Three Essays Applying Dynamic Models in Economics, Finance, and Machine Learning

<p>This dissertation is a composition in three parts. Collectively, these essays investigate dynamic methods and their application in the fields of Economics, Finance, and Machine Learning. It pulls liberally from all three. In particular, this dissertation makes repeated use of multi-state modeling frameworks popular in Economics to bring a faceted view to the underlying data and detect its hidden heterogeneity. The challenge of modeling financial assets and estimating their dependence is another focus. For stimulus, concepts in the Machine Learning field are brought in to aid or compete with established econometric techniques. <strong></strong></p> <p><strong>Econometric Applications of the Hierarchical Mixture-of-Experts</strong></p> <p>In this essay, a novel mixture model is studied. Named the Hierarchical Mixture-of-Experts (HME) in the machine learning literature, the mixture model utilizes a set of covariates and a tree-based architecture to efficiently allocate each observation to the most likely local regression. The nature of the conditional weighting scheme provides the researcher a natural interpretation of how the local (and latent) sub-populations are formed. Marginal effects, robust standard errors, and model selection are also discussed. The model is demonstrated by estimating a Mincer wage equation using US census data and occupational skills data from the Occupational Information Network. Several Monte Carlo exercises are carried out to better understand the behavior of the model on simulated datasets with varying degrees of heterogeneity.</p> <p><strong>A Comparison of Time-Varying GARCH Models</strong></p> <p>This essay uses three separate tools to model the evolution of the dependence structure between major global currencies from 2000 to 2018. It combines the ARMA-GARCH approach to univariate times series modeling with the conditional Copula, a flexible class of distributions with unique properties. Turning the flexibility to eleven, a selection of multi-state modeling strategies are discussed, enabling changes of dependence between finance assets over time. The framework is applied to the log returns of the exchange rates between the US Dollar and the European Euro, British Pound, and Japanese Yen. <strong></strong></p> <p><strong>Estimating Dynamic Time Warping in the Presence of Non-Stationary Time Series</strong></p> <p>This essay offers an overview of Dynamic Time Warping as a measure of time series similarity and its application in the field of Finance. Pearson's coefficient of correlation, indispensable to the field in its own right, provides a useful foil. Particular attention is given to the presence of random-walks in financial time series and their impact on the distribution of the dynamic time warping distance.</p>

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Economics

Grantor

The Graduate School and University Center of The City University of New York

Year dc:date.available

2024

Author dc:creator

• Dowiak, Lucas C.

Advisor dc:contributor.advisor

• Wim P.M. Vijverberg

Committee members dc:contributor.committeemember

• Chu-Ping Vijverberg
• Sebastiano Manzan

• Data Science
• Economics
• Finance
• Econometrics
• Multistate Models
• Pair-Trading
• Dynamic Time Warping
• Feedforward Neural Networks

Repository record dc:identifier

https://academicworks.cuny.edu/gc_etds/5805

OAI identifier oai:identifier

oai:academicworks.cuny.edu:gc_etds-6954