> Far as I can tell, M-t-M accounting is perfectly legit accounting
> method in some circumstances. Essentially it allows a company to
> book as profit at the time the a deal is made or loss total value
> of a deal the gains or losses of which have not been fully realized.
There's a much more significant use in this case: estimating a prudent level of capitalization. With fractional-reserve banking, you have to have a way of answering the question: how much is enough? Well, it needs to be *some* percentage of *some* amount. What's the amount? The future is uncertain, so you have to estimate; then the question becomes: as of when? It's pretty easy to see that the value can be subject to volatility if you sample too often; and can be meaningless (or at best be a lagging signal) if you do it not often enough.
The problem in this case is there's a nasty feedback loop: even if you expect to hold a bond to maturity, MTM says "It's only worth N% today" -- and if it's going down, your capital reserves have to go up. And the usual way to raise cash is to sell assets. So by looking "too closely" you've made a fictional calculation become reality. The other danger, especially in thinly traded markets, is that a single data point can become meaningful, tripping another cycle: lather, rinse, repeat.
/jordan