Selecting from amongst non–nested conditional variance models: information criteria and portfolio determination

We consider the finite sample properties of model selection by information criteria in conditionally heteroscedastic models. Recent theoretical results show that certain popular criteria are consistent in that they will select the true model asymptotically with probability 1. To examine the empirical relevance of this property, Monte Carlo simulations are conducted for a set of non–nested data generating processes (DGPs) with the set of candidate models consisting of all types of model used as DGPs. In addition, not only is the best model considered but also those with similar values of the information criterion, called close competitors, thus forming a portfolio of eligible models. To supplement the simulations, the criteria are applied to a set of economic and financial series. In the simulations, the criteria are largely ineffective at identifying the correct model, either as best or a close competitor, the parsimonious GARCH(1, 1) model being preferred for most DGPs. In contrast, asymmetric models are generally selected to represent actual data. This leads to the conjecture that the properties of parameterizations of processes commonly used to model heteroscedastic data are more similar than may be imagined and that more attention needs to be paid to the behaviour of the standardized disturbances of such models, both in simulation exercises and in empirical modelling.

Modelling heterotachy in phylogenetic inference by reversible-jump Markov chain Monte Carlo

The rate at which a given site in a gene sequence alignment evolves over time may vary. This phenomenon-known as heterotachy-can bias or distort phylogenetic trees inferred from models of sequence evolution that assume rates of evolution are constant. Here, we describe a phylogenetic mixture model designed to accommodate heterotachy. The method sums the likelihood of the data at each site over more than one set of branch lengths on the same tree topology. A branch-length set that is best for one site may differ from the branch-length set that is best for some other site, thereby allowing different sites to have different rates of change throughout the tree. Because rate variation may not be present in all branches, we use a reversible-jump Markov chain Monte Carlo algorithm to identify those branches in which reliable amounts of heterotachy occur. We implement the method in combination with our 'pattern-heterogeneity' mixture model, applying it to simulated data and five published datasets. We find that complex evolutionary signals of heterotachy are routinely present over and above variation in the rate or pattern of evolution across sites, that the reversible-jump method requires far fewer parameters than conventional mixture models to describe it, and serves to identify the regions of the tree in which heterotachy is most pronounced. The reversible-jump procedure also removes the need for a posteriori tests of 'significance' such as the Akaike or Bayesian information criterion tests, or Bayes factors. Heterotachy has important consequences for the correct reconstruction of phylogenies as well as for tests of hypotheses that rely on accurate branch-length information. These include molecular clocks, analyses of tempo and mode of evolution, comparative studies and ancestral state reconstruction. The model is available from the authors' website, and can be used for the analysis of both nucleotide and morphological data.

Bayesian analysis of an inverse Gaussian correlated frailty model

In survival analysis frailty is often used to model heterogeneity between individuals or correlation within clusters. Typically frailty is taken to be a continuous random effect, yielding a continuous mixture distribution for survival times. A Bayesian analysis of a correlated frailty model is discussed in the context of inverse Gaussian frailty. An MCMC approach is adopted and the deviance information criterion is used to compare models. As an illustration of the approach a bivariate data set of corneal graft survival times is analysed. (C) 2006 Elsevier B.V. All rights reserved.

A bagging-based correction for the mixture model estimator of population size

Estimation of a population size by means of capture-recapture techniques is an important problem occurring in many areas of life and social sciences. We consider the frequencies of frequencies situation, where a count variable is used to summarize how often a unit has been identified in the target population of interest. The distribution of this count variable is zero-truncated since zero identifications do not occur in the sample. As an application we consider the surveillance of scrapie in Great Britain. In this case study holdings with scrapie that are not identified (zero counts) do not enter the surveillance database. The count variable of interest is the number of scrapie cases per holding. For count distributions a common model is the Poisson distribution and, to adjust for potential heterogeneity, a discrete mixture of Poisson distributions is used. Mixtures of Poissons usually provide an excellent fit as will be demonstrated in the application of interest. However, as it has been recently demonstrated, mixtures also suffer under the so-called boundary problem, resulting in overestimation of population size. It is suggested here to select the mixture model on the basis of the Bayesian Information Criterion. This strategy is further refined by employing a bagging procedure leading to a series of estimates of population size. Using the median of this series, highly influential size estimates are avoided. In limited simulation studies it is shown that the procedure leads to estimates with remarkable small bias.

Bootstrap prediction intervals for autoregressive time series

The calculation of interval forecasts for highly persistent autoregressive (AR) time series based on the bootstrap is considered. Three methods are considered for countering the small-sample bias of least-squares estimation for processes which have roots close to the unit circle: a bootstrap bias-corrected OLS estimator; the use of the Roy–Fuller estimator in place of OLS; and the use of the Andrews–Chen estimator in place of OLS. All three methods of bias correction yield superior results to the bootstrap in the absence of bias correction. Of the three correction methods, the bootstrap prediction intervals based on the Roy–Fuller estimator are generally superior to the other two. The small-sample performance of bootstrap prediction intervals based on the Roy–Fuller estimator are investigated when the order of the AR model is unknown, and has to be determined using an information criterion.

Meta-analysis of relationships between enteric methane yield and milk fatty acid profile in dairy cattle

Various studies have indicated a relationship between enteric methane (CH4) production and milk fatty acid (FA) profiles of dairy cattle. However, the number of studies investigating such a relationship is limited and the direct relationships reported are mainly obtained by variation in CH4 production and milk FA concentration induced by dietary lipid supplements. The aim of this study was to perform a meta-analysis to quantify relationships between CH4 yield (per unit of feed and unit of milk) and milk FA profile in dairy cattle and to develop equations to predict CH4 yield based on milk FA profile of cows fed a wide variety of diets. Data from 8 experiments encompassing 30 different dietary treatments and 146 observations were included. Yield of CH4 measured in these experiments was 21.5 ± 2.46 g/kg of dry matter intake (DMI) and 13.9 ± 2.30 g/ kg of fat- and protein-corrected milk (FPCM). Correlation coefficients were chosen as effect size of the relationship between CH4 yield and individual milk FA concentration (g/100 g of FA). Average true correlation coefficients were estimated by a random-effects model. Milk FA concentrations of C6:0, C8:0, C10:0, C16:0, and C16:0-iso were significantly or tended to be positively related to CH4 yield per unit of feed. Concentrations of trans-6+7+8+9 C18:1, trans-10+11 C18:1, cis- 11 C18:1, cis-12 C18:1, cis-13 C18:1, trans-16+cis-14 C18:1, and cis-9,12 C18:2 in milk fat were significantly or tended to be negatively related to CH4 yield per unit of feed. Milk FA concentrations of C10:0, C12:0, C14:0-iso, C14:0, cis-9 C14:1, C15:0, and C16:0 were significantly or tended to be positively related to CH4 yield per unit of milk. Concentrations of C4:0, C18:0, trans-10+11 C18:1, cis-9 C18:1, cis-11 C18:1, and cis- 9,12 C18:2 in milk fat were significantly or tended to be negatively related to CH4 yield per unit of milk. Mixed model multiple regression and a stepwise selection procedure of milk FA based on the Bayesian information criterion to predict CH4 yield with milk FA as input (g/100 g of FA) resulted in the following prediction equations: CH4 (g/kg of DMI) = 23.39 + 9.74 × C16:0- iso – 1.06 × trans-10+11 C18:1 – 1.75 × cis-9,12 C18:2 (R2 = 0.54), and CH4 (g/kg of FPCM) = 21.13 – 1.38 × C4:0 + 8.53 × C16:0-iso – 0.22 × cis-9 C18:1 – 0.59 × trans-10+11 C18:1 (R2 = 0.47). This indicated that milk FA profile has a moderate potential for predicting CH4 yield per unit of feed and a slightly lower potential for predicting CH4 yield per unit of milk. Key words: methane , milk fatty acid profile , metaanalysis , dairy cattle

Joint data assimilation of satellite reflectance and net ecosystem exchange data constrains ecosystem carbon fluxes at a high-elevation subalpine forest

We utilized an ecosystem process model (SIPNET, simplified photosynthesis and evapotranspiration model) to estimate carbon fluxes of gross primary productivity and total ecosystem respiration of a high-elevation coniferous forest. The data assimilation routine incorporated aggregated twice-daily measurements of the net ecosystem exchange of CO2 (NEE) and satellite-based reflectance measurements of the fraction of absorbed photosynthetically active radiation (fAPAR) on an eight-day timescale. From these data we conducted a data assimilation experiment with fifteen different combinations of available data using twice-daily NEE, aggregated annual NEE, eight-day f AP AR, and average annual fAPAR. Model parameters were conditioned on three years of NEE and fAPAR data and results were evaluated to determine the information content from the different combinations of data streams. Across the data assimilation experiments conducted, model selection metrics such as the Bayesian Information Criterion and Deviance Information Criterion obtained minimum values when assimilating average annual fAPAR and twice-daily NEE data. Application of wavelet coherence analyses showed higher correlations between measured and modeled fAPAR on longer timescales ranging from 9 to 12 months. There were strong correlations between measured and modeled NEE (R2, coefficient of determination, 0.86), but correlations between measured and modeled eight-day fAPAR were quite poor (R2 = −0.94). We conclude that this inability to determine fAPAR on eight-day timescale would improve with the considerations of the radiative transfer through the plant canopy. Modeled fluxes when assimilating average annual fAPAR and annual NEE were comparable to corresponding results when assimilating twice-daily NEE, albeit at a greater uncertainty. Our results support the conclusion that for this coniferous forest twice-daily NEE data are a critical measurement stream for the data assimilation. The results from this modeling exercise indicate that for this coniferous forest, average annuals for satellite-based fAPAR measurements paired with annual NEE estimates may provide spatial detail to components of ecosystem carbon fluxes in proximity of eddy covariance towers. Inclusion of other independent data streams in the assimilation will also reduce uncertainty on modeled values.

Three-level main-effects designs exploiting prior information about model uncertainty

To explore the projection efficiency of a design, Tsai, et al [2000. Projective three-level main effects designs robust to model uncertainty. Biometrika 87, 467-475] introduced the Q criterion to compare three-level main-effects designs for quantitative factors that allow the consideration of interactions in addition to main effects. In this paper, we extend their method and focus on the case in which experimenters have some prior knowledge, in advance of running the experiment, about the probabilities of effects being non-negligible. A criterion which incorporates experimenters' prior beliefs about the importance of each effect is introduced to compare orthogonal, or nearly orthogonal, main effects designs with robustness to interactions as a secondary consideration. We show that this criterion, exploiting prior information about model uncertainty, can lead to more appropriate designs reflecting experimenters' prior beliefs. (c) 2006 Elsevier B.V. All rights reserved.

A radial basis function network classifier to maximise leave-one-out mutual information

tWe develop an orthogonal forward selection (OFS) approach to construct radial basis function (RBF)network classifiers for two-class problems. Our approach integrates several concepts in probabilisticmodelling, including cross validation, mutual information and Bayesian hyperparameter fitting. At eachstage of the OFS procedure, one model term is selected by maximising the leave-one-out mutual infor-mation (LOOMI) between the classifier’s predicted class labels and the true class labels. We derive theformula of LOOMI within the OFS framework so that the LOOMI can be evaluated efficiently for modelterm selection. Furthermore, a Bayesian procedure of hyperparameter fitting is also integrated into theeach stage of the OFS to infer the l2-norm based local regularisation parameter from the data. Since eachforward stage is effectively fitting of a one-variable model, this task is very fast. The classifier construc-tion procedure is automatically terminated without the need of using additional stopping criterion toyield very sparse RBF classifiers with excellent classification generalisation performance, which is par-ticular useful for the noisy data sets with highly overlapping class distribution. A number of benchmarkexamples are employed to demonstrate the effectiveness of our proposed approach.