Sunday, May 1, 2016

Sir Charles Bean Report on GDP Under Reporting and Spurious US Employment Data

Today a good friend and I visited the Metropolitan Art Museum and after we had finished with the American Wing we sat down in the courtyard in front of the entrance to the federalist building facade.

My friend in the in the alternative industry and he relayed there is an ever building buzz or unease with the current depiction of GDP, that the group he hangs with are suspecting it is under reporting.  He cited the  Independent Review of UK Economic Statistics by Sir Charles Bean as the main reference for this group.  Bean maintains that GDP is missing the impact of the digital age.  This is not a new idea and first came to light in the 90s with the work on "weightless" by the work of Dr. Danny Quah and even Greenspan was citing "weightless" as the reason productivity at the time was being understated at the time.  Quah and Greenspan turned out to be right event though it ended in a rather messy bubble, but still Google is with us to this day.

Sir Charles Bean has taken this idea (here a summary in Vox Dec 2015) and applied greater granularity and applied it to GDP measurement. His work is exhaustive in the larger study The Independent Review of UK Economic Statistics It is a fascinating read and will require more hours of study - and I admit I am late as missed this when it first came out in the end of last year.  It has chock full of great analysis and graphs, one of the first such studies to full embrace "data science".

After I received the email from my friend that provided the Vox link to Bean, I responded that I have noted a similar under reporting in employment and sent him this email:

I think only employment now offers a good read into the current strength of the economy, and only claims non seasonally adjusted provides that read.
The largest part of the hit in 09 tool place in January 2009 and a fair amount of that hit was considered to be seasonal.  I suspect this was political and not a misread.  Unemployment likely hit 12% at the worst but the UER 3 headline was capped at 10% so as to avoid panic.  Remember a key part of the TALF ABS were auto loans and the generally accepted wisdom was that default of the auto loans would equal unemployment and most Auto ABS would start to fall apart at 11% or higher, though supposedly able to survive up to 18% UER.
The difference between 12% ad 10% is about 2.6  million employed and I think the "fix" was achieved by shoving the larger unemployed into a reduction of the Labor Force Participation Rate.  The same seasonal jiggering was made to claims data at the same time. 

While the seasonal and nonseasonal could be tampered with via the LFPR reduction, and thereafter fixed into the ARIMA seasonal machine, claims data could not be so tampered with for long in seasonal and certainly not at all with non-seasonal claims.  When the economy rebounded each change was fairly accurately reported for the monthly survey, but the labor force had to be gaining at a faster rate than reality, to bleed back in the political fix.  The recovery in emplyment has been so strong they are just starting to get caught up but I calculate there is about 1 million under reporting in the monthly data and Seas adj claims data is pretty well caught up given the 5 year memory in the ARIMA seasonal machine, and of course non seasonal adjusted claims data has always been accurate, by definition.

So I examine the year over year change in non seasonal claims data and it shows complete recovery was reached last year and the recovery has been an unusually strong and swift recovery in claims, as shown by examining the data since 87.  The best way to show this data is via a heat map provided here (I cap the levels reached for both claims series so I have more shades of granularity).  Looking across a row gives you a specific week from week 1 to week 52 since 1980.  I find this startling, almost worthy of starting an Inspector General investigation given how out of whack claims are to the monthly data at 5% currently:


The BLS/CES website notes that weekly claims data trumps the survey and once a reasonable time period has gone by to avoid weekly noise, the survey is to re-calibrate to the claims data.  Remember the huge payroll jump we had after the 82 Reagan Recession, or the lesser amount but still shock in 1994?  That was this re-calibration kicking in but this time, even though the recession was equal or worse than the Reagan Recession we have not had the shock re-calibration.

This discrepancy also shows up where there was not a heavy hand in seasonal fudging or manipulating the Labor Force size.  The Beveridge Curve for insured labor force - JOLTS v insured unemployment rate shows a normal recession followed by a normal recovery.  But when a Beveridge Curve is put together for JOLTS v headline unemployment one can see the date they put the fix in to cap headline unemployment at 10% in2009 and 2010.  This is so obvious I stopped keeping the graphs current past last year and here is my last one (I have the axis reversed to usual Beveridge Curves having the JOLTS as causal (x)):

This all fits with the Bean thoughts on GDP for the above political fix could only be allowed if GDP was slow "secular stagnation" type of pattern.
I think the actual headline monthly household unemployment rate is now no more than 4 1/8% and that means GDP is under reported by 1% to 2%.

Feel free to share.....  "

No comments:

Post a Comment