Abstract:
This study explored modelling and forecasting of wholesale green bean monthly prices at the
Pettah market using asymmetric Generalized Auto Regressive Conditional Heteroscedasticity
Model (GARCH). Firstly, outliers which occur due to natural disasters in the data set identified
using a boxplot and the stationary check of time series data was performed by using (ADF) test.
The (ACF) and PACF are used to identify the conditional mean model. Breusch-Godfrey serial
correlation LM test and Jarque – Bera test are used for residual diagnostic checking. (ARCH) LM
test checks whether the volatility clusters exist in the residuals and Bayesian Information Criteria
(BIC) is considered in model selection and comparison. Finally, MAPE value decides the adequacy
of the final model. The Coefficient of variation indicates green bean prices at the Pettah market are
highly volatile represented by value of 56.29 percent. Based on the minimum BIC values, MA (4)
was selected as conditional mean model of green bean prices. Ljung-Box test applied to detect the
autoregressive conditional heteroscedasticity in residuals of conditional mean model. Conditional
variance model GARCH (1, 1) adequately captured this effect. But, according to sign bias test,
GARCH (1, 1) failed to capture the asymmetric effect in the green bean prices. Finally, based on
the minimum BIC values, the asymmetric GARCH model MA (4) – EGARCH (1, 1) was selected as
the best fit model for forecasting the wholesale prices of green bean at the Pettah market. The
MAPE in the MA (4)-EGARCH (1, 1), based on long term out of sampling forecasting, is 15.82
percent. The MAPE of MA (4)-EGARCH (1, 1), based on short term out of sampling forecasting, is
12.17 percent. Therefore, this model is capable for capturing the volatility, the time-varying
conditional variance, and errors on the wholesale price of green beans at the Pettah market.
