Watch the trend, not just the latest data point

Some excellent thinking from Barry Ritholtz in his recent post NFP Report: Trend vs Single Data Point discussing on how best to interpret the latest job numbers.   

The key ideas I take away from this are to 1) watch the trend, 2) ask the right questions to establish context, 3) examine a wider range of metrics (not just the headline numbers),  and especially 4) not to over focus on the latest data point. 

Related to some of our recent posts, one of the things that makes following Barry's advise more difficult these days is the way that the standard reporting modes for key metrics almost universally use Month-over-Month, and/or or Year-over-Year figures.   Other ways of reporting the data can (such as net change over 2 years or 5 years) can reveal details that would otherwise remain hidden.  

Playing with the data, keeping context in mind, and looking for the trends and the answers to our deepest questions is the best path to more complete understanding and improved decision making. 

Showing a key metric with multiple views: a nice example

Bill McBride's Calculated Risk blog has some crisp charts showing the latest new unemployment claims. The main chart shows this key metric since January 2000.

A second chart shows the same metric going all the way back to January 1971.



Both charts use a 4 week moving average to smooth out the more erratic week to week behavior. Bill's use of a dual chart approach helps present a much more complete picture of this important metric that puts recent behavior in context. Of course, even his "short" period is almost 11 years long so doesn't suffer from the common weakness of plotting too few data points.

Additional employment related charts showing other metrics and other views can be found in the Employment tab of Calculated Risk's Graph Gallery. Bill is prolific and posts some of the best looking, most unique charts related to economics and finance. Check out his gallery for yourself. You won't be disappointed

Despite these two excellent charts, one weakness I see in Calculated Risk's presentation of this important unemployment metric is that the verbal storytelling is weak. Bill's charts have potential explanatory power with important stories to tell, especially combined with the other charts in the Employment tab of the gallery, but these stories are left mostly as an exercise for the viewer.

In the blog post, the "story" told is mostly quotes from the dull boilerplate in the Department of Labor's UNEMPLOYMENT INSURANCE WEEKLY CLAIMS REPORT. This text discusses this metric with a very short term focus of only the preceding 4 weeks.

A second weakness is that the reporting (like almost all other reporting on the subject) only talks about and shows charts for this one Headline Initial Claims metric from the report while other complementary metrics are shunted aside. For example, some key missing metrics that are mentioned in the DOL report and whose short and long term time series charts could help us better understand the unemployment situation include:

• insured unemployment rate - the percentage of "covered" workers collecting regular state benefits
• insured unemployment - the number of people currently collecting regular state benefits
• total persons claiming benefits in all programs

Some other metrics from other sources might also be added to the mix for fuller understanding such as:

• total persons unemployed
• percentage of total unemployed who are collecting benefits in all programs
• total unemployed who are NOT collecting benefits

Note that Calculated Risk's Employment Tab does include these useful and complementary metrics shown in easy to digest graphic form but a story line to tie all these metrics together remains a challenge for another day.

• headline unemployment percentage
• employment population ratio
• participation rate
• number of workers who are part time for economic reasons
• number unemployed for over 26 weeks
• number unemployed for over 26 weeks as percentage of civilian labor force

What other employment related metrics would you like to see?

Do you know of others posting on the initial claims number who are crafting more complete stories than the standard laid down by the DOL report?

Digging into the Job Opening Data

As mentioned in the previous post, the Bureau of Labor Statistics makes detailed trend data available on all the key Job Opening and Labor Turnover Survey (JOLTS). You can take a look yourself at the BLS data page and then scrolling down to select one of the option buttons (e.g. for most requested statistics) for Job Openings and Labor Turnover Survey.

You can set the time period for the trend graphs created to be the entire time since the year 2000 that JOLTS data has been collected. Here are a few examples. It's somewhat puzzling to me why the Openings rate seems to be going steadily up from 12/2003 onward while the Hiring rate stays steady beginning around 12/2004.





There's a ton more data at the BLS web site. The biggest difficulty for me and I bet for others is just how time consuming and inflexible the BLS' trend visualization application (TVA) is to work with.

Job Openings and Labor Turnover Report

Here's the two reasonably readable graphs from this morning's JOLT (Job Openings and Labor Turnover) report from the Bureau of Labor Statistics (BLS). Looks like there has been a moderately strong trend at work in the job openings picture for the past 3 years.



As has been typical in the BLS formal reports like this, the time span is too short to be able to place recent behavior in context and the number of factors shown in visual form is far too few to fully grasp what is going on in the world of employment.

In the full monthly report itself which runs to 15 pages, there are numerous tables with literally hundreds of important factors. Unfortunately, these covering an even shorter two year time period with even fewer sample trend sample points per factor. This again matches the current BLS standard not-particularly-reader-friendly approach for their "printed" reports.

In my view, this standard BLS approach is far from the best way to present important trend data. All the time consuming work of searching for and extracting meaning is left to reader (which means in most cases it will never happen) or it is left to the expert pundits who will typically comment on a few of the sub factors that they find most interesting and perhaps present one or two charts.

Of course, back at the BLS web site, just about all the trend data for all the factors for all the time periods is available for those who have the time and skill to track it down.

Selective Perception, Cognitive Bias, & the Recency Effect

Barry Ritholtz has an excellent post this morning commenting on NonFarm Payroll (NFP) increase of 180,000 from yesterday's Bureau of Labor Statistics Employment Situation Report.

His comments on selective perception, cognitive bias, the "recency effect" and the soft prejudice of low expectations apply well beyond the NFP example.

Let's begin with a quick word on cognitive bias. Humans are guilty of this -- selectively perceiving and recalling what agrees with their world view. We are all guilty of this, and while we cannot escape it, being aware of it at least allows some measure of recognition, and perhaps, adaptation to the phenomena.

Let's use the NFP data as an example: Consider another bias, the Human tendency to overemphasize more recent data versus the totality of information and the overall trend. This is a cognitive bias known as the recency effect. Despite the overwhelming evidence showing this to be a generally weak jobs recovery (the worst since WWII), our primate brains interpret a single good data point as proof of something better. ...

Regardless, we also see some of the soft prejudice of low expectations in yesterday's data: 180k is hardly a rockin' strong number, relative to population growth,. Put that into the context of recent expansions such as the 1990s. Oh, and in that more recent period, there was no BLS Birth/Death adjustment, responsible for nearly a million fictitious jobs in 2006.

Unfortunately, the two trend chart he shows only go back three years (suffering themselves from the recency effect) so we can't compare the relatively weak current numbers to the NFP increases of the 1990s. The two charts shown in the Bureau of Labor Statistics Employment Situation Report also only go back 3 years.

So, below I have posted a 15 year view showing year over year percentage change for the seasonally adjusted value of the growth in NonFarm Payroll. It's pretty clear from this longer term chart that our growth in non-farm employment during this latest expansion never got back to the steady, multi-year 2.5% growth rate of the late 1990's and appears to be declining again.

Of course, in a subject area as complex as employment, an equally important danger in my opinion is the almost universal tendency these days to stick with just a few headline factors. If we truly want to understand employment, I believe we have to make it easy to visualize and then examine a full range of factors over a substantial time period.

For example, the recent drop off shown above does not seem to square with the reported drop in the headline unemployment rate. Maybe some other factor of the thousands of factors recorded by Bureau of Labor Statistics (BLS) can help explain it.

Too often, really important factors aren't mentioned at all, and the ones that are mentioned are not accompanied by their corresponding trend charts.

Here's a link to a short list of the most popular BLS statistics which might be a useful starting point for a deeper analysis.