Back to Community
Hierarchical Clustering: HowTo

enjoy

Loading notebook preview...
Notebook previews are currently unavailable.
4 responses

Interesting to read, a comparison between stock clusters and sectors:

http://badassdatascience.com/2014/12/31/clustering-stocks-1/
http://badassdatascience.com/2015/01/01/clustering-stocks-2/

If stocks aggregate in clusters that cross sector boundaries, then what value might have a sector neutral strategy? And so the sector risk exposure? Shouldn't the algorithms be "cluster neutral"?

If stocks aggregate in clusters that cross sector boundaries, then what value might have a sector neutral strategy? And so the sector risk exposure? Shouldn't the algorithms be "cluster neutral"?

I suppose the idea behind limiting sector exposure is one of risk management through diversification in a way that is familiar to investors and has a historical basis codified by Morningstar and others. And I'm figuring that traditional hedge funds will have analysts, traders, directors, etc. specializing in industry sectors, so that the risk manager has to keep them all in balance, to keep any one interest from getting too big a piece of pie.

Regarding making algorithms cluster-neutral, I would think just the opposite. If the cluster concentration risk is not captured in the Quantopian risk model, then load up on a cluster if it might be advantageous. Roll the dice. The whole point here is to pull one over on Wall Street, and so if Quantopian is following an invalid traditional path, then it might be an opportunity for algo authors, by introducing unexplained risks that can be packaged as new alpha.

My idea comes from "factor based equity investing, is the magic gone?:

To avoid concentration risk, as certain industries often dominate a
factor at a given period, the factor model spreads its exposure
systematically across the equity market. A natural choice is to use
industry classifications such as GICS, but from a risk management
perspective we are more interested in correlation clusters

this is something I've been looking into for awhile. this paper is similar to the research above but uses fundamentals to cluster stocks rather than time series of returns

www.cfapubs.org/doi/pdf/10.2469/faj.v73.n3.5