - 35% for the highest rated class (AAA) to as low as 12% - 15% in 2005. This thesis examines the trend in subordination levels of CMBS from 1995 to 2005 and the impact of specific risk and diversification factors as well as some deal specific characteristics (such as floating rate versus fixed, or large loans) and some market factors (yield spreads 5 between CMBS and corporate bonds) on subordination levels. This is an interesting area as there hardly exists any research in this area, though there are a few papers in the working. The analysis is based on a sample of CMBS deals or transactions from 1995 to mid-2005. The data, sourced from Trepp LLC, presents a summary of every deal issued including the features of the bond class, the details of the loans in the underlying pool, and the credit characteristics of the transaction. This thesis examines specific factors such as loan to value ratio, debt service coverage ratio, property type diversity, geographic location diversity, floating rate, and large loans, that may provide an explanation for this trend. The balance of this chapter provides background information on the CMBS market, the structure of the security and the subordination levels, how these credit enhancement levels are determined by rating agencies. Chapter 2 presents a literature review of the relevant research. Chapter 3 presents a description of the data as well as the methodology used in analyzing the data. Chapter 4 discusses the results of the data analysis. The thesis will close with the interpretation and implications of the results. 1.2 History and Background of CMBS The CMBS market has grown dramatically from its modest beginnings in the mid 1980s. Issuance, liquidity and the number of investors have all increased dramatically since and reached record levels in 2004, with 2005 expected to be an even better year. Last year the global CMBS issuance crossed the $100 billion mark for the first time in the decade plus history of the market. This year, it is expected to cross the $200 billion mark (US Issuance = $72 billion in the first half of 2005). The following graph depicts the explosion taking place in the US CMBS market. According to the Commercial Mortgage Alert1, loan demand continues to be fueled by low interest rates and rising property values. In fact, borrowers are currently financing securitized mortgages, indicating that the benefit of taking out larger loans with lower coupons exceeds the defeasance costs associated with prepayments. Global capitalization of CMBS currently exceeds $500 billion. 1 Commercial Mortgage Alert, July 1, 2005 6 Annual US CMBS Issuance $100.00 $93.11 $90.00 $77.85 $74.33 $80.00 $72.11 $67.15 $ Billion $70.00 $56.57 $60.00 $50.00 $36.79 $40.00 $26.37 $30.00 $20.00 $52.07 $46.89 $15.75 $10.00 2004 1H 2005* Year 2003 2002 2001 2000 1999 1998 1997 1996 1995 $- Source: Commercial Mortgage Alert, July 2005 * CMBS issuance for 1st half of 2005 (January to June) During the 1980s, a strong economy, the deregulation of the financial services industry, and preferential tax treatment led to an explosion in the level of capital flows into the commercial real estate markets. The real estate boom in the 1980s led to extreme overbuilding, which finally caused the bubble to burst in the 1990s. During the 1980s, primary sources of real estate funding included tax shelter syndicates, savings institutions, commercial banks, and life insurance companies. The Tax Reform of 1986 and the major devaluation of commercial property values in the early 1990s resulted in sizeable losses and led to a major retrenchment of lending activity by the traditional sources of financing. The largest factor contributing to the development of the CMBS market was the creation of the Resolution Trust Corporation (RTC). The RTC was created by the Congress to bail out the ailing thrift industry. The RTC would liquidate the assets, mostly mortgages, it acquired from insolvent thrifts as quickly and efficiently as possible. To “monetize” its investment, the RTC issued approximately $15 billion multifamily and mixed property CMBS. 7 Issuance has grown tremendously since 1993 and contribution from the RTC has fallen since. Witnessing the success of RTC in CMBS, many insurance companies, pension funds and commercial banks began to use the CMBS market as a source of restructuring their balance sheets. Non-RTC issuance began to build in the mid-to-late 1990s and could be classified into two broad categories – securitization of existing loan portfolios and securitization of new loans originated for the purpose of contributing to the underlying pool. Existing portfolio securitization consisted of both performing and non-performing loans, and thus, provided an exit strategy not just for the RTC but also for some insurance companies and pension funds that used it to liquefy their portfolios. The CMBS market was used as a means of liquidity for disposing the unwanted assets, to receive better regulatory treatment for holding securities in lieu of whole loans, or to raise capital for underwriting more loans. The total of commercial mortgages outstanding measured approximately 14.4% of gross domestic product, well above the long term average of 11.2%. A study2 on conduit diversity indicates that diversity has increased slightly. However, the largest deals were not necessarily the most diverse. Some transactions have become large by adding more loans (more diversity), while others have become large by adding larger loans (less diversity). 1.3 CMBS Structure and Subordination Levels A CMBS transaction is formed when an issuer places commercial mortgages into a trust, which then issues classes of bonds backed by the interest and principal of the underlying pool of mortgages. The basic building block of the CMBS structure is a mortgage that was written to finance a commercial purchase of property or to refinance a prior commercial mortgage. Each of the bond classes in a CMBS issue reflects a different risk – return profile. The objective of creating such a structure is to shift the investment risk from the highest rated class down to the lowest ranked class. There are many types of CMBS. A CMBS transaction can be backed by a pool of mortgages (conduit) or by a single asset. However, when underwriting a single property, the risk of 2 Moody’s Special Report, US CMBS 1Q 2005: Another warning light on the Credit Dashboard, April 28, 2005 8 unanticipated events that could affect the property’s cash flow is high. This is in contrast to pooled transactions, in which the diversification of properties and leases mitigates this risk. CMBS is very different from residential mortgage backed securities (RMBS) as there are certain aspects of the underlying pool that require different considerations in creating CMBS versus RMBS. First, prepayment terms differ; while residential mortgages permit prepayment without penalty, commercial mortgages usually require penalties or defeasance for prepayment. Second, in CMBS, even an imminent default has consequences as the special servicer (defined on Page 11) has wide discretion in modifications in such circumstances. Finally, a major distinction between CMBS and RMBS deals is the role of the junior bond class buyers. In CMBS, no deal occurs without first finding a buyer for the subordinate classes. This provides an extra layer of security for the senior buyer, particularly because the buyers of the junior classes tend to be knowledgeable real estate investors. The majority of CMBS transactions utilize a senior / subordinate structure, whereby the cash flow generated by the pool of underlying commercial mortgages is used to create distinct classes of securities. In such a structure, the lower priority classes provide credit enhancement for the senior securities. For example, credit support for Class A is the sum of classes B, C and D, and the credit support for class B is the sum of classes C and D. The credit support for a class must be fully extinguished before any default loss can affect the class. Thus, with subordination, the greater the credit support the higher rating the class receives. The most junior class has no credit support and is usually unrated. Traditionally, the lower rated (below BBB) and unrated classes are either privately placed or retained by the issuer, although this is slowly changing as the market matures and become comfortable with the characteristics of CMBS. Shown below is a simple diagrammatic representation of the senior / subordinate structure of a CMBS issue (based on the 2004 subordination levels). Credit support is the buffer or cushion from losses that each tranche requires for a specific rating category. For example, in the above structure 19.22% credit support at the ‘AAA’ level, the pool would have to suffer 19.22% of the loan balance before the ‘AAA’ certificates would be impacted. Similarly, the pool would have to 9 suffer losses that exceed 2.66% of the loan balance before the ‘BB’ certificates would be impacted. Table 1 Class (% Subordination) AAA (19.22%) AA (12.23%) A (8.39%) BBB (4.77%) BB (2.66%) B (1.55%) UR (unrated) (0%) Source: Data from Trepp LLC (based on average 2004 deal structures) The CMBS tranches retire sequentially beginning with the highest rated bond paying off first. Thus, any return of principal caused by amortization, prepayment or default is used to repay the highest rated tranche. Interest on principal outstanding will be paid to all the tranches, based on their outstanding face or nominal value, or is accrued for the lower rated classes and used to pay down the senior classes more quickly. The coupon rate of the security is always less than or equal to the lowest interest rate on any individual mortgage in the collateralized pool. This ensures that there is enough interest available to pay all the investors. 10 An important structural feature of CMBS transactions is the presence of the master and special servicers. Master servicers manage the routine, day-to-day administration functions required by all structured securities or collateralized transactions, while the special servicers are used to handle delinquent loans and workout situations.3 In case of a delinquency that results in insufficient cash to make all scheduled payments, the transaction’s servicer advances both interest and principal. This process continues till such time that the amounts are deemed unrecoverable. Often, the servicer’s interests are aligned with that of the CMBS investors as servicers now invest tin the non-rated and subordinate tranches of the deals they service. Losses arising from loan defaults will be charged against the principal balance of the lowest rated tranche and move upwards. The total loss charged includes the amount previously advanced as well as the actual loss incurred in the sale of the loan’s underlying property. A unique feature of the senior / subordinate structure is that the credit enhancement can grow over time. As principal is paid to the senior classes first, if no losses occur, these classes will pay down faster than the mezzanine4 or B-pieces5. This results in increasing the amount of subordinate classes as a percentage of the entire deal and thus, provides a higher credit enhancement of the remaining senior classes. Rating agencies play an important role in the CMBS market. The role of a rating agency is primarily to provide a third – party opinion on the quality of each bond class in the CMBS structure as well as the requisite amount of credit enhancement or subordination level to achieve the desired rating level. In essence, rating agencies help in deciding the appropriate structure for the deal. Subordination levels are set to attain AAA credit rating on the senior class. This is the highest rating possible, signifying that the bonds are deemed to have minimal credit risk. This makes the AAA tranche most attractive to investors. According to Standard and Poor’s, the credit rating assigned to a CMBS transaction is “an opinion on the ability of the collateral to pay interest on a timely basis and to repay principal by the rated final distribution date, according to 3 Frank Fabozzi, The Handbook of Commercial Mortgage Backed Securities Mezzanine Bonds: Non-senior securities receiving investment grade ratings 5 B-piece: Non-investment grade bonds 4 11 the terms of the transaction. The rating does not reflect the impact of prepayment or any other factors that may affect investors’ yields”.6 The credit risk is assessed by determining the probability of default and the severity of loss given default occurs. The credit risk captures the uncertainty in the timing and the magnitude of cash flow receipts created by a borrower’s default option. Since default generally results from a low collateral asset value, the risk lies in the volatility of the value of the security’s collateral i.e. the underlying real estate. Thus, the loan to value ratio as well as the debt service coverage ratio of the underlying pool of loans, among many other parameters, is also evaluated. If the target level of such parameters is not attained, the bond gets a lower rating. It is also important to estimate losses in the event a loan defaults and is foreclosed or restructured. The severity of loss should equal the property sale proceeds plus the property revenue, less principal owed upon default, foregone interest and expenses. 1.4 CMBS Rating Methodology According to a report by Moody’s on rating of conduit transactions7, the credit enhancement needed to achieve a rating level for a proposed securitization typically depends on the expected frequency, severity, and timing of future losses. An estimation of frequency and severity of losses is usually based on a statistical analysis of historical performance data for assets like residential mortgages, which can be homogeneous in character and, for which historical data is available. However, commercial mortgages are not uniform in character, and relevant historical loss information is limited. As a result, Moody’s analyzes the fundamental real estate credit risk of each asset to estimate the frequency and severity of losses within the legal and structural framework of structured finance. Moody’s reviews the portfolio diversification aspects of a pool, which have an impact on the volatility of expected loss for the pool, in turn affecting the level of credit enhancement needed for the rated bonds. In the case of non-recourse lending, which is typical of U.S. conduits, the default probability is assumed to be highly dependent on the debtservice coverage ratio (DSCR), and the loan to value ratio (LTV) associated with the underlying mortgage loan. DSCR is the main driver of frequency of loss, while LTV is the main determining 6 7 Standard & Poor’s Structured Finance CMBS Property Evaluation Criteria Moody’s Approach to Rating US Conduit Transactions, September 2000 12 factor for the expected severity of loss. Moody’s believes that the stability of cash flows and asset values of the major property types, ranked in order from best to worst, is as follows: multifamily, anchored retail, industrial, unanchored retail, office, and hotel. Moody’s also considers several portfolio characteristics, in addition to those of the underlying collateral. One such characteristic is the portfolio diversity; which could be by property type, geographic location, economic diversity and loan concentration. Moody’s views portfolios with multiple property types and geographic locations as more stable. Different property types have different risk profiles and market dynamics. Thus, property type diversification mitigates the expected losses in a pool, whether a pool of loans or a loan secured by a pool of properties. Geographic diversification helps mitigate the risk of single market declines, and serves to smooth the variability around an expected loss. It helps offset the impact to a pool from a regional downturn. For rating large loan CMBS transactions, the focus is on the credit characteristics of the asset, the structural features of the transaction, and the impact of diversification in a cross-collateralized pool situation. Moody’s considers the going-in and the balloon loan to value ratio and the actual and stressed debt service coverage ratios, in addition to the structural and legal issues. Large loans are typically pooled with other loans to reduce concentration risk. If a deal has eight to ten loans, the diversification benefit can be significant. Even though the expected loss for an underlying single loan does not change, the volatility of the expected loss for the pool is reduced due to higher diversity. Moody’s rating approach compares the credit risk inherent in the underlying asset with the credit protection offered by the structure. The credit risk of the underlying asset is determined primarily by two factors: the frequency of default, which is largely driven by DSCR and the LTV of the underlying loan, which impacts the severity of loss in the event of default. The structure’s credit enhancement is quantified by the maximum loss of value on the asset the securities are able to withstand under various stress scenarios without causing an increase in the expected loss for various rating levels. 13 Diversity is usually beneficial for the senior classes of bonds but may hurt the lowest rated class, all else being equal. Also, as more loans are pooled, there is a greater likelihood of the pool experiencing a defaulted loan. Since these transactions are tranched the increased likelihood of default is disproportionately concentrated in the junior most classes. To determine the appropriate credit enhancement, S&P8 first adjusts the cash flow of the underlying pool of properties to derive an estimate of the stabilized net cash flow that the property can be expected to sustain over the life of the securitized transaction. Sustainable net cash flow is derived by making adjustments to the property’s current revenues (typically trailing 12 months) and current expenses to produce the net operating income. These adjusted cash flows are used as the basis for modeling credit support for the transaction. The loss model measures the default frequency for each loan and the severity of loss that can be expected as a result of a default. The amount of recommended credit support is a function of the aggregate characteristics of the loan pool and will depend on the projected losses for each loan during various economic stress environments. Since the ‘AAA’ rating category is highest, the ‘AAA’ loss assumptions and related credit support should be sufficient to survive the worst possible economic stress. Conversely, the assumptions and the resulting credit support for a ‘B’ rating are less severe. Fitch’s model9 to determine the appropriate subordination levels for any given pool consists of three main components – default probability, loss severity, and pool composition factors. Fitch begins by calculating the debt service coverage ratio and loan to value ratio assuming an “AA” stress environment reflective of the real estate environment of the early 1990s. The default probability and loss severity assumptions, based on the DSCR and LTV for each loan, are adjusted for certain property and loan features to determine the credit enhancement based on individual loan characteristics. Finally, the composition of the pool is analyzed to identify any concentration risks, and the structure and the parties to the transaction are evaluated and incorporated into the ratings. 8 9 Based on Standard and Poor’s CMBS Property Evaluation Criteria Fitch Ratings Commercial Mortgage Criteria Report – Rating Performing Loan Pools 14 Table 2 Fitch Model Overview Default Probability X Loss Severity = Base Subordination Base Subordination X Pool Composition Factors = Final Subordination The default frequency of a specific loan is a function of its respective DSCR and LTV. The loss severity is the sum of the principal balance that may be lost during the liquidation and foreclosure period, accrued interest on the defaulted mortgage balance, as well as legal and other costs associated with the foreclosure and liquidation of the asset. Thus, the credit enhancement is a function of the default probability and loss severity, in addition to other pool composition factors. Subordination Level = ƒ (default probability, loss severity, pool composition factors) The foreclosure frequency at each rating category is multiplied by the loss severity at each rating category to provide the required credit support at each rating category. When determining the losses associated with any loan, the analysis incorporates the amount of amortization that has occurred by the assumed default date; the presence of additional debt; the quality of the real estate; the property type; the loan structure; whether the property is located in a judicial foreclosure10 or power-of-sale state11; and the availability of credit for the related asset type, or other factors that may affect liquidity and credit and thereby heighten or reduce the risks of potential losses associated with any loan. Thus, while the default frequency of two loans within the same property type with the same DSCR and LTV is the same, the losses associated with each of these two loans may differ, based on their individual characteristics as outlined above. Other factors are also considered in deriving credit support or subordination levels, which may affect the final levels that are recommended for the pool. For instance, environmental and property condition, seismic reports, underwriting standards and the lending environment in 10 Judicial foreclosure necessitates that the court order the foreclosure and supervise both the process and ultimate distribution of funds. 11 This is associated with a deed of trust mortgage instrument that allows an independent party (attorney or trustee) to conduct the sale in the event of foreclosure. 15 which the loans were originated are also factored into the recommendations. Additionally, the absence of reserves, cash management, and other structural features may adversely affect the overall sizing of the transaction. 1.4 Summary With the CMBS market outperforming expectations, the outlook still continues to be positive for the future. Market factors such as low interest rates, the entry of an increasing number of investors and the tightening spreads continue to make CMBS a competitive alternate to whole loans and other investments. The objective of this thesis is to investigate the risk and diversification factors as well as other pool characteristics such as pool type, interest rate type and market factors such as the spread differential between CMBS and corporate bonds that affect subordination, and if they are significant in explaining the change in subordination levels of CMBS. 16 Chapter 2 LITERATURE REVIEW Since subordination levels are determined by the probability of default and loss severity of the loans, it is important to look at factors that impact these very aspects. Thus, the relevant research includes studies on the default and loss severity characteristics of commercial mortgages. The default and loss severity study of mortgages is relevant as subordination levels are based on the probability of loss (due to default) of the mortgages in the underlying pool and the severity of loss, once a loan defaults. There is extensive literature on defaults of commercial mortgages and a few on the loss severity of commercial mortgages. Mortgage defaults have an impact on the credit risk of the mortgages, which my affect the subordination levels of the CMBS transaction. Also, as an attempt is being made to understand any diversification benefits that may affect subordination, studies on diversification benefits in real estate are also discussed. However, due to the relatively adolescent nature of the CMBS market, there have been few studies directly on the topic of subordination. 2.1 Mortgage Default and Loss Severity Research Default risk affects CMBS in different ways. Subordinate classes absorb losses first and they are generally most vulnerable to default. Interest only bond holders are subject to risk in that any decrease in the notional principal balance results in the loss of interest income without any compensating prepayment penalties. Mezzanine class investors are generally protected against any principal losses but need to understand their exposure to extension risk, which is the risk that balloon loans will not be able to pay off their principal balance at maturity. Finally, even the senior class investors are exposed to default risk. CMBS with lower quality collateral have greater cash flow risk. One of the most important aspects in assessing default risk in a CMBS transaction is the debt service coverage ratio (DSCR), which is the ratio of the available income from the property to its required debt service. The closer the DSCR is to one, the higher is the chance that the property income may fail to cover the mortgage payments. Some of this is alleviated by reserves which cover periods when income is interrupted or expenses are unusually high. 17 Another important aspect is the loan to value ratio (LTV), which is the ratio between the loan balance and the value of the property. In theory, the LTV should be more important than DSCR because a borrower should default only if the property value is less than that of the loan. Even if the DSCR falls below one, the borrower should be able to sell the property for more than the value of the loan and will not default. While LTV is an indication of the borrower’s incentive to support the debts service, the DSCR is an indicator of the property’s ability to do so. There have been several studies that have examined the relationship between default of commercial mortgages and the LTV and DSCR. A brief summary of findings and conclusions of the studies are discussed below. The incidence of default rises with the LTV12; thus, holding all other factors constant, the probability of default for a loan increases as the LTV increases. However, the increases are not equal. For example, an increase in original LTV from 60% to 70% resulted in a 50% increase in the probability of default from 0.54% to 0.80%. An increase in original LTV from 50% to 60% increased the probability of default by 0.18%; and an increase in LTV from 80% to 90% drove up the probability of default by 0.58%. Unlike the LTV, the probability of default decreases with corresponding increases in DSCR. However, studies have shown that the relationship between DSCR and probability of default is weaker than the relationship between LTV and probability of default. One explanation for the above is that borrowers have an incentive to negotiate the payment rescheduling and the debt restructuring with the lenders, but the incentive wanes quickly when the LTV is greater than 100%. Vandell, Barnes, Hartzell, Kraft and Wendt (1993) were the first to use loan level commercial mortgage data from a life insurance company and the results confirm the effect of loan terms and property value trends affecting default. Ciochetti, Deng, Gao and Yao (2002) used a similar data 12 Frank Fabozzi, The Handbook of Mortgage Backed Securities 18 set and applied the competing risk framework developed by Deng, Quigley and Van Order (1996, 2000) in the residential mortgage market to commercial mortgages. According to a study of defaults and loss severity in CMBS by Fitch Investors Service (1996), commercial mortgages with relatively little excess cash flow after servicing loan payments are more likely to default. DSCR is among the most significant variables in commercial mortgage default and loss severity. According to Pamela Dillon, Fitch found that loans with low DSCR had relatively higher annual rates of default. The study found significantly different average default rates and loss severity experience among different bands of DSCR. At one end, loans with DSCR below 0.5 showed annual default rates of 7.1% and average loss of 55.3% while the default rate for loans with a DSCR of 2.1 was just 1.8% with a loss of 12.8%. However, when the property’s income falls below the debt service, borrowers temporarily support the payments if the property is viewed as having a long term ability to pay the debt service. In another study of CMBS monitored by Fitch, Dillon and Belanger found significantly higher defaults and loss severities for floating rate compared with fixed rate loans, despite a period of “relatively stable and low interest rates”. The average annual default rate in the 1996 study was 6.7% for balloon loans, 2.4% for fully amortizing loans. The severity of loss was 31.1% for fixed rate loans and 41.2% for floating rate loans, with no information on whether these were balloon or non-balloon loans. A study by FITCH on default rate of commercial mortgages (1991 to 1997) by property type reported that loans secured by warehouses had the lowest annual default rate at 2.5%, followed by multifamily properties at 3.9%. The default rates for these property types were considerably lower than all other property types. The reason could be attributed to the fungibility and low capital cost of warehouse space and the very slow obsolescence and stability of multifamily properties. The default rate for the lodging sector was 4.2%, while office and retail properties had the highest default rates at 4.8% and 4.9%. Office properties have high capital costs relating to tenant improvement, leasing commissions, replacement reserves, etc, which may cause added stress on the property during times of significant tenant roll. The high retail default rates could be attributed to a significant percentage of unanchored retail and retail strip centers, which are more 19 volatile and subject to risk than malls and anchored community centers that were prevalent in the thrift loan portfolios. The loss by property type also varies with multifamily properties at 46%. This could be due to the fact that many of the loss observations were high leverage loans from RTC/thrift transactions, several of which were concentrated in real estate depressed areas such as California. These rates have changed since then but do give an insight into the importance of diversification in a CMBS pool of mortgages. The study also revealed that loans with low DSCRs, floating rate loans, and balloon loans had relatively higher annual rates of default. A study by Ciochetti13 (1997) showed that state foreclosure laws appeared to impact the performance of the assets, measured by net loss recovery. The data consisted of loans originated by fourteen life insurance companies between 1986 and 1995. The net loss recovery over the period averages 69%. The study finds significant evidence that the jurisdictional foreclosure method is strongly related to not only mortgage loss recovery, but the level of foreclosure cost and imputed interest as well. Loans foreclosed judicially have average recoveries that are 11% lower than loans foreclosed through power-of-sale. Additionally, judicially foreclosed loans have foreclosure costs nearly double, and total foreclosure costs nearly 50% greater, than for nonjudicially foreclosed loans. According to Fitch14, another variable that may explain loss severity is the method of loan resolution in each state. Foreclosures handled judicially take longer, on average, and result in higher losses than non-judicial foreclosures, which take place in power-of-sale states. The method of loan resolution had significant impact on loss severity, but very little discernible impact on average annual default. Power-of-sale states had an average annual default rate of 4.2%, while judicial states had an average annual default rate of 4.4%, with the results not be statistically significant. Archer et al. (2001) point out that the downside of applying contingent claims modeling to commercial mortgages is that variables such as LTV and DSCR are endogenous to the loan origination process. Lenders often decide on the LTV and DSCR based on the risk associated 13 14 Brian A Ciochetti, Loss Characteristics of Commercial Mortgage Foreclosures, Real Estate Finance, Spring 97 Faboozi, Frank, The Handbook of Commercial Mortgage Backed Securities 20 with a particular loan as opposed to applying a single LTV or DSCR across all loans. If this is the case, it is difficult to observe an empirical relationship between default and LTV and DSCR. The results of the study by Ambrose and Sanders (2003) confirm that default and prepayment are directly affected by changes in the economic environment, specifically by the future expectations of interest rate. Also, mortgages with higher LTV at origination are more likely to prepay, but the study doesn’t find a statistical relationship between LTV and the hazard of default. Researchers, Rod Dubitsky and Kumar Neelakantan (2001) found that current LTV, rather than original LTV, is a much more significant indicator of loss severity. The data suggest that when adjusted for other factors, higher current LTV at payoff experience a higher loss severity. Dubitsky credited the fact that low original LTV loans tend to prepay in full out of delinquency as one of the reasons for the lack of significance of original LTV in explaining loss severity. In summation, CSFB says that other things being equal, pools with lower average current loan balance, greater seasoning, and higher current LTV loans would be expected to have higher loss severity. A study by Ciochetti et al (2003) used the proportional hazards model to understand commercial mortgage defaults. The results of the study indicate that contemporaneous loan to value ratios are of less importance than contemporaneous debt service coverage ratios in explaining commercial mortgage defaults, though both these factors are of considerable significance in explaining the quarter-to-quarter variations in default rates. Both the LTV and DSCR may change over time – the first due to change in investors’ risk preferences and the latter due to decrease in rents or increase in vacancy rates. The results from the model do confirm that the contemporaneous DSCR is negatively correlated with default, while the contemporaneous LTV has a positive nonlinear relationship with commercial mortgage default. The study also evidences a greater increase in default with a decrease in DSCR than with an increase in the LTV; thus implying that the ability to meet the debt service obligations is the most relevant measure of commercial mortgage default. An analysis on loss severity by Pendergast and Jerkins of RBS Greenwich Capital (2003) analysis the performance of 33,666 loans with an original balance of $185 billion securitized in 21 fixed rate conduit and fusion CMBS transactions issued from 1995 through 2001. They examine the impact of property types, year of origination (vintage), loan size, loan to value ratio, and loan liquidation on CMBS loss severity. They mention that the most compelling result of the study is that the loan size matters – larger loans ($20 - $30 million) that have been liquidated during the period of the study have lower loss severities than the smaller loans ($2 - $3 million) that have been liquidated. To study the impact of loan to value ratio on loss severity, the 194 loans that were liquidated were also broken into various LTV buckets. According to the study, the 70% to 80% LTV bucket experienced the highest percentage of liquidated loans at 44%. However, it should also be noted that 50% of all the loans in the study fall in the LTV bucket of 70% to 80%. Loans with higher original LTV, in general, experienced higher loss severities, though there were some exceptions. Loans in the 60% - 70% LTV range experienced a weighted average loss severity of 41.2% but the loans with an LTV in the >90% range experienced a weighted average loss severity of 74%. At the same time, loans in the <60% range had a relatively high loss severity of 50%. The reason for such exceptions is that the study is based on original LTV, as recent appraisals were not available to determine the LTV just prior to disposition of the loan. However, LTV does seem to impact the loss severity to a certain extent. A subsequent study by Titus and Betancourt (2003) researched losses incurred on liquidated loans in non-agency CMBS transactions from 1993 through 2002. The results of the study indicate a cumulative loss severity of 40.11% over the nine year period. Also, the degree of loss severity is directly related to the condition of the property markets. Loss severities increased in the more recent years of study as the property market fundamentals had deteriorated. An econometric analysis was performed to examine factors affecting loss severity. The results show a statistically significant relationship between loss severity and the length of time to liquidate an asset. However, the relationship between the loans in the study and the original debt service coverage ratio and the original loan to value ratio were found to be statistically insignificant. This result is consistent with earlier studies that suggest contemporaneous LTV and contemporaneous DSCR as having greater explanatory power for delinquency and default. 22 The study by Chen and Deng (2004) states that the loan’s current equity share (calculated as one minus LTV) has a dominant effect on the probability of the underlying collateral value dropping below the critical value and hence, to default. The study also expects the net operating cash flows (as evidenced by DSCR) to have a negative relationship with foreclosure. The study reveals that the two loan specific financial variables, LTV and DSCR, are not significant, with LTV having the correct sign and DSCR having the incorrect sign. Chen and Deng attribute this to some errors in variable problem. Since both variables are largely estimated from market level indices, they may not reflect the true financial conditions of the properties. They state that both LTV and DSCR in their model are imperfect (noisy) measures of financial characteristics of an individual property. In their analysis, there appears to be big differences between what they refer to as “bad” loans and “good” loans in terms of average LTV and market occupancy rates. The differences in DSCR are also as expected, but not as pronounced as in LTV. Higher LTV leads to more defaults, which is what the option pricing theory suggests and appears to conform to the findings of Chen and Deng. The pool level weighted average LTV and DSCR is analyzed for the data, over time. However, it should be kept in mind that the weighted average statistics can be deceiving. For instance, two deals may have the DSCR below 1.00 and one of the deals has loans with LTVs above 100%. The lower quality of these pools is usually recognized by the rating agencies in the form of higher subordination. Since loan in conduit pools are not cross-collateralized, defaults are expected in such pools. While the AAA tranche may not experience losses due to adequate subordination, there still exists a great average life uncertainty associated with such pools. This is because recoveries from foreclosures are first paid to the senior classes. Thus, it becomes difficult to predict when the AAA bond holder receives the principal. Thus, investors in the AAA class from a low quality pool need extra spread to compensate for the average life uncertainty. A Moody’s report (2004) analyzes the relationship between loan delinquency and leverage. Delinquencies are found to be higher among loans with higher leverage, and the slope of the increase is non-linear, rising sharply and by multiples at certain levels. The same holds true with 23 respect to the DSCR. While, even among the best cohorts, delinquency is never zero, loans with a DSCR greater than 1.70 and LTV lower than 50% have very low delinquency rates (less than 1%). High leverage loans (greater than 90%) begin to separate from the pack as early as the first full year of life, experiencing delinquency at two to three times the rate of the lower LTV loans. The predictive power of initial DSCR varies with time. In the initial years, it matters a lot when coverage is low. However, by seven to eight years into the life of the loan, the initial DSCR is not strongly consistent with performance. A recent study by Downing and Wallace (2005), based on Monte Carlo simulations of expected defaults, indicate that the optimal subordination levels lie below those currently seen in the market. One interpretation of this result is that in the absence of a track record of performance of securitized commercial mortgages, subordination levels in the early years of CMBS were conservative and high. Several studies by been conducted by Esaki on commercial mortgage defaults. The first study included loans from 1972 through 1997; the subsequent study added three more years of data through 2000. The latest study (2005) includes loans through 2002. The addition of the new loans alters the previous conclusion (1972 – 2000) slightly. In the past, all investment grade bond classes of CMBS were well protected from the magnitude of losses experienced by the life insurance company loans. The second study revealed that with the reduction in subordination levels, some of the BBB classes would be vulnerable to a downturn of the magnitude of the late 1980s and the early 1990s. The latest study suggests that a greater proportion of the BBB classes are now vulnerable to a downturn of the magnitude on the late 1980s and early 1990s. The loss on the worst cohort, 1986, exceeds the average BBB subordination level on conduit and fusion CMBS transactions being issued today, and may also result in default of some A tranches. However, it is still below the lowest credit support levels for the AAA bond classes. Since BBB subordinations are about 5%, most investment grade CMBS are still protected against the average loss of origination cohorts of the last 30 years. The results show an average lifetime cumulative default rate (based on loan balances) for cohorts with at least ten years of seasoning decreased from 20.5% to 19.6% over 2000 to 2002. The 1986 24 cohort of originations continues to be the worst cohort in the past 30 years with nearly 33% of the total loan balance eventually defaulting. The average loss severity on liquidated loans was approximately 33%. The loans (as in the earlier study) are geographically well diversified, with the largest in the West (23%) and Northeast (22%). Highest default rates were in the South Central Region (25%) with the lowest in the West (10%). According to a Moody’s study15, across all asset types, default rates increase progressively as contemporaneous DSCR declines. Above a DSCR of 1.30, little variance in default rate occurs; default would occur only under extraordinary circumstances. Default rates begin to accelerate at a DSCR in the range of 1.20 to 1.30, a baseline commonly used for loan origination; implying that the probability of default is not zero even if the loan has a currently adequate DSCR. As the DSCR decreases, the probability of default progressively increases. The study states that default probability almost doubles when the contemporaneous DSCR from 1.0 to 0.80 to 0.90. These levels are likely to indicate a profound change in the property level conditions. Thus, a review of the above literature indicates that the debt service coverage ratio and the loan to value ratio just prior to the loan defaulting are important indicators of default probability. Additionally, the average loan size also seems to have an effect on the default and loss severity of commercial mortgages. These factors will be studied and analyzed in the thesis. 2.2 Real Estate Diversification Research The two sources of risk relating to the underlying pool of loans are the prepayment risk and the default or delinquency risk. Diversification of the underlying collateral is one of the ways to mitigate default risk. CMBS with better diversification in the underlying pool of loans have less exposure to economic events that affect particular regions or property types. CMBS investors focus on a number of portfolio issues, especially those involving the composition of the total collateral pool. The benefits of the CMBS structure derive from an ability to make the real estate investments without risks associated with the direct mortgage or equity placements (i.e. diversification by property type, geographic location, loan size, borrower, 15 Moody’s Special Report, US CMBS: DSCR Migration and Contemporaneous Probability of Default, June 7, 2005 25 etc.). This study focuses on the two aspects of diversification namely property type and geographic location. One of the factors considered when analyzing the risk of CMBS deal is the diversification of the underlying pool of mortgages across space. The rationale for this “spatial diversification” is that the default risk of the underlying mortgages is reduced if the loans are made on properties in different regions of the country. Most benefits come from geographic diversification, since regional recessions are more common than national recessions, and national recessions tend to affect regions differently. The diversified portfolio of the loans can spread the risk across many economies as opposed to having the entire portfolio of loans subject to an idiosyncratic risk factor. For instance, the impact of a collapse in the Houston real estate market (which may lead to higher defaults of loans) will be lessened if the commercial property markets remain strong in New York, Seattle and Chicago. According to FITCH, geographic diversity within a pool is a positive attribute that lessens some of the risk of regional economic and business cycle stress. In a study of defaults by region by Ciochetti (2003), the Southwest (mostly Texas) accounts for most of the foreclosed loans between 1986 and 1991. The Northeast moved in the opposite direction – only six loans had been foreclosed between 1986 and 1989. However, this number peaked in 1992 due to rising costs of living, and the economic weakness in the early 1990s. Thus, the presence of geographic diversification is important in a CMBS pool. In addition to spatial diversification, CMBS pools can be diversified across property types. Rating agencies tend to assign a lower credit enhancement or subordination level to deals that contain diversification across property types since a pool that is diversified across residential, office, hotel, retail and industrial will most likely avoid the potential of a national glut in any one of the sectors. Also, it is not possible to predict with certainty as to which property type will enjoy the best performance going forward, and it is better to be adequately diversified. Different property types vary widely in their default performance reflecting degrees of volatility in each property type’s income and value. The property types in order of most to least stable are regional malls, multifamily, anchored community shopping center, industrial / warehouse, unanchored community shopping center, office and finally hotel. 26 A study by Grissom et al. (1987) studies the effects of portfolio diversification on the reduction of unsystematic real estate investment risk by examining calculated returns of actual real estate assets in two geographically different markets. Ex post data were collected from 170 different income properties in two Texas markets between 1975 and 1983: Houston (40 assets) and Austin (130). To test for the effects of portfolio diversification, 4 types of assets in each market are identified: apartments, offices, industrial properties, and shopping centers. The empirical results indicate that the variance of returns of real estate portfolios decreases rapidly with diversification into a larger number of assets, across property types, and across markets. The average reduction of total unsystematic risk is 92.5% for diversification across markets and 97.8% for diversification across assets. The study concludes that diversification may be even more effective for real estate assets than for common stocks. A study by Laura Quigg (1997) indicates that the benefits of property type diversification are not as great as geographic diversification since the performance of different property type trends tends to be more highly correlated than that of properties in different locations. Multifamily is diversified in that tenants comprise of different economic and demographic backgrounds, so a 100% multifamily can still be considered diverse. For other property types, there is a benefit to mixing property types, because it gives a variety in tenant base as well as other factors that affect property performance. A study by Taylor and Fuchs (2004) assessed the impact of geographical dispersion on risk in multifamily mortgages by generating different random portfolios for a given number of apartment loan assets. All assets are granted equal weight and all have similar loan characteristics, i.e., DSCR of 1.25 and LTV of 80%, 10-year maturity and extended lockout provisions. The portfolio’s degree of diversification is subject to the randomness of the metros selected for inclusion. The Diversification Ratio (DR), defined as the ratio of unexpected loss for a diversified portfolio over the unexpected losses for the same portfolio given no credit for diversification (i.e. ignoring potential benefits flowing from diversification), is then calculated for each portfolio. For 27 example, a DR of .6 would indicate that unexpected losses for a particular portfolio are reduced 40% through diversification. The results show that as the number of assets or metros grow the DR declines more or less monotonically with marked reduction in the dispersion of observed DRs. Incremental additions (of assets) to the portfolio initially reap large diversification benefits, but there is a substantial marginal utility to each incremental addition. The simulations used in the study appear to indicate that the vast majority of diversification benefit is obtained by portfolios consisting of 14-15 assets. Additional assets continue to yield benefits, but at a declining rate. The risk mitigation benefits of maintaining a diversified portfolio are very attractive. Simulations show unexpected loss calculations for a diversified portfolio are 55% of those of the same portfolio assuming no diversification benefits. The study also extends this analysis to geographically focused portfolios such as census regions, and results show that loan portfolios concentrated in any particular region also reap diversification effects. From highest diversification benefits to lowest, the regional results are: the South (.48); Midwest (.51); Northeast (.60); and West (.64). Some of these portfolios offer relatively more diversification than a randomly selected national portfolio, some less. The difference among regional DRs including the number of metros represented in the region, the overall market condition or credit performance of those metros and/or the observed correlation of the real estate markets within each region. A review of the above studies indicates that diversification in a pool of mortgages helps mitigate risk associated, and thus an optimum diversification is always beneficial. Thus, the thesis will also attempt to look at the impact of pool level diversification on subordination levels. 2.3 Summary The above referenced research tends to indicate that the current / contemporaneous loan to value ratio and the current / contemporaneous debt service coverage ratio provide significant explanatory power for defaults and loss severity. Also, the average loan size (large versus small loans) seems to have an impact on default and hence needs to be considered for their possible impact on subordination levels. And finally, the impact of pool diversification will also be analyzed. 28 However, in addition to the above variables, other market forces may be playing an important role in determining the subordination levels. Thus, the thesis will also attempt to analyze factors such as spread differential between corporate bonds and CMBS, and the additional liquidity and flow of investors into the CMBS market that may have affected the subordination levels. 29 Chapter 3 DATA AND METHODOLOGY The objective of this chapter is to discuss the data and variables used in analyzing the trend in the subordination levels over time. The subordination levels of all the rated tranches of a CMBS transaction have been considered in the analysis. 3.1 The Data Data were collected on commercial mortgage backed securities issued between 1995 and mid2005. This reflects a decade of transactions in the CMBS market, thus enabling an analysis of the credit enhancement levels for a reasonable time frame. However, the universe of data has been narrowed to exclude FNMA, Freddie Mac and other Government Sponsored Enterprise (GSE) securities, due to the implied government support embedded in the subordination levels of such deals. Similarly, there are certain deals that have an implicit support of other government agencies, and the deals do not reflect any required subordination level. Such deals have also not been considered in the analysis. Thus, the final data set consists of 430 commercial mortgage backed securities issued in the past decade. Prior to collecting the data, the data fields that were thought to be most relevant in explaining the subordination levels were established. This includes the risk factors – loan to value ratio, the debt service coverage ratio, interest rate type (floating rate versus fixed rate deals), deal type (conduit, large loans); and diversification factors – property type diversification and geographic location diversification. Additionally, the spreads between corporate bonds and CMBS, and the increased liquidity and demand for CMBS will be examined. The dependent variable is the subordination level of each bond class in the transaction. The data for this thesis was collected from Trepp LLC. Trepp, LLC, is the leading provider of CMBS and commercial real estate information, analytics and technology to the investment management industry. Trepp meets the needs of both the primary and secondary market with a host of innovative products and services designed specifically for CMBS and commercial mortgage participants. A summary of every deal issued is available through the Trepp database. 30 It provides information summary information of the bond classes i.e. coupon rate and type, current rating, original rating, etc. It also provides information on the pool characteristics in terms of concentration of specific property types, of geographic locations, or metropolitan statistical area (MSA), etc. Other additional information include details of the top loans in the pool, details of delinquent loans in the pool, if any, top tenants in the properties in the underlying mortgages, information on defeased loans, if any and the collateral summary. Some of the descriptive statistics of the data are discussed below. From Table 3, the steady increase in the number of deals issued is evident, indicating the growth in the CMBS market. Table 4, further shows the classification of deals as per the deal type. The category “others” includes deals that are classified as “miscellaneous” (typically loans on golf courses, car garages, etc.) and credit tenant loans (as the name suggests, properties with a credit tenant makes up most of the underlying pool of mortgages). Table 4 Table 3 Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Total # of deals 15 24 37 45 44 36 41 38 52 66 32 430 Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Total Conduit 8 19 25 36 36 31 36 33 40 55 28 347 Large Loan 2 2 5 2 2 0 1 2 10 10 4 40 Deal Type Seasoned Franchise Loan 4 0 3 0 3 0 0 4 1 2 1 2 2 0 1 0 1 1 0 0 0 0 16 9 Others 1 0 4 3 3 2 2 2 0 1 0 18 In terms of fixed versus floating rate deals, there are a total of 396 fixed rate CMBS deals and 34 floating rate deals in the entire data set. The year wise break-up is given below. This shows an increasing number of floating rate deals in the recent years. This could be due to the low interest rate environment. 31 Table 5 Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Fixed 15 22 36 45 44 36 40 36 41 55 26 Floating 0 2 1 0 0 0 1 2 11 11 6 The following table shows the trend in the weighted average loan to value ratio and the weighted average debt service coverage ratio of the CMBS transactions from 1995 through mid 2005. Table 6 Year 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Avg. WALTV Avg. WADSCR 65.62 1.45 66.89 1.46 67.43 1.47 68.33 1.51 68.76 1.45 67.28 1.40 67.82 1.44 66.91 1.81 62.51 2.73 64.65 2.26 65.48 1.84 Out of the 430 CMBS transactions considered, 354 were diversified while 76 were nondiversified. Additionally, of the 32 CMBS deals that had a AAA subordination level of more than 35%, 19 were not diversified (Exhibit 1). The basis used for diversification is that the deal should consist of at least three property types with not more that 50% concentration in any one property type. The information used specifically for the analysis of subordination includes, first and foremost, the subordination levels of every bond class of every deal. Additionally, information on the 32 weighted average loan to value ratio and the weighted average debt service coverage ratio is used for the analysis. The deal type and the coupon type (fixed or floating) are also used. 3.2 Methodology The objective of this thesis is to analyze the trend in subordination levels for all bond classes over the past decade. The underlying hypothesis is that the reason for lower subordination levels in recent years is that the loans in the underlying pool may have become less risky. Risk attributes of a loan could be considered as a function of the loan to value ratio and the debt service coverage ratio. Subordination Level = ƒ (Risk factors) = ƒ (loan to value ratio, debt service coverage ratio, interest rate type, deal type) Another hypothesis is that the underlying pools of loans have become more diversified over the years, hence reducing the risk in the transaction. Standard linear regression has been used to analyze the data with subordination levels as the dependent variable and the variables described below as the independent or explanatory variables. Based on the review of the literature and the factors that affect default probability and loss severity, the specific variables that used for analysis of the subordination levels are described below. The variable used for loan to value ratio is the WALTV. This is the weighted average loan to value ratio of the pool of mortgages just prior to being securitized. The expected result for this variable is that a positive relationship exists between the loan to value ratio and the subordination levels. This is because, a higher loan to value ratio implies a higher risk of the loan, which would then impact the bond classes. Thus, a higher subordination level would be required for the different tranches to account for the higher risk in the underlying pool of loans. Subordination level = α + β (weighted average loan to value ratio) The variable used for debt service coverage ratio is WADSCR. This is the weighted average debt service coverage ratio of the pool of mortgages in the transaction prior to being securitized. The expected relationship between subordination levels and debt service coverage ratio is an inverse 33 relationship. The higher the debt service coverage ratio, the lower the risk associated with the loans, and thus, lower is the subordination or credit support levels required for a specific bond class. Subordination level = α - β (weighted average debt service coverage ratio) To analyze the diversity in the pool of mortgages in each deal, both property type and geographic diversification are considered. The model has been set up such that if a specific deal has mortgages on more than three property types with no more than a 50% concentration in any one property type, and is in at least four large cities across the United States, then the underlying pool can be referred to as diverse. A dummy variable, DDiverse, has been created such that if a specific CMBS transaction is diverse, the dummy variable attributed is 1 else it is 0. The expected relationship between diversity and subordination levels is also expected to be an inverse relationship. The greater the diversity is a pool, the lower the risk due to the spread across property types and geographies and thus, lower is the required subordination level. Subordination level = α - β (diversity) Conduit deals may have a lower subordination with respect to other deal types, such as franchise loans16, or miscellaneous loans (which may include properties like golf courses or garages). A dummy variable, DConduit, has been created to check if, in fact, conduit deals would indicate a lower subordination level as opposed to the other deal types. The intuition for such a relationship to exist is that conduit deals typically include loans on regular properties such as office, multifamily, etc. as opposed to more exotic and maybe, riskier properties such golf courses, etc. Thus is a specific deal is a conduit deal, then the dummy variable is 1 else it is 0. The expected relationship is a negative or inverse relationship, because conduit deals imply lower risk and thus, lower subordination levels. Subordination level = α - β (Conduit deal) Another variable that may affect subordination levels is the nature of the coupon i.e. fixed or floating. A floating rate deal, just like any floating rate mortgage, is assumed to have a higher 16 An example of a franchise loan is a retail property that is owned by a auto dealership of say, Toyota. They may also include properties that are restaurants or gas stations. 34 risk than a fixed rate deal. This is because, there is a degree of uncertainty associated with the movement in future interest rates that might impact the performance of the mortgages, and thus, their mortgage backed securities adversely. Such risk may be mitigated by a higher subordination structure for a deal. To examine this effect, a dummy variable DFloating, has been created such that deals that have a floating rate have a dummy variable of 1 and fixed rate deals have a dummy variable of 0. The expected relationship is that subordination levels should be higher for the floating rate deals. Subordination level = α + β (floating rate) Apart from the independent variables specified above pertaining to the risk associated with the loans, there may be other factors that may affect the subordination level for a particular deal in a particular year. For instance, since the data is time series, in that subordination levels have been considered over the past decade, time needs to be controlled for. This is because, arguably, other factors specific to the real estate cycle, or investment climate for other investments such as stocks and bonds, interest rates, etc may have also impacted the subordination levels. Thus, a dummy variable has been created for the year to control for the influence of time on the subordination levels. The dummy variables for the years are DNUM(yr) i.e. DNUM96, and so on. The analysis is on the CMBS transactions from 1995 through 2005. The year 1995 is the omitted variable; and if a deal pertains to a specific year, the dummy variable is 1 else it is 0. For instance, for the year 2003, all deals (data points) in the year 2003 have a dummy variable equal to 1 and all deals in the years other than 2003 have a dummy variable equal to 0. Similarly, since the thesis analyzes the change in subordination levels of all the bond classes, a dummy variable is created for the bond classes. All bond classes are compared to the AAA bond class which is the omitted variable in the regression. The dummy variable, D(tranche), with the values of 0 / 1 are used for the analysis. The following chapter discusses the results and the interpretation of the regression results. The spread differential between CMBS and corporate bonds is not used in the regression but is analyzed in the following chapter. 35 Chapter 4 DATA ANALYSIS The objective of the thesis is look at possible explanations for falling subordination levels in CMBS. The hypothesis is that the risk of the loans in the mortgage pools has gone down, thus giving room for lower subordination levels. Specifically, the proxy used to indicate if a loan is risky or not is the loan to value ratio and the debt service coverage ratio. These variables have been considered as they emerged as two important factors after a review of the default and loss severity literature on commercial mortgages. Pool level diversification has also been considered as increased diversification helps to mitigate risk. The relationship between the subordination levels and all the independent variables explained in Chapter 3 is discussed below. 4.1 Subordination Levels The first task in the analysis was to examine the subordination levels over the past decade for all the tranches in the commercial mortgage backed securities issued in this period. The average subordination level was calculated for each of the bond classes. A significant change in the structure of the AAA tranches has taken place since late 2004. CMBS issuers have started carving the senior triple-A tranches of many offerings into “super-senior” and, more recently, “super-duper” tranches that have higher credit support levels than the ratings agencies have reached for traditional senior tranches (sample deal in Exhibit 2). Thus, the AAA tranche of the CMBS deal is further split into “super-duper AAA”, “super-senior AAA” and the “regular AAA tranches”. The regular AAA tranches have a subordination level in the range of 11% to 15%. The super-duper AAA tranche still have subordination levels in the range of 30%. Many in the industry attribute this to the growing influence of crossover investors17 from the corporate bond market who are less comfortable with today’s lower subordination levels. Since the “regular AAA” tranche represents the credit risk in the transaction (and the underlying pool of mortgages), to be consistent in comparison of the subordination levels of the AAA 17 Such investors typically do not have an in-depth understanding of the real estate market, and are drawn by the higher yields offered by the commercial mortgage backed securities. Many of these investors are also foreign investors. 36 tranches of the earlier years in the study, subordination levels of the “regular AAA” tranche are considered for 2005 and some of the deals in 2004. Table 7 summarizes the trend in subordination levels for all bond classes from 1995 through mid-2005. Table 8 provides a comparison of the subordination levels (including the percentage decrease) in1995 versus those in 2005. Table 7 AAA AA A BBB BB B 1995 33.50 26.82 21.23 15.00 6.68 3.08 1996 33.25 27.22 20.57 15.59 7.79 4.14 AVERAGE SUBORDINATION LEVEL 1997 1998 1999 2000 2001 2002 30.81 28.78 26.68 22.81 21.72 20.86 24.55 23.38 21.48 18.70 17.90 16.62 18.32 17.78 16.68 14.48 13.86 12.44 12.96 12.38 11.38 10.03 9.24 8.28 6.51 6.30 5.74 5.14 4.76 4.28 3.95 3.24 2.79 2.47 2.45 2.10 2003 19.68 14.65 10.27 6.20 3.66 2.11 2004 16.34 12.23 8.39 4.77 2.66 1.55 2005 13.16 10.12 7.18 4.08 2.36 1.54 Table 8 Rating 1995 2005 Difference % Change AAA 33.50 13.16 17.17 -61% AA 26.82 10.12 14.59 -62% A 21.23 7.18 12.84 -66% BBB 15.00 4.08 10.24 -73% BB 6.68 2.36 4.01 -65% B 3.08 1.54 1.53 -50% It is evident from the above that subordination levels for all the bond classes have fallen dramatically from1995 to 2005. The subordination level of the AAA tranche has fallen from 33.50% in 1995 to 18.94% in 2005. It is also interesting to note that the subordination level of the BBB tranche, the lowest investment grade tranche, has fallen by the highest percentage, 73%, from a level of 15% to a current level of 4.08%. The following graphs further elucidate the falling trend in the credit enhancement levels of the commercial mortgage backed securities. 37 Class A Subordination Levels 40.00 35.00 33.50 33.25 30.81 28.78 Subordination Level 30.00 26.68 26.82 27.22 25.00 24.55 22.81 23.38 21.72 20.86 21.48 20.00 18.70 21.23 20.57 18.32 17.78 15.00 19.68 16.34 17.90 16.62 14.65 16.68 14.48 13.86 10.00 13.16 12.23 10.12 12.44 10.27 8.39 5.00 7.18 1995 1996 1997 1998 1999 2000 AAA 2001 2002 2003 2004 AA 2005 A Class B Subordination Levels 18.00 16.00 15.00 15.59 12.96 Subordination Level 14.00 12.38 11.38 12.00 10.03 9.24 10.00 8.28 8.00 7.79 6.68 6.20 6.51 6.00 6.30 5.74 4.77 5.14 4.76 4.00 4.14 2.00 4.08 4.28 3.66 3.95 3.24 3.08 2.66 2.79 2.47 2.45 2.10 - 2.11 1.55 2.36 1.54 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 BBB BB B 4.2 Discussion of the Regression Results The next step in the analysis was to analyze the subordination levels with respect to the loan to value ratio with an expected positive relationship implying a positive co-efficient for the loan to 38 value ratio in the regression result. Similarly, the relationship with the debt service coverage ratio is expected to be an inverse relationship. A summary of the results of the regression analysis is presented in the following table. The regression model has an R square of 0.9326 adjusted R square of 0.9320. This implies that the model explains 93% of the variability in the subordination levels. The detailed regression output is shown in Exhibit 3. Variable Regression Co-efficient T-Statistic WALTV 0.3917 43.27 WADSCR 2.3044 12.67 DAA -5.2178 -18.81 DA -9.7982 -35.22 DBBB -14.2459 -51.36 DBB -18.2441 -62.16 DB -20.3075 -67.86 DConduit -1.7266 -4.92 DNUM96 0.8139 1.20 DNUM97 -0.5939 -0.89 DNUM98 -1.7039 -2.63 DNUM99 -3.2126 -4.85 DNUM00 -4.2889 -6.43 DNUM01 -5.05515 -7.67 DNUM02 -6.4861 -9.76 DNUM03 -8.3123 -12.63 DNUM04 -9.6065 -14.87 DNUM05 -9.8814 -14.87 DFloating 2.4858 4.55 DDiverse -0.3821 -1.33 39 The regression results seem to suggest a positive relationship between the subordination levels and the weighted average loan to value ratio. This indicates that as the loan to value ratio increases, the subordination level of the bond classes is higher – for a unit increase in the loan to value ratio, the subordination level increases by 0.39. The loan to value ratio is statistically significant at the 95% confidence level in explaining the subordination levels. Surprisingly, there seems to exist a positive relationship between the debt service coverage ratio and the subordination levels i.e. with a unit increase in debt service coverage ratio, the required subordination increases by 2.30 (as per the statistical result). What is even more interesting is the fact that the variable is actually statistically significant in explaining the subordination level. This relationship as explained by the model may be ignored, as it also does not make intuitive sense. It should be noted that some of the research on the loan to value ratio and the debt service coverage ratio indicate that the latter is a better indicator of default. By that measure, the debt service coverage ratio should have had a negative co-efficient in the regression. This result of the regression could be attributed to the fact that the data on debt service coverage ratio, in its current form as collected, may not be appropriate and further research needs to be done. Another reason that may be responsible for a positive co-efficient for the debt service coverage ratio is the manner in which it may have been calculated. The data collected from Trepp represents the DSCR as calculated by the underwriters, and different underwriters make different assumptions to calculate the DSCR. The DSCR for some of the deals may be based on a cash flow figure accounting for reserves and other deductions, while other deals may be using the net operating income to calculate DSCR. This may cause a discrepancy in the DSCR reported across the deals. It may be more appropriate to use the rating agencies stressed DSCR (page 14 describes how S&P arrives at the net income to finally calculate DSCR). Also, the fact that the weighted average DSCR is being used instead of the full range of ratios for all the properties, which is more appropriate, may be a plausible explanation for the not so intuitive regression results. Finally, changing interest rates and cap rates over the period of study, which may influence the DSCR, have not been considered in this study. 40 As far as the effect of pool diversity on subordination levels is concerned, the regression yielded an insignificant (t-stat = -1.33) relationship between diversity and subordination levels. At the same time, it should be noted that most of the deals with very high subordination levels are not diversified (Exhibit1). However, the co-efficient of the variable seems to suggest that as diversity increases, the subordination levels fall. It should be noted that most of the CMBS deals considered in the data are diverse. If not so much across property types, they are well diversified as per geographic locations. Also, over the years, the composition of the diversity has changed. While the earlier deals constituted more multifamily, hotel and healthcare, the deals in the latter years mostly include office, retail and multifamily. Thus, there has existed a certain level of diversity in these transactions. As expected, the subordination levels are lower by about 1.72 if the deals are conduit as opposed to less common deals such as miscellaneous or franchise loan deals. Also, floating rate deals have a higher subordination level than the fixed rate deals. The subordination level seems to be higher by 2.48 if the CMBS transactions are floating rate deals. Both these variables are statistically significant in explaining the subordination levels. As far the dummy variables for the years are concerned, they seem to indicate that the subordination levels for all the bond classes have fallen with respect to the subordination levels of 1995. The variables are statistically significant for all the years except for 1996 and 1997 (where results indicate that the subordination levels increase in 1996 while those in 1997 dropped with respect to the 1995 levels). With respect to the AAA subordination levels, subordination levels for all other bond classes are lower. This is an expected result as the CMBS bonds are structured so that the tranches reflect the priority of payments and AAA being the highest rated tranche. As one moves down the deal structure, the subordination levels keep falling with the lowest tranche (B) having the lowest subordination level. All the variables are statistically significant. This indicates that the market structures the commercial mortgage backed securities efficiently. 41 Thus, based on the regression results above, loan to value ratio, conduit deals, interest rate type seem to affect the subordination levels of the bond classes in the CMBS. 4.3 Market factors affecting subordination levels It is evident from the above results that some of the factors affecting subordination include the loan to value ratio, the deal type (which depends on the type of loans in the underlying pool) and the interest rate type. However, in addition to the factors tested in the model above there may be other market factors that have played an important role in pushing down subordination levels to where they are currently. One of the explanations for the falling subordination levels is the increasing demand by investors for these kinds of securities. The public real estate debt market (primarily CMBS), is further integrating with the broader fixed income market, forcing greater conformity on the real estate industry. The historical performance of CMBS in recent years makes a strong case that risk has been overpriced i.e. spreads have been too wide and subordination levels have been too high for the realized risk reflected in the low delinquency rates. Typically, the yield on the investment grade CMBS tranches is quoted with respect to the spread on the swaps, while the yields on the non-investment grade CMBS tranches are quoted with respect to the spread on the US treasury. A swap spread is defined as the difference between the negotiated and fixed rate of a swap. The spread is determined by characteristics of market supply and creditor worthiness. The swap rate (the fixed rate) may be seen as the price for obtaining the floating rate payment. Hence the swap rate is the price which makes demand equal to supply in the plain vanilla market. Assuming a downward (upward) sloping demand curve (supply curve), an increase in demand for floating payments would raise the swap rate. An increase in supply of floating payments would lower the rate. For instance: in an environment of unusually high interest rates, economic agents may expect interest rates to decline, wanting to pay a floating rates instead of a fixed. The supply of floating payments would rise and the swap rate would fall, ceteris paribus. 42 The following two graphs show the yield spreads over the 10 year US treasury for all the CMBS bond classes on a monthly basis for the period August 1996 to July 2005. They indicate the tightening of spreads in the CMBS market. Investment Grade Spreads 350 300 250 200 150 100 50 AAA AA A Feb-05 Aug-04 Feb-04 Aug-03 Feb-03 Aug-02 Feb-02 Aug-01 Feb-01 Aug-00 Feb-00 Aug-99 Feb-99 Aug-98 Feb-98 Aug-97 Feb-97 Aug-96 0 BBB Source: Moody’s Investor Service Non-Investment Grade Spreads 1200 1000 800 600 400 200 BB Feb-05 Aug-04 Feb-04 Aug-03 Feb-03 Aug-02 Feb-02 Aug-01 Feb-01 Aug-00 Feb-00 Aug-99 Feb-99 Aug-98 Feb-98 Aug-97 Feb-97 Aug-96 0 B Source: Moody’s Investor Service In this study, to enable an apples-to-apples comparison of the spreads on CMBS and corporate bonds, the treasury spread is used. The following graphs show the trend in the yield spreads of 43 the AAA/AA18, A and BBB tranches of corporate bonds and CMBS. The spreads are with respect to the 10 year treasury. Corporate Bonds "AAA/AA" Yield Spreads CMBS 200 180 160 140 120 100 80 60 40 20 Apr-05 Dec-04 Apr-04 "A" Yield Spreads Aug-04 Dec-03 Aug-03 Apr-03 Dec-02 Aug-02 Apr-02 Dec-01 Aug-01 Apr-01 Dec-00 Apr-00 Aug-00 Dec-99 Aug-99 Apr-99 Dec-98 Aug-98 Apr-98 Dec-97 Aug-97 Apr-97 Dec-96 Aug-96 0 Corporate Bonds CMBS 300 250 200 150 100 50 18 Feb-05 Aug-04 Feb-04 Aug-03 Feb-03 Aug-02 Feb-02 Aug-01 Feb-01 Aug-00 Feb-00 Aug-99 Feb-99 Aug-98 Feb-98 Aug-97 Feb-97 (50) Aug-96 - Since the data for corporate bonds is a blended AAA/AA spread, the same has been calculated for the CMBS spreads. 44 "BBB" Yield Spreads Corporate Bonds CMBS 350 300 250 200 150 100 50 Feb-05 Aug-04 Feb-04 Aug-03 Feb-03 Aug-02 Feb-02 Aug-01 Feb-01 Aug-00 Feb-00 Aug-99 Feb-99 Aug-98 Feb-98 Aug-97 Feb-97 Aug-96 0 Source: Citigroup Capital Markets The spread differential between CMBS and corporate bonds indicates that commercial mortgage backed securities offer higher yields (as evident from the higher yield spreads of CMBS over the treasury), thus attracting investors. According to an industry source, most of the CMBS deals in today’s environment and market are oversubscribed by 50%. This may also explain the increasing number of corporate bond investors looking at the CMBS market for higher returns. This market trend implies a tremendous demand for CMBS, thus resulting in lowering of subordination levels. 4.4 Summary Thus, while the basic risk and diversification factors play some role in explaining the variability in subordination levels, the spread differential between CMBS and corporate bonds, compounded with the favorable interest rate environment and great property market performance seem to be affecting subordination levels. Various reasons have been cited for this trend. 45 One school of thought on the lower subordination levels is that, the subordination levels are probably not low and are just right. The reasoning behind such an argument is that the initial years of CMBS followed the debacle in the real estate markets of the late 1980s and the early 1990s. Most of the deals then included distressed properties and other such assets and thus, subordination levels were much higher then. Since then, the performance of the real estate markets has improved considerably due to low interest rates, rising property values, etc. Thus, since property level performance has increased, the underlying pools of mortgages are performing well, thus, justifying the lower subordination level for the securities issued today. At the same time, an industry source said that these are probably as low as the subordination levels can get. Other reports cite the “great performance” of the property markets. The robust capital inflows from the public and private debt markets, the current liquidity in the real estate capital markets, improving property market fundamentals, an expanding economy and generally lackluster investment opportunities elsewhere make real estate a lucrative investment option. Data from the real estate debt markets seem to show no signs of trouble. Delinquencies among seasoned CMBS (those aged one year or more) fell slightly to 1.78% in February, according to Morgan Stanley, and are expected to decline to about 1.30% by year-end. Loan delinquencies in life insurers’ commercial mortgage portfolios have also continued their downward trend. According to the American Council of Life Insurers (ACLI), loan delinquencies at year-end 2004 were barely measurable at just 0.08%. That said, delinquency and default data are, by definition, backwardlooking. And since loans rarely encounter problems in the first year or two, today’s low rates may not say very much about how the loans originated now will perform in the future. A recent report by Fitch19 addresses the concerns of lower subordination levels in CMBS. The market overall has benefited from the low interest rates and increasing values and CMBS transactions have benefited from improved loan and sponsor quality. The study states that the earliest CMBS transactions were conservative in many respects to attract investors. Early transactions consisted on distressed assets, often with little information on the collateral. Loans 19 Fitch Special Report, CMBS Subordination Levels: Too High, Too Low or Just Right, 2004 46 were not being originated for securitization, but rather securitization was undertaken as a means to clear balance sheets. Since then, CMBS performance has exceeded everybody’s expectations. Commercial real estate has also been systemized and disciplined through the capital markets. Flow of information has improved and detailed information of each property and its sponsor is available at origination. The lower subordination levels required by rating agencies are based on the outstanding performance of CMBS as well as intensive tracking of the bonds and the underlying collateral. Other factors such as low interest rates, inclusion of large loans in conduit structures, and improved collateral and sponsorship quality, have contributed to today’s lower levels. Thus, it can be argued, that while the more fundamental variables have an impact on in the subordination levels of the commercial mortgage backed securities, there are other market forces that impact subordination levels as well. For lack of time for the thesis, the other market forces could not be examined, and have potential for future research. 47 Chapter 5 CONCLUSION The objective of the thesis is to understand the determination of subordination levels in structuring commercial mortgage backed securities. Over the past decade, the subordination or credit enhancement levels of each of the bond classes have been falling. Through a quantitative analysis of 430 securities issued from 1995 through 2005, an attempt has been made to analyze the trend in subordination levels as well as to identify various factors that have contributed to this trend. The essence of subordination levels assigned to the various bond classes or tranches of the security is to ensure that the highest rated tranches are protected in the event of defaults and losses on the underlying pool of mortgages. Subordination is a means of mitigating the credit risk that exists in such a bond structure. The fundamental questions underlying the research of subordination levels are: have the underlying pools of loans become less risky so as to support the lower subordination levels or has the pool level diversity increased so as to justify the subordination levels? To answer the first question, loan to value ratio and debt service coverage ratio have been used to determine if the loan can be classified as risky or not. Additionally, the interest rate type and the nature of the deals (conduit, etc.) are also used as proxies for “risk” in the underlying pool of mortgages. Since subordination levels are determined by loan defaults and loss severity, the literature review makes an attempt to look at previous studies that have identified the variables that affect default and losses of commercial mortgages. Most of the research confirms the predictability of the contemporaneous debt service coverage ratio and the contemporaneous loan to value ratio as important variables. Additionally, the literature review also focuses on diversification benefits arising from both property type as well as geographic locations in real estate. Based on the factors that seem to influence probability of default and loss severity, and hence subordination levels, data were collected and analyzed. The results seem to indicate that over the 48 years, subordination levels have indeed dropped since 1995. Subordination levels seem to be affected by the loan to value ratio and other factors such as deal and interest rate type. However, more broadly, the average loan to value ratio has remained in the range of 60% to 70% while the subordination trends have declined. Thus, there seem to be other market forces that have influenced this downward trend. The thesis also looks at the difference in spreads between corporate bonds and commercial mortgage backed securities. The CMBS market has outperformed expectations and there has been an influx of investors into this area. The spread differential of the commercial mortgage backed securities over the corporate bonds seems to be luring the corporate bond investors to the CMBS market. This increasing demand for CMBS may be driving down the subordination levels. Another reason could be the time period for comparison of subordination levels. The CMBS market is a relatively nascent compared to its counterpart investments such as bonds and stocks. Also, the CMBS market emerged as a result of the real estate bust of the late 1980s and the early 1990s. The initial issues included loans on distressed assets, and other such loans which may have resulted in the higher subordination levels. Property market fundamentals have improved since, and this may be a reason for the lower subordination levels seen in today’s market. It may be likely that the ratings in the early years of CMBS were conservative, following the real estate cycle, and the current subordination levels are a reflection of the true inherent credit risk in the securities. Due to the time constraints of the thesis, the impact of the market forces on subordination levels could not be tested. Going forward, future research on the change in property market fundamentals, the composition of the underlying pools of mortgages (if the deals in the early and mid 1990s consisted of distressed loans) may throw some light on the reasons for this trend. 49 BIBLIOGRAPHY Ambrose, Brent W., and Sanders, Anthony B., Commercial Mortgage Backed Securities: Prepayment and Default, Journal of Real Estate Finance and Economics, 26:2/3, 179-196, 2003 Chen, Jun and Deng, Yongheng, Commercial Mortgage Workout Strategy and Conditional Default Probability: Evidence from Special Serviced CMBS Loans, February 2004 Ciochetti Brian A., Deng, Yongheng, Lee, Gail, Shilling, James D., Yao, Rui, A Proportional Hazards Model of Commercial Mortgage Default with Originator Bias, Journal of Real Estate Finance and Economics, 27:1, 5-23, 2003 Ciochetti, Brian A. and Shilling, James D., Default Losses and Commercial Mortgages, October 1999 Commercial Mortgage Alert, July 1, 2005 Eichholtz, Piet M.A., Hoesli M, MacGregor, Bryan D., Nanthakumaran N., Real estate portfolio diversification by property type and region, Journal of Property Finance, 1995 Esaki, Howard, “Commercial Mortgage Defaults: 1972 – 2000”, Real Estate Finance, Winter 2002 Esaki, Howard and Goldman, Masumi, Commercial Mortgage Defaults: 30 years of History, CMBS World, Winter 2005 Fabozzi, Frank , The Handbook of Mortgage Backed Securities Fabozzi, Frank J., and Jacob, David P., The Handbook of Commercial Mortgage Backed Securities, 1999 50 Fitch Special Report, CMBS Subordination Levels: Too High, Too Low or Just Right, 2004 Fitch Ratings Commercial Mortgage Special Report, US CMBS: Where Have All the Good Loans Gone?, January 10, 2005 Fitch Ratings Commercial Mortgage Special Report, Rating performing Loan Pools, November 4, 2004 Grissom, Terry V., Kuhle, James L., Walther, Carl H., Diversification Works in Real Estate, Too, Journal of Portfolio Management, Winter 1987, Vol. 13, Issue 2 Moody’s Special Report, CMBS: Moody’s Approach to Rating US Conduit Transactions, September 2000 Moody’s Special Report CMBS: Moody’s Approach to Rating Large Loan / Single Borrower Transactions, July 2000 Moody’s Special Report, US CMBS 1Q 2005: Another Warning Light on the Credit Dashboard, April 28, 2005 Moody’s Special Report, US CMBS: DSCR Migration and Contemporaneous Probability of Default, June 7, 2005 Nomura Fixed Income Research, Commercial MBS 2005 Outlook / 2004 Review, December 16, 2004 Pendergast, Lisa and Jerkins, Eric, CMBS Loss Severity Study: Portfolio Theory Aside, Size Matters, CMBS World, Spring 2003 Quigg, Laura, Default Risk in CMBS Bond Classes, 1998, Trend in Commercial Mortgage Backed Securities – Frank J Fabozzi (editor) 51 Standard & Poor’s Structured Finance, “CMBS Property Evaluation Criteria”, January 2004 Staples, Ed, Fitch Probes Commercial Mortgage Loss Factors, Real Estate Finance Today, 1996, Vol.13, Iss.25, p16 Taylor, Michaell and Fuchs, Randy, Diversification Benefits from Geographical Dispersion in Multifamily Mortgage Portfolios, Boxwood Means Inc., 2004 Vandell, Kerry D., Predicting Commercial Mortgage Foreclosure Experience, AREUEA Journal vol. 20 (1992): 55-88 Wheeler, Darrell and Chua, Gilbert, The Growth of the CMBS Market: An Overview, CMBS World, Spring 2004 Asset Securitization Report, CSFB offers first ever HEL Loss Severity Model; Notes “Darwinian Theory” of Sector, Impact of LTV, Vol 1, Issue 41, p16, 2001 52 Exhibit 1 Diversity of bonds with a High AAA Subordination Level Sr No Year 1 1995 2 1995 3 1995 4 1996 5 1997 6 2001 7 2003 8 2003 9 2003 10 2003 11 2003 12 2003 13 2004 14 2004 2004 15 2004 16 2004 17 2004 18 2005 19 2005 20 1995 21 1995 22 1996 23 1996 24 1996 25 1998 26 2002 27 2003 28 2003 29 2004 30 2004 31 2004 32 2004 Deal CUSIP CSFB 1995-M1 CSFB 1995-WF1 PSSF 1995-MCF2 SBM7 1996-C1 MSC 1997-ALIC JPMCC 2001-A COMM 2003-FL9 CSFB 2003-TF2A CSFB 2003-TFLA HVBMC 2003-FL1A JPMCC 2003-FL1A MSC 2003-XLF BSCMS 2004-BBA3 BSCMS 2004-HS2A BSCMS 2004-HS2A CGCMT 2004-FL1 COMM 2004-HTL1 LBFRC 2004-LLFA COMM 2005-F10A CSFB 2005-TFLA DLJMA 1995-CF2 NASC 1995-MD3 DLJMA 1996-CF1 PMLIC 6-PML SASC 1996-CFL CCAO 3A BASST 2002-X1 BALL 2003-BBA2 GSMS 2003-FL6A CSFB 2004-TF2A CSFB 2004-TFLA GCCFC 2004-FL2A MSC 2004-XLF Deal Type Miscellaneous Conduit Conduit Conduit Seasoned Miscellaneous Large Loan Large Loan Large Loan Seasoned Large Loan Large Loan Large Loan Large Loan Large Loan Large Loan Large Loan Large Loan Large Loan Large Loan Conduit Large Loan Conduit Seasoned Seasoned Miscellaneous Miscellaneous Large Loan Large Loan Large Loan Large Loan Large Loan Large Loan Fixed / Floating Fixed Fixed Fixed Fixed Fixed Fixed Floating Floating Floating Floating Floating Floating Floating Floating Floating Floating Floating Floating Floating Floating Fixed Fixed Fixed Fixed Fixed Fixed Fixed Floating Floating Floating Floating Floating Floating 53 Rating AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA AAA Subordination 35.01 35 37 38 43 49.25 41.88 41.98 43.29 48.62 38 48.79 47.27 56.32 41.76 61.22 40.3 41.51 44.88 42.7 35.99 35 37.02 35 41 37 39.75 42.89 41.27 41.1 42.59 42.1 42.86 Diversified (y/n) n n n n n n n n n n n n n n n n n n n n y y y y y y y y y y y y y Exhibit 2 Sample of a Multiple AAA Tranche Deal Source: Trepp Class A1 A1P ADP A2 A3 ABA ABB A4 A4A A4B A1A AJ B C D E F G H J K L M N O P XP* XC* Total Balance Original 100,000,000 50,000,000 166,616,000 349,848,000 288,705,000 207,259,000 29,609,000 500,000,000 1,171,595,000 167,371,000 169,634,000 300,060,000 65,013,000 35,007,000 75,015,000 40,008,000 55,011,000 45,009,000 40,008,000 20,004,000 20,004,000 20,004,000 10,002,000 10,002,000 10,002,000 55,011,486 3,831,315,000 4,000,797,487 4,000,797,486 GS Mortgage Securities Corp. II 2005-GG4 (GSMS 2005-GG4) BOND SUMMARY Credit Enhancement Original Ratings Original Ratings Original Ratings Securitized MDY S&P FII Coupon Coupon Type 4.369 Fixed Rate 20 Aaa AAA AAA 5.285 Fixed Rate 20 Aaa AAA AAA 3.452 Fixed Rate 20 Aaa AAA AAA 4.475 Fixed Rate 20 Aaa AAA AAA 4.607 Fixed Rate 20 Aaa AAA AAA 4.68 Fixed Rate 20 Aaa AAA AAA 4.756 Fixed Rate 20 Aaa AAA AAA 4.761 Fixed Rate 20 Aaa AAA AAA 4.751 Fixed Rate 20 Aaa AAA AAA 4.732 Fixed Rate 20 Aaa AAA AAA 4.744 Fixed Rate 20 Aaa AAA AAA 4.782 Fixed Rate 12.5 Aaa AAA AAA 4.841 Fixed Rate 10.88 Aa2 AA AA 4.89 Fixed Rate 10 Aa3 AAAA4.939 Fixed Rate 8.13 A2 A A 5.078 Fixed Rate 7.13 A3 AA5.415 Fixed Rate 5.75 Baa1 BBB+ BBB+ WAC/PT-4bp 4.63 Baa2 BBB BBB WAC/PT-0bp 3.63 Baa3 BBBBBB4.462 Fixed Rate 3.13 Ba1 BB+ BB+ 4.462 Fixed Rate 2.63 Ba2 BB BB 4.462 Fixed Rate 2.13 Ba3 BBBB4.462 Fixed Rate 1.88 B1 B+ B+ 4.462 Fixed Rate 1.63 B2 B B 4.462 Fixed Rate 1.38 B3 BB4.462 Fixed Rate 0 NR NR NR IO, Other Non-Fixed Aaa AAA AAA IO, Other Non-Fixed Aaa AAA AAA * - Notional Balance 54 Exhibit 3 Regression Output Source SS df MS Model Residual 521776.97 37688.55 20.00 2267.00 26088.85 16.62 Total 559465.53 2287.00 244.63 subordinat~n Coef. waltv wadscr daa da dbbb dbb db dconduit dnum96 dnum97 dnum98 dnum99 dnum00 dnum01 dnum02 dnum03 dnum04 dnum05 dfloating ddiverse Std. Err. 0.3917 2.3044 -5.2179 -9.7982 -14.2459 -18.2441 -20.3075 -1.7266 0.8140 -0.5940 -1.7040 -3.2127 -4.2890 -5.0552 -6.4861 -8.3123 -9.6066 -9.8814 2.4858 -0.3821 0.0091 0.1818 0.2774 0.2782 0.2774 0.2935 0.2992 0.3507 0.6811 0.6705 0.6471 0.6624 0.6674 0.6590 0.6645 0.6581 0.6460 0.6646 0.5460 0.2874 t Number of obs F( 20, 2267) Prob > F R-squared Adj R-squared Root MSE P>t 43.2700 12.6700 -18.8100 -35.2200 -51.3600 -62.1600 -67.8600 -4.9200 1.2000 -0.8900 -2.6300 -4.8500 -6.4300 -7.6700 -9.7600 -12.6300 -14.8700 -14.8700 4.5500 -1.3300 55 [95% Conf. 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2320 0.3760 0.0090 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.1840 0.3740 1.9479 -5.7618 -10.3438 -14.7899 -18.8196 -20.8944 -2.4144 -0.5216 -1.9089 -2.9730 -4.5116 -5.5978 -6.3474 -7.7893 -9.6029 -10.8734 -11.1847 1.4150 -0.9456 2287 1569.27 0.0000 0.9326 0.9320 4.0774 Interval] 0.4095 2.6610 -4.6739 -9.2526 -13.7020 -17.6685 -19.7207 -1.0389 2.1496 0.7210 -0.4350 -1.9138 -2.9801 -3.7629 -5.1829 -7.0218 -8.3397 -8.5781 3.5566 0.1814