No abstract

Depression has been linked with seasonal affective disorder (SAD), a condition that affects many people during the seasons of relatively fewer hours of daylight. Experimental research in psychology has documented a clear link between depression and lowered risktaking behavior in a wide range of settings, including those of a nancial nature. Through the links between SAD and depression and between depression and risk aversion, seasonal variation in length of day can translate into seasonal variation in equity returns. Based on supportive evidence from psychology which suggests SAD is linked closely with hours of daylight, we consider stock market index data from countries at various latitudes and on both sides of the equator. We model differences in the seasonal variation of daylight across countries to capture the in uence of daylight on human sentiment, risk tolerance, and hence stock returns. Our results strongly support a SAD effect in the seasonal cycle of stock returns that is both signi cant and substantial, even after controlling for well-known market seasonals and other environmental factors. Patterns at different latitudes and in both hemispheres provide compelling evidence of a link between seasonal depression and seasonal variation in stock returns: higher-latitude markets show more pronounced SAD effects and results in the Southern Hemisphere are six months out of phase, as are the seasons.The remainder of the paper is organized as follows. In Section I, we discuss SAD, depression, and equilibrium market returns. In Section II, we introduce the international data sets. In Section III, we explain the construction of the variables intended to capture the in uence of SAD on the stock market. We document in Section IV the signi cance of the SAD effect, both statistical and economic, and provide an example of the excess returns that arise from trading strategies based on the SAD effect. In Section V, we explore the robustness of the SAD effect to changes in variable de nitions and estimation methods. Section VI considers SAD in the context of segmented versus integrated capital markets. Section VII concludes.

(2002) revisits the issue of daylight-saving-time changes impacting nan

No abstract

Companies sometimes use statistical analysis to anticipate their bond ratings or a change in the rating. However, different statistical models can yield different ratings forecasts, and there is no clear rule for which model is preferable. We use several forecasting methods to predict bond ratings in the transportation and industrial sectors listed by Moody's bond rating service. A variant of the ordered-logit regression-combining method of Kamstra and Kennedy 1998 yields statistically significant, quantitatively meaningful improvements over its competitors, with very little computational cost.

No abstract

We develop a II(!lV procedllre to forecast flltllre casbjlolVsfrom ajillallcial asset alld tbell lise tbe presellt vallie of ollr casb jlOlV forecasts to calCIIlate tbe asset's f,mdamelltal price. As all example, we COllstrllct a 1I0lllillear AJljfA.-ARCH-Artificial Nellral Network model to obtaill Ollt-Ofsample dividelldforecastsf01'1920 alld beyolll/, tlsillg OIlly ill-sample dividelld data. The presellt vallie of ollr forecasted dividellds yield fimdamelltal prices tbat J'eprodtlce tbe magllittlde, timillg, mid time-series bebavior of tbe boom mid crasb ill 1929 stock p,ices. We tberefore reject tbe poplliar claim tbat tbe 1920s stock market cOlltailled a bubble. Many empirical tests of asset price behavior call for the comparison of an asset's market price to its This anicle is a ~uhstjJntial revision of ollr earlier paper. "l"sing Dividenu Forecasting i\todels 10 Rejcl1 Buhbles in Asset Prices: The Cast: of 1929"s Stock CrJsh," whith rcccin:d the i'cw York Stock Exchange prize for best paper on eqUity Ir.luing presented at the 199-1 \X"estern finance Association nlt:ctings. In rc\"ising this article we reech"eo numerous useful suggestions from a great nuny of our colleagues. Extended discussion~ and correspondence with FrJnklin Allen (the editor). John Camphell. !lunon lIollifieid. Allan Kleidon. Lisa Kr.lmer. Alan KrJ.us, Rick ;\Iishkin. Gregor Smith, and an anonymous referee of this journal were pankularly helpful. \X'c ~llso gmtefully acknowledge financial suppon from the Social ~ciences and Ilumanitics Research Council of Canada. "fe arc responsihle for ~my remaining errors, omi'i.

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. KEY WORDS forecast combing; artificial neural network; encompassing test When combining n individual forecasts II' ... , In' the single combined forecast F is traditionally obtained by selecting 13 weights in the linear model F = 130 + Lt. I 13Ji' a popular example being the simple average across forecasts (Le. 130 = 0, f3 i = 1/ n"'l i). I However, a linear combination may not be optimal if the individual forecasts come from nonlinear models or if the true underlying conditional expectation is a nonlinear function of the information sets on which the individual forecasts are based. Consider, for example, the case of a dependent variable y = exp(Itsl Xi) + E, where E is an innovation and XI' "" Xn are n explanatory variables known to us. If each of the i= 1, ... , n individual forecasts are produced by Ii: aiexp(Xi), then any linear combination of the n individual forecasts will be inferior to the nonlinearly combined forecast F = 11:. I /;/ a i • In this paper we investigate the incremental value of going from traditional linear forecast combining procedures to a particular class of nonlinear combining procedures based on Artificial Neural Networks (ANNs). Since ANNs have the ability to approximate arbitrarily well a large class of functions, they provide considerable flexibility to uncover hidden nonlinear relationships between a group of individual forecasts and realizations of the I The forecast combining literature is much too vast to adequately cite here. For excellent reviews of the forecasting literature and discussions of traditional weight-selection techniques, however, see Clemen (1989), Granger (1989) and Min and Zellner (1993).