Money Can Grow on Trees: Forecasting Ornamental Shrub Demand

Abstract

This paper examines how sales and demographic data can be used to forecast demand and optimize inventory allocation for ornamental shrubs. Using aggregated point-of-sale (POS) data from a regional ornamental shrub grower and national retailer across 442 stores and ten years (2016-2025), I develop store and item level forecasting models to demonstrate how such models could enhance grower shipping and production decisions, increasing profits and decreasing waste. Although demand forecasting has been studied for related perishable goods such as groceries and cut flowers, this paper is the first to examine hardy ornamental shrub demand. I first show that current industry allocation decisions can be substantially improved by incorporating historical sell-through rates and local economic variables in autoregressive (AR) time series models. At the store level, an AR(2) model augmented with demographic data achieves an out-of-sample cross-sectional R-squared of 0.35, and a random forest model using the same features achieves a similar level of accuracy with less variance. At the item level, I find that plant genus and category are significant predictors of sell-through, with hardy evergreens experiencing higher expected sell-through rates than fast-growing varieties; however, the profitability impact of this phenomenon may be negligible due to the tendency of high margin plants to realize a lower sell-through rate than low margin plants. Using a linear approximation of sell-through rates, I estimate allocating inventory based on a random forest model would have increased annual revenues by approximately 1.6% in 2020-25, conservatively translating to a 10-20% increase in profits. I propose a field experiment to empirically estimate marginal sell-through elasticities. Lastly, I suggest that investigating the tendency of analysts in other industries to shrink forecasts to averages and conducting similar analysis on related products such as annuals, carving pumpkins, and Christmas trees would likely produce meaningful, actionable insights that increase profits and reduce spoilage.