Relative Valuation with Machine Learning

We use machine learning for relative valuation and peer firm selection. In out-of-sample tests, our machine learning models substantially outperform traditional models in valuation accuracy. This outperformance persists over time and holds across different types of firms. The valuations produced by machine learning models behave like fundamental values. Overvalued stocks decrease in price and undervalued stocks increase in price in the following month. Determinants of valuation multiples identified by machine learning models are consistent with theoretical predictions derived from a discounted cash flow approach. Profitability ratios, growth measures, and efficiency ratios are the most important value drivers throughout our sample period. We derive a novel method to express valuation multiples predicted by our machine learning models as weighted averages of peer firm multiples. These weights are a measure of peer–firm comparability and can be used for selecting peer-groups.     

OUTPUT,OUTCOME, AND QUALITY ADJUSTMENT IN MEASURING HEALTH AND EDUCATION SERVICES

This paper provides an overview of measuring price and volume changes of the output of health and education providers. In the national accounts, outputs should reflect the results of production and these cannot normally be captured by outcome, the state of health or education of the population. However, we show that outcome information is required when it comes to quality adjustment of output measures. The paper clarifies terminology, and discusses output measurement and quality adjustment methods with a focus on health and education services.     

Low-Risk Anomalies?

This paper shows that low-risk anomalies in the capital asset pricing model and in traditional factor models arise when investors require compensation for coskewness risk. Empirically, we find that option-implied ex ante skewness is strongly related to ex post residual coskewness, which allows us to construct coskewness factor-mimicking portfolios. Controlling for skewness renders the alphas of betting-against-beta and betting-against-volatility insignificant. We also show that the returns of beta- and volatility-sorted portfolios are driven largely by a single principal component, which in turn is explained largely by skewness.     

On the comparison of dollar-unit and stratified mean-per-unit estimators

Abstract. The paper discusses the findings reported by Smieliauskas (1986) and expands on them. It focuses on the dominance criterion for cost efficiency and on the usefulness of power curves for stratified mean-per-unit estimators. The analyses lead to the conclusions that: (1) while an audit plan which dominates another is associated with lower cost, a plan with lower cost does not necessarily dominate: (2) the distribution function of the t statistic becomes normal for sufficiently large samples, but the approach to normality is not necessarily uniform as sample size increases; (3) large sample size is required for nominal confidence levels to be a good approximation to the true confidence level, and the true confidence level can move temporarily farther from the nominal level as the sample size increases; and (4) the lack of reliability of ex ante power curves for the stratified mean-per-unit estimator makes it difficult to decide which power curve dominates. Résumé. Cet article discute des résultats publiés par Smielauskas (1986) et les extrapole. Il met l'accent sur le critère de dominance pour l'efficacité des coûts et sur l'efficacité des courbes de puissance pour les estimateurs de moyenne-par-unité stratifiée. Les analyses portent à conclure que: (1) même si un programme de vérification dominant par rapport à un autre est associé à un coût inférieur, un programme à coût inférieur n'est pas nécessairement dominant; (2) la fonction de distribution de la statistique t tend vers la normale pour des échantillons suffisamment grands, mais ce rapprochement vers la normalité n'est pas nécessairement uniforme à mesure que la taille de l‘échantillon devient plus grande; (3) un échantillon de grande taille est requis afin que les seuils de confiance “nominaux” constituent une bonne approximation du seuil de confiance “réel”, et le seuil de confiance “réel” peut s’éloigner temporairement du seuil de confiance “nominal” à mesure que s'accroît la taille de l‘échantillon; et (4) l'absence de fiabilité des courbes de puissance ex ante pour l'estimateur de moyenne-par-unité stratifiée rend ardue la sélection de la courbe de puissance dominante.     

Immigrant Generation and Income in Australia*

Microdata for adult men from the 1981 Australian Census are used to study the determinants of income by immigrant generation. The overseas born, 30 per cent of adult males, have 5 per cent lower incomes than the native born, and ceteris paribus, 7 per cent lower incomes. Schooling and pre-immigration labour market experience have smaller effects for the overseas born. Among the Australian born, those with overseas-born parents have 4 per cent higher incomes overall, but, ceteris paribus, there is no difference. The means and partial effects of the explanatory variables among the native born are not related to the parents' nativity.