Forecasting food trends using demographic pyramid, generational differentiation and SuperLearner

The objective of this paper is to predict food consumption patterns for future decades by different social groups while taking generational change into account in the modelling. Using over 20 million observations of households in Switzerland from 1990 to 2017, we develop and apply four forecasting techniques that shift from referenced linear forecasts to population-driven forecasts. Each method considers values of selected household characteristics to define a "social group", derives the proportion of each social group in society for the years 1990-2050, forecasts the future consumption of 75 food items in each social group in its unique way, and weighs these consumption patterns to obtain a future consumption for the total population. Although the results vary for each of the 75 food items and each method, altogether and in general, they define a narrow interval of future consumption development until 2050. All aspects of the approaches and the comparison of the outcomes contribute to knowledge about possible and nontrivial forecasting techniques on big data and foresight about the future of home food consumption.

Daria Loginova, Stefan Mann