PlanIQ first removes the trend components from the data and then identifies the seasonality.
Seasonality analysis
PlanIQ identifies repeated performance. It checks for seasonal patterns on weekly, monthly or yearly timescales.
For example, to work out a cycle for a typical day, PlanIQ looks at the performance data for each Wednesday of the year.
PlanIQ identifies seasonality patterns based on the historical data timescale:
Time scale | Weekly seasonality | Bi-weekly seasonality | Monthly seasonality | Quarterly seasonality | Yearly seasonality |
Daily | Yes | Yes | Yes | ||
Weekly | Yes | Yes | Yes | ||
Monthly | Yes | Yes |
Here's the minimum amount of historical data you need for each type of seasonality analysis:
Time scale | Weekly seasonality | Bi-weekly seasonality | Monthly seasonality | Quarterly seasonality | Yearly seasonality |
Daily | 5 weeks | 6 weeks | 3 months | ||
Weekly | 3 months | 3 quarters | 3 years | ||
Monthly | 3 quarters | 3 years |
Even with postive historic values, you can get negative values in your seasonality analysis. This is due to how the analysis is carried out:
- The trendlines are subtracted from the historic data
- Then the seasonality data is calculated from the remaining historic data points, which can produce some negative seasonality points.
If PlanIQ cannot identify any seasonal patterns, it can return noisy values.
Tip: The more historical data you provide, the more accurate your seasonality analysis.
Trend analysis
Trend is calculated as a moving average performed over a period of time. In PlanIQ:
- Quarterly trends are calculated over the last 91 days
- Yearly trends are calculated over the last 365 days
For trend analysis to happen:
- You need at least 50% of filled values of historical data for a set period.
If more than 50% of historical data is missing then it's not possible to calculate trends for the item for that period. - Items with a shorter history or new to the dataset do not produce trend data .
Over time, this changes as more historic values for these items comes into the dataset.
Trend analysis is lagging
To calculate quarterly trends, you need to provide the previous 3 months of historic data.
To calculate a yearly trend for 2021, you need to provide the previous 12 months of historic data.
For example, to calculate the yearly trend for March, provide data from March 2020 to March 2021
Find out how to include seasonality and trend analysis in your forecast results.