Friday, June 29, 2012
BROOKE GLADSTONE: Both she and Texas Tribune reporter Matt Stiles believe in the power of data. They only wish it were more abundant and more relevant. But data can be dangerous, even deadly, as OTM producer Jamie York reports in these two cautionary tales.
JAMIE YORK: Tale number one: There are few more obvious indicators of urban blight than fires. Burned-out buildings and empty lots announce devastation and neglect. By the 1970s, New York City announced that it was going broke –
[SIRENS/TRAFFIC SOUNDS] - its residents were fleeing to the suburbs, and it was very clearly on fire.
MALE CORRESPONDENT: The fire statistics for New York are staggering. New York has more fires than Chicago, Detroit, Los Angeles and Philadelphia put together.
JAMIE YORK: Firemen called this decade and a half “the war years,” but the devastation was confined to poor neighborhoods, and most New Yorkers had no idea what was happening in their own city.
MALE CORRESPONDENT: And the Yankee defense -
JAMIE YORK: That is, until 1977. Joe Flood is author of The Fires. [SIRENS/BASEBALL GAME HUBBUB/UP AND UNDER]
JOE FLOOD: So it was the second game of the 1977 World Series - Los Angeles Dodgers, New York Yankees, Bronx, New York, about 60 million people watching at home. And before the game, a fire breaks out in an abandoned area of the Bronx. By the end of the '70s, this particular neighborhood had lost over 99 percent of its buildings to fire and abandonment. On this night, a fire broke out in what turned out to be an abandoned schoolhouse, and throughout the game, they cut to it. Famously, Howard Cosell intones to the tens of millions of people at home, there it is, ladies and gentlemen, the Bronx is burning. It turns out he actually never said this.
HOWARD COSELL: That is a live picture. And obviously the fire department in the Bronx have their -
JAMIE YORK: Nevertheless, it marks the moment America first becomes aware that this wasn't just about a few buildings burning in the Bronx.
JOE FLOOD: They see what surrounds it, which is block after block of burned-out and abandoned buildings, abandoned lots strewn with rubble, and, and this becomes a tale of decay, of the crumbling inner city.
JAMIE YORK: Dennis Smith wrote the definitive book of the war years, called Report from Engine Company 82. As a young firefighter in the late 1960s, he sought out the busiest firehouse in the city, Company 82 in the South Bronx.
[VOICE ON P.A. SYSTEM]
DENNIS SMITH: You could look out, you’d see the next job that you were going to go to because the smoke and the fire was already coming out of the windows. And we went from one fire to the next, in an average of about 40 alarms a day.
JAMIE YORK: So why was the Bronx burning? Many of the buildings in poor neighborhoods had been built in the early 20th century. They were old, unsafe and overcrowded. When the landlord, say, cut back on heating oil, people used space heaters, which either overtaxed the wiring or actually set something ablaze. If a few apartments were burned out, maintaining the rest of the building wasn't a priority. Junkies set up shooting galleries in the abandoned spaces. Where once a fire would lead to increased vigilance by neighbors, it was suddenly the opposite. Fire became an indicator of more fire. And then in the late 1960s, New York City made a fateful choice and hired the RAND Corporation to give advice to the fire department about how best to allocate its resources. RAND was the original military think tank. It has its roots in World War II, when the Air Force recruited a stable of intellectuals who created systems analysis, a clear, exacting way of exposing inefficiencies and reducing the messiness of war to tables, charts and spreadsheets. And it worked. Fresh off their success in World War II, this group of whiz kids created the RAND Corporation, and by the late 1950s, RAND’s approach, apolitical, data-driven, was revolutionizing business and public policy. Joe Flood.
JOE FLOOD: Technology creates the nuclear bomb. That leads to America’s victory in Japan. That leads to its international dominance, geopolitically. In the early 1960s, the White House’s economists year after year were coming within .2 percent accuracy of predicting the employment rate, of predicting inflation, of predicting gross domestic product. The world seemed to be a predictable place, a place that could be understood and solved by numbers.
JAMIE YORK: New York City had a progressive mayor and a highly innovative fire chief, both of whom were enthralled to the idea of data-driven solutions to their management and fire woes. The only problem? RAND knew nothing about fighting fires. So they did what they usually did. They broke the problem into manageable parts. First up, how long does it take a fire engine to get to a fire?
JOE FLOOD: What they do is they give stopwatches to about a dozen fire companies, a dozen stopwatches for around 300 in total fire companies, so margin of error’s already going to be a problem.
JAMIE YORK: Another problem? The system relies on firemen to operate the stopwatches on the way to a fire. Needless to say, participation is a little spotty. What’s more, Joe Flood says, RAND fails to accurately figure traffic into its model, estimating response time as the same at rush hour in midtown and a residential neighborhood at 4 a.m. Most importantly, according to Joe Flood, RAND assumed that firemen respond from their firehouses. But when you can see the next fire you’re going to be fighting, you might not get back to your firehouse for hours. To compensate, the busiest firehouses had what they called a second section, a backup company. But the models saw these backups as redundancies and recommended that they be closed in the most fire-prone neighborhoods.
JOE FLOOD: When these second sections are closed, you see the number of fires that the first company that’s left behind as much as quadruple within a couple of years. When you’re able to respond to a fire inside of a few minutes, you can often prevent it from spreading. So what is a serious problem becomes something of an epidemic of fire in a lot of these neighborhoods, and it, it feeds and grows on itself as this continues.
DENNIS SMITH: Empty buildings, like Dresden.
JAMIE YORK: Firefighter Dennis Smith.
DENNIS SMITH: In a report from Engine Company 82, I said there’s a Dresdenization of this neighborhood. All you see are these empty windows. They're like hollow eyes that are coming out of these buildings, thousands of them, and they serve no purpose except to remind you of the utter destruction of the neighborhood.
JAMIE YORK: It was a striking case of bad data having the worst kind of real-world consequence. RAND denies many of these errors in judgment, but Joe Flood stands by his analysis. He does not doubt that the people involved were guided by the best of intentions, that is, to solve problems. But the data are seductive. They may seem to be solutions, but they all too often create more problems. Cautionary tale number two: Ed Thorp was a math whiz, a professor at MIT who in the 1960s became the first person to go to Vegas and beat the house at blackjack. This was probability theory or, put another way, counting cards. Thorp called his 1962 book Beat the Dealer, and it inspired a whole generation of math geeks to apply math to gambling. But Thorp was already off to his next challenge.
SCOTT PATTERSON: He quickly alighted upon Wall Street. He looked at all sorts of investing strategies and came upon some fairly complex stuff in the convertible bond world. Ed went to work on it, he got his IBM supercomputers out, he fed in reams of data and he - he learned that there were a lot of anomalies in this at the time very obscure corner of Wall Street. He quickly started making money.
JAMIE YORK: Michael Lewis was a more traditional trader on Wall Street who watched the game change. He went on to write Liar’s Poker, an account of the Street in the mid-1980s.
MICHAEL LEWIS: In a matter of a few years, it was overrun by people who had PhDs in physics and math from MIT, and there was a sort of general view that managing risk was complicated, so complicated the ordinary mortal couldn't understand it.
JAMIE YORK: The Quants’ most elemental strategy, Patterson says, was to use the immense amounts of data from the past to predict the future. Patterson says that the strangest illustration of this comes from the most successful Quant group, a fund named Medallion, which is and has been staffed with voice recognition experts from IBM.
SCOTT PATTERSON: The idea is to predict with some probability what the most likely next syllable or word may be, based on historical patterns. So if it detects a phrase from The Star-Spangled Banner or something, it has a high probability of predicting what the next word is going to be.
JAMIE YORK: This capability applied to finance was something like magic, alchemy. The model seemed to have found a way to identify anomalies in the market and perfectly predict the potential losses of any gamble. But it insulated them to the very real unpredictability of life. And the Quants didn't just snow investors. The complexity of what they'd created befuddled their managers, and ultimately the Quants themselves.
MICHAEL LEWIS: The models told the bosses that, ah, this was actually - what they were doing was riskless.
JAMIE YORK: Michael Lewis.
MICHAEL LEWIS: The models were wrong. The important point about them was they all lost their money alongside everybody else. They believed what they were doing. What the complexity has done is make the risk opaque. In this era of Wall Street, complexity is the new opacity. It’s what prevents most people from exercising their judgment when presented with financial risk, 'cause they're just persuaded they don't, they don't understand it; it’s all too complicated.
JAMIE YORK: When Wall Street imploded, first in the so-called Quant Quake of 2007, and then in plain sight in 2008, the Quants lost a tremendous amount of money, mostly other people’s. And they were chastened, but only a little.
SCOTT PATTERSON: Now we have this new investing strategy, or I don't really know what you would call it, called high-frequency trading, which is estimated to account for upwards of 70 to 80 percent of the U.S. stock market.
JAMIE YORK: Scott Patterson.
SCOTT PATTERSON: And it’s completely computer-driven. These firms are pushing the limits of technological computer speed. It boggles your mind. They're measuring their trades, the - the speed which they can send an order to an exchange, at microseconds, which is a billionth of a second. And now they're talking about picoseconds. Is that a trillionth of a second or something?
JAMIE YORK: Then on May 6th, 2010, an event now widely known as the Flash Crash.
MALE CORRESPONDENT: We have seen a flight to safety within…
MALE CORRESPONDENT: This market is dropping precipitously. It just went negative 500, and is now in the 500…
FEMALE CORRESPONDENT: The market’s coping with continued fear over failure.
JAMIE YORK: The stock market fell about 10 percent in the space of just a few minutes. And we now know that it was almost completely the result of computers reacting to instability in the news and panicking.
SCOTT PATTERSON: And we saw crazy things happen. We saw – the order books for our national stock markets vanished. The order book is where all the trades take place at these data centers, most of which are in New Jersey. We saw exchange-traded funds that are representing billions of dollars of shareholder value trade for a penny. So [LAUGHS] it was a complete break in the market structure, because when the market is falling rapidly, that’s when you need market makers to step in and create order. That’s what the specialists on the floor of the New York Stock Exchange’s job has been for decades. But now, because of the proliferation of secret computer strategies, you have a market that is completely blind to itself.
JAMIE YORK: Data is always susceptible to our very human, very analog limitations. In New York City it was the well-intentioned desire to find clarity in a city of immense complexity, an earnest desire to help solve a problem. On Wall Street it was almost completely the opposite. The data allowed smart people to make a lot of money, even as it blinded them to how much risk was involved.
[MUSIC UP AND UNDER] There’s a pattern here, Joe Flood says. Initially we use data as a way to think hard about difficult problems. But then we over-rely on data as a way to avoid thinking hard about difficult problems. We surrender our better judgment and leave it to the algorithm. Right now, computers are flitting through unprecedented stores of information, systems that undergird so much of our lives, looking for answers and hidden patterns. It’s much too late to stand in that way of that process. Maybe Warren Buffett was right when he said, beware of geeks bearing formulas. For On the Media, I'm Jamie York.