"Experiments for Improving Statistical Classification of Feedforward Neural Networks"

ABSTRACT: Properly constructed neural networks are known to function as statistical classifiers. Neural networks are flexible-form approaches that use the available data to determine the most appropriate structure for the application. In this project, methods of incorporating data for statistical classification using feed-forward neural networks are discussed.

"The Relationship between Firm Innovativeness and CEO Compensation"

ABSTRACT: This study empirically examines the impact of firm innovation, as operationalized by the management initiative programs of activity-based costing and the balanced scorecard, and their relationship to CEO executive compensation. CEO total executive compensation is decomposed into its structural components or levels. This study provides further explanation as to which levels are most significantly associated with firm innovation. This study determines if there is a relationship between firm innovation measures and stock and option holdings and also adds to the literature by examining the change in the portfolio value of the stock and option holdings. Agency and contracting theory predicts that there should be a positive and significant relationship between these measures and each of the structural levels of CEO compensation.

"Memory in Eurobond Returns: Evidence From Hurst Exponents"

Abstract: This paper reports on the testing of electronic Eurobond markets for long memory effects. We calculate the Hurst exponent of the returns on the best-bid price for each bond traded on MTS for the first quarter of 2004. MTS is an electronic trading platform for Eurobonds. 219 unique ISIN codes are identified as traded on MTS during the first quarter of 2004. A mean Hurst exponent of .24 is found, with a weighted mean of .13. Results indicate that inter-dealer electronic trading of Eurobonds is non-stationary, with the pricing processes of the best bids time- dependent. The return series of almost all the bonds investigated show anti-persistent behavior. The degree of anti-persistence is pronounced, indicating that the European electronic bond markets are zealously mean reverting. To our knowledge, this is the first example in the literature of the long-term memory properties of hundreds of bonds using tick-by-tick data. Previous studies have looked at much smaller samples of bonds and have typically used daily data

“A Measurement of the Domestic Equity Premia, An Abnormal Earnings Model"

Abstract: This research replicates a research design originally put forth by Claus, James, and Jacob Thomas, (2001), "Equity Premia as Low as Three Percent? Evidence from Analysts' Earnings Forecasts for Domestic and International Stock Markets", Journal of Finance 56, (No. 5, October) pp. 1629-1666. The equity premium is the commonly used expression for the difference between the expected return on the market portfolio of common stocks and the risk-free interest rate. The average return on a broad portfolio of stocks is typically used to estimate the expected market return. The average real return for 1872 to 2000 on the S&P index is between eight and nine percent per year for U.S. equities. This large spread between the average stock return and the interest rate is the source of the so-called equity- premium puzzle: stock returns seem too high given the observed volatility of consumption. Put another way, estimates of the equity premium derived from asset pricing models are quite different from observed average excess rates of return. Rather than examining historical returns, this paper uses an abnormal-earnings model to estimate the equity premium from a calculated discount rate. This discount rate is the rate that equates market valuations with values implied by estimates of future cash flows. Results show that the US equity premium is 4.40% for the period 1985–2002.

“Probabilistic Analysis of Trading Techniques in Continous time"

Abstract: Mathematical models of "trading the line" and "stop loss order" Trading techniques were considered. We were interested in optimization of these trading techniques based on some utility function given by an investor. Black-Scholes model were studied. We have found probabilistic characteristics of the underlying stochastic processes, i.e. distributions, expectations, and variances. One can use these results to optimize trading techniques based on investor's utility function. These theoretical results were applied to the historical price processes in continuous time setting. The historical price processes of the most popular indexes Dow, Nasdaq, and S&P 500 and IBM stock were studied. Statistical analysis of these processes will be discussed too. Including volatility and drift estimation, statistical R/S analysis.

“Probabilistic Analysis of Trading Techniques in Discrete time"

Abstract: Mathematical models of "trading the line" and "stop loss order" trading techniques were considered. We were interested in optimization of these trading techniques based on some utility function given by an investor. We were assuming that this utility function depends on the expectations and variances of the gain and stopping time. Cox-Ross-Rubinstein market model were studied. We have found probabilistic characteristics of the underlying stochastic processes, i.e. distributions, expectations, and variances. One can use these results to optimize trading techniques based on investor's utility function. These theoretical results were applied to the real life price processes in continuous time setting.