R. Glen Donaldson

Forecast combining with neural networks

This paper investigates the use of Artificial Neural Networks (ANNs) to combine time series forecasts of stock market volatility from the USA, Canada, Japan and the UK. We demonstrate that combining with nonlinear ANNs generally produces forecasts which, on the basis of out-of-sample forecast encompassing tests and mean squared error comparisons, routinely dominate forecasts from traditional linear combining procedures. Superiority of the ANN arises because of its flexibility to account for potentially complex nonlinear relationships not easily captured by traditional linear models.

An artificial neural network-GARCH model for international stock return volatility

Abstract We construct a seminonparametric nonlinear GARCH model, based on the Artificial Neural Network (ANN) literature, and evaluate its ability to forecast stock return volatility in London, New York, Tokyo and Toronto. In-sample and out-of-sample comparisons reveal that our ANN model captures volatility effects overlooked by GARCH, EGARCH and GJR models and produces out-of-sample volatility forecasts which encompass those from other models. We also document important differences between volatility in international markets, such as the substantial persistence of volatility effects in Japan relative to North American and European markets.

Price barriers in the Dow Jones Industrial Average

This study tests the popular claim that the DJIAs movements around key reference points affect “investor sentiment” and thus price behavior. It is found that the DJIAs rise and fall is indeed restrained by “support” and “resistance” levels at multiples of 100 (e.g., 2800, 2900, 3000, etc.) but that, having broken through a 100-level, the DJIA then moves by more than otherwise warranted. A Monte Carlo study and comparisons with other indices confirm the significance of these findings. This suggests that some agents may trade on the basis of the DJIA but does not necessarily suggest that the market is inefficient.

Costly liquidation, interbank trade, bank runs and panics

Abstract Standard models of panics do not allow for trade between banks and assume that a run banks liquidation costs are determined exogenously. We develop a model in which banks trade with each other and liquidation costs are determined endogenously in a strategic environment. The model reproduces a panics characteristic fall in stock prices, rise in interest rates, and wave of bank runs. It also accounts for the seasonal timing of panics, the change in interest rates across panics, and international differences in panic frequency. The roles of withdrawal suspension, lender of last resort, and deposit insurance are also investigated.

Sources of panics: Evidence from the weekly data

Abstract We use weekly data (1867–1933) to investigate the sources of banking and financial panics. We find that, while a panics exact starting date is unpredictable, one can identify environments conducive to panics; a lack of liquidity in financial markets seems to be a key source. Results also reveal the effectiveness of the Aldrich-Vreeland Act inelastizing the money supply and shortening the panic of 1914 and suggest that the panic of 1933 may have been caused by a speculative attack on the gold-backed US dollar via the banks, as opposed to a run on the banks directly.

A Strategic Analysis of Corners and Squeezes

We develop a dynamic game-theoretic model of a futures market in which prices can be manipulated by corner and squeeze. We investigate equilibrium trading strategies and the price dynamics these strategies produce. Price paths produced by our model can mimic observed prices for potentially comerable commodities and explain the volatility of certain prices even when no manipulations occur. Our model also generates occasional apparent price bubbles and accounts for the existence of normal backwardation in futures markets even when players are risk neutral.

Financing banking crises : Lessons from the panic of 1907

Abstract A model of strategic asset pricing is used to investigate the increase in interest rates and the formation of cash syndicates that can occur during privately financed banking crises. Results produced help to explain the Federal Reserve Systems creation following the panic of 1907 and to clarify the Feds role in financing contemporary banking crises.

Estimating the Equity Premium

Existing empirical research investigating the size of the equity premium has largely consisted of a series of innovations around a common theme: producing a better estimate of the equity premium by using better data or a better estimation technique. The equity premium estimate that emerges from most of this work matches one moment of the data alone: the mean difference between an estimate of the return to holding equity and a risk-free rate. We instead match multiple moments of U.S. market data, exploiting the joint distribution of the dividend yield, return volatility, and realized excess returns, and find that the equity premium lies within 50 basis points of 3.5%, a range much narrower than was achieved in previous studies. Additionally, statistical tests based on the joint distribution of these moments reveal that only those models of the conditional equity premium that embed time variation, breaks, and/or trends are supported by the data. In order to develop the joint distribution of the dividend yield, return volatility, and excess returns, we need a model of price and return fundamentals. We document that even recently developed analytically tractable models that permit autocorrelated dividend growth rates and discount rates impose restrictions that are rejected by the data. We therefore turn to a wider range of models, requiring numerical solution methods and parameter estimation by the simulated method of moments.

Forecasting Fundamental Stock Price Distributions

This paper develops a procedure for forecasting the distribution from which future stock prices/returns will be drawn fundamentally. Our methodology therefore provides a means for forecasting future stock prices and return volatilities and thus for fundamentally valuing assets such as stocks and stock options. Statistical tests reveal the desirability of our procedure over those currently in use. For example, our fundamental volatility forecasts outperform standard ARCH-based models in forecasting the volatility of stock returns, and fundamental prices from our procedure reject bubbles and excess stock market volatility while prices from traditional fundamental models fail to do so.