Systematic Trading research and development, with a flavour of Trend Following
Au.Tra.Sy blog – Automated trading System header image 2

Back Online

June 28th, 2011 · 4 Comments · Blog

Hibernation over – I am back from my “self-imposed” off-the-grid isolation.

I will gradually resume posting on the blog, starting with an overdue update on Trend Following Wizards performance tomorrow (covering the results from May). It is so late in the month that there is not much point running a May edition of the State of Trend Following report: I will post a combined May+June State of Trend Following report in the first few days of July. Other posts will follow once I am back “in full swing”.

For those interested, I’ll leave you now with the answers to the “Summer puzzle” from my last post.

It is indeed possible to add a strongly anti-correlated instrument to a portfolio, without improving either return or risk of the portfolio. The “trick” was in the definition of “risk”: not the usual volatility/standard deviation but rather Max Drawdown.

The answer with two simple pictures:

And with the extra portfolio component:

Both the base portfolio and extra component have an identical (arithmetic) average monthly return, and their monthly returns are strongly anti-correlated, with a Pearson correlation coefficient of -0.87. However, the Max Drawdown of the combined portfolio is higher than that of the base portfolio.

I also attach the Excel spreadsheet for readers wanting to poke at, and check the numbers.

I obviously designed this example purely to play around and fit within the conditions of the “puzzle” but it does show that low or even anti-correlation is not necessarily the “magic bullet” – if calculated on a single timeframe only, as highlighted by Pumpernickel in the comments section below.

His/her comment actually represent a much better conclusion to this post. In summary: When looking for low or negatively correlated additions to a portfolio, do not focus on correlation calculations on a single timeframe (as per this “puzzle) but instead on multiple timeframes. Check the comment for a more detailed explanation from Pumpernickel.

Finally, the mountain pictured on the last post is Mount Kailash (as rightly guessed by some readers), one of the most famous mountains in Tibet – where I spent most of the time off in the past month.

Related Posts with Thumbnails

Tags:

4 Comments so far ↓

  • RiskCog

    Welcome back!

  • Ali

    good to see you back…

  • Pumpernickel

    ” … it does show that low or even anti-correlation is not necessarily the “magic bullet”.

    Take your Excel spreadsheet and calculate the correlations using 6-period returns rather than 1-period returns. You’ll find that the “base” and the “addition” are strongly positively correlated on this time scale. Therefore it is completely unsurprising that the combination of the two shows no improvement in performance.

    It’s wise to do this same thing when you’re thinking of combining two trading managers (or two trading systems). Calculate the correlation of their returns on (a) Yearly; (b) Monthly; (c) Weekly; (d) Daily timescales. Make sure the two returns series are uncorrelated or negatively correlated on all timescales.

    It’s problematic to do this with trading managers because (A) there are very few Annual returns datapoints so correlations may be meaningless; (B) you probably can’t get Weekly or Daily returns data; trading managers’ published track records are Monthly returns ONLY.

    However you *can* get daily, weekly, semi-monthly, monthly, quarterly, and annual returns data from backtest simulations. So if you’re thinking of combining two systems that you’re simulating yourself, calculate returns correlations on LOTS of timeframes. Build a mountain of evidence; compute time is very cheap!

  • Jez Liberty

    Pumpernickel,
    Thanks for dropping by, and providing a conclusion to this post, which is actually much better than mine.

    I have just updated the post conclusion based on your input..
    Cheers,
    Jez

Leave a Comment