Financial Decision-Making Using Data

The growth in data logging and availability allow for decisions which are increasingly informed through empirical study. The velocity and veracity of data, and the relative efficiency of financial markets enables decisions to be made with unprecedented speed and accuracy. In this thesis we outline three ways in which new data can be used to make decisions across three spheres of finance, namely in financial markets, consumer finance and policy decision making. These three areas show how the depth and breadth that new data and empirical methods can contribute to functional and timely decision making. We also explore information theory and financial markets. We discuss information transmission and processing, asymmetric information, individual behavior, and adverse selection. In Chapter 2 we use high frequency tick data to ascertain the optimal market structure for information discovery and transmission in world gold markets. We investigate which of the two main centers of gold trading—the London spot market and the New York futures market—plays a more important role in setting the price of gold. Using intraday data during a 17-year period we find that although both markets contribute to price discovery, the New York futures play a larger role on average. This is striking given the volume of gold traded in New York is less than a tenth of the London spot volume and illustrates the importance of market structure on the process of price discovery. We find considerable variation in price discovery shares both intraday and across years. The variation is related to the structure and liquidity of the markets, daylight hours, and macroeconomic announcements that affect the price of gold. We find that a major upgrade in the New York trading platform reduces the relative amount of noise in New York futures prices, reduces the impact of daylight hours on the location of price discovery, but does not greatly increase the speed with which information is reflected in prices. Chapter 3 uses the information contained in the bank transactions of borrowers to infer their time varying credit risk. We analyze the informativeness of non-mortgage bank transactions on mortgage default for a major retail bank. We find that short-term interest coverage, income changes, home maintenance expenses, and cash withdrawals are strong predictors of future default. Using transaction data improves model prediction by 8.74% from a standard baseline model, allowing for earlier intervention that can assist consumers and lower bank losses. We also develop new model fit measures which allow for the comparison of competing statistical models. These new measures compare already adequate models to clearly distinguish which model provides a better fit considering the magnitude of the improvement (IMPMAG) and the reduction in improvable area (IMPRED). The fourth chapter uses data from stock and options markets to ascertain the effect a policy announcement can have on the value of firms, while measuring and controlling for the unresolved uncertainty present when the announcement does not guarantee implementation. We examine the wealth effects of the solar and washing machine (SOLAR), and steel and aluminum (STEEL) tariffs enacted by President Trump in January and March 2018. Using a new approach formalized by Barraclough et al. (2013), Ball and Brown (1968), and Han et al. (2019), we estimate the effect of these announcement controlling for unresolved uncertainty. We find a negative value effect of approximately 546 and 358 billion dollars in the case of the solar and steel announcement respectively, with a statistically insignificant news effect in line with the market anticipating the announcements. This study outlines implications on the protectionist tariffs. This dissertation illustrates how modern data and methods can be used to inform decisions regarding the structure of markets, the credit risk of a borrower and the effect of a potential policy on firm value.