Case studies - 10 August 2023

Accuracy of annual forecasts for the cosmetics retail market

Accuracy of annual forecasts for the cosmetics retail market

We determined the forecast accuracy by using the mean absolute percentage error (MAPE), demonstrating how much PMR’s forecasts differed, on average, from the actual values: 



where n is the number of periods (years) for which the variable is forecast. 


The key obstacle to the proper verification of the forecast’s accuracy was a partial change in the historical data caused by the inclusion of new information and the revision of some market value estimates. These changes are not relevant to the market picture but affect the forecast accuracy and render it difficult to calculate the MAPE. 


We therefore present MAPEs which are determined in two ways. The first, unadjusted, MAPE, is based directly on data and forecasts from our 2019-2021 reports and current data. These errors are potentially magnified by changes in historical data. The magnitude of error caused by the changes in historical data is also presented. This is the mean error of the historical data adjustment.  


In addition we present the adjusted MAPE, after eliminating the impact of the historical data changes. Changes in the data mainly affected the values and, to a lesser extent, the rate-of-change, of the forecast indicators. We believe that it is reasonable to adjust the old historical data to match the new data by adopting the same values for the historical data up to the start of the forecasts, for the years 2018-2021 respectively, and then calculating the forecast values by means of the projected change, to compare them with the actual values.  


With this approach, the adjusted MAPEs for the value of the cosmetics market usually fall considerably short of 3%. At the same time, and also in the absence of historical data adjustment, no error of more than 3% is recorded.  


In terms of the value of the entire market from a technical, i.e. statistical, viewpoint, this means that PMR’s forecasts are accurate. 

The predictive accuracy assessment was based on data from our report entitled: “Cosmetics retail in Poland 2022”, and its previous versions . 


Adjusted MAPE, unadjusted MAPE and mean historical data adjustment error for the value of the cosmetics market in Poland, January 2019 – October 2021: 

Publication dateUnadjusted MAPE Mean historical data adjustment error Adjusted MAPE
Jan 20191,3%0,4%1,2%
Oct 20191,8%0,4%1,8%
Feb 20202,9%0,8%3,1%
Nov 20200,9%1,0%1,3%
Mar 20210,2%0,7%1,1%
Oct 20210,2%0,6%0,1%

Average adjusted MAPEs for all product subgroups have low values. Despite the fact that in the report editions published prior to the COVID-19 pandemic, forecast errors were more substantial, at around 4-6%, this reflected the unpredictability of that event. In subsequent editions of the reports, despite the still vibrant situation, the average error values for all subgroups no longer exceed 4%, which also supports the conclusion that the forecasts are accurate. 


Average adjusted MAPE for all product subgroups in Poland, January 2019 – October 2021: 

Publication date Unadjusted MAPE
Jan 20193,9%
Oct 20194,4%
Feb 20206,2%
Nov 20203,9%
Mar 20211,7%
Oct 20212,1%

For sales channels, the average adjusted MAPEs are, in each case, slightly higher, but still statistically acceptable. As is the case for the subgroups, these errors are more substantial for reports issued prior to the COVID-19 pandemic, which was a consequence of the unpredictability of the pandemic itself, along with the restrictions and limits put in place in the various channels. In turn, in more recent reports these errors are, accordingly, less severe and do not exceed 5%.  


Average adjusted MAPE for all sales channels in Poland, January 2019 – October 2021: 

Publication date Unadjusted MAPE
Jan 20198,9%
Oct 20196,3%
Feb 20205,5%
Nov 20204,1%
Mar 20212,9%
Oct 20212,6%
Do you need more data?

In PMR, we advise and assist in the operations of your company

Maciej Gazda

Head of Forecasting