The M5 Competition

The M5 extends the results of the previous four competitions into three important directions.

  • First, it uses hierarchical sales data, generously made available by Walmart, starting at the SKU level and reaching the level of departments, stores, and three large geographical areas. Besides the sales data there is also information on explanatory variables that may influence sales.
  • Second, in addition to point forecasts, the distribution of uncertainty will also be assessed by asking participants to provide information on four prediction intervals and the median.
  • Third, appropriate methods will need to be utilized, given the magnitude of the task of forecasting close to 100K series, as well as estimating the uncertainty around these forecasts.

These methods will have to include machine and deep learning, in addition to the traditional statistical ones, and will require fast computers, probably using GPUs, to complete the tasks of the M5.