6
WHEN IT COMES to speed, markets tend to follow a kind of “Goldilocks principle”: they mustn’t be too hot or too cold.
We’ve just seen how too-fast transactions can hurt a market. But excessively slow transactions can hurt markets, too. Surprisingly, markets can be too slow, or congested, even on the Internet. Although the Net operates at the speed of computers, the people using it still need time to consider and act. That’s why, if you really want to operate at digital speeds, you need to take people out of the middle of the process. One way to do this is by moving their deliberations to an earlier time. (Hence the emerging Internet of Things, in which devices learn your preferences, talk with one another, and make decisions for you.)
Markets that involve offers and responses require easy two-way communication. This is why the rise of mobile communications has been so important for the development of many Internet markets: smartphones shorten response times.
Consider Airbnb, which makes a market between travelers looking for a nice, cheap place to stay and hosts who want to rent out their underused guest rooms and apartments.
When Airbnb started in San Francisco in 2008, most people communicated with the Internet via computers. So if you wanted to make your guest room available for visitors the following week, you might use your laptop to post it on Airbnb in the morning before leaving for work. When you came home in the evening, you would check to see whether anyone had expressed an interest—and if so, you would confirm his or her booking. Easily done.
Now look at the other side of this transaction. As a potential guest, you might have had to wait a whole day to find out whether the room you wanted was still available, and if you learned at the end of the day that it wasn’t, you would have had to start over.
You can see the problem. Airbnb’s business model worked well enough in the beginning, when the market was small and the travelers were intrepid young people on tight budgets who were willing to take the time to find a good deal. Airbnb’s competitors in those days were similar Web services, such as the London-based Crashpadder (acquired by Airbnb in 2012) and the Toronto-based iStopOver (acquired by the Berlin-based 9Flats, also in 2012). Competition in those days was based largely on attracting more and more hosts and travelers in order to make the market thicker. That’s why bigger fish acquired littler fish.
But as Airbnb became huge, with lots of hosts and travelers, it became increasingly common to have to make multiple attempts to nail down a reservation. Meanwhile, Airbnb’s main competitors were no longer other small Internet businesses, but giant hotel corporations such as Hilton, Marriott, and Best Western. And one huge advantage these huge hotel chains offer to travelers is speedy confirmation. Their transactions are fast: by phone or on the Web, you can quickly find out whether rooms are still available and book one for the night you want. That’s because all the rooms in, say, a Hilton are managed by a central computer system, so one call lets you check all the rooms at the same time.
Imagine instead if you had to call Hilton to inquire about each room individually. On any given call, the only thing the reservation clerk could tell you was whether, say, room 1226 at the San Francisco Hilton was available for the night you wanted. If not, you had to make another call to find out about room 1227, then another for room 1228. Booking a room with an Airbnb host was a little like that.
So Airbnb had to figure out how a market with many hosts offering one room at a time could compete more effectively with hotels. Price was obviously important. But it was the spread of smartphones that helped Airbnb close the speed gap, and that may have mattered even more than price. Today, as hosts manage their reservations on their smartphones, they don’t have to wait until they return home to confirm a booking—they just check their phones. They can also, as soon as the room is booked, immediately update their Airbnb listing to remove its availability. That in turn makes it easier for a traveler searching for a room to find one that’s available, even though he or she still has to query one room at a time.
Thus smartphones make the home hosting market work better not just because hosts can respond faster but also because they can update their bookings, which makes them more informative. This, too, reduces congestion (fewer rooms appear to be available, and a room that looks available is more likely to actually be so), and as a result helps travelers search more efficiently, with fewer time-wasting false leads.
Just as Airbnb competes with hotels by making a market for bedrooms, the transportation company Uber competes with taxis by making a market for limos and private cars. In most cities, only taxis have the right to pick passengers up on the street, while limousine services can pick customers up only if they have made a reservation in advance. Those reservations used to take time to arrange. So although limos worked fine for a scheduled trip to the airport or for a fleet of black cars for a conference, if you stepped outside and it was raining, or checked out of a hotel after a leisurely breakfast, hailing a taxi was much easier.
Once again, smartphones changed all that. Now you can call a limo almost as easily as a taxi. So a lot of limos that once sat idle are now readily available. And that’s just the beginning. UberX and companies such as Lyft are even more like Airbnb: they are starting to make a market for the vacant passenger seats in private cars. Speed is of the essence in making these markets work differently than the “day ahead” market that already existed for limos. In some respects, these services are faster even than taxis: you don’t have to take time to pay when you reach your destination, since the same app through which you called the car will also pay the bill automatically through your credit card. This works because your original call goes through a central clearinghouse that archives your credit card details.
Perhaps you are beginning to see a pattern here and can anticipate (even build) marketplaces that don’t yet exist. Limousines have been around for a long time, and there have always been spare bedrooms that you could arrange to rent through friends. But computers and smartphones have helped Uber and Airbnb build multibillion-dollar businesses by making those markets thicker and quicker, bigger and less congested.
Opportunity Knocks
The opportunity to do well by building a good marketplace can arise whenever there are desirable but underused resources that take too much time to find and transfer.
How about the tickets you bought for a game or a show that you can no longer attend—or tickets that you really want but are sold out? StubHub is now making a market for those tickets. In 2007, StubHub was acquired by eBay, which may be the first Web-based company to make a market for things that were sitting idle in someone’s garage or attic. (By the way, speed is becoming important to eBay, too. Whereas most items were originally sold by auction, today most are sold at a fixed price. That’s a faster way to do business, because you can buy what you want as soon as you want it, without having to wait for an auction to end—and taking the chance of losing and having to try again in another auction.)
Reservations at fancy restaurants, which are often booked far in advance, are similar to unwanted tickets. No companies have yet made a large-scale market for restaurant reservations as I write this in mid-2014—you’re still pretty much stuck with scalpers and concierges—but some start-ups are giving it a try, in cooperation with the restaurants whose reservations they’ll offer. Stay tuned.
And just as your spare bedroom might have languished unused before Airbnb made a market for it, your home Wi-Fi might be lying idle during the day while you are at work. If I’m your guest, I can ask you for your password. But if I’m driving by your house, even though my phone may detect your protected Wi-Fi, I can’t access it, much less pay you for its use. Yet as I write this, a start-up called BandwidthX (I’m on its advisory board) is figuring out how your cell phone provider might be able to access this “Dark Wi-Fi” automatically. This could get travelers better connections when they have a weak signal or the system is overloaded, and it could get home owners reduced rates on their Wi-Fi by letting them sell access to it.
In all of these new markets, the entrepreneurs building them have had to figure out
how to make the market thick by attracting lots of buyers and sellers;
how to overcome the potential congestion that could result—that is, how to make the market quick even when it was thick; and
how to make the market safe and trustworthy (we’ll return to this later).
For Uber and Airbnb, dealing with congestion was a make-or-break issue. If making a match between cars and passengers became too frustrating, lots of passengers would go back to using taxis. If it became too hard to book a room in someone’s house, travelers would go back to booking hotel rooms. Congestion threatened the very thickness of the market that makes them both big businesses.
It can be hard to notice how dangerous congestion is to a market, because most successful markets have found a way to deal with it, and most markets that can’t deal with it fail to become big and thick enough for us to notice them. But congestion makes markets work badly even when it doesn’t threaten their ability to remain thick, with an abundance of participants on both sides.
Consider the matching market for putting children in public schools. New York City has lots of schools and lots of students, and the New York City Department of Education doesn’t have to worry too much about losing either. But the DOE did have to find a way to deal with congestion before it could do a good job of deciding who gets assigned to which schools. By looking at New York, we can see clearly the problems that congestion can cause and how they made summers hotter and longer for students waiting to be assigned to schools.
If You Can Make It There . . .
If markets can be too slow even using the Internet and smartphones, imagine how congested they can be when conducted by mail. In comparison to newer forms of communication, the mail has a leisurely quality to it. Even if people wait impatiently for an important letter to arrive, they may still feel that they can take a little time before replying.
I learned of just such a problem when my phone rang one day in 2003. On the line was Jeremy Lack, director of strategic planning for the New York City Department of Education. Jeremy’s boss, Chancellor Joel Klein, had been charged by Mayor Michael Bloomberg with reasserting mayoral control over the largest municipal school system in the country.
New York’s public schools had operated in a decentralized way for years, with autonomy given to individual school principals and community school boards. Mayor Bloomberg and Chancellor Klein aimed to give the almost 90,000 new ninth graders each year a meaningful choice of which of the city’s hundreds of high schools they might attend.
Shortly after that, Parag Pathak, a star graduate student who’d taken my course on market design at Harvard, came to my office in search of a project. He told me that he wanted to combine economic theory with “something real.” I suggested that he help me dig into the New York schools. Today he’s a professor of economics at MIT and an expert on school choice. We also drew in Atila Abdulkadiroğlu, who’s now a professor at Duke University but was then conveniently located in New York at Columbia University.
With so many students and hundreds of schools, the market for school places in New York was plenty thick. But it was also unbelievably congested.
In those days, New York ran a complicated paper-based high school admissions system. Children preparing to enter high school, assisted by their families, filled out forms listing up to five high schools they liked in rank order. The DOE collected these forms and sent copies of them to each of the schools listed. These became the students’ applications to the schools.
Some schools were required to admit students by lottery, but others could choose whom to admit or wait-list. After the schools had decided, the DOE sent letters informing the students where they’d been admitted. Each letter asked the applicant to choose a school (if the student was admitted to more than one) and a waiting list (if the student wished to remain on such a list for a school he preferred over the one to which he’d been admitted).
The rules also stated that students could accept no more than one offer and one waiting list. Schools that had some first-round offers rejected could then make new offers, and the DOE sent out a second round of letters. After students replied, there was a third and final round of letters and replies. Students without seats after the third round were assigned by the DOE, typically to the school with open seats closest to their home.
The result of this complicated system was a mess. Many students couldn’t get seats in any of the schools to which they’d applied—and when they were assigned, it was usually in August, just before school began. Many students didn’t take part in the process at all but instead, amid the chaos and congestion, slipped into schools through unofficial channels.
The reality was that three rounds of processing applications for 90,000 students just couldn’t place everyone.
Meanwhile, the system was also congested: there wasn’t time for more than three rounds. Only about 50,000 students received offers in the first round, and of them about 17,000 received multiple offers that had to be accepted or rejected before the next round could start.
You can see how that would slow things down. Even if you were accepted to your first choice, sending in your letter of acceptance in a hurry might be a low priority for you. Instead, you might celebrate for a day (or three) before mailing your reply to the DOE. And if you weren’t accepted to your first choice, you might want to consult with neighbors, friends, teachers, and others before deciding which waiting list you should stay on—and that could take time, too. So the problem wasn’t just that “snail mail” can be slow but also that decisions take time. And a lot of decisions had to be made by a lot of people as the process played out.
By the time the third round concluded, about 30,000 students still hadn’t been accepted to a school on their list. So they had to be assigned somewhere by the central office. Think about that: 30,000 kids and their parents, sweating out one of the most important events to date in their lives, right down to the last hot moments of the summer before entering high school.
That was frustrating enough, but congestion wasn’t the only problem. To many parents, the whole system seemed risky, unsafe, and untrustworthy. There was, officially, an appeals process and an “over-the-counter” process for assigning students who had moved or were otherwise unassigned before school began. But savvy parents knew that they could also appeal directly to principals, since schools didn’t have to include all their available places in the centralized process. The result was a robust “gray market” of motivated parents circumventing a system they viewed as opaque at best, and biased or corrupt at worst.
“Joel sensed that kids were getting in schools based on things other than the merits of their applications,” says Tony Shorris, deputy chancellor for operations. “Well-connected kids had an advantage.”
Another aspect of the old system that made choosing schools tricky was that the principals saw the students’ preference lists, so they knew where the students had ranked their schools. In response, a lot of schools would not admit any student who had not ranked them first.
Given the specialized nature of some of New York City’s high schools, this might seem sensible. Say you were the principal of Aviation High School, located between New York’s LaGuardia and Kennedy airports. Your mission was to prepare students for careers in aviation. You might have wanted to accept only those kids who felt called to that industry, and you might have assumed that they’d indicate this interest by ranking Aviation first. But if you restricted admission to those students, no one who ranked Aviation second could get in, and some students for whom Aviation was in fact their second or lower choice would face the strategic decision of whether to rank the school first nevertheless.
A student whose genuine top school was hard to get into would give up any chance of going to Aviation if she revealed her true preferences. The DOE implicitly acknowledged this in its 2002–3 directory of high schools, which advised that students, when ranking schools, should “determine what your competition is for a seat in this program.” In other words, the school system was telling kids and their parents to calculate and strategize, not just think about which schools they liked. This meant that even if Aviation’s principal admitted only students who ranked the school first, he had no guarantee that it really was their first choice. In truth, it was only the best school they thought they could get into if they ranked it first.
Families weren’t the only folks strategizing. The system wasn’t safe for school principals either, who often felt pressed to game it by concealing capacity, and holding back empty seats, until after students had been assigned. They then tried to fill those seats with students who weren’t happy with their assignments. The New York Times later quoted a deputy chancellor as saying, “Before you might have had a situation where a school was going to take 100 new children for 9th grade, they might have declared only 40 seats and then placed the other 60 children outside the process.”
But the most pressing problem, the one that prompted the call to me for help, was the massive crowd of 30,000 students who could not go to any school they’d chosen and instead had to be assigned, administratively, at the last minute. And that, too, was a problem of congestion, of not enough time to make enough offers of admission, and enough acceptances and rejections, to reach every kid who needed a spot, while still allowing families to have input on their kids’ fate.
Decongestant
Thick markets need to be quick, but it’s hard to be quick—no matter how fast the technology—if people have to wait for other people to make and act on their decisions. At the beginning of the New York City high school choice process, schools could make offers without waiting for anyone. But once they made some offers, they had to wait for replies before they could make new offers, and that caused the market to be congested, and too slow to process every student in time.
Consider how this works in the market for houses. When you’re selling a house, the asking price is offered to everyone. And when you’re buying a house, you can consider every house on the market.
But suppose you decide to make an offer on a particular house. A typical offer comes with “earnest money” and is a signed offer (backed by a deposit) to buy the house at a specified price. To give the seller time to consider the offer, it typically also comes with a specified duration, perhaps twenty-four hours—a little longer in a slow market and shorter in a hot market like the one in which I now find myself in Silicon Valley.
During the time the offer is open, most buyers can’t afford to make an offer on another house, so they must wait for a response. But in a hot market, while you’re waiting, another house on which you might want to make an offer may get sold. Similarly, if the seller makes a formal, signed counteroffer, he will have to wait for your decision, and he doesn’t want to leave it open any longer than is necessary. So although the market starts with everyone talking to everyone else, it abruptly turns into a series of private conversations between one buyer and one seller.
We’ve seen how job markets can be like that, too. A company looking to fill an important position can survey the market and interview a lot of candidates, then consider to whom to make an offer. But once the offer is made, the company often has to give the candidate a little time to consider it, and during that time the other candidates might take jobs elsewhere and become unavailable. As we saw in chapter 5, that’s especially true in a market in which offers are exploding and everything happens faster, so that companies may feel compelled to make exploding offers themselves.
Matching markets often have to deal with congestion, since each offer isn’t just a set of terms, it’s a proposal of a match to a particular counterpart. (In the market for houses, a buyer may not care deeply whom he buys from, but he certainly cares which house he buys, so markets for houses need to make matches, too.) Commodity markets usually have somewhat less trouble with congestion, since the offer to buy or sell shares of stock or bushels of wheat can be made to the whole market, and the buyer or seller can change a bid or an ask at any time, without having to wait for anyone else. But in matching markets, some offers have to wait for the decisions of others.
The key to speeding up the New York school choice system wasn’t just to design a computerized process, although that was part of it: computers are fast. But people are slow, and both the decisions that people make and the information they have about which schools would be best for their kids are critical to the market’s success.
The solution to decongesting New York’s high schools turned out to be letting people indicate their school preferences all at one time and then using those preferences to process decisions quickly. When preferences are submitted in advance, a computerized clearinghouse already knows which school a student would prefer if he was given a choice of two.
The same system can work in other matching markets, such as job markets, in which an applicant can indicate her preferences among the jobs she has applied for all at one time.
With various colleagues, I have helped design computerized clearinghouses that are now used to process preferences for schools, jobs, and workers that people decide on before the computer is turned on. The computer can then run through offer-rejection-offer chains fast enough that there’s time for every offer that someone wants to make to be made.
The key to such a clearinghouse is making it safe for people to state their preferences honestly. So before I start to describe the clearinghouses we’ve built, let’s talk more about safety.