The role of tweets in agricultural export: an approach from text-mining and time-series analyses

Abstract

PurposeThis study aims to investigate the relationship between consumer awareness of Zespri International Limited (Zespri™) and its sales promotion in Japan and the recent expansion of New Zealand (NZ) kiwifruit imported into Japan.Design/methodology/approachTweets mentioning Zespri™ were utilised as a proxy of such awareness. They were first summarised using two text-mining techniques: tf-idf scoring and a co-occurrence network graph. Afterwards, the authors estimated a tri-variate vector autoregression (VAR) model consisting of the net imports of NZ kiwifruit in Japan, unit import price and number of tweets. Additionally, the occurrence frequency of tweets with “Kiwi Brothers”, promotional characters for Zespri™’s sales, was added to the model, and a tetra-variate VAR model was estimated. Finally, Granger-causality tests, an estimation of the impulse response function and forecast error variance decomposition was conducted.FindingsAll these variables were found to Granger-cause each other. Furthermore, a shock in the document frequency of “Kiwi Brothers” significantly affected Japan’s kiwifruit imports from NZ, explaining approximately 20% of future imports. Zespri™’s distinctive sales promotion was, thus, found to contribute in part to the recent increase in NZ’s kiwifruit export to Japan.Originality/valueThis paper is the first to apply text-regression methodology to food consumption research; it contributes to food consumption research by proposing a practical way to combine tweets with outcome variables using a time-series analysis.