# How-To: Interpreting Results Cubes

Below is a detailed reference to the commodity price forecast cubes and their component fields, in order to facilitate users' interpretation and use in the continuous improvement of planning processes in terms of accuracy and efficiency.

## Qmfcst-week: Forecast Cube

It collects weekly forecasts and backtests (evaluations of historical data against model predictions) for each item and date in the forecast period, suggesting the most likely prices for the operation, supply or S&OP processes.

Name Column | Description | Interpretation |
---|---|---|

Date | Historical date and forecast. | |

Week | Number of the week of the year. | |

Item | Unique identification of the items or references that make up the product portfolio. | |

Target | Actual observation or actual historical behaviour of a particular item for a specific date. | |

SuggestedForecast | Forecasted purchase values for a particular item and specific date. | It should be taken as the most likely purchase value to be observed in the forecast period or date. |

SuggestedInterval | Forecast interval associated with the forecast purchase values. | Allows to identify whether the most likely purchase prices are closer to a sub-forecast (Up60, Up80, Up95) or to an over-forecast (Lo60, Lo80, Lo95). |

NextSuggestedForecast | Over-predicted purchase values for a particular item and specific date. | In S&OP sessions it allows to propose a higher most likely forecast value - by considering the SuggestedForecast to be conservative. |

NextSuggestedInterval | Forecast interval associated with the number of over-predicted units | Allows to identify whether the most likely values to buy, above, is closer to a sub-forecast (Up60- Up80- Up95) or to an over-forecast (Lo60- Lo80- Lo95). |

BackSuggestedForecast | Under-predicted purchase values for a particular item and specific date. | In S&OP sessions, it allows proposing a lower most likely forecast value by considering the SuggestedForecast as optimistic. |

BackSuggestedInterval | Forecast interval associated with the under-predicted purchase values. | Allows to identify whether the most likely values to buy below is closer to an under-forecast (Up60, Up80, Up95) or to an over-forecast (Lo60, Lo80, Lo95). |

ForecastNaive | Predicted purchase values, according to the naive forecast. That is, the forecast is equal to the last actual observation. | Backup forecast for those items and dates with errors greater than the control value (e.g. 50%) and the comparative error measure MASE is greater than 1. |

ForecastPoint | Predicted purchase values estimated by the model for the point interval. | Central forecast value generated by the model for a specific item and date. The forecast has a 50% probability of being located at the Forecast Point. |

ForecastLo95 | Predicted purchase values estimated by the model for the Lo95 interval. | Lowest possible forecast value generated by the model for a specific item or date. The forecast has a 95% probability of lying between Lo95 and the Forecast Point. |

ForecastLo80 | Predicted purchase values estimated by the model for the Lo80 interval. | Second lowest possible forecast value generated by the model for a specific item or date. The forecast has an 80% probability of falling between the Lo80 and the Forecast Point. |

ForecastLo60 | Predicted purchase values estimated by the model for the Lo60 interval. | Third lowest possible forecast value generated by the model for a specific item or date. The forecast has a 60% probability of falling between the Lo60 and the Forecast Point. |

ForecastUp60 | Predicted purchase values estimated by the model for the Up60 interval. | Third highest possible forecast value generated by the model for a specific item or date. The forecast has a 60% probability of falling between the Forecast Point and Up60. |

ForecastUp80 | Purchase values estimated by the model for the Up80 interval. | Second highest possible forecast value generated by the model for a specific item or date. The forecast has an 80% probability of falling between the Forecast Point and Up80. |

ForecastUp95 | Predicted purchase values estimated by the model for the Up95 interval. | Highest possible forecast value generated by the model for a specific item or date. The forecast has a 95% probability of lying between the Forecast Point and Up95. |

WMAPE | Measure of forecast error, Weighted Mean Absolute Percentage Error. | It allows to evaluate the prediction performance of the model for each item. It is usually evaluated in control bands: 0-30% (excellent), 30-50% (acceptable), +50% (poor). |

MASE | Measure of comparative forecast error, Mean Absolute Scaled Error. | Allow to evaluate the prediction performance of the model for each item by comparing the error of the proposed model against the error of the naive forecast (Forecast Naive). Errors less than 1 show that the proposed model performs better. |

Ranking | Union of ABC, FSN and XYZ classifications, according to the criteria of income/profitability - movement and stability. | The items of highest value to the business are AFX, AFY and AFZ. The items of lowest value to the business are CNX, CNY and CNZ. |

** Forecast intervals** are the forecast impact zones, measured as the deviation of the forecast from the actual observation. These zones are delimited by: Forecast Point, Forecast Lo95, Forecast Lo80, Forecast Lo60, Forecast Up60, Forecast Up80 and Forecast Up95.

## Qmfcst-mo: Prediction Cube

Gather forecasts and backtests for the month (evaluations of historical data against model predictions) for each item and date in the forecast period, suggesting the most likely prices for the operation, manufacturing, supply or S&OP processes. for the operation, manufacturing, supply or S&OP processes.

## Qimportance: Importance cube

It shows the impact of the most relevant variables that are being included in the forecast, this cube is organised from highest to lowest the items or variables that have the highest percentage importance or incidence in the forecast and in the other items when they present some variation.

## Qcorrelation: Correlation Cube

In this cube it indicates which items whose history is directly or inversely related to a particular variable, rising or falling in the same proportion. When they are inversely correlated as one variable goes up, the other usually tends to go down in the same proportion.

Correlation is measured from 1 to -1 where 1 is a highly correlated value and -1 is inversely correlated and any value between 0.6 or -0.6 are weakly correlated variables, meaning that they are not significantly correlated with each other.

It is important to mention that in the cube you will find the results organised from highest to lowest correlation to facilitate the analysis of the data.

## Qcausality: Cube of causality

Finally, the causality cube shows when one variable has a direct effect on another variable in the forecast. Its upward or downward behaviour is a cause of the behaviour of the variable of interest. That is, when there is a variation in an item or variable A, it affects an item B. This binary section defines whether the variable is causal or non-causal.

Causality is measured as follows, if the item or variable is greater than 0 it is said to be causal, - 0 not causal. However, in the cube the results are presented in order to facilitate the analysis of the results.