Title: Differentiating asset classes
Authors: Dilip B. Madan
Addresses: Robert H. Smith School of Business, University of Maryland, College Park, MD 20742, USA
Abstract: Representing continuously compounded returns in seven asset classes, by their four bilateral gamma parameter estimates, a multiclass classification support vector machine is trained, on a sample of less than one percent of the data, to predict the asset class from which the returns were obtained. The asset classes considered are equities, volatility, commodities, foreign exchange, credit and bond indices and returns of hedge funds. Linear classification is observed to perform poorly. The use of seven binary learners makes some improvement and twenty one, one on one, binary learners deliver a good classification algorithm, also performing well out of sample.
Keywords: bilateral gamma model; digital moment estimation; asset allocation.
DOI: 10.1504/IJPAM.2018.092647
International Journal of Portfolio Analysis and Management, 2018 Vol.2 No.2, pp.99 - 113
Received: 06 Sep 2017
Accepted: 21 Feb 2018
Published online: 26 Jun 2018 *