Trend of Historic Averages algorithm

Trend of Historic Averages (ToHA) forecasts sales or transaction volumes by comparing the trend of this year's actual data against the trend of last year's actual data. This forecast method accounts for seasonality by compiling data from the current fiscal year and previous fiscal year during the same calendar period as the date being forecast.

This section describes how the ToHA algorithm generates a forecast for a single day. In this example, the application is forecasting week 14 in this fiscal year, starting with Monday:
  1. The trend for the current fiscal year is calculated.
    1. The actual or forecasted data is retrieved for Mondays for the previous eight weeks.
      • If actual data is not available for a similar day this year, then the application uses the forecasted value for that day instead.
      • If any of the days are tagged as skip special day, then the application skips those days and uses the previous similar day, going back up to 53 weeks before the day being forecasted.
        Note: If the application does not find any similar days within the 53-week range, then it returns a zero value (0). The forecast generation does not fail, and you can manually edit the forecast after forecast generation.
    2. If there are at least six dates in the set, then the lowest and highest values from the set of current year figures are dropped.
    3. The remaining current year figures are averaged to create the current year trend.
  2. The trend for last fiscal year is calculated:
    1. The matching day for last year is found for the forecasted date. This is done by finding the matching day of the week in the corresponding week of the previous fiscal year. In this example, that would be Monday of week 14 of the previous fiscal year.
      Note: Week numbers are counted from the start date of each fiscal year, which is configured in the Fiscal Year Calendar maintenance form.
    2. The actual data is retrieved for the matching day and the Mondays for the eight weeks preceding the matching day.
      • If any of the current year dates are tagged as skip special days, then the corresponding days are skipped for last year. The previous similar day is used instead, up to 20 weeks before the matching day.
      • If any of the previous year dates are tagged as skip special days, then these days are skipped, and the previous similar day is used instead. When this happens, the corresponding date in the current year is also skipped and replaced by the previous similar day.
    3. If there are at least six dates in the set, then the highest and lowest values from the set of last year's figures are dropped.
    4. The remaining previous year figures are averaged to create the trend for last year.
  3. The change ratio is calculated by dividing the current year trend by the last year trend.
    Current Year Trend / Last Year Trend = Change Ratio
  4. The forecast for the date is calculated by multiplying the actual value for the matching day from last year by the change ratio.
    Monday Week 14 Last Year x Change Ratio = Forecast

To create the forecast for the rest of week 14, the application repeats the steps above for each day in the week.