Handling data gaps in Maltese balance sheet data compilation

Abstract

Missing data is a major issue that may jeopardize the quality of the results achieved in a study. In particular, it is an inevitable problem in almost all studies involving longitudinal data. Linear mixed models (LMMs) have superseded the classical repeated measures ANOVA, since the former models are better equipped to handle both repeated type of measurements and missing data. In this dissertation, we study the theoretical framework of LMMs with special focus directed to the Restricted maximum likelihood (REML) estimation technique. This technique is used to estimate the parameters of the LMM as it is known to produce unbiased estimates for variance components of the LMM unlike Maximum likelihood (ML) estimation. A number of LMMs are applied to a real life longitudinal data set obtained from the Central Bank of Malta (CBM). This data set also contains missing data. In this study we demonstrate how LMMs can be fitted to model change in the mean level of debts and equity over time for specific groups of companies in the economy and consequently explain why certain companies within the economy have higher amounts of debts and equity than others. Moreover, we compare the predictive accuracy of LMMs with the last observation carried forward (LOCF) method that is in current use by the bank to handle such data.