Scaling Standards

In two recent posts, I blogged about the proposal from my article, Random Selection for Scaling Standards, With Applications to Climate Change, but in the blog posts, I used COVID-19 ventilator production as an example instead of the climate change examples from the article. This highlights that the proposal is quite general, a technique for distributing money to claimants when there may exist many factors that ought to interact to determine how much each should receive. Usually, in these circumstances, we depend on rules, even if the best rules that we can craft are crude approximations that ignore many relevant considerations.

I promised to return with answers to commonly asked questions. I'll start, though, with a brief recap, but using a different application, also related to the current pandemic, to illustrate the core idea behind the proposal. Suppose that the government wishes to distribute $500 billion to the American people. The simplest way would be to send fixed payments to all Americans; if we later decide that some received too much, we could always increase taxes on some groups later. But let's suppose that we genuinely wanted to distribute payments in proportion to need. There are many factors arguably relevant to assessing need: how much someone's income declined (if at all), how much of this decline was attributable to COVID-19, how much of a nest egg someone had, whether the person was capable of obtaining some other job, whether the person faces eviction, where this person lives, and so on. Meanwhile, people might disagree about which factors are most relevant or even about which way they cut. If Congress and the President cannot come to an incompletely theorized agreement about these issues, we are likely to end up either with a crude solution (like fixed payments) or no solution at all.

Ex post decisionmaking might seem impossible in this context. We certainly do not want a bureaucracy to process claims from every American through individualized hearings, however informal. That is too much work and would lead to too much inconsistency. We could, however, imagine that a governmental panel (perhaps meeting a few years into the future) could make ex post retrospective assessments of the needs of a small number of claimants, randomly selected. With only a small number of cases to consider, the panel could take into account a wide range of factors. Of course, we can't wait for this to occur before people receive their checks, and we also might worry that the panel will make some idiosyncratic decisions. Still, we might think that objective consensus predictions today of what such a panel would decide on average will be about right. The question then becomes how can we make sure that people are paid quickly based on expectations of what this panel would decide.

One approach (not my proposal, but still in its spirit) might be for the government to select some number companies (say, 100), each assigned a randomly allocated small percentage of the American people. Each would then be able to distribute funds according to whatever criteria it sees fit. In a few years, the government panel would use its ex post assessments to rate each of these companies on its performance, distributing some much smaller amount of money (say, 1%, or $5 billion). Companies that performed well according to the panel would receive more money; companies that performed badly would receive little (or might even need to pay money to add to the $5 billion being distributed to the other companies). These companies would thus have incentives to ask of claimants questions that it thinks will be relevant to the government ex post and to determine the weight to which to assign to different answers. I believe that such a proposal would lead to funds being distributed in a more tailored and thoughtful way than a solution like fixed amounts for each citizen. That might well be worth the reward money distributed to these companies. Still, there are some challenges, such as determining how many companies to authorize, how to regulate those companies to ensure that they do not accept bribes or have conflicts, and so on.

My proposal is simpler. The government simply commits to distributing the entire $500 (or $505) billion to the holders of claims from a small number of randomly selected Americans, say 100, in proportion to the ex post valuation of their needs. That is, each citizen would be able to sell his or her right to a claim to an intermediary. The intermediaries would have incentives to pay more for claims that they expect would on average receive more money on average ex post. Intermediaries would likely build diversified portfolios of claims. Competition among intermediaries would cause them to bid up the prices they would offer to claimants. The article explains that we might have some regulations to enhance competitiveness and reduce the possibility of discrimination. But the strength of the proposal is that it allows the government to distribute money with a minimal decisionmaking and regulatory burden while still ensuring that a wide range of factors are considered. It is especially useful in an emergency where it just is not possible for the government to scale up a bureaucracy in a short amount of time.

With this application spelled out, I will return to some questions asked or hinted at in responses to my previous blog post.

Don't the usual objections to government spending apply? Sure. There is a deadweight loss of taxation. My proposal simply assumes that the government should be spending money for the relevant purpose, whether on new ventilators or to help people get by in a crisis. It addresses how the government can distribute money to many claimants, some with stronger claims on the money than others, when it is difficult to craft ex ante rules. In a hypothetical world without externalities and perfectly efficient markets, we would not need this system. Similarly, I am not making a claim that I have found the best thing to spend money on (although ventilators seem to me to be quite cost-effective now).

Isn't this just a lottery, where completely undeserving people might receive the funds? This is a lottery for the intermediaries, not for the ultimate claimants, who receive money from the intermediaries in exchange for their claims before the lottery takes place. The money that the government distributes is distributed to intermediaries in proportion to their measured need, so claimants with claims that are expected to be more valuable will receive more from intermediaries. One might worry about completely frivolous claims--for example, that many who haven't even built ventilators might file claims on the ventilator fund--but the ex post panel would be likely to rate their contributions at close to zero. So intermediaries won't pay much for those claims. Moreover, if one worries that a high percentage of the ex post claims might be frivolous, one can address that by requiring an intermediary to add a certain amount of money into the fund for each claim to be eligible for the lottery.

Should claims be selected proportional to dollar expenditure? No, that shouldn't be necessary so long as we use the mechanism above for eliminating frivolous claims. It would be possible to have such a mechanism, but if claim A has twice the chance of claim B of being selected, then, if both in fact are selected, the social welfare evaluation of claim A should be divided by two before allocating the fund proportionately.

Isn't there too much risk for intermediaries? This is a major issue considered in depth in the paper, so I won't discuss it much here. The risk that intermediaries face can be divided into two sorts: risk associated with the random selection and risk associated with uncertainty about government decisionmaking. The first type of risk can be costlessly insured away (I explain how the government can offer insurance on random selection at actuarially fair rates), and the second type of risk is not so different from many types of risk absorbed through capital markets.

What if the ex post decisionmaking is corrupt? One might worry that someone on the panel would give a high valuation for a claim owned by an intermediary that either bribes the ex post decisionmaker or, say, that employs a friend of the ex post decisionmaker. These are legitimate dangers, but this can be addressed with anti-corruption and recusal rules. In general, judicial decisionmaking is less susceptible to special interests and corruption than other forms of decisionmaking. Meanwhile, even if some corruption did occur ex post, it would affect the ultimate claimants only if intermediaries could predict ex ante which claims would corruptly receive high valuations.

Won't people try to scam the system? Sure, but that explains why the system is so simple. (The system may seem unusual, but it is much simpler than any existing administrative programs.) The best scam idea that any commenter had was to submit huge numbers of claims in the hope that each would be worth a few pennies, but the approach described above of charging a fee per application addresses that. It's hard to scam a system where you need to convince someone to buy your claim from you. Of course, claimants may try to commit fraud to deceive intermediaries (an issue discussed in depth in the paper), but the intermediaries at least have incentives to try to identify such fraud and won't spend much money on claims that they are skeptical of.

Don't we need to have experience with the system before using it? Certainly, it would make sense to implement a system like this on a small scale before expanding it. Starting with $500 billion at stake might be a mistake. But one virtue of the system is that it's easy to scale up without greatly increasing the need for government employees, as the number of claims to be randomly selected can stay fixed.

How do we know a market will form? If the government has credibly committed to giving away $500 billion, private interests can be expected to spend close to that amount trying to get that money, particularly if the cost of entry into the market is low. This is basic rent seeking theory, but here the rent seeking is largely socially valuable. Of course, private firms might wonder how much investment other firms have made. One can require that intermediaries make periodic disclosures about how much they have spent on claims, so that intermediaries can gauge the total size of the market easily.

What if the ex post decisionmakers don't have a proven track record? All that matters is that they get the right result in expectation. If there is a chance that a claim might be adjudicated a thousand dollars too high or a thousand dollars too low, those possibilities cancel out, and the claimants will receive the right amount from the intermediaries.

What if the ex post decisionmakers are especially political? Ideally, the ex post decisionmakers won't be chosen until years after the intermediaries have bought up all claims. That way, at the time the intermediaries buy up claims, they won't know what political party and which officials will be in control. It's possible that ex post decisionmakers may behave differently based on political party, but, once again, that roughly cancels out in expectation. And because the ex post decisionmakers will only be affecting how intermediaries are compensated, rather than on what priorities the money will be spent, the decisionmaking might be lower salience and less political.

What is the difference between this and grants? Grants are given ex ante, so if the grant makers have some systematic biases, that will affect grants. Moreover, a grant mechanism can't scale to millions of potential claims without creating a huge bureaucracy.

Does this provide sufficient accountability? If one believes that accountability is the most important thing, one might prefer to just allow the President to distribute the money as he likes. My view is that it's better to reduce the role of political considerations in this process. Ex post, retrospective analyses are less likely to be politicized.

Does the fund need to be fixed? Not necessarily. We could make the fund dependent on some ex post measure of social benefit -- for example, lives or QALYs saved by additional ventilator production. The fixed fund, however, minimizes the risk that the government, having induced production, will effectively refuse to pay up by rating contributions artificially low. If there is a sufficiently good process for measuring social welfare benefits ex post, a fixed fund is not necessary.

Many of the above questions are addressed in some form in the article, as are many other questions that might have been raised. Thanks to all commenters for their contributions.

Comments

[The Turtle Dove]

I refer the honorable gentleman to the comment I made some time ago. These generic questions don't really seem to reach the questions that were raised by commenters. Good luck with the article.

Mr. D.

[Michael Abramowicz]

Turtle Dove, Thanks for your comment and your earlier one, which I had thought I replied to but will reply to more directly. Your core concern is that "intermediary funding would seem to favor a very large portfolio of ferry businesses without actual boats, but with the most convincing appearance of boats that's humanly possible." If I understand you correctly, intermediaries will knowingly buy up many claims that superficially seem attractive but that in fact represent minimal contributions to social welfare. This seems unlikely to me for two reasons. First, the few randomly selected claims will be adjudicated very carefully; this is a luxury that we can have with random selection. So, if the claims are only superficially attractive, that will become clear, and the valuations should become low. Thus, such claims will receive much less money than claims representing genuine social contributions. Second, as I noted here, one could require a per-claim contribution to the fund to discourage frivolous claims.
Perhaps you are worried that intermediaries' only incentive is to secure outside capital and that outside capital might be fooled by "ferry businesses without actual boats." But outside capital at least has some incentive to monitor the intermediaries, for example by examining some portion of the claims being funded.
You also worry about the possibility of "duplication of R&D and manufacturing setup." This is more of a concern, and one familiar both from the world of patent races and from market competition more generally. But a standard-based approach can make things better. If a firm releases a new ventilator design or other information useful to other companies, it should be able to receive some credit for this in the system that I describe.
Thanks again for commenting!

[Armchair Lawyer]

No.

Intermediaries take a cut.

You're making this too complicated and open for corrupt abuse. Just send the checks. Make it up with taxation afterwards (or more realistically, inflation).

[Michael Abramowicz]

Sure, in equilibrium, total intermediary investments will be less than payouts, with the difference being profit. That's how it generally works in a market economy.

My project was not to compare the purchase of ventilators with just sending checks to the entire population. But I think there are many reasons that sending checks will not lead to a sufficient increase in the production of ventilators. I will discuss that soon in another post.

[bernard11]

ISTM that the article sort of waves away the objection that this procedure merely imposes the costs of the adjudication on the intermediaries - and hence the claimants - rather than the government. It may be true that the private evaluation of the claim will be less costly, because of simpler procedures, than government adjudication, but it's not that simple.

First, this assumes that a more formal procedure is no more accurate - that the procedural steps do nothing to improve accuracy. Maybe, but no evidence is adduced in support.

More important, if the market for claims is truly going to be competitive then claims will be evaluated by several intermediaries, rather than one agency or court. So even if the process is cheaper when done once, doing it three or four times will be expensive, and that cost will come, eventually, out of the pockets of the claimants.

Of course, intermediaries might just resort to the kind of rules you are trying to avoid, but then what is accomplished?

And is there any doubt that some intermediaries take advantage of uninformed claimants? Sure, you propose some safeguards, but they too will be costly, and will surely be imperfect.

I'm also unimpressed by the insurance scheme, which seems to be based on the belief that the intermediary will know exactly what the claim will be worth when selected.

[Michael Abramowicz]

I agree that formal adjudication ex post can improve accuracy relative to informal market evaluation. Indeed, a central point of the proposal is that with a relatively small number of adjudications, ex post decisionmaking may be fairly effective. But especially if the number of claims is very large, this isn't practical. It's not entirely clear that would be so with ventilators, since there would be a small number of producers, but it would likely be so with a wider range of potential contributions (hand sanitizer, personal protective equipment, etc.)

More specifically, you are right that informal market evaluation can involve multiple different people examining claims. I'm not sure that's different from formal adjudication, where multiple lawyers will examine claims. But one can also reduce this, for example by requiring public release of evaluations and offers by intermediaries. Thus, if the first intermediary to make an offer has a reputation for accuracy, the second intermediary may take less effort in examining the claim.

I discuss safeguards for uninformed claimants in detail in the article. I think the market is much better for uninformed claimants than most, because claimants just need to look for the highest offer. But still, there are a number of things one can do to make it better.

On the insurance scheme, the intermediary doesn't need to know exactly what the claim will be worth. There are two sources of risk--the first, from random selection, the second, from decisionmaker unpredictability even with careful adjudications. The insurance scheme responds only to the first.

[bernard11]

One interesting thing about this proposal is that it might be the case that many claimants will do better than they would under conventional methods. If there really are a lot of intermediaries bidding for claims, then the winner's curse comes into play.

Of course sophisticated, not to mention colluding, intermediaries will allow for this, but that again introduces an element of guesswork.

[Michael Abramowicz]

Yes, the system works best in equilibrium, where everyone has some sense of how much everybody else is investing. In the absence of this, we might get the winner's curse (good for society, bad for intermediaries) or underinvestment (the reverse). We can reduce this somewhat by requiring some public disclosures.

[jst1]

I really wanted to give this article a chance. Then...
"It is especially useful in an emergency where it just is not possible for the government to scale up a bureaucracy in a short amount of time."
Please. We are not starting from scratch.
It might be easier to just argue the pandemic case instead of referencing climate change into the equation. But hey, I am all for the intermediaries as long as I get to be one.

[Michael Abramowicz]

Sure, we're not starting from scratch, but I don't think there is a government bureaucracy that is especially well suited to evaluate different ventilator designs and coordinate production in a short amount of time. The military may be best suited to accomplish this, but even the military generally does not get production ramping up in mere days. Procurements are much slower than the approach identified here.

[Peter Gerdes]

Looking at your article I see that your ultimate solution to the problem of big valuable claims being undervalued with respect to smaller more numerous claims is to allow subdividing into shares.

But this immediately creates a problem. My incentive is to subdivide my claim into as many shares as possible so you immediately result in the absurd situation where people are issuing a google (the number) shares in their claims. Moreover, it doesn’t solve the initial problem (even if you limit the number of shares) because every claim will be a subdivided and a billion dollar claim is at the same disadvantage relative to a billion $1 claims when both are divided up into a billion pieces.

Now maybe you want to prevent claims from being subdivided below say $1/ share. That works but that is just a different way of saying select proportional to cost.

—

Look here is a more general argument that you must select proportional to dollar value.

Ultimately, we don’t want to weight actions based on how easily they subdivide. For instance, if ventilators are far more important in the current crisis than respirators we want that to be reflected in the incentives even if anyone can start making respirators at home but you need 10 million dollars and a factory to make ventilators.

Now suppose social value is such that each ventilator trades off against 10,000 respirators. We want the incentives to reflect that so it better be that each claim for making a ventilator has an expected value 10,000 times as much as each claim for a respirator. That’s what happens if you evaluate claims relative to the purported value they claim (with a mechanism that reduces the reward proportional to any overclaim of the value...so you lose money by exaggerating how much you did). Any mechanism that doesn’t weight claims in exactly this fashion will distort the result.

[Peter Gerdes]

I’d add that with the tweak of selecting proportional to claimed value (with overclaiming penalty) this mechanism seems workable for situations with a linear utility function, eg, each additional ventilator adds 10k utils and each respirator adds 1 util and relatively straightforward evaluation criteria.

But the minute your utility function is non-linear you have problems. For instance, suppose we have two goods x, y and overall utility is given by y +10,000-10,000/(x+1), ie, x has diminishing marginal utility as many real goods do (think hand sanitizer).

Now an efficient outcome dictates that if I can make x or y at the same cost I should make x provided that the partial derivative d[util]/dx is greater than d[util]/dy, ie, x < 99. But consider how the system rewards people who make x and y.

Since everyone making x contributes the same the overall value of x production will be 10,000-10,000/(x+1) and if we assume that the total payout for x claims is proportional to 10,000-10,000/(x+1) and for y claims proportional to y you end up with an expected value of 1 for each y claim and 10,000/x - 10,000/(x^2+x). But this means you maximize return by making x goods rather than y goods as long as x is less than 9999.

So in the toy model of ten thousand people each of which can make one unit of x or one unit of y at the same cost the ideal outcome is 100 x units and 9900 y units. But the incentive for producers here is to maximize profit by making 9999 x units and 1 y unit.

Now if you allow claim subdividing or weight odds of claim selection you can change this a bit but the basic point remains.

[Peter Gerdes]

Oops off by 1 error there in that final statement. Social msc is when x=99 and y= 9901.

[Peter Gerdes]

To put the point more simply the problem is that in paying off claims ex-post you end up with overproduction of goods that have decreasing marginal utility (and underproduction of goods whose marginal utility declines less drastically) because each producer of a good gets an equal share of the total so even if you know someone else can make so many ventilators each additional one is worth pennies you’ll have an incentive to make them too to get your cut of the overall ventilator rewards.

[Peter Gerdes]

I’d add that the idea that insurance/investment would allow pooling risk is in direct conflict with the suggestion this minimizes the costs of evaluating the measures taken.

In order to buy a claim an aggregator needs to be able to fairly price that claim, ie, they need to predict (statistically) how it will be adjudicated. But you could have just used whatever process the aggregator picks to price claims it buys and had the government adopt it for a strictly more efficient outcome. After all an aggregator needs to price in the risks of unknown standards applied later so you’ll have to pay a more to have the system run through the intermediary rather than just implementing that scheme directly.

Now you might object that there will be many aggregators and a better mechanism than what the government picked will be selected by the market. But each aggregator is just as likely as the government to fuck this up so the more likely you think it is that the government would pick badly the more risk each aggregator is subject to and the greater premium you pay.

Your strictly better off hiring the investment banks to directly implement a mechanism for immediate government payouts of claims based on their best ex-ante predictions (if the gov wants they can run several different buying schemes just like the market would).

[Peter Gerdes]

Also note that aggregators can’t just use a quick and dirty heuristic because they suffer an adverse selection problem. The creator of a claim has detailed info about it and will know if their claim will do better or worse in the detailed adjudication than a surface analysis suggests. Claims that look stronger than they are will be preferentially unloaded while those that have hidden strengths will be held onto.

[Peter Gerdes]

Ohh and your argument about the effects of the adjudicators politics only works if all actors have equal information about the views of future adjudicators and no mechanism to influence them.

In actual fact what you should expect is that big aggregators would spend a huge chunk of the money on building complex models about the future political outcome (eg hire Nate Silver types to build complex models, perform surveys and use contacts to acquire info about who might run etc) and spend truly large amounts of money on lobbying to influence the process.