Terms of Trade and Investment Behaviour in Indian Agriculture: A Cointegration Analysis

Abstract

The issue of capital formation in Indian agriculture has generated substantial research interest in recent years (Mishra and Chand, 1995; Misra and Hazell, 1996; Dhawan, 1996 a, b; Alagh, 1997; Rao, 1997; Gulati and Bathla, 2001). The main focus of this literature has been on three aspects, viz., (i) the declining trend in public investment since the early 1980s; (ii) the complementary relationship between the public and private sector investment; and (iii) the positive influence of domestic terms of trade on private investment levels in agriculture. In a recent study, Chand (2001) has indicated that both the private and public investment series in India are non-stationary sequences, i.e., they have a unit root in their respective data generating processes. The distinction between deterministic and stochastic trend (unit root) models has considerable bearing for understanding the time-series behaviour of a variable (Nelson and Plosser, 1982; Gil-Alana and Robinson, 1997; Murray and Nelson, 2000). If the private and public agricultural investment variables in India are truly characterised by unit root processes, it implies that as such there do not exist any long-run trend values of these variables. Since regular shocks in the economy can potentially send the variables off on a wholly different path for the rest of the time, there may be a tendency for the variables not to return to their respective long-run trend, and instead drift apart over time. The presence or absence of unit root in investment series also assumes importance in analysing its relationship with terms of trade (henceforth TOT) in agriculture. So far, the underlying data generating processes of both the investment and TOT series have been assumed to be stationary (absence of a unit root). However, in case this assumption is not valid, standard asymptotic distribution theory cannot be used for the purpose of drawing inference. Because traditional regression analysis in models that include variables with unit root can produce spurious regression results (Granger and Newbold, 1986; Maddala, 1992; Greene, 1997).