Share Email Print
cover

Proceedings Paper

Non-equilibrium thermodynamics theory of econometric source discovery for large data analysis
Author(s): Rutger van Bergem; Jeffrey Jenkins; Dalila Benachenhou; Harold Szu
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Almost all consumer and firm transactions are achieved using computers and as a result gives rise to increasingly large amounts of data available for analysts. The gold standard in Economic data manipulation techniques matured during a period of limited data access, and the new Large Data Analysis (LDA) paradigm we all face may quickly obfuscate most tools used by Economists. When coupled with an increased availability of numerous unstructured, multi-modal data sets, the impending 'data tsunami' could have serious detrimental effects for Economic forecasting, analysis, and research in general. Given this reality we propose a decision-aid framework for Augmented-LDA (A-LDA) – a synergistic approach to LDA which combines traditional supervised, rule-based Machine Learning (ML) strategies to iteratively uncover hidden sources in large data, the artificial neural network (ANN) Unsupervised Learning (USL) at the minimum Helmholtz free energy for isothermal dynamic equilibrium strategies, and the Economic intuitions required to handle problems encountered when interpreting large amounts of Financial or Economic data. To make the ANN USL framework applicable to economics we define the temperature, entropy, and energy concepts in Economics from non-equilibrium molecular thermodynamics of Boltzmann viewpoint, as well as defining an information geometry, on which the ANN can operate using USL to reduce information saturation. An exemplar of such a system representation is given for firm industry equilibrium. We demonstrate the traditional ML methodology in the economics context and leverage firm financial data to explore a frontier concept known as behavioral heterogeneity. Behavioral heterogeneity on the firm level can be imagined as a firm's interactions with different types of Economic entities over time. These interactions could impose varying degrees of institutional constraints on a firm's business behavior. We specifically look at behavioral heterogeneity for firms that are operating with the label of ‘Going-Concern’ and firms labeled according to institutional influence they may be experiencing, such as constraints on firm hiring/spending while in a Bankruptcy or a Merger procedure. Uncovering invariant features, or behavioral data metrics from observable firm data in an economy can greatly benefit the FED, World Bank, etc. We find that the ML/LDA communities can benefit from Economic intuitions just as much as Economists can benefit from generic data exploration tools. The future of successful Economic data understanding, modeling, simulation, and visualization can be amplified by new A-LDA models and approaches for new and analogous models of Economic system dynamics. The potential benefits of improved economic data analysis and real time decision aid tools are numerous for researchers, analysts, and federal agencies who all deal with increasingly large amounts of complex data to support their decision making.

Paper Details

Date Published: 19 June 2014
PDF: 9 pages
Proc. SPIE 9118, Independent Component Analyses, Compressive Sampling, Wavelets, Neural Net, Biosystems, and Nanoengineering XII, 911804 (19 June 2014); doi: 10.1117/12.2054927
Show Author Affiliations
Rutger van Bergem, George Mason Univ. (United States)
Jeffrey Jenkins, George Mason Univ. (United States)
Dalila Benachenhou, George Washington Univ. (United States)
Harold Szu, The Catholic Univ. of America (United States)


Published in SPIE Proceedings Vol. 9118:
Independent Component Analyses, Compressive Sampling, Wavelets, Neural Net, Biosystems, and Nanoengineering XII
Harold H. Szu; Liyi Dai, Editor(s)

© SPIE. Terms of Use
Back to Top