DM 검정을 이용한 냉동 고등어 소매가격 예측력 비교

Abstract

Forecasting has a key role in economics and especially in fishing management policies. Thus, this study firstly aims to estimate which model is superior to predict frozen mackerel prices using statistical approaches, and secondly aims to provide policy implications of frozen retail mackerel market. The study used daily frozen mackerel prices of the retail market from 2006 to 2015, the prices were estimated and predicted by ARMA model, GARCH model and VAR model. The two ARMA models are selected according to the information criteria standard, and one of them is estimated as a GARCH model because it does not satisfy homoskedasticity assumption. Also, VAR model is estimated to include other related variables such as frozen wholesale, fresh wholesale and retail prices. In addition, the predictive accuracies of each models are tested through Diebold & Mariano test (1995) by comparing actual prices and predicted prices for the next year. The summary of the study is as follows. Firstly, ARMA(4,3), ARMA(2,1) models are selected by AIC, SC and HQ information criteria, and ARMA(2,1) model is estimated as GARCH(1,1). Secondly, the restricted VAR model of selected t-2 lag is estimated to forecast the frozen mackerel retail prices. Thirdly these models are used to compare prediction accuracy among models. According to prediction accuracy test, ARMA(2,1)-GARCH(1,1) model has the most excellent accuracy by MSE and MAE standard, respectively. The study found that this model is superior to predict the retail mackerel prices in a conservative manner. The study is significant because it estimated forecasting models of frozen mackerel, and compared prediction accuracy using statistic approaches in the retail market. Besides, the study is not only useful for the purchasing and storing projects by government but also for fisheries companies and retailers.