Comparative Analysis on Neural Networks and ARIMA for Forecasting Heterogeneous Portfolio Returns

In the domain of portfolio management theory, the principle of separation postulates that all investors will achieve an identical optimal risk portfolio given the same inputs. However, the realization of true optimality depends on the accuracy of technical analysis, conducted by portfolio managers or investors in predicting the rate of return for the financial assets, encompassed in the portfolio. This article leverages the Nonlinear AutoRegressive with Exogenous inputs Neural Network (NARXNN) for the purpose of predicting the prices of ten financial instruments, facilitating the computation of their respective rates of return. Subsequently, a multi-objective optimization problem is formulated to construct an optimal risk portfolio that concurrently maximizes return while minimizing risk. The resulting portfolio is then compared with a similar portfolio derived from the same dataset, utilizing Autoregressive Integrated Moving Average (ARIMA) models to forecast the rates of return for the assets.