September 15, 2013
“It is my understanding that the Constitution of the United States allows everybody the free choice between cheesecake and strudel.”
– Sky Masterson (“Guys and Dolls”)
“One of these days in your travels, a guy is going to show you a brand-new deck of cards on which the seal is not yet broken. Then this guy is going to offer to bet you that he can make the jack of spades jump out of this brand-new deck of cards and squirt cider in your ear. But, son, do not accept this bet, because as sure as you stand there, you’re going to wind up with an ear full of cider.”
– Sky Masterson (“Guys and Dolls”)
“So long as they continued to work and breed, their other activities were without importance. … Heavy physical work, the care of home and children, petty quarrels with neighbors, films, football, beer and above all, gambling filled up the horizon of their minds. To keep them in control was not difficult.”
– George Orwell, “1984”
“If God did not exist, it would be necessary to invent Him.”
“Mistah Kurtz – he dead.”
– T.S. Eliot, “The Hollow Men”
Growing up in Alabama, gambling was part of the fabric of my life. Sports and card games of all sorts were constants, but the truth is that anything was fair game for a bet, from the behavior of lightning bugs to the duration of a thunderstorm. I learned important market-oriented lessons within this particular sort of educational framework, from the power of leverage in playing gin rummy with my uncle at a penny a point and Hollywood scoring (ouch!) to the Eureka moment when I gave my father 5 points on a college football team, took 7 points from my grandfather on the same team, and realized what would happen if they lost by 6. Maybe the most valuable lesson was one of my grandmother’s bridge dictums – “there is no such thing as a good re-double” – which has served me well time and again in a wide range of real-life situations.
Of all the gambling activities that permeated my life, however, only one was licensed and sanctioned by the state of Alabama – dog racing. I’m sure it’s hard for anyone under the age of 40 or anyone who grew up in an economically-privileged geography to understand the appeal of dog racing, but it used to be a pretty big deal. More to the point, it was the only game in town for my entire region, as the Gulf Coast casinos and even a state lottery were still decades away from realization, and horse racing might as well have been Formula I racing given the expense involved of the facilities and the animals. In fact, horse racing eventually came to Alabama in 1987 with a Birmingham-area race track, but it was an unmitigated economic disaster and was ultimately converted into a simulcast and live dog racing venue. Dog racing may not have been “the sport of kings”, but it was definitely the sport of the kings of Alabama, as the industry was controlled by Paul Bryant, Jr. – son of legendary Alabama football coach Bear Bryant – who amassed an enormous fortune by virtue of this state-sponsored monopoly over legal gambling.
Is dog racing systematically abusive to greyhounds? Yes, I believe it is, although I can promise you that there were far more gruesome “sports” going on with dogs throughout the area. Would I ever go back to a dog track? No, I would not, but I’m not sorry that I did. The truth is that as a 17-year old kid out on a Friday night with his buddies in Greene County, I was blissfully unencumbered by concerns over the training methods employed or the post-race conditions endured at the kennel. The truth is that I enjoyed the rush of gambling, the excitement of hearing the announcer say “Heeere comes Lucky!” as the mechanical rabbit sped by the gates, and I learned some important lessons about both markets and human nature from the experience.
Now supposedly it’s possible to handicap the potential risk and reward of a dog race as expressed in the odds posted at race time, but this always struck me (even at age 17) as a poor proposition. First of all, unless you are a trainer, and maybe not even then, it seems enormously difficult and costly to acquire private information. And even if you had useful private information, not only is it substantially less than a zero-sum game as the track managers take their cut, but also there is no market-maker to provide liquidity or lock-in the odds/price of your bet. You have some expectation of what the odds are likely to be, based on the tout sheets and the track’s pre-race odds estimations, but the actual pay-outs are entirely dependent on the distribution of actual bets placed by your fellow spectators and have no direct relationship to the pre-race estimations. You may correctly deduce that the pre-race odds on a particular dog are wrong and present an asymmetric risk/reward opportunity, but if everyone else makes the same calculation that asymmetry vanishes (or reverses) in a heartbeat, usually after you’ve placed your bet! Pari-mutuel betting is a wonderful example of the Common Knowledge game in action, where a dynamic assessment of what everyone thinks that everyone thinks is the most useful way to predict attractive pay-out odds, but even with perfect game-playing the enormous transaction costs and shallow liquidity make it a losing proposition. Still worse, though, is to believe that your “fundamental analysis” or straightforward handicapping of the limited public information actually provides you with an asymmetric risk/reward opportunity when everyone else is doing exactly the same thing.
Is today’s stock market the functional equivalent of a dog race, where fundamental analysis of the risk and reward associated with a security based on limited public information is a sucker’s game? No. The fundamental investor is rescued over time by the non-zero sum nature of the stock market, as well as the presence of non-fundamental investors with different preference functions.
But is there a dog racing aspect to today’s stock market that is growing in size and influence, aided and abetted by powerful institutions that want you to respond atavistically to the market equivalent of a mechanical rabbit and the “Heeere comes Lucky!” call? Absolutely.
This past Thursday, September 12th, the weekly unemployment claims number came out as usual at 8:30 AM with a shockingly low result – only 292,000 Americans had filed for initial unemployment benefits in the past week, the lowest number in more than seven years and a decline of more than 30,000 applicants from the prior week. As it turns out, neither Nevada nor California reported complete data to the Labor Department, so the data release was significantly under-reporting actual claims, but this error was not part of the initial news release. Instead, the Labor Dept. only informed a few reporters of the mistake in an embargoed fashion, which meant that the “news” of the embedded error in the filing was announced in an ad hoc fashion by private news agencies after the official release. Both the substance and the process of the error were entirely avoidable, and it was such an egregious mistake that you had past and current officials of the Labor Dept.’s Bureau of Labor Statistics (BLS), who are apparently not directly responsible for collecting this data from the states, criticizing other Labor Dept. employees with more direct responsibility. To use a sports analogy, this is like the defensive line of a professional football team holding a press conference to criticize the offensive line … it never happens.
It’s no surprise that the Labor Dept. made a ridiculously large error with the weekly initial unemployment claims, because this is a systematically biased and flawed data source, particularly under the current Administration. The chart below is a histogram (frequency distribution) of the errors made in the reporting of weekly unemployment claims during the Bush Administration from September 30th, 2005 through the 2008 election.
The horizontal axis consists of different “buckets” of errors as measured by the difference between the revisions that are reported one week later and the original report. For example, if the original report said that 300,000 people filed for unemployment benefits in a given week, but the revised report one week later said that 310,000 people filed for benefits, then that original report would fall into the 9-10k error bucket. A positive error means that the original report under-reported the true number of unemployment claims, and vice versa for a negative error. The more accurate the original reporting, the more the errors should be at or near the 0 mark on the horizontal axis, and the less biased the original reporting, the more the error bars should look like a bell curve centered on the 0 mark. In this chart, 36% of the original reports were error-free (no difference between the revised and original reports) and 47% of the original reports under-reported the actual unemployment claims by 1,000 to 2,000 people. There are a few weeks with slight over-reporting of unemployment claims and a few more weeks of more pronounced under-reporting of unemployment claims, but by and large this is a picture of a reasonably accurate and only slightly biased data report.
And now here’s the frequency distribution of the errors made in the reporting of weekly unemployment claims during the Obama Administration from September 30th, 2009 through the 2012 election.
I mean … I could go through all the statistics that show how accuracy declined and bias increased in the weekly unemployment claim reports from the Bush Administration to the Obama Administration, but this is a good example of how one picture is worth a thousand words.
In the period 9/30/2009 through the November 2012 election, original Labor Dept. reports understated initial jobless claims by 858,000 relative to initial revisions. Compared to final revisions, the original estimates look even worse, understating jobless claims by 884,000. In the period 9/30/2005 through the November 2008 election, on the other hand, original Labor Dept. reports understated jobless claims by 292,000 relative to initial revisions. Compared to final revisions, the original reports understated jobless claims by only 5,000 people! Were there more initial jobless claims in the Obama time period than in the Bush time period? Yes, but only approximately 25% more. The total errors in initial revisions, on the other hand, increased almost 300% over the comparable time periods (and almost 180x versus final revisions!), and the skew towards under-reporting actual claims is even more pronounced.
The systematic error and bias is almost certainly due to the same bureaucratic procedure that caused the most recent snafu – when individual states report incomplete or stupidly estimated data, the Labor Dept. takes the reports at face value and makes no effort to validate or correct them. The error and bias are obvious to the BLS, which is staffed with very smart people, and they could easily fix it if they wanted to. The BLS adjusts raw data all the time, and there are obvious statistical adjustments that could be applied to this obvious systematic error. But the BLS chooses not to fix it because the systematic bias favors the incumbent Administration and is embedded in standard operating procedures. Why make bureaucratic waves to fix this broken data series when turning a blind eye to its structural flaws works in your favor? Only now with this very public embarrassment and no re-election campaign ahead does Keith Hall, a former BLS commissioner, and “several” unnamed economists (i.e., current BLS commissioners) tell the WSJ that the methodology and bureaucratic oversight need to be improved.
As intriguing as this data and shifting bureaucratic Narrative might be from a political perspective (let’s just say that Jack Welch was perhaps not totally wrong in his well-publicized comments regarding the suspicious nature of labor data prior to the 2012 election), there’s a still more interesting question here from an Epsilon Theory perspective. How did this tertiary data series … a labor statistic that maybe 1 in 1,000 professional investors was aware of pre-crisis … a labor statistic that the Bureau of Labor Statistics deemed too minor to administer (!) … come to be the subject of such incredible hand-wringing and scrutiny. Why are we paying such close attention to weekly initial unemployment claims? What does it mean for today’s market that this report is treated as a useful informational signal?
Part of the answer is wrapped up in Fed communications policy and its explicit linkage of monetary policy to labor conditions. I’ve written about this at length in “2 Fast 2 Furious”, and the linkage explains both a generic market and media interest in all things labor-related as well as a very specific interest in the labor statistics that the Fed has noted in its communications. But those communications have never included weekly initial unemployment claims data as a meaningful signal, and it is inconceivable that the Fed – which is staffed with even smarter people than the BLS – would give this data series any credence or weight whatsoever, except in the very broadest sense. Moreover, there are plenty of indications of labor conditions that we know the Fed cares about – the quality or permanence of new jobs, for example – that do not receive one iota of the media attention given to weekly initial unemployment claims.
Here’s the outstanding quality of the initial unemployment claims data from a media presentation perspective – it’s weekly. The data release occurs precisely at 8:30 AM ET every Thursday. With its patina of labor-relevance it’s not an obviously stupid subject to highlight, and with its predictable timing and frequent occurrence it’s tailor-made for media scheduling. The media treatment of weekly initial unemployment claims is representative of the dog race-ification of capital markets, where artificial “events” are established and promoted in hopes of generating investor attention and activity.
In dog racing, there is no meaning to the activity itself other than to provide a context for gambling on the outcome of some pseudo-analyzable event. College football games would still take place even if no one gambled on them, but that’s not the case for dog races. They exist only to excite a gambling animus, and everything about the creation and production of the race is geared to that end. Dog racing is an artificial and constructed practice that fills the coffers of the state Treasuries and the pockets of the Paul Bryant, Jr.’s of the world by providing a regularly scheduled and officially sanctioned venue for the satisfaction of powerful human desires. So, too, are many CNBC segments and WSJ articles.
The weekly unemployment claims number is probably the most egregiously artificial or constructed data series, but it is by no means unique. For example, within the broad category of housing conditions we now have enormous attention paid to weekly data on mortgage applications, as well as monthly data on housing starts, housing permits, new home sales, existing home sales, new home prices, existing home prices, Case-Shiller, and homebuilder sentiment. In general, there has been a pronounced increase in the level of financial media attention paid to regularly scheduled macro data releases – both US and non-US – over the past few years, and it’s a trend that shows no signs of abating. Whether it’s Jobs Friday™ (discussed extensively in “What We’ve Got Here … Is Failure to Communicate”) or Bloomberg’s latest branding effort around consumer sentiment, these macro “signals” are being created and promoted at an accelerating pace, to the benefit of any institution that relies on investment activity, regardless of what that activity is. Those institutions, of course, include every sell-side and media firm, not to mention most political and regulatory bodies. The fact is, once you start looking for artifice in the data signals we are bombarded with, you will find it everywhere.
The point of this note and its identification of the constructed nature of so much of what passes for “data” these days is NOT to encourage you to ignore the reports. This is, unfortunately, what it means to live in a Common Knowledge world – even if you privately believe that much of what you hear is hokum, it is necessary and rational to act as if these are meaningful signals so long as everyone thinks that everyone thinks the signals have meaning. The point is to encourage you to recognize these data reports for what they are, to call them by their proper name, so that you limit your reaction to a rational Common Knowledge response rather than make the mistake of believing that these reports are any more analyzable in a fundamental fashion than the $5 tout sheet at the dog track in Greene County, Alabama. And the next time you hear a CNBC anchor breathlessly describe the importance of the upcoming data report, complete with a clock ticking down the seconds to its release, I hope you hear that broadcast for what it’s really saying – Heeere comes Lucky!