Development of a Model for Predicting CEO Compensation Sensitivity Using Metaheuristic Algorithms: Genetic Algorithm and Particle Swarm Optimization

The aim of this study is to propose a model for predicting CEO compensation sensitivity by employing metaheuristic algorithms, including the genetic algorithm and the particle swarm optimization algorithm. The statistical population of this study consists of all companies listed on the Tehran Stock Exchange during the period from 2011 to 2021. A systematic elimination method was used for sample selection, resulting in a final sample of 110 companies. This research is classified as applied research in terms of its objective and as quasi-experimental in terms of its nature and methodology. Furthermore, it falls within the category of descriptive research of a non-experimental survey type. The required data were collected through document analysis, internet searches, and library studies. In this study, 12 influential variables on CEO compensation sensitivity were selected as input variables for the data mining model. These variables include institutional ownership, family ownership, financial statement comparability, earnings management, conditional conservatism, revenue-expense matching, market value added, corporate acquisition, debt contracts, and cost behavior with three indicators (changes in return on assets, changes in sales revenue, and changes in operating costs). Additionally, CEO compensation sensitivity was considered as the output variable of the data mining model. To analyze the data, three data mining models based on cost behavior parameters were designed, and for comparison purposes, three linear regression models were also utilized. Among the 12 examined parameters, seven variables, including institutional ownership (X1), financial statement comparability (X3), revenue-expense matching (X6), market value added (X7), changes in return on assets (X101), changes in sales revenue (X102), and changes in operating costs (X103), demonstrated a significant relationship with CEO compensation sensitivity. Accordingly, these parameters were selected as input variables for the data mining model. The analysis results indicated that the deep neural network model optimized with the particle swarm optimization algorithm recorded the lowest mean squared error (MSE) of 0.0458 and the highest coefficient of determination (0.9853), highlighting its superior performance compared to other examined methods. The deep neural network model optimized with the genetic algorithm ranked second in predictive performance. Ultimately, the findings demonstrate that the deep neural network model outperforms the linear regression model in terms of the coefficient of determination and error index (MSE).