How Good are RiskMetrics' Tools?

Since I wrote about Glass-Steagall, CDSs [Credit Default Swaps], Blythe Masters and the Gramm-Leach-Bliley act in an older article, I had to reply to several private e-mails on how stupid the pricing of CDSs were. Now, most CDSs are priced and transacted in an unregulated environment. In fact, most of the larger investment banks used RiskMetrics' (RMG) tools.

Most educated investors know about the details of the Paramax/UBS lawsuit - where UBS (UBS) “insured” $1.3 billion in sub-prime loans for a premium of 0.155% per year. That is a premium of $1.55 million a year for a notional value of $1.3 billion. In fact, the numbers are in-line with actuarial numbers for insuring a house against the possibility of destruction from earthquake - but that has a 15% deductible. In the world of CDS’s, there are no deductibles. The insurer pays for all losses - including those from “markdowns”.

What bothers me here is whether Paramax used RMG’s software to figure out the premium associated with this CDS. If they did - how good is/are RiskMetrics’ tool[s]?  What if we regulated CDSs as insurance products [I am sure that the insurance commissioners will be clueless - compared to the Blythe Masters' of the world]?

Let’s play the devil’s advocate here.  Assuming that RMG’s models [till now] were excellent:

1. CDS pricing would have been perfect.  All of the “insurers” would have collected in premium - the present value of all possible payoffs multiplied by the probability of default at each time-point in the future.
2. If the above statement was true, then, all CDS insurers using RMG’s models would have collected “proper premiums” - that would have reflected the correct amount of risk taken on by the insurer.

Obviously, this wasn’t true in round #1. JPMorgan (JPM)| used RMG’s tools, and they sure did lose a lot of money [though they did not lose as much as Bear (BSC) or Lehman (LEH)]. RMG sure isn’t claiming that their software is adaptive [read - with the ability to guess correctly after the fact]. From the outside - looking in,

1. RMG’s software to predict CDS premiums wasn’t good the first time around.
2. RMG is not the only game in town as most companies have internal software pricing tools that are similar to RMG’s.  But I am confident that Goldman Sachs’ tools were better than the others’.
3. RMG is a new issue [less than a year listed]. While they do not have a track record of earnings, they sure do have excellent revenue growth and gross margins for what I could read filed at the SEC.

Bottom-line:  This is an issue that is worth keeping on one’s watch list.  With all the bad press regarding CDSs, I’m sure that RMG does not have an easy road ahead, but they are backed by some of the biggest names in the derivatives business, and they are relatively unknown [till now].

Disclosures: No positions in RMG or any banks for that matter.

Comments

[zenstockman:]

When I saw your title, "How Good are RiskMetric's Tools?" I assumed you were actually a user of those tools or had some statistical data to back up your evaluation. My assumptions were wrong, as could be your assumptions that RMG’s tools are somehow flawed or poor models for predicting risk.

In almost any statistical model the long-tail, which seems unheard of to investment professionals in smooth-sailing markets, is always a possibility. What is clear is that the more esoteric derivatives will always be a step ahead of those hoping to evaluate their pricing. I think this is part of an investment banker’s mantra (i.e. “Liar’s Poker” by Michael Lewis).

Please keep this “on one’s watch list” and let us know when you have something concrete. Thanks.

2008 Sep 26 04:58 PM

[Market Spec...:]

It would be nice if you actually knew the tool to have an opinion about it. The tool is only as good as the inputs you put in...for someone who has intimate knowledge of risk modeling, rest assured it was the tool that was flawed. It was who was doing the inputs that was flawed.

2008 Sep 26 10:32 PM

[whidbey:]

The tools are illusory at best. The assumption that one can capture all the risks and value them has been proven wrong time and again. In some years of teaching finance at a good university my research shows that the markets come very close to assessing risk best, but not in advance, hence the quest for tools. But, if one makes back casts the flows jump out and seeming defy capture in the model. Not hopeful and quants and add much except to explain HOW things failed.

2008 Sep 28 01:47 PM

[MichaelSchm...:]

Looking post hoc at the loss incurred versus premium collected with respect to a single loss is not a reasonable or fair way to determine the propriety of the premium. Almost any single insured loss will look improperly priced when viewed in that way. Only in the context of an entire portfolio, probably over an entire cycle, can the propriety of premium be fairly evaluated.

Actuarial analysis works reasonably well with life insurance, property loss, auto and a few other lines in which you have large, diverse pools of insureds and excellent long-term data. It is less than reliable when it comes to complex types of financial risk (D&O, E&O, surety), especially when the degree to which the individual insured risks will correlate is unknown. That said, and in the perspective of actual insurance (not CDS, which aren't technically "insurance"), $1.5M in premium for up to $1.3B in risk looks very low compared to how other types of complex risk is priced.

2008 Oct 20 12:53 PM