Portfolio optimization with wavelet analysis and neural fuzzy networks

Abstract

Recently, Robo-advisors have come to the fore in the investment management business both in the world and in Turkey. Robo-advisors can be defined as financial algorithms that generate trading signals and optimize financial assets with machine learning. Robo-advisors, who perform asset management by analyzing data without human intervention, have the potential to provide better returns than traditional portfolio management. Assets under management in the Robo-Advisors segment are projected to reach US$3.2 tn in 2027. According to latest surveys, 63% of Americans are open to using a robo-advisor to manage their investments, with millennials being the most open (75%). Financial asset forecasts are of great importance in portfolio management and the performance of forecasts plays a key role in the success of portfolio managers. This situation has led to an increased interest in models. While most of the models were based on statistical techniques in the past, new modeling techniques have been used recently. The most notable of these are artificial intelligence models such as artificial neural networks and fuzzy logic. In this study, the daily values of Borsa İstanbul 30 Index, Gold and USD / TL exchange rate are tried to be estimated by using Wavelet Analysis and Neural Fuzzy Networks method. Buy / Sell signals are generated from the estimates created by the model. The performance of the portfolio was analyzed assuming that the underlying asset was invested on the days when the model predicted an increase and the investment was not made on the days when it predicted a decrease, and it was evaluated in overnight risk-free interest when not invested. In addition, the model has been tested in artificial indices and stock market indices of other developing countries. While the model showed a successful performance in Russia and China, it remained below the index in South Korea stock exchange. Then, the optimal portfolio was created by using wavelet fuzzy network model estimates and the performance of the portfolio was examined. With the return and standard deviation values produced by the model, optimization was made to obtain the largest Sharpe ratio, and the performance of the portfolio of three assets was compared with the assets' own performances and risk-free interest by re-balancing at different time intervals. The results show that the model created by using wavelet analysis and fuzzy neural networks together gives successful results in predicting the future values of financial assets and further research has potential. Wavelet Neural Network method, in which Artificial Neural Networks are used together with wavelet transform, can be used to predict the future price of assets traded in financial markets such as BIST30, Gold and USD/TL exchange rate. In the future, estimates can be made using the model in this study for financial assets other than gold, USD/TRL and BIST-30. The performance of the model in different market conditions can be tested by repeating the study at different time intervals. In the study, wavelet transform is done using Haar wavelet, financial series can be decomposed into its components by using different wavelets.