Profitability estimation with artificial neural networks and comparison with panel data analysis: The case of real estate investment trusts

Making profitability predictions by fund owners to make investment decisions, and by firms to make strategic decisions regarding the financing of their investments, is highly important. In addition, predicting profitability in practice can be challenging and requires expertise. Profitability estimates made with traditional econometric models may face limiting factors. In recent years, artificial intelligence techniques have come to the fore for profitability estimation with their flexible calculation, ability to work with large and different data types, and ability to predict instant changes. The main purpose of this study is to determine the best profitability prediction model using Artificial Neural Networks (ANN) algorithms, one of the artificial intelligence techniques. Additionally, the findings obtained from the ANN prediction model are compared with the findings obtained from applying panel data analysis to the dataset, allowing for an interpretation of the prediction success of the ANN technique. The scope of the study includes the quarterly financial statement values of Real Estate Investment Trusts (REITs) traded on the Borsa Istanbul (BIST) between 2010 and 2019. The best R^2 overall performance values obtained from numerous trial models run with ANN for the dependent variables considered in profitability prediction, namely asset profitability and equity profitability, are respectively obtained as 87% and 98%. As a result of comparing these values with the findings of panel data analysis, it is concluded that the ANN prediction models are able to make more successful predictions compared to panel data analysis models.