Streetwise Professor

Update. I followed up on the “note to self” on Internet pricing (see below). There were indeed several papers on Internet pricing schemes that were intended to address the congestion issue. Salient examples include Jeffrey MacKie-Mason & Hal Varian, “Some Economics of the Internet” and the same authors’ “Pricing the Internet.” Alan Wiseman’s “The Internet Economy” (2000) summarizes several pricing proposals. Perhaps not surprisingly, several Internet pricing proposals parallel the trading system capacity market schemes that I outline in this post. The tâtonnement process is similar to the MacKie-Mason & Varian auction approach. The capacity right system is analogous to David Clark’s (1997) expected capacity pricing mechanism. End Update.

After a series of embarrassing technology-related stumbles in 2005, the Tokyo Stock Exchange outdid itself with its early closing on Wednesday, 18 January 2006. The TSE’s electronic trading system has a built-in constraint that prevents it from handling more than 4 million transactions (regardless of their size) in a day. A flood of sell orders washed over the exchange when it was announced that Japanese authorities had raided the offices of internet highflier Livedoor during an investigation of securities fraud.

This TSE constraint is a new one on me. I was well aware of what one might term flow constraints—the number of messages (bids, offers, cancellations, transactions, confirmations) that a system can handle per second or per minute, but I was heretofore unaware of the possibility of an “integral constraint”—specifically, that the sum of trades over a day not exceed some preset limit. I’m not a hardware expert, but I would imagine that this constraint arises due to a limitation on system storage. Each trade requires storage of a record that details all relevant characteristics of each transaction—price, quantity, date, time, broker, buy/sell, party, counterparty, etc.—regardless of trade size. Thus, whereas flow constraints are attributable to limits on bandwidth or processing capacity, the TSE constraint is likely due to a limitation on the ability to store all of the transaction records.

Any electronic trading system is subject to capacity limits. Eurex had some capacity issues in the late-90s. NYMEX is considering an upgrade to its electronic trading systems and a possibility of using GLOBEX due to the capacity constraints inherent in its legacy ACCESS system (which experienced problems when volume surged after the London terror attacks in July, 2005).

Every open outcry system is also subject to capacity limits: recall the bottleneck created by slow printers at NYSE specialists’ posts during the October ’87 crash, or the “back office crisis” of the 1960s. Open outcry exchanges are constrained by their capacity to direct orders to the floor, and the capacities of floor brokers and market makers to process, execute, and financially carry trades. (OTC markets have capacity constraints too—cf. the CDO market.)

Capacity is limited because it is costly. It is uneconomic to have capacity to handle all possible contingencies. Therefore, exchanges must decide on how much capacity to install. This raises two issues.

First, how much capacity is optimal? Second, how does one allocate scarce capacity?

These questions are closely related. In particular, if there is a price mechanism that allocates capacity efficiently, and which discovers the shadow value of a unit of capacity, exchanges will be able to make good capacity choices. In the absence of a price-based allocation mechanism, it is much harder for exchanges to choose the right capacity.

The trading industry is not the only one that must answer these questions. These have been key questions in the electricity business since Edison first flipped the switch on the Pearl Street Station in New York in 1882. Since 124 years later the electricity industry is still experimenting with different institutional approaches to capacity allocation and planning, we shouldn’t be too sanguine about the prospects for a quickie solution in financial trading.

Interestingly, there are some similarities between the trading and power capacity conundrums. In particular, at least historically in the electricity business the ultimate consumers of electricity have not paid prices that fluctuate in real time to reflect fluctuations in the scarcity of generating and transmission capacity arising from demand changes, generation outages, etc. Steven Stoft calls this the “First Demand Side Flaw” in electricity markets. Although there are numerous proposals to improve metering and measurement technologies to permit this, widespread adoption of such measures is not likely any time soon.

Similarly, at present, securities and derivative trading systems do not have system usage fees that fluctuate with system usage. Indeed, most systems charge fees on the basis of completed transactions and charge zero prices for other actions that utilize system resources, such as the submission of bids and offers, changes in standing bids and offers, and cancellations. In essence, users pay a zero price at the margin for their consumption of system resources.

This is an issue that has been analyzed in the property rights economics literature that builds on Coase’s insight that it is costly to operate a price system. In the electricity context, real time pricing involves investment in metering equipment. This is a non-trivial cost. In the exchange context, it is costly to monitor, record, store, and bill every system user action. Indeed, creating such a capability would require additional investment in system capacity. Moreover, such a system imposes costs on users, such as, inter alia, the cost of monitoring the accuracy of billing. The question arises, then, as to whether the costs of implementing a capacity pricing system are smaller or larger than the benefit. The benefit arises from the fact that when a scarce resource is unpriced it is overutilized. (Cf. Barzel’s analysis of excessive search.)

Given that the marginal cost curve of using a system is likely close to zero when demands on it are below its capacity, a zero price is right most of the time. When demand bumps against the capacity constraint, however, the zero price is very costly. The exchange must find some non-price means to allocate capacity—such as a shutdown or a system slowdown. The hue and cry over the TSE’s action suggests that a shutdown is very costly (especially as it is likely to occur when information is flowing rapidly and therefore people desire to trade a lot.) Even a slowdown is costly, as traders who due to a slowdown have submitted orders only to find that they cannot cancel or revise them, or who do not know whether their orders have executed, will attest. Traders are typically control freaks who flip out when system problems reduce or eliminate their ability to control their trades and positions.

The problem with non-price rationing is that it impedes the efficient allocation of system resources among different customers with differing reservation prices: those with high willingness to pay may be cut off, but those with a low WTP may be able to trade. Moreover, absence of prices deprives exchanges of information about the value of capacity, which means that they are flying blind when deciding how big to make a trading system.

A price system could conceivably address these problems, and it is interesting to contemplate how such a system would work. It seems challenging—perhaps more challenging than in electricity markets. In electricity, except when generating capacity is completely committed, the value of an additional unit of generation is given by the marginal cost of production, which can be measured with some accuracy in real time (through generators’ bids, for instance). In the exchange context, however, the shadow price of capacity Q is given by the willingness to pay of the marginal consumer. How can this be determined in real time?

One can envision (theoretically) a tâtonnement process. Users submit messages (quotes, cancels, etc.) assuming that the system is unconstrained and the price of system resources is zero. The system evaluates whether these messages exceed system capacity. If so, the price of utilization of system resources is raised slightly, and users resubmit. The price is increased until the capacity constraint is just binding.

The operation of such a tâtonnement process in a continuous market seems impractical. Are there alternatives? One can also imagine the creation of capacity rights and a capacity market. For instance, market users could be required to have capacity rights (which in aggregate equal system capacity) in order to trade. These rights could be traded in a bilateral market—or perhaps the exchange could also provide a mechanism to trade these rights in a continuous market. During most periods the price of the capacity right would be zero. When capacity is constrained, or there is an appreciable probability that capacity will be constrained (one can think of the capacity right as being similar to an option because the value of the right to use capacity is a convex function of capacity utilization, so the capacity price will have some extrinsic value), those who have a high demand to trade can acquire it from those less interested in trading.

Given that settling transactions in anything—including rights to capacity—takes time, I am skeptical that this is a practical way to ensure that those with the highest willingness to pay are not rationed out during a capacity crunch, particularly one that is unexpected (as is likely to be the case in financial markets.) Certainly such a mechanism can ensure the allocation of trading rights to those who anticipate having a high demand to trade during a capacity shortage, so this is better than nothing. That is, those who anticipate having the high WTP can by the capacity option—the right to trade—ex ante. To the extent that ex ante evaluations of WTP track real time WTPs, this mechanism will allocate scarce system resources pretty efficiently and generate information on the value of capacity that exchanges can use to make investment choices.

Implementation of this approach also requires enforcement of the right: how do you know somebody has exceeded their limit, and how do you prevent them from doing so? This seems to me to be a non-trivial problem. Things are likely to be particularly complicated for integral constraints like the one that caused the TSE problems.

To my knowledge, neither the tâtonnement nor the capacity right approach have been even proposed in financial markets, let alone developed. (I do recall reading about pricing of the Internet, which raises similar issues. Note to self: check this out.) This suggests that for the foreseeable future that exchanges and market users will have to live with non-price capacity allocation during periods of market stress.

The absence of a price mechanism that discovers the value of capacity deprives exchanges of information needed to make efficient capacity investment decisions. So how much capacity should exchanges invest in and how do they figure this out? Here again the electricity experience is somewhat informative, and somewhat discouraging. Utility planners traditionally utilized the concept of “value of lost load” (“VOLL”) for system planning purposes. VOLL is an estimate of the cost of “shedding load,” i.e., of turning off customers’ lights in the event that electricity demand exceeds the physical capacity available to produce it. Due to the First Demand Side flaw, there is no equilibrium in prices when the (vertical) demand curve is to the right of system capacity, so it is necessary to shut people off. As Stoft shows, if you assume you can set VOLL correctly (insert economist joke here), market participants will make the efficient capacity choice. After all, VOLL is the opportunity cost of a unit of capacity.

The problem arises, of course, with determining the right VOLL. The VOLL in power markets is typically set arbitrarily. (Australia uses VOLL pricing and has taken a somewhat more systematic approach to estimating it.) Set VOLL too high, and you get too much capacity. Set VOLL to low, and you get too little.

So what is the VOLT—the “Value of Lost Trades”—in a financial market? Beats me, but it is worth some investigation by those who have skin in the game. I’m also curious as to the criteria that exchanges use when choosing system capacity. Do they have some measure of VOLT in mind? Where does this number come from?

In the meantime, it may be worthwhile to cut the TSE some slack. After all, all we know for certain is that an exchange that never hits its capacity limit has inefficiently invested in too much capacity. Without knowing the VOLT, it is much more difficult to evaluate whether the observed frequency of outages is too high or too low. Maybe TSE has the right capacity. Customer complaints about outages aren’t that informative about VOLT, because they don’t directly pay the price for capacity—it’s easy to bitch when somebody cuts off your freebies.

I also wonder how political and regulatory processes will impact exchange capacity planning and investment. The Japanese government has made its disappointment with TSE very clear. Japan’s Economy Minister, Kaoru Yosano has said “[a] Stock exchange that can’t carry on trading simply doesn’t deserve to exist.” There is reason to be concerned that governments will pressure exchanges to overinvest in capacity. Zero tolerance plays well in the press, but ignores the economic reality that trade-offs are part of life. It would be better to step back and apply good economics to evaluate these trade-offs.