There was a lot of scandal, a lot of exposure of bad people behaving badly, in Michael Lewis' 1990 book about his years at Solomon Brothers in the mid 1980s, but very little of that stuck with me. No, the part of Liar's Poker that has stuck with me for almost twenty years since I first read it, the reason I still occasionally recommend it to people, is what Lewis' former bosses on Wall Street taught him about one of the oldest, and thorniest, and still hardest problems in economics, and that's what's called "the question of value." I was well primed for it when I saw it in Liar's Poker. I had a far, far better teacher of economics in high school than almost anybody ever got; not only had most American high schools ceased teaching economics long before I got to that age, but of all the people teaching economics in America in the 1970s at any level, he was one of the only ones who kept correctly demonstrating his economic abilities by making money at them. (Teaching at a Christian fundamentalist college-prep private school was just a hobby for him, part of his contribution to the takeover of the Republican Party by the Christian anti-communist caucus that I've written about before. Fascinating man.) But even before that, I saw it where many of you probably saw it for the first time, in Robert Heinlein's award winning (and still controversial) science fiction novel Starship Troopers, where Heinlein's knee-jerk anti-Marxism resulted in his authorial-stand-in character taking what I think is the wrong side of the question of value. It stuck with me, it was one of the very few notes in that book that rung false to me, so the question has fascinated me ever since I was a fairly small child:
Does a thing have an actual value that can be measured or computed, a correct price that actually exists in the real world that can be determined as a matter of objective fact? Or, is it true, as the opposite side of the question says, that "the value of a thing is the price it will bring"?
When Michael Lewis was first hired, right out of college, as a salesman at Solomon Brothers, his instinctive belief that things have an actual value was something that got relentlessly mocked by his coworkers. In his early days on Wall Street, they explained to him that there is a pecking order in the financial services industry. For a variety of cultural and regulatory reasons, investment firms are required to have people working for them who are experts at calculating the actual value of an investment based on current mathematical models and best available data. Their department is called "Quantitative Analysis," and the people who work there are derisively called "quants." Quants have the lowest prestige jobs in the entire industry, draw remarkably low salaries considering the level of education you have to have to get those jobs and the long hours and the awful working conditions, and Lewis says that they are routinely and cruelly and ruthlessly snubbed by the other half of the business. Those are the people whose specialty is sales. And at the absolute top of the pecking order, Lewis taught us, are the people who were called the Big Swinging Dicks, people who demonstrated the superiority of their manhood by being able to sell anything, however worthless the quants said it was, for however much the company needed it to sell for.
How does a BSD make his money? First, he starts by asking, "what do I have available to me to sell?" How much of it is there? Now divide that into his assigned sales goal. Add in the cost to sell it. That tells him what the thing he has to sell has to be worth; for it not to be worth that is flatly unacceptable. Which leads to the next, and last, and hardest part: what lies does he have to tell to himself to convince himself that it really is worth what he needs to sell it for, and how does he fool himself into believing his own lies? That's the only use that any BSD has for a quant: once in a very, very rare while a quant says something to him that, if he ignores the caveats and footnotes and fine details, he can twist into a justification. The quant doesn't have to have been right. Even if the quant was right, the salesman doesn't even have to accurately remember, let alone accurately pass on to others, what the quant said. He just needs to be able to believe that he is honestly representing what the quant told him when he quotes some bit of (what is to him) quant bullcrap about what the investment is "really worth" when he's on the phone to the client. And he really, really has to believe it, himself, even if he used to know that he was lying, even if he used to know that he was making this stuff up out of whole cloth. Humans kid themselves that they're good at detecting it when people tell them things that aren't true. They aren't. But to the limited extent that they are, they're really only good at detecting one thing: whether or not the other person believes what they're saying. So the ultimate BSD is the guy with the greatest talent for figuring out what he needs to believe in order to get you to give him money, and then making himself believe that, and believe it really hard.
After the savings and loan crisis. After the dot-com meltdown. After the mortgage bubble. People who trust me to follow these stories, friends of mine who were only vaguely following the news but still knew enough not to buy into earth-shatteringly dumb investments have always asked me, "how did anybody not know?" How in the hell does a lie, a lie that says that an investment that's actually worth zero dollars and zero cents is actually worth tens or hundreds of thousands of dollars, get believed? And once I explain the above to them, they admit that okay, J. R. "Bob" Dobbs was right: as dumb as the average person is, half of 'em are even dumber than that. And that means that as long as there are con men, there will be marks. But when I explain to them how the BSDs first have to lie to themselves about what the quants said, once I explain to them that the lies originate (in a twisted way) with quantitative analysts with some of the best mathematical and scientific minds in the country, the question then becomes how did people as smart as the quants ever give such bad advice to the BSDs? And then I tell them the story of one of the above three crashes, and what the quants actually said, and how the BSDs got it wrong.
The single most entertaining person I've ever told one of these stories to is phierma, a master-class costumer and genuinely talented artist whose day job is working on advanced medical imaging software. Like me, his degree isn't a completely worthless rag like a company's product certification, or a mostly worthless scrap of paper like a degree in "information technology," or even merely a degree in something vaguely skilled like "computer programming." No, Phierma has the same degree I do, even if I don't put mine to any use any more, in the much harder and rarer discipline of computer science. (That people who boast of those other three credentials are ever allowed to touch a compiler is why none of the software on your computer works right.) And I cherish the moment when I get to the point in the story where I explain where the breakdown occurred, because the look of horror (and, being Phierma, barely throttled rage) is priceless: it's something that computer scientists are taught, in their first freshman class at the age of 18, never to do.
You see, every computer program that does any calculation at all is, at its heart, a mathematical model of part of the world. And like any compact mathematical model of the world, it has what are called "boundary conditions." That is to say, we know that for a certain range of inputs, this set of calculations will produce honest and reasonably accurate and useful results. But we also know, if we understand the math at all, or understand the limits of the computer hardware we're running it on, that there are inputs that just should never be put into that model, because the results will be wrong. Maybe we know the math will create errors, like divide by zero errors. In today's world, more often it's that we know that the model was statistically derived, and we know the range of inputs in our statistical model, and we know that we haven't studied what happens when the numbers are outside that range. If your degree was in "information technology," this is probably the first time you've ever heard of this. If your degree was in "computer programming," this got hand-waved, or so they tell me, and you were told that it was enough to just put a note in the documentation with the code, to be put into the user manual by the tech writers, telling people to never input a value there that is out of bounds. Computer scientists, on the other hand, are told early in their training, by brutal instructors who will not brook disagreement or even argument on this subject on pain of flunking you out of any reputable program, that they have a moral obligation as computer scientists to test every input to their programs, and if a number gets entered that the program can not reliably produce a useful result from, to "error out" right there: to stop the program, and refuse to go any further, unless the user puts in a valid number. "Wait a minute!" he shrieks in that tone of outrage his friends all know so well, "Are you telling me that they lost all those billions of dollars on something as simple as an 'out of bounds' error?" Yes, Phierma (and the rest of you). That's what I'm telling you. Not only that, but they've done it three times in the last thirty years. The best minds on Wall Street have trashed the entire US economy three times in thirty years because of boundary errors.
1980s: A quantitative analyst named Ivan Boesky takes up, almost as a hobby, the job of calculating what the "risk premium" should be for every class of common investment on Wall Street. To a quantitative analyst, one of the fundamental laws of physics is that if you estimate what the inflation rate will be over the length of an investment, and you can estimate what the percentage chance of not getting your money back is, you can plug those two numbers into a simple equation and calculate what interest rate you have to charge to break even; the difference between the inflation rate and that number is called the "risk premium." So given the better database and spreadsheet tools we started putting on every desk in the early 1980s, he crunched the numbers to determine just what the default rate was on all kinds of investments and calculated actual risk premiums for each of them, and then compared them to the risk premiums that had been set on them the old fashioned pre-computer way, by trial and error, by glorified guesswork. And he discovered one really glaring error: junk bonds. They were a class of investment that was designed specifically for small, startup businesses, almost all of them family-owned, just starting to expand, who'd never had access to "commercial paper" loans before. Since they'd never borrowed that kind of money before, nobody knew how to tell if this particular business was going to pay back its loans, so the risk premium they got charged was plenty high. What Boesky found, when he went out and talked to some people to confirm the results from his spreadsheets, was that these loans mostly went to people who understood that not just their business, but their entire personal future right to ever borrow money ever again, depended on them making these payments, even if they had to sell their kids into sex slavery in Thailand to do it (figuratively speaking), that there was nothing they wouldn't do and no hours they wouldn't work and no effort they would spare to make those payments. And he happened to mention this at a conference to a bunch of sales guys, most famously Michael Milken, who just happened to be sitting on a bunch of junk bonds that (in order to make their numbers) they had to sell for way, way more than junk bonds had ever sold for. And if that had been where the story stopped, it would have been "how capitalism is supposed to work" in perfect form. But the problem is that they did make good money doing it, at a time when almost everybody was losing money, which created huge demand for junk bonds to resell. So the people writing junk bonds started floating junk bonds to the kind of people that weren't in Boesky's original statistical model, and the biggest demand was from "corporate raiders" who wanted to borrow the money to buy up a majority of a company's stock so they could run it (or more specifically loot it) however they wanted. Obviously those two groups are going to have dramatically different default rates. Didn't matter to the BSDs, though; all they heard, out of everything quants like Boesky said, was "junk bonds always make money and lots of it."
I could tell you a similar story about the dot-com bubble of the 1990s, about the difference between actual e-commerce websites with plausible business models, even if only a few of them were going to make it, and (once the demand for e-commerce shares outstripped supply, because no other class of investment was offering the same returns or making the same cheery promises) the proverbial three guys who knew nothing about the business and had nothing to show but a 6-slide PowerPoint presentation and a logo some friend of theirs drew on the back of a napkin. But I don't have to go into as much detail, because most of you remember those days. Just remember this: as fuzzy as the models were, even the most optimistic of the venture capitalists who created the dot-com business model were explicit about the difference between "has a plan to actually make money doing this some day" and "doesn't." But the people selling shares in dot-com startups didn't want to hear that caveat, not if it stood in the way of making their numbers. So they just remembered the part about how some companies in some investment pools were going to make huge money down the road, enough to pay off the losses on the investments in the ones who didn't, and brainwashed themselves into forgetting the rest of the caveats so that they could make their numbers.
Now remember that ever since the first experiments in widespread home loans for first-time home buyers during the Great Depression, one thing that the industry has learned is that working class and middle class people, once they're in a home, will relentlessly work 120 hour work weeks, and if that isn't enough they will murder random strangers and feed their entrails to their children, rather than default on the mortgage, that people who've never taken out home loans before have long been charged risk premiums that are way out of line with their defaults. Remember that in the late 1990s and early 2000s after waves of Clinton and Bush administration deregulation, investment firms' quants came up with a way to spread the risk of the few (but big) defaults across so many people that they'd never notice it, so long as we didn't change our model of who was buying homes and why they were buying them. And then compare that to this paragraph, from Lewis' "The End:"
Eisman knew some of these people. One day, his housekeeper, a South American woman, told him that she was planning to buy a townhouse in Queens. “The price was absurd, and they were giving her a low-down-payment option-ARM,” says Eisman, who talked her into taking out a conventional fixed-rate mortgage. Next, the baby nurse he’d hired back in 1997 to take care of his newborn twin daughters phoned him. “She was this lovely woman from Jamaica,” he says. “One day she calls me and says she and her sister own five townhouses in Queens. I said, ‘How did that happen?’” It happened because after they bought the first one and its value rose, the lenders came and suggested they refinance and take out $250,000, which they used to buy another one. Then the price of that one rose too, and they repeated the experiment. “By the time they were done,” Eisman says, “they owned five of them, the market was falling, and they couldn’t make any of the payments.”I guarantee you that in every firm involved in this market, somewhere in some windowless cubicle in the basement, there was at least one quantitative analyst screaming his head off in emails about this, about how the mathematical modeling that underlaid the calculations that determined pricing on collateralized debt obligations was based on statistical analysis of customers who were buying their primary home, not as an investment vehicle but to live in, and pricing those homes based on reasonable expectations of what someone in their social class could afford to live in and paying no more for them than three times their income. And no BSD in the entire industry wanted to hear that caveat. He didn't dare. He almost certainly paid people to intercept those emails and keep them from him. Because if he let himself get drawn into what must have seemed to him like an arcane and wrong-headed argument about what something is "actually worth," when there really is no such thing, when everybody knows that the only accurate way to value something is to put it in the hands of a talented salesman and see what he can sell it for? He wouldn't have made his numbers.