Conventional analysis predicts that workers pay part of their on-the-job training costs by accepting a lower starting wage and subsequently realize a return to this investment in the form of greater wage growth. Missing from the conventional treatment of on-the-job training is a discussion of the process by which heterogeneous workers are matched to jobs requiring varying amounts of training. This matching process constitutes a key feature of the on-the-job training model presented in this article and tested with a unique data set containing extensive information concerning on-the-job training, employer search, wages, and wage and productivity growth.

Abstract. In conventional dollar unit sampling (DUS) the “alpha” risk of erroneously rejecting an acceptable accounting population generally can be held to reasonable levels by working with a sufficiently large audit sample. Kaplan (1975) has discussed how such a DUS sample size may be chosen when controlling both the alpha and beta risk. However, as illustrated in this paper, neither Kaplan's procedure nor other previously developed DUS procedures provide a workable method that effectively controls for both the alpha and beta risk with small sample sizes or with accounting populations that have comparable understatement and overstatement errors. This paper presents a hypothesis‐testing approach to DUS that provides a solution to this problem. “Power curves” based on a large number of simulation studies of the new approach are presented in order to illustrate the reliability of the approach. Comparisons are also made with the corresponding power curves for two Stringer‐based confidence bound methods of DUS. In these comparisons, the new hypothesis‐testing approach to DUS appears to provide more favorable audit performance with smaller sample sizes than do the conventional DUS methods, although the primary advantage of the approach lies in its ability to control for both effectiveness ( beta risk) and efficiency ( alpha risk). Résumé. Dans les méthodes traditionnelles de sondage des unités monétaires (SUM), le risque «alpha», soit le risque de rejeter à tort une population comptable acceptable, peut en général être maintenu à des niveaux raisonnables si l'on travaille avec un échantillon de vérification suffisamment grand. Kaplan (1975) a traité de la façon de choisir une taille d'échantillon suffisamment grande aux fins du SUM dans les cas où l'on contrôle à la fois le risque alpha et le risque bêta. Toutefois, comme le démontrent les auteurs, ni la méthode de Kaplan ni les autres méthodes de SUM mises au point précédemment n'offrent de technique simple permettant de contrôler efficacement le risque alpha en même temps que le risque bêta avec de petites tailles d'échantillon ou des populations comptables présentant des erreurs de sous‐estimation et de surestimation comparables. Les auteurs suggèrent une méthode de test d'hypothèse, relativement au SUM, qui résout ce problème. Ils présentent des «courbes de puissance», fondées sur un grand nombre de cas de simulation de la nouvelle méthode, dans le but d'en illustrer la fiabilité. Ils établissent également des comparaisons avec les courbes de puissance correspondantes pour deux méthodes de limite de confiance du SUM, fondées sur les travaux de Stringer. Dans ces comparaisons, les nouvelles méthodes de test d'hypothèse en ce qui a trait au SUM semblent offrir un meilleur rendement de vérification lorsque les tailles d'échantillon sont petites que les méthodes traditionnelles de SUM, bien que l'avantage primordial de la méthode réside dans le fait qu'elle permet de contrôler à la fois l'efficacité (risque bêta ) et l'efficience (risque alpha ).

This paper examines the Beaver, Lambert, and Morse (1980) valuation model that capitalizes ‘ungarbled’ earnings. The analysis identifies the model's unspecified capitalization factor, and it emphasizes that a complete model of ungarbled earnings must focus on the joint stochastic behavior of earnings and dividends. It is shown that expected ungarbled earnings, scaled for a constant, equal expected dividends. Except for the scale factor, no apparent economic reasons suggest that ungarbled earnings are any different from dividends.

This paper provides a simple method to account for heteroskedasticity and cross-sectional dependence in samples with large cross sections and relatively few time-series observations. The method is motivated by cross-sectional regression studies in finance and accounting. Simulation evidence suggests that these estimators are dependable in small samples and may be useful when generalized least squares is infeasible, unreliable, or computationally too burdensome. We also consider efficiency issues and show that, in principle, asymptotic efficiency can be improved using a technique due to Cragg (1983).

Résumé. Dans le présent article, les auteurs font état des résultats d'une recherche dans le cadre de laquelle des vérificateurs expérimentés ont interprété les critères du Statement of Financial Accounting Standards No. 5 (Norme n° 5): Accounting for Contingencies (Comptabilisation des éventualités). Cette recherche est axée sur deux questions: 1) la nature et le degré du consensus qui se dégage des interprétations des vérificateurs et 2) la mesure dans laquelle ces interprétations dépendent de la nature de la perte éventuelle. Quarante‐cinq vérificateurs expérimentés (chefs de groupe, directeurs et associés) issus de cabinets d'experts‐comptables figurant parmi les «huit grands» se sont prêtés à un exercice visant à recueillir leur interprétation des critères de probabilité de la Norme n° 5. Nous avons centré notre analyse sur le seuil auquel s'établit la distinction entre une perte «improbable» et une perte «plausible», et entre une perte «plausible» et une perte «probable». Les résultats de nos recherches ont indiqué: 1) un seuil moyen respectif de 0, 16 et 0,68; 2) un consensus plus manifeste chez les vérificateurs dans le cas du premier seuil que dans le cas du second; 3) un premier seuil beaucoup plus bas que ne l'avaient laissé supposer les précédents travaux de recherche; et 4) dans l'ensemble, l'absence de lien entre les seuils et la nature de la perte éventuelle.

Abstract. This paper reports the results of research in which experienced auditors interpreted the criteria of Statement of Financial Accounting Standards No. 5 (SFAS 5): Accounting for Contingencies. The research focuses on two issues: (1) the nature and degree of consensus in the auditors' interpretations, and (2) the extent to which these interpretations depend upon the type of contingent loss. Forty‐five experienced auditors (managers, principals, and partners) from “Big 8” CPA firms responded to a research instrument that elicited their interpretation of SFAS 5 probability criteria. Our analysis focuses upon the thresholds between the “remote” and “reasonably possible” criteria and between the “reasonably possible” and “probable” criteria. Our results indicate: (1) threshold means of 0.16 and 0.68, respectively; (2) more auditor consensus for the first threshold than for the second; (3) the first threshold was significantly lower than indicated by previous research; and (4) the thresholds were generally not dependent upon the type of contingent loss.

A common finding is that the forward discount is a biased predictor of future exchange rate changes. We use survey data on exchange rate expectations to decompose the bias into portions attributable to the risk premium and expectational errors. None of the bias in our sample reflects the risk premium. We also reject the claim that the risk premium is more variable than expected depreciation. Investors would do better if they reduced fractionally the magnitude of expected depreciation. This is the same result that many authors have found with forward market data, but now it cannot be attributed to risk.

We derive and test an alternative closed-form general equilibrium model of the term structure within the Cox, Ingersoll, and Ross theoretical framework in which yields are nonlinear functions of the risk-free rate. We show that equilibrium bond prices and the risk-free rate are not always inversely related and that bond risk need not be strictly increasing in maturity. Using Hansen's generalized method of moments to obtain parameter estimates, this nonlinear model outperforms the Cox, Ingersoll, and Ross square root model in describing actual Treasury bill yields for the 1964–1986 period.