Modelling and Forecasting Portfolio Inflows: A Comparative Study of Support Vector Regression, Artificial Neural Networks, and Structural VAR Models

Background

In the literature, the effect(s) of strong wave of portfolio inflows are highlighted; under ordinary conditions capital flows have valuable impacts for developing economies. In a few events, floods of strong portfolio flows have gone before scenes of money related instability, for instance, the Mexican emergency of 1994 and the Asian emergency 1997 (Lo Duca, 2012). As this is the case, the negative effect of portfolio inflows to receiving economies calls for appropriate policies to be put in place, in which case the drivers of these flows may be used in developing these policies.

There are several applications of Support Vector Regression in solving forecasting problems in many fields where the model was successfully applied such as atmospheric science forecasting (Hong, 2009) and financial time series (stock index and exchange rate) forecasting (Cao, 2003). Chen and Wang (2007) employed Support Vector Regression, back-propagation neural networks (BPNN) and Autoregressive Integrated Moving Average (ARIMA) to forecast tourism demand and genetic algorithm was employed to select the optimal parameters of the Support Vector Regression model and show that Support Vector Regression outperforms other selected models. Hong (2009) also employed chaotic particle swarm optimization (CPSO) for choosing parameters for the Support Vector Regression model and showed that CPSO outperforms both the genetic algorithm (GA) and simulated annealing algorithm. Kazem et al. (2013) forecasted stock market prices employing a model based on Support Vector Regression, chaotic mapping and firefly algorithm using a time series data of stock prices, bank shares and intel. They compared their proposed model with Genetic Algorithm based Support Vector Regression (SVR-GA), Chaotic Genetic Algorithm based Support Vector Regression (SVR-CGA), Firefly based Support Vector Regression (SVR-FA), Artificial Neural Networks (ANNs) and Adaptive-Network-based Fuzzy Inference Systems (ANFIS), and revealed that the proposed model outperformed other models. Also, Adebiyi et al., (2014) compared artificial neural networks (ANNs) and Autoregressive Integrated Moving Average (ARIMA) models as far as anticipating precision of the stock market data sourced from New York Stock Exchange and disclosed that the Artificial neural networks (ANNs) model outperformed Autoregressive Integrated Moving Average (ARIMA) model.