Delay of Release of PPI for January 2004 and February 2004

Bureau of Labor Statistics, US Government ^ | March 12, 2004 | Bureau of Labor Statistics

[doug9732]

U.S. Department of Labor Bureau of Labor Statistics Producer Price Indexes

As announced on February 17, the release of the Producer Price Index (PPI) for January 2004 has been delayed from the originally scheduled date of February 19, 2004. In addition, as announced on March 5, the release of February data originally scheduled for Friday, March 12, has also been postponed.

The delays have been caused by unexpected difficulties in the conversion of PPI data from the Standard Industrial Classification system to the North American Industry Classification System. These difficulties have taken far longer to resolve than we originally anticipated.

We have made substantial progress towards completing the calculations underlying the January 2004 PPI. When revised release dates for the January and for the February 2004 Producer Price Indexes have been determined, we plan to announce them at least one day ahead of time on this web page and through news advisories.

The Bureau of Labor Statistics expresses its sincere apologies to those who have experienced any problems as a result of this continuing delay.

[doug9732]

The US Government has suddenly "lost" its ability to calculate the Producer Price Index. Does anyone know if this has ever happened before? Could it be that the numbers are politically too bad to announce?

[lelio]

Place your bets on when this number will be released. Any sales of government securities coming up? My guess will be shortly after the sale the number will magically appear.

[So Cal Rocket]

No, this has to do with the switchover from the SIC Codes to the NAICS Codes for collecting, sorting and analyzing data. I do econometrics for a large company and we're struggling with the switchover as well.

[ambrose]

the math geeks who crunch these numbers are not political appointees.

[Our man in washington]

No.

Let's just say I have inside knowledge on this. What we have is a collosal screwup. It's unprecedented as far back as anyone remembers.

[lelio]

No, this has to do with the switchover from the SIC Codes to the NAICS Codes for collecting, sorting and analyzing data

This is why I think the government should just release the raw data and let the market figure out an algorithm to combine it all together. That way everything's out in the open and more can be gained from the data instead of just one singular number.

Competing methods will come up, but I believe it will offer a better picture of the economy. Who knows how that figure is calculated now?

[Starwind]

What is it, technically or logistically, that makes this switchover difficult, please?

[Our man in washington]

The problem is confidentiality rules. For most surveys, companies aren't required to give the BLS data. Companies give the BLS data on the understanding that it's for compilation purposes only.

BLS could take company names off the data, but the reality is that it wouldn't be that difficult to identify certain big companies. If a company has 100,000 employees in metropolitan area X and industry Y, it wouldn't be that difficult to tell its identity.

[Our man in washington]

Let me take a stab at that.

If you change measurement systems, you have to change your entire production process for a survey. First of all, companies often miscode what industry they're in, so you have to retrain the data collectors. You also have to redesign all the forms and vehicles for entering data.

Weighting and nonresponse adjustment is based on industry, so the statisticians have to redesign the processes for that. You then have to take all the new processes from the statisticians and survey designers and create new software to compile the numbers.

However, BLS has known for years they had to make the switchover. BLS needs to be put on the hot seat and should explain how they'll insure this historic failure will never happen again.

[So Cal Rocket]

Reader's Digest version:

You have to be able to compare the current month's results using NAICS data to previous month's results which were formulated using SIC data - on an "apples to apples" basis.

That means understanding what went into each of the SIC datapoints and what is going into the new NAICS datapoints, and developing new metrics using the NAICS data to replicate what we were modelling using the SIC data. For me, that means researching the data behind several dozen metrics and formulating a new model.

My company is concerned with only a small portion of the overall economy (aerospace products). The econometricians at BLS have to do this analysis on thousands of metrics which represent the entire economy.

It's quite obvious that they underestimated the complexity of their task.

[Starwind]

Thanks,

However, BLS has known for years they had to make the switchover.

Yes, I would have thought most of the tasks you listed would have been done and tested in advance. I could understand some last minute problems related to different/wrong survey responses necessitating some survey and software tweaks...but not two months.

Yet, So Cal Rocket says his company is likewise encountering difficulties, so what is it about this cutover that couldn't be known/tested in advance?

[lelio]

and software tweaks

From what I read they're also running on "30 year old computers" which might mean they have to dig up the punchcard reader to change the software. Also how many data points are we talking about? I read numbers of 150-400k. Perhaps their new algorithms are so complex that their systems couldn't handle it. I sense a Seti-At-Home project in the works!

I can see a problem if a percentage of their customers (for lack of a better word) gave them data in half SIC and half NAICS formulations and then the BLS would have to go through manually and correct each item. Man what a pain.

Thanks OurMan and SoCal for the detailed explainations.

[Starwind]

Thanks,

You have to be able to compare the current month's results using NAICS data to previous month's results which were formulated using SIC data - on an "apples to apples" basis.

Is this comparison problem a one-time conversion thing, or anytime old data is compared?

Is this because the SIC data (at BLS and your company both?) is not being converted? Only new data entered is NAICS, and so any comparison with data from Dec '03 back has a special comparision conversion problem, whereas any comparisions among new NAICS data only is straight forward?

How do you know when your new model is right?

[So Cal Rocket]

There was a period of time in which the SIC data and NAICS data were both published. We've been revamping our models and comparing the results over this time period - and tweaking the models to get consistant results... unfortunately, for industry, we don't have access to much of the "raw data" behind the metrics, so we have to do it by "trial and error".

[mewzilla]

This stuff is why I prefer chicken entrails for economic forecasting...

[Our man in washington]

You are right. All that I described could be tested ahead of time. And it was.

According to my sources, BLS is organized along functional lines. The computer people have a separate hierarchy from the people who manage the survey, and the data collectors have their own hierarchy. The problem with that approach is that people get ahead by learning their one process and not any other process.

When everyone has to work together to build a system, few people are available who can truly understand the big picture. Everyone understands their little piece. Thus collaboration is difficult, and lack of communication leads to massive failures.

[Our man in washington]

By the way, I bookmarked your profile page. Great set of links you've compiled! Everyone interested in economic data should check out Starwind's profile page.

[getgoing]

any news?

[SCR1]

Does anyone care that the number has not been released. What about Wall Street?