Revolving Doors and Robot Appraisers

Also Gary Cohn, Martin Shkreli, Additional Tier 1 capital, volatility and unicorns.

Revolving doors.

Why do banks hire former regulators? There is a popular view that they do it to reward the regulators for being nice to them: If regulators know that, after a few years of regulating, they can leave for a cushy job in the private sector, they are more likely to be lenient in their regulation. But this view is wrong. It is wrong as a description of what regulators do. We talked several years ago about a New York Fed report finding "a positive association between the intensity of strict actions over time and across states and the net inflows into the regulatory sector": Strict regulation increases the demand for ex-regulators in the private sector, so a purely self-interested regulator should want to regulate strictly in order to maximize her chances of getting paid later.

It is apparently also wrong as a description of what banks do. Here (via Tyler Cowen) is a paper (free version here) by Sophie Shive and Margaret Forster of Notre Dame, who "investigate the motivations and effects of financial firms' hiring of former U.S. financial regulatory employees":

We seek to understand how ex-regulators might add value to financial firms. A first possibility is that the ex-regulator is hired for his or her social connections in order to obtain preferential treatment from the regulatory agency (the “quid pro quo” hypothesis). In this case, we might expect to see the firm take on higher levels of risk subsequent to the hire, especially for hires of executives with experience at prudential regulators, who focus most of their supervision efforts on risk levels. The ex-regulator could also be hired as a reward for past preferential treatment. In that case, we would expect higher levels of risk, especially in a period prior to hire, but perhaps continually, if the firm sequentially targets new employees of the regulator. An alternative hypothesis is that ex-regulators are hired for their expertise (the “regulatory schooling” hypothesis), perhaps after receiving regulatory pressure to manage risk. Under this hypothesis, we might expect high levels of risk prior to hire and a decrease in the level of risk subsequent to hire, along with additional indications that the firm is making efforts to decrease risk.

"The evidence we find is most consistent with the schooling hypothesis," they write: "In the quarter after hire, market and balance sheet measures of firm risk decrease significantly and measures of risk management activity increase." If a prudential regulator leaves to become a bank executive, she is likely to bring the mindset of a prudential regulator into banking -- not to bring the mindset of a banker into regulation.

On the other hand, the revolving door, as the name implies, opens in both directions. Here's a profile of Randal Quarles, a former Treasury official who became a Carlyle Group dealmaker and who is now the nominee to be the Federal Reserve's vice chairman of supervision, where one assumes he will be light on the supervision:

Having spent his career studying the banking business and regulations, Quarles declared himself unimpressed by the 2010 Dodd-Frank Act, the U.S. response to the financial crisis, calling it a “failure of ambition” and a “concession to inappropriate pressures,” during a Bloomberg Television interview in 2015.

One story you could tell about the regulatory revolving door is that incentives probably matter less than acculturation. If you work as a regulator, you are surrounded by other regulators, and regulation is the cool thing to do, and you think regulatory thoughts. It might occur to you that you could make a lot more money one day by undermining regulations, but that is not a big part of your everyday thinking. And then when you leave your regulatory job for the private sector, you carry your regulatory thoughts with you, at least for a while. On the other hand, if you work in finance, you are surrounded by financiers, and ... optimizing ... for regulation is the cool thing to do, and you do that. And then when you leave your finance job for a regulator, you carry your finance-friendly thoughts with you, at least for a little while. 

Elsewhere in Fed nominees.

Here is a Politico story reporting that "President Donald Trump is increasingly unlikely to nominate Federal Reserve Chair Janet Yellen next year for a second term," and that "National Economic Council Director Gary Cohn is now the leading candidate to succeed Yellen as the world’s most important central banker." That ... seems ... right? Like, Janet Yellen seems like a pretty good Fed chair, and Gary Cohn seems like a pretty good manager who is unusually competent and pragmatic for the Trump White House, but who is not particularly qualified to be a Fed chair, given that he is not an economist. They are perhaps the two senior people in government most qualified for their current roles, so of course they can't stay in those roles. Moving Cohn from the White House to the Fed would reduce the competence of both the Fed and the White House, which I have come to assume is the goal.

Robot appraisers.

There are financial markets where algorithms are taking over from humans. Stocks and currencies trade pretty much continuously, and there is a ton of data to analyze, and computers are better at crunching data than humans are, and crucially they are constantly being given new out-of-sample data to test their analysis on: If your stock-trading algorithm is wrong, it will lose money in the next minute, and you can turn it off. On the other hand, there are financial markets where it seems like algorithms will have a tougher time: Private equity firms, for instance, buy companies fairly infrequently, and data about private companies is often non-public and non-standardized, and it's hard to tell if your decision to buy a company was correct until many years have passed. So while quant firms are common in the stock market, no one is going to set up an automated algorithmic private equity firm any time soon.

Still you have to expect automation to, as it were, move up the value chain, into less frequently traded and more opaque asset classes. Here is a fun article about the automation of real estate appraisal:

Advances in big data and computing are helping automation creep into knowledge-based professions, threatening to knock off jobs in much the same way robots have been doing at factories for decades.

In the real estate business, Zillow Group Inc. says its algorithms are learning to capture not only the crude facts about values in the surrounding neighborhood but also more sophisticated price indicators, such as whether the living room has hardwood floors or the kitchen has granite countertops.

As always, the counterargument is that there are some things that only a human can do, like go out and actually look at houses. But Zillow engineers are already "mining the website’s trove of photos provided by real estate agents, looking for interior features that make one home more valuable than another," and "trying to teach their algorithms to recognize the curb appeal—a mature tree casting shade over the front lawn, for example—that makes buyers ready to pay more." And the robots can examine much more data than the humans can, and can form intuitions that the humans can't, because the humans are data-constrained:

There are all kinds of untapped data that could make models work better, Humphries says, including some that Zillow probably doesn’t know about yet. “Maybe it will be ambient street noise or airport flight paths,” he says.

Human appraisers might have some sort of gut instinct about the valuation effect of flight noise, if they happen to do an appraisal when a plane is overhead. But the robots will know all the flight paths, and how flight paths correlate with value, and their gut instinct will be more reliable.


For the models to work, they need similar, recently sold homes for comparison. The more remote or unique a property, the worse the algorithms are at estimating its value. “They won’t perform well on a geodesic dome in the middle of Montana,” says Susan Allen, senior vice president for valuation solutions at research company CoreLogic Inc.

But that is also what human appraisers need! Why would a human appraiser's appraisal of a geodesic dome in the middle of Montana be an accurate guide to its future selling price? There is no innate human ability to recognize the value of real estate; that is not a skill that evolved in our ancestors on the savanna. Human real estate appraisers try to estimate the value of a house based on recent selling prices of similar houses. They are in the pattern recognition business. The argument for artificial intelligence in financial markets is basically that computers are better -- more rigorous, more able to use large amounts of data -- at recognizing patterns than humans are, and that in the long run pattern-recognition jobs will be automated.

Elsewhere, here is Fabio Ciucci on artificial intelligence and machine learning:

Unlike some other sciences, you can't verify if an ML is correct using a logical theory. To judge if an ML it is correct or not, you can only test its results (errors) on unseen new data. The ML is not a black box: you can see the "if this then that" list it produces and runs, but it's often too big and complex for any human to follow. ML it's a practical science trying to reproduce the real world's chaos and human intuition, without giving a simple or theoretical explanation. 

How's Martin Shkreli doing?

When Martin Shkreli was first arrested for securities fraud, I was pretty impressed that he managed to take an initial investment of $700,000 down to $700 - a loss of 99.9 percent -- and then somehow make it back and pay off his investors with profits. "If you're down 99.9 percent, it's hard to come back," I said, "but it's reeeeeeally easy to lose that last 0.1 percent." It turns out I underestimated him!

Jurors saw that after Shkreli made a disastrous trade on Feb. 1, 2011, his hedge fund’s net asset value also plummeted, from high of $1.1 million the day before to just $55,786 on Feb. 28, recording a balance of minus 33 cents by July of that year.  

It is fairly easy to turn $700,000 into $700, or $1.1 million into minus 33 cents (a loss of 100.00003 percent). It is harder to turn $700 into millions. But it is very, very difficult indeed to turn minus 33 cents into millions, as Shkreli more or less did. Negative 33 cents might be the very hardest amount of money to turn into positive millions. For instance, if you told me that someone had a fund with negative $100 million, I would say: That guy is going places! Having negative a lot of money gives you a decent chance of one day having positive a lot of money. But I suspect that no one had ever turned negative 33 cents into millions, until Shkreli did it. That is just not a lot of money, even in absolute-value terms. "If you owe the bank $100, that's your problem; if you owe the bank $100 million, that's the bank's problem," but if you owe the bank 33 cents you are just not on the bank's radar. 

Bank capital.

Here is a good piece from Thomas Hale about the odd treatment of bank capital securities in the Banco Popular Espanol SA failure. In particular, he notes that Additional Tier 1 capital securities were meant to be "going concern" capital, that is, bonds that flip into equity when bank capital ratios fall, rather than only being written off when the bank actually fails. But in fact Popular's AT1s were wiped out when it failed, exactly the same as its Tier 2 capital securities, which are senior to the AT1s and are meant to be "gone concern" capital. Hale writes: 

The AT1s were basically treated as though they were tier 2s – in other words, the gradual move through the capital structure, which is the whole point of several forms of bank capital, could not happen.

The problem is that bank failure is not usually about capital levels gradually falling over time, which could be solved by gradually flipping bonds into equity. Instead it is usually about liquidity -- about deposit runs -- and recapitalizing banks using AT1s can actually hurt liquidity:

One flaw with additional tier 1 bonds is that, while they have in-built triggers which boost a bank’s capital ratio, they also act as extremely effective triggers, or warning signals, for depositors to place their cash elsewhere.

The effect of the secondary market for AT1 here exacerbates the existing problem of publicly traded banks, which may be vulnerable to retail or corporate withdrawals in the instance of a sudden share price collapse. If AT1s were hit while the bank were still running, its easy to imagine an extremely negative response from depositors.

Elsewhere: "Bank Rescues Leave Sweden Questioning Point of Post-Crisis Union."

People are worried that people aren't worried enough.

"Can't Complain About a Lack of Volatility Now," writes my Bloomberg Prophets colleague Robert Burgess, but sure you can! Yes, there was some excitement yesterday: "The Dow Jones Industrial Average erased about 160 points in 20 minutes and volume in the most widely traded S&P 500 futures tripled as a gut check landed on traders just after 11 a.m. in New York Tuesday," in the form of Donald Trump Jr.'s weird Twitter confession. But the Dow closed up 0.55 points on the day -- up 0.0026 percent -- and the VIX volatility index actually fell, closing at 10.89. The market reacted to political news, then remembered that we're not doing that any more, and quickly un-reacted to the news.

People are worried about unicorns.

The biggest worry for unicorns this week comes from the most recent class of unicorn alumni, Snap Inc. and Blue Apron Holdings Inc., which "have swiftly gone from two of the year’s most anticipated IPOs to poster children for companies whose rich private valuations haven’t withstood the scrutiny of public markets." Snap fell below its $17 initial public offering price on Monday, and closed at $15.47 yesterday after an analyst at Morgan Stanley, the lead underwriter on the IPO, cut his price target from $28 to $16. (Oops!) Blue Apron, meanwhile, fell below its already-discounted $10 IPO price pretty much as soon as the greenshoe ran out, and closed yesterday at $7.14. Both companies are trading below their last private valuations, which is not encouraging for other unicorns:

Dick Costolo, Twitter Inc.’s former chief executive officer, said in a Bloomberg TV interview that he expects the valuation declines at the IPO stage to persist in the next two years. He predicts “lots of money in the private markets chasing few deals” and driving up prices.

“When these things finally see the light of day in the retail markets, it sort of -- probably rightly -- turns out to be the case that these are not worth as much as their last private rounds were,” said Costolo, now head of his fitness startup Chorus.

It is hard to draw too many conclusions from these data points: Each unicorn is a unique magical creature, and there's no certainty that the weak reception for these two is a bad sign for others. But the basic worry of "people are worried about unicorns" has always been that frothy private markets were funding technology companies at valuations that could not be justified in the public markets, and now we are starting to see sustained evidence that that might be right. So what will happen?

Fear of underperforming on lofty private valuations has already gummed up the exit pipeline. Pre-IPO companies that don’t need the cash may hold off listing in the hope they can grow into their private market value, Gordon said.

Typically price bubbles are resolved through price: People realize that private markets are overvaluing companies, and so private-market valuations go down. But it is not quite so simple in the Enchanted Forest, where valuations don't update every day, and where people are emotionally invested in never having a down round. So the price bubble might be resolved through time: Rather than trading at a low price, you just wait, and hope to trade at a higher price ("grow into your valuation"). Obviously this plan doesn't work if the trend is permanently down, but they are all optimists, out in the Enchanted Forest.

The other big unicorn worry is of course gender discrimination:

A team of management and psychology researchers examined how venture investors interacted with entrepreneurs at TechCrunch’s New York startup competition over seven years and found that VC funds typically asked male founders questions about how their companies will succeed, while posing questions to women founders about how they would avoid failure.

Also apparently there's a deer-unicorn in Italy.

Things happen.

Paul Singer Could Finally Get Warren Buffett to Punch Back. Wanda to Lend Borrowed Billions to Sunac to Close Deal. China’s $800 Billion Sovereign Wealth Fund Seeks More U.S. Access. Temasek’s Divestments Outpace Investments for First Time in Nine Years. Dimon warns EU could force banks to move staff out of UK. Paying Professors: Inside Google’s Academic Influence Campaign. Segmented money markets and covered interest parity arbitrage. Time Inc. Explores Renaming the Company, Seeking a Refresh. Ex-Credit Suisse Trader Raises $66 Million for Bitcoin Push. Swedish Security Company Boss Declared ‘Bankrupt’ After Identity Stolen. Airline Passenger Checks Single Can of Beer As His Only Piece of Luggage. 

If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Thanks! 

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

    To contact the author of this story:
    Matt Levine at

    To contact the editor responsible for this story:
    James Greiff at

    Before it's here, it's on the Bloomberg Terminal.