Fuzzy panel data analysis

Abstract

In statistical and econometric researches, three types of data are mostly used as cross-section, time series and panel data. Cross-section data are obtained by collecting the observations related to the same variables of many units at constant time. Time series data are data type consisted of observations measured at successive time points for single unit. Sometimes, the number of observations in cross-sectional or time series data is insufficient for carrying out the statistical or econometric analysis. In that cases, panel data obtained by combining cross-section and time series data are often used. Panel data analysis (PDA) has some advantages such as increasing the number of observations and freedom degree, decreasing of multicollinearity, and obtaining more efficient and consistent predictions results with more data information. However, PDA requires to satisfy some statistical assumptions such as “heteroscedasticity”, “autocorrelation”, “correlation between units”, and “stationarity”. It is too difficult to hold these assumptions in real-time applications. In this study, fuzzy panel data analysis (FPDA) is proposed in order to overcome these drawbacks of PDA. FPDA is based on predicting the parameters of panel data regression as triangular fuzzy number. In order to validate the performance of efficiency of FPDA, FPDA, and PDA are applied to panel data consisted of gross domestic production data from five country groups between the years of 2005-2013 and the prediction performances of them are compared by using three criteria such mean absolute percentage error, root mean square error, and variance accounted for. All analyses are performed in R 3.5.2. As a result of analysis, it is observed that FPDA is an efficient and practical method, especially in case required statistical assumptions are not satisfied.