In the past I've reviewed Eliezer Yudkowsky's "Inadequate Equilibria" book. My main complaint was that while it explains the problem of suboptimal Nash equilibria very well, it doesn't propose any solutions. Instead, it says that we should be aware of such coordination failures and we should expect ourselves to fare better than the official institutions in such cases. What Yudkowsky is saying (if I understand him correctly) is that given that the treatment of short bowel syndrome in babies is stuck in an inadequate equilibrium, there's no way to fix the problem on the system level. However, you can spot the problem and acquire the medication needed to keep your kid alive from abroad yourself.
In my review I've sketched a couple of ideas how to approach the problem, but that was just me trying to be clever. Something backed by evidence would make me much more happy.
So I've decided to have a look at how people are getting out of suboptimal equilibria in the real world. And that's how I've got to the Elinor Ostrom's book "Governing the Commons (The Evolution of Institutions for Collective Action)".
The book is very explicitly covering the same problem as Yudkowsky. As Ostrom say about her research in her Nobel lecture: "One of the key questions that we've been addressing is 'Are rational individuals hopelessly trapped in dilemmas?'"
Ostrom is an economist with a game-theoretic bent. But she's not a theoretician. She's a field researcher. I've expected some good read and I wasn't disappointed.
The book focuses on special subset of coordination problems, problems of management of what Ostrom calls "common pool resources". Common pool resource is something that, unlike private property, is not exclusionary, something that one cannot easily restrict people from taking advantage of, but that, unlike public property, can be depleted by excessive usage. So, for example, it's hard to exclude others from fishing in the ocean, yet, the fish stock can be depleted by overfishing. Ocean fishery is a common pool resource. Compare that with a private fish farm where others are prevented from fishing by a barrier. And on the other side, contrast it with a public weather forecast which doesn't get depleted as more people turn on the radio.
On the game theoretic level, these scenarios can be though of as "tragedy of the commons" or "prisoner's dilemma" problems. At the heart of the problem is the fact that while cooperating and limiting everyone's rate of extraction helps to sustain the resource, individual players get better short-term reward if they defect and overuse the resource. For example, limiting the number of cows on a common grazing land keeps the pasture sustainable. However, each villager is incentivized to get as many cows as possible which will in turn result in overgrazing and destruction of the grazing land.
As Ostrom notes, people reading about this kind of problem tend to jump to one of the two conclusions. One group believes that the grazing land should be nationalized and the laws should be enacted to prevent the overgrazing. The other group thinks that the land should be split into small plots and privatized.
The problem with the latter solution is that it's often not feasible. For example, there's no reasonable way to privatize an ocean fishery. Fish stock is migratory and cannot be split among stakeholders. So, if you invest invest in the growth of the fish stock by temporarily fishing less, the fish will eventually migrate elsewhere and get caught by other fishers.
In other cases the resource can be split but doing so results in suboptimal performance. For example, if conditions on the grazing land vary, some parts may be dry this year, but lush the next year. Some parts may have been underwater the last year but are all right this year. In such cases splitting the grazing land gives suboptimal outcome for every participant. Each of them experiences a sequence of good and bad years. Cows die of hunger one year, but then the grazing land is underutilized the next year. Each herder's ability to plan ahead is badly compromised.
As for the nationalization approach, Ostrom demostrates the problem using a simple game. She starts with the classic tragedy of the commons scenario. Everybody is incentivized to overgraze and the pasture will be eventually destroyed.
But then the state steps in. Land is nationalized and overgrazing is prevented by law. Participants who do overgraze are fined. And when we take the fine into account in our game-theoretic calculation the outcome suddenly changes. Overgrazing in no longer the optimal strategy. The pasture will be sustained. Hooray for nationalization!
But not so fast! As Ostrom notes, law enforcement is not perfect. Sometimes offenders are not caught and sometimes people are fined unjustly. She assumes a certain error rate, she does the calculation again and lo and behold! The incentives to overgraze are back!
And we haven't yet considered the cost of policing. If we spent too little on policing, we'll either have less policemen, resulting in higher error rate, or the policemen are underpaid and tempted to accept bribes which will in turn compromise the entire system.
And, undoubtly, there are many more facets of the policy that play a role and can affect the result of the calculation.
It turns out that thinking about common pool resources in absolutist ways is not helpful. In reality, there's a broad continuous multi-dimensional range of policy options. The villagers can split the land into private and public parts. They can spend more on law enforcement. They can police each another. They can limit the usage of common resource based on effort spent maintaining it. They can use rotational allocation of land, or maybe a lottery. And so on and so forth.
But it's even more complex than that. It not just policies that determine the outcome. The nature and the particularities of the resource itself may determine the optimal policy.
Take, for example, the beach seine fishery in Mawelle, Sri Lanka. The village beach is divided into two launching sites, one on the harbor side and one on the rock side. There are different rules for those two sites. The rules also vary depending on the time in the day. The study in question notes:
The Mawelle fishers provide a coherent explanation for why they use this complex set of authority rules, rather than a simple rotation system, to equalize the opportunity to make a big catch. Four environmental or technological considerations affect the problem of equalizing access: (1) The harbor side produces the really big catches, but the rock side is more consistently productive when there are fewer fish. (2) The first catch of the morning is most likely to be the biggest catch of the day, and the prices are highest in the morning. (3) The weather affects the number of hauls that can be made in the day, and any system assigning a set hour of the day would be inefficient. (4) Beach-seining involves high labor inputs to prepare a net for use and to restack it afterward, and simple rotation systems allowing all nets to be used only once per rotation would involve higher labor cost.
As can be seen, the goegraphy, the shape of the seabed or maybe the sea currents, the weather patterns, the details of the fishing technology and so on, they all affect the optimal policy for governing the fishery. And it turns out that the optimal policy differs for two sites barely kilometer and a half apart.
There's no universal "correct" solution to the problem. A host of minor details matter. And, let's be frank, we have yet to see an economist who incorporates a map of sea currents at a particular place at the Sri Lankan shore into his economic model.
The absolutist position is also untenable becuase the optimal policy may change over time. Consider climate change. It may cause the weather patterns in Mawelle to change. And suddenly, the governance system that has worked for centuries stops working even though none of the rules were changed.
All in all, it seems that organically grown institutions are a lot like Hayek's free markets. They are information-processing machines. They aggregate countless details, too small and numerous for any central planner to take into account, and generate a set of efficient governance rules.
In her discussion of the governance systems Ostrom picks several cases that proved to be sustainable over a long period. She analyzes communal tenure of high mountain meadows and forests in Switzerland and Japan, the "huerta" irrigation systems in Spain and the "zanjera" irrigation system in Philippines. All of them are practiced for hundreds of year, in some cases even for thousand years. I am not going to go into details, but do read the book. At the end she concludes that sustainable common pool resource governance systems tend to have the following features:
I am not going to reproduce the entire argument, but I cannot resist quoting few interesting paragraphs from the book and expressing some thoughts of my own (in no particular order).
I always thought that social capital (or common knowledge, if you will) is expensive to create yet very easy to destroy. The following tidbit seems to indicate that it can, in fact, be rather resilient. It would be nice to see more research on this aspect of the common knowledge. Note how it was sustained via a two-sided consensus about the temporary nature of the violations of the system:
[During the depression in 30's] almost all the villagers knew that almost all the other villagers were breaking the rules: sneaking around the commons at night, cutting trees that were larger than the allowed size, even using wood-cutting tools that were not permitted. This is precisely the behavior that could get a tragedy of the commons started, but it did not happen in Yamanaka. Instead of regarding the general breakdown of rules as an opportunity to become full-time free riders and cast caution to the winds, the violators themselves tried to exercise self-discipline out of deference to the preservation of the commons, and stole from the commons only out of desperation. Inspectors or other witnesses who saw violations maintained silence out of sympathy for the violators' desperation and out of confidence that the problem was temporary and could not really hurt the commons.
As for the "graduated sanctions" point it seems that fines for occasional breaches of the rules are very low. So low, in fact, that it still may be profitable for the player to break the rule. One study that comes to mind here is the one by Gneezy and Rustichini that shows that when parents were fined for picking their kids late from the day-care center they become more likely to be late. The fine have normalized otherwise socially unacceptable behaviour. One has to ask whether this effect plays a role in the systems described by Ostrom.
One observation that came as a surprise to me was that moderate fine (as opposed to large fine which may cause resentment) can in fact increase the trust in the system. The fined participant can experience fist-hand that breaking the rules doesn't go unnoticed and thus gains confidence that others are not breaking the rules. Think of it as a probe mechanism: Every participant has a cheap way to test the monitoring and enforcement system and make sure that it's still working as intended. If the fines were very high the ability to monitor the enforcement system would be compromised.
From the chapter about Zanjera irrigation system in Philippines:
A few parcels, located at the tail end of the system, are assigned to the officials of the association as payment for their services. The system not only provides a postive reward for services rendered but also enhances the incentives for those in leadership positions to try to get water to the tail end of the system.
There's not much to say about it. We should use this kind of approach in politics much more often.
I am including the following quote because the overpopulation problem is one of the examples of suboptimal equilibria. In Hirano, Nagaike and Yamanaka villages in Japan:
Rights of access to the comunally held lands were accorded only to a household unit, not to individuals as such. Consequently, households with many members had no advantage, and considerable disadvantages, in their access to the commons. Population growth was extremely low (0.025% for the period 1721-1846), and ownership patterns within villages were stable.
If you are interested in the topic look also here.
Finally, I've enjoyed finding out that resentment-free fair-division algorithms are actually being used in the wild. When harvesting wood from communal forests in Törbel, Switzerland:
The first step is that the village forester marks the trees ready to be harvested. The second step is that the households eligible to receive timber from work teams and equally divide the work of cutting the trees, hauling the logs, and piling the logs into approximately equal stacks. A lottery is then used to assign particular stacks to the eligible households.
It seems that a similar lottery system is used in Zanjera irrigation system in Philippines.
All of that is nice and interesting, but hey, let's remember what we are after here! We are not looking for the best institution. We are trying to find out how to escape a local maximum. How to jump from an institution that sucks, but happens to be a Nash equilibrium, to a better institution. The question thus is not how such a better institution looks like but rather how it gets created.
When discussing the problem Ostrom shifts the focus from the long-lived institutions (we have very little, if any, data on how they were created centuries ago) to the governance of water basins in Southern California. These institutions were crated not that long time ago, in forties, fifties and sixties, and a lot of detailed information is available.
The problem is as follows. Under today's Los Angeles metropolitan area there are natural underground water reservoirs, a water-bearing strata made of sand and gravel. These reservoirs get replenished at a constant rate by the rains that fall in the foothills and upper valleys. They are also used by many parties as a source of water. This naturally leads to a tragedy-of-the-commons-style problem. If everyone pumps as much water as they can they are going to suck the reservoir dry. Even worse, low level of water in the reservoir means that sea water starts to seep in and will eventually destroy the reservoir.
The situation was made worse by the standing law. One was entitled to specific amount of water based on how much of it they were using. Thus, lawyers advised everyone to pump as much as they could.
The first change occurred in Raymond Basin. After the attempt to reach a voluntary settlement failed, city of Pasadena initiated legal proceedings against city of Alhambra and 30 other producers.
That changed the dynamics of the system. It was found out that the water is being pumped out at a rate exceeding the replenishment rate by 38%. The ruling would, presumably, redistribute the water so that it was pumped out at at most the replenishment rate. However, due to complex lagal and ownership arrangements it was not at all clear what the ruling is going to be and who's going to take the worst hit. Every participant had to consider the scenario where they would be the ultimate loser. That provided an incentive to try to reach a negotiated settlement.
Within six months the parties had drafted a stipulated agreement to share the cutback on proportional basis signed by all but 2 of the 32 participants of the litigation. Rather than imposing his own solution the judge issued a final judgement based on the stipulated agreement.
West Basin came next. Unlike in the case of small Raymond Basin, the litigation had almost 500 parties. Also, when the numbers came in it turned out that pumping rate was three times the natural replenishment rate of the basin. Thus, each participant had to face drastic cutbacks if the case was settled by the judge. A forum was created for negotiation of the settlement. Although everyone had an incentive to agree on anything better than two thirds cutback, it took two years of negotiations and a threat of court to achieve an interim agreement to cut back the pumping back to the levels from the year 1949. The interim agreement was used for seven years while the producers pursued other strategies to enhance the local water supplies, to replenish the basin, and to try to convince non-signatories to agree to the curtailment.
In 1961, after 16 years a trial was held and proposed judgement was passed to the court.
Later on similar process happened for Central Basin.
Now here's a quote that I find interesting:
No one really knows the exact costs involved in the West Basin litigation, given the large number of parties and the length of time involved, but the best available estimate is $3 million. […] Amortizing the costs of the litigation over a 50-year period (as one would expect to do for the construction of a major physical facility), […] the adjudication costs in West Basin amounted to an annualized cost of $2.50 per acre-foot of water rights.
I like the idea of treating the creation of an institution (i.e. the rules governing the usage of the basin) as an infrastructure project, similar to a dam. If you read the piece by Scott Alexander you'll get an impression that you have to at least sacrifice a black goat and hire an exorcist to solve a coordination problem. After reading Ostrom though, it doesn't look any more exotic than hiring a couple of lawyers and an accountant.
It should be also taken into account that construction of an institution is an information good of a kind and thus the cost can be reduced by having some prior information. For example, the litigation process in Central Basin happened after the litigation in West Basin and the participants were thus able to learn from the existing experience, eliminate unnecessary steps (for example, it was immediately clear that they are shooting for a negotiated settlement, not a court ruling) and the cost of the project, despite Central Basin being bigger and having more stakeholders, were estimated at mere $450,000.
Similar project had, however, failed in San Bernardino county. Ostrom lists several reasons for the failure, including that San Bernardino county is much bigger that the other basins, that it may actually consists of several physical basins and so on. But listen to this:
No voluntary water associations were created to facilitate discussion of these issues, and no consensus emerged over time about any of them. Conflicts emerged between the large and small water pumpers, between advocates for development and advocates for no-growth policies, between industry and agriculture, between locals and "external experts", and between appointed personnel and elected officials. The lack of fundamental agreement led to acrimonious political conflict, including several recall elections, front-page stories in the local papers that pushed aside stories on the Watergate scandal, and finally the suspension of the litigation in 1974. No action has since been taken to limit groundwater pumping.
It seems to me that the real cause of the failure was that the model from West Basin was imposed on the participants, in top-down manner, without much discussion. The fora that existed in West Basin weren't created in San Bernardino. The result was one ugly coordination failure.
Which makes me wonder whether the process of creation of an institution may be instrumental in its eventual success. What if the same set of rules in the same circumstances may succeed or fail depending on how they were conceived? What if the process of creation, the one where people discuss their options, argue about them and bounce ideas off one another serves as a mechanism to establish common knowledge among them? That's a really interesting idea and I would love to see some experimental results to either confirm it or disprove it.
Obviously, Californian water producers had many advantages on their side. They lived in a democratic country. They had functional legal system and law enforcement. Corruption was low. The level of social trust was high.
Would they be able to succeed elsewhere? It's hard to say, obviously, but let's have a look at what Ostrom writes about Sri Lankan irrigation system.
She discusses the topic in the chapter devoted to coordination failures. She describes the situation in Kirindi Oya. The population was heterogenous, composed of individuals coming from different regions, castes and kinship groups, all of whom are initially poor. The upstream farmers were using more water than they needed (because having rice fields flooded helps with weed control) and left little, if any, for downstream farmers. The central regime was unwilling to enforce rules impartially. Police treated water offences as trivial. Monitoring was non-existent and disputes were sometimes solved by violent means. The infrastructure was damaged:
Gates are missing, structures damaged, channels tapped by encroachers and others. When asked why they don't prevent some of the most blatant offenses, two young technical assistants replied "that they were afraid to because of the fear of being assaulted".
In short, the situation was as bad as it gets.
But then she recounts a positive story from the left bank of Gal Oya irrigation project. The situation was similar to that in Kirindi Oya. The fact that upstream farmers were mostly Sinhalese while downstream cultivators were mostly Tamil haven't made things better.
Cooperation among farmers was minimal. Social relations among settlers, who came from different areas of the country were strained… Relations between farmers and Irrigation Department (ID) officials were marked by mistrust and recriminations. Farmers had no confidence in the competence or the trustworthiness of the ID's staff… Many field-level officials … were notorious for their corruption and thuggery. The main obstacle to efficient water management, from the farmers' point of view, was the local-level officials, who had political and bureaucratic power behind them. On the other hand, the ID officials, especially irrigation engineers, believed that farmers could not use water responsibly and carefully. Therefore, they argued that it was necessary to organize, educate, and discipline the farmers to do what ID asked them to do. Thus farmers were considered a part of the problem while the latter constitute the solution.
Then an experiment was made. The idea was to introduce "catalysts" into the situation of mutual mistrust and unpredictability. "Institutional organizers," (IO) mostly college graduates who also had farm backgrounds were dispatched to the area. They've received a six week training on how to approach and motivate farmers and on technical subjects related to irrigation. Each went to a small area served by one distributory canal. Their purpose was not to impose a particular policy but rather to organize the farmers to plan self-help strategies. At the same time they had status to deal effectively with Irrigation Department officials.
Instead of establishing a predefined organization, the IO tried to form a working committee to solve particular problems, such as repairing a broken gate or desilting a field channel. Further, IOs identified problems beyond those that could be solved by local farmers working together, problems that had to be articulated to ID officials and others. Once farmers were used to working together and had achieved benefits from group action, the IO would then help form a local organization and select, through consensus, a farmer-representative. This representative could articulate the interests of the other farmers on his field channel at larger meetings and report back to the others what had happened in larger arenas.
When the farmers started working on rehabilitation of the field channel the attitude of irrigation officials toward them started to change.
In the areas where the new system was introduced farmers started to use water rotation procedures quite generally. There were even deliberate efforts to make water available to the farmers downstream.
On the Sinhalese-Tamil boundary, for example, maintenance haven't been done in years. Farmers talked about previous murders over water disputes. Within few months after introducing the new system, Sinhalese and Tamil farmers began to work on clearing out the channels.
That's not to say there were no problems in Gal Oya project. However, the example shows that there are ways to get out of suboptimal equilibria even in a highly damaged and non-functional environment. Sometimes, though, it seems to require a little nudge from the outside.
Common pool resource problems are a strict subset of inadequate equilibria problems. However, I don't see why they would be inherently easier to solve than other types of problems. Maybe there's a viable solution for each such problem. Maybe there's a way to escape any suboptimal Nash equilibrium. Maybe all we have to do is to try and when we fail to try again.
May 25th, 2018
Discussion at LessWrong