JoeyDVivre Wrote: ------------------------------------------------------- > Start with the problem of we don’t really know how > to model interest rates, default probabilities, > “correlation” of defaults and then add onto that > complex structures with prepayments and tranching > and it’s murderous. It’s easy to come up with a > model, even 500 models. It’s just impossible to > know if the model is any good. Yeah, a big institution wanted to contract me to come up with a pricing model for CDOs. I read up a bunch on the stuff and then told them I didn’t feel comfortable doing it. People said I was crazy - that this was my big break. All I could see is that I couldn’t figure out if the model was a good one, and I didn’t want to be the one being sued by a big institution because they made a bunch of trades with my model and the system blew up like it’s doing now.
bchadwick Wrote: ------------------------------------------------------- > JoeyDVivre Wrote: > -------------------------------------------------- > ----- > > Start with the problem of we don’t really know > how > > to model interest rates, default probabilities, > > “correlation” of defaults and then add onto > that > > complex structures with prepayments and > tranching > > and it’s murderous. It’s easy to come up with > a > > model, even 500 models. It’s just impossible > to > > know if the model is any good. > > > Yeah, a big institution wanted to contract me to > come up with a pricing model for CDOs. I read up > a bunch on the stuff and then told them I didn’t > feel comfortable doing it. People said I was > crazy - that this was my big break. All I could > see is that I couldn’t figure out if the model was > a good one, and I didn’t want to be the one being > sued by a big institution because they made a > bunch of trades with my model and the system blew > up like it’s doing now. That’s my thought too. I can make aa whole bunch of cool models but they all start with “Assume that interest rates can be modeled as in Black and Karanski and we will calibrate this to the swaptions market and that default correlation can be modeled using the copula found in …” That’s all very nice for academic papers but means nothing when you are betting the ranch on it.
Another problem with securitizing mortgages is the reduced ability of the customer to work out a payment plan with the bank. With a local savings and loan, when a customer can’t afford to pay, the local bank might allow them to skip a couple of payments (and tack it onto the principal). But when your loan was originated by a mortgage broker and eventually sold off to multiple IB’s or HF’s, the contract on the securitized mortage is much more rigid and foreclosure processes will start sooner without the opportunity to work out payments, reducing your willingness to pay further.
bchadwick Wrote: ------------------------------------------------------- > I don’t know those details, since I’ve been > staying away from that stuff like the plague. > However, Warren Buffet’s analogy to Weapons of > Mass Destruction is amazingly good. Anyone who > was anywhere near the anthrax letters in 2001 > knows that this stuff gets EVERYWHERE. > > Maybe someone who is a MBS specialist can tell us > more about recourse and whether monthly payments > were what gave it the AAA rating. I suspect it > may have more to do with the fact that principal > payments were coming in too (i.e. amortizing > loans), and therefore there wasn’t some huge cash > flow at the end that either came in or didn’t. Getting the AAA rating only really has to do with the amount of enhancement lies underneath each bond. For example, if your default assumtion is 5%, your excess spread assumption is 2%, you “stress” that assumption by a loss multiple. The loss multiple corresponds to the statistical distribution of loss probabilities. For example, the probability of a AAA loss may be .05%, thus, you need to model the bond to correspond to a .05% chance of losses. To get that probability you may need to stress losses by 5x and stress excess spread by 50%. Thus, your expected losses would be 25% and your excess spread, over the life of the bonds, is 1% per annum. If your WAL is 5 years, your spread covers 5% losses, giving you 20% losses. So, a AAA bond with a loss proxy of 5%, would require 20% enhancement underneath it before it defaulted. Now, the biggest problem is what your loss proxy is. If you underestimate it, you blow the whole analysis. That’s what happened. They had no history for low-credit, Alt-A, no-doc loans in highly bubbled economies. So they guessed.
JoeyDVivre Wrote: ------------------------------------------------------- > spierce Wrote: > -------------------------------------------------- > ----- > > > > Largely, it isn’t hard to model these things. > > However, you need to good historical data, that > > resembles the pool being securitized. > > > I don’t know what you mean by that, but I’ve been > to oodles of conferences and talks, read up on it > a bunch, am a reasonably smart and educated guy > and I think it’s a tremendous problem. > Again, assuming that underwriting standards and the underlying asset hasn’t changed much (rapid appreciation leading to depreciation) then the models should, reasonably, hold up. > Start with the problem of we don’t really know how > to model interest rates, default probabilities, > “correlation” of defaults and then add onto that > complex structures with prepayments and tranching > and it’s murderous. It’s easy to come up with a > model, even 500 models. It’s just impossible to > know if the model is any good.
equity_analyst Wrote: ------------------------------------------------------- > JoeyDVivre Wrote: > -------------------------------------------------- > ----- > > spierce Wrote: > > > -------------------------------------------------- > > > ----- > > > > > > Largely, it isn’t hard to model these things. > > > > However, you need to good historical data, > that > > > resembles the pool being securitized. > > > > > I don’t know what you mean by that, but I’ve > been > > to oodles of conferences and talks, read up on > it > > a bunch, am a reasonably smart and educated guy > > and I think it’s a tremendous problem. > > > > Start with the problem of we don’t really know > how > > to model interest rates, default probabilities, > > “correlation” of defaults and then add onto > that > > complex structures with prepayments and > tranching > > and it’s murderous. It’s easy to come up with > a > > model, even 500 models. It’s just impossible > to > > know if the model is any good. > > One of the big issues is that the historical data > going into the models to predict the future–I’m > thinking about home prices specifically–never > contained a down year in prices. People just > straight-lined the past. Linear perception to the > Nth degree. Some did include declines. For example, CA and Boston had their own declines.
juventurd Wrote: ------------------------------------------------------- > Another problem with securitizing mortgages is the > reduced ability of the customer to work out a > payment plan with the bank. With a local savings > and loan, when a customer can’t afford to pay, the > local bank might allow them to skip a couple of > payments (and tack it onto the principal). But > when your loan was originated by a mortgage broker > and eventually sold off to multiple IB’s or HF’s, > the contract on the securitized mortage is much > more rigid and foreclosure processes will start > sooner without the opportunity to work out > payments, reducing your willingness to pay > further. The Servicer has a lot of leeway in working out these problems. Usually the documents allow them to use “customary practices of the Servicing Standard”, more or less, provided that they aren’t doing f’d up stuff, they are OK. However, they are limited somewhat by modifying the terms, or deferring payments (extending the term). however, the Servicer can change their Credit and Collections Policies, they usually need to inform the Trustee, or get a 50% or 66% assention from the noteholders.
spierce Wrote: ------------------------------------------------------- > equity_analyst Wrote: > -------------------------------------------------- > ----- > > JoeyDVivre Wrote: > > > -------------------------------------------------- > > > ----- > > > spierce Wrote: > > > > > > -------------------------------------------------- > > > > > > ----- > > > > > > > > Largely, it isn’t hard to model these > things. > > > > > > However, you need to good historical data, > > that > > > > resembles the pool being securitized. > > > > > > > I don’t know what you mean by that, but I’ve > > been > > > to oodles of conferences and talks, read up > on > > it > > > a bunch, am a reasonably smart and educated > guy > > > and I think it’s a tremendous problem. > > > > > > Start with the problem of we don’t really > know > > how > > > to model interest rates, default > probabilities, > > > “correlation” of defaults and then add onto > > that > > > complex structures with prepayments and > > tranching > > > and it’s murderous. It’s easy to come up > with > > a > > > model, even 500 models. It’s just impossible > > to > > > know if the model is any good. > > > > One of the big issues is that the historical > data > > going into the models to predict the > future–I’m > > thinking about home prices specifically–never > > contained a down year in prices. People just > > straight-lined the past. Linear perception to > the > > Nth degree. > > Some did include declines. For example, CA and > Boston had their own declines. Either way, I guess we now know that underlying home price data–national, or, local, exhibits non-stationarity, a quality that was not reflected in the models.
spierce- Do you work with CDOs? I do not. I just joined CRM at my bank doing forecasting.
> Either way, I guess we now know that underlying > home price data–national, or, local, exhibits > non-stationarity, a quality that was not reflected > in the models. I certainly agree. I also think a lot of people knew the data didn’t reflect reality and hedged by increasing the loss proxy, or the loss multiple. However, it’s painfully obvious they didn’t increase it enough.
juventurd Wrote: ------------------------------------------------------- > spierce- > > Do you work with CDOs? I do not. I just joined > CRM at my bank doing forecasting. I am familiar with CDOs and have analyzed CDO and CLO structures and securities. However, I’ve never been an investor, or structurer, of a CDO transaction.