Nonlinearity in Indian Stock & Commodity Markets: A Pre-Diagnostic Investigation

Abstract

While there are many evidences of nonlinearity in developed markets, there has not been many works in this direction in Indian financial markets. In this study we wish to bridge this gap by testing for nonlinearity in the Indian stock and commodity market. We consider the index movements in Indian financial markets. The stock indices under consideration are S&P CNX NIFTY, S&P CNX 500 based on National Stock Exchange, SENSEX, BSE 500 based on Bombay Stock Exchange, India from August 2000 to August 2009. The commodity indices under consideration are MCX-COMDEX, MCXENERGY, MCX-METAL, and MCX-AGRI indices based on trading data from the Multi Commodity exchange of India from June 2005 to August 2010. We take the time series representing the daily close value of the indices as our input value for the tests. We first use the test method developed by Brock, Dechert, and Scheinkman (BDS) and test for nonlinearity in each of the time series. Additionally we also perform the Keenan’s test for nonlinearity. Another popular non-linear test is the Hinich bispectrum test, which involves estimating the bispectrum of the observed time series. We also use this test to find out whether it detects nonlinearity in these tine series. To reinforce our findings we also conduct the White’s neural Network tests on the same data set. Another linearity test for time series was introduced based on concepts from the theory of neural networks. Terasvirta et al. developed its power fully. We use this Terasvirta Neural Network test as a final reinforcement of our findings. Our findings lead us to the conclusion that all the time series developed from data in the Indian Stock and Commodity Market exhibit significant nonlinearity. On one hand, the results highlight the fact that researchers cannot take the linear assumption as granted, especially dealing with financial market time series data. On the other hand, it points to the need to test for nonlinearity as a preliminary diagnostic tool to determine the nature of the data generating process before any further empirical analysis.