PhD Oral Exam - Hanieh Amjadian, Mathematics and Statistics

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Three Essays on Model Risk in Insurance and Finance Hanieh Amjadian Abstract This thesis examines what can be concluded from a statistical model when it is not assumed to coincide with the data-generating process. The first study develops a framework for model risk using policy-level property and casualty insurance data. Out-of-sample prediction errors correct the bias and quantify the predictive uncertainty, which separates model risk from estimation risk, and the same errors support a formal comparison of competing models. Poisson, negative binomial, normal, and XGBoost specifications illustrate the method.

The second study extends the framework to Value-at-Risk for large portfolios. Fitting a model and assessing its risk on the same data understates VaR, particularly under overfitting or misspecification. A separate calibration sample removes this circularity, and a bootstrap procedure handles heavy-tailed losses where the Gaussian approximation is inaccurate.

The third study concerns identification rather than estimation. Once the Poisson assumption on latent order counts is relaxed, several structural parameters of the Probability of Informed Trading model are no longer separately identified, but PIN itself remains recoverable from observable parameter combinations. This motivates closed-form method-of-moments and GMM estimators that target those combinations directly. It also yields a restriction that can be tested without estimating the model, since the baseline specification requires nonpositive buy–sell covariance. In cryptocurrency data that covariance is positive for every asset examined.