Hoocoodanode will be going offline for security related maintenance at 11 PM PST tonight, and may be down for several hours. Sorry for the inconvenience.
If I read this correct, the models work with the latest correct current data but not with the old former correct current data that went south because the assumptions were wrong.
The equity market certainly seems to have decided that, whatever all that subprime stuff was about (which was really too boring and complicated to understand), it was a bunch of overrated noise and is well over now.
So, the debacle is now predictable--hey, that's good news! "The knife has been falling for some time, so we can predict with confidence that it will continue to fall." Comforting!
Okay. Now that the models are calibrated and correctly predicting current conditions and the data is verified can you please use the model as intended? Telling us the model verifies current reality is not the purpose of models. What is the model saying about Apr '08?
well, it all depends if conditions in Apr 08 are exactly like the conditions now, doesn't it? I'm still not convinced they've figured out how to run alternative scenarios. I just got a flash of parakeets staring into and pecking their mirrors repeatedly, but I'm not sure I could explain why.
Have you ever seen any parameters for the models for rating different slices of CDOs? I am wondering what they used as variables in the equation. For example did they use geographic location as a way of improving diversification. What about age? The only ones that are obvious to me would be FICO, D/E & perhaps ICR. (Of course those might not even br right:)) Are the models black boxes? Or do the openly share them with debt buyers?
I was chuckling the other day thinking about that equities shake up a month or so back and the stats guys said the odds were one and umphteen billion. Come on.....
I think scav is exactly right, in that what they need to do is run scenario analysis. Run a good case, a bad case, and a "those guys who horde guns and ammunition and gold were right" case. It's a more thorough approach, as it provides a starting point to do contingency planning--just in case everything doesn't come up roses, is there an exit strategy and/or hedging strategy that will limit losses?
But, of course, the models will get broken again, at which point they'll likely throw the models out the window and swear never to touch mortgages again, leaving the market open to someone who is ready to do serious analysis, as opposed to the back-of-the-envelope "yeah we're going to make lots of money" stuff that passed for quantitative analysis over the last few years.
Ok, dumb question, but I did my retirement planning with a Monte Carlo simulation widget. Don't these guys do the same, or are they just way smarter than me?
Reminds me of a con where 10000 marks were sent complementary predictions using a 'computer model' on who would win a football game. 1/2 of the marks were sent one name and 1/2 the opponent. The marks sent the wining team name were sent the names for the teams in the next game 1/2 got one 1/2 got the other name, and repeated the next time-each time only the winners received the next complementary predictions. So after about 7 times one set of 40 folks had 7 100% accurate predictions. They had been set the winning team's name each time while 9960 folks had one or more losing teams and had been eliminated from the scam.
The 40 were sent a notice that they had been sent 7 successful picks and reminded how much they could have made betting on those 7 games. They were told that for a mere $1000 they would be provided the winning teams name for the season. Most sent in the money.
Mike Bykhovsky's models are primarily for Agency-backed collateral, you know, FNMA, FHLMC and GNMA. There are tweaks for non-agency, but he is most likely not referring to those. These are not the CDO mark-to-model models that everyone is talking about. These are assumptions used during the projection of underlying cashflows. Nobody uses these models as-is except in the agency-backed world. Typically, a trader will put his/her own assumptions about prepayments, defaults and recoveries into their cashflow model. This is where the biggest problem lies.
I'm not in "the biz", but to this outside observer these models seem like just big curve-fitting exercises. Even the parameters that get plugged in for future events are based on historical data. If they aren't used by a very sophisticated person it ends up just yielding raw extrapolation. Sure, that works until the 2nd derivative changes and then look out. But then, 6 months after your model is blown, plug in the most recent data and LOOK the model works again!
Is a freshman year book on numerical methods all you need to be a top-flight financial analyst? Am I missing something here?
Haldane, the U.K. central bank's head of market infrastructure, called for a ``fundamental rethinking of the way we do stress testing,'' in a speech last week in Chicago.
"If we feed current [Home Price Indices] and projected HPIs into the model, the resulting delinquency output is very much consistent with what we are observing."
Uh, so the Monte Carlo six-sigma stress test never plugged in the case of housing prices going down?
Seriously, there's a lot of mockery of modeling here, but if this is true, then the models themselves ain't the problem, it's the lack of reasonable stress testing when passed to the simulations group. If true, the models would have worked, had they not been offered only rosy historical scenarios.
This sounds persnickety, but it's actually an important point, because it means with more realistic scenarios, we could actually price some of this junk again...
(Shame that realistic testing would likely show that the junk... is junk.)
Vader,
Many stock brokers get their start doing the same thing. My friend knew of one years ago who used tennis tournament draw sheets to track who he told stock A was going up or down, B up or down, C up or down.
He tried to get the semi-finalists to indicate their level of income through small talk. The lower income people were sold the dogs to try to pump them up a bit. They attempted to maintain a longer relationship with the higher income people.
"And so it goes" - Slaughterhouse 5
hello i think this is great to you
Costa Blanca estate agent offers spanish property for sale from apartments to villas new builds and resales. spanish property for sale
If I read this correct, the models work with the latest correct current data but not with the old former correct current data that went south because the assumptions were wrong.
First?
S.
The equity market certainly seems to have decided that, whatever all that subprime stuff was about (which was really too boring and complicated to understand), it was a bunch of overrated noise and is well over now.
Amazing, isn't it, Vader? Put plausible numbers into the machine, and damn us, that mark-to-model thing actually works! Yawn.
So, the debacle is now predictable--hey, that's good news! "The knife has been falling for some time, so we can predict with confidence that it will continue to fall." Comforting!
Okay. Now that the models are calibrated and correctly predicting current conditions and the data is verified can you please use the model as intended? Telling us the model verifies current reality is not the purpose of models. What is the model saying about Apr '08?
Zucchini's I noticed at the market were 1.99 /lb, verifying the prediction I made when I saw the sign.
Why can't I be in print?
well, it all depends if conditions in Apr 08 are exactly like the conditions now, doesn't it? I'm still not convinced they've figured out how to run alternative scenarios. I just got a flash of parakeets staring into and pecking their mirrors repeatedly, but I'm not sure I could explain why.
Boring? Not to me. This is fun to watch. Can you make it move a little faster? Maybe that would raise the excitement level.
Tanta, (or others)
Have you ever seen any parameters for the models for rating different slices of CDOs? I am wondering what they used as variables in the equation. For example did they use geographic location as a way of improving diversification. What about age? The only ones that are obvious to me would be FICO, D/E & perhaps ICR. (Of course those might not even br right:)) Are the models black boxes? Or do the openly share them with debt buyers?
I was chuckling the other day thinking about that equities shake up a month or so back and the stats guys said the odds were one and umphteen billion. Come on.....
I think scav is exactly right, in that what they need to do is run scenario analysis. Run a good case, a bad case, and a "those guys who horde guns and ammunition and gold were right" case. It's a more thorough approach, as it provides a starting point to do contingency planning--just in case everything doesn't come up roses, is there an exit strategy and/or hedging strategy that will limit losses?
But, of course, the models will get broken again, at which point they'll likely throw the models out the window and swear never to touch mortgages again, leaving the market open to someone who is ready to do serious analysis, as opposed to the back-of-the-envelope "yeah we're going to make lots of money" stuff that passed for quantitative analysis over the last few years.
Ok, dumb question, but I did my retirement planning with a Monte Carlo simulation widget. Don't these guys do the same, or are they just way smarter than me?
First?
Sebastia
Robert Coté
Reminds me of a con where 10000 marks were sent complementary predictions using a 'computer model' on who would win a football game. 1/2 of the marks were sent one name and 1/2 the opponent. The marks sent the wining team name were sent the names for the teams in the next game 1/2 got one 1/2 got the other name, and repeated the next time-each time only the winners received the next complementary predictions. So after about 7 times one set of 40 folks had 7 100% accurate predictions. They had been set the winning team's name each time while 9960 folks had one or more losing teams and had been eliminated from the scam.
The 40 were sent a notice that they had been sent 7 successful picks and reminded how much they could have made betting on those 7 games. They were told that for a mere $1000 they would be provided the winning teams name for the season. Most sent in the money.
So much for models.
Still in the denial stage. Wow.
Wow, I'll have to remember that one vader. Oh, and remind me not to take any game picks from you..
apparently wachovia gets different remits than everybody else.
Mike Bykhovsky's models are primarily for Agency-backed collateral, you know, FNMA, FHLMC and GNMA. There are tweaks for non-agency, but he is most likely not referring to those. These are not the CDO mark-to-model models that everyone is talking about. These are assumptions used during the projection of underlying cashflows. Nobody uses these models as-is except in the agency-backed world. Typically, a trader will put his/her own assumptions about prepayments, defaults and recoveries into their cashflow model. This is where the biggest problem lies.
I'm not in "the biz", but to this outside observer these models seem like just big curve-fitting exercises. Even the parameters that get plugged in for future events are based on historical data. If they aren't used by a very sophisticated person it ends up just yielding raw extrapolation. Sure, that works until the 2nd derivative changes and then look out. But then, 6 months after your model is blown, plug in the most recent data and LOOK the model works again!
Is a freshman year book on numerical methods all you need to be a top-flight financial analyst? Am I missing something here?
where did Tanta go? i bet she fell asleep modeling prepayments in the next few months for abx 06-2. yawn.
Haldane, the U.K. central bank's head of market infrastructure, called for a ``fundamental rethinking of the way we do stress testing,'' in a speech last week in Chicago.
Greenspan Sees `Rethinking' on CDOs After Subprime Collapse - Bloomberg.com
"If we feed current [Home Price Indices] and projected HPIs into the model, the resulting delinquency output is very much consistent with what we are observing."
Uh, so the Monte Carlo six-sigma stress test never plugged in the case of housing prices going down?
Seriously, there's a lot of mockery of modeling here, but if this is true, then the models themselves ain't the problem, it's the lack of reasonable stress testing when passed to the simulations group. If true, the models would have worked, had they not been offered only rosy historical scenarios.
This sounds persnickety, but it's actually an important point, because it means with more realistic scenarios, we could actually price some of this junk again...
(Shame that realistic testing would likely show that the junk... is junk.)
Vader,
Many stock brokers get their start doing the same thing. My friend knew of one years ago who used tennis tournament draw sheets to track who he told stock A was going up or down, B up or down, C up or down.
He tried to get the semi-finalists to indicate their level of income through small talk. The lower income people were sold the dogs to try to pump them up a bit. They attempted to maintain a longer relationship with the higher income people.
"And so it goes" - Slaughterhouse 5
hello i think this is great to you
Costa Blanca estate agent offers spanish property for sale from apartments to villas new builds and resales.
spanish property for sale
Looks like the building boom in Spain may have overshot the demand too....