Bayesian updating

Here are some interesting articles that I ran across.  First one from the Economist on the real (actual) minimum wage—zero:

Perhaps not coincidentally, the number of unpaid internships has grown just as hiring has become riskier, pricier and more complex. In recent years anti-discrimination and unfair-dismissal rules have been tightened, and minimum wages raised, in many rich countries. The growing cost of benefits such as pensions, health care and maternity leave makes employees more expensive. Interns have therefore become an appealing alternative.

I’ve recently argued that CEOs should be paid far more than they were paid in the 1960s, because their jobs are much more consequential.  Here’s evidence that they are not overpaid:

So, if the shares rise on an executive’s death, that means he was overpaid; if they fall, he was not. By this measure only 42% of the bosses studied were overpaid; furthermore, those with the most eye-popping rewards were found to be giving the best value for money, as measured by the share-price slump when they passed away.

The study also reckons that of the increase in value that results from a firm hiring an executive, he gets 71% and the shareholders therefore get 29%. In the sense that investors at least get some positive reward from the relationship, executives as a whole are not overpaid.

Followers of Mr Piketty are unlikely to be convinced. They would say that even when bosses add more value than the amount by which their pay exceeds the average, they are still overpaid because the average is itself excessive; and that it is inherently indecent for bosses to get such a big share of the gains from their relationship with their firms.

Yeah, they would say that, wouldn’t they?

Recently I did a long post at Econlog saying education wasn’t very important (at the margin in developed countries) and that spending more money wouldn’t have much impact.  I did point to one strong counterargument, a study showed that having a single good teacher at a young age can substantially impact a person’s life outcome.  I found that result quite surprising, but now it looks false:

Estimates that adjust for changes in students’ prior achievement find evidence of moderate bias in VA scores, in the middle of the range suggested by Rothstein (2009). The association between VA and long-run outcomes is not robust and quite sensitive to controls.

Also check out Bryan Caplan’s excellent post on the dubious merits of compulsory attendance laws.

Update:  Tyler Cowen cites a study that conflicts with my prior belief.  But is it scalable? And does it go beyond improving test scores, to improving life outcomes?

I’ve also argued that most anomaly studies in finance are merely data mining, and hence are essentially worthless.  Look at the last sentence in this abstract:

Hundreds of papers and hundreds of factors attempt to explain the cross-section of expected returns. Given this extensive data mining, it does not make any economic or statistical sense to use the usual significance criteria for a newly discovered factor, e.g., a t-ratio greater than 2.0. However, what hurdle should be used for current research? Our paper introduces a multiple testing framework and provides a time series of historical significance cutoffs from the first empirical tests in 1967 to today. Our new method allows for correlation among the tests as well as missing data. We also project forward 20 years assuming the rate of factor production remains similar to the experience of the last few years. The estimation of our model suggests that a newly discovered factor needs to clear a much higher hurdle, with a t-ratio greater than 3.0. Echoing a recent disturbing conclusion in the medical literature, we argue that most claimed research findings in financial economics are likely false.

I’m skeptical of proposals to “regulate” the financial system.  Here’s one example:

“DON’T bail out the big banks on Wall Street another time,” thundered Richard Durbin, an American senator, “Once in a political lifetime is enough!” His amendment to the Dodd-Frank financial reform of 2010 capped the fees banks can charge merchants to process debit-card transactions, on the grounds that banks were gouging businesses and their customers. But the limits on “interchange fees”, as the financial jargon has it, have not worked out as planned. They have resulted, by one calculation, in the transfer of between $1 billion and $3 billion annually from poor households to big retailers and their shareholders. These were not the beneficiaries Mr Durbin had in mind when the amendment came into effect three years ago this week.  .  .  .

Meanwhile the banks, which are in even worse shape, have tried to make up for the lost revenue with higher charges for other things, including monthly fees for having a debit card, or even a current account. In 2009 banks provided 76% of America’s current accounts free of charge; last year the figure was only 38%. The higher charges in turn, have pushed 1m Americans out of the formal financial system””not the result Mr Durbin was aiming for.

I predicted that Hollande’s socialist policies would fail, and he’d do a U-turn just like Mitterrand:

The new team is engineering a shift in economic policy not unlike that under Mitterrand, who made a sharp U-turn in 1983, also after two years in office. Like Mitterrand, Mr Hollande has so far spent most of his time making matters worse. Having declared during his campaign that the “world of finance” was his enemy, and promised his 75% top tax rate, Mr Hollande increased taxes by €30 billion ($40 billion) in his first year. He reversed some of Mr Sarkozy’s popular work-friendly policies, such as tax-free overtime. He sent out mixed messages to foreign investors and entrepreneurs. He failed to curb public spending. And he brought in new rules that choked growth in sectors such as construction.

On Mr Hollande’s watch, the overall tax take grew from 43.7% of GDP in 2011 to 46% in 2013. Annual income growth in 2012-14 has averaged a mere 0.4%. Unemployment, which Mr Hollande had promised to bring down, edged up to over 10%. Confidence collapsed, investment was put on hold, and many of the rich left for Brussels or London. To take but one example of the damage Mr Hollande has wrought, new rent-control rules designed by Ms Duflot (who refused to serve under Mr Valls because she considered him too right-wing) have battered the construction industry. In the two years to January 2014, new housing starts fell by nearly a quarter.

Now the government has gone into reverse. It has embraced a business-friendly mix of policies in a bid to revive the private sector. This may stop short of what the economy needs to get back on its feet, but it contains a decent dose of common sense. In 2015 a cut in the hefty social charges paid by employers will come into full effect, in an attempt to encourage hiring. Savings of €21 billion will be squeezed out of public spending, including €9.5 billion from the social-security system. Perhaps most symbolic of all, the 75% top tax rate, set up initially as a temporary two-year measure, will be quietly allowed to die.

If the Piketty/Krugman soak the rich policies won’t work under a socialist government in France, when and where will they work?

In other posts, I’ve expressed concern over eco-terrorists (think unibomber) who believe the world is overpopulated. Soon they’ll have a weapon:

Nearly 50 cities, mostly in America and Europe, are now home to groups of biohackers or amateur laboratories where they can meet and experiment. Besides Open Wetlab, these include Biocurious in Sunnyvale, California, Genspace in New York and La Paillasse in Paris. The number of biohackers around the world is anybody’s guess, but the movement’s main online-mailing list boasts nearly 4,000 members and is growing rapidly.

What drives the movement is the belief that “biology is technology” (to quote the title of a book by Rob Carlson, a DIYbio pioneer): that DNA is a form of software that can be manipulated to design biological processes and devices. But some people worry that amateur laboratories could create killer bugs or provide training for bio terrorists. For the moment, at least, such fears seem premature.

For the moment . . .

Someday I’ll change my mind, and stop being so dismissive of new theories that challenge my prior beliefs.  But not today.

HT:  Tyler Cowen



23 Responses to “Bayesian updating”

  1. Gravatar of Kevin Erdmann Kevin Erdmann
    25. October 2014 at 19:14

    Here’s a post I did on crackdowns on unpaid internships:

    Here’s one on guaranteed free compulsory education

  2. Gravatar of Andrew_M_Garland Andrew_M_Garland
    25. October 2014 at 21:16

    “If the shares rise on an executive’s death, that means he was overpaid; if they fall, he was not overpaid.”

    That is wrong. One executive salary is insignificant to profits. Further, the only savings or losses on salary would be the difference between the salary of the dead executive and his successor. How much can that be?

    The significance of an executive death or resignation is the estimate of his knowledge and ability. If he was considered much better than any known replacement, the stock goes down. If he is thought to be worse than an available replacement, the stock goes up.

    As usual, the overwhelming factor in stock price is an estimate of what the company can do in the future.

  3. Gravatar of tesc tesc
    25. October 2014 at 22:07

    So I shouldn’t go into financial econ PhD?

  4. Gravatar of Brett Brett
    25. October 2014 at 22:48

    1. That’s not really true. The unpaid internships are going heavily towards people with families rich enough to support them while they do them – the article pointed out that the family income backgrounds of people in some industries with lots of unpaid interns are creeping upwards. And those internships are at least theoretically part of a climbing process, not just menial work.

    3. Discussion of CEO pay in general is really strange and overly focused on the pay earned by top executives in a couple hundred big firms. It’s easy to forget that the average CEO earns $178,400 a year, and that even the CEO-to-lowest-paid-worker ratio among the top firms is being really heavily skewed by the “restaurant and hospitality” sector.

    4. That Durbin law was incredibly stupid. What would have been smarter would be to allow merchants to pass the cost of it on to customers, either by charging an additional fee for credit card use if they wanted to, or by offering a “cash discount”. A lot of places I go to won’t take a credit card payment if the charge is under $3.00 anyways.

    5. I wonder if Montebourg got fired because of his stupidity when France was trying to find a buyer for an Arcellor-Mittal plant that was going to close. In any case, France really needs investment and new business, and they’re under heavy pressure from Germany and the ECB to do structural reforms (i.e. make it easier for businesses to fire people and get businesses started). They might trim some of that around the edges, but the last time they tried to do a major reform to French labor laws in 2006, it generated massive protests and a backlash.

    6. It’s pretty hard to safely create a deadly disease (especially in a cheap laboratory) and then distribute in such a way that it’s effective as a weapon.

  5. Gravatar of BC BC
    26. October 2014 at 03:19

    Re: Brett, #6. Is it hard to unsafely create a deadly disease and then allow it to infect a lot of victims at random (including possibly the disease’s creator inadvertently), even if those victims are not explicitly chosen as targets of a weapon? A bio-terrorist, who wants to solve the “problem” of overpopulation, has different requirements for a bio-weapon than does a conventional military that wants to target a specific enemy while minimizing “collateral damage”.

  6. Gravatar of Nick Nick
    26. October 2014 at 03:29

    Professor sumner,
    I’ve left a comment to this effect before, but I’m pretty sure your ‘many worlds utilitarianism’ is pro dangerous bio experimentation–even if it has a substantial risk of annihilating all of humanity. After all, it won’t kill us all in at least one quantum universe … And it could be cool.
    I think it might be a slightly probelematic philosphy.

  7. Gravatar of ssumner ssumner
    26. October 2014 at 04:24

    Thanks Kevin.

    Andrew, Well at least we know from the study that CEOs are really important, and that it’s worth paying a lot if it helps you get a good one.

    Brett, Some good points, but I don’t see your first point contradicting the quote, and as far as your final point “It’s pretty hard” is exactly what scares me:

    1. It has to be more than “pretty hard” when you are talking about billions of deaths.

    2. It’s getting easier every year.

    3. We know there are lots of eco-nuts worried about “overpopulation.”

    Sure, it’s hard enough that it hasn’t happened yet, but is that supposed to reassure me? It’s also pretty hard to destroy the Twin Towers with explosives.

    NIck, Don’t worry, I’m not confident enough about the many worlds concept to experiment with it. 🙂

    Also, some MW proponents claim that it has no implications for our behavior. I’m not smart enough to disagree with them. (I seem to recall Eliezer Yudkowsky is one.)

  8. Gravatar of MaynardGkeynes MaynardGkeynes
    26. October 2014 at 05:37

    I imagine that if all of the employees died, the stock price of most companies would also go down. It’s nice to live in a country where no one is overpaid.

  9. Gravatar of mbka mbka
    26. October 2014 at 05:57

    I haven’t read these CEO pay studies and I am sympathetic to the idea that CEO pay overall makes sense, through various kinds of reasoning. But: the statistical argument here does not make sense to me. Please enlighten me on the following: Assume CEOs have zero influence on a company’s fate. Statistically, upon CEO death, half of the companies affected would see their stock prices fall initially, and half would see them rise, on spurious market beliefs. So, a 50/50 split is the null hypothesis (CEOs are entirely uncorrelated with company performance). Now assume CEOs have 100 % influence on company performance but CEO performance is perfectly normally distributed. We would still expect a 50/50 split in stock price performance at CEO death. So, not only does a 42/58 split not point to overwhelming evidence for CEO significance. Rather, this kind of data does not seem to prove anything to me. Where (if any) am I going wrong with this line of thought?

  10. Gravatar of Becky Hargrove Becky Hargrove
    26. October 2014 at 07:11

    To me, the CEO studies would make sense particularly for tradable goods which represent product that is mostly a matter of choice, rather than necessity. That is, the CEO would most positively impact decision making processes in this kind of marketplace, for additional profits. In particular, I don’t think the correlation would be good for CEOs of what is essentially skills management.

  11. Gravatar of TravisV TravisV
    26. October 2014 at 07:39

    Matt O’Brien: “Why Europe is doomed, in 3 paragraphs”

  12. Gravatar of Pietro Poggi-Corradini Pietro Poggi-Corradini
    26. October 2014 at 09:29

    But would Hollande’s plans have succeeded had the BCE done the right thing?

  13. Gravatar of Brett Brett
    26. October 2014 at 10:07

    “Succeed” is a little strong, but I don’t think they’d be feeling the heat to pare them back as much.

  14. Gravatar of BC BC
    26. October 2014 at 17:06

    mbka, a company’s stock price reflects the market’s assessment of the net present value (NPV) of the firm’s future cash flows. Shareholder value is maximized by the firm taking on all projects with positive NPV. A CEO’s NPV equals the NPV of the additional cash flows resulting from his/her hiring less the NPV of that CEO’s compensation. A fairly compensated CEO, just like any fairly priced investment project, has an NPV of zero. (If that seems strange, consider that NPV discounts cash flows with required rates of return that reflect the riskiness of those cash flows.) An overpaid CEO has a negative NPV, and an underpaid CEO has a positive NPV. A firm loses the CEO’s NPV if he dies unexpectedly. Thus, upon CEO death, share price rises for overpaid CEOs (firm loses negative NPV), share price falls for underpaid CEOs (firm loses positive NPV), and share price remains largely unchanged for fairly compensated CEOs.

  15. Gravatar of Benjamin Cole Benjamin Cole
    26. October 2014 at 18:57

    BTW, OT to Scott Sumner:

    You have been confident about mainland China.

    But you also have some faith in markets.

    The Hong Kong Hang Seng Index is now trading at 6.6 times forward 12m earnings.

    Now, if China grows for the next 20 years, the HSI is likely a screaming buy.

    The HSI seems to be saying there is trouble in China.

  16. Gravatar of mbka mbka
    26. October 2014 at 20:23

    BC, Becky,

    I was making a point about statistics: If CEO performance is normally distributed, or even just symmetrically distributed, then half the CEOs are above average in their performance and half are below. Therefore, upon CEO death half the companies have now lost an underperformer and half have lost an overperformer. In my view, therefore, half the companies should see their stock rise and half should see their stock fall, given the expectation that the next CEO will most likely be exactly average (this stems from probability). So the split in stock response should be 50/50 in gains / losses. Factoring in that death of a CEO is to some extent always a burden and a rsik, and should therefore skew stock performance negatively, the actual outcome (58% stock lost value and 42 % gained value) is reasonable. But it tells us nothing about the extent to whether the CEO was beneficial, harmful, or otherwise.

  17. Gravatar of Scott Sumner Scott Sumner
    27. October 2014 at 04:45

    Maynard. “All employees” are generally paid more than CEOs, unless I’m mistaken. And there are lots of overpaid people in America, many are called government employees.

    mbka, That’s a good point. I assumed they were doing event studies. If the price moves sharply on the death of a CEO (up or down), that’s pretty strong evidence that CEOs matter a lot, and that it’s worth expending a lot of money trying to get the best. But I didn’t look at the study closely. Did they find statistically significant moves right after the deaths?

    I agree that it doesn’t tell us whether they are overpaid, just that they should be paid a lot, so I take that as showing that a finding they are paid a lot is not evidence they are overpaid. (Which many seem to assume.)

    Peitro, He would have failed more gradually.

    Ben, I am confident China will eventually become a developed country (say within 30 years or less.) I’m not confident at all that they will avoid economic turmoil/debt crises in the years ahead.

    I don’t have strong views on the HK market, but yes, I do assume the EMH is true.

  18. Gravatar of Floccina Floccina
    27. October 2014 at 06:53

    Perhaps the overpopulation crowd will focus on a virus that makes people sterile.

  19. Gravatar of cthorm cthorm
    28. October 2014 at 12:32


    I’m with you on the existential threat presented by biohackers. It is an underrated threat in my estimation. I like the example from Twelve Monkey’s, where the instigator of the apocalypse was a rogue biologist/biochemist/geneticist that created an especially virulent virus and released it on a number of global flights.

    I’m at odds on the issue though. Biotechnology, especially the DIY movement, has enormous potential to improve our lives. Or end them. I’m leaning towards the attitude of Cody Wilson, founder of Defense Distributed (who’ve introduced 3D printed guns): the technology is out there, you can’t ban it now. We’ll have to look to biotech itself for defense mechanisms.

  20. Gravatar of Vivian Darkbloom Vivian Darkbloom
    28. October 2014 at 12:38

    @cthorn (and Scott)

    But, over at Econlog, Bryan Caplan would argue that, because there have been zero casualties *in the past* this presents no risk!

  21. Gravatar of cthorm cthorm
    28. October 2014 at 13:46


    Not to go full Taleb, but the fact that something has (or has not) happened tells us almost nothing about it’s ex-ante probability.

    Even if it is an ultra low probability event (say 1 in a billion), we know that the potential damage is catastrophic based on prior pandemics like the Spanish flu. The Spanish Flu had a death rate of ~2.5%, twenty-times higher than the common flu strains we have today. An engineered virus could be a further 20-times more deadly, which seems to be on par with Ebola in developed countries.

    That said I agree with Caplan that conventional counter terrorism efforts are likely to be futile in preventing such damage. Our best bet is some sort of immune system enhancement (engineered antibodies? I don’t know.)

  22. Gravatar of Nick Nick
    28. October 2014 at 14:06

    The technology for creating a small nuclear accident is out there, but we try and stop them anyway. There was an interesting Superfund site created bc a disturbed enthusiast had created a pile from smoke alarm batteries. He had no agenda, just an enthusiasm for fission.
    Let’s try and stop some of this stuff, even if some will evade the effort.

  23. Gravatar of ssumner ssumner
    30. October 2014 at 05:15

    cthorm, Just to be clear, I’m not advocating banning DIY, just worrying about it.

    Vivian, Good point. In fairness, I suppose if one was discussing conventional terrorism (say Al Qeada) the risk is fairly low (even with 9/11). The real risk would be future scientists who are terrorists.

Leave a Reply