Essays in Macroeconomics and Finance

Abstract

This dissertation consists of three essays in macroeconomics and finance. The first and second chapters analyze the impact of the financial shocks and anti-corruption campaign on Chinese firms through the bank lending channel. The third chapter provides a new method to predict the cash flow from operations (CFO) via semi-parametric estimation and machine learning. The first chapter explores the impact of the financial crisis and sovereign debt crisis on Chinese firms through the bank lending channel and firm borrowing channel. Using new data linking Chinese firms to their bank(s) and four different measurements of exposure to the international markets (international borrowing, importance of lending to foreign listed companies, share of trade settlement, and exchange/income), I find that banks with higher exposure to the international markets cut lending more during the recent financial crisis. In addition, state-owned bank loans are more pro-cyclical compared with private bank loans. Moreover, banks with higher exposure to the international markets cut lending more when there is a negative shock in OECD GDP growth. With regard to firm borrowing channel, I find that firms with higher weighted aggregate exposure to the international markets through banks have lower net debt, cash, employment, and capital investment during the financial crisis. Firms with higher weighted aggregate exposure to the global markets have higher net debt and lower cash, employment, and capital investment when there is a negative shock in OECD GDP growth. This paper also provides a theoretical model to explain the mechanism in a partially opened economy like China. The second chapter discusses the impact of the anti-corruption campaign on Chinese firms through the bank lending channel. Using confidential data linking Chinese firms to their bank(s) and prefecture-level corruption index, I find that banks located in more corrupted prefectures offer significantly less credits before the anti-corruption investigation, and this effect changes the direction after the investigation. Moreover, banks located in more corrupted prefectures tend to use higher interest rates, longer maturity, and more collateral before the campaign, all of these effects change the direction after the campaign. This paper suggests that the banks located in more corrupted prefectures have stronger monopoly power (or higher markup, and lower efficiency). This monopoly effect could be proved by that the bank concentration ratio is higher, and the bad loans of the banks are higher in the more corrupted areas, and all of these effects disappear after the campaign. The third chapter considers the methods of prediction of Cash flow from operations (CFO). Forecasting CFO is an essential topic in financial econometrics and empirical accounting. It impacts a variety of economic decisions, including valuation methodologies employing discounted cash flows, distress prediction, risk assessment, the accuracy of credit-rating predictions, and the provision of value-relevant information to security markets. Existing literature on statistically-based cash-flow prediction has pursued cross-sectional versus time-series estimation procedures in a mutually exclusive fashion. Cumulated empirical evidence indicates that the beta value varies across firms of different sizes, and the cross-sectional regression can not capture an idiosyncratic beta. However, although a time series based predictive model has the advantage of allowing for firm-specific variability in beta, it requires a long enough time series data. In this paper, we extend the literature on statistically-based, cash-flow prediction models by introducing an estimation procedure that, in essence, combine the favorable attributes of both cross-sectional estimation via the use of "local" cross-sectional data for firms of similar size and time-series estimation via the capturing of firm-specific variability in the beta parameters for the independent variables. The local learning approach assumes no a priori knowledge on the constancy of the beta coefficient. It allows the information about coefficients to be represented by only a subset of observations. This feature is particularly relevant in the CFO model, where the beta values are only related to cross-sectional data information that is "local" to its size. We provide empirical evidence that the prediction of cash flows from operations is enhanced by jointly adopting features specific to both cross-sectional and time-series modeling simultaneously.