APPLICATION OF VECTOR AUTOREGRESSIVE MODELS AND MACHINE LEARNING TECHNIQUES TO ESTIMATE THE EFFECT OF MACROECONOMIC INDICATORS ON PORTFOLIO INVESTMENT IN NIGERIA: A DATA ANALYTICS APPROACH

Abstract

Portfolio investments in emerging markets such as Nigeria have been very volatile although the policy has been used to stabilise capital markets. The intricate interaction between the macroeconomic indicators and the portfolio investment is of key importance in the sustainable economic development and policy making, which require the understanding of the data. This paper uses sophisticated econometric and machine learning models to examine how portfolio investment and shares in Nigeria are affected by the macroeconomic variables (interest rate, exchange rate, and inflation rate) between 1981 and 2019 and compare the results of using traditional VAR/VECM model and modern machine learning techniques. The sources used to obtain data were the Central Bank of Nigeria (CBN) and World Development Indicators (WDI). Our hybrid analytical model adopted a combination of: (1) Inferential analysis, (2) Vector Autoregressive (VAR) and Vector Error Correction (VECM) models, (3) Stationarity check, (4) long-run relationship, (5) Johansen cointegration test, and (6) comparative predictive performance by choosing: (a) Random Forest, (b) XGBoost, and (c) Long Short-Memory (LSTM) networks. Akaike Information Criterion (AIC), Log-likelihood, RMSE and MAE were used as the measures of model performance. The cointegration analysis showed that there is significant long-run relation between all variables (Trace test statistic = 155.6993, p < 0.0001). The results of the VECM indicated that the interest rate has a negative significance impact on portfolio investment (p = 0.0225, β = -172.571), whereas the inflation rate has a positive significant impact (p = 0.0429, β = 250,988). The Log-likelihood was equal to -4706.210 and AIC = 61.78053. The predictive accuracy of machine learning models was better and LSTM had the lowest RMSE (0.1243) than traditional VAR (0.2876). Indicators of the macroeconomic show that they have a great long-run connection with the portfolio investment in Nigeria. The hybrid econometric machine learning model offers greater predictive power in the portfolio investment forecasting. We suggest the Central Bank of Nigeria to adopt dynamic interest rate levels and come up with data-driven monetary policy levels to enhance sustainable portfolio investments.