Update on paper

Ok, so the paper will now be focused on entropy based diversification measures and models. I've noticed some searches on the website for this topic and it's interesting. The results I've seen so far indicate that using a entropy style diversification/concentration constraint results in a diversified portfolio will low turnover. Of course I need to grind out a complete set of results and check the statistics. I will post a draft of the paper to the website in the next few days.

Portfolio Optimization Paper

I've had some time to generate some results for my paper and I'm not really liking what I'm seeing so far. I'm hoping it's all due to some horrible programming mistake. I'm using a variety of methods and manipulations on the data and just about everything winds up with the same ex post sharpe ratio of 0.49. That is a fine number, I was just hoping to see some differences in the various results and they are all much closer together than I would have thought, even across dimensions like diversification and turnover.

I came across some new papers the other day and one has a simple algorithm for Simulated Annealing as applied to cardinality constrained portfolios. Of course I'm working on this right now. I already have an entropy based cardinality constrained method, I treat it just like a recommend level of diversification, but in the end it turns out to force the diversification to the constraint. The one thing tough to do with interior points is minimum size, something like an asset either has to have weight 0 or else it has to > 5%. This can be done pretty handily with the Simulated Annealing method.

What is this blog?

I realized anybody who just comes upon this might not have any context for this blog.

I'm working on an open source project on Asset Allocation and investment tools in Java. The project named akutan, for a volcano in Alaska, is hosted by sourceforge. I've got sample applets on my website www.jayw.com, one which shows off the variety of algorithms I've been working with the last 2 years, and another which shows off some work on the Black-Litterman model I've done.

I've also written a paper on Black-Litterman trying to yet again de-mystify the model by walking through all the equations from square 1. It is the good part about publishing on the web, there is no issue with too many equations and not enough inches of text!

Writing a new paper

Now that I have a first pass at the ranking/quintile/state preference optimization working I am going to write up a small paper demonstrating how it works and comparing it to some of the other algorithms which I've been working on. It seems to have two advantages, diversification and stability/robustness to errors in the estimates. These are the areas I will touch on.

I've been wanting to start a new paper for awhile and now I have a topic. I'm still thinking about my big book of Asset Allocation, but that I never did get started on that because I found more interesting things

Optimization Update

Tonite I checked in some of the State/Preference logic I've been working on lately, haven't built it into the applet for running on the site, but should get to that soon. There are two methods, one which optimizes entirely in state/preference space and then transforms into risk/return space and another which minimizes state/preference covariance subject to a constraint on return. Both methods can lead to kinked efficient frontiers unfortunately.

State/Preference Theory

So my latest interest is in State/Preference theory and how it relates to Portfolio Optimization. I read William Sharpe's new book, but it mostly had to do with asset pricing and the market, and very little with regards to trying to just use quintiles to optimize a portfolio. I'm working on building some code to do this. It is supposed to be more robust to estimation error, and more stable over time.