The Effects of Temporal Data Aggregation on Price Transmission Analysis

Agricultural economists often use temporally aggregated data for price transmission analyses due to data availability but the used vector error correction models (VECM) indirectly assume that all price information is included in the analysis. In consequence, internal dependency structures are not captured correctly, and estimated parameter values can be biased. The model with temporally aggregated data requires a moving average dynamic as a remedy because the error terms are autocorrelated. We show how the correct aggregated parameter values can be derived. With Monte Carlo experiments, we compare the true parameter values with estimated ones. For the estimations, we use maximum likelihood method by Johansen and Juselius for VECM and the method by Yap and Reinsel for error correction vector autoregressive moving average (ECVARMA) models. To demonstrate that temporal aggregation matters in real world data, we perform a price transmission analysis for the French and German hog market. We ourselves aggregate the data and observe a moving average dynamic in the aggregated data sets. Furthermore, we see an improvement in biasness and efficiency for the adjustment parameter with the EC-VARMA models. We conclude that aggregated data need moving average parameters for a correct estimation, but EC-VARMA models are challenging to specify and to estimate, especially in small data sets due to the identification problem.