Question
Why doesn't the historical segment of the forecast equal the corresponding historical series?
Answer
Definitions
Our forecast time series each have two parts, a historical segment and a forecasted segment. Each forecast series is based on one or more historical drivers. The historical segment is derived from the driver(s), whereas the forecast segment is calculated by our model, and the two segments are concatenated. The date of the last period of the historical driver is exposed as the last historical value metadata. The historical segment is intended to provide context to our forecast, but cannot fully substitute for historical data (which is one reason we maintain separate series).
Hereafter, "pair of series" refers to a historical series and the historical segment of the corresponding forecast series.
The series are mismatched
If the pair of series do not match, at least one of these conditions applies:
Synchronization. A change to a driver is not immediately copied to the forecast series; rather, such updates occur only during our regular forecast production cycle. (If you need the value immediately, use the historical series.)
Different frequencies. Our typical forecast series has a quarterly native frequency, but if for example the driver is monthly, the values of the historical segment are an interpolated result. This conversion is irreversible, and the historical segment cannot be "unscrambled." (If you need values verbatim, use the historical series. If you need to align the historical and forecast series, convert the former to match the latter.)
Multiple drivers. Although the description or mnemonic of the forecast may so imply, there is no single historical series to compare against. (Please contact us to clarify the nature of the forecast.)
Mutation. You have modified the forecast series by applying a frequency conversion, analytic transformation, rounding, etc.
The series match exactly
Conversely, the pair of series will match, if and only if these strict conditions apply:
- You consider only the interval of the historical series that has been copied to the forecast series. (Consult the last historical value metadata of the forecast series.)
- Said interval has not been revised since production of the forecast.
- The forecast series is taken natively; it has not undergone a frequency conversion or analytic transformation.
- The historical series has the same frequency as the forecast series (or has been converted to the same frequency).
- A single historical series serves as the driver (or, if multiple series, we provide the formula to combine them).