At this point, most readers are likely aware of my skepticism towards broad headlines and general government statistics. For those seeking greater insight into the ways GDP misrepresents economic well-being or other similar issues, I suggest reading the free book, Mismeasuring Our Lives: Why GDP Doesn't Add Up by Joseph Stiglitz, Amartya Sen and several others. In the meantime, Peter Tchir of TF Market Advisors, details a few of the recent headlines and dives into the actual calculations that should provoke such skepticism. Although I don't have a specific tally, weekly initial jobless claims have been revised higher nearly every week for the past year (if not more). An interesting side note on the topic, during the past two weeks these claims were supposedly higher than normal due to a significant number of Verizon employees filing for unemployment insurance. While the report only counts claims filed, I'm curious whether any employees that willingly went on strike received unemployment benefits.
Tchir, whose posts I look forward to daily, also offers pertinent information about channel stuffing, inventory builds and the statistical significance of non-farm payroll reports. In my opinion, these topics only scratch the surface or questionable data mining and reporting. Here are a few other topics that deserve further consideration:
1) GDP is calculated in two different formats by the BEA. Why do the measures deviate from one another?
2) The BEA also uses a GDP deflator to represent inflation and convert nominal GDP to real GDP. The BLS calculates inflation as CPI. The Fed prefers to use PCE (personal consumption expenditures) when measuring inflation. All three tend to deviate from one another and at least recently the GDP deflator has been the lowest (CPI or PCE measures would likely have shown negative real GDP growth in Q1 and Q2). Why are there several different measures, used for different purposes?
3) On average, S&P 500 companies tend to report quarterly earnings that exceed analyst estimates 60 to 75% of the time regardless of the macroeconomic environment. Why are these estimates consistently wrong in the same direction?
4) Analysts and investors frequently highlight cash on the balance sheet as a sign of strength in corporate America. However, a company that borrows $1 million in cash certainly does not have a stronger balance sheet than another company with no cash or debt. Why is the size of debt never considered when discussing the amount of cash?
There are certainly countless other statistics that could be included in this list and are deserving of further research and transparency. Too frequently these statistically insignificant data points are directing policy and allocation of capital to the detriment of the larger economy. Continuing the discussion Stiglitz and others hoped to incite will ideally lead to policies that better address the economic and personal well-being of society.
Guest Post: One Death is a Tragedy, One Million is a Statistic:
Submitted by Peter Tchir of TF Market Advisors
One Death is a Tragedy, One Million is a Statistic
Another day of statistics, where the headlines are widely published, some details are somewhat explored, and in-depth analysis is next to nil.
The
initial jobless claims number has become a farce. It is virtually statistically impossible for this many upward revisions unless the data is purposely under reported. I can understand the desire to smooth data, or make it seem better, but at some point the line of credibility is crossed. Not only do they screw with the main statistic, but they seem to use continuing claims as a secondary diversion. Last week’s 3641k continuing claims seemed statistically implausible, yet it was cheered. The doom and gloom crowd argued that it must be from people using up eligibility, in the end, it was just wrong by over 100k, according to today’s release. How is that possible?
Next we move to
auto sales. It is hard to avoid hearing about auto sales today, probably, because the headline numbers seem good. I can almost ignore the fact that the first thing mentioned is the percentage change from a horrible period last year, but I am shocked the focus is still on total sales. Since at least 2005, the problem with car companies has been selling cars at a profit, not just selling cars. Nothing from the data indicates how profitable the sales are. So we can cheer this headline, but to a large degree it is meaningless. Then, making it more meaningless, is the fact that it includes fleet sales and is really based on sales to dealers. It doesn’t give a clear picture of how many cars were driven off the lot by bona fide, actual, human owners. If anything, the hype surrounding these figures rewards channel stuffing. There seems to be a degree of confusion by those spouting the numbers about the lack of follow through in the stocks. Maybe stocks have finally learned the lessons from these numbers, but it would be great if the masses were presented with details and useful statistics rather than just what the auto companies want to hype.
Then there was the
ISM Manufacturing data. The sighs of relief from trading floors shook the buildings almost as much as last week’s earthquake. Where to begin with this data? It is a “diffusion” index. So it treats each respondent’s answer the same. It doesn’t matter if a company has 5 employees, or 5,000, their answer counts the same in the survey. If the often unreliable ADP report is correct that most of the hiring is occurring at small and medium companies, does that impact ISM?
If 2 small companies report better conditions, and 1 large company reports worse conditions, then the diffusion index would be 66, but the real world impact might be a lot different. Size does matter. Maybe that is part of the reason we see a discrepancy between regional surveys and the national survey? Does ISM report diffusion indices based on size? It would be interesting, at the very least, to see if there is a dramatic difference between big and small companies.
The next thing about the ISM methodology that I find interesting, is that the positive responses include positive responses and ½ of unchanged responses. I guess that is necessary to make a diffusion index, but I would like to see if it is realistic. Do “unchanged” responses have a tendency to follow the trend the following month? I could easily see someone who reported improving conditions one month being inclined to report unchanged the next month, even if conditions were actually worse. These are surveys done by people like you and me, well actually by people with “survey filler outer” included in their job description.
I’m not saying that having the data broken down more precisely would change the market reaction, but I don’t see why it isn’t available, and I don’t see why more analysts aren’t demanding that data. We can all look at a headline, but the value comes from those who can figure out what is going on behind the scenes. If there have been structural changes in the economy (and I believe there have been), then knowing more details would be helpful. The old rules of thumb may be deceiving us.
Of the 5 components, I am most confused by inventories. Inventory growth strikes me as highly suspect. I can see times where it is indicative of future economic growth as companies prepare for increased demand, but equally, it strikes me that it could represent a sudden slow down in final demand resulting in an unexpected inventory build. Had inventories remained unchanged, we would have seen a sub 50 print. You can’t convince me easily that inventory build last month was a positive indicator, yet that is what the headline would have you believe.
This is all in advance of tomorrow’s
NFP report. NFP holds a special place in the dubious statistic category. First we have the fact that there are two separate surveys. We have the establishment and the household. The establishment survey is statistically significant for changes of 100,000; whereas the number is 400,000 for the household survey. In this day and age where virtually everything is done on the internet, I bet Google or Facebook could probably produce a more accurate report within a few months if either one bothered. Until that time, we are stuck with 2 sources of data, both of which have wide margins for error. Then we have the fact that for many months, the birth/death model generates more jobs than the headline itself. That wouldn’t be bad if it didn’t seem to require annual revisions lower. Once again the consistency of the annual revisions indicates that the reports are designed to produce numbers more positive than the reality in the hopes that by the time it is adjusted down, the market has moved on. Having said all of that, the market, or at least the analysts will try and distinguish between 40,000 and 80,000 when the difference is not actually significant or verifiable. They will latch on to whatever survey provides the most positive spin. Those who said the household survey matters more than the establishment survey, will find equally compelling reasons why it is now the establishment report that matters, or vice versa. Obama will be talking about jobs next week, so no matter what number comes up, expect it to be cited often over the next week.
I just realized, the president will be speaking about jobs and more handouts right before the start of the NFL season. Even Stalin might blush at trying to use bread and circuses so obviously.