Carl Noble At Pit.AI Technologies we do not have the pretension to assume that we know what model a real-world phenomenon truly follows, let alone make a general claim on whether the i.i.d. Gaussian model represents any real-world phenomenon. Instead, we believe that all models are wrong, but some are more useful than others. The point of this two-part post is to illustrate using empirical evidence that Pearson’s correlation is not as useful a measure of financial risk as practitioners often think, and to propose an information-theoretic alternative that addresses its limitations. Part I (this post) establishes the connection between Pearson’s correlation, linear regression, CAPM’s beta, and i.i.d. linear factor models to illustrate how pervasive the issue is. I invite you to read Part II, where we empirically illustrate the limitations of Pearson’s correlation as a measure risk, and propose the information-adjusted correlation as an alternative that addresses said limitations.

    Yves-Laurent Kom Samo

    Written by

    Mathematician, Coder, ML PhD (Oxford), Google PhD Fellow in ML (2016), ex-Goldman Quant. Solving Intelligence for Investment Management at Pit.AI Technologies.