Where Will Treasury Find Its Model?

Steve Hsu has some ideas on how Treasury's reverse auction might work; a new paper from NERA also considers the practicalities. The big question, of course, is how on earth one goes about valuing unique and hard-to-understand MBSs, CDOs, CDOs of CDSs, CDO-squareds, and other such exotica.

The NERA paper gives an idea of how it might be done, if it can be done at all. You start by finding relatively simple pass-through securities which are amenable to a standard reverse auction. Then you extrapolate:

Holding an auction may not be feasible if ownership of a distinctive and therefore unbundleable asset is too concentrated. For these non-auctioned assets, financial or statistical analysis may be used to estimate the value. This would involve calibrating pricing models to the newly available market data generated by auctions, to estimate the contribution to value of the various asset characteristics. Applying the estimated models to the non-auctioned assets would then yield a predicted price for each asset.

Is this realistic? I asked Marcia Mayer, one of the authors of the NERA paper, and she responded:

Indeed, pricing of multi-sector CDOs (which contain tranches of other CDOs) and CDO2s poses difficult modeling challenges. Prices for these securities have historically been model-based, but those model-generated results were benchmarked against recent transactions in similar securities. Since the summer of 2007, model-based values, even using assumptions adjusted for rising delinquencies and defaults, became increasingly difficult to validate in the face of the drying up of the liquidity in CDOs and other component securities. The auctions would generate publicly-reported transactions in component securities, which would greatly facilitate the pricing process. We recommend that the auctions begin with simpler products and move toward more complex products. For assets that are narrowly held and too idiosyncratic to bundle, prices obtained at auction for simpler assets--and their implications for discount rates, default rates, prepayment rates, etc.--could be used to estimate fair values.

In other words, if you have prices for simpler assets, and you have a workable model, then you can end up with decent prices for more complex assets. And the auction should generate nice transparent prices for the simpler assets.

But where is the workable model going to come from? One of the defining characteristics of this financial crisis has been the collapse of models which worked up until 2007 or so, and then failed, with devastating effects.

There are couple of people who seem to have built models which worked quite well: John Paulson and Andrew Lahde, for starters. Do you think they might be persuaded to give those models to Treasury, now that they don't need them any more?