Estimation of Missing Farmgate Prices Using Regression Imputation
Author: Marlene S. Ymana
Of the data available for computing agricultural indices, the use of farmgate prices is deemed most practical as basis for evidenced-based policies on agricultural programs, and agricultural production processes. However, missing observations occur in the farmgate price data series given the availability of data depends only on the unreliable timeline of when producers dispose their produce. This study then aims to estimate missing farmgate prices using regression models with autocorrelated errors. Datasets on the monthly average price quotations from the Farmgate Prices Survey (FPS) and the wholesale price monitoring of the Bureau of Agricultural Statistics (BAS) are used. Two cases are used in estimating farmgate prices: (Case 1) estimation using highly correlated series (farmgate prices vis-à-vis wholesale prices of palay; and ( Case 2) estimation using less correlated series (retail prices of bangus vis-à-vis retail prices of pork). Further, simulations are used to assess the procedure employed in the study. Results showed that satisfactory estimates for missing values from regression models with autocorrelated are obtained when the related series used in estimating observations behaved very similarly to that of the incomplete series (Case 1). When a related series is not available (Case 2), an alternative procedure in estimating missing observations is the use of time series forecasting or backcasting techniques.