Boduroğlu, İ. İlkay2022-05-112022-05-112019978-3-030-05348-2978-3-030-05347-50302-97431611-3349https://doi.org/10.1007/978-3-030-05348-2_23https://hdl.handle.net/20.500.11776/612312th International Conference on Learning and Intelligent Optimization (LION) -- JUN 10-15, 2018 -- Kalamata, GREECEA risk measure that specifies minimum capital requirements is the amount of cash that must be added to a portfolio to make its risk acceptable to regulators. The 2008 financial crisis highlighted the demise of the most widely used risk measure, Value-at-Risk. Unlike the Conditional VaR model of Rockafellar & Uryasev, VaR ignores the possibility of abnormal returns and is not even a coherent risk measure as defined by Pflug. Both VaR and CVaR portfolio optimizers use asset-price return histories. Our novelty here is introducing an annual Desirability Value (DV) for a company and using the annual differences of DVs in CVaR optimization, instead of simply utilizing annual stock-price returns. The DV of a company is the perpendicular distance from the fundamental position of that company to the best separating hyperplane H-0 that separates profitable companies from losers during training. Thus, we introduce both a novel coherent surrogate risk measure, Conditional-Desirability-Value-at-Risk (CDVaR) and a direction along which to reduce (downside) surrogate risk, the perpendicular to H-0. Since it is a surrogate measure, CDVaR optimization does not produce a cash amount as the risk measure. However, the associated CVaR (or VaR) is trivially computable. Our machine-learning-fundamental-analysis-based CDVaR portfolio optimization results are comparable to those of mainstream price-returns-based CVaR optimizers.en10.1007/978-3-030-05348-2_23info:eu-repo/semantics/closedAccessPortfolio optimizationMachine learningRisk managementDownside riskConditional value at riskLinear programmingFundamental analysisInternational financial reporting standardsModelConference Object11353257270N/AWOS:0006119492000232-s2.0-85059948504Q3