Marjorie B. Platt

Probabilistic Neural Networks in Bankruptcy Prediction

Abstract Prior application of neural networks in finance and accounting, notably in bankruptcy prediction, are limited to back-propagation neural networks. Their well-known disadvantages, however, limit the practical usefulness of neural discriminant models. Instead, a probabilistic neural network with fewer of these difficulties is proposed. Using data from the U.S. oil and gas industry, alternate methodologies are compared. Whereas probabilistic neural networks without pattern normalization and Fisher discriminant analysis achieve the best overall estimation results, discriminant analysis produces superior results for bankrupt companies.

A note on the use of industry-relative ratios in bankruptcy prediction

Abstract The potential for industry-relative financial ratios to improve the prediction of firms in financial distress motivated this comparison of model specifications based on either unadjusted or industry-relative ratios. Both specifications yielded stable parameter estimates over the time periods examined. However, the industry-relative specification appeared to add incremental information not contained in the model based on the unadjusted financial ratios; the converse case did not hold. In addition, with the industry-relative specification, ex post forecast accuracy was slightly improved relative to the ex ante forecast, while with the unadjusted model specification, ex post forecast accuracy declined from that obtained ex ante.

Predicting corporate financial distress: Reflections on choice-based sample bias

Financial distress precedes bankruptcy. Most financial distress models actually rely on bankruptcy data, which is easier to obtain. We obtained a dataset of financially distressed but not yet bankrupt companies supplying a major auto manufacturer. An early warning model successfully discriminated between these distressed companies and a second group of similar but healthy companies. Previous researchers argue the matched-sample design, on which some earlier models were built, causes bias. To test for bias, the dataset was partitioned into smaller samples that approach equal groupings. We statistically confirm the presence of a bias and describe its impact on estimated classification rates.

Classroom participation and discussion effectiveness: student-generated strategies

Classroom discussion is one of the most frequently used and often embraced pedagogical strategies. In attempting to enhance participation quality and discussion effectiveness, there is concern over what to do about students who are less inclined to participate voluntarily. We examined the context of intensive graduate business classes—in which the instructor had high expectations for participation, placed significant weight on the participation grade, and cold called (i.e., called on students whose hands were not raised). In a questionnaire, we asked students to identify what enhanced the quality of participation and the effectiveness of discussion in this class. Qualitative content analysis indicated that student responses clustered in several areas: (1) required/graded participation, (2) incorporating ideas and experiences, (3) active facilitation, (4) asking effective questions, (5) supportive classroom environment, and (6) affirming contributions/constructive feedback. The results strongly endorse the practice of cold calling. The class instructor utilized student responses to formulate future teaching strategies.

Business cycle effects on state corporate failure rates

Abstract The effects of economic conditions, business costs, and new business formations on business failures were estimated for the time period 1969–1982. A cross-sectionally correlated and temporally autoregressive model was used to estimate model parameters. Significant differences were uncovered in the causes of business failure across the 48 states in the continental U.S. and Washington, DC. Four subgroups of states with common business failure conditions form the core of the analysis. Within each subgroup of states, economic conditions, business costs, and new business formations were found to be significant determinants of corporate failure rates with common parameter estimates.

Predicting Failure of Savings & Loan Associations

Early-warning systems are intended to provide regulators with identification of problem institutions sooner than is possible with the present system of call reports and periodic on-site examinations. Given an earlier indication of potential problems, resources can be focused on those institutions most in need of monitoring, thus potentially reducing both the number of failures and FSLIC financial assistance in the remaining failures. While some early-warning systems were developed in the mid-1970s for use by commercial bank regulators, very little attention has been focused on developing similar systems for the thrift industry. To address this issue, multiple discriminant analysis was used to develop an early-warning system for savings and loan associations in the Boston district of the Federal Home Loan Bank system. The results suggest that use of this model would have given signals of impending trouble well before the actual failure occurred. Consequently, an early-warning system could provide information to regulators sooner, permitting scarce resources to be allocated more effectively. In addition, earlier intervention could reduce the amount of FSLIC financial assistance required. By intervening earlier, FSLIC may be able to arrest the failure-promoting activities in which the association is engaged. Copyright American Real Estate and Urban Economics Association.

Nonvoluntary Class Participation in Graduate Discussion Courses: Effects of Grading and Cold Calling.

Classroom discussion is a frequently used “active learning” strategy; however, a dilemma this strategy produces is what to do about students who are less inclined to volunteer. This study examines an environment which uses cold calling and graded participation to include more students in discussions. The results suggest that despite widespread concerns about cold calling, students are not uncomfortable. This environment also is shown to increase participation frequency and marginally affect preparation, which in turn lead to increased comfort. Practical guidance is provided to help instructors design cold calls to ensure students have a positive experience and fully benefit from inclusion in the discussion.

Sustainable growth rate of firms in financial distress

Sustainable growth rate defines the rate at which a company’s sales and assets can grow if the company sells no new equity and wishes to maintain its capital structure. The traditional formula assumes that the firm can increase its indebtedness. Many private firms and most firms in financial distress have limited or no access to debt markets. While distressed firms may prefer a no growth strategy, external pressures such as inflation or demand increases may cause their sales to rise exogenously. A new sustainable growth rate formula is developed that describes how much growth the firm with no new debt capacity can endure.

Impact of Cold-Calling on Student Voluntary Participation

Classroom discussion is perhaps the most frequently used “active learning” strategy. However, instructors are often concerned about students who are less inclined to participate voluntarily. They worry that students not involved in the discussion might have lower quality learning experiences. Although instructors might consider whether to call on a student whose hand is not raised (“cold-call”), some instructors resist cold-calling fearing that the student will feel uncomfortable. This study examines the impact of cold-calling on students’ voluntary participation in class discussions and their comfort participating in discussions. The results demonstrate that significantly more students answer questions voluntarily in classes with high cold-calling, and that the number of students voluntarily answering questions in high cold-calling classes increases over time. Furthermore, students in classes with high cold-calling answer more voluntary questions than those in classes with low cold-calling; this also increases over time. Finally, in classes with high cold-calling, students’ comfort participating in class discussions increases while in classes with low cold-calling, students’ comfort participating does not change. Research findings show that cold-calling can be done fairly extensively without making students uncomfortable. Thus, the research reported here provides support for using this instructional strategy to engage more students to participate in discussions.

Cognitive Integration and Referential Communication: Effects of Information Quality and Quantity in Message Decoding

Four experiments are reported in which respondents were presented with one or more written passages, describing the face of an actor who was portraying a particular emotional expression (e.g., surprise). After reading each message ensemble, the respondents attempted to select the appropriate targetreferent from an array of 24 photographs showing the same actor in different emotional poses. The results indicated that performance was consistently affected by the quality of the message ensembles (as quantified by the percentage of relatively accurate descriptions in the set) and by the quantity of information that was available (the number of different descriptive passages in the set). When a message-set was expanded while its overall quality was reduced (by including more and more low-quality descriptions), performance deteriorated. Finally, there was no convincing evidence of information overload; despite substantial expansions in the size of the message-sets that ultimately included as many as 24 different passages, when the quality of the ensembles was held constant, performance did not decline. In a recent series of experiments, Manis and his colleagues (Manis & van Rooijen, 1973; Manis & Platt, 1975, 1976) explored the process of information integration. In contrast to many studies in this domain, these experiments included a criterion of correctness, which made it possible to assess the e~rectiveness with which discrete informational cues were combined. In this work (and in the experiments described below), the respondents task was to decode a number of message-ensembles, each of which contained one or more written descriptions; our goal was to explore the performance levels (hit-rates) that would result when respondents were given several descriptions, all based on the same referent-object, and were asked to select the appropriate referent from a large array of similar stimuli. The procedure devised for these studies is an extension of the