JD Opdyke, Chief Analytics Officer at DataMineit, LLC, will be speaking about "The Highly Versatile Angles Space of Positive Definite Dependence Measures: Causal Discovery, Inference, Sampling, and Generalized Entropy" on Wednesday, January 28th from 3:30 - 4:30pm on Zoom.
Zoom Link:
https://ucsb.zoom.us/j/86037405872
Meeting ID: 860 3740 5872
Title:
The Highly Versatile Angles Space of Positive Definite Dependence Measures:
Causal Discovery, Inference, Sampling, and Generalized Entropy
Abstract:
The angles space of positive definite dependence measures possesses many desirable qualities that make it useful and of interest for a wide range of empirical inquiry, including: 1. causal discovery; 2. inference for and sampling of these dependence measures; and 3. distance measurement using these measures. Positive definite dependence measures arguably span all those in widespread usage, so this approach maintains a broad and relevant range of application. Angles correspond one-to-one with cells of all-pairwise dependence measure matrices, placing them at the right level of granularity for most analyses of dependence structure, as opposed to spectral approaches related to eigen decompositions that remain an order of magnitude less granular, not to mention less robust for estimation when matrices approach singularity (which is the rule rather than the exception for financial portfolios in many settings). Angles distributions are multivariate independent, well bounded, and generally well behaved (i.e. unimodal, not extremely asymmetric, etc.), allowing for reliable estimation and more robust inference. They also remain robust to challenging data conditions, because their only requirement is the positive definiteness of the dependence measure matrix. Real-world financial returns data, for example, is characterized, simultaneously, by marginal distributions with different and varying degrees of asymmetry, heavy-tailedness, serial correlation, and non-stationarity: these conditions in themselves pose no issues for the definition of angle variables or their distributions. Finally, because angles are a multivariate bijection of any positive definite dependence measure matrix, they provide a unifying framework allowing for ceteris paribus comparisons across dependence measures and/or across estimators applied to the same measure, often where no such comparative analyses previously were tractable or even possible. Relying on the angles space, I provide examples of original derivations and empirical implementations of 1., 2., and 3. herein.
Professional Bio:
JD Opdyke, Chief Analytics Officer at DataMineit, LLC, is a senior data scientist of over 30 years in the investment and risk analytics space. He has strong and extensive experience across major financial verticals (capital markets (exotics/complex derivatives products, sovereign wealth fund, venture capital), banking (corporate, retail mortgage, credit card), insurance (enterprise portfolio risk)), as well as decades of data science expertise as a consultant. JD has built and led several senior quant teams, published 14 peer reviewed journal papers and book chapters, several of which were voted ‘Paper of the Year’ by panels of experts, and was invited to write a financial portfolio risk / causal modeling book currently under editorial review. He is a frequently invited speaker/presenter at top quant and risk conferences globally.
Most recently JD was Chief Analytics Officer and Senior Managing Director at Sachs Capital Group Asset Management, LLC where he hired and lead a team of senior modelers to optimize the construction and allocation of SCG’s equities-based complex derivatives portfolios. He also derived the firm’s IP for transforming discrete, infrequent “looks” into continuous time series and robust oracle-property parameter estimates for fully systematic multi-period alpha capture. JD came to SCGAM from the Abu Dhabi Investment Authority (ADIA) where he was quant lead in a rapidly growing alpha factory. Prior to that, he was Head of Enterprise Risk Analytics at Allstate, where he was responsible for modeling and estimating enterprise-level economic capital for the firm’s entire portfolio, across all risk silos and business lines. JD held a similar role at GE Capital and has consulted to numerous banks and financial organizations, including Barclays Capital, Wells Fargo, American Express, Northern Trust, and Deloitte Consulting.
JD earned his Bachelor's, with honors, from Yale University, his Master’s degree from Harvard University where he was awarded two paid, competitive Fellowships, and he completed a post-graduate fellowship in MIT’s graduate mathematics department as an Advanced Study Program Fellow. He serves as review editor of several journals, including Artificial Intelligence in Finance.