How did that pricing help you in 2008? My experience pricing MBS was that it ended up all being BS and none of the standard pricing algorithms did anything sensible.
The models are more or less capable and were. Of course over time models improve but the real issue is and always has been assumptions. That can be simple parameterizations or more complex PDE boundary conditions or in the mortgage market prepayment assumptions, loan loss provision assumptions,... We are very bad at producing sensible assumptions because we attempt to apply a long term average on all current environments, even stressed environments. I don't know whether we just don't know better or what it is. My point is that the models are overall pretty sound, many pricing models are used today that were used pre 2008 and during the crisis. Until better models come along this won't change. Always see the final valuation as an expected value rather than "price", because it actually is nothing but an expected value, an approximation given all the shaky assumptions that go into a model.
2008 was my best year, followed by 2011 - I tend to be on the right side of a lot of fuck-ups. BTW, the problem with MBS and CDO models was, predictably, the inputs As an up and coming quantitative trader, you are also depending on models and their assumptions, you just perceive them differenty. Models, none the less, are important - they tell you (a) that you are getting competitive pricing, for example when comparing dealer prices or when asking for an unwind or novation (b) give you a general sense of risk, including sensitivity to the model parameters (including non-observable ones) (c) allow you to compare risk factors rather then trying to wrap your head around what exactly is happening in your book if you have anything non-linear there
A very good point about the models being capable. I'm curious more about specific experiences. Like you said, garbage assumptions in, garbage values out. I'm wondering about those who had the correct assumptions input into their models and got reasonable valuations out and what they did that was different.
Funny you mention 2008. A big chunk of my pnl came from broker arbs. I mention it because you quite directly mentioned the importance of models. When I say broker arb I am of course talking about more liquid products, mostly otc index options, variance swaps,... It was insane how off different brokers were. There were days in 2008 where I booked 300k or 400k usd pnl risk free 2 hours into the day. Simple stuff such as Nikkei straddles. I mention it because everyone had/has a pretty solid and simple valuation model yet the input assumptions (implieds) were so off from one shop to the next that incredible amounts of money could be made off those mispricings.i remember I booked around 9 or 10 mil that year exclusively off the back of arbing brokers aside my then bread and butter.
This is why I was suggesting FINCAD's F3 to a poster elsewhere. F3 gives you first (and I think second) order sensitivity to your model parameters as well as market data. The only problem is that they are terrible at advertising this. I had to click around for 5 minutes to find this blurb: http://www.fincad.com/technology/features/risk-analytics "Comprehensive Greeks/sensitivities to every relevant risk factor" Anyway, an "edge" as a quantitative trader could be that you make more accurate assumptions as opposed to fancy schmancy models.
It's less about inputing the correct assumptions and more about (a) knowing the sensitivity to those assumptions (b) boundaries and historical levels for those assumptions Let's take the CDO disasters as an example. After all said and done, the correlation assumptions were off and that led to it being priced very aggressively. And it's not like they could not have figured it out - for example, for synthetic CDS it was very easy to realize that correlation was f*cked up (it was also surprisingly hard to short those things politically).