A0779
Title: Identify the mean reverting properties and prediction of milk prices in EU countries
Authors: Yushu Li - University of Bergen (Norway) [presenting]
Bjorn-Gunnar Hansen - TINE SA (Norway)
Johan Lyhagen - Uppsala University (Sweden)
Abstract: The mean-reverting or long memory property of a time series can be captured by autoregressive fractionally integrated moving average models (ARFIMA). Previous studies have argued that time series with a structural break can easily be misidentified as a long memory process. The comprehensive empirical analysis is carried out based on monthly cow raw milk prices for 23 EU countries from Jan. 2003 to Nov. 2018 to investigate the structural change and long memory properties of the price process. Three hypotheses are proposed and are verified by rigorous statistical tests, model construction and estimation process. To estimate the fractional difference parameter before and after structural change adjustment, a novel technique is used based on a state-space model. The forecasting result is also carried out and compared for the countries whose milk price series are stationary and mean reverting before and after structural break adjustment. The result shows it is of essential importance to identify possible structural breaks and estimate the model based on the structural break-adjusted process. Several experts are also interviewed in the diary field and give explanations and interpretations of the estimation result. The findings have consequences for analysis and can offer certain guidelines for predicting raw milk prices in the countries included in the study.
