FORECASTING FARM-GATE CATFISH PRICES IN UGANDA USING SARIMA MODEL

Over the last ten years, the aquaculture subsector has emerged as a vibrant sector in Uganda and is being considered as a strategic sub-sector for promoting agricultural diversification. With improved market prices, government intervention for increased production, and stagnating supply from capture fisheries, aquaculture has attracted entrepreneur farmers seeking to exploit the business opportunity provided by the prevailing demand. As production has adjusted over time so have prices. Farm-gate prices have been highly volatile and are potentially a major factor in explaining why catfish farmers and processors are operating at narrow net margins. Price fluctuations translate into significant price risk, since the magnitude and the direction of the month-to-month changes are often unknown to producers.

The farmer has to frequently assess whether to harvest now to capture a known price, or to continue to feed to deliver a larger catfish at an unknown future price. Thus, the knowledge of future prices and factors influencing prices would be helpful to fish producers in decision making. It is against this background that a seasonal ARIMA forecasting model was developed to improve the prediction of catfish prices in Uganda.

Monthly farm-gate prices for African catfish covering the period 2006-2013 were obtained from Aquaculture Management Consultants in Uganda. Catfish prices, expressed in Uganda Shillings per kilogram, were deflated using a consumer price index. Stationarity of the series was examined using the ADF and PP test statistics. The optimal number of lags was determined using the Schwarz information criteria. Investigation was also done by examining the ACF and PACF functions. Using the AIC and BIC, six tentative SARIMA models were tested and the best model SARIMA (1,1,1) (0, 1,1)12 was selected by picking the model with the least values.

The estimates parameters are 0.6532 for the non-seasonal AR term and 0.9012 and 0.8095 for the non-seasonal and seasonal MA terms, respectively (Table 1). Based on 95% confidence level, we conclude that all estimated coefficients are significantly different from zero, and thus the model is ideal for forecasting catfish prices.

The forecasted in-sample results are reported in Table 2. It can be noticed from the results that forecasted catfish real prices are close to their actual values. Moreover, all the actual prices fall within the 95% Confidence Intervals of the forecasts, which further confirmed the reliability of the fitted model.