Standard & Poor\u2019s Leveraged Commentary and Data (LCD Comps) hints at these phenomena in its data set of BB\/B-rated leveraged loans. Figure 1<\/strong> shows cumulative default curves by tranche size and reveals that smaller tranches default at a far lower rate than larger tranches. We believe our explanation that smaller borrowers with fewer lenders tend to negotiate around payment defaults, bankruptcy, and distressed exchanges accounts for much, if not all, of this difference. (See the appendix for other explanations.)<\/p>\n FIGURE 1 CUMULATIVE DEFAULT CURVES BY TRANCHE SIZE<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/p>\n<\/div>\n Recognizing the limited applicability of the ratings agency definition of default to the specifics of middle-market direct lending, CA broadened the definition of default to include: (1) all material modifications of loan documents; (2) PIKing not<\/em> at the borrower\u2019s option (i.e., excluding PIK toggle structures); and (3) cessation of accrual of interest and distressed covenant waivers. In general, material modification refers to the \u201csacred rights\u201d of credit documents\u2014to wit, any term or condition that affects yield and which requires unanimous lender approval (e.g., term, interest rate, amortization, commitment, etc.). CA\u2019s database, therefore, does not track actual defaults as material loan modification, which we interpret as evidence of general credit stress in a portfolio. Put another way, lack of material modifications in loan documents usually, but not always, implies a healthy borrower that can make its interest and principle payments in a timely manner and in compliance with all existing covenants. Naturally, the rate of stress in our sample will exceed default rates observed in the BSL market, and this has significant implications for recovery analysis (see the appendix). Nevertheless, we believe it offers a more searching calculus to underpin prudent capital allocation. We will refer to modifications and credit stress interchangeably.<\/p>\n However, we recognize that our approach also has drawbacks. Just as the ratings agency definition may render false negatives, our definition of stress may yield false positives: instances qualifying as stress where in fact none or very little exist. For example, the unexpected opportunity to purchase a competitor or a new factory may require both a capital expenditure covenant waiver and an amortization holiday. Similarly, a borrower slated for sale just prior to a loan\u2019s impending maturity may see the sales process stalled through no fault of its own, requiring an immediate extension of the maturity pending resolution of the obstacles to the sale. These events would require material modifications to a credit document that would be caught in our definition of credit stress, when in fact the borrower may be performing to plan or better.<\/p>\n Understanding that stress can mean almost anything from outright business failure to virtually immaterial documentary changes is vital to reading our analysis. The broad definition is, therefore, perhaps most helpful in its counterfactual: loans experiencing no reported credit stress very likely performed to or above plan at underwriting. When reading the stress rate analysis below, the reader should consider this alternate perspective.<\/p>\n When comparing recoveries, we used publicly available information from Moody\u2019s because they calculate recoveries based on trading price and recoveries based on ultimate recovery. The former is calculated as the discounted (at the coupon rate) trading recovery price as a percentage of the original par value. The latter seeks to identify actual recoveries and is \u201cthe value creditors realize at the resolution of a default event. For example, for issuers filing for bankruptcy, the ultimate recovery is the present value of the cash or securities that creditors actually receive when the issuer exits bankruptcy, typically one to two years following the initial default date.\u201d<\/span> 2<\/a><\/sup> <\/p>\n CA recognizes that replicating this level of detail for middle-market loans is impracticable. As a result, we gathered data reflecting the total amount of principal collected excluding interest and fees. CA further recognizes that senior debt funds calculate losses and recoveries differently and sought to implement a standard method with minimal scope for manipulation. Our loss and recovery rates, therefore, exclude any recovery from interest and fees.<\/p>\n Another difference between our method and that of the ratings agencies is their focus on individual instrument recoveries. While this is theoretically the best way to aggregate recovery data, we believe that collecting this level of data from senior debt funds would prove onerous. As a result, we calculate losses and recoveries on aggregate vintages, generating a directionally accurate average. Vintage losses are calculated by dividing the par value of losses incurred by a vintage by the aggregate reported par value of that vintage.<\/span> 3<\/a><\/sup> Recovery rates are calculated by subtracting that rate from one.<\/p>\n We hope to provide an estimated range of recoveries for middle-market loans and to compare them to the information provided by CRAs compare the relative risk of loss for BSLs and middle-market loans.<\/p>\n Importantly, CA did not audit the data provided and relies solely on what was reported by cooperating senior debt funds. As a result, we rely on the honesty and forthrightness of participating senior debt funds. Our interaction with these lenders, their detailed questions, desire for elaboration, and specification of our methods and criteria lead us to conclude that they are trustworthy partners in this exercise.<\/p>\n As noted, we received data on first-lien senior loans, including unitranche loans, from 11 senior debt funds totaling 2,728 loans with $61.7 billion in aggregate par value and average original par value of $22.6 million. Loans were categorized by origination year and then tracked by stress year and loss year. Our data set shows aggregate loans issued in each year from 2002 to 2017 and aggregate stressed loans and losses of each vintage. CA can therefore report, for example, total loans issued by number and par value in 2002 (the 2002 \u201cvintage\u201d) and total number and par amount of stressed loans and losses of that vintage in years 2002 through 2018.<\/p>\n When calculating credit stress, we rely on loan count, and when calculating losses, we rely on value. We believe that this reflects the maxim that borrowers default and instruments recover. In addition, this approach comports with that of LCD Comps, which offers a very comparable methodology and data set.<\/p>\n While we believe that our total sample size of loans is robust, we recognize that it represents a small sample of the entire universe of middle-market loans. Moreover, we recognize that the reporting funds create two biases. First, some funds that declined to participate may fear that their performance is poor relative to peers. If that fear is valid, then their absence improves the overall data set. (We do not suggest that non-participating senior debt funds all have inferior modification and loss experience\u2014merely that the possibility exists.) Second, reporting funds in existence prior to 2008 create a survivorship bias. In other words, we do not have data from those funds that did not survive the global financial crisis (GFC).<\/p>\n In Figure 2<\/strong>, our findings and sample size are broken out by vintage. We analyzed the data by vintage and across all vintages on an annual and cumulative basis. We also divided the data into cohorts by vintage depending on the likelihood that the loans would have lasted through the GFC.<\/span> 4<\/a><\/sup> <\/p>\n FIGURE 2 RESULTS BY VINTAGE YEAR<\/strong><\/p>\n Source: Cambridge Associates LLC. Figure 3<\/strong> shows the cumulative credit stress rate for all of the loans in our sample. In generating this curve, we employed the same arithmetic approach as LCD Comps in generating the curves presented above: the cumulative observations of stress (by borrower count) for each year are divided by 2,728, the total number of loans made from 2002 to 2017. Recognizing that the entire sample includes loans from 2013 to 2017, which have not seasoned, we show a curve with vintages from 2002 to 2012.<\/p>\n FIGURE 3 CAMBRIDGE ASSOCIATES CUMULATIVE “CREDIT STRESS” RATE<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/p>\n<\/div>\n The curve clearly shows that the incidence of stress<\/em> is far higher in our sample than the highest default<\/em> of 4% in the LCD Comps sample used to generate Figure 1. We hesitate to compare actual ratings agency default incidence to our incidence of stress because our approach should capture everything from a benign documentation change described above all the way to liquidation.<\/p>\n We believe the best reading of this curve concludes that approximately 85% of total borrowers in the seasoned cohort did not seek and were not granted material loan modifications by year 10 and therefore experienced little to no credit stress.<\/p>\n Within that cohort, the incidence of material modifications ranged from 6.4% (2012) to 22.2% (2007). By comparison, LCD Comps reports default rates ranging from 0.7% (2009) to 12.0% (2007). As expected, the implied rate of credit stress exceeds default rates. However, we believe that the data suggest that credit stress, broadly defined by material modifications, occurs less frequently than many may believe, affecting one in five borrowers at the height of the GFC (i.e., the 2007 vintage). The appendix further breaks out each individual vintage, as well as cohorts of vintages relative to their position before, during, and after the GFC.<\/p>\n We recognize that stress rates do not answer the burning question of how much a senior debt fund can expect to lose. We note the aggregate losses in Figure 2 and the par value weighted loss curves are presented below. Figure 2 shows that recovery rates in the event of stress can range from 100% to as little as 34% (in the 2009 vintage) and that vintages have historically lost between 0% and 10% of their aggregate principal balance. We further note that these loss rates were not experienced by particular fund vehicles.<\/span> 5<\/a><\/sup> In the case of 2009, 14.1% of the total portfolio encountered stress and 9.3% of the total portfolio was lost. Our loss-given-stress calculation divides the loss rate by the stress rate to show that approximately two-thirds of the value of stressed (or modified) loans<\/em> were lost.<\/p>\n The stress rate here is critical when considering loss given stress. Two of the three worst recovering vintages, 2009 and 2010, raise practitioners\u2019 eyebrows because these vintages should have offered the best opportunities to lend, yet their loss-given stress levels are very high, and their implied recoveries are very low. However, Figure 4<\/strong> shows the relationship between stress and loss is critical when deriving loss estimates. For example, in 2010 a greater percentage of modified loans incurred losses, even if absolute losses were low. This may be attributable to the fact that fewer loans made in 2010 struggled at all (suggesting a healthy credit environment), but those that did struggled mightily, with losses of $109 million on $168 million of modified loan value.<\/p>\n FIGURE 4 MODIFICATION RATES AND LOSS RATES OVER TIME<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/p>\n<\/div>\n A comparison of 2009 and 2004 sheds further light on the importance of the relationship between modifications and losses. For 2009, our sample shows 52 loans made with six modifications (total value of $120 million) generating a stress ratio of 11.5%. Four of those loans, however, incurred losses of $79 million, or 9.3% of total par value. By comparison, 2004 saw 45 loans made with nine incidents of stress, a rate of 20%, almost double that of 2009. Losses in the 2004 vintage were 8.9%, roughly in line with 2009. The big disparity between stress rates generates a very large difference in loss-given stress.<\/p>\n There are a couple possible explanations for this phenomenon. For 2009, one may simply be sample size. The 2009 vintage had one of the lowest loan counts in the sample, exposing it to greater variation of outcomes. Another may be that some loans may have been committed to later in 2008 and so were made before the full force of the GFC impacted borrowers. The 2010 phenomenon is more difficult to explain. The majority of losses in this vintage were actually incurred in 2017 ($59 million of $109 million), more than six years after origination. Loans tend to sour in the first two to three years after origination. It is possible that this vintage may have been overly exposed to sectors that deteriorated later and for reasons unrelated to the GFC (e.g., energy, retail, etc.). We would, therefore, suggest that investors focus on gross losses rather than losses as a percentage of stressed assets.<\/p>\n Furthermore, our database offers certain insights into losses that can help investors form an opinion about the risk of loss in middle-market loans. We frequently hear concerns that middle-market companies can just \u201cgo away,\u201d leaving lenders with little or no recovery. Our data hint at this risk. At the same time, there is more direct evidence of robust recoveries. For example, of the three vintages reporting one loan loss, all recovered more than 98% of principal. While we realize that middle-market companies, not unlike their larger peers, can \u201cjust go away,\u201d we resist the commonly held belief that their disappearance is the norm.<\/p>\n Building on our previous work \u201cOrigination Year Defaults: A Canary in the Credit Coal Mine?,\u201d which demonstrated the importance of vintage even among identically rated loans, we broke out loss curves by origination year for our senior debt fund loans. As expected, the vintages with the highest cumulative loss rates are 2006 through 2009 because they are clustered around the GFC. The 2004 vintage is particularly interesting, as losses spiked in 2012 and 2013 to 2008 peak levels, which is likely related to small sample size (Figure 5<\/strong>).<\/p>\n FIGURE 5 CUMULATIVE LOSS CURVES BY TOTAL VALUE OF LOANS<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/span><\/p><\/div>\n Figure 6<\/strong> categorizes all the vintages into four cohorts: (1) before the GFC, 2002\u20132004, whose loans likely seasoned before 2008; (2) through the GFC, 2005\u20132008, whose loans were made just prior to the GFC and therefore were serviced during the GFC; (3) during the GFC, 2009, which were made when the crisis was at its worst; and (4) post-GFC, 2010\u20132013. CA recognizes that some of these loans may still be outstanding and could still incur losses.<\/p>\n FIGURE 6 LOSS CURVES BY TOTAL VALUE OF LOANS: GFC RELATIVE COHORTS<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/span><\/p><\/div>\n Finally, CA attempted to create a model portfolio of loans in our data set to simulate the actual year-to-year performance (Figure 7). We did this by chronologically adding each reported annual par value to the net sum of the previous year\u2019s existing outstanding loan balance, less actual losses in that year, and estimated repayment. Annual losses rose to approximately 4% in the teeth of the GFC and then declined as old loans repaid and were replaced by new, unseasoned, performing loans.<\/p>\n The simulation in Figure 7<\/strong> does not guarantee performance for senior debt funds and is based on assumptions that may be faulty. However, it helps to frame an analysis of senior debt fund performance and provides a superior analytical lens compared to individual loan losses.<\/p>\n FIGURE 7 SIMULATED LOSS RATE<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/p>\n<\/div>\n Finally, we formulated a hypothetical \u201cworst case\u201d scenario senior debt fund with a five-year investment period from 2005 to 2009, the years leading into and through the GFC. We further assumed that the losses happened immediately, generating no interest or amortization to cover losses and forcing the \u201cfund\u201d to rely on the performing loans to cover losses and generate returns.<\/p>\n Figure 8<\/strong> suggests that despite lending into and through the GFC at a small spread over LIBOR, a senior debt fund would very likely not have lost LP capital at the portfolio level (as denoted by the multiples of less than 1.0x). These loans would have probably generated a safe, if unspectacular, return on invested capital of approximately 1.1x at the portfolio level. Moreover, if these loans were made at the average prevailing one-month LIBOR rate with no spread (i.e., L+0.0% coupon), the performing loans\u2019 interest could have compensated for the losses incurred in 2005, 2006, and 2007 (when average one-month LIBOR calculated on a daily basis was 3.3%, 4.9%, and 5.1%, respectively), and those vintages could also have compensated the portfolio for losses incurred in 2008 and 2009 (when average one-month LIBOR calculated on a daily basis was 2.6% and 0.3%, respectively). Our analysis does not forecast or guarantee performance of senior debt funds through the next credit cycle. Instead, it is meant to strongly suggest that LPs would run a very low risk of losing capital invested exclusively through one of the worst economic downturns of the last century.<\/span> 6<\/a><\/sup> <\/p>\n FIGURE 8 PERFORMANCE OF SENIOR DEBT FUNDS THROUGHOUT THE GFC<\/strong><\/p>\n Source: Cambridge Associates LLC. The analysis confirms our belief that senior debt funds have historically demonstrated resilience in the face of economic stress and have offered LPs a low volatility, yield-generating investment opportunity. We believe that many of these attributes will persist. However, we also recognize that deterioration in loan terms, higher leverage, and other pernicious developments in the middle market will erode that historic resilience. Nevertheless, there are some senior debt fund managers that can negotiate the treacherous currents within the middle market and could emerge with performance similar, if not superior, to historic observations. \u25a0<\/p>\n We posit that smaller tranche sizes exhibit lower default incidence because the loans are made by smaller groups of lenders that can alter documents to accommodate borrowers and thereby avoid defaults. Another plausible explanation for this profound difference could simply be that the borrowers of tranches less than $100 million are as large as those borrowing more, but are less levered and therefore less likely to default. LCD data do not appear to disclose borrower size or leverage level. Based on experience and discussions with senior debt fund managers, we believe that our explanation is more likely.<\/p>\n APPENDIX FIGURE 1 CUMULATIVE DEFAULT CURVES BY TRANCHE SIZE<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/p>\n<\/div>\n As with larger loans, vintage can play an important role. The curves suggest that through the GFC, 80% of loans by number of loans and 75% of loans by total value of loans exhibited little to no financial stress.<\/p>\n We repeat the caveat that while we view our sample size of loans in terms of number and outstanding principal as generally robust, we recognize that the reporting universe of 11 senior debt funds is small relative to the outstanding population. We also realize that the majority of reporting funds did not lend through the GFC.<\/p>\n APPENDIX FIGURE 2 CREDIT STRESS<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/p>\n<\/div>\n Concerns raised over distortions wrought by small sample sizes do not appear to plague our data. The figure below shows that instances of stress do not correlate to the number of issued loans.<\/p>\n APPENDIX FIGURE 3 REPORTED STRESS RATES AND NUMBER OF LOANS ISSUED<\/strong><\/p>\n Source: Cambridge Associates LLC.<\/p>\n<\/div>\n While we include losses by number of loans, we suggest focusing on losses by value. The data underscore the importance of vintage in assessing loss probability, but also show that staggered vintage deployment mitigates losses.<\/p>\n What does it mean for recoveries when stress exceeds defaults?<\/em><\/p>\n The cornerstone calculation for any credit investor is the product of the probability of default (PD) and the loss-given default (LGD), which yields expected loss (EL).<\/p>\n PD X LGD = EL<\/p>\n In our sample, we use \u201closs-given stress\u201d as a proxy for LGD, but in practice it would be very difficult for a loan to lose principal without suffering a default as defined by the ratings agencies in the BSL market.<\/p>\n Our analysis notes that the probability of credit stress (PCS) exceeds PD observed in the BSL market and suggests that BSL EL (ELBSL<\/sub>) is broadly in line with middle-market EL (ELMM<\/sub>).<\/p>\n PDBSL<\/sub> X LGDBSL<\/sub> ~ PCS X LGSMM<\/sub><\/p>\n Expected losses should not change. However, if we remove the parts of credit stress that fall short of actual defaults, then PCS should decline in the equation above. For the identity to survive, LGDMM <\/sub>must increase.<\/p>\n![\"FIGURE](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-Fig-1-new.png\")
<\/a><\/p>\nLoss Definition\u2014Very Basic<\/h3>\n
Caveats and Methodology<\/h3>\n
Findings, Results, and Conclusions<\/h2>\n
![\"FIGURE](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-Fig-2-new.png\")
<\/a><\/p>\n
\nNote: The Estimated Vintage Recovery Rate is calculated as 1 \u2013 (Loss Rate by Value\/Stress Rate by Value).<\/p>\n<\/div>\nCredit Stress Analysis<\/h3>\n
![\"FIGURE](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-Fig-3-new.png\")
<\/a><\/p>\nMeasuring Losses: The Challenge of Vintage Data<\/h3>\n
![\"FIGURE](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-Fig-4-new.png\")
<\/a><\/p>\nInsights from Vintages<\/h3>\n
![\"FIGURE](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-Fig-5-new.png\")
<\/a>![\"FIGURE](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-Fig-6-new.png\")
<\/a>Model Portfolio Creation and Simulation<\/h3>\n
![\"FIGURE](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-Fig-7-new.png\")
<\/a><\/p>\n![\"FIGURE](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-Fig-8-new.png\")
<\/a><\/p>\n
\nNotes: Excludes management fees and carry and impact of fund level leverage. Average one-month LIBOR calculated daily from 2005\u201309 averaged 3.25%.<\/p>\n<\/div>\nConclusion<\/h3>\n
Appendix<\/h2>\n
LCD Default Curves: Alternative Explanations<\/h3>\n
![\"APPENDIX](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-App-Fig-1-new.png\")
<\/a><\/p>\nCurves Broken Out by CA Data Set Vintage<\/h3>\n
![\"APPENDIX](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-App-Fig-2-new.png\")
<\/a><\/p>\nSample Size<\/h3>\n
![\"APPENDIX](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-App-Fig-3-new.png\")
<\/a><\/p>\n
![\"APPENDIX](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-App-Fig-4-new.png\")
<\/a><\/p>\n![\"APPENDIX](\"https:\/\/publishedresearch.cambridgeassociates.com\/wp-content\/uploads\/2019\/07\/2019-07-Recovery-DB-App-Fig-5-new.png\")
<\/a><\/p>\n