In Monday’s blog we talked about the natural bankers’ tendency to judge credit risk tools by how they do on firms, sovereigns or individuals who failed to pay. How they performed on the best credits, those who did pay, gets short shrift. Today’s blog talks about the real business costs of default models that have been inflated to make the default probabilities on the defaulters even higher, knowing that the credit analyst will ignore the inflated default probabilities on the non-defaulters. As we explained Monday, this is a common problem aggravated by the moral hazard of model builders who take advantage of this knowledge.
Coming off the 2007-2010 credit crisis, when rating agencies and older Merton credit model technology performed poorly, the normal bankers’ tendencies have gotten even more reinforcement. “I don’t care about false positives, warning of risk when it’s not there, it’s the false negatives that tell me a risky counterparty is safe that really cost me money.” That sentiment is very easy to understand and the costs are very easy to measure: we’re still totaling the $1 trillion plus losses from the last credit cycle. The cost of false negatives is easy to measure, but as Paul Jackson points out in this very interesting post, it’s important to remember that most of the CDO-related losses were due to inaccurate models of correlation, along with fraud (poor underwriting) and greed (selling AAA ratings too cheaply with no thought of the long run consequences). Here are Paul’s comments:
Are false positives a real problem in credit modeling? Does it matter if a vendor of credit risk technology inflates all default probabilities since analysts will spend 98% of their time “pre-sale” looking at defaulters? We think the answer is a resounding YES.
The biggest difficulty with measuring the cost of false positives in credit risk is that they prevent attractive transactions from happening, and this opportunity cost is tracked by next to no one. Take the current tragedy of the Gulf oil spill. The initial response in a thinly traded credit default swap market for BP plc (66 CDS trades the week of May 3-7, 2010 according to www.dtcc.com) was to push spreads for BP way up, from 42.75 basis points on April 20 to 83 basis points today, according to CMA. If we assume that these 83 basis points a good default indicator, many analysts would say “we only buy investment grade securities,” think of the historical default rate on BBB securities of 35 basis points or so, and reject an investment in BP. A cold hard analysis of 2 million observations and 2,000 defaults like that embedded in Kamakura Risk Information Services version 5 default probabilities, however, rates the probability of default for BP on a cumulative basis over 5 years as only 10 basis points, only 2-3 basis points on an annualized basis. This gives a completely opposite signal than one gets from the view that the CDS spread is essentially pure default risk. BP has low leverage, lots of cash, and lots of insurance. Time will tell if today’s 83 basis point CDS quote is a false positive. If over the next 5 years BP does not default, the opportunity cost would have been 83 basis points minus the expected loss on an annualized basis, 3 basis points times (1 – the recovery rate on BP)…few will remember that they passed on this credit and on what the return would have been on average.
Other examples on the cost of false positives are much simpler to write about and much more personal, as told by some of the participants in the meeting of the International Association of Credit Portfolio Managers held in London, May 5-6, 2010. Unlike BP, these stories are about false positives in retail credit risk measurement. Here are just a few stories:
Mr. A has always been a risk-taker in his personal finances. Finally, his ship came in and he received a large reward on a few decades of hard work, financed on charge cards and other borrowing. On one happy day just prior to the IACPM meetings, he paid off completely almost six figures of credit card debt. “It felt so good to pay off Citibank at 29.99%, knowing they were funding their loans to me at 50 basis points or less.” “What happened after you paid them off?” I asked, “My guess is that they’d cut your rate in half and double your credit line.” He said “That’s what I expected too, but that’s not what happened. Three days after I paid them off, they cut my credit line by 90 percent.” I couldn’t help but say “They were probably mad that you’re a better credit risk than Citibank is now. What did you do?” He said, “I cancelled the card.” He prompted me to do the same, as part of my campaign against banks that are too big to fail. False positives cost Citi a banking relationship with Mr. A, now a sterling credit of high net worth, the normal prime target for most banks.
Mr. B is a very international person and traveled from the U.S. to visit the IACPM meetings. “What did you do during the 45 minutes you had free yesterday afternoon?” I asked him. “I had a curious experience,” he said. “I went to Harrod’s to buy some small gifts for my 3 children. I figured I might have some trouble using my normal US credit cards, so I pulled out my university-branded card, because it’s run by Barclays.” “Uh-oh,” I said, “I can guess what’s coming next.” “You’re right,” he replied, “The transaction was completely rejected. Plenty of room on the card, but they predicted I was a thief and I wasn’t. To make matters worse, they e-mailed me the next day and said there was suspected fraud on my account, asking me to call a US toll-free number. Of course that’s impossible to do from overseas so I ignored it until my return.” Another false positive, one that guarantees my friend Mr. B will not be using his card for the lucrative international airline fares and hotel bills that every charge card bank judges to be prime business.
Mr. C had another variation on the same theme, as he listened to Mr. B tell his story. “Here’s what happened to me. I went out to do my family souvenir shopping too, and I figured I’d set off all kinds of fraud alarms using my US credit card oversees,” he explained, noting that his US bank was actually a subsidiary of BNP Paribas. “So I just used my cash card. The good news is that I got the transaction done, but I had to wait 20 minutes for a telephone authorization.” The postscript came in an e-mail after he returned to the States. “You’ll never believe this. The morning after my return to the States, my wife sent me off to the supermarket. $150 of groceries later, the bank rejected the charge on my cash card even though there was plenty of cash in the account. I called the bank to complain and they asked me to call them before every overseas trip I make so this doesn’t happen again. They also had sent me an e-mail, just like Mr. B got, asking me to call a toll free number.” Mr. C’s getting an American Express card to use when he’s traveling.
On the one hand, we all know that losses and fraud are a huge problem in the credit card market. But to see three prominent members of the IAPCM treated like fraudsters by their long-standing bankers, in spite of their own credit expertise, means that at least for some clients, the false positives are happening way too often. Whether it’s corporate credits, retail credits, or sovereign credits, the costs of default probabilities being too high for good credits are just as important as the cost of default probabilities being too low for bad credits.
Donald R. van Deventer
Honolulu, May 18, 2010