In the below paragraph, what is the meaning of “benchmark revisions”?
Could someone please explain it in simpler terms and also what is the said para trying to convey
“Furthermore, one or more official revisions to initial data values are common. Sometimes these revisions are substantial, which may give rise to significantly different inferences. Often only the most recent data point is revised. Other series are subject to periodic “benchmark revisions” that simultaneously revise all or a portion of the historical data series. In either case—routine updating of the most recent release or benchmark revision—the analyst must be aware that using revised data as if it were known at the time to which it applies often suggests strong historical relationships that are unreliable for forecasting.”
I found it in official notes of CFA Institute in subject “Asset Allocation”, chapter " Capital Market Expectations, Part 1: Framework and Macro Considerations" in Los “Challenges to Forecasting”
The original publication (in July) of the US CPI for June might be 1.2% annually. As more data come in, in August the June number might be revised to 1.4% annually, and in September it might be revised to 1.55% annually.
This is from memory, so there may well be some inaccuracies, and if so, hopefully someone can post the accurate version.
As an example, in the US, if you have a job, payments for UI (unemployment insurance) are likely deducted from your salary. Along with taking your money, the government collects a lot of data at the same time about employment, salaries, and highly accurate reports are published quarterly.
BLS (Bureau of Labor Statistics) also publish a less accurate monthly jobs report, which is produced using statistical jiggery pokery using stratified sampling from a small (6% rings a bell) subset of the respondents to the larger sample. The weights for the stratified sampling come from the March quarterly report (the benchmark)
Once a year, the March monthly report is compared to the March quarterly report (which is the benchmark) and if necessary the weights for the sampling used in the monthly report are adjusted, which will change the monthly data. When that happens, the weights for the monthly reports for the previous 11 months are also changed, and it is assumed that the changes are linear in time.
Those are benchmark revisions: the historical monthly data is revised because the benchmark has changed.
If I am getting it right, suppose we are in the Jan-Mar quarter, then the monthly BLS reports for Jan, Feb and March, are less accurate and are more of statistical predictions while the quarterly reports for the March quarter are more accurate as they are based on data actually collected.
Now since Mar quarter data would be available at the earliest only in June, we need to revise in June the weights used in the stratified sampling for each of the month from Jan to March and this “revision” in Jan to Mar estimates is called as Benchmark Revisions.
It’s my understanding that, although the more accurate data is collected quarterly, they only use the March quarterly data as the Benchmark, and not the other 3 quarters.
You’re correct about the delay in availability of the data, and I’d forgotten about that.
When the March 2024 quarterly data becomes available, 2 sets of benchmark revisions will be done:
i) revisions to the monthly data for March 2024 and the 11 months BEFORE, so April 2023, May 2023,…, Feb 2024, March 2024
ii) revisions to the 9 months AFTER March 2024, so April 2024, May 2024, …, December 2024.