WHEN THINGS START TO THINK | Chapter 6: Smart Money
May 15, 2003
- Neil Gershenfeld
Originally published by Henry Holt and Company 1999. Published on KurzweilAI.net May 15, 2003.
Barings Bank was founded in 1762. In its long history it helped to finance the Louisiana Purchase (providing money Napoleon needed to keep fighting his wars), and counted the Queen among its loyal customers. In January of 1995 a twenty-eight-year-old trader for Barings in Singapore, Nick Leeson, lost most of what eventually proved to be $1.4 billion by trading futures in the Japanese Nikkei Index. That was twice the bank’s available capital; by February the bank had folded, and in March it was sold to the Dutch bank ING for £1.
In July of that same year, Toshihide Iguchi, a trader for Daiwa Bank in New York, confessed to the bank that he had lost $1.1 billion trading apparently harmless U.S. Treasury securities. By November Daiwa was forced out of the United States for concealing the losses.
The next year Yasuo Hamanaka, a copper trader for Sumitomo, dwarfed these scandals by admitting to having lost $2.6 billion in copper trading. In the preceding years he had single-handedly kept the copper market artificially inflated, leading to the digging of unnecessary new mines and a subsequent glut of copper. After his manipulations were revealed the price collapsed by 25 percent.
Losing large sums of money no longer requires even the specialized skills of a rogue trader; Robert Citron, the duly elected treasurer of Orange County, California, lost $1.7 billion by investing in derivatives. This led to the largest municipal default in U.S. history, by one of its richest counties, no less than the birthplace of Richard Nixon.
As quickly as enormous amounts of money are vanishing into economic thin air they are reappearing elsewhere. Markets are valuing bits far more than atoms. In 1997 U.S. Steel had twenty thousand employees, sales of $6.8 billion, and assets of $6.7 billion. The total value of U.S. Steel stock was $2.9 billion. Last year Yahoo, the company with the popular Internet index site, had just 155 employees, sales of $67 million, and assets of $125 million. But the total value of Yahoo stock was about the same as U.S. Steel, $2.8 billion. Netscape already had kicked off the frenzy of public offerings of stock in Internet companies, with a first day of trading that brought the net worth of the company to $2.2 billion.
The market’s conversion rate between atoms and bits has been running at about ten to one. The stock market values Ford and GM—which together have a million employees and assets of $500 billion—at $100 billion. The stock for Microsoft and Intel, which jointly have about a hundred thousand employees and assets of $50 billion, is worth $300 billion.
The distribution of this new wealth is distinctly uneven. According to the Census Bureau, in 1968 in the United States the top 5 percent of households received 16.6 percent of the total household income; by 1994 that had climbed to 21.2 percent. Similar redistributions are happening globally: World Bank numbers show that poor- and medium-income countries held 23 percent of the world’s wealth in 1980, dropping to 18 percent eight years later.
There’s something happening here. The economy appears to be divorced from the laws of physics, creating and eliminating great wealth with little apparent connection to the material world. And that is exactly the case: money is increasingly digital, packets of data circulating in global networks. The electronic economy is weaving the digital and physical worlds together faster than anything else. The demand for more sophisticated ways to manipulate money is forcing computing to go where money goes, whether in a trader’s workstation, a smart card in a wallet, or a thinking vending machine.
Freeing money from its legacy as a tangible asset carries with it great promise to make the economy more accessible to more people, but we’ll be in trouble if we continue to act as if money is worth something. The bits of electronic cash still retain a vestigial reflection of their origin in the atoms of scarce resources.
Just after World War II, the Bretton Woods Agreement fixed a conversion rate between dollars and gold at $35 an ounce, and in turn set exchange rates between the dollar and other currencies. The U.S. government was obligated to convert dollars into gold upon request, using the cache held at Fort Knox. In 1971, saddled with a dwindling gold supply, a persistent recession, and an expensive war in Vietnam, President Nixon took the dollar off the gold standard. Since then, the value of currencies has been fixed solely by what the markets think they are worth. We maintain a polite fiction that money stands for a valuable commodity, but it now represents nothing more than beliefs.
The implications of the end of the gold standard reach far beyond the passionate debates of that time, because of an apparently unrelated event that was quietly happening in a handful of laboratories: in 1971 the ARPANET entered into regular service. This was a project supported by the U.S. Defense Department Advanced Research Projects Agency (DARPA) to create a computer network that had no central control and multiple ways to route messages, just the thing to survive a nuclear attack. In 1971 it had all of fifteen host machines. Its original research function was soon forgotten as the network filled with people sending e-mail, and connecting more computers. With millions of hosts on it by 1989, DARPA gave up on trying to manage the experiment, and the Internet was born.
At the end of the gold standard, computing was done by relatively isolated mainframes; computers were needed to record financial trends but did not create them. With the arrival of instantaneous global connectivity and distributed computing, money can now travel at the speed of light, anywhere, anytime, changing its very nature. The only physical difference between a million and a billion electronic dollars is the storage of ten extra bits to fit the zeros; money has become one more kind of digital information.
The apparent function of money is clear. It has a transferable value; I can exchange a dollar bill for a carton of milk because the grocer can then exchange the dollar for something else of comparable value. An electronic representation alone does not change this. Digital money has been around for years in the form of electronic funds transfers done by financial institutions, transactions that differ in scale but not in kind from buying milk. They could in principle be done equally well by moving around (very) large amounts of physical currency. The spread of the Internet has now brought electronic commerce into the home. The grocer can send a packet of data describing the milk to your Web browser, and you can send a packet of data back to the grocer with your credit card information.
There’s a popular misconception that it is dangerous to transmit a credit card number over the Internet. This dates back to the time when network traffic was not encrypted, so that a malicious or even just curious person with the right software could connect to a network and read the contents of the packets sailing by. This was an eavesdropper’s dream, because the data came to you instead of your having to go to it. Although I’ve had taxi drivers lecture me on the dangers of the Internet, the reality now is that there are solid cryptographic protocols that make this kind of network sniffing impossible. In a “public-key” cryptosystem you can openly advertise an ID much like a phone number that lets people easily encode a message before sending it to you, but that makes it impossible for anyone other than you to decode it unless they know your secret key associated with the public key. This can be done by using what are called secure socket network protocols, built into most Web browsers.
Modern cryptography goes much further, splitting hairs that you might not have realized you had. If you don’t trust the grocer, a “zero-knowledge proof” lets you give the grocer enough information to be able to bill you without letting the grocer learn either who you are (so you don’t have to worry about getting junk mail) or what your account is (so you don’t have to worry about later unauthorized charges). Conversely, by using a digital signature, the grocer can prove that an advertisement came from the store and not someone trying to spoof store business. Electronic certificates permit a third party whom you trust to remotely vouch for the grocer’s reliability before you sign a contract for milk delivery.
Smart cards build these cryptographic capabilities into your wallet instead of your Web browser. A smart card looks like a credit card but acts like a dollar bill. A chip in the card stores electronic money that can be spent directly, unlike the account number stored in a credit card that needs access to a central computer to be used. The first widespread use of smart cards came in pay phones in Europe, where they eliminated the need to fumble through a Pocket-of-Babel’s worth of change to make a local phone call. They’re now expanding into retailing so that both buyers and sellers can stop carting around currency.
After all, what could be more useless than a penny? They travel between bowls on shop counters and jars on bureaus because it’s too much trouble to take them back to the bank. There’s a solid economic argument that the net cost of handling pennies exceeds their worth, therefore they should be eliminated as the smallest currency unit. The future of the penny does not look bright.
Qualify that: the future of a physical penny does not look bright. The value of electronic pennies is just beginning to be appreciated. On the Internet the packets of data representing financial exchanges are a drop in the bucket compared to high-bandwidth media such as video. There’s plenty of room to send much more of them. The total amount of money in circulation can’t go up, but the size of a transaction can certainly go down. The overhead in selling a product sets a floor on what people are willing to purchase. Either you buy a CD, or you don’t. If, however, the CD arrives as bits on the network, you could pay for it a track at a time. Rather than buying an expensive video game cartridge, you could be billed for each level of the game as you play it. Your taxes that pay for garbage collection, or highway maintenance, could be collected each time you take out the trash or drive your car. The incentive for a consumer to consider buying this way is to be able pay for just what gets used rather than paying in advance for the expected lifetime cost of something. The advantage for a merchant is the opening up of new markets by decreasing initial purchase costs, the ongoing revenue stream, and the detailed and continuous usage information that gets provided (if the buyer agrees to make that available).
As interesting as it is to introduce micropayments and cryptography into commerce, these are just ways to repackage our old model of money, analogous to your buying milk in lots of tiny containers, or visiting the store with a paper bag over your head to protect your privacy. Most electronic representations of old media start out by emulating their functionality and then later generalize beyond the hidden assumptions. For example, the first formats for digital pictures stored an array of intensities to match each part of an image, before people realized that a picture could be stored in order of importance so that transmitting a small amount of data provides enough information to produce a small image that can be refined as more information is received, or identifying data can be hidden in the image to catch forgery, or the parts of a picture can be encoded separately so that they can later be rearranged. The familiar notion of a picture has grown to encompass many new functions. Likewise, moving money as bits instead of atoms creates an opportunity to expand its identity far beyond the familiar forms that it has taken in the long history of commerce.
Each day, over a trillion dollars circulates in the global currency market. That’s a lot of money. The bulk of this activity has nothing to do with getting francs for a trip to Paris; it is a massive hunt for profit and security in the inefficiencies of the market. Now this is very different from buying a gallon of milk. There is no intention to use what is bought or sold; the real content of a currency trade is the information that motivates it. What’s being exchanged are opinions about whether the yen is overvalued, say, or whether the Fed will lower interest rates. Financial speculation isn’t a new idea, but the size is. The sums need to be so vast because the trades are exploiting minuscule differences in the value of currencies that show up only on the scale of a national economy. And they can be so big because the combination of an electronic representation of money with a global computer network makes taking a $100 million foreign exchange position no more difficult than sending an e-mail message.
The dollars used in a large financial trade have a very different character from the dollars used to buy a gallon of milk, but we continue to presume that they are identical. The dollars that you use to purchase milk are useful in small quantities, and essentially all of their value goes into paying for the milk (there’s just a bit of extra demographic information that the seller can collect about where and when you bought the milk). The dollars that a trader uses are useful only in big quantities because of the small margins and the large minimum size of the trades, and there’s no desire to actually use what is being bought. This distinction can be thought of as atom-dollars and bit-dollars.
People, and countries, that don’t have access to bit-dollars are increasingly being left behind, unable to compete in the digitally accelerated economy. The corner grocer has to move a lifetime of milk to match the resources committed by a few keystrokes on any trader’s keyboard. And people who can participate in the new economy are seeing bigger and faster swings between spectacular riches and ruin because normal economic fluctuations can be magnified almost impossibly large.
Just as it was once thought that printed dollars needed to be backed up by a gold standard, we now assume that bit-dollars must be interchangeable with atom-dollars. The end of the gold standard was an economic leap of faith that remains controversial, but it arguably is what enabled the rapid ensuing economic development in many parts of the world. Now, given the spread of computation and a digital representation for money, the conversion between electronic and tangible assets can depend on the nature of a transaction. It’s possible to end the atom standard.
We already accept that the grocer cannot receive a dollar for a carton of milk and then directly exchange it for a currency position, because of the large minimums and rules for foreign exchange market access. The converse would be to have $1 million made in a currency transaction buy something less than a million gallons of milk. This could be implemented by making the value of a dollar depend on its context. Its purchasing power could vary based on how many dollars it was earned with and then how many dollars it is being spent with.
The $1 million earned in a trade could still buy a million-dollar dairy, but for individual purchases at the corner market it would be worth less than dollars that had been earned more slowly in smaller quantities. Such a scaling is not a simple devaluation or transfer of wealth from the rich to the poor; it’s a recognition that the money is being used in entirely different ways and it is not at all obvious that the conversion factor should be one-to-one. Considering such a change raises enormous economic, political, and social questions, but now that it is technically possible to do it, choosing not to is itself an even more significant decision.
There’s a precedent for valuing something based on a rule that depends on its history: that’s what derivatives do, the new financial instruments that are now so painfully familiar to Orange County. A derivative sets a price based on the behavior over time of an asset, rather than the worth of the underlying asset itself. For example, an option could be designed that would pay off if the activity in a given market increases, regardless of which way the market moves. These instruments were developed because the markets are so random that it is all but nonsensical to forecast the daily values but quite reasonable to forecast changes in attributes such as the volume of trades. What’s interesting about options is this introduction of algorithms to the price of something, so that value can be associated with change.
Derivatives themselves are not dangerous. The reason that they are associated with so many billion dollar boo-boos is that it’s not possible to assess their present worth, or your vulnerability to changes in the market, simply by tallying up an account balance. In a financial firm the traders are at the top of the pecking order, surrounded by the most advanced computers and data analysis techniques to guide and make trades. Then there’s a risk assessment group that trails after them trying to determine the firm’s exposure to risk. The latter is generally a sleepy enterprise located far from the trading floor, lacking the tools of the former. This imbalance is a recipe for disaster, because committing resources to a position is easy compared to evaluating the financial risk of a portfolio of derivatives, which requires a detailed understanding of the state of the markets and how they are likely to move. In this unequal competition, the trading side almost always wins out over the risk assessment side. Some firms effectively give up and ask the traders to evaluate themselves. Tellingly, Leeson at Barings, and Iguchi at Daiwa, both kept their own books because their managers from an earlier generation did not feel capable of supervising them.
The solution is to recognize that the further that assets get divorced from underlying resources, the more necessary it becomes to merge spending with monitoring. Each new algorithm for valuing something must be mated to a new algorithm for assessing it. Back-end accounting then moves to the front lines, drawing on the same data feeds and mathematical models that are used for trading. Too many people still think of smart money in dumb terms, assuming that like gold bars it has an intrinsic worth that won’t change very much if you look away from it. Understanding the implications of a derivative outstrips human intuition, particularly as computers, instead of people, increasingly initiate trades. Since physicists have spent centuries trying with limited success to understand the behavior of complex interacting random systems, it’s too much to expect one trader to be able to do so (the physicists on Wall Street certainly haven’t been able to). The money must do a better job of supervising itself. This is a cultural problem more than a technological one, requiring an acceptance that the bits representing money should come packaged with bits representing algorithms for tracking their worth.
Stepping back to recognize that it is more natural to view electronic money as comprising the combination of data describing quantity and algorithms specifying valuation, it then becomes possible to create combinations that are useful for purposes far from high finance. Think what could happen if ecash contained the means for ordinary people to add rules. A child’s allowance might be paid in dollars that gain value based on the stability of the child’s account balance, to make the benefits of saving apparent in the short term instead of just through the long-term accrual of interest. Or a store’s price guarantee could be implemented by pricing an item in dollars that stay active after the transaction, and that have a value derived from the market for that item. You don’t need to search for the lowest price; the money can. If a competing car dealer lowers the price on a car the day after you bought one at another dealer across town, that information can automatically initiate a transfer between your dealer’s bank account and yours to refund the difference. The dealer gets your business if it thinks it can always beat the competition; you get the freedom from worrying about finding the best deal.
If money can contain algorithms then this kind of oversight could be built into every transaction rather than being a matter of fiscal policy alone. Central banks have just a few coarse levers with which to manage a nation’s economy, primarily controlling the prime interest rate that the government charges banks. If it’s low, new money gets printed at less than its real worth, in effect making a loan from the future to the present. That may be a good idea in a depression, but otherwise it saddles your children with your debt. If the rate is high, then money costs too much and people are forced to sacrifice now for the sake of saving for the future. This one number applies to everyone, everywhere.
Once money is manipulated by machines, then it can be personalized to reflect both national and local interests. Dollars could have a latency period when they’re not valid after transactions, to slow down panic spending. In a time of economic turmoil this would mean that you might have to wait a day after withdrawing your life savings from a bank before you could put all the money into a get-rich-quick scheme. To encourage regional economies, dollars could have a value that decreases with distance from their origin. Boston dollars would not go as far in Texas, and vice versa. Or, to reflect true environmental costs, dollars might be worth less when purchasing nonrenewable resources than renewable ones. These are not matters to be settled once for everyone everywhere; they are dynamic decisions to be made by buyers and sellers on an ongoing basis.
Such goals are currently implemented by familiar mechanisms such as tax policies and sales contracts, without needing something so ambitious as a redefinition of the nature of money. But as the money, and the information about how the money can be and is being spent, merge in packets of economic information running around networks, the distinction between the money and the supporting information becomes less and less meaningful, until it becomes simpler to recognize the financial data packet as a new definition of money.
Just like any other digital media, for this to succeed there must be a vigorous community of money developers working with open standards. If any economic or technological lesson has been learned over the last few hundred years, it’s that central planning cannot substitute for the creative anarchy of relatively free markets. It would be a shame if the dollar bill had to relearn all of the mistakes of previous computer standards, which have a nasty habit of rediscovering old errors and not anticipating future growth. A number of computer operating systems started out using 16-bit addresses to identify memory locations. This is enough to use 65,536 addresses, which at one time seemed unlimited. As hardware and software both expanded beyond this boundary, awkward patches were needed to use some of the precious addresses to point to other blocks of addresses that could be swapped in and out as needed. Because the number of address bits was fixed by the operating system, this problem continued to recur long after it was clear that sixteen bits wasn’t adequate. Similarly, most every network standard (including the IP protocol that runs the Internet) has run out of addresses and needs a redesign. The solution to this problem of lack of foresight is not better foresight, it is more humility. Since it’s not possible to anticipate future growth, standards are increasingly designed to let independent developers add new capabilities. One reason that the computer game Doom has been so popular is that the authors released it with a specification that lets other people add new worlds and new capabilities to it. There’s a lively little industry that has grown up around the game, continually reinventing it in a way that one team of programmers alone could never do.
Opening up fiscal policy to the people who brought us Doom is admittedly a terrifying prospect, but it’s instructive to compare the history of the Internet and Bitnet. In the dark ages of computing one didn’t send e-mail, one “sent a Bitnet.” The Bitnet was an international network of mainframes organized along the line of the Soviet economy with central planning and management; the Internet was (and still is) a relatively anarchic collection of distributed switches and collectively supported evolving protocols. Guess which one survived? Bitnet was shut down in 1996, a centralized monoculture that could not keep up with the frenetic growth and distributed innovation of the Internet. Even with the occasional glitches, the Internet is perhaps the largest and most reliable system ever constructed, precisely because there is no central location to fail or mismanage and so many people can contribute to its development.
The history of computing is littered with unsuccessful standards that sought reliability through specification rather than experience. When the car companies and other heavy industries wanted to install computer networks in their factories they decided that what everyone else was using was not reliable enough, so they came up with their own standard that was supposed to be foolproof. They now use the open standards for networking that everyone else does. They found that reliability depends much more on having a diverse community working on developing and deploying the networks than on trying to control their design. As a result, intelligence is now distributed throughout factories rather than being centralized in a few computers, or people.
Conversely, there are many bad standards that have been very successful because they came along at the right time. If a standard gets specified too soon, it reflects early ignorance rather than useful experience. If it comes too late, it’s irrelevant no matter how good it is. In between is a time when a standard appears too soon to really know how to write it well, but soon enough to be able to have an impact. MIDI, the protocol for connecting electronic musical instruments, is a good example of a bad standard (because it can’t scale up to handle installations larger than what was originally envisioned), but it arrived at just the right moment to create a new industry. You can hear the result in almost any popular music recording.
The economy is now ready for something similar. Just as technological improvements in the tools to create sounds or send messages have led to a transformation of our notion of what it means to make music or interact with others, the increasing manipulation of money by machines rather than people presents an opportunity to transform the global economy to better reflect personal needs. It’s not yet clear exactly how to write an open standard for smart money that merges value with behavior, but markets are already doing this imperfectly, and history provides strong guidance for how to go about creating standards for reliable, scalable, digital systems. These are increasingly questions for programmers, not economists.
Commerce is where the physical world and the digital world will always meet. A purchase must become embodied to be usable. As electronic information appears in more and more of those transactions, it is essential that money be freed to complete its journey from a tangible asset to a virtual concept. Buying and selling is increasingly mediated by smart systems that still use dumb money. Abstracting money from its legacy as a thing will make it easier for still smarter things to manipulate it, helping them better act on our desires.
WHEN THINGS START TO THINK by Neil Gershenfeld. ©1998 by Neil A. Gershenfeld. Reprinted by arrangement with Henry Holt and Company, LLC.