A joint project of the Graduate School, Peabody College, and the Jean & Alexander Heard Library
Title page for ETD etd-03272017-164330
|Type of Document
||Nay, John Jacob
|Author's Email Address
||A Machine Learning Approach to Modeling Dynamic Decision-Making in Strategic Interactions and Prediction Markets
||Interdisciplinary Studies: Integrated Computational Decision Science
|Jonathan M. Gilligan
|George M. Hornberger
|Mark D. Abkowitz
- game theory
- prediction market
- machine learning
- genetic algorithm
|Date of Defense
My overarching modeling goal for my dissertation is to maximize generalization – some function of data and knowledge – from one sample, with its observations drawn independently from the distribution D, to another sample drawn from D, while also obtaining interpretable insights from the models. The processes of collecting relevant data and generating features from the raw data impart substantive knowledge into predictive models (and the model representation and optimization algorithms applied to those features contain methodological knowledge). I combine this knowledge with the data to train predictive models to deliver generalizability, and then investigate the implications of those models with simulations systematically exploring parameter spaces. The exploration of parameter space provides insights about the relationships between key variables.
Chapter 2 describes a method to generate descriptive models of strategic decision-making. I use an efficient representation of repeated game strategies with state matrices and a genetic algorithm-based estimation process to learn these models from data. This combination of representation and optimization is effective for modeling decision-making with experimental game data and observational international relations data.
Chapter 3 demonstrates that models can deliver high levels of generalizability with accurate out-of-sample predictions and interpretable scores of variable importance that can guide future behavioral research. I combine behavioral-game-theory-inspired feature design with data to train predictive models to deliver generalizability, and then investigate interactive implications of those models with optimization and sensitivity analyses.
Chapter 4 presents a computational model as a test-bed for designing climate prediction markets. I simulate two alternative climate futures, in which global temperatures are primarily driven either by carbon dioxide or by solar irradiance. These represent, respectively, the scientific consensus and the most plausible hypothesis advanced by prominent skeptics. Then I conduct sensitivity analyses to determine how a variety of factors describing both the market and the physical climate may affect traders’ beliefs about the cause of global climate change. Market participation causes most traders to converge quickly toward believing the “true” climate model.
|| Approximate Download Time
| 28.8 Modem
|| 56K Modem
|| ISDN (64 Kb)
|| ISDN (128 Kb)
|| Higher-speed Access
If you have more questions or technical problems, please Contact LITS