Like most geek-Americans, I've been taking the opportunity when folks start throwing up their hands and jumping out of windows to understand what went wrong. I've only got a keyhole-eye view to the drama behind the doors of Wall Street, but a little sleuthing is fun, no?
What I'm thinking about right now is how easy it might be to take a huge portfolio of loans, say 1500 or 15,000 or even 150,000 into an Essbase model and run a bunch of valuations on them and then have Essbase deliver in a quick query, whichever ones comply with the requirements you have. Technically, the product is well-suited to do that. To give you an idea, I bet a good analyst could run three or four of these every hour iteratively. And I could probably build the model in a week or two. The only variable would be all the handshaking that I would need to get from point A on this blog to point B onsite at your financial institution.
Oh and one more thing which would be that if there were proprietary valuation algorithms involved, I'd need a Java programmer's time and permission to use the algorithm(s). However the aggregation and selection of n- criteria? That's practically a no brainer, and we can scale as big as you like in terms of portfolio size on fairly ordinary hardware.
So I did some simple Googling and I found this deal called Cox proportional-hazards regression. Apparently that's one of the algorithms used to figure out valuations. I basically stole the idea from a TIBCO-like company's marketing data. The strength in Essbase is that it does aggregations and drilldowns very swiftly, and allows you to create very nice and interactive query sets. What you select on is really up to you. NuOPT, S-PLUS, S+Bayes, you call it.
It has been some time since I personally worked with an economist to generate some model algorithms, but if it can be explained mathematically, then it can be programmed. The niceness is that putting all of this into a multidimensional model on the Oracle/Hyperion Essbase platform is that you can communicate it all very quickly to any number of people, you can run models all day, you have full security and a mature product. The biggest advantage is that it's an open development platform for communicating financials of all sorts, so when you want to tweak your models, you're at an advantage. The downside is that this is not a special-purpose platform, so you have to take some time and explain exactly what you mean.
Considering what's been going on, that may not be such a bad idea. Let me know when you're ready to throw out your old financial models in this new world we're living in.
Speaking of which, I did have a preliminary discussion with a regional banker this past spring about what this software might do to speed up his processes. I was fairly convinced that I could reduce his time to process a customer's books in the loan process - you know munching everything out of MAS or Solomon in an hour or so instead of days of retyping.
I expect, that within the year, I will have worked with folks in my company to come up with a nice cloud model. I've already built the framework for data management and I know it scales up as large as it needs to. So it's just a matter of time and priority to see if we can come up with a profitable hosting solution.
To that end, I'm going to spend a bit more time when I have it, to deal with modeling. I'm going to hold my own personal modeling workshops with me and figure out what kinds of cool models I can come up with even before we have customers. So this idea of loan portfolio valuation, and interactive aggregation and multivariate selection is just a quick spinoff off the general idea.
How do I learn about this stuff! That sort of analysis sounds very interesting!
Posted by: Anthon | January 27, 2009 at 05:54 PM