|
on Forecasting |
| By: | Valeri Voev (Aarhus University and CREATES) |
| Abstract: | We analyze the applicability of economic criteria for volatility forecast evaluation based on unconditional measures of portfolio performance. The main theoretical finding is that such unconditional measures generally fail to rank conditional forecasts correctly due to the presence of a bias term driven by the variability of the conditional mean and portfolio weights. Simulations and a small empirical study suggest that the bias can be empirically substantial and lead to distortions in forecast evaluation. An important implication is that forecasting superiority of models using high frequency data is likely to be understated if unconditional criteria are used. |
| Keywords: | Forecast evaluation, Volatility forecasting, Portfolio optimization, Mean-variance analysis |
| JEL: | C32 C53 G11 |
| Date: | 2009–11–24 |
| URL: | http://d.repec.org/n?u=RePEc:aah:create:2009-56&r=for |
| By: | Karsten R. Gerdrup (Norges Bank (Central Bank of Norway)); Anne Sofie Jore (Norges Bank (Central Bank of Norway)); Christie Smith (Reserve Bank of New Zealand); Leif Anders Thorsrud (Norges Bank (Central Bank of Norway)) |
| Abstract: | Forecast combination has become popular in central banks as a means to improve forecasts and to alleviate the risk of selecting poor models. However, if a model suite is populated with many similar models, then the weight attached to other independent models may be lower than warranted by their performance. One way to mitigate this problem is to group similar models into distinct `ensembles'. Using the original suite of models in Norges Bank's system for averaging models (SAM), we evaluate whether forecast performance can be improved by combining ensemble densities, rather than combining individual model densities directly. We evaluate performance both in terms of point forecasts and density forecasts, and test whether the densities are well-calibrated. We find encouraging results for combining ensembles. |
| Keywords: | forecasting, density combination; model combination; clustering; ensemble density; pits. |
| JEL: | C52 C53 E52 |
| Date: | 2009–11–11 |
| URL: | http://d.repec.org/n?u=RePEc:bno:worpap:2009_19&r=for |
| By: | Florian Heinen (Leibniz University of Hannover); Philipp Sibbertsen (Leibniz University of Hannover); Robinson Kruse (Aarhus University and CREATES) |
| Abstract: | We consider the problem of forecasting time series with long memory when the memory parameter is subject to a structural break. By means of a large-scale Monte Carlo study we show that ignoring such a change in persistence leads to substantially reduced forecasting precision. The strength of this effect depends on whether the memory parameter is increasing or decreasing over time. A comparison of six forecasting strategies allows us to conclude that pre-testing for a change in persistence is highly recommendable in our setting. In addition we provide an empirical example which underlines the importance of our findings. |
| Keywords: | Long memory time series, Break in persistence, Structural change, Simulation, Forecasting competition |
| JEL: | C15 C22 C53 |
| Date: | 2009–11–17 |
| URL: | http://d.repec.org/n?u=RePEc:aah:create:2009-53&r=for |
| By: | Sébastien Laurent; Jeroen Rombouts; Francesco Violente |
| Abstract: | A large number of parameterizations have been proposed to model conditional variance dynamics in a multivariate framework. However, little is known about the ranking of multivariate volatility models in terms of their forecasting ability. The ranking of multivariate volatility models is inherently problematic because it requires the use of a proxy for the unobservable volatility matrix and this substitution may severely affect the ranking. We address this issue by investigating the properties of the ranking with respect to alternative statistical loss functions used to evaluate model performances. We provide conditions on the functional form of the loss function that ensure the proxy-based ranking to be consistent for the true one - i.e., the ranking that would be obtained if the true variance matrix was observable. We identify a large set of loss functions that yield a consistent ranking. In a simulation study, we sample data from a continuous time multivariate diffusion process and compare the ordering delivered by both consistent and inconsistent loss functions. We further discuss the sensitivity of the ranking to the quality of the proxy and the degree of similarity between models. An application to three foreign exchange rates, where we compare the forecasting performance of 16 multivariate GARCH specifications, is provided. <P>Un grand nombre de méthodes de paramétrage ont été proposées dans le but de modéliser la dynamique de la variance conditionnelle dans un cadre multivarié. Toutefois, on connaît peu de choses sur le classement des modèles de volatilité multivariés, du point de vue de leur capacité à permettre de faire des prédictions. Le classement des modèles de volatilité multivariés est forcément problématique du fait qu’il requiert l’utilisation d’une valeur substitutive pour la matrice de la volatilité non observable et cette substitution peut influencer sérieusement le classement. Nous abordons ce problème en examinant les propriétés du classement en relation avec les fonctions de perte statistiques alternatives utilisées pour évaluer la performance des modèles. Nous présentons des conditions liées à la forme fonctionnelle de la fonction de perte qui garantissent que le classement fondé sur une valeur de substitution est constant par rapport au classement réel, c’est-à-dire à celui qui serait obtenu si la matrice de variance réelle était observable. Nous établissons un vaste ensemble de fonctions de perte qui produisent un classement constant. Dans le cadre d’une étude par simulation, nous fournissons un échantillon de données à partir d’un processus de diffusion multivarié en temps continu et comparons l’ordre généré par les fonctions de perte constantes et inconstantes. Nous approfondissons la question de la sensibilité du classement à la qualité de la substitution et le degré de ressemblance entre les modèles. Une application à trois taux de change est proposée et, dans ce contexte, nous comparons l’efficacité de prédiction de 16 paramètres du modèle GARCH multivarié (approche d’hétéroscédasticité conditionnelle autorégressive généralisée). |
| Keywords: | Volatility, multivariate GARCH, matrix norm, loss function, model confidence set, Volatilité, modèle GARCH multivarié, norme matricielle, fonction de perte, ensemble de modèles de confiance. |
| Date: | 2009–11–01 |
| URL: | http://d.repec.org/n?u=RePEc:cir:cirwor:2009s-45&r=for |
| By: | José Julián Sidaoui; Carlos Capistrán; Daniel Chiquiar; Manuel Ramos Francia |
| Abstract: | This note studies the causal relationship that may exist between the producer price index (PPI) and the consumer price index (CPI). In contrast with previous international studies, the results suggest that, in the case of Mexico, information on the PPI seems to be useful to improve forecasts of CPI inflation. In particular, CPI inflation responds significantly to disequilibrium errors with respect to the long-run relationship between consumer and producer prices. These results are based on in-sample and out-of-sample tests of Granger causality, in the context of an error correction model. |
| Keywords: | Cointegration, forecast evaluation, Granger causality, vector error correction. |
| JEL: | C32 C53 E31 E37 |
| Date: | 2009–11 |
| URL: | http://d.repec.org/n?u=RePEc:bdm:wpaper:2009-14&r=for |