Semiparametric Spatiotemporal Model with Mixed Frequencies

TL;DR

This paper introduces a semiparametric spatiotemporal model that effectively handles mixed-frequency data from different sources, improving prediction accuracy over traditional models in agricultural applications.

Contribution

It develops a novel semiparametric model utilizing backfitting for mixed-frequency spatiotemporal data, enhancing information use from high-frequency variables.

Findings

Model outperforms simple GAM with aggregated predictors.

Simulation studies confirm optimality of the proposed approach.

Better predictive accuracy with remotely-sensed data in agriculture.

Abstract

In modelling time series data coming from different sources, frequencies can easily vary since some variable can be measured at higher frequencies, others, at lower frequencies. Given data measured over spatial units and at varying frequencies, we postulated a semiparametric spatiotemporal model. This optimizes utilization of information from variables measured at higher frequency by estimating its nonparametric effect on the response through the backfitting algorithm in and additive modelling framework. Simulation studies support the optimality of the model over simple generalized additive model with aggregation of high frequency predictors to match the dependent variable measured at lower frequency. With quarterly corn production and the dependent variable, the model is fitted with predictors coming from remotely-sensed data (vegetation and precipitation indices), predictive ability is better compared to two benchmark generalized additive models.