Abstract:

A factor that directly affects investment decisions made on capital markets, the distribution of funds provides clues about the future performance of the enterprise as much as it shows its past performance. In this study in Turkey, the prediction of the supplies distributed by public enterprises in Turkey is to test the availability of the Marsh&M (M&M) model and apply machine learning techniques, aimed at developing a better model than M&M. In the study, the estimate of the cash-distributed asset ratio of 139 enterprises in the manufacturing sector operated in Borsa Istanbul (BIST) using the data between 2003-2012, the M&M model and the multi-layer detector (One and Two Hidden Layer CKA), Radial-Based Function Networks (RTFA), Support Vector Machines (DVM) and Adaptive Nervous Bullying Extrusion Systems (ASBÇS) compared five different models. The study found that the ASBÇS model was a successful machine learning method for certain tolerance values. In general, ASBÇS and RTFA models have shown better performance than M&M models, CKA models have shown results close to M&M models, and DVM models have shown worse results than M&M models.

Keywords:

Abstract:

Dividend payment is a factor that affects investment decisions in capital markets. Although dividend payments indicate past performance of corporate, they also give some clues about company’s future performance. In this study, feasibility of Marsh&Merton (M&M) model is tested in Turkey tried to develop a better model than M&M model by applying machine learning techniques. For this study payout ratios between 2003 and 2012 from 139 manufacturing companies which are quoted on ISE are selected. M&M model and five machine learning models namely Multi-Layer Perception (MLP), Radial Based Function Networks (RBFN), Support Vector Machines (SVM) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) are compared with each other. Generally it is occurred that RBFN models produces similar results with M&M model, MLP models cannot forecast low paid dividend and SVM model executes worse than M&M model. As a result ANFIS model is observed the most successful method in forecasting dividends.