Robust response transformations based on optimal prediction

Nonlinear regression problems can often be reduced to linearity by transforming the response variable (e.g., using the Box-Cox family of transformations). The classic estimates of the parameter defining the transformation as well as of the regression coefficients are based on the maximum likelihood criterion, assuming homoscedastic normal errors for the transformed response. These estimates are nonrobust in the presence of outliers and can be inconsistent when the errors are nonnormal or heteroscedastic. This article proposes new robust estimates that are consistent and asymptotically normal for any unimodal and homoscedastic error distribution. For this purpose, a robust version of conditional expectation is introduced for which the prediction mean squared error is replaced with an M scale. This concept is then used to develop a nonparametric criterion to estimate the transformation parameter as well as the regression coefficients. A finite sample estimate of this criterion based on a robust version of smearing is also proposed. Monte Carlo experiments show that the new estimates compare favorably with respect to the available competitors.

It's still 2%: evidence on convergence from 116 years of the US States panel data

A new debate over the speed of convergence in per capita income across economies is going on. Cross sectional estimates support the idea of slow convergence of about two percent per year. Panel data estimates support the idea of fast convergence of five, ten or even twenty percent per year. This paper shows that, if you ``do it right'', even the panel data estimation method yields the result of slow convergence of about two percent per year.

Combining multivariate density forecasts using predictive criteria

This paper combines multivariate density forecasts of output growth, inflationand interest rates from a suite of models. An out-of-sample weighting scheme based onthe predictive likelihood as proposed by Eklund and Karlsson (2005) and Andersson andKarlsson (2007) is used to combine the models. Three classes of models are considered: aBayesian vector autoregression (BVAR), a factor-augmented vector autoregression (FAVAR)and a medium-scale dynamic stochastic general equilibrium (DSGE) model. Using Australiandata, we find that, at short forecast horizons, the Bayesian VAR model is assignedthe most weight, while at intermediate and longer horizons the factor model is preferred.The DSGE model is assigned little weight at all horizons, a result that can be attributedto the DSGE model producing density forecasts that are very wide when compared withthe actual distribution of observations. While a density forecast evaluation exercise revealslittle formal evidence that the optimally combined densities are superior to those from thebest-performing individual model, or a simple equal-weighting scheme, this may be a resultof the short sample available.

Recent misconceptions about the "database search problem": a probabilistic analysis using Bayesian networks

This paper analyses and discusses arguments that emerge from a recent discussion about the proper assessment of the evidential value of correspondences observed between the characteristics of a crime stain and those of a sample from a suspect when (i) this latter individual is found as a result of a database search and (ii) remaining database members are excluded as potential sources (because of different analytical characteristics). Using a graphical probability approach (i.e., Bayesian networks), the paper here intends to clarify that there is no need to (i) introduce a correction factor equal to the size of the searched database (i.e., to reduce a likelihood ratio), nor to (ii) adopt a propositional level not directly related to the suspect matching the crime stain (i.e., a proposition of the kind 'some person in (outside) the database is the source of the crime stain' rather than 'the suspect (some other person) is the source of the crime stain'). The present research thus confirms existing literature on the topic that has repeatedly demonstrated that the latter two requirements (i) and (ii) should not be a cause of concern.

Does money matter in shaping domestic business cycles? An international investigation (with appendices)

We study the contribution of money to business cycle fluctuations in the US,the UK, Japan, and the Euro area using a small scale structural monetary business cycle model. Constrained likelihood-based estimates of the parameters areprovided and time instabilities analyzed. Real balances are statistically importantfor output and inflation fluctuations. Their contribution changes over time. Models giving money no role provide a distorted representation of the sources of cyclicalfluctuations, of the transmission of shocks and of the events of the last 40 years.

Japan's lost decade: Does money have a role?

We study the contribution of the stock of money to the macroeconomic outcomesof the 1990s in Japan using a small scale structural model. Likelihood-basedestimates of the parameters are provided and time stabilities of the structural relationshipsanalyzed. Real balances are statistically important for output and inflationfluctuations and their role has changed over time. Models which give moneyno role give a distorted representation of the sources of cyclical fluctuations. Thesevere stagnation and the long deflation are driven by different causes.

Female labour supply in Spain: The importance of behavioural assumptions and unobserved heterogeneity specification

We estimate four models of female labour supply using a Spanish sampleof married women from 1994, taking into account the complete form of theindividual s budget set. The models differ in the hypotheses relating tothe presence of optimisation errors and/or the way non-workers contributeto the likelihood function. According to the results, the effects of wagesand non-labour income on the labour supply of Spanish married women dependon the specification used. The model which has both preference andoptimisation errors and allows for both voluntarily and involuntarilyunemployed females desiring to participate seems to better fit the evidencefor Spanish married women.

Finance versus costs for teaching hospitals in Spain

In this paper we analyse the observed systematic differences incosts for teaching hospitals (THhenceforth) in Spain. Concernhas been voiced regarding the existence of a bias in thefinancing of TH s has been raised once prospective budgets arein the arena for hospital finance, and claims for adjusting totake into account the legitimate extra costs of teaching onhospital expenditure are well grounded. We focus on theestimation of the impact of teaching status on average cost. Weused a version of a multiproduct hospital cost function takinginto account some relevant factors from which to derive theobserved differences. We assume that the relationship betweenthe explanatory and the dependent variables follows a flexibleform for each of the explanatory variables. We also model theunderlying covariance structure of the data. We assumed twoqualitatively different sources of variation: random effects andserial correlation. Random variation refers to both general levelvariation (through the random intercept) and the variationspecifically related to teaching status. We postulate that theimpact of the random effects is predominant over the impact ofthe serial correlation effects. The model is estimated byrestricted maximum likelihood. Our results show that costs are 9%higher (15% in the case of median costs) in teaching than innon-teaching hospitals. That is, teaching status legitimatelyexplains no more than half of the observed difference in actualcosts. The impact on costs of the teaching factor depends on thenumber of residents, with an increase of 51.11% per resident forhospitals with fewer than 204 residents (third quartile of thenumber of residents) and 41.84% for hospitals with more than 204residents. In addition, the estimated dispersion is higher amongteaching hospitals. As a result, due to the considerable observedheterogeneity, results should be interpreted with caution. From apolicy making point of view, we conclude that since a higherrelative burden for medical training is under public hospitalcommand, an explicit adjustment to the extra costs that theteaching factor imposes on hospital finance is needed, beforehospital competition for inpatient services takes place.

Adaptive model selection using empirical complexities

Given $n$ independent replicates of a jointly distributed pair $(X,Y)\in {\cal R}^d \times {\cal R}$, we wish to select from a fixed sequence of model classes ${\cal F}_1, {\cal F}_2, \ldots$ a deterministic prediction rule $f: {\cal R}^d \to {\cal R}$ whose risk is small. We investigate the possibility of empirically assessingthe {\em complexity} of each model class, that is, the actual difficulty of the estimation problem within each class. The estimated complexities are in turn used to define an adaptive model selection procedure, which is based on complexity penalized empirical risk.The available data are divided into two parts. The first is used to form an empirical cover of each model class, and the second is used to select a candidate rule from each cover based on empirical risk. The covering radii are determined empirically to optimize a tight upper bound on the estimation error. An estimate is chosen from the list of candidates in order to minimize the sum of class complexity and empirical risk. A distinguishing feature of the approach is that the complexity of each model class is assessed empirically, based on the size of its empirical cover.Finite sample performance bounds are established for the estimates, and these bounds are applied to several non-parametric estimation problems. The estimates are shown to achieve a favorable tradeoff between approximation and estimation error, and to perform as well as if the distribution-dependent complexities of the model classes were known beforehand. In addition, it is shown that the estimate can be consistent,and even possess near optimal rates of convergence, when each model class has an infinite VC or pseudo dimension.For regression estimation with squared loss we modify our estimate to achieve a faster rate of convergence.

Efectos de la reforma del IRPF sobre la oferta laboral de las mujeres

The aim of this paper is to simulate the effects of the Spanish 1999 taxreform on the married women s labour behaviour and welfare in a partialequilibrium context. We estimate by maximum likelihood two models of laboursupply which take into account of the characteristics of the budgetconstraint. The simulation exercises suggest that the new tax can havesignificant effects on female s labour supply decisions and seems toincrease the individual s welfare.

A low dimensional Kalman filter for systems with lagged observables

This note describes how the Kalman filter can be modified to allow for thevector of observables to be a function of lagged variables without increasing the dimensionof the state vector in the filter. This is useful in applications where it is desirable to keepthe dimension of the state vector low. The modified filter and accompanying code (whichnests the standard filter) can be used to compute (i) the steady state Kalman filter (ii) thelog likelihood of a parameterized state space model conditional on a history of observables(iii) a smoothed estimate of latent state variables and (iv) a draw from the distribution oflatent states conditional on a history of observables.

The risk of divorce and household saving behavior

We analyze the impact of an increase in the risk of divorce on the savingbehaviour of married couples. From a theoretical perspective, the expected sign of theeffect is ambiguous. We take advantage of the legalization of divorce in Ireland in 1996as an exogenous increase in the likelihood of marital dissolution. We analyze the savingbehaviour over time of couples who were married before the law was passed. We proposea difference-in-differences approach where we use as comparison groups either marriedcouples in other European countries (not affected by the law change), or Irish familieswho did not experience a significant increase in the expected risk of divorce (such as veryreligious families, or single individuals). Our results suggest that the increase in the riskof divorce brought about by the law was followed by an increase in the propensity to saveof married couples, consistent with a rise in precautionary savings interpretation. Anincrease in the risk of marital dissolution of about 40 percent led to a 7 to 13 percent risein the proportion of married couples reporting positive savings.

Firm investment in imperfect capital markets: A structural estimation

We set up a dynamic model of firm investment in which liquidity constraintsenter explicity into the firm's maximization problem. The optimal policyrules are incorporated into a maximum likelihood procedure which estimatesthe structural parameters of the model. Investment is positively related tothe firm's internal financial position when the firm is relatively poor. This relationship disappears for wealthy firms, which can reach theirdesired level of investment. Borrowing is an increasing function of financial position for poor firms. This relationship is reversed as a firm's financial position improves, and large firms hold little debt.Liquidity constrained firms may be unused credits lines and the capacity toinvest further if they desire. However the fear that liquidity constraintswill become binding in the future induces them to invest only when internalresources increase.We estimate the structural parameters of the model and use them to quantifythe importance of liquidity constraints on firms' investment. We find thatliquidity constraints matter significantly for the investment decisions of firms. If firms can finance investment by issuing fresh equity, rather than with internal funds or debt, average capital stock is almost 35% higher overa period of 20 years. Transitory shocks to internal funds have a sustained effect on the capital stock. This effect lasts for several periods and ismore persistent for small firms than for large firms. A 10% negative shock to firm fundamentals reduces the capital stock of firms which face liquidityconstraints by almost 8% over a period as opposed to only 3.5% for firms which do not face these constraints.

Phylogenetic analyses of C4 photosynthesis evolution in grasses: genetics of convergence and convergence of genes

Summary : During the evolutionary diversification of organisms, similar ecological constraints led to the recurrent appearances of the same traits (phenotypes) in distant lineages, a phenomenon called convergence. In most cases, the genetic origins of the convergent traits remain unknown, but recent studies traced the convergent phenotypes to recurrent alterations of the same gene or, in a few cases, to identical genetic changes. However, these cases remain anecdotal and there is a need for a study system that evolved several times independently and whose genetic determinism is well resolved and straightforward, such as C4 photosynthesis. This adaptation to warm environments, possibly driven by past atmospheric CO2 decreases, consists in a CO2-concentrating pump, created by numerous morphological and biochemical novelties. All genes encoding C4 enzymes already existed in C3 ancestors, and are supposed to have been recruited through gene duplication followed by neo-functionalization, to acquire the cell specific expression pattern and altered kinetic properties that characterize Ca-specific enzymes. These predictions have so far been tested only in species-poor and ecologically marginal C4 dicots. The monocots, and especially the grass family (Poaceae), the most important C4 family in terms of species number, ecological dominance and economical importance, have been largely under-considered as suitable study systems. This thesis aimed at understanding the evolution of the C4 trait in grasses at a molecular level and to use the genetics of C4 photosynthesis to infer the evolutionary history of the C4 phenotype and its driving selective pressures. A molecular phylogeny of grasses and affiliated monocots identified 17 to 18 independent acquisitions of the C4 pathway in the grass family. A relaxed molecular clock was used to date these events and the first C4 evolution was estimated in the Chloridoideae subfamily, between 32-25 million years ago, at a period when atmospheric CO2 abruptly declined. Likelihood models showed that after the COZ decline the probability of evolving the C4 pathway strongly increased, confirming low CO2 as a likely driver of C4 photosynthesis evolution. In order to depict the genetic changes linked to the numerous C4 origins, genes encoding phopshoenolpyruvate carboxylase (PEPC), the key-enzyme responsible for the initial fixation of atmospheric CO2 in the C4 pathway, were isolated from a large sample of C3 and C4 grasses. Phylogenetic analyses were used to reconstruct the evolutionary history of the PEPC multigene family and showed that the evolution of C4-specific PEPC had been driven by positive selection on 21 codons simultaneously in up to eight C4 lineages. These selective pressures led to numerous convergent genetic changes in many different C4 clades, highlighting the repeatability of some evolutionary processes, even at the molecular level. PEPC C4-adaptive changes were traced and used to show multiple appearances of the C, pathway in clades where species tree inferences were unable to differentiate multiple C4 appearances and a single appearance followed by C4 to C3 reversion. Further investigations of genes involved in some of the C4 subtypes only (genes encoding decarboxylating enzymes NADP-malic enzyme and phosphoenolpyruvate carboxykinase) showed that these C4-enzymes also evolved through strong positive selection and underwent parallel genetic changes during the different Ca origins. The adaptive changes on these subtype-specific C4 genes were used to retrace the history of the C4-subtypes phenotypes, which revealed that the evolution of C4-PEPC and C4-decarboxylating enzymes was in several cases disconnected, emphasizing the multiplicity of the C4 trait and the gradual acquisition of the features that create the CO2-pump. Finally, phylogenetic analyses of a gene encoding the Rubisco (the enzyme responsible for the fixation of CO2 into organic compounds in all photosynthetic organisms) showed that C4 evolution switched the selective pressures on this gene. Five codons were recurrently mutated to adapt the enzyme kinetics to the high CO2 concentrations of C4 photosynthetic cells. This knowledge could be used to introgress C4-like Rubisco in C3 crops, which could lead to an increased yield under predicted future high CO2 atmosphere. Globally, the phylogenetic framework adopted during this thesis demonstrated the widespread occurrence of genetic convergence on C4-related enzymes. The genetic traces of C4 photosynthesis evolution allowed reconstructing events that happened during the last 30 million years and proved the usefulness of studying genes directly responsible for phenotype variations when inferring evolutionary history of a given trait. Résumé Durant la diversification évolutive des organismes, des pressions écologiques similaires ont amené à l'apparition récurrente de certains traits (phénotypes) dans des lignées distantes, un phénomène appelé évolution convergente. Dans la plupart des cas, l'origine génétique des traits convergents reste inconnue mais des études récentes ont montré qu'ils étaient dus dans certains cas à des changements répétés du même gène ou, dans de rares cas, à des changements génétiques identiques. Malgré tout, ces cas restent anecdotiques et il y a un réel besoin d'un système d'étude qui ait évolué indépendamment de nombreuses fois et dont le déterminisme génétique soit clairement identifié. La photosynthèse dite en Ça répond à ces critères. Cette adaptation aux environnements chauds, dont l'évolution a pu être encouragé par des baisses passées de la concentration atmosphérique en CO2, est constituée de nombreuses nouveautés morphologiques et biochimiques qui créent une pompe à CO2. La totalité des gènes codant les enzymes Ç4 étaient déjà présents dans les ancêtres C3. Leur recrutement pour la photosynthèse Ç4 est supposé s'être fait par le biais de duplications géniques suivies par une néo-fonctionnalisation pour leur conférer l'expression cellule-spécifique et les propriétés cinétiques qui caractérisent les enzymes C4. Ces prédictions n'ont jusqu'à présent été testées que dans des familles C4 contenant peu d'espèces et ayant un rôle écologique marginal. Les graminées (Poaceae), qui sont la famille C4 la plus importante, tant en termes de nombre d'espèces que de dominance écologique et d'importance économique, ont toujours été considérés comme un système d'étude peu adapté et ont fait le sujet de peu d'investigations évolutives. Le but de cette thèse était de comprendre l'évolution de la photosynthèse en C4 chez les graminées au niveau génétique et d'utiliser les gènes pour inférer l'évolution du phénotype C4 ainsi que les pressions de sélection responsables de son évolution. Une phylogénie moléculaire de la famille des graminées et des monocotylédones apparentés a identifié 17 à 18 acquisitions indépendantes de la photosynthèse chez les graminées. Grâce à une méthode d'horloge moléculaire relâchée, ces évènements ont été datés et la première apparition C4 a été estimée dans la sous-famille des Chloridoideae, il y a 32 à 25 millions d'années, à une période où les concentrations atmosphériques de CO2 ont décliné abruptement. Des modèles de maximum de vraisemblance ont montré qu'à la suite du déclin de CO2, la probabilité d'évoluer la photosynthèse C4 a fortement augmenté, confirmant ainsi qu'une faible concentration de CO2 est une cause potentielle de l'évolution de la photosynthèse C4. Afin d'identifier les mécanismes génétiques responsables des évolutions répétées de la photosynthèse C4, un segment des gènes codant pour la phosphoénolpyruvate carboxylase (PEPC), l'enzyme responsable de la fixation initiale du CO2 atmosphérique chez les plantes C4, ont été séquencés dans une centaine de graminées C3 et C4. Des analyses phylogénétiques ont permis de reconstituer l'histoire évolutive de la famille multigénique des PEPC et ont montré que l'évolution de PEPC spécifiques à la photosynthèse Ça a été causée par de la sélection positive agissant sur 21 codons, et ce simultanément dans huit lignées C4 différentes. Cette sélection positive a conduit à un grand nombre de changements génétiques convergents dans de nombreux clades différents, ce qui illustre la répétabilité de certains phénomènes évolutifs, et ce même au niveau génétique. Les changements sur la PEPC liés au C4 ont été utilisés pour confirmer des évolutions indépendantes du phénotype C4 dans des clades où l'arbre des espèces était incapable de différencier des apparitions indépendantes d'une seule apparition suivie par une réversion de C4 en C3. En considérant des gènes codant des protéines impliquées uniquement dans certains sous-types C4 (deux décarboxylases, l'enzyme malique à NADP et la phosphoénolpyruvate carboxykinase), des études ultérieures ont montré que ces enzymes C4 avaient elles-aussi évolué sous forte sélection positive et subi des changements génétiques parallèles lors des différentes origines de la photosynthèse C4. Les changements adaptatifs sur ces gènes liés seulement à certains sous-types C4 ont été utilisés pour retracer l'histoire des phénotypes de sous-types C4, ce qui a révélé que les caractères formant le trait C4 ont, dans certains cas, évolué de manière déconnectée. Ceci souligne la multiplicité du trait C4 et l'acquisition graduelle de composants participant à la pompe à CO2 qu'est la photosynthèse C4. Finalement, des analyses phylogénétiques des gènes codant pour la Rubisco (l'enzyme responsable de la fixation du CO2 en carbones organiques dans tous les organismes photosynthétiques) ont montré que l'évolution de la photosynthèse Ça a changé les pressions de sélection sur ce gène. Cinq codons ont été mutés de façon répétée afin d'adapter les propriétés cinétiques de la Rubisco aux fortes concentrations de CO2 présentes dans les cellules photosynthétiques des plantes C4. Globalement, l'approche phylogénétique adoptée durant cette thèse de doctorat a permis de démontré des phénomène fréquents de convergence génétique sur les enzymes liées à la photosynthèse C4. Les traces génétiques de l'évolution de la photosynthèse C4 ont permis de reconstituer des évènements qui se sont produits durant les derniers 30 millions d'années et ont prouvé l'utilité d'étudier des gènes directement responsables des variations phénotypiques pour inférer l'histoire évolutive d'un trait donné.

Development and validation of a clinical decision rule for the diagnosis of influenza.

INTRODUCTION: A clinical decision rule to improve the accuracy of a diagnosis of influenza could help clinicians avoid unnecessary use of diagnostic tests and treatments. Our objective was to develop and validate a simple clinical decision rule for diagnosis of influenza. METHODS: We combined data from 2 studies of influenza diagnosis in adult outpatients with suspected influenza: one set in California and one in Switzerland. Patients in both studies underwent a structured history and physical examination and had a reference standard test for influenza (polymerase chain reaction or culture). We randomly divided the dataset into derivation and validation groups and then evaluated simple heuristics and decision rules from previous studies and 3 rules based on our own multivariate analysis. Cutpoints for stratification of risk groups in each model were determined using the derivation group before evaluating them in the validation group. For each decision rule, the positive predictive value and likelihood ratio for influenza in low-, moderate-, and high-risk groups, and the percentage of patients allocated to each risk group, were reported. RESULTS: The simple heuristics (fever and cough; fever, cough, and acute onset) were helpful when positive but not when negative. The most useful and accurate clinical rule assigned 2 points for fever plus cough, 2 points for myalgias, and 1 point each for duration <48 hours and chills or sweats. The risk of influenza was 8% for 0 to 2 points, 30% for 3 points, and 59% for 4 to 6 points; the rule performed similarly in derivation and validation groups. Approximately two-thirds of patients fell into the low- or high-risk group and would not require further diagnostic testing. CONCLUSION: A simple, valid clinical rule can be used to guide point-of-care testing and empiric therapy for patients with suspected influenza.