Title: Exploring the usage of econometric techniques in nonlinear machine learning and data mining

Authors: P. Lakshmi; S. Visalakshmi

Addresses: Department of Management Studies, National Institute of Technology, Tiruchirappalli 620015, Tamil Nadu, India ' VIT Business School, VIT University, Vellore 632 014, Tamil Nadu, India

Abstract: The present study, investigates the inter-linkage of the Indian spot market with other global markets and the predictability of S&P CNX NIFTY Index returns with a set of five new international market returns as input variables in artificial neural networks (ANNs). Identifying the right set of exogenous input variables using conventional techniques like OLS, Granger causality and cross correlation substantially increased the predictability of financial time series like stock return in the Indian context. The performance of the ANN model in forecasting NIFTY index returns is evaluated by comparing it for different sample periods in terms of forecasting error functions with statistical measures like mean absolute error, root mean square error, mean absolute percentage error and mean square error. The findings suggest that higher accuracy of the predictive power of neural network is largely influenced by the input variables.

Keywords: cross correlation; Granger causality; ordinary least squares; OLS; artificial neural networks; ANNs; NIFTY; NASDAQ; KOSPI; KLSE; SENSEX; SHANGAI; econometrics; nonlinear machine learning; data mining; India; spot markets; market returns; forecasting.

DOI: 10.1504/IJMOR.2016.078825

International Journal of Mathematics in Operational Research, 2016 Vol.9 No.3, pp.349 - 362

Received: 11 Oct 2014
Accepted: 17 Dec 2014

Published online: 03 Sep 2016 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article