Research

Abstract: Fintech firms are often viewed as competing with banks. Instead, more recently, there has been growth in partnership and collaboration between fintech firms and banks. These partnerships have allowed banks to access more information on consumers through data aggregation, artificial intelligence/machine learning (AI/ML), and other tools. We explore the demographics of consumers targeted by banks that have entered into such partnerships. Specifically, we test whether banks are more likely to extend credit offers (by mail) and/or credit originations to consumers who would have otherwise been deemed high risk either because of low credit scores or lack of credit scores altogether. Our analysis uses data on credit offers based on a survey conducted by Mintel, as well as data on credit originations based on the Federal Reserve’s Y-14M reports. Additionally, we analyze a unique data set of partnerships between fintech firms and banks compiled by CB Insights to identify the relevant partnerships. Our results indicate that banks are more likely to offer credit cards and personal loans to the credit invisible and below-prime consumers — and are also more likely to grant larger credit limits to those consumers — after the partnership period. Similarly, we find that fintech partnerships result in banks being more likely to originate mortgage loans to nonprime homebuyers and that they increase the mortgage loan amounts that banks grant to nonprime buyers as well. Overall, we find that these partnerships could help to move us toward a more inclusive financial system.

Abstract: There exists an increasing number of methods of estimating stochastic volatility from price returns within the financial econometrics literature. Combining results from different models has proved to be beneficial in forecasting across fields in economics. In this paper we synthesize the work done in forecast combination with estimation methods for integrated volatility in high frequency financial data, as well as using machine learning methods to to estimate optimal volatility. We test for the utility and accuracy of our combined estimates by applying the Volatility Feedback Effect, which highlights the negative relation between volatility and returns, as well as in Monte-Carlo simulated data. The efficacy of combining estimators is compared by using economic criteria in the form of trading strategy profits from a strategy utilizing the negative relation outlined from the Volatility Feedback Effect. We find that combining methods of integrated volatility yields positive results for the stock returns analyzed. In particular, we observe a few things of interest. Volatility calculated at the 5-minute level is not always optimal for computing volatility, at least for the trading strategy used in this paper. The more useful volatility estimator, as well as frequency at which its estimated, varies amongst the stocks analyzed; there is no one-size-fits-all volatility estimation method. Combination volatility estimation is not always superior to the individual estimation methods, it does tend to produce a more useful estimate of volatility than the vast majority of other estimators, making it a consistently solid volatility estimation choice. Finally, there exists additional utility in estimating volatility via OLS and LASSO regression methods, implying machine learning algorithms may yet find a place in the volatility literature.

Abstract: Systemic risk is present across markets, and may be measured by analyzing differing stock indices to see where jumps occur across multiple sets of data. Using the presence of these co-jumps versus total jumps, we construct a number of jump ratio factors. We further reduce use data reduction methods to identify relevant predictive set of factors for macroeconomic events.