Sales Forecasting With A Great Number Of Variables: The Case Of Intra- And Inter- Category.
Name: JOÃO PEDRO ARAUJO DOMINGUES
Publication date: 04/05/2018
Advisor:
Name![]() |
Role |
---|---|
HÉLIO ZANQUETTO FILHO | Advisor * |
Examining board:
Name![]() |
Role |
---|---|
HÉLIO ZANQUETTO FILHO | Advisor * |
MARCELO MOLL BRANDÃO | Internal Examiner * |
Summary: In the present work the main objective was verify if time series of Intra and Inter-category were able to improve sales forecasting model in short term for retail. This is a study case that used multiple regression and the methodological selection of variables Lasso (Least Absolute Shrinkage And Selection Operator). The specific objectives were: (1) Confirm empirically the existence of complementarity and substitutability in Intra and Inter-category; (2) Propound a sales forecasting model that uses time series from both Intra and Inter-category; (3) Compare the results between the model which uses only one time series and the model which uses Intra and Inter-category; (4) Identify if there are any difference between the results using Intra or Inter-category. Therefore, the major results have shown the presence of complementarity and substitutability among Intra and Inter-category at weight level. Furthermore, the outcomes have shown the more prevalence of complementarity, with 88.8% of the interactions, and the rest 11.2% were substitutability. Other results displayed that 83.8% of the improvements of RMSE comes from time series from Intra-category, which represents the prime majority. Inside this percentage, the decrease average of RMSE was 56.60%. Meanwhile, the research highlighted that Inter-category was capable to improve 16.2% of accuracy, showing the reduction of the error and proving the series interaction cross-category. Finally, the study concluded that the usage of time series from Intra-category can improve the accuracy in the majority of the cases, and the reduction reached satisfactory outcomes.