Here’s a story that the Securities and Exchange Commission tells about Standard & Poor’s 2010 holiday party. Specifically, it’s about Barbara Duka, the managing director in charge of S&P’s ratings of commercial mortgage-backed securities at the time, who wanted to make a change to S&P’s ratings methodology:

At S&P’s holiday party, she and one or two other members of the CMBS Group approached the new CMBS criteria officer, who had just joined S&P earlier on the same day, and pushed him to agree to use blended constants. When he demurred, Duka approached the chief of S&P’s structured finance criteria organization with the same request early the next morning. After a brief meeting, Duka unilaterally concluded that she had obtained his approval for use of the blended constants, but she made no record of the meeting or this decision.

We’ll get to blended constants in a moment, but don’t you just love this scene? For one thing, I mean, what a fun party. And ambushing the new guy at a party on his first day to get him to sign off on your aggressive changes is really just such a power move. Herodotus says of the Persians:

It is also their general practice to deliberate upon affairs of weight when they are drunk; and then on the morrow, when they are sober, the decision to which they came the night before is put before them by the master of the house in which it was made; and if it is then approved of, they act on it; if not, they set it aside. Sometimes, however, they are sober at their first deliberation, but in this case they always reconsider the matter under the influence of wine.

I was pleased to learn that this was also the practice at S&P.

Today the SEC fined S&P a total of $58 million – and two state attorneys general took another $19 million – for doing an assortment of bad stuff in its rating of mortgage-backed securities. The SEC is also suing Duka personally^{1} for, basically, being awesome at parties.

Standard & Poor’s is perhaps best known for giving bad ratings to residential mortgage-backed securities before the financial crisis, and the big lawsuit over that is ongoing, but today’s fines are mostly for*commercial *mortgage-backed securities ratings,^{2} and are for post-crisis behavior, not for getting the global financial crisis wrong.

Actually one of the fines is for S&P getting the Great Depression wrong. The idea is that S&P wants an “AAA” rating to mean that a security would survive the Great Depression. That is a good thing to advertise, but it does sort of require you to figure out how the security would have done in the Great Depression. That is in a sense a silly question, but in 2012 S&P went and did it for commercial mortgages, publishing an article called “Estimating U.S. Commercial Mortgage Loan Losses Using Data From The Great Depression.” The SEC has some complaints about the article:

The Great Depression Article was flawed, in part because it suggested “about 20%” losses in periods of “extreme economic conditions” without adequately disclosing certain significant assumptions, including the following:

a) S&P’s analysis of purported Great Depression losses and defaults included analysis of performance of commercial mortgages originated between 1900 and 1935, many of which were not affected by the extreme economic stress of the Great Depression;

It goes on. This one cost S&P $15 million. Fifteen million dollars, for getting its history wrong.^{ 3}

The big fine, though — $42 million to the SEC^{4} – stems from the “blended constant” stuff that Barbara Duka tried to push through at that party. This had to do with how S&P calculated the debt service coverage ratio, a measure of whether a commercial building generates enough money to pay off its debts with some room to spare. This ratio fed into S&P’s estimate of default likelihood, and thus into how conservative, or not, its ratings were.

Conceptually, the way you calculate debt service coverage ratio for a property is:

- You figure out how much cash the property will generate every year.
- You figure out how much the property owner will have to pay back on its loans every year.
- You divide the first number by the second number.

The SEC has no problems with how S&P calculated the numerator, the future cash flow.^{5} The denominator — the debt service cost — is where S&P ran into trouble. This is a bit weird, because unlike the numerator — future cash flows444 this is just a number that you can know. You’re rating a loan. You know how big it is. You know its interest rate.^{6} You know its amortization schedule. From that, you can figure out how much the building’s owner has to pay each year on the loan.

But S&P does some arithmetic manipulation: Instead of [Cash Flow *divided by *Debt Service], it calculates the coverage ratio as [Cash Flow *divided by *(Loan Constant *times* Loan Amount)]. Naively, the loan constant is just the debt service divided by the loan amount, so it should all cancel out. You can convert any loan payment into a “loan constant.” Just for fun I calculated the loan constant on my mortgage. It’s 6.44 percent. So if I had a $1 million mortgage, I’d be paying $64,400 a year in interest and principal.

Here’s the methodology I used to calculate the loan constant on my mortgage:

- Take my yearly payment, and divide by the original principal amount of my mortgage.

Easy! Here’s the methodology S&P used to calculate its loan constant, before its holiday party:

- Take the yearly payment, and divide by the original principal amount of the loan. (This is called the “actual constant.”)
- Take an arbitrary number from Table 1 on page 5 of this 2009 publication, 7.75 percent to 10 percent, depending on the property type. (The SEC calls this the “criteria constant” or “Table 1 loan constant.”)
- Use whichever is higher.

I mean … I guess? The arbitrary numbers tended to be higher than the actual numbers, and a higher loan constant gets you a lower debt service ratio, and a lower debt service ratio gets you a lower rating. So this methodology was conservative. On the other hand, it’s maybe a little weird? To just use an arbitrary number?^{7}

Duka pushed to change the methodology to a “blended constant” model, which worked like this:

- Take the yearly payment, and divide by the original principal amount.
- Take the arbitrary number from that Table 1.
- Take the average of #1 and #2 to get a third, literally semi-arbitrary number.
- Take whichever of #1 and #3 is higher.

So now, if the arbitrary number was higher than the actual debt service, it would only have half as much impact as it did before. This was a change! Toward … accuracy? I don’t even know.