Multi-tranche securitisation structures: more than just a zero-sum game?

ABSTRACT

This paper explains multi-tranche structuring and the yield that securitisation bonds offer by incorporating several factors into a comprehensive model. Results indicate that the degree of complexity of multi-tranche securitisation structures is related to market completeness and solving information asymmetry problems. We also find that the complexity of multi-tranche structure enables the yield offered by triple-A bonds to be reduced but not the average yield, concluding that tranching is a zero-sum game. This research uses a database comprising of all the MBS and ABS issues (1993–2011) in Spain, one of the world’s main securitisation markets. Analysing this long period has allowed us, for the first time, to contrast the Great Financial Crisis (GFC) disruptive effect on the analysed relationships in the securitisation market.

KEYWORDS:

• Securitisation
• primary yield
• tranche
• asymmetric information
• Great Financial Crisis

JEL CLASSIFICATIONS:

• G12
• G21
• G24

Acknowledgments

A great many specialists and institutions ranging from universities to the world of finance have provided us freely with assistance in securing data and documents, and in resolving occasional issues: Eduardo Blanco (Deputy Head of Statistics at the CNMV), Carmen Barrenechea (Head of Intermoney Titulización, SGFT and a member of the Executive Committee of ESF), Mari Luz Garrido (University of Vigo), Mercedes Pérez and Andrés Alaña (Treasury and Capital Markets Dept. of Kutxabank), Pilar Fernández-Ferrín (UPV/EHU), etc. Thanks are also due for comments received from Ramiro Losada (Studies Technician), Elías López (Deputy Head of Studies) and María del Rosario Martín (Dept. of Studies, Statistics and Publications) at the CNMV (Spanish Securities and Exchange Commission, Madrid, June 2014). Finally, we thank the anonymous reviewers and the editor for their careful reading of our manuscript and their many insightful comments and suggestions. The findings and conclusions published here are the sole responsibility of the authors. This research was supported by FESIDE Foundation and UPV/EHU under Grant NUPV12/01.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Securitisation Data Report Q2-2012 – Association for Financial Markets in Europe (2012).

2. The generation of high-quality bonds is supported by the flexibility associated with this financial technique, which enables issues to be structured into several differentiated series of bonds subordinated one to another. This means that certain subordinate (or ‘equity') tranches take up most of the risk, thus releasing other ‘senior' tranches and making it easier for them to obtain the highest ratings.

3. However it is mainly the originators who hold on to most risky series (equity tranches), in line with US and European regulations, which require the withholding of the equivalent of 5% of the total risk (Guo and Wu 2014; Benendo and Bruno 2012).

4. Ashcraft and Schuermann (2008) detail the frictions associated with information asymmetries in the securitisation chain and indicate that, according to Moody’s, the standard of management by those responsible of funds may influence actual losses by as much as 10%.

5. In spite of the controversy sparked in the wake of the subprime crisis as regards the reliability of ratings, most studies opt to use the average rating awarded by CRA as an ex ante proxy to measure risk in issues, so it can be considered as a correct indicator of inherent risk (Peña-Cerezo, Rodríguez-Castellanos, and Ibáñez-Hernández 2015; Bodenstedt, Rösch, and Scheule 2013; Rösch and Scheule 2012).

6. According to fund diversity, all securitisation issues in Spain take place with a single type of collateral. Therefore, for each type of collateral (dummy variables are used to identify them) the main form of fund diversification is size, because the greater the volume is, the more extensive the geographical, demographic and activty base of the original loan is. The size of securitisation funds can be larger for either of two main reasons: (1) because the organisation involved is large and operates in many territories, sectors or markets (in these cases the loan portfolio comes from a single organisation), or (2) because the loan portfolio transferred to a securitisation fund contains loans assigned by many financial organisations. Both lead to the same result: greater diversification of the portfolio through greater size.

7. FTA-CM = Multi-seller asset securitisation funds. These funds are excluded because of the following specific characteristics: (i) they are not backed by a specific asset pool but instead by all the originatoŕs mortgage-backed assets; (ii) they are the only ones that offer a constant coupon in the majority of cases; (iii) they are not set up on the basis of multi-tranche structures; and (iv) they do not generate series of bonds with clearly differentiated yield/risk profiles, as evidenced by the fact that they always obtain the highest (AAA/Aaa) rating. FTA-CM are therefore considered as atypical securitisation structures (Carbo-Valverde, Degryse, and Rodríguez-Fernández 2015).

8. Because in these cases the primary yield of the securitisation bonds issued in association with such funds, measured as the differential over the benchmark interest rate, is only known at the time of their setting up and not at any time subsequent to their issue. It must be noted that not counting FTA-CM, only six such funds were set up, for a total amount of 4,873 million euros.

9. The protection effect generated by junior tranches cannot be observed in securitisation funds with a single tranche because there are no junior tranches. Not counting FTA-CM and funds that do not offer variable yield, only 27 ABS and MBS-type single-tranche funds were set up with private credit institutions as their originators, for a total amount of 19,283 million euros.

10. The design of securitisation operations in Spain has been subordinated to the interests of wholesale investors, present in the process of negotiation and setting up conditions (road shows prior to issues) which prioritise the acquisition of riskless bonds (though they may offer a low yield premium) rather than more speculative bonds, which are generally held back by the originators as a sign of quality (Uhde and Michalak 2010). This has led to an effort to maximise the relative weight of AAA tranches in Spain.

11. Thus, one variable (in our case Ntranches) can act as an explanatory (or independent) variable in regard to the yield of issues (YieldAAA or YieldAVE) and at the same time as a variable to be explained.

12. The results of the Mann-Whitney U test shows that on average YieldAVE exhibits lower levels in the pre-crisis period (Mn = 21.01, Mdn = 20.00, SE = 9.79) than during the crisis (Mn = 51.07, Mdn = 42.00, SE = 27.95, U = 2178, p-value < 0.01, r = −0,7971). Similarly, YieldAAA also exhibits lower levels in the pre-crisis period (Mn = 16.92, Mdn = 15.00, SE = 10.10) than during the crisis (Mn = 39.74, Mdn = 30.00, SE = 25.81, U = 2900, p-value < 0.01, r = −0.7801).

13. Paradoxically, in 2008–2011 the yield premium associated with the bonds with the worst categories (lowest ratings) were, on average, lower than those for 1993–2007, whereas in categories with ratings higher than Baa3/BBB the yield premium (differential) in 2008–2011 was higher than in 1993–2007.

14. Other control variables linked directly or indirectly to risks for issues (and issuers) and to the quality of the ratings awarded are the following: Nrating (Number of rating), Guarantee (offered by a public institution), Types of collateral, Types of originator, Nissuers (Number of issuers or originators), etc.

15. In the parsimonious models (4, 8, 12 & 16) maximum correlation levels between the explanatory variables are given in the pairs Size&Nrating (0.65) and Size&Ntranches (0.41) in the pre-crisis period, and in the pairs Size&Guarantee (−0.39) and MTC&Maturity (0.35), in the crisis period. The average of the correlations between all the explanatory variables considered (in absolute values) is 0.224 and 0.128 in the pre-crisis and crisis period, respectively.

16. As a further analysis to validate the model, a simple cross-validation is carried out by splitting the data into two subsets. In particular, we have chosen to carry out two types of simple cross-validation. First, the population analysed was randomly divided into two subsamples, with the adjustment level (R2) of the refined models (4, 8, 12 and 16) in both samples being calculated. It is observed that in the pre-crisis period the stability of the models is acceptable (maximum difference in R2 lower than 4.5%), while in the post-crisis period the model loses stability (maximum difference in R2 higher than 14%). Second, the population analysed is randomly divided into two subsamples: training sample vs. test sample, with each of the four refined models in the training sample being tested and the coefficients obtained being saved. Subsequently, using these coefficients, the expected value of the dependent variables in the sample test is calculated and compared with the observed values. From these two series (observed values vs. estimated values) the Pearson correlation coefficient (r) is obtained. Following this second method, similar results are obtained: in the pre-crisis period the stability of the models is acceptable (maximum difference in r less than 6%), while in the post-crisis period the model loses stability (maximum difference in r greater than 20%).

Additional information

Funding

This work was supported by University of the Basque Country (UPV/EHU): [Grant Number NUPV12/01]; FESIDE Fundation: [Grant Number Securitization13/01/25].