A carregar...
Publicação
Proposta de Metodologia para Seleção de Modelos de Previsão de Vendas para a Sumol+Compal
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.27 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Em Portugal, o consumo das bebidas não alcoólicas tem vindo a crescer nos últimos anos o que gerou um aumento da competitividade das empresas que trabalham neste setor de atividade.
A presente dissertação foi desenvolvida na empresa Sumol+Compal com o objetivo de analisar o processo de previsão de vendas e apresentar uma proposta de metodologia que permita melhorar a precisão do cálculo da previsão de vendas. Para o desenvolvimento da dissertação foram estudadas 3 Famílias de referências nos 4 canais de distribuição da empresa, nomeadamente, Horeca, Alimentar, Distribuidores e C&C. Uma vez que a concorrência neste setor é elevada a Sumol+Compal necessita de ter um processo de previsão de vendas preciso para ter uma maior eficiência.
Na previsão de vendas de cada referência por canal de distribuição, são aplicados 2 tipos de modelos, o Modelo de Holt-Winters e o Modelo de Decomposição Clássica, considerando as formas funcionais, multiplicativa e aditiva, e três hipóteses de aplicação que depende do nível de agregação das referências por canal e da sequência de aplicação. Na hipótese 1 é desenvolvido um modelo de previsão de vendas ao nível da Família de referências, sendo depois desagregadas sequencialmente as previsões para cada referência e canal. Na hipótese 2, são desenvolvidos modelos de previsão de vendas para cada referência que compõe a Família de referências, independentemente do canal de distribuição. Depois os valores são desagregados para cada canal. Na hipótese 3 são desenvolvidos modelos de previsão de vendas para cada referência em cada canal. É objetivo em termos metodológicos identificar qual a hipótese de modelação mais precisa.
A seleção do modelo e hipótese mais vantajosa para cada Família de referências é realizada tendo em conta a análise do erro associado à amostra de validação, tendo sido escolhido o modelo e hipótese que apresentavam um menor erro global e um maior coeficiente de determinação relativamente à amostra de validação. Assim, o modelo de previsão de vendas para a Família X é o modelo de Holt-Winters aditivo, para a Família Y é o modelo de Holt-Winters multiplicativo e para a Família Z são os modelos de Holt-Winters multiplicativo e aditivo, uma vez que apresentam a mesma precisão.
Através das análises elaboradas conclui-se que cada Família de referências analisada deve ser tratada individualmente, uma vez que apresentam comportamentos distintos entre si. Tal deve-se ao facto de cada uma das Famílias ser muito específica de um dos canais de distribuição da empresa.
In Portugal, the consumption of non-alcoholic beverages has been growing in recent years, which has generated an increase in the competitiveness of companies that work in this sector of activity. This dissertation was developed in the company Sumol + Compal in order to analyse the sales forecasting process and present a methodology proposal that will improve the accuracy of the sales forecast calculation. For the development of the dissertation were studied 3 reference families in the 4 distribution channels of the company, namely, Horeca, Food, Distributors and C&C. Since competition in this industry is high, Sumol + Compal needs to have an accurate sales forecasting process for greater efficiency. In the sales forecast of each reference by distribution channel, two types of models are applied, the Holt-Winters Model and the Classic Decomposition Model, considering the functional, multiplicative and additive forms, and three application hypotheses depending on the level. aggregation of references by channel and sequence of application. In hypothesis 1, a family of sales forecast model is developed, and then the forecasts are sequentially disaggregated for each reference and channel. In hypothesis 2, sales forecasting models are developed for each reference that makes up the Reference Family, regardless of the distribution channel. Then the values are disaggregated for each channel. In hypothesis 3 sales forecast models are developed for each reference in each channel. It is methodologically objective to identify the most accurate modelling hypothesis. The selection of the most advantageous model and hypothesis for each family of references is made taking into account the analysis of the error associated with the validation sample. The model and hypothesis that presented the smallest overall error and the highest coefficient of determination in relation to the sample was chosen. validation Thus, the Family X sales forecast model is the additive Holt-Winters model, for Family Y is the multiplicative Holt-Winters model, and for Family Z the multiplicative and additive Holt-Winters models, a since they have the same accuracy. Through the elaborated analyses it is concluded that each family of references analysed must be treated individually, since they present distinct behaviours among themselves. This is because each Household is very specific to one of the company's distribution channels.
In Portugal, the consumption of non-alcoholic beverages has been growing in recent years, which has generated an increase in the competitiveness of companies that work in this sector of activity. This dissertation was developed in the company Sumol + Compal in order to analyse the sales forecasting process and present a methodology proposal that will improve the accuracy of the sales forecast calculation. For the development of the dissertation were studied 3 reference families in the 4 distribution channels of the company, namely, Horeca, Food, Distributors and C&C. Since competition in this industry is high, Sumol + Compal needs to have an accurate sales forecasting process for greater efficiency. In the sales forecast of each reference by distribution channel, two types of models are applied, the Holt-Winters Model and the Classic Decomposition Model, considering the functional, multiplicative and additive forms, and three application hypotheses depending on the level. aggregation of references by channel and sequence of application. In hypothesis 1, a family of sales forecast model is developed, and then the forecasts are sequentially disaggregated for each reference and channel. In hypothesis 2, sales forecasting models are developed for each reference that makes up the Reference Family, regardless of the distribution channel. Then the values are disaggregated for each channel. In hypothesis 3 sales forecast models are developed for each reference in each channel. It is methodologically objective to identify the most accurate modelling hypothesis. The selection of the most advantageous model and hypothesis for each family of references is made taking into account the analysis of the error associated with the validation sample. The model and hypothesis that presented the smallest overall error and the highest coefficient of determination in relation to the sample was chosen. validation Thus, the Family X sales forecast model is the additive Holt-Winters model, for Family Y is the multiplicative Holt-Winters model, and for Family Z the multiplicative and additive Holt-Winters models, a since they have the same accuracy. Through the elaborated analyses it is concluded that each family of references analysed must be treated individually, since they present distinct behaviours among themselves. This is because each Household is very specific to one of the company's distribution channels.
Descrição
Palavras-chave
Previsão de vendas Modelo de Holt-Winters Modelo de Decomposição Clássica Retalho alimentar Canais de distribuição Famílias de referências