Choosing Seasonally Adjusted Data for Inflation Models
Summary
The document considers whether an inflation factor built from the GDP deflator, finished-goods producer prices, and a spot commodity index should use seasonally adjusted or unadjusted historical series. The answer ties the choice to the forecast target. Seasonal patterns can matter when modeling or forecasting inflation levels within a year, because weather, purchasing behavior, industrial activity, and tax effects can influence observed data. Unadjusted series may be adequate when the model focuses on year-over-year changes.
The response emphasizes practical data limitations. Seasonal adjustment factors are recalculated regularly, and revisions can alter earlier observations, complicating reproducibility and introducing potential modeling error. It therefore recommends understanding the construction and revision history of each series and weighing the benefits of adjustment against the time and data resources required. The discussion offers general guidance rather than a definitive rule: the appropriate treatment depends on the intended use, series availability, and the model’s data collection schedule.
Key ideas
- Choose seasonal adjustment based on whether the model targets within-year levels or year-over-year changes.
- Seasonal effects can reflect weather, purchasing patterns, industrial activity, and tax influences.
- Unadjusted data may be a practical choice for models focused on year-over-year movements.
- Revisions to seasonal factors can change historical observations and reduce reproducibility.
- Inspect each series’ construction and revision schedule before incorporating it into a model.
Tags
Full text
# Do you use seasonally or non seasonally adjusted index in analysis # Do you use seasonally or non seasonally adjusted index in analysis I am constructing an inflation factor that includes the gdp deflator, the PPI for finished goods, and a spot commodity index. Do I uses seasonally or nonseasonally adjusted historical series? Thank you ## Answer by Emma Muhleman CFA CPA (score 3) https://quant.stackexchange.com/a/43464 It depends on the intended end-use of your model, but generally-speaking, if you were solely trying to measure and forecast inflation levels or the GDP deflator over the course of a year (including the use of, say, the GDP deflator percentage change in March, as a factor that somehow goes into your April forecast), you would need to consider seasonal adjustment factors due to the impact weather and other related events have on purchasing patterns, commercial and industrial activity levels, the influence of taxes on PPI, etc. That being said, it might be a lot easier to use non-seasonally-adjusted data, which could be okay if you're only using it as a factor to forecast and analyze year-on-year (YoY) changes (as part of a time series, for instance). While seasonally-adjusted data is great in theory, data limitations and BLS revisions to the seasonal adjustment factors (which occurs every January) may limit the practicality of this approach. For more detail, as you will likely want to get a strong understanding of the data series you are going to use if you want it to be of any value to a model, see https://www.bls.gov/cpi/seasonal-adjustment/intervention-analysis-seasonal-adjustment-2018.pdf. Here, the BLS explains in detail why and how it performs what it calls 'Intervention Analysis in Seasonal Adjustment,' which is something you will want to understand. Also, note the BLS recalculates its seasonal adjustment factors every January for the preceding year's data and notes that this routine annual calculation "may result in revisions to seasonally adjusted indexes for the previous 5 years." This could just add another nuance to your modeling and introduce potential errors, depending on the regularity of your data collection, etc. Note also that BLS will make available recalculated seasonally adjusted indexes, as well as recalculated seasonal adjustment factors, for the period January 2014 through December 2018, on Monday, February 11, 2019 (updated data will be released here), so you might want to look out for that updated data series if you do use seasonally adjusted figures. Hope this helps. In short, some things are great in theory but aren't all that practical in reality. Then take into account all the assumptions and other potential sources of error in an econometric model and it may seem more effective to use the unadjusted prices. It is ultimately your choice and depends on the amount of time and resources you will devote to the project, but generally, if you are attempting to build a robust model, you might want to find a way to avoid the revised, seasonally-adjusted data.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.