Search

We investigate (1) how well a machine learning algorithm can predict one-year ahead effective tax rates (ETRs) and (2) which items in the financial statements and notes are most useful for these predictions. We compare our machine-generated ETR predictions with those from ETRs implied by analysts' earnings forecasts and find the algorithm's predictions are less biased, more precise, and explain more of the variance in future ETRs. We then use Explainable AI (based on Shapley values) to measure the usefulness of each disclosure item in the algorithm's predictions. We find that while some tax-related items are useful, others offer minimal value. Using the machine learning algorithm's use of information as a benchmark, we then further use Shapley values to examine which information is underweighted or overweighted by analysts. Overall, our results help inform standard setters on the relevance of certain tax disclosures in achieving the objective of predicting future ETRs. JEL Classifications: G17. [ABSTRACT FROM AUTHOR]

This paper presents a do-it-yourself algorithm to generate the historical GVKEY-CIK link table. The proposed algorithm features a technique of pre-classifying sample data into different treatment subgroups and utilizing historical firm information available from the source data to increase (reduce) matching efficiency (errors). Simulation results show that our algorithm is superior to applying only conventional name matching operations over the whole sample: 57.5 percent of the overall matching results are error free ex ante, and for the remaining 42.5 percent of data, records without Type I errors (with Type II errors) increase (decrease) by 34.0 percent (59.4 percent) when the optimal threshold is used. JEL Classifications: C89; M40; G10; G18. [ABSTRACT FROM AUTHOR]

Many auditing expert systems have been developed for individual components of the audit planning process such as inherent risk assessment and internal control evaluation, The audit planning process involves integrating judgments from each step in the process into subsequent steps. Based on a task analysis of the audit planning process, this paper presents a generalizable knowledge-based framework for the development of expert systems encompassing the entire audit planning process from assessment of overall audit risk to selection of appropriate audit procedures. The framework, which incorporates the audit risk model, suggests the use of knowledge-bases for the heuristic reasoning components of audit planning expertise and a database for the factual knowledge components. Knowledge of how to select appropriate audit procedures given a particular audit situation is best represented using a pattern matching algorithm. The paper also describes an attribute-based design approach which facilitates flexibility in the resulting expert system. [ABSTRACT FROM AUTHOR]

This paper provides a brief review of the salient economic, behavioral and quantitative literature in accounting and transfer pricing. Further, it relates this review to a relatively new programming formulation, suggesting the greater descriptive implications potential inherent in the new technique generalized goal decomposition programming. The paper does not develop new ideas, but attempts to relate existing material in a manner that will bring the reader's attention to the potential of the mathematical programming models in hierarchical organizations. The expanding literature in this area clearly is deserving of attention as it provides a means by which the behavioral and algorithmic approaches to resource transfer problems may be integrated. It is a step in the direction of models integration and thus, a step toward more realistic mathematical formulation of the decision-making environment. One rely on the scholar T. Ruefli model in this presentation as it provided a relatively easily explained formulation. Interested readers will find on pursuing this literature, that certain of the Ruefli limitations have been overcome in alternative algorithms.

This paper constructs a model to explain the determinants of the extent of product cost bias which results when a firm employs a conventional two-stage overhead cost allocation system. The extent of product cost bias has previously been hypothesized to be a function of (1) production technology heterogeneity, (2) unit input costs, and (3) product mix. Our model provides an explicit framework within which to analyze such factors and their interactions. In particular, we derive an expression for the firm's economic loss from product cost distortion as a function of each product's squared bias and the extent of product market competition. In turn, the squared bias is shown to be a function of the heterogeneity of the production technology, unit input costs, and the product mix. In addition, the paper uses the model described above to develop and evaluate algorithms which managers can use as practical guides in choosing the best allocation base at each overhead cost pool. Such algorithms are potentially valuable to firms in two ways. First, in designing the firm's cost system, particularly the overhead cost pool structure, the algorithms can help managers to quantify the magnitude of cost distortion induced by alternative allocation bases, thus allowing improved tradeoffs to be made. Second, once an overhead cost pool design has been chosen, the proposed algorithms can also be used in selecting the best allocation base at each cost pool. [ABSTRACT FROM AUTHOR]

This paper examines whether the use of algorithms in accounting education provides effective professional training for students. Based on empirical research findings, the paper argues that algorithmic methods may be appropriate only for selected topics. The paper identifies the characteristics of such topics and points out that though substantial benefits are associated with proper use of algorithms, accounting instructors should be aware of some significant risks posed by algorithmic approaches. Finally, the paper offers some suggestions to minimize the risks and improve the effectiveness of algorithm-based pedagogies. [ABSTRACT FROM AUTHOR]

In their recently published intermediate accounting text, Danos and Imhoff show how one can capitalize R&D expenditures and (after a period of convergence) still show the same R&D expense as occurs now under FASB Statement No. 2, which requires immediate expensing of R&D. This paper develops an analytical tool to present and analyze this capitalization scheme: a coordinate grid model. This coordinate model enables one to see relationships between expenditures and expenses visually. As an indication of the power of this visual representation of pattern, an algorithm is presented that was derived from the coordinate grid. The algorithm calculates the net asset value of R&D under the conditions of the capitalization model. [ABSTRACT FROM AUTHOR]

An "ultimate learning algorithm" is one that produces models that closely match the real world's underlying distribution of functions. To try to create such an algorithm, researchers typically employ manual algorithm design with cross-validation. It has been shown that cross-validation is not a viable way to construct an ultimate learning algorithm. For machine learning researchers, "meta-learning" should be more desirable than manual algorithm design with cross-validation. Meta-learning is concerned with gaining knowledge about learning methodologies. One meta-learning approach involves evaluating the suitability of various algorithms for a learning task in order to select an appropriate algorithm. An alternative approach is to incorporate predictions from base algorithms as features to be evaluated by subsequent algorithms. This paper reports on exploratory research that implemented the latter approach as a three-layer stacked generalization model using neural networks, logistic regression, and classification tree algorithms to predict all categories of financial fraud. The purpose was to see if this form of meta-learning offered significant benefits for financial fraud prediction. Fifteen possible financial fraud predictors were identified based on a theoretical fraud model from prior research. Only public data for these possible predictors were obtained from U.S. Securities and Exchange Commission filings from the period 1995-2002 for a sample of 50 fraud and 50 non-fraud companies. These data were selected for the year prior to when the fraud was initiated. These variables were used to create a variety of neural network, logistic regression, and classification tree models while using holdout sample and cross-validation techniques. A 71.4 percent accurate neural network model was then stacked into a logistic regression model, increasing the prediction accuracy to 76.5 percent. The logistic regression model was subsequently stacked into a classification tree model to achieve an 83 percent accuracy rate. These results compared favorably to two prior neural network studies, also employing only public data, which achieved 63 percent accuracy rates. Model results were also analyzed via probability-adjusted overall error rates, relative misclassification costs, and receiver operating characteristics. The increase in classification accuracy from 71 percent to 83 percent, the decline in estimated overall error rate from 0.0057 to 0.0035, and the decline in relative misclassification costs from 2.79 to 0.58 suggest that benefits were achieved by the meta-learning stacking approach. Further research into the meta-learning stacking approach appears warranted. [ABSTRACT FROM AUTHOR]

Controls must be changed when information systems are modified. Audit, assurance, and quality-control (AAQC) personnel must evaluate the reliability of controls in the new system versions. Based on Bunge (1977, 1979) and Wand and Weber (1989a, 1990), we describe a model and search-space algorithm that AAQC personnel can use to determine where required control changes are likely to be located in the new system version, thereby mitigating the need for an exhaustive evaluation of all controls. To use the model and algorithm, AAQC personnel must have (1) accurate and complete requirements specifications for the old and new versions of the system, (2) a controls specification for the old version that covers all errors and irregularities that might occur, (3) evidence to conclude all controls for the old version are in place, adequate, and working, and (4) specifications for the new version expressed as a level structure of systems and subsystems. [ABSTRACT FROM AUTHOR]

Despite the clear prescription from economic theory that a firm should set price based only on variable costs, firms routinely factor fixed costs into pricing decisions. We show that full-cost pricing (FCP) can achieve the optimal price. FCP marks up variable cost with the contribution margin per unit, which, in equilibrium, includes the fixed cost. FCP converges to the optimal price when the firm can estimate its equilibrium income. We compare FCP to alternative pricing algorithms that require less information, but converge to optimal price under more narrow conditions than FCP. [ABSTRACT FROM AUTHOR]

ABSTRACT: Although many firms now prepare a cash-based statement of changes in financial position, reported "funds from operations" is typically a measurement of working capital. The statement reader wishing to determine cash flow from operations must use a series of indirect adjustments to do so. Our purpose in this paper is to identify the inherent difficulties the reader encounters when using this indirect method to convert reported funds from operations to a cash flow amount. We demonstrate how the indirect method will in fact not equal actual cash flow from operations because of numerous conceptual and practical problems encountered when applying the necessary adjustments. We develop illustrations of these problems, and we show how cash flow from operations calculated by the indirect method is, at best, an estimate of actual cash provided by operations. [ABSTRACT FROM AUTHOR]

Among many things accountants deal with in their daily work, there is one item that has had the largest growth rate in volume over the last ten years. It is not the volume of pronouncements by the Financial Accounting Standards Board nor that of regulations by the Internal Revenue Service of the U.S., but calculators. Prices have plunged as the volume exploded. Almost exactly ten years ago, as of 1983, the first hand-held scientific calculator was marketed by Hewlett-Packard at the price of $395 a unit. Such calculators are now available for under $20 and are owned by millions of students. Experts seem to agree that the next wave of computer technology coming into the business and consumer product markets is personal computers. If the past trend is any indication of the learning curve applicable to computer products, those units currently selling for $300 to $3000 will come down in price to the $30 to $300 range in several years. Dozens of accounting software packages have been developed for personal computers, dealing with general ledgers, accounts receivable and payable, payrolls, and inventories, along with general-purpose report generators.

In contrast with classical, randomization-based strategies, model-based sampling is purposive, i.e., not random in nature. Unfortunately, the optimality of model-robust procedures depends heavily on the assumption that the true form of the superpopulation model is known to the sampler prior to sampling. To date, model-based sampling has not found wide application due to concerns about the robustness of the approach in the presence of superpopulation model misspecification. This paper evaluates the robustness of model-based sampling strategies in audit settngs. In doing so we introduce the concept of model-robust sampling, an extension of model-based sampling which provides some protection against model misspecification. An efficient algorithm for sample selection is presented. Simulation is used to measure the robustness of the various model-based approaches to changes in assumptions about the target population. We conclude that while substantive gains in efficiency are possible through model-based sampling, randomization-based strategies should be preferred in the absence of reliable prior information as to the assumed form of the variance function. [ABSTRACT FROM AUTHOR]

The article discusses the factors that might influence reliance on decision aids, including monetary incentives, feedback, expertise, experience, accountability, justification, competition and predictive ability. It focuses on the experiment, including the subjects and operationalization of decision aid explanation and framing. It addresses the development and testing of hypothesis. The authors recommend that knowledge of the algorithm used to develop the aid would increase confidence and should, therefore, increase reliance on the decision aid. The users should shift to greater reliance on the aid as the level of decision difficulty increases.

This study compares the performance of six popular statistical and machine learning models in detecting financial statement fraud under different assumptions of misclassification costs and ratios of fraud firms to nonfraud firms. The results show, somewhat surprisingly, that logistic regression and support vector machines perform well relative to an artificial neural network, bagging, C4.5, and stacking. The results also reveal some diversity in predictors used across the classification algorithms. Out of 42 predictors examined, only six are consistently selected and used by different classification algorithms: auditor turnover, total discretionary accruals, Big 4 auditor, accounts receivable, meeting or beating analyst forecasts, and unexpected employee productivity. These findings extend financial statement fraud research and can be used by practitioners and regulators to improve fraud risk models [ABSTRACT FROM AUTHOR]

Influence diagrams are presented as a complement to analytical methods for solving capacity-planning and pricing problems under uncertainty. An influence diagram model that permits continuous decision variables is used to determine optimal decision rules for a single product monopolist making capacity and price selections once per period. Solutions under different scenarios based on the timing of actual demand observations and the ability to adjust initial capacity are compared to previous solutions in the literature. The results demonstrate that optimal decision rules for capacity and price determined using influence diagrams closely approximate those found using analytical approaches. The influence diagram is used to determine the effect of changing demand function parameters on the optimal solutions. Additionally, the influence diagram model can be adapted to solve capacity-planning problems that do not meet the assumptions required to obtain analytical solutions or heuristics in previous studies. [ABSTRACT FROM AUTHOR]

While behavioral research has consistently documented the advantages of using decision aids, decision makers have been reluctant to rely on them, generally with unfavorable results. We offer a model of aided decisions that incorporates both the level of reliance on the decision aid and the agreement of decisions with the decision aid's advice. In the model, the level of reliance is an attribute of the decision strategy adopted by the aided decision maker, while the extent of the decision maker's agreement with the decision aid is observed by comparing the decisions with the aid's advice. The model also includes a classification scheme for factors that affect decision strategy. The categories in this classification scheme are decision aid features, decision-maker characteristics, decision task characteristics and factors affecting the evaluation of decision strategies. In an experiment testing the sensitivity of decision aid reliance, neither the extent of algorithm revelation (revealed or bidden) nor decision framing (positive or negative) affected the level of reliance on the decision aid. In a comparison of aided and unaided decision makers, the decision aid had greater influence in the harder tasks than in the easier tasks. [ABSTRACT FROM AUTHOR]