Direct Lending Returns

I examine the performance of US business development companies (“BDC”). BDCs have produced returns in line with those of private funds engaged in direct lending. Leveraged loan and small-cap value equity returns explain a significant part of BDC performance, and the alpha of BDCs is zero on a market-value basis but a statistically significant 2.74% per annum based on net asset value (NAV) valuations. I find no evidence of an illiquidity premium, which suggests that the alpha could result from regulatory arbitrage or a peso problem. Cross-sectional BDC returns are widely dispersed and exhibit strong persistence in top- and bottom-quartile manager performance.

T he size of the private credit market consisting of investment companies, funds, and other structures providing corporate debt financing outside of banks and liquid capital markets has almost quintupled in size since the financial crisis of 2007-2009 and exceeds USD 1 trillion of assets as of 2021. 1 The growth of the market, alongside other alternative assets characterized by varying degrees of illiquidity, complexity, and attractive headline returns have caught the attention of investors and regulators alike.Relative to the importance of private credit, little academic research has been published on the market to date.This article examines the returns, risk exposures, and performance persistence of business development companies (BDCs) engaged in direct lending to middle-market and lower-middle-market companies in the United States.Publicly listed BDCs, which are SEC-registered, closed-ended investment companies, provide an interesting source of data on the returns, risks, and premia available in private credit markets to complement the analysis of the performance of private credit funds that is typically based on net asset values and cash flows reported to commercial databases.Specifically, my data set enables investigation into factor exposures in direct lending beyond univariate performance indicators, such as public market equivalent (PME) commonly used in private asset research.Factor-based analysis is useful in capturing not only the multidimensional return drivers in direct lending but also investment company-level leverage that is a common feature in both BDCs and other private credit investment vehicles.
I construct a sample of 47 listed BDCs over the period December, 2009, to June, 2022, with around $110 billion of aggregate invested assets as of year-end 2021.For each BDC, I create a time series of total returns including dividends based on, first, market value (as observed in the stock market) and, second, reported net asset value (NAV, based on company 10-K and 10-Q filings.)The net asset value series could be considered similar in nature to fund NAV data used in previous private asset studies.I then calculate two market capitalization-weighted indices illustrating the average performance of BDC market values and NAVs, respectively.I also gather data on BDC BDCs link closely to the nascent body of literature on private debt performance.In a pioneering study, Munday et al. (2018) report an internal rate of return (IRR) of 9.7% and Sharpe ratio of 1.55 (adjusted for return autocorrelation) for direct lending funds in 2004-2016 vintages based on data in the Burgiss database.They find a statistically significant quarterly alpha of 1.80-2.00% in univariate regressions of direct lending fund returns against high-yield bonds, BDC market price index, and leveraged loans.B€ oni and Manigart (2022) examine the Preqin private credit database and discover direct lending IRR of 8.78% per annum over the sample period 1996-2020.They also report mean PMEs of 1.04 and 1.05 against investment grade and high-yield bond indices, respectively.Furthermore, the authors discover statistically significant performance persistence in follow-on fund performance, particularly in the top and bottom quartiles of funds.
More broadly, this study builds on previous research of private asset performance that has focused mostly on private equity; a comprehensive overview of the literature is provided in Doskeland and Str€ omberg (2018).Kaplan and Schoar (2005) present the PME as a measure of private equity performance and report significant dispersion in manager returns as well as evidence of performance persistence.Harris, Jenkinson, and Kaplan (2014) find private equity outperformance of more than 3% per annum compared to public markets.Brown and Kaplan (2019) provide supporting evidence with more recent performance data.L 'Her et al. (2016); Ilmanen, Chandra, and McQuinn (2019); and Stafford (2021), among others, suggest more muted relative performance once higher leverage and factor exposures in private equity are properly accounted for.Asset pricing tests using common "style" factors are more difficult to conduct on the infrequent cash flow-based data commonly encountered in private equity research, but a handful of studies have nevertheless sought to address the issue.Anson (2013) finds a growth bias and mixed evidence of a size effect in private equity index data, whereas Jegadeesh et al. (2015) discover a statistically significant size effect but an insignificant value factor loading in private equity fund of funds and listed private equity returns.Ang et al. (2018) report a statistically significant positive loading on the Fama-French value factor as well as the Pastor and Stambaugh (2003) liquidity factor (but a statistically insignificant size loading) in buyout funds.Stafford (2021) uses listed equity data to create portfolios replicating PE buyout returns and finds statistically significant exposure to the size and value factors.
This article contributes to the existing literature on several fronts.First, the results provide a comprehensive look at the returns in the BDC sector using a battery of asset pricing tests and cross-sectional analysis.The findings support the use of leveraged loan data as a starting point for direct lending analysis, adding to the inaugural study by Munday et al. (2018).Furthermore, this study is, in my knowledge, the first one to establish the important link between equity factors and direct lending returns.Also, while long statistically significant market factor lags and resultant aggregate market betas have been found in private equity and real estate research (e.g.Anson, 2013 andMladina, 2018, respectively), they have not been extensively reported in prior private credit studies.The lack of evidence of funding or market liquidity impacting direct lending returns casts some doubt on the common notion of private credit as a source of an illiquidity premium and thus contributes to broader private asset research.Linking my results to recent banking and finance literature, I propose regulatory arbitrage profits as a plausible explanation for direct lending outperformance.Finally, I find evidence of persistence in BDC manager performance in line with earlier results from Kaplan and Schoar (2005) in private equity, and B€ oni and Manigart (2022) in private credit, and suggest several conduits through which manager skill, scope, and experience may manifest themselves in realized returns.
These results have practical relevance for portfolio construction, risk management, manager selection, and performance attribution in private credit assets.I propose a framework for the estimation of realistic bounds for direct lending risk based on BDC market values.As far as investment policy and asset allocation are concerned, the wide dispersion in BDC returns suggests that market-average direct lending returns may be a poor guide to actual investment outcomes with specific managers.

BDCs and Their Relevance for Private Credit
Business development companies are SEC-registered, closed-ended investment companies providing both debt and equity financing to middle-market 2 and lower-middle-market companies in the United States.BDCs are required to invest at least 70% of their assets in non-public equity and debt of US corporations.Eligible investments also include US government securities, cash, and listed securities of companies with a market capitalization of less than $250 million.BDC investment holdings are subject to diversification requirements.Companies owned by private equity funds are an important segment of BDC borrowers.BDCs finance themselves through equity and debt issuance in public and private capital markets, and the market consists of both publicly listed and private companies.BDCs often raise capital initially as a private company and then convert into a publicly traded entity on a later date.The focus of this article is exclusively on publicly listed BDCs.
As discussed in the next section, senior secured loans form the majority of BDC portfolio assets today.BDCs may also hold subordinated debt as well as equity warrants and direct equity ownership in their portfolio companies.Ninety percent of a BDC's income must be derived from dividends and interest, and the companies are required to provide managerial assistance to their portfolio companies for the duration of ownership.Under the Investment Company Act of 1940 (40 Act), BDCs are allowed to use up to 1:1 leverage (debt-to-equity).The Small Business Credit Availability Act, passed in 2018, further permits BDCs to increase leverage up to 2:1 debt-to-equity subject to specific corporate governance criteria.In order to qualify as a regulated investment company (RIC) under the 1986 Internal Revenue Code enabling pass-through taxation, 90% of the BDC's taxable income must be distributed to shareholders.Consequently, high dividend yields reaching double digits are a salient feature attracting income investors to BDCs.
BDCs may be internally managed, when the company directly employs the executives responsible for the management of the investment operation, or externally managed, when the BDC contracts a third-party manager to operate the investment business.Either way, the managers are typically compensated through a combination of fixed and incentive-based management fees (percentage of net interest income and realized gains.) 3   Several observations suggest that BDCs are representative of the wider universe of private credit and especially of funds engaged in middle-market direct lending.First, Section 17d of the 40 Act prohibits registered investment companies and private funds of the same manager from co-investing in the same securities unless exemptive relief is granted by the SEC.As noted in Nesbitt (2019), almost all managers that operate a BDC request and are granted 17d relief, enabling their BDCs to co-invest alongside the manager's private credit funds.Consequently, he explains that "the 17d exemptive relief also means that BDC loan assets will look very much like the assets found in BDC managers' private fund offerings.This also means that investors can look to the performance of the BDC as indicative of private fund performance." 4  Second, the above observation by Nesbitt (2019) is supported empirically by the analysis in Munday et al. (2018), who find the Cliffwater Direct Lending Index (CDLI) to be the closest performance benchmark for private direct lending funds.The CDLI is an asset-weighted index that seeks to measure the unlevered, gross of fees performance of US middlemarket corporate loans, as represented by the underlying assets of BDCs.The CDLI is calculated based on financial statements and other information contained in SEC filings of BDCs. 5   Third, data in the recent private debt survey by Block et al. (2023) further illustrates the relevance of BDCs for private credit.Forty-five percent of their US respondents manage a BDC (possibly alongside other private debt funds), and the median BDC manages approximately 1.5 times the amount of assets in private debt compared to non-BDCs.BDC respondents also report longer experience in private debt investment than non-BDCs (14.3 vs. 11.5 years).The authors note that most of their other survey results are similar for BDCs and other types of private lenders.
Fourth, as further anecdotal evidence, a recent industry report of the private credit market 6 lists the current top-ten private debt funds.Eight of the top-ten managers operate (or have operated) a public BDC, seven of which are included in the sample of this study.BDC data is also used as a publicly available indicator of developments in the wider private credit market. 7  In terms of relative market size, as of year-end 2020, the public BDCs in my sample had approximately $90 billion assets under management.Overall, funds engaged primarily in direct lending (the main activity of BDCs) managed $260 billion at the same time (plus almost $150 billion of dry powder.) 8Thus, BDCs represent a meaningful proportion of the aggregate assets in direct lending.

Data
I collect data on 47 publicly listed BDCs (as reported in Table A.1 in the Appendix), with a combined equity market capitalization of $51.1 billion and total assets of $112 billion as of year-end 2021.The average length of the market value time series of the BDCs during the full sample period (December, 2009, to June, 2022) is 9.6 years and the median 10.2 years, ranging between 2.3 and 12.5 years.Eight of the 47 BDCs were acquired, delisted, or withdrew their election to be treated as a BDC during the sample period.Further details of the sample selection are provided in the Appendix.
The choice of sample period is driven by three main considerations.First, the majority of public BDCs were only listed post-2010 (see Figure A.1 in the Appendix), and the aggregate market value of the BDCs in the sample only reached $10 billion in Q2 2010.The industry then experienced rapid growth with the number of BDCs in the sample increasing from sixteen in 2009 to forty in 2014.Second, as will be further detailed in the following section, the asset allocation of BDCs has changed after the financial crisis from a majority in equity and subordinated debt to a majority in senior debt (see also Davydiuk, Marchuk, and Rosen, 2022a).As the emphasis of this study is on direct lending, the post-crisis period forms a more representative sample.Third, direct middle-market corporate loans are "level 3" assets subject to fair value accounting in accordance with Financial Accounting Standards Board ASC 820 (previously, FAS 157) introduced from 2006 onward.Prior to ASC 820, non-traded assets such as private loans were valued at historical cost adjusted for write-ups or write-downs. 9Consequently, the accuracy of BDC NAV data (which I rely on in some of the forthcoming analysis) in the pre-ASC 820 period may be questionable.Notably, ASC 820 was effective for entities with fiscal years beginning after November 15, 2007.Hence, the methodology was in effect fully adopted for annual financial statements by December 2008, a point in the middle of the global financial crisis that represented the trough of BDC valuations and market value discounts to NAV.In other words, inference from a sample starting in December 2008 would be distorted by the exceptional circumstances prevailing during the financial crisis.
BDC prices, dividends, and market capitalization are sourced from Eikon Datastream and total assets and liabilities data from Compustat (Wharton Research Data Services.)As BDC net asset value data in common financial databases contain gaps and inaccuracies, I have collected quarterly NAV per share data from 10-K and 10-Q filings of the 47 individual BDCs reported in the EDGAR database.Similarly, BDC portfolio allocations, average portfolio yields, as well as fees and interest expenses have been obtained from EDGAR filings.
Equity and leveraged loan benchmarks and ETF data are sourced from Refinitiv Eikon, and bond index data from Bloomberg.The Fama-French factor data sets are sourced from Professor French's website. 10 The Pastor and Stambaugh (2003) liquidity factor is sourced from Professor Stambaugh's website, 11 and the He, Kelly, and Manela (2017) intermediary capital factor from Professor He's website. 12Descriptive Statistics.improvement in BDC loan seniority evidenced previously in Figure 1.The average BDC yield spread over leveraged loans during the sample period was 5.7%, ranging between 0.9% and 7.9%.For comparison with the broader private debt market, Block et al. (2023) report an average target unlevered IRR of 8.16% for US private debt managers in Q3 2021, whereas the market value-weighted average portfolio yield for the BDCs in my sample was 8.82% at that time.
Weighted average BDC debt to equity is shown in Figure A.3 in the Appendix.The ratio averaged 71% during the sample period, increasing almost monotonically since 2010.The increase in leverage has coincided with the more conservative asset allocation evidenced in Figure 1, and more recently it reflects the relaxation of regulatory rules in the 2018 Small Business Credit Availability Act.Since 2019, the debt-to-equity ratio has averaged 102%, versus 61% in 2009-2017. For comparison, Block et al. (2023) report an average debt-to-equity ratio of 56% for non-BDC private credit managers and 79% for BDCs.
The average annual management fee expense (including incentive fees) over the sample period was 3.19% of total assets, corresponding to 5.46% of net assets assuming an average debt-to-equity ratio of 71%. 14 This compares with an average management fee of 3.14% per annum of net assets for private direct lending funds, as reported by Cliffwater (2022).The average financing expense was 4.36% per annum of total debt, or 3-month LIBOR þ 3.58% given an average LIBOR rate of 0.78% during the period. 15

BDC Market Performance
To illustrate the broad performance of the BDC universe I calculate two market capitalization-weighted indices reflecting the total return of the market values of the 47 listed BDCs (MVIndex), and their quarterly NAVs (NAVIndex). 16Both indices include reinvested dividends and are rebalanced annually to reflect market capitalization weights at previous calendar year end.Figure 2 shows the returns of these indices since December 2009.
The difference between the two indices reflects the average market capitalization-weighted premium or discount to NAV of BDC market prices.The figure shows the tendency of the broad BDC market to generally trade close to the NAV, except during times of market stress when the discount can widen substantially-best illustrated by performance during the 2020 pandemic.It is important to note that the index series represent the market value and NAV "as of" a given valuation date.As an example, the NAV of each BDC as of December 31, 2021, is reported alongside the BDC's market value as of December 31, 2021.In reality, the NAVs of the BDCs underlying the index are unknown on such dates and will only be published in the financial reporting of the BDC several weeks after the NAV date.I will follow this convention consistently throughout the study, as the resultant foresight bias will not affect the analysis.
Further statistics of the market value and NAV indices and various comparable benchmarks are presented in Table 1 and Figure 3 The BDC market value index produced an annualized total return of 8.63% and Sharpe ratio of 0.38 over the period, lagging the NAV index (9.41%and 1.73,  respectively 17 ) based on quarterly data.As shown in Figure 2, the MVIndex suffered a sharp drawdown in the first half of 2022.Over the period Q4/2009 to Q4/2021 (i.e.excluding the recent drawdown) the compounded returns of MVIndex and NAVIndex were 9.97% and 9.60% per annum, respectively.Consistently with the requirement to distribute at least 90% of income to investors, the positive total returns of the BDC indices are derived entirely from dividend income, as the market value and NAV returns excluding dividends (not reported in the table) were -1.53% and -0.61% per annum, respectively, over the full sample period.
Both the absolute and risk-adjusted performance of MVIndex were higher during the period 2010-June 2022 than in Beltratti and Bock (2018), who report an average annual (market value) return of 6.43% and Sharpe ratio of 0.19 for their BDC index in 2004-2016.However, similarly to their results, the MVIndex Sharpe ratio has lagged both high yield bonds and leveraged loans over the past decade despite the greater absolute return.On the other hand, the MVIndex Sharpe ratio is broadly in line with small-cap equities (Russell 2000) and regional bank stocks.
As far as the higher moments of the distribution, the difference in the volatility of BDC market value and NAV returns is notable.MVIndex volatility is broadly in line with that of small-cap equities, but lower than the volatility of regional banks.BDC returns exhibit more negative skewness and excess kurtosis than the equity benchmarks, but less than the leveraged loan index.NAVIndex and CDLI indicate even greater non-normality in their returns.
The correlation of BDC market value returns with both high-yield debt and leveraged loans-both natural comparisons a priori given the composition of BDC portfolios-is materially higher in my sample compared to Beltratti and Bock (2018) (0.85 vs. 0.65, and 0.92 vs. 0.43, respectively.)The increase in correlation with leveraged loans is particularly notable and consistent with the shift in BDC asset allocation toward senior secured debt assets.Regarding the equity market benchmarks, the MVIndex exhibits the highest correlation with the Russell 2000 index.
The BDC returns in Table 1 appear consistent with long-term private credit fund performance documented in previous research.Munday et al. (2018) report an average IRR of 9.7% for direct lending funds in their full sample period, 2004-2016.Furthermore, they calculate an excess return over cash of 8% for direct lending funds (vs.an excess return of 8.14% [market value] and 8.91% [NAV] for BDCs in Table 1.)In 2009-2016 vintages, the average IRR was 9.3%.B€ oni and Manigart (2022) show a mean IRR of 8.78% (median 7.9%) for direct lending funds in the longer sample period 1996-2020.

Market and Factor Exposures
In this section, I will examine the common factor exposures of BDC returns.One would expect BDC market value returns to have sensitivity to equity market returns as well as the size factor, given the characteristics of BDC investee companies.As far as the value factor, previous BDC research has not addressed the question, and the evidence from related private equity studies is mixed.Given their asset allocation, BDCs would be expected to have exposure to non-investment grade credit markets.As middle-market loans are typically secured and carry a floating rate of interest, leveraged loans which share these characteristics would appear to be the most relevant benchmark. 18These observations are supported by the high-level correlation data presented in Table 1.BDCs seek to add value through opportunistic investment in small or distressed firms.Their value proposition, like that of private equity, is based in part on active engagement with portfolio companies.Consequently, it is also interesting to understand whether the companies outperform passive benchmarks.when explaining BDC market value returns (MVIndex.)While the first two exposures are intuitive, the value loading is a novel finding that has not been discussed in prior studies.A plausible explanation for the value tilt stems from Campbell, Hilscher, and Szilagyi (2008), who find that distressed firms have high market betas but also high loadings on the size and value factors. 19  As illustrated in the second column, leveraged loans are also a statistically significant explanatory factor in a univariate regression.In the third column, the leveraged loan index is added to the Fama-French model.Leveraged loans retain a positive and statistically significant loading, the significance of the three FF3 factors is maintained, and the r-squared of the model rises from 72% to 80%.In column 4, I move to a parsimonious model where the equity market and style exposures are bundled and captured by the Russell 2000 Value index of small-cap value stocks.
In addition to simplifying the regression model, the single index has the benefit of being a traded, longonly investment, as opposed to the Fama-French size and value research factors which are constructed as long-short portfolios and are not themselves traded or investable.Combining the equity factors results in a marginal improvement in the r-squared of the model.Notably, the alpha term in the three regressions is consistently negative, although statistically insignificant.
For a further and arguably fairer examination of riskadjusted performance, column 5 presents the results of a regression of MVIndex on directly traded and fully costed factors, namely ETFs tracking small-cap value and leveraged loans indices. 20ETF excess returns are reported net of their management fees and thus represent a like-for-like comparison with BDCs that are similarly exchange-traded and expressed net of all fees.The r-squared of the regression is marginally lower, but the ETF returns remain highly statistically significant explanatory variables.The net-of-fees intercept term, which has a direct, implementable alpha interpretation, flips to mildly positive but remains statistically insignificant.
In Table 3 I turn to the factors explaining the performance of quarterly BDC NAVs represented by NAVIndex.In essence, this analysis is analogous to modeling the performance of private credit funds based on their quarterly NAV and cash-flow data.
Having established the small-cap value equity and leveraged loan indices as both economically justified (based on the BDC asset allocation data) and statistically significant variables with a high explanatory power of BDC market value returns, it is natural to use them also as a basis for the NAV analysis.As the first column illustrates, a simple regression of NAVIndex on contemporaneous market benchmarks explains 68% of NAV return variation and results in a beta of 0.57 to leveraged loans, a nonsignificant small-cap value equity beta, and constant term of 6.28% per annum.the variation in NAVIndex returns with a beta of 1.45 and an annualized intercept of -4.56%, reflecting the structural leverage and fee/expense drag of BDCs.It is important to emphasize that while the CDLI is a commonly used benchmark in private credit, it is derived from accounting valuations rather than traded market prices and is not directly investable.
One obvious question to consider is whether the equity market exposure in BDC NAVs is simply a reflection of the equity and warrant holdings of BDCs that was reported in Figure 1.I test this by examining the factor exposures of the CDLI that tracks middle-market loan assets.The results (which are not reported in detail) show that a parsimonious model of contemporaneous and lagged small-cap equity value and leveraged loan returns discovers a statistically significant small-cap value equity beta that sums up to 0.09 and a leveraged loan beta of 0.58 even in the CDLI returns.The r-squared of the model is 0.87, and the intercept is economically and statistically significant at 5.6% per annum.Again, omission of the equity factors would result in overstatement of CDLI outperformance over liquid markets.This analysis further supports the finding that equity market factors affect direct lending performance.
Private credit investments are commonly marketed as a source of an illiquidity premium that can be harvested by patient, long-term investors.If such compensation for lack of liquidity was indeed a driver of direct lending performance, one could expect observed returns to show some dependence on market-wide liquidity measures.I test the sensitivity of BDC market value and NAV returns on three liquidity proxies, namely changes in the 3-month Treasury-Eurodollar (TED) spread, which is a common measure of funding liquidity; the Pastor and Stambaugh (2003) market liquidity factor; and the intermediary capital factor of He, Kelly, and Manela (2017).The intermediary capital factor is derived from tions to the market equity ratio of primary dealers and, while not a pure liquidity measure as such, the authors show that the factor has significant explanatory power for the cross-sectional variation in returns across asset classes.
The results of the analysis which are not reported in detail show no statistically significant relationship between BDC market value or NAV returns and contemporaneous or lagged liquidity proxies when such variables are added to the baseline regressions (with small cap value stocks and leveraged loans as explanatory variables) presented in Tables 2 and 3. Thus, I fail to find robust evidence that direct lending returns are in part compensation for illiquidity risk.

Individual BDC Performance
This section moves from index-level analysis to individual BDC data.Table 4 gives an overview of the dispersion in BDC performance based on several market value-and NAV-based metrics.Panel A shows the average, median, minimum, and maximum of annualized market value and NAV total returns of individual BDCs, along with average returns within performance quartiles.The results are pooled across all BDCs in the sample over their individual return series of varying length and are not, unlike the MVIndex and NAVIndex, weighted according to market capitalization.Consequently, the return metrics would not be expected to correspond to those found in the earlier index-level analysis.All the average returns are lower than the corresponding medians, resulting from a small number of very poorly performing BDCs pulling down the equal-weighted average performance.For obvious reasons relating to data availability, most existing private fund literature has focused on persistence in the performance of follow-on funds from a given manager.The structure of the BDC database allows an investigation in line with the seminal article by Carhart (1997).In this section, I will calculate the annual performance of individual BDCs based on three measures presented previously, namely market value total return, NAV total return, and alpha based on market value excess returns over benchmark indices.Regrettably, given the infrequency of data (quarterly) combined with the statistically significant lag structure reported earlier, I cannot conduct the study meaningfully with NAV-based alphas.Table 5 presents the probabilities of each of the performance measures in a subsequent year being in quartiles 1-4 (based on cross-sectional ranking, first quartile being the highest performance), conditional on the quartile in the initial (previous) year.
Panel A of Table 5, which is based on market value total returns, shows a relatively equal distribution of outcomes.First-quartile funds are slightly more likely to either remain in the first quartile or fall to the bottom quartile in the subsequent year than to migrate to the second or third quartile.Fourth-quartile BDCs exhibit some above-average tendency to rise to the second or third quartile in the subsequent year,  BDC Performance and Characteristics.

Individual BDC Performance
BDC selection and portfolio construction is not the focus of this article, but it is nevertheless interesting to look at some possible drivers of performance.In Table 6, I examine five high-level BDC characteristics: BDC size measured by total assets, 22 risk profile proxied by debt-to-equity ratio and portfolio yield, expenses measured as the ratio of management expenses to total assets, as well as the market priceto-NAV ratio serving as market-based valuation measure.Panel A lists the descriptive statistics of these characteristics across the 47 BDCs.Panel B then reports the results of multivariate regressions of the BDC performance measures on the characteristics in Panel A. The five characteristics explain between 36% and 53% of the cross-sectional variation in the performance metrics, but market price-to-NAV is the only consistently significant characteristic.
Interestingly, the relationship between returns and price-to-NAV ratios result is not confined to just market value-based performance measures where a rising market price would tautologically result in a higher price-to-NAV.To illustrate, market price-to-NAV ratios and NAV alphas of individual BDCs are plotted in Figure 4. Two possible interpretations of the finding would be, first, that price-to-NAV is a signal of a quality-like factor that is rewarded.The second plausible explanation arises from two institutional details of the BDC market.BDCs that elect to be taxed as regulated investment companies must distribute at least 90% of investment income to shareholders annually, which severely restricts the companies' ability to accumulate retained profits.Furthermore, BDCs are not allowed to issue shares to the public below NAV without annual shareholder approval.Consequently, market price-to-NAV has direct implications on a BDC's ability to grow and acquire new investments. 23I present further results from panel of rolling annual BDC return metrics on BDC characteristics in the Appendix.The results are noisier and less uniform than those from the cross-sectional study, but generally confirm the link between price-to-NAV and returns.However, the statistical link disappears when BDC fixed effects are included in the model, suggesting that a high price-to-NAV coincides with outperformance by a select group of BDCs, as opposed to price-to-NAV valuations fluctuating over time in the cross-section.In some specifications, higher BDC assets and lower leverage are also associated with greater returns.Overall, the results of this section should be interpreted with some caution due to the small sample size and impact of outliers on the findings.

Discussion and Implications
Direct Lending Returns and Premia.My results support the findings of Munday et al. (2018) and B€ oni and Manigart (2022) on the performance of private credit funds, as over the sample period from 2010, BDC NAV excess returns have outperformed liquid benchmarks on a risk-adjusted, net-of-fee basis with an alpha of 2.74%.The implications for non-BDC private credit funds are accentuated by the difference of over two percentage points between the management fees of BDCs and those of institutional private credit funds, as reported in previous research.
Three observations should be made at this point.First, the result only holds for BDC NAV returns but not for market values that exhibit zero alpha from a statistical standpoint (consistent with the equilibrium framework discussed in e.g.Berk and van Binsbergen, 2017). 24 NAV returns are analogous to the data used in private credit fund studies and thus corroborate findings in earlier research.The fundamental question is which measure is the correct one-the noisy but observable and tradable market price, or the stable but potentially biased and stale NAV valuation?In the context of the present study, it could be argued that finite-life debt valuations are subject to less potential manipulation than quasi-perpetual assets such as equities or real estate, because in the absence of continuously abundant availability of refinancing, a systematic overvaluation of loan assets will eventually transform to realized credit losses.One could therefore interpret the NAV alpha-calculated with appropriate lags and using both credit and equity benchmarks-as a measure of the long-term fundamental value in direct lending, whereas the nonsignificant alpha of BDC market values could be seen as a manifestation of excess volatility in traded BDC prices. 25In other words, once traded in the listed market, BDCs adopt the volatility of common stocks and may deviate from their fundamental value due to changes in investor risk aversion and market liquidity.Lee, Shleifer, and Thaler (1991) link the variation of closed-end fund discounts to changes in investor sentiment, and further conclude that as a result of such changes in sentiment closedend funds are riskier than the assets they hold.In the context of BDCs, fluctuations of market values relative to published NAVs, which may be driven in part by investor sentiment, contribute to the higher volatility of MVIndex vs. NAVIndex, as illustrated earlier in Figure 2. The dynamics of such premiums and discounts to NAV in the BDC industry are an interesting topic for future research.
Second, the sample period of my study (and the live history of much of the broader private credit market) does not include a prolonged recessionary period characterized by a spike in defaults and severe tightening in liquidity and credit conditions.Hence, the evidence of outperformance (based on NAV data) could in part reflect a "peso problem," that is, the returns embody a compensation for exposure to a latent and hitherto unrealized risk factor and the true success of non-bank lending business models can only be evaluated over a much longer time horizon. 23third related observation is that proxies of funding and market liquidity, such as changes in the TED spread, the Pastor and Stambaugh (2003) market liquidity factor, or the intermediary capital factor of He, Kelly, and Manela (2017) do not explain direct lending returns in a significant manner.Consequently, the common characterization of private credit as a source of an "illiquidity premium" cannot be empirically validated from the data.Is there another plausible explanation for the apparent premium offered by direct lending?
In a recent study, Chernenko, Erel, and Prilmeier (2022) calculate that public middle-market firms pay on average 4.35% higher interest rates to non-bank lenders compared to bank loans; however, when adjusted for different borrower and loan characteristics, the spread reduces to 1.67%, and the difference is not explained by higher realized default rates.The authors conclude that non-bank lenders improve access to capital for riskier firms that are unable to borrow from traditional banks due to banking regulations, and that their lending technologies impact other non-price items such as loan covenants and the use of warrants.This argument is further supported by the findings of Campbell, Hilscher, and Szilagyi (2008), and my discovery of a value bias in BDC returns which is consistent with their reported characteristics (size and value tilt) of distressed firms.
Davydiuk, Marchuk, and Rosen (2022a) find a difference of a similar magnitude (4-5%) in the spread of BDC loan interest rates compared to traditional bank commercial and industrial (C&I) lending (unadjusted for differences in credit quality and loan terms.)The authors note that their results support the view that small and risky firms have been credit rationed following the financial crisis and BDCs have in part helped to fill the gap.BDCs may also offer their borrowers other valuable benefits, such as more flexibility, loan tailoring, quicker execution, and looser covenants, than those available from traditional lenders.
It is therefore plausible that non-bank lenders have been successful in earning superior risk-adjusted returns in the decade following the global financial crisis by filling a void created by tighter bank regulations and reduced risk-taking capacity of traditional lenders.Obviously, even if the characterization of private credit premia as essentially a regulatory arbitrage profit or compensation for the benefits brought to the borrower by superior lending technology is accurate over the recent history, the question remains how long such profits can endure given the flow of capital to the non-bank sector.
Direct Lending Analysis.Given the growth in private credit investing, the need for representative and robust benchmarks for performance monitoring and attribution, as well for risk management purposes, is very apparent.Previous research on other private asset classes has predominantly focused on IRR, return multiples, and public market equivalent (PME) versus selected liquid market benchmarks as key performance indicators. 24One problem with such measures is that they do not accurately separate genuine asset class or manager outperformance from the impact of leverage without careful adjustment of the chosen benchmark.As noted in, for example, Block et al. (2023), the use of fund-level leverage in private credit is a common practice, with average debt-to-equity ratios not very dissimilar to those of the BDC market.The second issue is the choice of appropriate benchmark for PME measures.Holdings-based analysis and statistical evidence from BDCs both suggest that leveraged loans might serve as the most relevant debt market benchmark for direct lending.Third, the traditional indicators do not adequately capture the possible multidimensional nature of direct lending factor exposures.This study has shown that BDC returns are dependent on both small-cap value equity and liquid leveraged loan performance.
Could BDCs act as a realistic proxy for the riskiness of other direct lending funds?Munday et al. (2018) characterize BDC market values as too volatile and mainly driven by equity market movements, and the CDLI as probably too stable to properly gauge private credit returns.One possible approach would be to consider the CDLI (adjusted for serial correlation in returns) as the lower bound for direct lending risk.On the other hand, BDC market value volatility adjusted for the average debt-to-equity ratio of the market could act as an upper bound (and a conservative measure for risk management purposes) for welldiversified, unlevered direct lending investments.As an example, Table 1 indicated that the de-smoothed CDLI volatility over the sample period of this study was 3.59%.The market-capitalization-weighted BDC market price index (MVIndex) had a volatility of 18.96%.Adjusting for the average 71% debt-toequity ratio over the period would give an upperbound estimate for unlevered direct lending volatility of 18.96%/1.71¼ 11.09%. 25  Similarly, BDC returns adjusted for leverage could be used as a realistic investable benchmark in regression-based return attribution of direct lending funds, especially since their holdings are presently much more representative of broader direct lending ness than during the pre-financial crisis era, which covers half of the sample of the seminal work by Munday et al. (2018).I leave this as a question for future research.
Manager Selection.The results on BDC performance persistence based on the NAV total return measure which is the closest comparison to IRR commonly used in private equity and private debt literature lend support to earlier findings of private fund performance persistence.Nesbitt (2019) characterizes the BDC market as two-tiered, with a group of strongly performing firms with shareholder-friendly management and large platforms at the one end, and another group of small boutique firms with unnecessarily high fees at the other.While I cannot establish a robust link between BDC performance and size (total assets), it is very conceivable that organizational experience, skill, and scope in deal origination, underwriting, negotiation of terms and covenants, debtor monitoring, and refinancing, restructuring, and workout processes are supported by economies of scale and result in the accumulation of investment success.The results of panel regressions of individual BDC returns further support the existence of superior skill (alpha) in some individual BDCs.In the BDC market, investors may have been able to identify the skilled managers, resulting in pricing at a premium to NAV, but not to the extent that the increased valuations would have eroded the cross-sectional performance differential.Again, attributes leading to superior investment performance appear as an interesting topic for future research.

Conclusion
I have presented evidence of US direct lending returns of listed business development companies, showing attractive, high-single-digit total returns that are commensurate with IRRs previously reported for private direct lending funds.Seen in the context of the significantly higher fee loads of BDCs compared to private funds, the results appear encouraging and supportive of further institutional investment in the sector.Linking the results to recent studies in the banking and finance literature, I suggest that recent direct lending returns might represent regulatory arbitrage profits arising from bank lending constraints following the financial crisis of 2007-2009 or a compensation for superior lending technology.Alternatively, private credit might be susceptible to a "peso problem."I discover sensitivity to equity market, size, and value factors in direct lending returns, and show that omission of such factors may lead to overstated estimates of outperformance in valuation-based returns.I also suggest the use of public BDC data in private credit risk analysis.Finally, the results show high levels of dispersion in individual BDC returns, as well as performance persistence by bottom-and top-ranked managers.
In addition to several practical applications, the evidence in this study brings up several potential ideas for future research.Subject to availability of robust data, the applicability of BDC-based measures as benchmarks of private credit funds could be examined further.One relevant issue that has been outside the scope of this study is the impact of fees, and specifically the asymmetry arising from incentive fees, on BDC returns.Manager performance persistence is another interesting topic that this study has highlighted.A closer analysis of BDC manager characteristics and portfolio composition as determinants of future returns should be of both academic and practical relevance.
16.Further details of the index construction and comparison with public BDC benchmarks are provided in the Appendix.17.The BDC NAV index, as well as CDLI, exhibit positive first-order serial correlation in quarterly returns.I adjust the Sharpe ratios of these indices in line with Getmansky et al. ( 2004).
18. Beltratti and Bock ( 2018) report a higher correlation for BDCs with high-yield bonds than with leveraged loans.I show in robustness tests in the Appendix that this result is specific to their sample that covers an earlier time period when BDCs were less exposed to senior direct lending.
19.The size and value factors.as well as BDC returns.could be all driven by a common latent factor, possibly related to interest rates.I exclude this possibility by examining the relevance of government bond returns as an explanatory variable for BDC returns.While BDCs have a statistically significant negative loading on bond returns in a univariate regression, the model r-squared is low at 0.10.Furthermore, the loading changes sign and becomes insignificant when the equity and leveraged loan factors are added to the regression (results not reported in detail.) 20.The indices tracked by the leveraged loan ETF are different from those used in the benchmark regressions (columns 2 and 3).The leveraged loan ETF tracks the S&P/LSTA Leveraged Loans 100 index referencing the largest and most liquid leveraged loans.Consequently, the differences between the benchmark and ETF regressions are not explained solely by ETF fees.As the leveraged loan ETF was listed only in February 2011, I backfill the data to December 2009 by deducting the ETF's management fees from the benchmark index returns over the pre-launch period.
21.The first, third, and fifth quarterly lags are statistically significant at the 10% level.The result is consistent with, for example, Anson (2013), who reports that statistically significant lagged equity returns of up to one year explain private equity performance, and Mladina (2018), who finds significant lagged REIT betas of up to eight quarters in private real-estate valuations.The choice of the reported lag structure is described further in the Appendix.
22.In the regressions, logarithm of total assets is used.
23. Davydiuk, Marchuk, and Rosen (2022b) present related analysis of the impact of the substantial reduction in institutional ownership of BDCs following their exclusion from major stock indices in 2014.
24.As a thought experiment related to the previous comment on BDC fees, if the returns of BDCs were adjusted upward by the 2.32% annual management fee difference between BDCs and private credit funds, the ETF regression in Table 2 would result in an economically but not statistically significant positive alpha of 2.91% p.a.On a gross-of-fees basis, the alpha is 6% p.a., which is economically very significant and statistically weakly significant (at the 10% level).This analysis is an approximation because the management fees are a combination of fixed and performance-based compensation.
25. Shiller (1981) discusses the concept of excess volatility in the context of the equity market as a whole.More recently, e.g.Rudin and Farley (2022) seek to reconcile the conundrum of the gap between risk-adjusted private and public equity returns.Note that an increase in the observed volatility of an asset will mechanically increase its estimated beta and decrease the alpha, assuming that the asset's correlation with the market factor is positive and the market factor return is positive in-sample.27.One important exception is Ang et al. (2018), who present an interesting methodology for deriving historical time series of returns from private equity cash flow data.
28.This result also relates to the ongoing debate around the role of illiquidity as a driver for private asset demand.Asness (2019) has famously suggested in the context of private equity that illiquidity and pricing opacity "may actually be a feature not a bug," and that investors may be willing to pay a premium for assets that offer smoothed and ostensibly stable returns.The difference between the lower and upper bounds of volatility quantifies the smoothing benefit in direct lending and gives one indication of what the volatility of private credit would be if it was subject to trading in the liquid markets.
29.The number is an approximation, as the data in the article is presented in chart format after the end of the sample period (from July 2022 to June 2023).
30.This BDC significantly underperformed the broader BDC market in the final six months of the sample (by 15%) but subsequently recovered two thirds of the underperformance after the end of the sample period to June 2023.respectively.indices are rebalanced annually at calendar year end according to prevailing market value weights of the components.BDCs that drop out of the sample as a result of a merger or withdrawal of BDC status are held in the index at the last observed market price or NAV until the end of the

Robustness Checks and BDC Factor Model Specifications
Representativeness of MVIndex of the BDC Market.Table A.2 presents a comparison of the returns of the MVIndex with two public BDC indices, namely the Cliffwater BDC Index and the S&P BDC Index.The Cliffwater BDC index (which, unlike the Cliffwater Direct Lending Index, is based on market prices of listed BDCs) is structurally closer to the MVIndex because it is market-capitalization weighted, whereas the S&P BDC Index applies a 10% cap on individual BDC weights.The table also shows the returns for two variations of the MVIndex.First, MVIndexEW represents the equally-weighted returns of the BDCs comprising the index.Second, MVIndexEx represents the returns of an index in which each year the largest BDC by market capitalization is removed from the index.These variations are motivated by the dominance of the BDC industry by a handful of very large BDCs, which obviously skews the market capitalization-weighted returns.At the same time, market capitalization weights arguably represent the most accurate picture in terms of practical implementation and investment opportunities in the sector.
Results from Extended Sample Period, Subperiods, and Alternative Specifications of the Regression Model.
The justification for the choice of the sample period in the main study is explained in the Data section of  The purpose of the third sample specification is to highlight the impact of the financial crisis on BDC returns.However, the results should be interpreted with caution, because BDC asset portfolios were considerably more focused on equities and junior debt during the pre-2010 era.
The fourth sample specification serves to demonstrate the impact of the increase in BDC leverage following the 2018 Small Business Credit Availability Act.The increase of leverage post-2018 is notable in the higher leveraged loan exposure, while the equity exposure is somewhat reduced.
Columns 2, 4, and 9 compare and contrast the results from an alternative factor model that uses high yield bonds rather than leveraged loans as the credit factor.While statistically significant, with slightly higher r-squared than the model with leveraged loans over the extended sample period from December 2009-June 2022, high-yield bonds lose their significance as explanatory factors for BDC performance in the final part of the sample.As discussed in the main study, leveraged loans most closely resemble the structure of core BDC asset, namely secured floating-rate loans.Despite the common credit risk factors behind both high yield and leveraged loan performance, returns of the two indices will inevitably differ at times of large interest-rate moves.
Lag Structure of NAVIndex Regressions.

Panel Regression of BDC Returns on Characteristics
Table A.5 shows the results of a panel regression of annual return measures of individual BDCs on company characteristics.

Figure 4 .
Figure 4. Market Price-to-NAV Ratio and Alpha from NAV Regressions on Benchmarks 26.I address this issue in robustness tests presented in the Appendix, where I examine MVIndex returns over a longer period from December 2004-June 2022.The total return of MVIndex in December 2004-June 2022 was 6.22% p.a. (5.1% p.a. in excess of T-Bills).During the five years including the financial crisis (December 2004 to December 2009) the total return was just 0.68% p.a. (excess return -2.13%).My main findings regarding BDC factor exposures and alphas are unaffected by the inclusion of data covering the period over the financial crisis.It is important to note, however, that BDC investment portfolios were significantly more focused on equity and subordinated debt during the period 2005-2009 and thus less representative of the current direct Direct Lending Returns Volume 80, Number 1 lending business model.I am to robustly extend the NAV analysis back in time due to the issues with NAV data quality prior to the adoption of ASC 820.

Figure
Figure A.1 illustrates the number of BDCs in the sample, along with their aggregate market values and total assets during the period December 2009-June 2022.A comparison with the comprehensive BDC market statistics in Davydiuk, Marchuk, and Rosen (2022a) suggests that the sample selection substantially captures the BDC market.The authors report approximately 26 50 public BDCs with aggregate assets of approximately $63 billion as of Q4 2017, whereas my final sample has 43 BDCs with $61 billion of assets at the same point in time.As a further data point, Nesbitt (2019) reports $55 billion of assets across 42 publicly listed BDCs as of Q4 2017.
Figure A.2 plots portfolio yield of the BDCs and Figure A.3 the debt-toequity ratio.Both series are weighted according to BDC market capitalization.Index Construction.The MVIndex and NAVIndex are calculated based on the total returns (i.e.including reinvested dividends) of individual BDC market values (monthly) and NAVs (quarterly),

Figure A. 1 .
Figure A.1.Total Assets, Market Value, and Number of Public BDCs in the Sample Q4/2004 -Q2/2022

Table 2
presents the empirical analysis, starting with the canonical Fama-French (1992) equity factor model.The equity market, size, and value factors all have positive and statistically significant loadings

Table 1 .
Performance of BDC Market Value and NAV Indices and Benchmarks December 2009-June 2022 Note: This table shows performance statistics of a market value-weighted portfolio of BDCs calculated based on market price return (MVIndex, Panel A and B) and net asset value return (NAVIndex, Panel B only) and selected comparables.All of the indices include reinvested dividends.TR per annum.is the total return of the index and ER is the return in excess of USD 1-month T-Bills.Adjusted SR is the Sharpe ratio corrected for first order serial correlation in the NAVIndex and CDLI returns.The correlation matrix of quarterly returns is presented at the bottom of Panel B. Equity and loan benchmark data is from Eikon Datastream, and the bond index and CDLI data from Bloomberg.BDC NAV data is from company 10-K and 10-Q filings sourced from EDGAR.Financial Analysts Journal | A Publication of CFA Institute

Table 2 .
Regressions of Monthly BDC Market Value Index Excess Returns on Fama-French (1992) and Traded Factors, December 2009-June 2022 Note: This table shows regressions of monthly excess returns (over 1-month T-Bill) of a market value-weighted portfolio of BDCs calculated based on market price return ("MVIndex") on Fama-French (1992) factors (column 1), the excess return of a leveraged loan benchmark (column 2), Fama-French factors plus the excess returns of leveraged loans (column 3), excess returns of benchmark indices (column 4) and traded ETFs (column 5).Mkt-RF is the equity market factor, SMB is the size factor, and HML is the value factor from the Kenneth H. French database.R2K Value is the Russell 2000 Value Index, and Leveraged Loans is the S&P/LSTA Leveraged Loan Index.The corresponding ETFs are the iShares Russell 2000 Value ETF and the INVESCO Senior Loan ETF.The benchmark and ETF data is from Eikon Datastream.Robust standard errors are in parentheses.Ã and ÃÃ indicate statistical significance at the 5% and 1% level, respectively.

Table 3 .
Regressions of Quarterly BDC Net Asset Value Index Excess Returns on Benchmarks, December 2009-June 2022 Note: This table shows regressions of quarterly excess returns (over compounded 1-month T-Bill) of a market value-weighted portfolio of BDCs calculated based on Net Asset Value return (NAVIndex) on contemporaneous and lagged excess returns of benchmark indices.R2K Value is the Russell 2000 Value Index, Leveraged Loans is the S&P/LSTA Leveraged Loan Index, and CDLI is the Cliffwater Direct Lending Index.CDLI is sourced from Bloomberg and other index data from Eikon Datastream.BDC NAV data is from company 10-K and 10-Q filings sourced from EDGAR.Robust standard errors are in parentheses.Ã and ÃÃ indicate statistical significance at the 5% and 1% level, respectively.
Alphas from regressions of BDC market value and NAV excess returns on small-cap value stocks and leveraged loans are presented in Panels B and C. For the market values, regressions with traded ETFs are shown alongside the benchmarks.Similar to column 2 of Table3, the regression of NAVs in Panel C includes five lags of small-cap value and one lag of leveraged loan returns.The median alphas in the BDC market value regression on ETFs as well as the NAV regression on benchmarks are positive and economically significant at 2.18% and 2.93%, respectively.While only three of the 47 market value alphas (regressed on ETFs) are weakly statistically significant at 10% level, 43% of the NAV alphas are positive and statistically significant at the 5% level or better, and 36% at the 1% level.The performance metrics in Table4are relatively highly correlated with each other, with an average Pearson correlation of 0.86.The highest correlation (0.99) is between the market value/benchmark and market value/ETF alphas, and the lowest (0.78) between market value/ETF alpha and NAV alpha.
(Kaplan and Schoar, 2005)e.Performance persistence is an important topic in private asset research.Unlike mutual funds, evidence from an early period of private equity funds(Kaplan and Schoar, 2005)and recently from private debt(B€ oni  and Manigart, 2022)shows persistence in manager returns, possibly supporting the existence of informed or skilled managers and (concave) economies of scale in asset management.
This table shows pooled return statistics for individual BDCs.Panel A summarizes the annualized market value and NAV total returns of the BDCs in the sample.Panel B shows the annualized alpha from regressions of monthly BDC market value excess returns (over 1-month T-Bill) on Russell 2000 Value Index and S&P/LSTA Leveraged Loan Index excess returns (Benchmark), and iShares Russell 2000 Value ETF and INVESCO Senior Loan ETF excess returns (ETF).Panel C shows the annualized alpha from regressions of quarterly BDC NAV excess returns (over compounded 1-month T-Bill) on contemporaneous and lagged Russell 2000 Value Index and S&P/LSTA Leveraged Loan Index excess returns.The total returns and alphas are calculated for each of the BDCs over the period December 2009-June 2022, or such shorter period where data for such BDC is available.BDC market value and other index data is from Eikon Datastream.BDC NAV data is from company 10-K and 10-Q filings sourced from EDGAR.
Kaplan and Schoar (2005)ound in earlier research by B€ oni and Manigart (2022) for private debt funds, orKaplan and Schoar (2005)for private equity.Firstquartile BDCs remain in the first quartile with a probability of 53%, and fourth-quartile BDCs remain at the bottom of the ranking with a probability of 59%.A chi-squared test rejects the null of equal probabilities at the 1% level.

Table 5 .
Contingency Table of Initial and Subsequent One-Year Performance Rankings Note: This table shows the probability of a BDC's performance quartile ranking (Q1 ¼ highest, Q4 ¼ lowest) in a subsequent year, given the ranking in the initial year based on three measures (market value, market value alpha, and NAV total return).All of the performance measures are calculated for each of the BDCs in the sample for each calendar year 2010-2021 where a full year's performance history is available.The alphas are calculated from regressions of monthly BDC market value excess returns (over 1-month T-Bill) on Russell 2000 Value Index and S&P/LSTA Leveraged Loan Index excess returns.BDC and benchmark data is sourced from Eikon Datastream and BDC NAV data is from company 10-K and 10-Q filings in EDGAR.

Table 6 .
Individual BDC Characteristics and Performance from Eikon Datastream and total assets from Compustat.BDC NAV, portfolio yield, and management expense data is from company 10-K and 10-Q filings sourced from EDGAR.Robust standard errors are in parentheses.Ã and ÃÃ indicate statistical significance at the 5% and 1% level, respectively.
Note: This table shows descriptive statistics of key characteristics of the 47 BDCs in the sample in Panel A. The characteristics are average total assets, debt-to-equity ratio, portfolio yield, annual ratio of management expense to total assets, and market price to NAV.The characteristics are calculated quarterly over the period Q4/2009-Q2/2022, except for management expense, which is calculated at year-end 2010-2021 (or during the BDC's reported history, if shorter).Panel B shows results of cross-sectional regressions of portfolio performance measures regressed on characteristics in Panel A. The performance metrics are the annualized total returns based on market value and NAV, annualized alphas from a regression of BDC market value excess return (over 1month T-bill) on Russell 2000 Value Index and S&P/LSTA Leveraged Loan Index excess returns (Mkt Val.BM Alpha), iShares Russell 2000 Value ETF and INVESCO Senior Loan ETF excess returns (Mkt Val.ETF Alpha), and the annualized alpha from regressions of quarterly BDC NAV excess returns (over compounded 1-month T-Bill) on contemporaneous and lagged Russell 2000 Value Index and S&P/LSTA Leveraged Loan Index excess returns (NAV Alpha), each calculated over the period where data is available for each BDC from December 2009-June 2022.BDC market values, debt-to-equity ratios, and other index data are

Table A
.1.List of BDCs in the Sample

Table A .
2. Performance Comparison of BDC Market Value Index with Public BDC Indices and Alternative Weighting Schemes, December 2009-June 2022 Panel A shows monthly performance statistics of a market value-weighted portfolio of BDCs calculated based on market price return (MVIndex) and two public BDC indices, the Cliffwater BDC Index and the S&P BDC Index.The table also includes an equally-weighted MVIndex (MVIndexEW) and a version where each year the largest BDC by market capitalization is excluded from the index (MVIndexEx).All of the indices include reinvested dividends.TR p.a. is the total return of the index and ER is the return in excess of USD 1-month T-Bills.A correlation matrix of monthly returns is presented at the bottom of the panel.Panel B shows regressions of monthly excess returns (over 1-month T-Bill) of the various BDC indices in Panel A on excess returns of benchmark indices.R2K Value is the Russell 2000 Value Index, and Leveraged Loans is the S&P/LSTA Leveraged Loan Index.Robust standard errors are in parentheses.Ã and ÃÃ indicate statistical significance at the 5% and 1% level, respectively.CWBDC index data is from Bloomberg and other data from Eikon Datastream. Note:

Table Alternative Sample
Periods and High-Yield Bond ExposureThis table shows regressions of monthly excess returns (over 1-month T-Bill) of a market value-weighted portfolio of BDCs calculated based on market price return (MVIndex) on excess returns of different benchmark indices over the period December 2004-June 2022 (columns 1 and 2) and during December 2009-June 2022 (the main sample period in the study, columns 3 and 4).Column 5 reports the results of the regression in column 1 over the period from December 2009-June 2022., excluding the global financial crisis (GFC) that is defined as June 2007 to May 2009, and column 6 shows the results during GFC.Columns 7 and 8 present the results from the period before and after the introduction of the 2018 Small Company Credit Availability Act, respectively, and column 9 shows the results from the period after 2018 using a high-yield bond index as the credit market benchmark.R2K Value is the Russell 2000 Value Index, Leveraged Loans is the S&P/LSTA Leveraged Loan Index, and High Yield is the Bloomberg US Corporate High Yield Total Return Index.The high-yield index is sourced from Bloomberg and the other data from Eikon Datastream.Robust standard errors are in parentheses.Ã the study.I present a of alternative periods alongside my main sample, plus alternative specification of the main regression model in Table A.3.Specifically, I consider the following periods: 1. December 2004 to June 2022 2. December 2009 to June 2022 (the sample period of the main study) 3. December 2004 to June 2022 including and excluding June 2007 to May 2009 (GFC) 4. December 2009 to December 2017 and December 2019 to June 2022 Note:

Table A .
4 presents a regression of NAVIndex on eight lags of small-cap equity value returns and four lags of leveraged loan returns, as well as the final chosen model.The final model is selected by testing down nonsignificant lags from the regression in column 1 one by one.Note that the first, third, and fifth equity lags in the chosen specification (column 2) are statistically significant at the 10% level.
Table A.4. NAVIndex Full Lag Structure, December 2009-June 2022 This table shows regressions of quarterly excess returns (over compounded 1-month T-Bill) of a market value-weighted portfolio of BDCs calculated based on net asset value return (NAVIndex) on contemporaneous and lagged excess returns of benchmark indices.R2K Value is the Russell 2000 Value Index, and Leveraged loans is the S&P/LSTA Leveraged Loan Index.Benchmark index data is from Eikon Datastream.BDC NAV data is from company 10-K and 10-Q filings sourced from EDGAR.Robust standard errors are in parentheses.Ã and ÃÃ indicate statistical significance at the 5% and 1% level, respectively. Note:

Table
Panel Regressions of Annual BDC Performance Metrics on Characteristics This table shows results of panel regressions of annual (calendar year) individual BDC performance measures regressed on five BDC characteristics (total assets, debt-to-equity ratio, portfolio yield, management expense to net assets, and price-to-NAV).The performance measures are the annualized total returns based on market value and NAV, and annualized alphas from a regression of BDC market value excess return (over 1-month T-bill) on Russell 2000 Value Index and S&P/LSTA Leveraged Loan Index excess returns, each calculated over the period where data is available for each BDC from December 2009-June 2022.BDC market values, debt-to-equity ratios, and other index data are from Eikon Datastream and total assets from Compustat.BDC NAV, portfolio yield, and management expense data are from company 10-K and 10-Q filings sourced from EDGAR.Robust standard errors adjusted for clusters in annual returns are in parentheses.Ã and ÃÃ indicate statistical significance at the 5% and 1% level, respectively.