zero marginal product

Zero marginal product workers

Matt Yglesias suggests the notion is implausible, but I am surprised to read those words.  Keep in mind, we have had a recovery in output, but not in employment.  That means a smaller number of laborers are working, but we are producing as much as before.  As a simple first cut, how should we measure the marginal product of those now laid-off workers?  I would start with the number zero.  If a restored level of output wouldn't count as evidence for the zero marginal product hypothesis, what would?  If I ran a business, fired ten people, and output didn't go down, might I start by asking whether those people produced anything useful?

It is true that the ceteris are not paribus,  But the observed changes if anything favor the hypothesis of zero marginal product. There has been no major technological breakthrough in the meantime.  If anything, there has been bad monetary policy and a dose of regulatory uncertainty.  And yet again we can produce just as much without those workers.  Think of "labor hoarding" yet without…the hoarding.

You might cite oligopoly models and argue that the workers can produce something, but firms won't hire them because they don't want to expand output, due to lack of demand.  That doesn't seem to explain that output has recovered and that profits are high.  And since there is plenty of corporate cash, it is hard to claim that liquidity constraints are preventing the reemployment of those workers.

There is another striking fact about the recession, namely that unemployment is quite low for highly educated workers but about sixteen percent for the less educated workers with no high school degree.  (When it comes to income groups, the lowest decile has an unemployment rate of over thirty percent, while it is three percent for the highest decile; I'm not sure of the time horizon for that income measure.)  This is consistent with the zero marginal product hypothesis, and yet few analysts ask whether their preferred explanation for unemployment addresses this pattern.

Garett Jones suggests that many unemployed workers are potentially productive, but that businesses do not, at this moment, want to invest in future productive improvements.  The workers only appear to have zero marginal product, because their marginal product lies in future returns not current returns.  I see this hypothesis as part of the picture, although I am not sure it explains why current unemployment is so much higher among the unskilled.  Is unskilled labor the fundamental capability-builder for the future?  I'm not so sure.

It's also interesting to look at the composition of the long-term unemployed (not the same as the composition of all the unemployed, of course).  Older workers with a college education are quite stuck, conditional on their being unemployed.  And in this group, more education predicts a longer spell of unemployment.  Is this ongoing "recalculation," optimal search theory, or is the roulette wheel simply coming up zero each time it is spun for these workers?  Maybe a bit of each.  If you want, call some of it age discrimination and relabel the idea "perceived zero marginal product."

In general, which hypotheses predict lots more short-term unemployment among the less educated, but among the long-term unemployed, a disproportionately high degree of older, more educated people?  This stylized fact seems to point toward search and recalculation ideas, with some zero marginal products tossed in.  Do aggregate demand theories yield that same data-matching prediction?  I don't see it, at least not without being paired with a theory of concomitant real shocks.

Nothing in the zero marginal product hypothesis requires that these marginal products be zero forever.  As the entire economy expands more rapidly (when will that happen?), the value of even a low quality worker can quickly become much higher.  If you are opening up a new building, suddenly you really need that extra janitor and he is indeed more productive at the new margin.

Some people identify the zero marginal product hypothesis with the "hopeless dregs of the earth" description, but the two are not necessarily the same.  Complementarity, combined with some fixed initial factors, can yield zero or near-zero marginal products of labor.  (You'll see the phrase "excess capacity" used in this context, though that matches the oligopoly hypothesis more closely.)  The "dregs of the earth" view is pessimistic, but the complementarity version of the zero marginal product idea can be quite optimistic, predicting a very rapid recovery in the labor market, once the interactions turn positive. 

The "dregs" and the "complementarities" views also have different policy recommendations.  The dregs view implies either hopelessness or a lot of fundamental retraining or ongoing assistance, while the complementarity view leads one to ask how we might mobilize positive complementarities (rather than leaving orphaned factors of production) more quickly.  Perhaps there are some fixed factors, such as managerial oversight, and entrepreneurs do not want to strain those fixed factors too hard.  How can we make such fixed factors more replicable or more flexible?

Addendum: Arnold Kling comments.

Productivity and market power in general equilibrium

Let’s assume away cyclical factors, so full employment always holds.

If every sector of an economy becomes monopolistic, output will contract in each sector, and it might appear that productivity will decline.  But for the most part this output reduction will not be achieved by burning crops in the fields.  Rather, less will be produced and factors of production will be freed up for elsewhere.  New sectors will arise, and offer goods and services too, perhaps with monopolies as well.  In any case, the consumer surplus gains from those new sectors might be especially high, because they will be selling to the highest points on demand curves that previously were unsatisfied altogether.

You can cite the deadweight loss of monopoly all you want, but we’re getting more outputs of other stuff.  Value-added could be either higher or lower, productivity too.

The Schumpeterian tradition, of course, suggested that market power would boost innovation.  There are at least two first-order effects pushing in this direction.  First, the monopoly has more “free cash” for R&D, and second there is a lower chance of the innovation benefiting competing firms too.  I don’t view the “monopoly boosts innovation” hypothesis as confirmed, but it probably has commanded slightly more sympathy from researchers than the opposite point of view.  Bell Labs did pretty well.  In any case, partial equilibrium output restrictions won’t get you to any kind of smooth conclusion about monopoly causing low productivity growth.

Let’s say instead that every sector is ruled by monopolistic competition, a’la Chamberlin.  In the model, that puts firm production at points on the AC curves above minimum AC for each firm or sector.  You might consider that to be a productivity problem, though of course you must compare it to the rise in product diversity that follows from monopolistic competition.

But under those same conditions, profits are zero and so the mark-up arguments from the DeLoeker and Eeckhout paper do not apply and indeed cannot hold.

I find most of what is written on monopoly and productivity these days to be under-theorized.

CEOs and Marginal Product

I think Tyler is being unnecessarily opaque in his two posts (here and here) on CEO pay and marginal product.

Contra Tyler, it is, of course, perfectly reasonable to compare a CEO with his or her likely replacement. Board of Directors do this all the time. If the Board finds that revenues would be the same with a new CEO the theory doesn’t say that the current CEO should get a zero wage (as Tyler oddly suggests). It says that the current CEO should not be paid more than the wage necessary to hire the replacement. If the current CEO is paid more, he may be out a job. If the current CEO is paid less, he may move.

Tyler argues that CEOs are paid far too little (less than 1% of their marginal product). I say that as a result of the process described above CEOs are paid more or less their marginal product on average. Indeed, according to one astute writer, Gabaix and Landier find that replacing “replacing the No. 250 chief executive with the No. 1 will increase the value of the company by only 0.014 percent.” More or less means that principal agent problems, risk aversion and uncertainty also matter but they matter within a range determined by the usual process. I also believe that the case for some CEOs being overpaid because they choose friendly board members is far stronger than Tyler’s case that CEOs on average are radically underpaid.

Addendum: Tyler’s points about tax incidence are well taken and need not rely on an underpayment argument.

What is the marginal product of a CEO?

If you read carefully my post from yesterday, you may have noticed what a tricky question this is. “CEO” of course is a discrete position, and while there are companies with “zero” or “two” CEOs, those comparisons are not the correct ones to define marginal product; in any case they would give you two very different numbers.

Nor is it correct to compare “this CEO” to “his likely replacement.”  That difference could be zero, but it does not mean the CEO adds zero value or will or should receive zero pay.  Keep in mind that we are already juggling a few margins here, including “getting this CEO to work harder or better” and “this CEO vs. another.”

Alternatively, imagine there are ten firms in the economy, of differing size and import, all bidding for managers in a pool of fifty people.  A credible CEO offer has to satisfy a participation constraint, namely getting the candidate to take the job over CEO at a lesser firm or working in a lesser job.  But if a CEO can add 50 percent of value to a firm, that CEO will not in general be paid fifty percent of the firm’s value and need not be paid anything close to that.  The shareholders know he will take the job for less and there are other candidates who might add forty-seven percent of value.  The firm can make a credible offer of “two percent of value added” and it might be accepted.

Unlike hiring widget-makers for “less than their marginal product,” there is no subsequent disruption of equilibrium which must follow from this apparent disjunction of CEO pay and marginal product.  For instance there is no firm-level incentive to further expand output or hire extra CEOs.  There is one discrete slot, a wide range of potential compensation values, and the final sum is set by a bargain, determined by the context of principal-agent theory.

CEOS who can add so much value will try to start and grow their own firms, holding lots of equity from the very beginning, as Mark Zuckerberg has done.  Those CEOs will indeed be paid something like their traditional marginal product, but they are a distinct minority and wealth and risk constraints limit their number.

In general, it is confusing to suggest that CEOs will be paid their marginal product.  The traditional notion of marginal product does not apply to a CEO in the simple “widgets per worker” way.  There are ways you can define “marginal product” to make the claim “CEOs are paid their marginal product” more or less true, but that is not my preferred way forward.  Instead we should get more used to thinking intuitively within the principal-agent model, even though it is harder to do.

Sumner and Krugman on zero MP workers

Scott's post is here, Krugman's is here.  (My first post on the topic is here, my last post here).  Let's start with Scott, excerpt:

This post by Stephen Gordon shows US employment in 2010:3 falling about 5% below its 2008:1 peak, while output seems to have declined only about 0.7%.  This is what Cowen finds puzzling. 

But I don’t see any puzzle at all.  If employment didn’t change, I’d expect US output to grow at about 2% a year, which is the trend rate of productivity growth.  Because we are looking at a two and a half year period, you’d expect output to grow roughly 5% with stable employment.  Now assume that employment actually fell 5%.  If the workers who lost jobs were similar to those who remained employed, I’d expect output to be flat over that 2.5 year period.  Because output fell slightly, it seems like the workers who lost jobs were slightly more productive than those who remain employed.

Do I believe this?  No, for several reasons I think they were less skilled than those who remained employed.  Labor productivity growth (assuming we were at full employment) probably slowed in the most recent 2.5 years, as investment in new capital declined.  Measured productivity continued to rise briskly, partly because technological progress continues in good times and bad, and partly because those workers still employed are somewhat higher skilled, or perhaps are trying harder in fear of losing their own jobs.  So Tyler is probably right that those workers who lost their jobs have a lower than average marginal product.  I just don’t see why zero is the natural starting point for consideration of the issue, as you only get that number by making some fairly extreme assumptions about technological progress coming to a screeching halt after 2008:

A few points:

1. The claim is that some workers have zero marginal product (net of employment costs), not all workers (as Krugman inexplicably ponders for two full paragraphs), and not even all unemployed workers.  Just as a hypothetical example, if unemployment is 9.5 percent and the natural rate is 5.5 and zero MP helps account for half of that difference, that's two percent of the labor force at zero MP.  Try hiring labor for a while and see how crazy that sounds.

2. Zero MP is a property which may hold in an AS-AD equilibrium, it is not a substitute for an AS-AD view.  Krugman mischaracterizes the hypothesis here, by identifying it with "AD denialism."  In my post which Krugman links to (and indeed for years), I've made it clear that demand matters in any coherent account of the equilibrium.  

Oddly, Krugman himself stated one of the coherent versions of the Zero MP view in July 2010, and then he considered it possible and appropriately, he expressed uncertainty about what might be going on.

3. The Zero MP hypothesis is simply another way of talking about "labor hoarding," a well-known and time-honored idea, except that the labor is not in fact hoarded.  It doesn't encounter strange paradoxes and it is more intuitive than when the labor is hoarded (which appears to violate first-order conditions).  There is plenty of very specific and indeed striking evidence that the "previously hoarded labor" isn't being hoarded any more.  That same link implies that Krugman's invocation of 1983 is a red herring and probably not a good instance for finding many zero MP workers.  The "oughties" job market differs in a number of other "real" ways from the 1980s, including the fact that we've had no net job growth over the last decade, not even pre-crisis.  Here is further evidence on how productivity patterns were different during the early 1980s.

4. For another take on #3, during the job-destroying periods of 2009, per hour labor productivity growth is rising at astonishing rates, try 3.4, 8.4, 7, and 6 percent, each quarter, annualized.  That's not just the regular accretion of technological progress (though some of it may be), it is an artifact from dumping lower quality and zero MP workers.  If I look in the second quarter and see labor hours go down 7.9 percent and see per hour productivity rise 8.4 percent, well, that's no proof but I sure go hmm….

5. I would not take it for granted that "normal" productivity growth continues in times of shock and crisis.  Maybe yes, maybe no. Scott's talk of the "trend rate" is assuming that the growth and cyclical components are separable and that is begging part of the question.

6. The zero MP hypothesis helps explain why unemployment is so much more severe among the less educated and the lower earners.  In contrast, Krugman writes: "As Mike Konczal points out, basically everyone’s unemployment rate has doubled, no matter their education level or location."  In reality, that kind of multiplicative relationship is very much consistent with joint AS-AD determination, including a zero MP for many workers in the equilibrium.  I'll write an entire blog post on that question soon, and then we'll see that this result actually discriminates against pure AD theories or is at best a neutral pointer.

7. What does the zero MP hypothesis add?  First, the zero MP hypothesis explains why wage adjustments can't do the trick for a lot of the unemployed, as wages won't fall below zero.  Second, the zero MP hypothesis explains why you need steady real growth, boosting the entire chain of demand, to reemploy lots of workers and reflation alone won't do the trick.  (I still, by the way, favor reflation because I think it will do some good.)  Those predictions are not looking terrible these days.

8. The AD-only theories, taken alone, encounter major and indeed worsening problems with the data.  Year-to-year, industrial output is up almost six percent, sales up more than six percent, but the labor market has barely improved.  How does that square with the AD-only hypothesis?  Has it been seriously addressed?  Arnold Kling also has relevant comments and in passing I'll note that the "increases in the risk premium" factor is being neglected in this discussion of Kling's points. 

9. Krugman (not Scott), in particular, is proposing an alternative view with a) upward-sloping AD, b) downward-sloping AS, c) the implication that huge boosts in the minimum wage would restore the economy and employment, and d) requires a tight liquidity trap when the theoretical literature distinguishes between "interest rates literally at zero" and "interest rates near zero."  His comparison of ZIRP and ZMP does not raise those issues on the other side of the ledger.

I think Scott is underplaying some of the more detailed facts about labor markets, such as mentioned in #3, #4, #6, and #8.  Krugman simply isn't considering the stronger versions of the zero MP hypothesis and thus he is dismissive rather than confronting the very real problems in the AD-only point of view.

Addendum: Arnold Kling has more.

Does marginal cost equal price?

We’re all taught that in a competitive industry price will equal marginal cost.  Well, what is a competitive industry?  There are lots of Chinese restaurants in or near Fairfax, and with a few noble exceptions they have more or less the same menu.  Each could serve an extra diner at essentially zero marginal cost, yet the price of the food is not zero.  Not even marginal meals are given away for free, except perhaps to the staff.  If price is equal to marginal cost, we have to ask equal to which marginal cost?  The marginal cost of one more Kung Pao Chicken?  The marginal cost of being known for giving some meals away?  The marginal cost of possibly setting off destructive price competition with rivals?  The concept of marginal cost relies on a definition of time horizon, strategic assumptions, and the counterfactual against which real world action is being compared.  Yikes.

Armen Alchian and Fischer Black are the guys to read on what cost really means (Buchanan and the Austrians only get you so far).  If you really want to get dizzy read Lester Telser on when there is a core, and wonder whether the industry you have in mind meets his screwy but essentially correct standards for MC, AC, and no coherent equilibrium.  It’s not just the airlines.  So when is price equal to marginal cost, average cost, or some blend of the two?  And which definitions of average and marginal cost? 

What about "reality"?  Toss a bone to social frictions, then ask for some micro-studies of how "competitive" industries price in the short run.  Use interviews and ethnography to supplement the formal models.  In practical terms, you might end up with some understanding of a) why prices can be sticky in apparently competitive industries, and b) why few businessmen — including high IQ types — will admit to pricing at marginal cost or even understand what that means.

The bottom line: I’ll say that MC is flat if truly all inputs are replicated.  But that’s never the case, so MC is usually zero under one set of counterfactuals and sloping upward dramatically under another set.  That’s not the end of the world, live with it.

The second bottom line: When it comes to teaching the students, just tell them that marginal cost slopes upward at some point.  After all, sooner or later they all stop studying.

The third bottom line: #13 out of 50.

The marginal product of NBA players

Most of you have read Moneyball by now, so why not measure the marginal products of NBA players as well? After all, playoff time and the MVP award are just around the corner. Wayne Winston, professor of decision sciences at Indiana University, has tried to crack the numbers:

“Basketball’s a team sport, and lots of things aren’t tracked,” Winston says. “Like taking the charge, going through a screen, tipping a ball to your teammate, saving a ball from going out of bounds. That’s where our system comes in. All these little things should translate into points.”
One problem: Traditional plus-minus systems tend to overrate average players on good teams and underrate good players on lousy ones. After all, a zero plus-minus rating on the Los Angeles Lakers is not the same as a zero rating on the Los Angeles Clippers, mostly because one team has Kobe Bryant and Shaquille O’Neal and the other has Marko Jaric and Chris Kaman.
To compensate, Winval’s ratings are weighted to take into account every other player on the floor. For every time segment a player is in a game, the system tracks the other nine players on the floor, the length of the segment and the score at the start and end of the segment.”

In other words, he tries to measure marginal product in terms of points, adjusting for the values of the other players. The system is called Winval, here is the article, the paper version has more information. And who comes out on top? Please sit down, the five best players in the NBA, according to this measure are:

1. Hedo Turkoglu (who? he plays for San Antonio but doesn’t even start)
2. Vince Carter (a well-known star, but universally considered soft and a choker)
3. Kevin Garnett (the likely MVP for this year)
4. Brad Miller (very good player, but not elite)
5. Manu Ginobili (very good player, perhaps headed toward elite status)

Shaq, Kobe, and Tim Duncan are not in the top ten. None of the five, except for Garnett, crack a recent USA Today straw poll for NBA mvp. (By the way, here is last year’s Winval list.) Now maybe these rankings are right and conventional wisdom is all wet. Marginal product is, well…marginal product. But what are some other options?

1. A player with a high rank could have a really bad replacement, thus boosting his measured marginal impact.
2. Some players are “momentum” players, they are put in when the tide is turning.
3. Some players are wonderful for the time they play but could not keep it up for the whole game. They look great when you see them, but they are not worth very much overall.
4. The econometric model is somehow misspecified, but of course you can always say this.

The bottom line: I’m still puzzled by how much the measurements diverge from common sense. The NBA offers gobs and gobs of measurable information. Yet intuition remains indispensable when we try to estimate marginal product. By the way, Winval predicts that Mitchell Butler (who?, 13.4 minutes per game) is the best Washington Wizard.

Two misunderstood movies, two Rorschach tests (not too many spoilers here)

American Sniper is one of the best anti-war movies I have seen, ever.  But it shows the sniper-assassin, and his killing, to be sexy, and to be regarded as sexy by women, while the rest of war is dull and stupid.  (Even the two enemy snipers are quite attractive and fantastic figures, and there is a deliberate parallel between the family life of the Syrian sniper and the American protagonist.  The klutziness of the non-assassin soldiers limited how many African-Americans and Hispanics they were willing to cast in those roles, as it is easiest to make white guys look crass in this way without causing offense.)  By making the attractions of war palpable, this film disturbs and confuses people and also occasions some of the worst critical reviews I have read.  It also, by understanding and then dissecting the attractions of blood lust, becomes a quite convincing anti-war movie, if you doubt this spend a few months studying The Iliad.  (By the way, Clint Eastwood, the director and producer, describes the movie as anti-war.)  The murder scenes create an almost unbearable tension, the sandstorm is a metaphor for our collective fog, and they had the stones to opt for the emotional overkill of four rather than just three tours of duty.  Iraq is presented as a hopeless wasteland with nothing of value or relevance to the United States, and at the end of the story America proves its own worst enemy.  It is not clear who ever gets over having killed and fought in a war (can anything else be so gripping?…neither family life nor sex…), even when appearances suggest a kind of normality has returned.  The generational cycle is in any case replenished.  I say A or A+, both as a movie and as a Rorschach test.

Two Days, One Night has some of the worst economics I have seen in a movie, ever.  It would be brilliant as a kind of Randian (or for that matter Keynesian) meta-critique of the screwed up nature of Belgian labor markets and social norms, and most of all a critique of the inability of the Belgian intelligentsia to understand this, except it is not.  It is meant as a straight-up plea for sympathy for the victim and as such it fails miserably, even though as a movie it embodies reasonably good production values.  Everything in the workplace of this solar power company is zero-sum across the workers and we never see why.  The protagonist campaigns to get her job back, but never asks or even considers how she might improve her productivity or attitude, asking only on the basis of need.  (And she is turned down only on the basis of need.)  At one point her employer states the zero marginal product hypothesis quite precisely, something like “when you took time off, we saw that sixteen people could do the work of seventeen.”  She never asks if there might be some other way she could contribute — but she does need the money — nor does the notion of a better job match somewhere else rear its head.  The depictions of financial hardship confuse wealth and income, basic survival and discretionary spending.  The rave reviews this movie has received represent yet another Rorschach test and one which virtually every commentator seems to have failed.

Assorted links

1. Further evidence for political insider trading, this time for hedge funds.

2. You call it “3-D printing optimism,” I call it “order transmission and cheap delivery pessimism.”  Amazon enters the game.

3. Google search and stock prices, further results.

4. Indian monkey costume markets in everything, Parliamentary edition.  And lingerie for Chinese peaches.

5. China will liberalize its internal passport system.  And The Woman in Black settles in Winchester,VA.

6. Broody octopus keeps four-year vigil.

7. In which we learn that Neel Kashkari is very good at impersonating a zero marginal product worker.  That skill usually does not operate in reverse.

8. Brad DeLong on Jeff Faux and NAFTA.  I mostly agree, but also think rural Mexican corn farmers need to be leaving that sector in any case.

Are the Long-Term Unemployed on the Margins of the Labor Market?

There is new Brookings research by Alan B. Krueger, Judd Cramer, and David Cho:

The short-term unemployment rate is a much stronger predictor of inflation and real wage growth than the overall unemployment rate in the U.S. Even in good times, the long-term unemployed are on the margins of the labor market, with diminished job prospects and high labor force withdrawal rates, and as a result they exert little pressure on wage growth or inflation.

Consistent with my earlier views, this work is suggesting that many of the long-term unemployed are/have become an economically segmented group.  This is noteworthy too, as it implies the problem is not merely initial discrimination:

…even after finding another job, reemployment does not fully reset the clock for the long-term unemployed, who are frequently jobless again soon after they gain reemployment: only 11 percent of those who were long-term unemployed in a given month returned to steady, full-time employment a year later.

I would consider that evidence for a notion of zero marginal product workers.  Furthermore, in my view (I am not speaking for the authors here), right now further inflation is as likely to harm as to help these individuals.  To ask whether the Fed “should give up” on the long-term unemployed is a biased framing which is more likely to mislead us than anything else.

There is a good piece up at 538:

Krueger and his coauthors, Princeton economists Judd Cramer and David Cho, find evidence that the long-term unemployed aren’t getting jobs even in parts of the country where the job market is comparatively healthy, suggesting that a stronger economic rebound won’t be enough to put them back to work.

Wassily Leontief and Larry Summers on technological unemployment

Here is a very interesting piece from 1983 (jstor), Population and Development Review, it is called “Technological Advance, Economic Growth, and the Distribution of Income,” here is one excerpt:

In populous, poor, less developed countries, technological unemployment has existed for a long time under the name of “disguised agricultural unemployment”; in Bangladesh, for instance, there are more people on the land than are needed to cultivate it on the basis of any available technology.  Industrialization is counted upon by the governments of most of these countries to relieve the situation by providing — as it did in the past — much additional employment.

If I may put this into my own terminology, Leontief is suggesting that at some margins fixed proportions mean many agricultural laborers, or would-be laborers, are ZMP or zero marginal product.

Haven’t you ever wondered how some traditional economies can have unemployment rates which are so high?  Those are “structural” problems, yes, but of what kind?

By the way, Brad DeLong cites Larry Summers on ZMP workers:

My friend and coauthor Larry Summers touched on this a year and a bit ago when he was here giving the Wildavski lecture. He was talking about the extraordinary decline in American labor force participation even among prime-aged males–that a surprisingly large chunk of our male population is now in the position where there is nothing that people can think of for them to do that is useful enough to cover the costs of making sure that they actually do it correctly, and don’t break the stuff and subtract value when they are supposed to be adding to it.

What are humans still good for? The turning point in Freestyle chess may be approaching

Some of you will know that Average is Over contains an extensive discussion of “freestyle chess,” where humans can use any and all tools available — most of all computers and computer programs — to play the best chess game possible.  The book also notes that “man plus computer” is a stronger player than “computer alone,” at least provided the human knows what he is doing.  You will find a similar claim from Brynjolfsson and McAfee.

Computer chess expert Kenneth W. Regan has compiled extensive data on this question, and you will see that a striking percentage of the best or most accurate chess games of all time have been played by man-machine pairs.  Ken’s explanations are a bit dense for those who don’t already know chess, computer chess, Freestyle and its lingo, but yes that is what he finds, click on the links in his link for confirmation.  In this list for instance the Freestyle teams do very very well.

Average is Over also raised the possibility that, fairly soon, the computer programs might be good enough that adding the human to the computer doesn’t bring any advantage.  (That’s been the case in checkers for some while, as that game is fully solved.)  I therefore was very interested in this discussion at RybkaForum suggesting that already might be the case, although only recently.

Think about why such a flip might be in the works, even though chess is far from fully solved.  The “human plus computer” can add value to “the computer alone” in a few ways:

1. The human may in selective cases prune variations better than the computer alone, and thus improve where the computer searches for better moves and how the computer uses its time.

2. The human can see where different chess-playing programs disagree, and then ask the programs to look more closely at those variations, to get a leg up against the computer playing alone (of course this is a subset of #1).  This is a biggie, and it is also a profound way of thinking about how humans will add insight to computer programs for a long time to come, usually overlooked by those who think all jobs will disappear.

3. The human may be better at time management, and can tell the program when to spend more or less time on a move.  “Come on, Rybka, just recapture the damned knight!”  Haven’t we all said that at some point or another?  I’ve never regretted pressing the “Move Now” button on my program.

4. The human knows the “opening book” of the computer program he/she is playing against, and can prepare a trap in advance for the computer to walk into, although of course advanced programs can to some extent “randomize” at the opening level of the game.

Insofar as the above RybkaForum thread has a consensus, it is that most of these advantages have not gone away.  But the “human plus computer” needs time to improve on the computer alone, and at sufficiently fast time controls the human attempts to improve on the computer may simply amount to noise or may even be harmful, given the possibility of human error.  Some commentators suggest that at ninety minutes per game the humans are no longer adding value to the human-computer team, whereas they do add value when the time frame is say one day per move (“correspondence chess,” as it is called in this context.)  Circa 2008, at ninety minutes per game, the best human-computer teams were better than the computer programs alone.  But 2013 or 2014 may be another story.  And clearly at, say, thirty or sixty seconds a game the human hasn’t been able to add value to the computer for some time now.

Note that as the computer programs get better, some of these potential listed advantages, such as #1, #3, and #4 become harder to exploit.  #2 — seeing where different programs disagree — does not necessarily become harder to exploit for advantage, although the human (often, not always) has to look deeper and deeper to find serious disagreement among the best programs.  Furthermore the ultimate human sense of “in the final analysis, which program to trust” is harder to intuit, the closer the different programs are to perfection.  (In contrast, the human sense of which program to trust is more acute when different programs have more readily recognizable stylistic flaws, as was the case in the past: “Oh, Deep Blue doesn’t always understand blocked pawn formations very well.”  Or “Fritz is better in the endgame.”  And so on.)

These propositions all require more systematic testing, of course.  In any case it is interesting to observe an approach to the flip point, where even the most talented humans move from being very real contributors to being strictly zero marginal product.  Or negative marginal product, as the case may be.

And of course this has implications for more traditional labor markets as well.  You might train to help a computer program read medical scans, and for thirteen years add real value with your intuition and your ability to revise the computer’s mistakes or at least to get the doctor to take a closer look.  But it takes more and more time for you to improve on the computer each year.  And then one day…poof!  ZMP for you.

Addendum: Here is an article on computer dominance in rock-paper-scissors.  This source claims freestyle does not beat the machine in poker.

ZMP workers and morale externalities

On Twitter, Bryan Caplan asks me to clarify why zero marginal product workers do not clash with the notion of comparative advantage.  The point is simple: some workers destroy a lot of morale in the workplace and so the employer doesn’t want them around at any price.

Most of us buy into “morale costs” as a key reason behind sticky nominal wages.  If your wage is too low, your morale falls, you produce less and so the wage cut isn’t worth it.  Well, what else besides low wages makes people unhappy in their workplace?  Very often the quality of co-workers is a major source of unhappiness; just listen to people complain about their jobs and write down how many times they are mentioning co-workers and bosses.  (I do not exempt academics here.)  A “rotten apple” can make many people less productive, and you can think of that as a simple extension of sticky nominal wage theory, namely that installing or tolerating a “pain in the ass” is another way of cutting wages for the good workers, they don’t like it, it lowers their productivity, and thus it is not worth tolerating the rotten apple if said apple can be identified and dismissed.

There is no particular reason to think that ZMP workers are especially stupid or in some way “disabled.”  If anything it may require some special “skills” to get under people’s skins so much.  (Of course there are some individuals who, say for health reasons, cannot produce anything at all but they are not usually in or “near” the active labor force.)  To draw a simple analogy, the lowest-publishing members of academic departments are rarely those who make the most trouble.

To the extent production becomes more complex and more profitable, ZMP workers are more of a problem because there is more value they can destroy.  The relevance of these morale costs also varies cyclically, in standard fashion.  A company is more likely to tolerate a “pain” in boom times when the labor itself has a higher return.

Note also the “expected ZMP worker.”  Let’s say that some ZMPers destroy a lot of value (that makes them NMPers).  You pay 40k a year and you end up with a worker who destroys 80k a year, so the firm is out 120k net.  Bosses really want to avoid these employees.  Furthermore let’s say that a plague of these destructive workers hangs out in the pool of the long-term unemployed, but they constitute only 1/3 of that pool, though they cannot easily be distinguished at the interview stage.  1/3 a chance of getting a minus 120k return will scare a lot of employers away from the entire pool.  The employers are behaving rationally, yet it can be said that “there is nothing wrong with most of the long-term unemployed.”  And still they can’t get jobs and still nominally eroding the level of wages won’t help them.

In the perceived, statistical, expected value sense, the lot of these workers is that of ZMPers.

One policy implication is that it should become legally easier to offer a very negative recommendation for a former employee.  That makes it easier to break the pooling equilibrium.  There also are equilibria where it makes sense to “buy the NMPers out” of workforce participation altogether, pay them to emigrate, etc., although such policies may be difficult to implement.  Oddly, if work disincentives target just the right group of people — the NMPers — (again, hard to do, but worth considering the logic of the argument) those disincentives can raise the employment/population ratio, at least in theory.

Addendum: Garett Jones offers yet a differing option for understanding ZMP theories.