Wharton Research Data Services Research Articles

Relationship between CEO Pay Slice and Firm Performance

This study investigates the relationship between the CEO Pay Slice (CPS) and firm performance in North America. The CEO Pay Slice (CPS) is defined as the proportion of the top five executives’ total compensation that is captured by CEO. This measure reflects the importance of the CEO as well as the extent to which the CEO is able to extract rents. In addition, CPS might also have an impact on the effectiveness of the board performance through its influence on cooperation among its members. Using data for firms in the North America from Wharton Research Data Services over 1992 to 2016 period, we find a negative relationship between CPS and firm performance as measured by Tobin’s Q. This negative relationship between CPS and firm performance is further verified by using Returns on Assets (ROA) rather than Tobin’s Q as firm performance measures. Our results are consistent with the view that pay disparity has a negative impact on firm performance by discouraging employees from making efforts, and therefore, high CPS is generally associated with bad firms’ performance. This also indicates that CPS can be a potential useful tool for studying the performance of firms.

Data Matching for Corporate Governance Research

Matching observations across multiple datasets for corporate governance research carries with it some well-known and other less known challenges. This paper highlights best practices for carrying out merging the widely used BoardEx, CRSP and Compustat databases accessed via Wharton Research Data Services (WRDS), utilizing their respective company and security identifiers. A companion Github repository of the associated SAS code is made available to assist researchers in consistently applying the discussed matching techniques.

The Impact of the Dodd-Frank Act on Executive Compensation

Congress passed the Dodd-Frank Wall Street Reform and Consumer Protection Act (Dodd-Frank Act) on July 21, 2010, which became effective for shareholder meetings held on, or after January 21, 2011. The Dodd-Frank Act requires U.S. public companies to conduct an initial advisory, non-binding vote on executive compensation called say-on-pay and a subsequent vote on the frequency of say-on-pay. This study investigates the voting pattern of the say-on-pay and the say-on-frequency and examines the impact of the Dodd-Frank Act on the amount and structure of executive compensation. The say-on-pay and say-on-frequency patterns of over 1,700 companies from the ExecuComp database are investigated using the U.S. Securities and Exchange Commission Forms DEF 14A, 8-K, and 10-Q. A further study of S&P 500 companies, from 1992 – 2009, found in the Wharton Research Data Services is used in the regression analysis to forecast executive compensation. Results show that 98% of the sample firms are supported by a majority of shareholders who voted on executive compensation, while 2% of the sample firms failed the say-on-pay. Preliminary results show that 60% of companies initially recommended every three years as the preferred say-on-frequency, while the shareholders voted in favor of every one year (90% of the companies). However, company recommendations for say-on-frequency have grown from 60% in favor of every three years in March 2011 to 63% in favor of every one year in December 2011, suggesting that companies are responding to shareholder pressure. Further results indicate that executive total cash compensation (salary plus bonus) has been decreasing since 2006. Additional analyses, comparing 2010 to 2009 total cash compensation, show that firms reduced total cash compensation in 2010 prior to the advisory vote on executive compensation.

Failure prediction of dotcom companies using neural network–genetic programming hybrids

This paper presents novel neural network–genetic programming hybrids to predict the failure of dotcom companies. These hybrids comprise multilayer feed forward neural network (MLFF), probabilistic neural network (PNN), rough sets (RS) and genetic programming (GP) in a two-phase architecture. In each hybrid, one technique is used to perform feature selection in the first phase and another one is used as a classifier in the second phase. Further t-statistic and f-statistic are also used separately for feature selection in the first phase. In each of these cases, top 10 features are selected and fed to the classifier. Also, the NN–GP hybrids are compared with MLFF, PNN and GP in their stand-alone mode without feature selection. The dataset analyzed here is collected from Wharton Research Data Services (WRDS). It consists of 240 dotcom companies of which 120 are failed and 120 are healthy. Ten-fold cross-validation is performed throughout the study. Results in terms of average accuracy, average sensitivity, average specificity and area under the receiver operating characteristic curve (AUC) indicate that the GP outperformed all the techniques with or without feature selection. The superiority of GP–GP is demonstrated by t-test at 10% level of significance. Furthermore, the results are much better than those reported in previous studies on the same dataset.

Failure prediction of dotcom companies using hybrid intelligent techniques

This paper presents a novel hybrid intelligent system in the framework of soft computing to predict the failure of dotcom companies. The hybrid intelligent system comprises the techniques such as a Multilayer Perceptrons (MLP), Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), Classification and Regression Trees (CART). The dataset collected from Wharton Research Data Services (WRDS) consists of 240 dotcom companies (also known as click-and-mortar companies), of which 120 are failed and 120 are healthy. Ten-fold cross validation is performed on the data set for all the techniques considered in their stand-alone mode. Further, two hybrid techniques viz., ensembling and boosting are employed to improve the accuracies. Moreover, t-statistic is performed on the dataset for feature selection purpose and the reduced feature subset with 10 features is extracted. The reduced feature subset is tested with all the techniques and then ensembling and boosting is also done for the reduced feature subset. Results supported by Receiver Operating Characteristic (ROC) curve indicate that the important features extracted by the t-statistic based feature subset selection yielded very high accuracies for all the techniques. Furthermore, the results are superior to those reported in previous studies on the same data set.