Clarence C. Y. Kwan

Neural network forecasting of quarterly accounting earnings

Abstract This study uses an artificial neural network model to forecast quarterly accounting earnings for a sample of 296 corporations trading on the New York stock exchange. The resulting forecast errors are shown to be significantly larger (smaller) than those generated by the parsimonious Brown-Rozeff and Griffin-Watts (Foster) linear time series models, bringing into question the potential usefulness of neural network models in forecasting quarterly accounting earnings. This study confirms the conjecture by Chatfield and Hill et al. that neural network models are context sensitive. In particular, this study shows that neural network models are not necessarily superior to linear time series models even when the data are financial, seasonal and non-linear.

Portfolio selection under institutional procedures for short selling: Normative and market-equilibrium considerations

Abstract In view of the acceptance of short selling of stocks as an investment tool in the portfolio context by a growing number of institutional investors in recent years, the present study considers both normative and market-equilibrium aspects of portfolio selection with short selling. Under the full-information covariance structure of security returns, the study accurately captures institutional procedures for short selling without sacrificing analytical tractability. While short selling enhances the portfolios risk-return trade-off from a normative perspective, the equilibrium analysis reveals that there is a continuum of market-clearing prices within two boundaries for each security. Economic implications of the equilibrium pricing relationship are also explored in the study.

Foreign-Exchange Rate Dynamics: An Empirical Study Using Maximum Entropy Spectral Analysis

This study examines whether foreign-exchange rates evolve as a random walk by directly comparing the predictive ability of autoregressive (AR) models of spot rates with that of the random walk. To reduce the influence of model specifications on test results, we neither specify the order of the AR process a priori nor assume that the order is necessarily the same over the entire sample period. For each subperiod, the AR model is estimated by maximum entropy spectral analysis, using Akaikes criterion of final prediction error for optimal order selection. In contrast to standard Box–Jenkins techniques, this analysis neither arbitrarily truncates the data in the time domain beyond the sample period nor imposes periodic extension in the frequency domain, and thus it mitigates against potential structural change in the time series. It is shown that for six currencies, relative to the U.S. dollar, past spot rates are irrelevant for predicting future spot rates, or in other words, spot rates behave as a random walk.

Optimal portfolio selection under institutional procedures for short selling

Abstract This study presents and formally justifies a simple ranking approach for optimal portfolio selection under institutional procedures for short selling. It also provides economic insights of the explicit solution of the portfolio problem. The analysis is applicable to different treatments of the short-sale proceeds and any margin deposits. In contrast to previous approaches, it does not require assumptions that overstate short-sale benefits for maintaining analytical tractability, and it can easily identify and filter out securities that are undesirable for investing or for short selling. Thus, this study can enhance the usefulness of portfolio modeling for assisting practical investment decisions.

Mean-Variance Utility Functions and the Demand for Risky Assets: An Empirical Analysis Using Flexible Functional Forms

In a recent study, Levy and Markowitz [15] demonstrate that, at least for some utility functions, expected utility can be approximated by a judiciously chosen function defined over mean and variance. In addition to resurrecting mean-variance analysis from the limbo into which it was placed by the criticisms of Borch [10] and others, the analysis by Levy and Markowitz yields a more direct approach to portfolio analysis than that provided by the current empirical literature. The current portfolio literature is concerned with notions of efficient sets and systematic risk rather than with utility functions and mean-variance. While much has been gained from a utility-free methodology, it is ultimately predicated upon a separation theorem and, hence, an environment with zero transactions costs. But security markets are not costless and the separation theorem may not hold. In that event, a utility-dependent approach to portfolio analysis could potentially lead to more powerful results especially if such an approach could be empirically implemented.

A note on market-neutral portfolio selection

Abstract Long–short equity strategies allow investors to benefit potentially from both undervalued and overvalued securities. The present study develops a normative portfolio model under the practical conditions that a market-neutral strategy entails. The offsetting long and short equity holdings are established jointly and without any constraints by the underlying market index. While accurately capturing institutional procedures for short selling, the model contains the analytical and economic properties as required for a ranking approach to filter out any undesirable securities under consideration. In view of its practical features, the analysis should be of interest to practitioners for assisting their long–short investment decisions.

A note on the transmission of public informtion across inetrnational stock markets

Abstract Previous studies have established that public information cannot explain the decrease in return volatility during the closing of a stock exchange. The present study shows that the failure of the public information hypothesis in explaining the trading/nontrading volatility differential is merely a domestic phenomenon. The manifestation of public information in an international setting is very different. Specifically, the absence of price information due to the closing of the US market affects both volatility and trading volume in the Canadian market.

Some Further Analytical Properties Of The Constant Correlation Model For Portfolio Selection

The constant correlation model is a mean-variance portfolio selection model where, for a given set of risky securities, the correlation of returns between any pair of different securities is considered to be the same. Support for the model is from previous empirical evidence that sample averages of correlations outperform various more sophisticated models in forecasting the correlation matrix, an important input component for portfolio analysis. To enable a better understanding of the constant correlation model, this study identifies some additional analytical properties of the model and relates them to familiar portfolio concepts. By comparing computational times for portfolio construction with and without simplifying the correlation matrix in a simulation study, this study also confirms the models computational advantage. This study is intended to provide further analytical support for the model as a viable, simple alternative to those portfolio selection models where input requirements and the attendant computations are more burdensome.

An Empirical Investigation of the Random Character of Annual Earnings

The conclusion that annual accounting earnings follow a random walk is fairly well established in the literature. Still, this result is not all that compelling. The early studies by Little (1962), Lintner and Glauber (1978), and Ball and Watts (1972) reached their conchusion primarily by examining the serial correlation of earnings changes.’ However, the analysis of serial correlation of changes in earnings makes the implicit and simplistic assumption that the underlying time series model is first-order autoregressive (AR[ 13). Later studies by Watts and Leftwich (1977); Lookabill(l976); and Albrecht, Lookabill, and McKeown (1977) reached the random walk conclusion by comparing the forecast accuracy of the random walk with the best fitted Box-Jenkins models. There are three problems with this approach. First, standard Box-Jenkins techniques generally truncate the data beyond the sample period in the time domain. Truncation leads to serious problems in the spectral estimation of the data sequence. This problem, called “aliasing,” is especially pronounced when the sampling period is relatively short, as in the case of annual earnings data. The effect can be compared with viewing an infinitely wide scenery through a window of finite width. The abrupt cutoff of the scene at the edge of the window introduces nonexistent high-frequency components in the spectral composition of the actual scene. As a result, relatively long time series are required. On the other hand, using long time series of annual earnings may result in poor forecasts because of structural change in the series. Second, and related to the first point, the random walk, unlike Box-Jenkins models, uses only the most recent year’s earnings data, thereby reducing the influence of extreme observations arising

Optimal portfolio selection without short sales under the full-information covariance structure: A pedagogic consideration

Abstract This paper applies Markowitzs critical line algorithm to Lintner tangency portfolios for optimal portfolio selection without short sales under the full-information covariance structure of security returns. An illustration is provided showing how the approach can be implemented using spreadsheets on microcomputers.