Strategy out of an academic paper. 16.5 % return at 2.36 sharpe, max drawdown 6.9% through 1999-2010. Is this good, medium or bad? Equity curve, year by year statistics and relative performance here
Looks ok, but really these 2 pages does not give much information.
Hss this been verified by independent auditors, or is this backtest results? if this is backtest results, then I have seen a lot like it.
You say this is an academic paper. Is this present and open source?
Nice graph though 🙂
The strategy (implemented out of sample) above is based on the two papers below
Adding the dynamic S&P strategy to the dynamic commodity strategies results in a significant reduction in volatility.
You are right the 2 pages give much information. What else would you like to know?
These are backtest results or you could say paper traded results for a while ago..
I skimmed through the above article and would say that your results seem pretty realistic. I missed the part in the paper where you have a sharpe of 2.36 not sure if that was for all ten years or your best year. Looks to me like for an active strategy applied to a particular commodity the sharpes are between 1 an 1.5. Incidentally I have an intermediate strategy for ETFs that produces somewhat similar results ie active management does better than long only posns and sharpes in the same range as you. What happens if you toss out your most volatile commodities from the mix and instead of having ten use say the least volatile 5-7. Just curious I am going to review your paper more closer it looks pretty good.
sorry i did not read the google docs paper first so now i see where the 2.36 comes from
(a) how were the weights chosen
(b) if you are shorting assets do you adjust the sharpes in any way to account for the short position ie is sharpe still appropriate measure for a long/short basket as opposed to long only or would you use an adjusted sharpe.
Initially I skimmmed your other papers
Thanks for your interest
(a). The strategy is a combination of 3 equally weighted strategies (2 for commodities, 1 – S&P 500). Weights are chosen on equal weighting principles rather using mean-variance optimization on historical data.
(b). Our sharpes are simply calculated as the ratio of (Mean return – risk free) and (sigma of return). I see your point about sharpe for shorts, because return (or more specifically ROI in this case) would change for a short position.
Other statistics of our strategy is that for the 11 year period, it the maximum weekly loss was -2.39 percent and there were only 4 occasions when return was less than -2%. And also only 25 weeks when return was greater than 2%.
the papers you posted were very interesting. Thank you for posting them.
For some reason I can’t view\download the pdf from Google Docs. Any way, to be concrete, On a simple scale of bad, fair, good then 6.9% is good, 16.5% is fair and 2.36 sharpe is also fair (that is, if you’re after setting up your own hedge fund).
Thanks for the reply. Wonder why arent you able to access the google docs document. I can email it you if you like?
The results are too good to be true. The Devraj Basu and Alexander Stremme paper uses data from 1993-2007 to find a suitable model and the uses 1993-1998 to estimate the predictive model and then used the 1999-2007 as out of sample data. It is like not having any out of sample data at all. I wonder what was the performance during 2008-2011.
From my experience strategies that uses linear regressions would show performance with long periods of really good and really bad performances.
The performance of the backtest over 2008-2010 is given in the document attached to the comment. The Sharpe ratios were 2008: 1.80, 2009:4.71 2010:2.02. For 2011 the strategy was up around 5% until the end of June. THe statement that the Basu and Stremme paper used data from 1993-2007 to find an appropriate model is not entirely accurate. An unconditionally efficient portfolio strategy was chosen ahead of time and the in-sample period was used for parameter estimation via the predictive regression and the fitted model was then evaluated out of sample.In effect it could have been run in real time.
I wasn’t clear enough. I would like to see the performance of the strategy only for SP500 during 2008-2011. You are right my statement is not entirely accurate but I see an issue related to variable selection (Why COT+VIX ?) using the entire sample.
the email with links to this article was dumped into spam (i use gmail) because it said it has text frequently used by spamers. i see what they are talking about with those cheap air jordan ads.
How large of leverage do you use?
which platform u r using for doing back testing and how robust is ur platform..
Thanks for sharing the papers. I was wondering how did you back test this system? Did you back tested from 1999 – 2010 or year by year ( 1999- 2000, 2000 – 2001, 2001- 2002,etc), You may get completely different results.
NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!