Corporate Financial Distress and Bankruptcy Prediction in North American Construction Industry
By Gang Li
This paper seeks to explore the application of Altman’s bankruptcy prediction model in the construction industry by measuring its percentage accuracy on a dataset consisting of 108 bankrupt & non-bankrupt firms selected across the timeline of 1985-2013. Another main goal this paper is to explore the predictive power of an expanded variable set tailored to the construction industry and compare the results. Specifically, this measuring process is done using machine learning algorithm based on scikit-learn library that transforms a raw .csv file into clean vectorized dataset. The algorithm provides various classifiers to cross-validate the training set, which produces mixed statistics that favors neither variable set but provides insight into the reliability of the non-linear classifiers.
Advisor: Connel Fullenkamp | JEL Codes: C38, C5, G33, G34 | Tagged: Bankruptcy, Corporate, Discriminant Analysis, Distress, Machine Learning
Forecasting Beta Using Conditional Heteroskedastic Models
By Andrew Bentley
Conventional measurements of equity return volatility rely on the asset’s previous day closing price to infer the current level of volatility and fail to incorporate information concerning intraday influntuctuations. Realized measures of volatility, such as the realized variance, are able to integrate intraday information by utilizing high-frequency data to form a very accurate measure of the asset’s return volatility. These measures can be used in parallel with the traditional definition of the Capital Asset Pricing Model (CAPM) beta to better predict the time-varying systematic risk of an asset. In this analysis, realized measures were added to the General Autoregressive Conditional Heteroskedastic (GARCH) framework to form a predictive model of beta that can quickly respond to rapid changes in the level of volatility. The ndings suggest that this predictive beta is better able to explain the stylized characteristics of beta and is a more accurate forecast of the realized beta than the GARCH model or the benchmark Autoregressive Moving-Average (ARMA) model used as a comparison.
Advisor: George Tauchen | JEL Codes: C0, C3, C03, C32, C53, C58 | Tagged: Beta, GARCH, GARCHX, High-Frequency Data, Realized Varience
Identifying Supply and Demand Elasticities of Iron Ore
By Zhirui Zhu
This paper utilizes instrumental variables and joint estimation to construct efficiently identified estimates of supply and demand equations for the world iron ore market under the assumption of perfect competition. With annual data spanning 1960-2010, I found an upward sloping supply curve and a downward sloping demand curve. Both of the supply and demand curves are efficiently identified using a 3SLS model. The instruments chosen are strong and credible. Point estimation of the long-run price elasticities of supply and demand are 0.45 and -0.24 respectively, indicating inelastic supply and demand market dynamics. Back-tests and forecasts were done with Monte Carlo simulations. The results indicate that 1) the predicted prices are consistent with the historical prices, 2) world GDP growth rate is the determining factor in the forecasting of iron ore prices.
Advisor: Gale Boyd | JEL Codes: C30, Q31 | Tagged: Demand, Iron Ore, Supply, Simulation, Simultaneous Equation
Volatility and Correlation Modeling for Sector Allocation in International Equity Markets
By Melanie Fan and Kate Yuan
Reliable estimates of volatility and correlation are crucial in asset allocation and risk management. This paper investigates Static, RiskMetrics, and Dynamic Conditional Correlation (DCC) models for estimating volatility and correlation by testing them in an asset allocation context. Optimal allocation weights for one year found using estimates from each model are carried to the subsequent year and the realized Sharpe ratio is computed to assess portfolio performance. We also study cumulative risk-adjusted returns over the entire sample period. Our ndings indicate that DCC does not consistently have an advantage over the other two models, although it is optimal in certain scenarios.
Advisor: Aino Levonmaa, Emma Rasiel | JEL Codes: C32, C51, G11, G15 | Tagged: Asset Allocation, Dynamic Correlation, Emerging Markets, Volatilita
Multi-Variable Regression Analysis For the Prediction of Equity Returns Over 10 Year Periods
by Arjun Singh Jaswal
Abstract
The use of 5 variables is examined in order to forecast ex ante the total return from holding equities over 10 year periods. The 5 variables are a moving average of Campbell and Shiller’s P/E ratio, Robert B. Barsky and J. Bradford De Long’s log price predictor, a function of James Tobin’s q, the rate of change of GDP over 30 years and the rate of change of cash flow over 10 years. The significance of these variables is explained by considering them individually, simultaneously and finally under the architecture suggested by David Hirshleifer.
Professor Edward Tower, Faculty Advisor
JEL Codes: C3, E22
