Now showing 1 - 1 of 1
  • Publication
    Essays in Environmental and AI Finance
    (University College Dublin. School of Business, 2022)
    Entitled "Essays in Environmental and AI Finance," this dissertation consists of three self-contained essays. The first essay avails of capital market price signals to assess the presence and magnitude of economic incentives for clean innovation relative to dirty innovation. Second essay examines the utility and ethics of incorporating national culture profiling in bank-level machine-learning informed alert models relating to financial malfeasance. And the third essay tests state-of-the-art model-agnostic explainable AI (XAI) methods to uncover algorithmic injustice in the bank lending space. Essay 1 that seeks to bring new insights to the corporate environmental – financial performance debates examines how Tobin's Q is linked to 'clean' and 'dirty' innovation and innovation efficiency at the firm level. While clean innovation relates to patented technologies in areas such as renewable energy generation and electric cars, dirty innovation relates to fossil-based energy generation and combustion engines. A global patent data set covering over 15,000 firms across 12 countries helps uncover strong and robust evidence that the stock market recognizes the value of clean innovation and innovation efficiency and accords higher valuations to those firms that engage in successful clean research and development activities. The results are substantively invariant across innovation measurement, model specifications, estimators adopted, select sub-samples of firms and the United States and European patent offices. Essay 2 examines the utility and ethics of incorporating national culture profiling in bank-level machine-learning informed alert models relating to financial malfeasance. On a globally significant financial institution, binary classifier type alert models are used to establish the utility of dimensions of national culture in formulating anti-money laundering predictions. For corporate (individual) accounts, Hofstede individuality (individuality, and national-level corruption perception and financial secrecy) scores of the country in which a customer is resident, or from which a wire is sent/received, are of paramount importance. When combined with extensive account and transaction data against an even proprietary institutional algorithm, national culture traits markedly enhance the models' predictive performances. Against a global standard, ethical implications of ascribing values to dimensions of national culture are examined. We posit an ethical framework for the use of national profiling in anti-fraud alert models. Essay 3 provides evidence of the validity of Shapley model-agnostic explainable AI methods’ on real-world datasets. This work contributes initial evidence on the usefulness of Global Shapley Value and Shapley-Lorenz methods, with respect to racial discrimination in lending. Using 157,269 loan applications from the Home Mortgage Disclosure Act data set in New York during 2017, it is confirmed that the methods reveal evidence of racial discrimination inherent in the predictions of a transparent logistic regression model. Thus explainable AI can enable financial institutions to select an opaque creditworthiness model which blends out-of-sample performance with ethical considerations.
      280