The Use of Credit Ratings in the Delegated Management of Fixed Income Assets

Fixed income markets are large and of central importance to the financial system, providing much of the external financing of corporations, financial institutions, and the public sector.¹ Most investment decisions in fixed income markets are not made by the ultimate asset owners (e.g., households and governments) but by agents such as mutual funds, insurance companies, and pension funds. Together, bond mutual funds and money market mutual funds managed $13.7 trillion worldwide in 2021 (Investment Company Institute 2022). The interaction between such asset managers and the ultimate asset owners is key to understanding fixed income markets. An important determinant of the relationship between owners and managers is the investment mandate, which stipulates how assets are to be managed and with which restrictions. For example, a mandate may define a fund’s broad investment strategy or dictate which types of assets a fund manager can purchase. In particular, investment mandates often restrict the risk of portfolio assets (Almazan et al. 2004, He and Xiong 2013). In this paper, we examine the structure of such investment mandates for fixed income asset managers and the role that credit ratings play in restricting risk taking by funds.

Using regulatory filings in the United States and Europe for the last 20 (United States) and 10 (Europe) years, we perform a textual analysis on investment mandates (contained in the filings) to classify how portfolio risk is characterized and constrained in these mandates. Most mandates limit the set of investable assets using credit ratings. In all, 60% of European funds and 93% of U.S. fund mandates refer to credit ratings in some way.² This is in line with earlier survey-based findings by Cantor et al. (2007) that credit ratings are used in the guidelines of some of the largest asset managers working with pension plans.³ References to credit ratings are implemented in different ways; some mandates specify that assets will be invested “primarily” in investment grade securities, and others determine that most funds will be invested in high-yield assets; some require ratings from a particular agency, whereas others allow the ratings of any agency certified by regulators. Mandates also differ in their amount of leeway (e.g., allowing 10% of assets to be outside the target rating range).

We collect data on security holdings of funds to establish that portfolios conform to mandates; the credit ratings of fund assets closely match mandates. For example, examining corporate bond portfolios of investment grade (“IG”) funds, we document that 96% of assets are IG. In contrast, high-yield (“HY”) funds hold 7% IG assets. Funds without ratings references of any kind hold 60% IG assets. The close match between mandates and holdings may reflect legal and commercial risks associated with deviations from investment mandates. Examples of lawsuits about this include a 2008 case in which investors in Schwab’s YieldPlus Fund filed a class action suit against Schwab for deviating from investment and concentration policies. The defendant paid $119 million to settle the charges (Laisse 2011). Similarly, in 2012, Evergreen Investment Management Company paid $25 million to settle a class action claiming that it misled investors about one of its fund’s exposures to mortgage-backed securities (Reuters 2012).⁴

The use of ratings in mandates extends to funds marketed to retail as well as to institutional investors (as in the Cantor et al. 2007 study), to ETFs as well as open-ended mutual funds, and across all large categories of fixed income funds (e.g., municipal bond funds and corporate bond funds). Younger funds, funds with more assets under management, corporate bond funds, and ETFs are more likely to use ratings, whereas index funds are less likely to do so. Broad-based use of ratings is consistent with an agency-based view of asset management (see, e.g., Admati and Pfleiderer 1997, He and Xiong 2013), whereby constraints on risk-taking are required to convince investors to invest with a manager.

Not only is ratings use frequent, but it has also been increasing. In our U.S. sample, a broad measure of ratings use in mandates increased from 90.0% of funds in 2010 to 94.4% in 2020. This implies that nonuse has been cut almost in half and suggests a net increase in ratings use by 0.4% of the fund stock annually (this net reflects both changes in mandates and turnover in the fund universe). In Europe, ratings use increased from 46.8% to 65.8% between 2012 and 2021, suggesting that nonuse has been reduced by one-third (equivalent to an annual increase in ratings use by 2.1% of the European fund universe). Increases in ratings references in mandates are observed after controlling for fund characteristics such as assets under management and asset class (corporate, municipal, etc.) and within individual funds over time. The pattern of increasing ratings use holds for the period after the global financial crisis in 2008 as well as in the decade preceding it.

This broad-based trend might seem puzzling given the widespread criticism of credit ratings following the financial crisis.⁵ There are at least three possible explanations (not mutually exclusive). First, despite the popular and regulatory backlash against credit rating agencies, the financial crisis may not have changed sophisticated investors’ perceptions of ratings quality much, at least for asset classes not associated with major losses during the financial crisis. The most significant losses were sustained in structured assets, which most funds do not hold; fixed income mutual funds largely invest in corporate and municipal bonds and in Treasury securities. European mandates most often use ratings when investing in U.S. assets, where coverage is deepest and where the history is the longest. Perhaps ratings are perceived as reliable metrics in this area and have no viable alternatives. Under this view, low quality ratings generate a negative externality on the financial system, given that they are effectively irreplaceable.⁶

Second, competition between asset managers for funds may force increased use of ratings. As Stein (2005) pointed out for open-ended fund structures and Donaldson and Piacentino (2018) for references to credit ratings, competition for funds under management accelerates the adoption of contractual commitments that reduce agency problems.⁷ Competition has been increasing in asset management (see, e.g., Khorana and Servaes 2012 and Gârleanu and Pedersen 2018), and this trend fits increasing ratings use in mandates.⁸

Third, ratings use in fixed income mandates and other private, nonregulatory settings may be “sticky,” so that the use of ratings remains the market convention even if better alternatives exist. For example, contracting conventions may lead to the ubiquity of ratings.⁹ Furthermore, network externalities may result in persistent and increasing use of ratings through market participants’ desire for consistency and comparability of credit risk metrics; thus, increasing ratings use may increase the utility of ratings to investors, issuers, and intermediaries.¹⁰ This view implies that there may be multiple equilibria (e.g., all funds use ratings or all funds use market prices to measure credit risk) and raises the possibility of welfare losses associated with the current equilibrium.

The widespread and increasing use of ratings in private contracting has implications for financial regulation. Credit ratings fulfill the same function in regulation as in private contracting: measuring credit risk. They have well-understood scales (especially the investment grade and high-yield categories), they have a long track record, and they are available for many securities free of charge to investors. Having a well-understood risk measure available broadly and at zero marginal cost to contracting parties allows regulators to make capital requirements of financial institutions dependent on the risk of their assets in a transparent manner, just as it allows mutual fund clients to allocate funds across risk categories.¹¹ Given this similarity between the various private and public uses of ratings, a lack of alternatives for one is likely informative about a lack of alternatives for the other. Among recent rulemaking in the United States, the Dodd-Frank Act instructed federal agencies to remove references to ratings wherever possible.¹² Our findings suggest that contracting on credit risk without ratings may be infeasible and replacing them difficult. In countries with less developed fixed income markets (such as those in Europe), formal contracting is used less by fixed income funds. Our results suggest that increased reliance on credit ratings may be unavoidable in these markets, because delegated asset management with formal mandates and competition for funds are likely to increase.

1. Data and Main Samples

1.1. General Aspects of the Textual Analysis

We construct a data set that quantifies textual information related to investment mandates in both U.S. and European fixed income mutual funds. For U.S. funds, this information is extracted from archived prospectuses of U.S. investment companies. The source of these documents is the EDGAR database of the SEC. Our primary sample comprises fund-specific summary prospectuses (filing type 497K) filed between 2010 and 2020 pursuant to rule 497(k) of the Securities Act of 1933.¹³ Summary prospectuses are typically short (2–3 pages), have standardized headings, and were specifically designed by the SEC with retail investors in mind. Because these documents describe specific funds, we can link them to observable portfolio characteristics from the CRSP Mutual Fund Database such as investment style classifications and holdings.

In addition to fund-specific summary prospectuses, we also consider prospectuses filed at the level of fund groups (filing types 485APOS and 485BPOS) between 1999 and 2020.¹⁴ Such groups of funds are typically a subset of an investment company’s funds that were launched on the same date. Although most of these fund group prospectuses encompass more than one fund, making cross-sectional comparisons across funds less clear-cut, they allow for an analysis of trends over a longer sample that covers both the pre- and post-financial crisis periods.

We remove filings that contain no text that is useful for our analysis.¹⁵ We use Series IDs in the case of 497K filings and Central Index Keys (CIKs) in the case of 485 filings to identify funds and fund groups, respectively. The CIK is a unique identifier for fund groups, and the Series ID is the unique identifier at the fund level. Each filing is associated with the date on which it was filed with the SEC. Whenever we are left with more than one filing at the CIK or Series ID level in a given year, we use the one that contains the largest number of sentences.

To construct text-based variables from the prospectuses, we first perform some basic cleaning steps and remove formatting and HTML code. Next, we identify and extract text passages that explicitly describe the reporting funds’ investment mandates. Finally, using dictionaries that we develop for this purpose, we perform text searches that capture references to credit ratings and several related concepts. For example, we record whether a given fund’s mandate explicitly refers to specific agencies and whether it mentions the terms “investment grade” and “high yield”.

In the case of 497K filings, we identify the relevant passages by focusing on the mandatory section “Principal Investment Strategies”. Following SEC regulation, this section contains the rules according to which the reporting funds invest.¹⁶ Fund group filings of the types 485APOS and 485BPOS tend to have a more idiosyncratic structure than 497K summary prospectuses, which are standardized. However, we can extract the same type of information from group prospectus filings by focusing on sentences that contain the following elements: (i) a relevant fund word (e.g., “we,” “fund,” “portfolio”), (ii) a relevant action word (e.g., “invest,” “hold,” “purchase”), and (iii) a mandate phrase (e.g., “we may,” “up to XX% of the portfolio”).¹⁷

For both U.S. filing types, we also exclude examples and consider only statements about credit quality.¹⁸ This ensures that we do not capture references that are unrelated to credit ratings (for example, references to S&P indices). Given the selected passages and the dictionaries we develop, we can run fully automatic searches that achieve a high classification accuracy and yield all the main text-based variables employed in the analysis of Section 2.¹⁹ We report these variables together with the corresponding dictionaries and exclusion lists in Table 1.

Table 1. Descriptions of the Main Text-Based Variables

For the analysis of European fixed income fund investment mandates, we use Key Investor Information Documents (“KIIDs”) that we obtain in PDF format from the Morningstar Direct database. KIIDs are two-page documents that are similar in nature to the 497K summary prospectuses used for the analysis of U.S. mutual funds. KIIDs are prepared according to the UCITS IV Directive of the European Parliament, implemented in January of 2009. We consider all European open-end fixed income funds that are domiciled in Luxembourg and which are available in Morningstar Direct as of mid-2021. Morningstar Direct publishes the latest KIID for each share class of a fund; in some cases, a limited set of historical KIIDs can also be retrieved. For each fund with nonmissing information on net assets and nonmissing ISIN (International Securities Identification Number, a unique share-class level identifier), we collect English-language KIID filings of the fund’s largest share class. We consider KIIDs from 2012 (there are few filings for prior years on Morningstar) to 2021.²⁰

In the KIIDs, the mandatory section that contains the funds’ investment mandates is typically entitled “Objectives and Investment Policy.” We extract these sections from the PDFs and convert them to machine-readable texts using optical character recognition. To construct the variables that we use in the empirical analysis, we then implement the same filters and definitions as in the case of the U.S. 497K summary prospectuses.

1.2. Sample of U.S. Summary Prospectuses

Our main sample consists of fund-specific summary prospectuses of U.S. funds (filing type 497K). Using the EDGAR-CRSP linking file, we combine information from the CRSP mutual fund database with information from funds’ SEC filings on EDGAR. Using this link, we add the funds’ Lipper objective codes from CRSP to the funds’ summary prospectuses. In our main tests, we retain the 497K filings of fixed income mutual funds according to the Lipper classification. We exclude from our main sample filings of money market mutual funds, because the investment opportunity set of such funds was circumscribed by ratings-based regulation until the end of 2016 (Rule 2a-7 of the Investment Company Act of 1940). We also exclude fixed income funds that invest only in U.S. government securities, because those assets de facto all carry the highest credit ratings. The fund categories in the sample thus comprise municipal debt funds, fixed income funds focusing on debt from international issuers, corporate debt funds, funds investing in mortgage-backed securities, and “other” fixed income funds. Table A.4 in the online Appendix lists the main fixed income categories examined by us, along with the constitutive Lipper objective codes.

Table 2 reports the number of summary prospectus filings by fund category; a given fund is represented at most once per year. The sample includes 19,304 filings by 2,642 fixed income mutual funds. The number of summary prospectus filings has increased over time, reflecting a rising number of reporting funds. The two largest fund categories in terms of filing volume are corporate debt funds (7,463 filings) and municipal debt funds (5,601), followed by “other” fixed income mutual funds (4,064). Fixed income funds primarily investing in foreign debt securities and those primarily investing in MBS contribute 2,011 and 165 filings, respectively.

Table 2. Summary Statistics: No. of Summary Prospectus Filings by Fund Type (U.S. Sample), 2010–2020

Since 2010, funds have been required to include a separate summary section in their fund group prospectuses (filing type 485). However, they can also release these summary sections as separate filings (497K). Therefore, the number of 497K filings in any given year does not necessarily reflect the number of active U.S. funds. In fact, based on our analysis of CRSP data, the number of fixed income mutual funds (defined using Lipper objective codes, excluding government and money market funds) with at least $1million in total net assets was 2,026 in 2011, increasing almost monotonically to 2,388 funds in 2020 (there were 3,352 unique funds over that period). In contrast, the number of fixed income funds filing summary prospectuses increased from 1,020 in 2010 to 2,076 in 2020. We control for possible changes in the composition of the sample by including fund fixed effects in some regressions (see Section 2).

Table 3 reports summary statistics for both the variables constructed using the mandate texts and the data from the CRSP Mutual Fund database.

Table 3. Variables From the U.S. Sample of 497K Filings, 2010–2020

1.3. Sample of U.S. Fund Group Prospectuses

Filings of the types 485APOS and 485BPOS encompass entire fund groups (which can include both equity and fixed income funds), and they are available for a longer period than the 497K summary prospectuses, namely from 1999 to 2020. We match the fund group’s CIK from the 485 filing to the CRSP Mutual Fund database using the EDGAR-CRSP linking file. We then determine whether the fund group includes a fund that is classified as a debt fund using Lipper objective codes (see Table A.4 in the Online Appendix). We retain in the sample those 485 filings that contain at least one debt fund.²¹ The resulting sample contains 13,194 prospectuses filed by 758 different fund groups over the period 1999–2020. Table 4 reports summary statistics for this sample. It shows the variables derived from the extracted investment mandate passages.

Table 4. Variables From the U.S. Sample of 485 Filings, 1999–2020

1.4. Sample of European Key Investor Information Documents

Table 5 reports the number of KIID filings by fund category (funds are grouped into five coarse categories using more detailed underlying information on fund types from Morningstar; for details on the grouping, please refer to Online Appendix Table A.5). The sample spans the years 2012 to 2021 and encompasses 12,382 KIIDs from 2,189 European fixed income funds.²² The KIIDs were downloaded from Morningstar Direct in August and September 2021. Because the database primarily covers recent filings and historical filings are not available for all funds, the sample size increases over time.²³ Table 6 reports summary statistics for the variables used in the analysis of European fixed income funds. These comprise text-based variables (see Table 1 for details) as well as variables from Morningstar Direct, with values as of 2021.

Table 5. No. of KIID Filings by Fund Type (European Sample), 2012–2021

Table 6. Variables From the Sample of KIID Filings (European Sample), 2012–2021

2. Empirical Analysis

2.1. Proof of Concept

Before we analyze in detail how portfolio risk is characterized and constrained in fixed income fund investment mandates, we verify that the textual data extracted from the regulatory filings accurately capture references to credit ratings both in the time series and the cross-section of funds. To conserve space, we report these checks in Section B of the Online Appendix.

We start by validating the time series properties of our measures. First, we consider a regulatory reform that affected money market mutual funds (MMMF). As discussed in Section 1.2, we exclude MMMF from our main sample because the investment opportunity set of such funds was circumscribed by ratings-based regulation until 2016 (Rule 2a-7 of the Investment Company Act of 1940); therefore, references to ratings in such funds’ mandates could differ in nature from nonregulated funds. Because regulation encouraged ratings references before 2016 and discouraged them afterward, any change in ratings use is likely a reflection of regulation, rather than a test of how funds evaluate credit ratings of their own choice. The event does allow us to test how well our text-based measures reflect expected changes in ratings use; we expect MMMF to be less likely to refer to credit ratings in their prospectuses after the rule change. Indeed, we observe that the fraction of MMMF that refer to credit ratings fell considerably following the implementation of the reform; for example, the share of MMMF referring to the term “NRSRO” dropped from 18% in 2015 to 1% in 2018. We report these results in Table B.1 in the Online Appendix, Section B.

For the second proof of concept examining the time series properties of our measures, we identify references to environmental, social, and governance (ESG) criteria in investment mandates. Given the rising interest in ESG issues in recent years, a positive trend would seem natural. Table B.2 in Online Appendix, Section B, reports the fraction of summary prospectus filings that mention ESG-related terms over the period 2010–2020. As expected, only a minority of fixed income funds discuss such matters. In addition to the modest overall level, we also observe the expected increase in ESG references over time (from 0.3% of fixed income funds in 2010 to 13.6% of funds in 2020). This provides additional evidence that the text-based analysis of mandates yields useful data on how fixed income funds operate.

Next, we consider the cross-sectional properties of the measures. We examine whether the text-based measures based on the investment mandates match the actual portfolio holdings of fixed income funds. We consider the ratings of corporate bonds held by three groups of fixed income funds, classified using a textual analysis of the investment mandates: high-yield (HY) funds, investment grade (IG) funds, and funds that do not have any investment restrictions based on credit ratings. We find that portfolios are very closely connected to investment mandates; assets of IG funds are 96% IG, 3% HY, and less than 1% unrated; assets of HY funds are 7% IG, 91% HY, and 1% unrated; and assets of funds with no ratings references are 60% IG, 37% HY, and 2% unrated. In other words, investment grade funds overwhelmingly hold high-rated securities, and funds classified as high yield hold a vast majority of lower-rated securities, whereas funds not restrained by rating-based mandates hold securities across the whole rating spectrum. This analysis, also reported in Online Appendix Section B (Figure B.1), illustrates that our text-based classification produces data with meaningful cross-sectional properties and that ratings-based investment restrictions are indeed reflected in the portfolios of fixed income funds.

Finally, in Online Appendix Section B, we perform an additional test to illustrate that the measures derived from the text analysis of fixed income fund investment mandates are reliable and are related to the portfolios of funds. We consider security sales and purchases by funds with different investment mandates. These tests show that high-yield funds are significantly more likely to buy newly issued high-yield securities, whereas investment grade funds are significantly less likely to do so. Furthermore, securities that are downgraded to high yield are less likely to be sold by high-yield funds and are more likely to be sold by investment grade funds. In summary, our analyses confirm that not only do funds refer to credit ratings in their investment mandates but that the ratings-based investment restrictions of the mandates are also reflected in funds’ actual portfolio holdings.

2.2. The Use of Credit Ratings in U.S. Investment Mandates, 2010–2020

Which types of U.S. fixed income funds use credit ratings in their mandates to delineate the investment opportunity set? Table 7 sheds light on this question. We report coefficients from OLS regression models of the following type, which can be interpreted as cross-sectional comparisons,

Table 7. Determinants of Ratings Use in U.S. Fixed Income Fund Mandates

where f denotes the fund and t the year. ${\mathit{\gamma }}_{\mathit{t}}$ is a vector of year fixed effects. Because we are interested primarily in fund characteristics, we cluster standard errors at the fund level in these specifications. X denotes a set of fund characteristics. Y denotes the dependent variables: All ratings references (columns 1 and 2), which capture any type of reference to a credit rating in the mandate; HY/IG (columns 3 and 4), a dummy variable that is one if the mandate refers to terms that denote the investment grade threshold; Big 3, which is one if the mandate refers to S&P, Moody’s, or Fitch (columns 5 and 6); NRSRO, which captures references to the term “nationally recognized statistical ratings organization” (columns 7 and 8); and finally, Letter rating, which takes the value of one if the mandate refers to a specific alphanumeric credit rating, such as “A+” (columns 9 and 10).

Based on Table 7, we make several observations about the characteristics of funds that use ratings. First, fixed income funds investing primarily in corporate bonds (Category-corporate is the omitted fund category variable in the regressions and thus serves as the reference point) are significantly more likely to use ratings terms in the mandate than other types of fixed income funds. This is consistent with the fact that ratings for corporate bonds have historically been the most reliable measure of default risk compared with ratings for other asset classes (Cornaggia et al. 2017). Second, younger funds and ETFs are significantly more likely to use ratings, whereas index funds are less likely to do so. Third, the positive and significant coefficients on the variables Fraction (.) suggest that ratings use is strongly correlated across funds within management companies. These patterns are consistent across various measures of ratings use in mandates (different dependent variables). There is also some evidence that funds with more assets under management, funds that have retail share classes, and those charging higher fees tend to rely more on ratings than other funds, but the corresponding coefficients on these variables are not statistically significant in all specifications.

2.3. Trends in the Use of Credit Ratings in U.S. Investment Mandates, 2010–2020

How has the use of credit ratings in U.S. fixed income investment mandates evolved over time? Has the financial crisis affected the private use of ratings in financial markets, mirroring regulatory efforts to pull back on the reliance on ratings? Table 8 reports the annual fraction of funds that make ratings-related references in their investment mandates over the 2010–2020 period. Eighty-eight percent of fixed income mutual funds refer to the investment grade threshold (the mandates refer to “investment grade,” “high yield,” or both); this fraction has increased from 84% in 2010 to 90% in 2020. We consider the investment grade threshold as an indirect reference to credit ratings. About 23% of debt funds refer to the term “NRSRO.” Fifty-seven percent of funds refer to specific alphanumeric ratings or agencies (variable Direct ratings reference in the table) in 2010, rising to 62% in 2020. Overall, Table 8 suggests that both direct and indirect references to ratings in fixed income mandates have modestly increased over the 2010–2020 period from a high initial level.

Table 8. Annual Averages of Ratings Variables, United States (2010–2020)

At the end of our sample, 94% of the fixed income funds contain a direct or indirect ratings reference (up from 90% in 2010). Given the near-universal use of ratings in U.S. investment mandates, this begs the question of which funds do not use ratings and why. Although a detailed investigation is beyond the scope of this paper, we have more closely analyzed U.S. funds that do not use ratings; 253 funds that belong to 133 fund groups do not reference ratings over the 2010–2020 period. Based on univariate comparisons, these funds are significantly more likely to be index funds (35% of ratings non-users vs. 10% of ratings users), more likely to be ETFs (26% vs. 11%), less likely to be mainly investing in corporate bonds (15% vs. 40%), and more likely to be investing primarily in MBS (10% vs. 0.2%); ratings non-users also have slightly lower expense ratios (0.7% vs. 0.8% on average) and tend to be larger (average TNA of $2.3 billion vs. $1.8 billion). Other differences in characteristics do not stand out as economically significant. In online Appendix Table A.1, we have reproduced investment mandates of three funds that refer to ratings and three funds that do not. The mandates that do not refer to ratings appear to be shorter and vaguer than other mandates based on this limited sample.

Investment mandates of fixed income funds regularly refer to specific rating agencies. Do trends differ across these different raters? Are there reversals in trends, perhaps because of reputational damage suffered by specific rating agencies in relation to the financial crisis? For example, in 2015 (2017), S&P (Moody’s) settled a collection of lawsuits filed by the U.S. government related to S&P’s (Moody’s) structured finance ratings prior to the financial crisis. S&P or Moody’s may have suffered reputational damage related to the quality of ratings produced in the run-up to the financial crisis. Consequently, fixed income funds may have switched to other raters for the purposes of defining their investment opportunity sets.

Table 8 also sheds light on this question. The table reports the unconditional averages of the variables S&P, Moody’s, and Fitch over the 2010–2020 period. S&P is referred to most often (on average, 30% of the funds refer to S&P), Moody’s only slightly less frequently. Fitch is mentioned by around 18% of the funds. Over the 2010–2020 period, the fraction of funds referring to Fitch has increased from 13% to 21%, a steeper increase than for the other two raters. The average of the variable Nb. agencies (number of unique credit rating agencies mentioned by name in the mandate) increased from 0.6 in 2010 to 0.9 in 2020, which suggests that funds have been adding Fitch as an additional rater in mandates rather than using it as a substitute for S&P or Moody’s.²⁴ Overall, the table suggests that ratings use in mandates is widespread and that there has been no substantial negative revision of the view of individual agencies since the financial crisis.²⁵

The aggregate time trends in the sample suggest a stable or somewhat increasing use of credit ratings in U.S. investment mandates. However, other variables may be changing over time, and this may make a clear interpretation of the findings in Table 8 difficult. To avoid drawing conclusions from time trends that may be affected by omitted variables bias, we introduce controls for key characteristics that are potentially related to ratings use. Perhaps most critical in this regard are entry and exit from the universe of reporting funds. The aggregate trend toward (moderately) increased use of ratings indicates some combination of (i) new funds using ratings more than the existing population, (ii) exiting funds using ratings less, and (iii) continuing funds changing their mandates from year to year.

To address these issues and to investigate trends in the use of ratings over time, we estimate OLS regression models of the following type:

where f denotes the fund and t the year. ${\mathit{\gamma }}_{\mathit{f}}$ is a vector of fund fixed effects, which we include in some of the specifications; these eliminate the impact of fund turnover on the time trend, isolating the effect of changes in mandates of continuing funds. Linear trend takes the value of 0 in the year 2010; it is 1 in 2011, 2 in 2012, 3 in 2013, etc. The coefficient $\beta$, therefore, captures trends in rating references by fixed income funds. We cluster standard errors at the year level in these specifications because we are interested primarily in the precision of the estimate for the coefficient on the variable Linear trend. Y denotes the dependent variables, of which we report two in Table 9: All ratings references (columns 1–3), which captures any type of reference to a credit rating or rating agency; and Big 3 (columns 4–6), which captures references to at least one of the three main rating agencies (S&P, Moody’s, and Fitch).

Table 9. Trends in Rating References in the United States

Table 9, column 1, reports the coefficients on linear trend from regressions without controls and fixed effects. Table 9, column 2, reports coefficients from regressions that include various fund-level control variables: dummy variables for the fund type, assets under management, fund age, indicator variables for the existence of institutional or retail share classes, dummy variables for index funds and ETFs, the fund’s annual expense ratio, and a variable denoting the fraction of other funds of the same management company having the respective rating reference in their mandates. Finally, Table 9, column 3, reports a specification with fund fixed effects.²⁶ Table 9, columns 4–6, reports coefficients from similar specifications studying the trend in references to the three big credit rating agencies (dependent variable Big 3).

Consistent with the simple averages reported in Table 8, the regressions reported in Table 9 suggest that there has been a moderate increase in various rating references in fixed income investment mandates between 2010 and 2020. For example, considering the variable All ratings references, the coefficients on linear trend range from 0.002 in column 3 to 0.005 in columns 1 and 2. This implies that the incidence of mandate use of ratings has increased by 0.2 to 0.5 percentage points per year over the period 2010–2020. A similar moderate, positive trend can be observed for the variable Big 3: up to 0.8 percentage points in specifications without fund fixed effects and 0.2 percentage points in regressions with fund fixed effects. These tests imply that about half of the increase is attributable to continuing funds adding references, the rest to entry and exit. In the Appendix (Table D.1), we report coefficients from regressions using three additional ratings-related dependent variables: NRSRO, HY/IG, and Letter rating. Consistent with the patterns documented in Table 9, we also observe a moderate but significant positive trend over the 2010–2020 period using those alternative ratings variables.

2.4. The Use of Ratings in U.S. Investment Mandates Over the 1999–2020 Period

The sample employed in Section 2.3 is based on annual summary prospectuses (filing type 497K). The advantage of this sample is that each filing is fund specific and that all filings contain standardized sections for funds’ investment mandates. Furthermore, using the unique Series ID identifier from the SEC for each fund, together with the EDGAR-CRSP linking file, we can match the summary prospectuses to the CRSP mutual fund database and retrieve additional information on the funds. This permits us, for example, to classify funds as fixed income funds using Lipper objective codes. A disadvantage is that 497K filings are available only from 2010 onward, the post-financial crisis period. However, it is conceivable that the use of ratings by mutual funds differed prior, during, or after the financial crisis. To shed light on this issue, we extend our analysis to the pre-2010 period using fund group prospectuses (filings of the type 485, see Section 1.3). Each of these filings typically encompasses a group of funds rather than a single fund, and each filing may contain various types of funds (fixed income, equity, etc.). Furthermore, given the lack of common structure of these documents, it is not always possible to link discussions of investment mandates to specific funds within the filing. We describe the construction of the sample of fund group prospectus filings in Section 1.3.

We first examine the annual averages of the ratings-based variables in Table 10 for this sample of fund group prospectuses, covering the period 1999–2020. The time series average of the variable All ratings references is 0.97. This implies that most fund groups that contain at least one fixed income mutual fund have at least one such fund that refers to credit ratings in its investment mandate. Most ratings terms became more widely used over the sample period. For example, references to the investment grade threshold (variable HY/IG) increased from being in 83% of the group prospectuses in 1999 to featuring in 96% in 2020, whereas references to the term “NRSRO” increased from 45% to 60% over the same period. The increase is almost monotonic in most variables. There are a few exceptions to this pattern, however. First, references to alphanumeric ratings (variable Letter rating) remained rather flat during the sample period. Second, it appears that there was a modest, temporary drop in references to S&P and Moody’s after 2009, whereas references to Fitch continued to increase during the same period. Although this hardly constitutes dramatic evidence, it is suggestive that there may have been some reputational repercussions for S&P and Moody’s after the financial crisis.

Table 10. Annual Averages of Ratings Variables, United States (1999–2020)

Table 11 reports regressions studying the trend in ratings references over the period 1999–2020 more systematically. We report coefficients from regression models of the following type:

where g denotes the fund group and t the year. ${\mathit{\gamma }}_{\mathit{g}}$ is a vector of fund group fixed effects. Linear trend takes the value of 0 in the year 1999; it is 1 in 2000, 2 in 2001, etc. Y is the dependent variable; we employ All ratings references in columns 1–4 and Big 3 in columns 5–8 (specifications using the dependent variables NRSRO, HY/IG, and Letter rating are reported in Table D.2 of the Appendix). Standard errors are clustered at the year level. Whereas the specification reported in column 1 does not contain any fixed effects, the coefficients reported in column 2 are from a regression that contains fund group fixed effects.

Table 11. Trends in Rating References, 1999–2020, United States

Based on the coefficient estimates on the variable Linear trend in columns 1 and 2, we infer that ratings references in fixed income investment mandates have increased by about 0.3 percentage points per year between 1999 and 2020. This trend estimate is comparable in size to the one based on summary prospectuses reported in Table 9. The regressions support the conclusion that ratings use has moderately increased over the 1999–2020 sample period from a high initial level.

We also investigate whether the rate of adopting ratings in investment mandates has changed since the global financial crisis (GFC) of 2008. During the GFC, banks and investors sustained large losses, in many cases on securities that had been rated in the highest categories.²⁷ This resulted in widespread criticism of rating agencies’ methods, business models, and market power.²⁸ Figure C.1 in the Online Appendix provides an illustration of this view based on the tone of news coverage of credit ratings and rating agencies in the financial press during the 2000–2019 period; the tone of news articles became significantly more negative following the GFC. Regulatory reforms after the crisis were aimed at reducing the risk of ratings inflation and limiting the impact of flawed ratings in the future. There was broad agreement that the financial system’s reliance on credit ratings should be reduced; references to credit ratings were removed from many regulations.²⁹ Reforms included several provisions in the Dodd-Frank Wall Street Reform and Consumer Protection Act, approved by the U.S. Congress in 2010. Furthermore, in both the U.S. and Europe, new agencies were instituted: the European Securities and Markets Authority (ESMA) and the SEC’s Office of Credit Ratings. Given the backlash against rating agencies after the GFC, a drop in the use of credit ratings by private parties, including in mandates, could be expected.

To investigate whether the rate of adoption of ratings in mandates changed after the financial crisis, we modify regression model (3) by estimating separate trends for the 1999–2007 and the 2008–2020 periods, respectively. Results are reported in Table 11, columns 3 and 4. We find that the trend in ratings use has been positive both in the pre- as well as post-GFC period; however, the positive trend is flatter in the post-GFC period. According to Table 11, column 3, the time trend coefficient is 0.005 for the pre-GFC sample and 0.003 post-GFC. Both coefficients are significantly different from zero, and the difference between the two subsample coefficients is significant (the p value of that difference is 0.004). An examination of the trends in references to the big three credit rating agencies (specifications reported in Table 11, columns 5–8, using the dependent variable Big 3) leads to a similar observation.

What can one conclude from the reduced increase in the rate of adoption after the financial crisis? The patterns documented in Table 11 (columns 3, 4, 7, 8) need to be interpreted with caution. Whether the slowdown is driven by Dodd-Frank, the crisis itself, reputational damage suffered by the raters, or by some other event cannot be answered conclusively using these time series regressions. The slowdown in the increase of ratings use in mandates may be due to these factors, but they may also be a purely mechanical effect; the use of ratings is capped at 100%, so the rate of increase must slow down as the market approaches universal use of ratings. Given the very high level at the end of the sample (see Tables 8 and 10), this must happen soon.

Overall, our analysis suggests that, over the period from 1999 to 2020, the trend in the adoption of credit ratings in investment mandates has been positive, but the rate of increase has been slowing down over the past decade as the U.S. investment management industry approaches near-universal ratings use.

2.5. Changing Contract Terms: Adding or Removing Rating References in Investment Mandates

Asset managers may change the fund’s contract terms, including their investment strategies and how the investment opportunity set is demarcated. Funds that refer to ratings in their investment mandate in one year may cease to do so in the following year and vice versa.³⁰

How persistent are contract terms in fixed income funds? Do funds frequently add and remove credit rating references in their investment mandates? Do new funds tend to use ratings? We examine these questions in Table 12, in which we report transition frequencies for funds with respect to their use of credit ratings. We classify funds into four mutually exclusive categories: (i) funds that do not refer to any ratings-related term in their investment mandate; (ii) funds that refer only to the investment grade threshold (i.e., the dummy variable Direct ratings reference is zero, whereas HY/IG takes the value of one); (iii) funds for which Direct ratings reference is one; or (iv) new funds, i.e., funds that file a summary prospectus (497K) for the first time. We observe that rating references are rather “sticky.” Funds that refer to ratings in a given year (either directly or indirectly by referring to the investment grade threshold) have a likelihood of more than 95% to do the same in the next year. Less than 0.5% of the funds that use ratings in their mandates in a given year stop doing so in the following year. We also find that more than 90% of the new funds make a direct or indirect credit ratings reference in their investment mandates.

Table 12. Transition Frequencies Between Rating References, United States

2.6. The Use of Credit Ratings in Investment Mandates of European Fixed Income Funds

Does the use of ratings in mandates of European fixed income funds differ from the use by U.S. funds? There are notable differences between the bond, ratings, and mutual fund markets of the United States and Europe. For example, in the United States, publicly listed firms obtain a larger share of their financing from bonds than from loans, whereas for European listed companies, the amount of loans outstanding is twice the amount of bonds (Becker and Josephson 2016). Consequently, as a fraction of GDP, the European corporate bond market is significantly smaller than that of the United States (10% of GDP in 2017 compared with 31% in the United States). The European market is also more heterogeneous and fragmented along national lines. For example, whereas most European corporate bonds are rated, only about half of Nordic corporate bonds are rated.³¹ However, there are also similarities. As in the United States, S&P, Moody’s, and Fitch have more than 90% market share in the European ratings market (source: ESMA).

As discussed in Section 1.4, European UCITS-regulated investment funds must publish documents aimed at investors (so-called Key Investor Information Documents, or KIIDs) that are very similar in structure and content—including a discussion of the investment mandate—to the 497K summary prospectuses. This enables us to compare the use of ratings in fixed income mandates in Europe to that in the United States. We first ask which fund characteristics are associated with the use of ratings in European fixed income fund investment mandates. Table D.3 in Appendix Section D reports coefficients from regression model (1) estimated using the European fund data. Consistent with the evidence from the U.S. sample, European corporate fixed income funds are significantly more likely to refer to ratings in their investment mandates than other fixed income fund types. Furthermore, funds with more assets under management and younger funds are more likely to refer to ratings, which also mirrors the evidence from the U.S. sample. The specifications reported in Table D.3 also show that Luxembourg-domiciled funds that invest mainly in emerging market (U.S.) debt securities are significantly less (more) likely to refer to ratings in their mandates than funds that purchase mainly European debt securities. Regarding the region of sale, there does not appear to be any specific pattern.

Next, we shed light on the trend in the use of ratings in Europe. Table 13 reports annual averages of the text-based variables for the European mandate sample, which covers the years 2012–2021. Compared with the averages of ratings variables reported in Table 8 for the U.S. sample, there are many similarities but also several notable differences. First, as in the U.S. sample, there is a clear upward trend in the use of ratings over time. The trend appears to be steeper than in the United States (a more formal test follows in Table 14). Second, although ratings are commonly used in European mandates, they are still significantly less common than in the United States. For example, in 2020, 65% of the funds refer to any ratings (variable All ratings references), whereas the corresponding figure in the United States is 94%. This pattern can be observed for all ratings variables.

Table 13. Annual Averages of Ratings Variables, Europe

Table 14. Trends in Rating References, Europe

Third, although references to S&P and Moody’s are equally common in U.S. mandates (see Tables 8 and 10), S&P is used more often in European mandates than Moody’s (e.g., 17% of the European funds refer to S&P in 2021, whereas only 11% refer to Moody’s). As in the United States, Fitch features considerably less often in European mandates, but it is on an upward trend (increasing from 2% of the mandates in 2012 to 7% in 2021). These patterns are qualitatively consistent with market share data from the European securities markets regulator, ESMA; in 2018, S&P had 42% market share in Europe compared with 33% for Moody’s and 17% for Fitch.³² This suggests that investor demand may be driving the use of ratings in mandates. Also, like the United States, other rating agencies are rarely mentioned in the mandates; whereas Kroll, Morningstar, and Duff & Phelps are never mentioned, Dominion is mentioned in only 10 out of 12,382 mandates during the sample period.

Table 14 examines trends systematically by estimating regression model (2) using the European sample. Specifications 1–3 employ the dependent variable All ratings references, whereas specifications 4–6 use the variable Big 3; we report similar regressions with the dependent variables HY/IG and Letter rating in Appendix Table D.4, which lead to the same conclusions. In Table 14, column 1 reports coefficients from a regression without control variables, specification 2 includes a set of control variables (in addition to the Linear trend variable),³³ and column 3 reports a specification with fund fixed effects. We find that the use of ratings has been trending up over time. Considering the dependent variable All ratings references in Table 14, columns 1–3, we find that references to ratings have become more common over the 2012–2021 sample period by 1.3 to 2.7 percentage points per year. Regressions employing the dependent variable Big 3 show the same patterns, although the annual increase is more modest (between 0.5 and 0.6 percentage points per year, depending on the specification).

How do the trends in ratings use in Europe compare with those in the United States? To make a comparison, we pool the observations from both samples. The combined U.S. and European sample spans the years 2012–2020 (the years for which the two geographical samples overlap). To compare trends, we estimate the following regression model (separately for U.S. funds, for European funds, and in the pooled sample):

where f denotes the fund and t the year. ${\mathit{\gamma }}_{\mathit{f}}$ is a vector of fund fixed effects. X is a vector of year fixed effects for the years 2013–2020; the dummy variable for the year 2012 is omitted and serves as the benchmark. Therefore, the coefficients in vector $\beta$ capture trends in rating references by fixed income funds after accounting for fund fixed effects. The fund fixed effects eliminate the impact of fund turnover on the time trend, isolating the effect of changes in mandates of continuing funds. These fixed effects also help account for the fact that the coverage in the European sample increases over time (see Table 5). Figure 1 reports the coefficients $\beta$ from regression model (4), including 99% confidence intervals based on heteroskedasticity-robust standard errors clustered at the year level. Overall, the trend in ratings use is steeper in Europe than in the United States, which is unsurprising given that the level of ratings use is already significantly higher in the United States than in Europe (see Tables 8 and 13). The time trend in the pooled sample is intermediate (steeper than in the U.S. sample but flatter than in Europe).

Finally, Table 15 reports transition frequencies between rating references in investment mandates for the European sample. As in the U.S. sample, the probability of a fund exiting the sample (for example, because of closure of the fund) is around 2–4% per annum. Also, like in the United States, funds that refer to ratings in a given year have a likelihood of more than 95% to do the same in the next year. However, there are notable differences between the U.S. and European fixed income funds. In the U.S. sample, 8% of the funds that do not use ratings will use ratings in the subsequent year; in contrast, only 5% of the European funds that do not use ratings will do so in the next year. Furthermore, less than 0.5% of U.S. funds that use ratings will not use ratings in the subsequent year, whereas the corresponding fraction for European funds is around 1.5%.

Table 15. Transition Frequencies Between Rating References in Investment Mandates (Luxembourg-Domiciled Funds)

In summary, ratings are widely used in both the European and the U.S. fixed income markets, but more so in the United States than in Europe. There is a modest but significant upward trend in ratings use by funds domiciled on both continents. S&P features most commonly in European mandates, with some distance to Moody’s and a large gap to Fitch; in the United States, S&P and Moody’s appear to be similarly widespread, with a large but less glaring distance to Fitch.

3. Conclusions

Fixed income securities constitute a large component of the financial system, of investor financial wealth, and of financial institutions’ assets. These markets are of critical importance to monetary policy and to the financing for governments and firms. Overwhelmingly, investment decisions for these assets are delegated to professional managers. How are principal-agent conflicts in this market overcome? We use textual analysis to classify the mandates of fixed income mutual funds in the United States and Europe to shed light on the features of the interaction between portfolio managers and investors.

We find that credit ratings are widely used in mandates. Ratings fulfill a unique role as ex ante constraints on the level of risk-taking by funds.³⁴ The use of ratings is almost universal in U.S. fixed income funds; in Europe, about two-thirds of the funds refer to ratings in their mandates. Whereas credit ratings have been in use in the United States for more than a century, ratings are a more recent phenomenon in Europe; this may be part of the explanation why delegated asset management of fixed income assets in Europe is less reliant on ratings than in the United States.

Not only is the frequency of ratings references in mandates high throughout our sample, it has also increased over recent years in both the United States and Europe. In the United States, the use of ratings went from very common (9 in 10 funds in 2010) to almost universal (16 in 17 funds in 2020). In Europe, ratings use went from half of funds to two-thirds between 2012 and 2020. The steady increase in the use of ratings in delegated management of fixed income assets may be due to a variety of factors, including competition, which has steadily increased in the mutual fund sector.³⁵ Ratings may be increasingly used by fund managers in their contracts with asset owners to attract investor capital in an increasingly competitive environment (consistent with Donaldson and Piacentino 2018).

The pattern of high and rising use of credit ratings contrasts with the negative view of ratings that emerged after the financial crisis. Even if credit ratings have important flaws, as the academic literature convincingly suggests,³⁶ they remain critical to fixed income investors, to the health of financial markets, and to the funding that flows through these markets. The continued and widespread private use of credit ratings may reflect either that financial market participants find them reliable enough or that there is a lack of appropriate substitutes. This has important implications for the ability to replace ratings. Therefore, any regulatory effort to curb the usage of ratings needs to recognize as a first-order challenge the need for viable alternatives.

Given that European asset management is increasingly competitive, reflecting the common currency, regulatory harmonization such as the introduction of ESMA in 2012, and specific efforts to raise competition in asset management (European Commission 2020), we would hypothesize that the use of credit ratings will continue to rise in Europe.