COMPARATIVE ANALYSIS OF QUALITY TO FORECAST MODELS OF EXCHANGE TRADED FUNDS DYNAMICS

Анотація

The article is devoted to the comparative analysis to forecast models of exchange-traded investment funds dynamics. The exchange traded investment fund was chosen for the study as a modern investment instrument, which managed to combine the best features of stocks and mutual investment funds. However, despite the advantages of this financial instrument, the main task of the investor is the ability to predict its dynamics. To date, there are many forecast models, but the ability to create a reliable and accurate forecast remains extremely important.Two exchange traded funds were selected for forecasting: SPDR S&P 500 ETF TRUST (the stock ticker is SPY) and VanEck Vectors Gold Miners (the stock ticker is GDX). Based on daily prices for the period from January 2016 to January 2020, forecast models of two types were built: linear and nonlinear. Namely, linear models of moving average and exponential smoothing (Holt model) were selected. The neural network model was chosen as a nonlinear model. It turned out that the quality assessment of all models is quite high. However, the constructed forecasts showed that despite the high quality of the obtained statistical models of moving average and exponential smoothing, forecasting with their help is possible only for the forecast horizon, which is one trading day. The neural network model, conversely, shows a worse forecast for the first forecast value, but captures the dynamics and direction of price changes in both the exchange-traded investment fund SPY and GDX. That is, with the help of training, the neural network is able to establish hidden nonlinear patterns of price dynamics. But the horizon of the neural network forecast is also limited: in research it is established that the forecast on the basis of a neural network model it is expedient to build no more than for one exchange week.

Класифікація

Ідентифікатори

Схожі роботи