Three essays on the application of machine learning for risk governance in financial institutions

Thumbnail Image
Owusu, Abena Fosua
Issue Date
Electronic thesis
Research Projects
Organizational Units
Journal Issue
Alternative Title
The first two essays of my dissertation focus on risk culture as an important element for risk governance and risk management in U.S. banks and insurance firms. The third essay identifies climate change risk as an emerging risk impacting the financial industry and explores insurance firms' assessment and adaptation to climate change risks. Due to the challenges in defining and measuring abstract concepts such as risk culture and risk response, I apply big data analytical tools, specifically textual analysis and machine learning techniques, to identify financial institutions' risk culture and response to climate change risks using information disclosed in their 10-K regulatory filings with the U.S. Securities and Exchange Commission (SEC).
The second essay examines the impact of regulation on the risk culture of U.S. insurance firms to address the call for federal regulations in the insurance industry after the 2007-2008 financial crisis. By applying machine learning techniques to define insurance firms' risk culture, I assess how risk culture changes in the insurance industry over time. My findings show a positive impact of the federal Dodd-Frank regulation on risk culture in the insurance industry. I find that the risk culture of insurance firms is significantly defined by their risk strategies, decisions, and recruitment and training practices. A spatial and temporal prediction analysis of risk culture shows that, over time, large insurers who downgrade into a poor risk culture state have a higher probability of remaining in this downward trend relative to large insurers that improved their risk culture status. Similarly, I observe an improvement in risk culture of large insurers after the Dodd-Frank Act was passed, compared to non-large insurers.
In the first essay, I apply text mining and unsupervised machine learning algorithms to identify the risk culture of U.S. bank holding companies using risk culture-sentiment features extracted from their 10-K annual filings. I find that the uncertainty, litigious and constraining sentiments associated with the leadership, strategy and portfolio of the banks are significant in defining the banks' risk culture. Cluster analysis of these features proposes three distinct risk culture clusters, which can be labeled as good, fair and poor. Examining the relation between risk culture and performance, I find that banks with a sound (good and fair) risk culture are characterized by high profitability ratios, bank stability, lower default risk and good governance, consistent with regulatory expectations.
This dissertation consists of three essays on risk governance in financial institutions that underline the importance of risk culture and risk assessment and response for effective risk governance. According to the International Finance Corporation (IFC), risk governance focuses on applying the principles of sound corporate governance to the identification, management and communication of risks. The 2008 global financial crisis highlighted the lack of firm values, ethics and governance structure that understands, identifies and manages risks as they evolve in the financial industry. With increasingly frequent climate disasters in recent years, investors are also interested in financial firm's governance of climate-related issues they face in their Environmental, Social and Governance (ESG) investments. Hence, efforts by governments and regulators to enhance risk governance in the finance industry have addressed topics related to the risk culture of financial firms and regulatory disclosure of firms' response to risk factors that pose threats to the industry's financial stability and soundness.
In the third essay, I assess and distinguish between insurance firms' response to climate change risks, and examine how their climate change risk exposures relate to their financial and governance characteristics. Using a text mining approach, I build a climate change dictionary with three sub-dictionaries - risk exposure, impact and response, and apply a `nested structure' feature extraction approach to extract, define and classify insurance firms' adaptation levels to climate change risk exposures. I find that insurance firms with high exposure to climate change physical risks present a high level of adaptation response to the pecuniary impact of the risks. These risks are event-driven (acute risk) and long term shift in climate change patterns (chronic risk). Furthermore, insurance firms with a high exposure to both climate change risks have lower financial performance measures compared to insurance firms with low exposure to acute and chronic climate change risks. Relating the climate change risk features to quantitative firm characteristics in a classification and regression tree analysis, I find in addition to the depreciation of tangible fixed assets, reserves and plant, property and equipment, that reinsurance liabilities and reinsurance assets of insurance firms, largely dictate climate-related risks of insurance firms.
August 2020
School of Management
Full Citation
Rensselaer Polytechnic Institute, Troy, NY
PubMed ID