This document analyzes tropical fruit price data from Malaysia from 1990 to 1998 using ARIMA models to generate forecasts. Five ARIMA models were identified and their performance was evaluated based on diagnostic statistics. The ARIMA(1,1,1) model performed best according to measures like MSE, RMSE and MAPE on both the estimation and evaluation data. However, a simple Holt-Winter univariate model achieved even lower errors and is concluded to be the best model for forecasting future tropical fruit prices.
2. INTRODUCTION
*The data will helps us to forecast the price of tropical fruits
for the next period.
*Box Jenkins ARIMA modeling approach is followed (Harvey,
1993) to generate the forecast of the monthly price of
tropical fruits.
*The final models that used for forecasting are determined
by a number of diagnostic statistics including the Mean
Squared Error (MSE), Root Mean Squared Error (RMSE),
Akaike Information Criterion (AIC) and Bayesian Information
Criterion (BIC).
3. DESCRIPTION DATA
Focused on the topic tropical fruits in Malaysia
from January 1990 to December 1998.
It divided into fitted and hold out parts ( January
1990 until September 1996 is for estimation part
while October 1996 up to December 1998 is for
evaluation part)
12. CONCLUSION
Based on error measure, the univariate
model which is Holt-Winter is shown
the smallest error measure. For MSE is
0.19, RMSE IS 0.43 and MAPE is 100.93.
We can say that the univariate model is
the best model for forecasting the
future price of tropical fruits.