
on Forecasting 
By:  Andrew J. Patton; Allan Timmermann 
Abstract:  Evaluation of forecast optimality in economics and finance has almost exclusively been conducted on the assumption of mean squared error loss under which forecasts should be unbiased and forecast errors serially uncorrelated at the single period horizon with increasing variance as the forecast horizon grows. This paper considers properties of optimal forecasts under general loss functions and establishes new testable implications of forecast optimality. These hold when the forecaster's loss function is unknown but testable restrictions can be imposed on the data generating process, trading off conditions on the data generating process against conditions on the loss function. Finally, we propose flexible parametric estimation of the forecaster's loss function, and obtain a test of forecast optimality via a test of overidentifying restrictions. 
Keywords:  forecast evaluation, loss function, rationality tests 
JEL:  C53 C22 C52 
Date:  2005–01 
URL:  http://d.repec.org/n?u=RePEc:cep:stiecm:/2005/485&r=for 
By:  Pedro N. Rodríguez,; Simón SosvillaRivero 
Abstract:  Previous empirical studies have shown that predictive regressions in which model uncertainty is assessed and propagated generate desirable properties when predicting outofsample. However, it is still not clear (a) what the important conditioning variables for predicting stock returns outofsample are, and (b) how composite weighted ensembles outperform model selection criteria. By comparing the unconditional accuracy of prediction regressions to the conditional accuracy conditioned on specific explanatory variables masked), we find that crosssectional premium and term spread are robust predictors of future stock returns. Additionally, using the biasvariance decomposition for the 0/1 loss function, the analysis shows that lower bias, and not lower variance, is the fundamental difference between composite weighted ensembles and model selection criteria. This difference, nevertheless, does not necessarily imply that model averaging techniques improve our ability to describe monthly upanddown movements' behavior in stock markets. 
URL:  http://d.repec.org/n?u=RePEc:fda:fdaddt:200603&r=for 
By:  Peter M Robinson 
Abstract:  Much time series data are recorded on economic and financial variables. Statistical modelling of such data is now very well developed, and has applications in forecasting. We review a variety of statistical models from the viewpoint of 'memory', or strength of dependence across time, which is a helpful discriminator between different phenomena of interest. Both linear and nonlinear models are discussed. 
Keywords:  Long memory, short memory, stochastic volatility 
JEL:  C22 
Date:  2005–03 
URL:  http://d.repec.org/n?u=RePEc:cep:stiecm:/2005/487&r=for 
By:  Andrew Leigh; Justin Wolfers 
Abstract:  We review the efficacy of three approaches to forecasting elections: econometric models that project outcomes on the basis of the state of the economy; public opinion polls; and election betting (prediction markets). We assess the efficacy of each in light of the 2004 Australian election. This election is particularly interesting both because of innovations in each forecasting technology, and also because the increased majority achieved by the Coalition surprised most pundits. While the evidence for economic voting has historically been weak for Australia, the 2004 election suggests an increasingly important role for these models. The performance of polls was quite uneven, and predictions both across pollsters, and through time, vary too much to be particularly useful. Betting markets provide an interesting contrast, and a slew of data from various betting agencies suggests a more reasonable degree of volatility, and useful forecasting performance both throughout the election cycle and across individual electorates. 
JEL:  D72 D84 
Date:  2006–02 
URL:  http://d.repec.org/n?u=RePEc:nbr:nberwo:12053&r=for 
By:  John H. Cochrane 
Abstract:  To question the statistical significance of return predictability, we cannot specify a null that simply turns off that predictability, leaving dividend growth predictability at its essentially zero sample value. If neither returns nor dividend growth are predictable, then the dividendprice ratio is a constant. If the null turns off return predictability, it must turn on the predictability of dividend growth, and then confront the evidence against such predictability in the data. I find that the absence of dividend growth predictability gives much stronger statistical evidence against the null, with roughly 12% probability values, than does the presence of return predictability, which only gives about 20% probability values. I argue that tests based on longrun return and dividend growth regressions provide the cleanest and most interpretable evidence on return predictability, again delivering about 12% probability values against the hypothesis that returns are unpredictable. I show that Goyal and Welch's (2005) finding of poor outofsample R2 does not reject return forecastability. Outofsample R2 is poor even if all dividend yield variation comes from timevarying expected returns. 
JEL:  G0 G1 
Date:  2006–02 
URL:  http://d.repec.org/n?u=RePEc:nbr:nberwo:12026&r=for 