I want to algorithmically add a new investment to an existing portfolio. The decision should be based on a low correlation to the existing assets.

E. g. the following situation

  • The portfolio already contains 5 stocks: A1 ... A5
  • My trading system generates 3 buy signals for new investments that aren’t in the portfolio yet: B1 ... B3
  • I want to pick one out of three investments to add to the portfolio

So I will calculate for each investment the correlation coefficient: correlation(Bx, Ay) which gives me a matrix with 3 rows (B1 ... B3) and 5 columns (A1 ... A5). What will then be the right decision criteria to pick the row with the least correlation?

  • The row with the lowest correlation average?
  • Or the lowest correlation average weighted by invest volume?
  • Or the row with the lowest maximum correlation?
  • Other variants?

Other topic: Am I right that I just have to take take the negative value of the correlation coefficient in case my system also creates short-signals?



2 Answers 2


If you want the lowest correlation then just short your portfolio. Correlation -1, now you have zero exposure. But I don't think that's really your objective function.

For a two asset portfolio ... $$ \sigma =\sqrt{w_1^2\sigma_1^2 + w_2^2\sigma_2^2+2w_1w_2Cov_1,_2} $$

And your real objective function incorporates your return expectation.

Bear with me now, because we're just stating the obvious so far. And I haven't been thinking through the generalization to more assets lately nearly as much as calling library functions to do this and as sunny as it is outside, I'm in no mood to do matrixes in markdown or stare at the equation until I'm sure I generalize right. I'm going to suggest you go have a look at this explanation with a spreadsheet.

On to some of the assumptions about where to look for your non-correlated positive return expectation.
You can't assume that inverting your signal generation will give you a set of low correlation signals in the form of shorts. It might look that way on daily returns, but my experience with long/short systems of this nature is that the long and short can be highly correlated on time frames longer than one day. Furthermore, you have different market structure issues with shorts. I see some feasibility issues in terms of share availability, costs issues (there's a highly profitable sub-industry thriving off of lending shares to folks like you and me), and stocks just behave differently going down than going up.

You have to diversify in terms of signal generation methods entirely. This might apply more to some strategies and less to others. You should certainly test for it in your strategies. Even hope you will find it if you must. But run the numbers and don't assume.

Keep diversifying. And I certainly don't mean to suggest you shouldn't trade some inverted signals, because most people I know do that. Just don't expect them to be negative correlations or even the best diversification for your longs. Uncorrelated alpha is the holy grail. Happy questing!


Personally I would look to increase the asset which has the highest positive co-skewness with your existing portfolio as this should both enhance your return and in general reduce your correlation. For example see the following spreadsheet http://www.academia.edu/attachments/31382573/download_file?st=MTQyNjcyODcyNywyMDIuMTc0LjE3MC4xNjIsMTIyMTAxMg%3D%3D&s=work_strip


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.