Do Portfolio Managers Model the Illiquidity Discount?

Neil Shah of Reuters brings up the perennial debate about "mark to model" behavior among fund managers. Such behavior is dangerous, because it gives those managers an incentive to be lazy. If the model gives them profits and the associated performance fees, then they have no incentive to start worrying about problems with it.

On the other hand, there are good reasons to "mark to model", and foremost among them is the fact that most of the instruments so marked are highly illiquid, which means that a "mark to market" system might well be even worse.

Shah, and Tanta, concentrate on one big weakness of the "mark to model" system, which is that the models can break. Either the data which got put into the model was internally flawed, or the model itself was flawed. Either way, you end up with a broken model, which can be very dangerous.

But I have a more basic question about these systems. "Mark to model" is, at heart, a replacement for a "mark to market" system, wherein the value of a portfolio is calculated every day. Losses can't be easily hidden in a "mark to market" system, because they show up as soon as the market falls. So my question is this: How much does the output of a "mark to model" system vary on a day-to-day basis?

Yes, I am worried about models breaking. But I'm also worried that a "mark to model" system might be really bad at reflecting many changes in the market, whether they're sudden and unexpected or not. The value of a CDO tranche is basically the present value of its future cashflows, discounted three times: once for the probability that those cashflows will not materialize (credit risk); once for the fact that it can't be sold (the illiquidity discount); and once for the possibility that those cashflows will materialize too soon (prepayment risk). On top of that is model risk – which is basically the chance that the model got one of those risks wrong.

What I'd like to know is how the markets model the illiquidity discount. If you're marking to model, it's easy to change the present value of your holdings according to changes in the Treasury yield curve, or even according to prepayment statistics. But how can you work out how much of a discount the market is requiring before it will buy illiquid paper? That discount changes over time, and should be modelled somehow.

In recent months and years, illiquid paper has traded at a much smaller discount than ever before. Clearly there's a risk that discount will widen out. Is that modelled? How? Do portfolio managers who "mark to model" take losses if the illiquidity discount goes up? Or do they simply declare that they're holding their investments to maturity, and therefore don't care what the illiquidity discount is? That would be intellectually dishonest, at best – because there's an opportunity cost to buying a security with a low illiquidity discount, in that you might be able to buy the same security at a higher illiquidity discount in the future. It's the kind of thing which really should be part of the model. Is it?