Article Preview
TopIntroduction
The mechanism of vegetable trade is very different as compared to the trade of other lesser perishable agricultural commodities. In case of vegetables, the consolidator collects (purchases) the vegetables from the farmers and brings it to the wholesales market, which follows the system of spot auction. The commission agents in the wholesales vegetable market participate in the auction on behalf of consolidators and retailers. These auctions happen only for few hours (e.g. 6am-11am). The consolidators are bounded to sell the vegetables by the end of the auctions, as there is almost no cold storage available and waiting for another day will incur extra cost. The perishable nature of the vegetables and lack of the infrastructure facilities can be attributed to the low bargaining power of the consolidator. This in turn results in low bargaining power and lesser profits to the farmers. One of the major reasons to this is the lack of demand visibility at the customer’s end. Generally the farmers are unaware of the market trends and follow the traditional product mix (variety and volume of produce to be planted). This scenario results in a push rather than a pull system, and thus, a mismatch between demand and supply. This causes either waste of excess produce or unsatisfied customers.
Vegetables are a seasonal produce and there is a lead time between the demand and supply. The harvesting decisions are based on experience or speculation rather than market demand. This is due to the lack of an accepted forecasting model. Due to these conditions the farmers are forced to sell the vegetables to the consolidators at a very low price. This is more so in the developing countries such as, India where the major percentage (~98%) of the vegetables are sold in the spot markets. Hence, there exists a need to forecast the demand of vegetables in the wholesales market to avoid wastage of vegetables and to increase the profits of all the stake holders.
There are very few papers that have studied the demand forecasting of vegetables in the wholesales market. Among these, most of the studies are either focusing on price forecasting or forecasting the demand on an aggregate level. Researchers have generally considered all the vegetables as a single commodity (e.g. Mutuc, Pan, & Rejesus, 2007) and have tried to forecast the demand of vegetables. But there is a need of forecasting the demand of an individual vegetable such as onion, potato, tomato. It is also found that the papers are generally forecasting the demand on weekly or monthly level (e.g. Zou, Xia, Yang, & Wang, 2007). But, in real life situation the farmers may need the daily demand to take their harvesting decisions due to the short selling horizon and perishable nature of the vegetables. Adding to it is the high price fluctuations and availability of substitute produce that further increases the uncertainty of forecasted demand of an individual produce on a daily basis. However, there exist no papers forecasting the demand of an individual fresh produce on a daily basis.
In real life situation, researchers and practitioners prefer models that are easy to operate and have a user friendly interface to operate in terms of set-up and data requirements (Bucklin & Gupta, 1999; Ali, Sayın, van Woensel, & Fransoo, 2009). There exist a large number of papers studying time series forecasting and comparing the results obtained from different forecasting techniques. It is found that AutoRegressive Integrated Moving Average (ARIMA) models are more preferred in literature for short term forecasting as compared to artificial intelligent models (Co & Boosarawongse, 2007; Zou, Xia, Yang, & Wang, 2007). The comparison shows that artificial intelligent models such as Artificial Neural Networks (ANN) performed almost equal but not better than ARIMA models (Church & Curram, 1996; Ntungo & Boyd, 1998). On the other hand, ARIMA models performed better than ANN models in short-term forecasting (Maier & Dandy, 1996; Kirby, Watson & Dougherty, 1997; Darbellay & Slama, 2000). Exponential smoothing models are a special case of ARIMA models (McKenzie, 1984). Box and Jenkins (1970) first introduced the ARIMA model building methodology. These models are highly accepted and applied due to the statistical properties (Zou, Xia, Yang, & Wang, 2007). The steps of ARIMA model building methodology is presented in a flow chart in Figure 1.