Steve Mayo![]() Legend ![]() ![]() ![]() ![]() Posts: 414 Joined: 10/11/2012 Location: Austin, TX ![]() | Hi Ian and Steve Thank you for sharing your spreadsheet. It does an excellent job of showing how the component strategies and symbols contribute to the overall portfolio performance. I like that it highlights how just a few stocks generate the bulk of the return in most portfolios. In the portfolio I tested, for example, it was interesting to note that 50% of the return came from only 28 stocks out of the 500+ that were traded! Mark Holstius and I have been working on portfolio optimization for almost a year now, using an approach that involves automated testing of thousands of strategy combinations. Others in the forum have reported using the recent short-term best-performers in sort of a momentum approach. Keith McIntyre has been promoting sequential strategy assembly. There are lots of interesting approaches. The difficulty I find with all these various approaches, however, is the question of predictability. Can we be confident that our candidate portfolio will again identify those 28 or so high-flyers next year? I note that the standard deviations of the monthly returns are multiples of the means on most, if not all, strats/ports I have studied. Taking ARM4 Margin as an example of a good portfolio, there is a 5% probability of a 16% (2 Sigma) drop each month for a portfolio that averages about 5% return/month. [This assumes a normal distribution whereas the distribution graph shows the port actually has fatter tails.] In short, there's a lot of variability in these portfolios. Of course, that's a good thing on the upside. :-) To get back on point, reviewing your spreadsheet, it appears your approach is to identify strategies for replacement within a candidate portfolio, seeking to maximize consistency foremost. For comparison purposes with other "optimizers", would you be willing to share some metrics from your best-performing optimized portfolio so we can compare? I would also be interested to learn how well your live-trading results match your optimized model. Meanwhile, I will share some of our candidate portfolios and risk measures; we haven’t yet traded these live yet. The attached graphs show some of the metrics used to compare 3 recent candidates against Nirvana’s ARM4 Margin portfolio (our benchmark) as well as the overall market. Our best candidate in this series (2013-1-1 v2) has a 6-year CAGR of 90%, MDD of 32% for a Calmar of 2.8. The Modigliani and Sortino ratios show it is the better candidate of this group on a risk-adjusted, benchmark-relative basis. However, the non-parametric box plot and the overlapping confidence intervals show that, statistically-speaking, the difference in mean monthly returns between these 4 portfolios are not significantly different (p=0.6 by ANOVA). Any material different result one might get trading any of these would most likely be random chance. Statistics aside, I think you are on the right track with making consistency your primary optimization metric. Optimizing for maximum return alone, in my opinion at least, is likely to result in a portfolio that will not repeat its luck going forward. [Edited by Steve Mayo on 1/27/2014 4:06 PM] ![]() |