I'll be in Chicago this Monday Aug 16 to Wednesday Aug 18. I'm visiting to try to find out where I should go after graduating college this year. Let me know if you want to get coffee/lunch/dinner. maxdama at berkeley.edu
Most people are unaware that standard functions for estimating correlation, variance, and standard deviation from languages like Matlab and R are biased:
Most people use SD as a measure of volatility, so their estimates of volatility will be too low.
Most people use correlation in portfolio construction or pairs trading, so this implies their portfolios and pairs will be far from optimal.
These problems simply compound the other problems of price data not following a Gaussian distribution, not being stationary, etc etc.
Genetic algorithms are a general type of parameter optimization procedure, applicable to almost any problem, although not usually very fast. This note has a clear explanation of some of the practical details of their implementation - http://www.nashcoding.com/?p=28 - and it gives an example of how to implement what I previously referred to as meta-optimization, under the sub-section heading 'Self-Adaptation'.
Big-O is the mathematical way of discussing an algorithm's efficiency. It's important to know how to optimize code, especially for backtesting quickly and for higher-frequency strategies. Here's a really good explanation of this notation so you can better understand things you might read or just to help you think in a more principled fashion about your program flow: http://stackoverflow.com/questions/487258/plain-english-explanation-of-big-o/487278#answer-487278
http://www.bjmath.com/bjmath/kelly/mandk.htm
And something else relating to a previous note on Kelly- it turns out the keyword that people use to discuss things like I covered in this Kelly = Markowitz note is 'Geometric Mean Approximations'; here's another example.
I just read this good concise history of money management / position sizing: http://finance.martinsewell.com/money-management/
And here's a mirror on my server since the page doesn't look too stable.
On the topic - Here's a good recent (2009) paper on Kelly optimization applied to portfolios: http://arxiv.org/abs/0712.2771. It has a very readable introduction section with references to all the major work on extending Kelly. And their results are interesting as well.
Subscribe to:
Posts (Atom)
