Forecasting fashion retail sales in Brazil: a case study before, during and after COVID-19

Abstract

Purpose

This study analyzes the sales behavior of a Brazilian fashion retailer before, during, and after the COVID-19 pandemic, aiming to generate short-term forecasts using machine learning models. The pandemic’s impact on the retail sector created a need for accurate sales forecasting.

Design/methodology/approach

Sales behavior was analyzed using Seasonal Autoregressive Integrated Moving Average (SARIMA) and Neural Network AutoRegressive (NNAR) models. Performance was tested during the Full-Price and Off-Price stages, considering eight clothing collections launched before and during the pandemic. Forecast accuracy was evaluated using the Root Mean Square Error (RMSE), Symmetric Mean Absolute Percentage Error (SMAPE), and Total Absolute Percentage Error (TAPE).

Findings

Sales before and after COVID-19 showed low volume and variability during the full-price period and high income volatility during the off-price stage. Collection 4, launched in February 2020, displayed stable sales with reduced promotional impact. NNAR slightly outperformed SARIMA, highlighting the importance of nonlinear models in capturing retail sales volatility. Sales showed greater variability before and after restrictions, particularly during discounts, which resulted in higher prediction errors.

Originality/value

This study helps fashion retailers to choose suitable models for forecasting sales during the full- and off-price stages, considering specific environmental conditions. It also provides insights into retail dynamics during disruptions.