Network Reliability As A Public Good, And What To Do About It

The ideas that have captured my attention recently (such as government funding of culture) have a central theme: the extent to which something is a public good, and what those public good characteristics imply for public policy. If a good (or service) has public good characteristics, does that fact necessarily imply that its provision should be centrally coordinated through regulation? Not necessarily.

One example I’ve been working through recently, and that I presented at the recent USAEE meetings in Washington, DC, is network reliability as a public good. Say there is a group of agents who are connected on a network, and their participation in the network generates benefits for each of them. In the case of electric power, being connected to the grid enables agents to buy power from agents who produce and sell power , meaning either that consumers don’t have to produce it themselves (which is prohibitively expensive historically, but becoming less so), or that they don’t have to buy it from the nearest producer.

Being connected through a network generates benefits for the connected agents, but these benefits come at a cost (duh). In addition to the cost of building the network, there is an additional cost of ensuring that the network will operate reliably for the agents using it. Ensuring a particular level of reliability requires investing in network capacity, particularly in peak hours, because one of the biggest causes of reliability reduction on all networks (electric, cell phone, Internet, highways, airports, etc.) is congestion. Does the network have sufficient capacity to provide reliable service in high-congestion periods?

Several different means exist for providing this capacity. One is to build more transmission wires, so the overall capacity constraint becomes less binding. Another is to build generation closer to consumer agents, using the network less. These approaches are costly in several ways. First, obviously, is high capital costs. Second, building more generation capacity closer to load has NIMBY costs, and in a large sense defeats the purpose of being interconnected on a network. But there are other means. We can use voltage-management technologies as springs or shock absorbers at various points in the network. We can ensure that we have good incentives and good capacity for reactive power to balance local network flows. And we can use demand response as a form of capacity building, empowering customers to choose from among a portfolio of contracts that could include real-time pricing, peak-load pricing, or fully-loaded flat rate pricing. Such pricing flexibility has proven and demonstrable benefits for networks, particularly in getting the most bang for the buck out of the existing network capacity without risking to o much loss of reliability. In other words, demand response promotes reliability by using price to increase the network’s load factor. Voltage management technologies and reactive power capacity also enhance network load factor without a cost of loss of reliability. They represent investments in reliability.

The thing about such investments, though, is that when one network agent invests in one of these reliability assets, all of the rest of the connected agents benefit, and the one who pays can’t always exclude the others from benefiting. So that makes reliability a public good, right? And if we apply standard neoclassical public good theory, the fact that reliability is nonexcludable and nonrival means that no one will want to invest in reliability because he/she will not capture all of the additional benefits from the additional investment. Thus we can expect underprovision of reliability unless we have some central coordination to require network agents to pay for the investments in reliability that benefit all of them.

This application of public good theory is misguided, and the resulting policy implications are likely to be inappropriate and costly (and therefore produce inefficient outcomes). For insight into why this pure public good treatment of reliability is misguided, I recommend to you an unjustly under-read paper in economics: James Buchanan and Craig Stubblebine, “Externality,” Economica 1962 [n.b. public goods are just a special case of externality]. Buchanan & Stubblebine start from a standard neoclassical model, where agents maximize utility subject to their budget constraints. But, as in the case of public goods, their agents have interdependent utilities in the sense that one good is consumed jointly, and the total amount available for consumption is a function of the choices of all of the agents. They then go on to derive efficient provision levels and optimality conditions for their model.

Buchanan & Stubblebine provide major insights in analyzing the reasoning underlying the “it’s a public good, and therefore underprovided in equilibrium” result, which I stated above as “no one will want to invest in reliability because he/she will not capture all of the additional benefits from the additional investment”. The flaw in this logic is a straightforward consequence of economic thinking: it’s incorrect because it overlooks the fact that if the agent’s marginal benefit from making the investment exceeds the marginal cost, she’ll do it, regardless of whether or not she can capture all of the marginal benefits. As long as she enjoys enough of them herself, she’ll invest in reliability assets, and there will be no underprovision in equilibrium.

The implications of this insight are profound. First, note that it takes advantage of the fact that different agents on the network are going to have different preferences over reliability. These different preferences mean that for a given level of reliability, some agents will have high marginal benefits and some will have low marginal benefits. That heterogeneity of preferences means that reliability is both a public good and a private good. Yes, individual production and consumption of reliability are interdependent, but when we recognize the diversity of preferences over reliability, the next logical step is to acknowledge that for high marginal value agents, reliability is a private good for which they are willing to pay beyond the lower levels that low marginal value agents would prefer.

That observation leads to the second major implication of the Buchanan & Stubblebine argument. If some agents are willing to pay for more reliability given that it does have private good aspects, then if left to their own investment choices, high marginal value agents would invest to a level beyond what would satisfy low marginal value agents. In other words, high value agents choosing to pay for higher reliability would satiate low value agents. Thus in the sense in which reliability is a public good, there are going to be cases in which the marginal value of additional reliability to a low value agent is essentially zero. In that case the low value agent is satiated, there is no externality at the margin, and any interdependency/externalities at the margin occurred at lower levels of reliability. High value agents wouldn’t stop there, though, if their private marginal benefit from reliability were still higher than their private marginal cost. Technically, this means that the interdependence experienced on the way to getting to equilibrium is a case of inframarginal externalities. Such externalities do not affect the amount of reliability provided in equilibrium, because at the margin the high value agents determine that level and the low value agents are satiated. The policy implication of this? If the low value agents are free riding, so what? It doesn’t affect the efficient outcome, if the environment is structured in such a way that the high value agents face incentives to “walk the talk” and invest in reliability (gee, like, maybe, markets?). And if the low value agents are forced to invest, then in the worst case you get over-investment in reliability, and in the best case it’s an income transfer from the low value agents to the high value agents. How fair is that?

The third implication of this argument is that in an efficient equilibrium, there will still be un-internalized externalities. It’s just that the un-internalized effects, which are inevitable consequences of interdependence on networks, are small enough at the margin that even if they were internalized, they would not change the actual outcome, the actual investment in reliability.

The fourth implication is that the only externalities that should matter, i.e., be policy relevant, are those that would affect the actual outcome in equilibrium. That’s what economists usually mean when we say “underprovision”. But — if there are effects on agents’ values at the margin (if my failure to invest in reliability has a significant effect on your utility), that means that there are unrealized gains from trade and we are leaving money on the table if we don’t negotiate and figure out how to internalize the effects. We could self-internalize by you paying me to invest, and if your payment is enough to get me over the hump, then I do it. If it’s not, then it shouldn’t be done anyway. If this idea sounds familiar to you, it should — this is where the Buchanan & Stubblebine and the Coase “Problem of Social Cost” arguments dovetail. Externalities are not policy relevant if transaction costs are sufficiently low that we can negotiate to self-internalize them. In such cases contractual approaches to policy will lead to superior outcomes relative to regulatory approaches like the imposition of mandatory reliability standards on all agents. The policy implication of this? Focus on rules that reduce transaction costs and foster the development of markets, formal and informal, through which network agents can self-internalize the relevant, inevitable, value effects of their interdependence.