Knowledge that Transforms

[Many accounting studies examine dichotomous choices (e.g., qualify/do not qualify an audit opinion or capitalize/do not capitalize a cost). These studies often involve small overall sample sizes, disparate response group sizes, and predictor variables that are skewed and collinear. These factors can cause distributional problems in test statistics for logit (or probit) regression models, which can lead to incorrect inferences. However, few empirical benchmarks exist for assessing the effect of these factors. The current paper determines how response group size and the number, distribution, and correlation of predictor variables affect empirical error rates and the minimum required sample size for using logit. Comparisons are made with the error rates obtained from an ordinary least squares (OLS) linear probability model, an alternative that has been suggested for small sample studies. Because accounting researchers choosing between logit and OLS may be concerned with more than the calibration of the models' test statistics, comparisons of the sensitivity of logit and OLS parameter estimates to the range of data sampled for the predictor variables and of the models' classificatory ability also are made. Both simulated and real accounting data are used. The results of Monte Carlo simulations show that logit test statistics are biased when the sample size is small. However, much of the bias is attributable to the skewness of the predictor variables, a problem that is characteristic of accounting research and that also affects OLS test statistics. In such settings, OLS may result in test statistics that are minimally better calibrated. The parameter estimates of the model, however, will be more sensitive to the sampling frame. Furthermore, experimentation with data on auditors' Statement No. 87 consistency judgments indicates that OLS also may result in higher Type I error rates when it is used for prediction or classification. These results are interpreted as indicating that, even for sample sizes as small as 50, logit rather than OLS still may be the preferable model for accounting choice studies.]

[Analyses of audit judgment have suggested or implied that frequency knowledge (i.e., knowledge of the base rates associated with error occurrences) acquired through experience is an important component of audit expertise. For example, auditors are assumed to use base-rate expectations about error occurrence (modified to reflect client-specific information) to allocate audit effort among financial statement accounts and to use likelihoods associated with alternative causes of errors that are identified by analytical review. This article reports two studies that examine both error-effect frequency knowledge, i.e., the frequency with which an individual financial statement account is affected by error, and error-cause frequency knowledge, i.e., the underlying reason for an error in a particular account. In the first study, auditors' knowledge of the frequencies with which errors affect financial statement accounts in five industries is compared with archival frequencies. In the second study, auditors' knowledge of both the causes and effects of errors in the manufacturing industry is compared with the archival frequencies of both error causes and their effects on financial statement accounts. The second study provides a within-auditor comparison of both types of error frequency knowledge and allows comparison with the empirical results on error causes (in manufacturing) reported in Libby (1985) and Libby and Frederick (1990). In both studies, the relation between error frequency knowlege and several measures of experience is examined. This article provides evidence on three research questions: (1) How many audits (in a particular industry) does an auditor experience? The answer to this question will help establish whether auditors learn error frequencies from direct, personal experience with financial statement errors or acquire their knowledge of error frequencies by other means. (2) What do auditors know about the relative frequencies actually associated with the population of financial statement errors discovered during the audit process? If error frequency knowledge is essential to audit expertise, then it is useful to understand the nature of that knowledge. Further, if expertise is to be measured, a valid empirical measure of knowledge must be developed. (3) Do more experienced auditors have more accurate error frequency knowledge than less experienced auditors? Discernible knowledge differences across experience levels would suggest that error frequency knowledge is gained through audit experience, consistent with the general psychological characterization of expertise. The results show, first, that even the most experienced auditors have limited direct experience with financial statement errors. Second, auditors seem to know only the most frequently occurring error effects and causes. Third, differences in auditors' knowledge of error effects across experience levels are not explained by differences in the length of either audit experience or industry-specific audit experience, or by the number of clients audited in an industry. Moreover, auditors with similar experience levels show large individual differences in knowledge of causes and effects. These results suggest that audit experience should be viewed as relating to specific audit tasks rather than as a singular, all-encompassing concept and that particular experience must be understood as it relates to a particular type of knowledge. Moreover, the results and the fact that financial statement errors are rare events raise questions about the value of normative and descriptive models that have been proposed as the basis for understanding audit judgment and expertise-models that assume knowledge of error frequencies.]

[The information furnished by management accounting systems aids top management in assessing the performance of lower levels and in setting proper incentives. These systems also provide information to lower levels which aids them in making operational decisions. Typically, detailed information is provided to lower levels in the organization, while only a summary of this information is furnished to top management. Therefore, in designing a management accounting system, a question arises as to the welfare effect of giving an employee access to detailed information, on which he can base his decisions, when such (detailed) information cannot be used in evaluating his performance. More generally, the question arises as to the welfare effects of increasing the informational asymmetry between upper and lower management by improving lower management's private pre-decision information system. Answering the above question can provide important insights into the proper design of firms' management accounting systems. We examine this issue using the principal-agent framework. In any given period, information reported in the managerial accounting system may be pre-decision or post-decision. When the system reports post-decision information and the contracts are complete, it is clear that the value of such information is non-negative. The value of pre-decision information is more difficult to assess. An agent who has access to better pre-decision information is able to use that information to make better decisions, given his objectives. However, the agent's objectives and the principal's objectives need not be the same. For example, the agent may use his better pre-decision information system to more successfully shirk on the job, making the principal strictly worse off. Thus, the principal is not necessarily better off by improving the agent's pre-decision information system. One way in which the principal can mitigate any negative effects of improving the agent's private pre-decision information system is to require the agent to communicate the private information. In this paper, we ignore the possibility of such communication. The reason for this is, as noted earlier, while large amounts of detailed information are provided to individuals at lower levels of the firm, only a small amount of that information is ever communicated to higher levels of the firm. Therefore, we view ignoring communication as an approximation. We examine a principal-agent model in which the principal can influence the extent to which the agent has superior private information on which the latter can base his action choice. We find sufficient conditions under which a strict Pareto improvement results from improving the agent's private pre-decision information system. This result arises because improving the agent's private pre-decision information system leads to improved coordination between the agent's information signal and action choice, which, in turn, results in an increase in the agent's average productivity. Although we do not find sufficient conditions under which a strict negative value might arise, we discuss some possible reasons and illustrate them with examples.]

[In this article, we examine the information content of announcements of increased reserves for loan loss by Citicorp and other banks, and the later write-off announcement made by the Bank of Boston. During 1987, most major U.S. banks, led by Citicorp on 19 May 1987, announced large increases in their loan loss reserves because of problem loans in lesser developed countries (LDC). With substantial flexibility in accounting rules for determining loss exposure, the banks announced varying levels of reserve increases. On 14 December 1987, the Bank of Boston began a second round of activity relating to LDC debt by announcing a $200 million write-off of LDC loans and further increase in loan loss reserves. Financial reporters suggested that these events could be interpreted differently. Because Citicorp was a leading money-center bank, its announcement could be interpreted favorably as a signal of willingness to deal with the LDC debt problem. This interpretation could similarly apply to other banks, especially the more exposed money-center banks. In comparison, the Bank of Boston announcement was portrayed in the press as detrimental to the money-center banks for two reasons. First, unlike a reserve increase, a write-off reduces a bank's capital adequacy ratio. Capital adequacy ratios are used by bank regulators in determining the need for, and the level of, supervisory intervention. Second, the write-off was construed as an effort by regional banks to exploit their relatively limited exposure to LDC loans as a competitive advantage in the domestic banking market. We find evidence consistent with the expectations of the financial press. The strongest stock-price increases associated with both the Citicorp announcement and the subsequent announcements of reserve increases by other banks were found for the banks with the greatest exposure to LDC debt. In contrast, those banks with the greatest exposure to LDC debt and with the largest reserves sustained the largest stock-price decreases at the Bank of Boston write-off announcement. The larger money-center banks sustained, on average, a three-day decline in value of 5 percent around the Bank of Boston announcement date.]

[SFAS 94 (1987) requires consolidation of all majority-owned subsidiaries (unless control is temporary or does not rest with the majority owners), including those of nonhomogeneous operations, large minority interests, or foreign locations. An effect of implementing the standard is that financial statements components other than net income and stockholders' equity will differ from those that would have been reported as if the standard did not apply. In this study, I use a sample of 72 companies to examine the security market reaction to the issuance of SFAS 94 that required consolidation of finance subsidiaries. The results indicate that the issuance of SFAS 94 was associated with significant negative excess stock returns. Of the hypotheses tested, this evidence is consistent with the prediction generated by the cash-flow effects hypothesis; but is inconsistent with the redistribution-effects hypothesis. In addition, no significant positive excess returns were obtained for nonconvertible debt securities of firms that did not consolidate prior to SFAS 94, which provides weak evidence of the dominance of the cash-flow effects of SFAS 94 over the redistribution effects.]

[Accounting Principles Board (APB) Statement No. 20 (1971) defines financial statement errors as items resulting "from mathematical mistakes, mistakes in the application of accounting principles, or the oversight or misuse of facts that existed at the time the financial statements were prepared" (APB 20, par. 13). This definition encompasses both intentional and unintentional misrepresentation by management. Errors affecting previously reported earnings are revealed as prior period adjustments as specified in Statement of Financial Accounting Standards No. 16 (1977). If the erroneous year's financial statements are presented, retroactive restatement is required (in accordance with APB 9 1966). Footnote disclosure of the nature of the error and its effect on earnings, earnings before extraordinary items, and earnings per share is also required. Although financial statement disclosures generally provide no indication that prior errors were intentional, they may be motivated by the same types of economic incentives influencing managers' choices of accounting methods or management of accruals. In this study, we examine the incidence of accounting errors revealed by prior period adjustments for 41 firms in comparison with a control group of another 41 firms. This comparison is used to highlight circumstances that are likely to motivate managers to use errors as an income management tool. While corrections of prior year earnings are rare for both over- and understatement errors, the latter are relatively less frequent. Our investigation revealed 41 overstatement firms but only three understatement firms, which is consistent with an income-increasing motivation. Because of the very small number of understatements, the analysis is limited to overstatement errors. We find that the earnings overstatements are negatively correlated with the growth in earnings. Analysis also indicates that earnings overstatements are more likely when firms have diffuse ownership, lower growth in earnings, and fewer income-increasing GAAP alternatives available. Overstatements are less likely among firms that have audit committees. These results are generally consistent with the view that overstatement errors are the result of managers responding to economic incentives.]

[This study investigates the measures of pension assets and liabilities disclosed under SFAS 87 to determine which most closely reflect those that investors implicitly assess when they value the firm. Several measures considered to have conceptual merit are disclosed under SFAS 87, but no single method has been deemed most appropriate. The attributes of the three asset and five liability measurement alternatives disclosed and the controversies surrounding the FASB's deliberations on these alternatives are considered in the research design. Because the pension asset and liability issues in SFAS 87 relate to measurement and reflect a variety of unresolved questions, an approach directly comparing measurement error across alternatives is used. The research design utilizes relevance and reliability, two primary accounting choice characteristics advanced by the FASB. These two attributes are operationalized by the variances and levels of differences between alternative measures and the implied investors' assessment. In most prior studies using cross-sectional valuation to address a variety of research questions, a model is posited and then tested by using book values as substitutes for unobservable variables. The measurement error is considered an econometric problem. Although the valuation equation assumed here is also based on unobservable market values, the measurement error is modeled, and the impact of the measurement error covariance structure on the bias in estimated regression coefficients is a fundamental research design feature. In prior research, this covariance structure is often either assumed to be zero (i.e., the error is "white noise") or left unspecified. The model is based on the attributes of historical costs for the differences between market values and book value amounts, including the candidate pension alternatives. With appropriate assumptions, a technique that exploits the implications of measurement errors in cross-sectional regression is used to compare levels of measurement error rather than purge or ignore it. Although several measures of pension assets and liabilities are found to be significant in explaining firm market value, differences among the alternatives are also significant. The fair value of plan assets and the accumulated benefit obligation each exhibit less measurement error than other alternatives for the entire sample. The projected benefit obligation has less measurement error variance for subsamples in which the salary progression rate includes expected inflation and productivity changes. These findings suggest that (1) footnote disclosures are closer to those assessed in market valuations than are the measures recognized in the balance sheet and (2) investors appear to include expectations about future salary progression in assessing pension liabilities, but view the projected benefit obligation measure as noisy.]

[This study addresses the issue of self-selection bias in the analysis of economic consequences of mandatory accounting changes. Self-selection bias arises from the use of truncated, nonrandom samples to assess the behavior of firms using different accounting methods at the time of the mandated change. Using ordinary least squares (OLS) to estimate regression models containing data generated by self-selected firms can yield inconsistent and inefficient estimates of regression parameters. The present study uses the case of SFAS No. 2, promulgated in 1974, to illustrate the effects of selection bias on studying the economic consequences of accounting regulation. The estimation method used to correct for self-selectivity is a two-stage switching regression procedure developed by Heckman (1976, 1979) and Lee (1976, 1978). Employing this research method requires developing a complete model that explains the accounting choice decision and R&D investment decision. The switching regression model is estimated with data from 1973 to correct for self-selection bias and to predict the likely economic consequences of SFAS No. 2 prior to its adoption. The unbiased estimates of the R&D equations are then used with the Wald test to examine structural changes in the R&D model after the implementation of SFAS No. 2. To examine the sensitivity of the results to self-selection bias, the analysis is replicated with OLS estimates. The results of the switching regression analysis indicate that selection bias exists in both the capitalizing and expensing groups. This bias is further shown in systematic differences between the results of OLS and switching regression estimates. The OLS estimates consistently understate the predicted values of R&D expenditures for both groups and appear to understate the negative impact of SFAS No. 2 on the capitalizers' R&D expenditures. The results of the Wald test show that observed changes in the capitalizers' R&D spending behavior after 1974 are attributable, at least in part, to general macroeconomic phenomena. However, after controlling for the effects of economywide changes, the analysis shows an incremental effect of SFAS No. 2 on the R&D expenditures of former capitalizers.]