Impact of the economic crisis on demand: how to improve sales forecasting?

The deepening of the economic crisis, evidenced by the 1,7% contraction of GDP in the third quarter of 2015, is directly affecting household consumption and, consequently, the volume of retail sales. If until the beginning of this year, the crisis was more evident in the industry, now it is possible to clearly identify it in the series of sales of consumer goods in retail.

In addition to the challenges inherent in the drop in demand, companies need to deal with the increase in sales forecasting errors. Although it is not possible to completely mitigate this effect, since historical data do not allow predicting a downward movement, it can be mitigated by using sales forecasting methods that are capable of “reacting” more quickly or that consider the impact of the economic cycle on the series.

We present, in the examples below, the forecast results for the Anfavea motor vehicle sales series (December/2011 to November/2015) and the IBGE retail sales index (December/2011 to September/2015), using the Classical Decomposition methods (with calculation of the economic cycle), Holt-Winters and a simpler third, where the forecast is given by the sales of the same month in the previous year plus the correction made by the average growth/decrease (trend) of the last year , widely used in day-to-day business, according to ILOS research.

New Vehicle Sales

Figure 1 - Analysis of Sales Forecast Methods for New Vehicles

Source: ANFAVEA and ILOS Analysis

When analyzing Anfavea's new vehicle sales series, it is observed that the drop in demand has been accentuated each year. The drastic reduction observed in sales between 2014 and 2015, of 22,55% until November, is a reflection of the withdrawal of tax incentives, in the form of the IPI reduction, as of January of this year.

This sharper drop in demand caused a considerable increase in the sales forecast error, indicated by MAPE, by the Classical Decomposition and Adjustment with Trend methods. However, we observed an improvement in the error result by the Holt-Winters method, which is able to respond more quickly to sudden changes in behavior in the historical pattern of sales. Even so, the Classical Decomposition with the economic cycle calculation was the method with the lowest error.

Supermarket Sales Index

Figure 2 - Analysis of Sales Forecasting Methods for the Retail Sales Index

Source: IBGE and ILOS Analysis

When we analyze the historical series of the IBGE supermarket sales index, we see that the drop in demand begins to be observed only in 2014 and becomes quite accentuated in 2015. Here, the reversal of the sign, from growth in sales in 2013 to decrease in 2014, caused a more intense detrimental effect on the sales forecast than the accentuation of the drop in 2015. The Classical Decomposition methods with analysis of the economic cycle and Holt-Winters improved their predictive capacity in relation to last year, with emphasis on this second method, which with its ability to quickly adapt has become the most assertive.

The results observed here in grouped sales series are similarly valid for forecasting product families and even SKUs in a company. That is, the practice of adjusting the previous year's values ​​based on short-term growth/decrease to obtain a sales forecast, very common in the Brazilian market, tends to present significantly greater errors than those obtained using methods capable of to consider the effect of the economic cycle on the sales series and, in moments of sudden change in market behavior, of methods that are able to adapt more quickly to these changes.

References

<http://seriesestatisticas.ibge.gov.br/lista_tema.aspx?op=0&no=2&de=3>

<http://www.anfavea.com.br/tabelas.html>

<https://ilos.com.br/web/analise-de-mercado/relatorios-de-pesquisa/planejamento-no-supply-chain/>