Article Preview
TopIntroduction
India ranks second in tea production, after China, and produces about one fourth of the world’s total production (ONICRA, 2014). India is ranked fourth based on export of tea (Das & Zirmire, 2018). Kenya, China, and Sri Lanka are ahead of India in tea export. India exports nearly 30 percent of tea produced in the country and holds nearly 12 per cent share of the world’s tea export. Tea export holds significance in India’s international trade for the following reasons, first, due to its potential of earning valuable foreign exchequer for India, and secondly, poor export performance would exert downward pressure on tea price in the domestic market. Therefore, proper forecasting of export for Indian tea planters and traders is essential to plan its production as well as inventory holding.
The crop-specific sensitivities to local climatic variability and progresses in technology could complicate the imputation of production variation to climate change (Adhikari et al., 2015; Lehner et al., 2016). Both high temperature and excess rainfall carry adverse impact on agricultural production in general (Poudel & Kotani, 2013; SreeVidhya & Elango, 2019). Tea is a perennial crop whose life may be divided in two broad phases - development phase and productive phase. During the development phase, the planter has to nurture the bushes till it reaches its early productive phase when one can pluck the leaves and twigs. The development phase may be as long as five to six years. Once the bushes have attained the productive phase output is realized in four flushes between March to November. The first flush is picked during February end to mid-April, second flush during May-June, followed by a rain flush over July-August and lastly, an autumn flush spanning over September to November. Therefore, rainfall seems to be a critical input in tea production as well as its export.
The effect of rainfall on tea production is well documented in the existing literature. As for example, Dutta et al. (2011), analysing the tea productivity in Northeast India, the major tea producer belt in the country, observe that monthly rainfall carry a significant positive impact of tea productivity. Waheed et al. (2013), working on the data from Shinkiari, Pakistan, also document a significant effect of precipitation on tea production. However, the attempt to understand the impact of rainfall on tea export is relatively scanty. Econometric investigations (see, Chatterjee, 2011; Gunathilaka & Tularam, 2016 for review) undertaken, thus far, to identify the factors affecting India’s tea export have summarily considered the relative international price as well as exchange rate, as regressors. Against the conventional time series approaches, namely cointegration and vector error correction models, employed in earlier research, Pal and Mitra (2015) have modelled asymmetric response of India’s tea export to international price realization through Quantile Autoregressive Distributed lag (QADL) suggested by Galvao et al. (2013). Existing literature has also examined India’s market power in international tea trading (Veeramani, 2012; Suresh & Mathur, 2016), the effect of the food safety standard on tea export from China (e.g., Wei et al, 2012) and fair trade in linking South African tea producers with consumers in the American markets (Raynolds and Ngcwangu, 2010; Lama, 2019). None of these studies have attempted to elucidate the impact of rainfall on tea export from India or from any other major tea producing nations. Further, the models employed to identify factors affecting tea export were observed to have relatively low predictive power. For example, the gravity model extended by Wei et al (2012) could explain only 32 per cent of the variability in the tea export. We attempt to fill this gap in the existing literature. Our objective is to forecast India’s tea export by employing new generation forecasting models of data mining techniques where we aim first, to segregate India’s monthly tea export data series into homogenous clusters based on certain self-similarity and second, to use these clusters for building predictive models based on the support vector machine.