FORECASTING DEFERRED TAXES IN INTERNATIONAL ACCOUNTING WITH MACHINE LEARNING

Abstract

The aim of this study is to estimate the possible deferred tax values and the TAS-TFRS profit/loss of 31 companies in three different sectors- the wholesale trade, retail trade and hospitality industry- whose shares are traded on Borsa Istanbul (BIST). This estimation is based on the companies' deferred tax values for the years 2015-2019 as well as twelve main economic parameters. Within the context of the study, the deferred tax output parameters, which companies will present in their annual financial reports in 2020, have been estimated using the following methods: the DTA value using the random forest method with an accuracy rate of 0,823, the net DTA value using the artificial neural networks method with an accuracy rate of 0,790, the DTL value using the random forest method with an accuracy rate of 0,823 and the net DTL value using the random forest method with an accuracy rate of 0,887. In addition, it has been discovered that the TAS-TFRS profit/loss, which is one of the output parameters, can be estimated using the random forest method with an accuracy rate of 0,629.