In Defense of Ratings Agencies

People don’t generally like interacting with algorithms, especially when it comes to something important. Algorithms are rigid, reducing the complexity of an individual situation to an inhumane and seemingly simplistic check list, and nobody likes thinking that their situation isn’t unique and special. Humans tend to be very judgmental when they see algorithms produce erroneous results, even if the overall accuracy level is higher than that of humans.

So you can understand why most investors despise credit ratings agencies. Investors feel that ratings agencies don’t understand the businesses they’re rating, that they ask formulaic, check-the-box questions, and that, above all, they lack nuance and an ability to deal with unique and complex individual cases.

But all this talk of the stupidity of the ratings agencies misses a fundamental truth: the ratings agencies are using empirically validated and highly accurate actuarial algorithms that work precisely because they compare individual companies to base rates rather than overweighting the specifics of each individual company. And there is no more highly replicated and robust finding than the superiority of algorithmic prediction models over clinical prediction models.

The process that ratings agencies follow is relatively simple, but consistent and repeatable, which makes it powerful. While specifics vary, each agency follows a scorecard process, assessing various parts of the business such as competitive position or financial strength. Below, we show an example of a scorecard from Moody’s that is illustrative of the broader process.

Figure 1: Example of Moody’s Scorecard for the Chemical Industry

Source: Moody’s

This process quickly sorts companies by rating, with companies at different ratings levels having strikingly different financial profiles. Below we show how three simple credit statistics, EBIT (size), EBITDA/Interest (ability to pay), and Debt/EBITDA (debt level) vary by rating.

Figure 2: Median Credit Statistics by Rating (Issuer Weighted)

Source: Verdad bond database

As you can see, there’s a vast quantitative difference between companies with different ratings. And it turns out, unsurprisingly, that this algorithmic process is excellent at doing what it’s supposed to do: predicting corporate default rates. Below are the average cumulative default rates by rating shown over the years from when the rating is issued.

Figure 3: Global Corporate Average Cumulative Default Rates by Rating (1981-2018)

Source: Standard & Poor’s

We believe these default statistics are so reliable that they are worth investors memorizing so as to quickly be able to translate ratings into estimated default rates and thus understand the true default risk of different stocks and bonds. Below we show Moody’s and Fitch’s estimates of five-year default rates and upgrade rates.

Figure 4: Realized 5-Year Default and Upgrade Rates

Source: Moody’s and Fitch. The numbers differ due to different time periods (there was a spike in default rates in the late 1980s that the Fitch data doesn’t capture) and breadth of coverage (Moody’s has broader coverage, including more sub-investment-grade debt).

The story is remarkably clear. Once into the single-B category, bad things (default) are just as or more likely than good things (upgrades). When the ratings agencies assign a single-B or CCC rating, the message is clear: you will experience default loss, and you will not earn your yield. Index returns by rating bear this out.

Figure 5: Average Yield Versus Return by Rating 12/1996 – 10/2019

Source: Verdad bond database

Expected returns in bonds are a function primarily of the entry yield and the expected default rate. The latter is what credit ratings agencies are extremely good at predicting. As the miserable performance of lower-rated credits in the above chart demonstrates, betting on yields to overcome default rates is a loser’s game.

One simple test of the rating agencies is to see what happens when the rating agencies and the market disagree. Do the rating agencies move toward the market-implied rating level, or does the market move toward the agency ratings? The expectation is that the market leads and the rating agencies follow. But that is not what happens. When the market suggests a downgrade, the agencies do indeed downgrade, on average, but the market also corrects toward the agency rating. When the agencies are above the market, the bonds rise toward the agencies.

Figure 6: The Agencies and the Market Split the Difference

Source: Verdad bond database

Ratings send valuation signals that the market does not fully incorporate. Or, as our friends Wes Gray and Tobias Carlisle put it much more memorably, “We think we know better than simple models, which have a known error rate, but prefer our own judgement, which has an unknown error rate.”

None of this should be surprising, yet the trend in private markets has been into exactly the kind of low-rated credit that does not earn its yield. Private equity finances buyouts in the single-B credit market. Private credit makes primarily single-B and CCC loans. Both areas have experienced explosive growth in the past several years, and the agencies have dutifully delivered single-B and CCC ratings. We believe investors in credit would be wise to start with the assumption that the ratings agencies’ assessments of default are likely to prove correct. Ignore the base rates of default at your peril.