Kortina // Radical Markets / RadicalxChange

Sorry to Bother You (2018), Boots Riley. In response to some recent posts (History of the Capital AI & Market Failures in the Attention Economy, Kinky Labor Supply and the Attention Tax) a few people pointed me to the RadicalxChange conference.

I did a very cursory bit of background research and found that RadicalxChange was philosophically aligned with many of my own core beliefs — mistrust of centralized institutions, belief in incentive design and markets as a decentralized coordination mechanism, acknowledgment of the failure of the current dominant form of democratic capitalism, etc — I decided to go (also, I had been intending to go visit the birthplace of Stevie Wonder, Detroit, for years…).

The conference was excellent and led me to many bits of homework which I wanted to share here (including the book which was the inspiration for the conference, Radical Markets, by Weyl and Posner, which I’ll post detailed notes about below).

Sections: Notes + Links from RadicalxChange Notes + Quotes from Radical Markets

Here are links from the conference and (homework) links I wrote down for myself.

Baratunde Thurston gave a keynote on his datafesto.

Boots Riley (director of Sorry to Bother You) said something that really resonated with me in one of the keynotes, namely, that there is this tension with art vs action — he summed this up with the question:

Riley also commented to the effect of:

Harberger Tax on Property

The Harberger tax on property (cf. Sam Lessin’s proposal for a buy-sell agreement as a way to tax wealth in the US). Here are the full details of the mechanics of the tax Weyl and Posner propose:

Following up on Vickrey’s concerns about the Venezuelan fiscal system, Harberger worried that appraisers were frequently bribed by homeowners to understate the value of property so as to minimize the tax burden. While he was apparently unaware of the historical precedents, his solution has a timeless elegance:

If taxes are to be levied … on … the value of … properties … it is important that assessment procedures be adopted which estimate the true economic value … The economist’s answer … is simple and essentially fool-proof: allow each … owner … to declare the value of his own property, make the declared values … public, and require that an owner sell his property to any bidder … willing to pay … the declared value. This system is simple, self-enforcing, allows no scope for corruption, has negligible cost of administration, and creates incentives, in addition to those already present in the market, for each property to be put to that use in which it has the highest economic productivity.

While Harberger designed his scheme as a way to raise government revenue, it offers an inspired solution to the monopoly problem we highlighted above. Harberger’s tax, later also proposed by the Nobel Prize–winning economist Maurice Allais, makes it costly to declare a high valuation and thus deter the purchase of assets. Therefore, it penalizes any attempt to exercise monopoly power over an asset. The higher the price the possessor demands, the more tax she must pay.

Harberger’s tax closely resembles the partnership scheme of Cramton et al. Suppose that the annual tax rate is set equal to the probability that a buyer who values the asset more than the seller materializes within a period of, say, a year. Anastasia owns a house, and likes it. But there is a certain probability that someone else will show up who likes the house more than Ana does, and is willing to pay more for it than Ana’s valuation or reservation price (we call this probability the “turnover rate,” meaning the rate at which assets of this sort typically move into another person’s hands). Suppose that the tax rate and the turnover rate are both 30%. If Ana raises her sale price above her reservation (that is, actual) value, she benefits from the higher sale price 30% of the time — when those higher-value buyers turn up. Her benefit from raising the price would thus be .3ΔP, where ΔP is the increment in the sale price. On the other hand, as long as she remains in possession of the house she must pay the tax of 30%, which, applied to this incremental value, forces her to pay an additional .3ΔP. Thus, the benefit from increasing the price above the reservation price is exactly offset by the cost. This stops owners from holding out for a high sale price by setting a price higher than their reservation value.

At the same time, Ana also wants to ensure that the asset is not taken from her at less than her reservation value. So, she will, of course, not declare a price below her actual valuation. This means that she can do only one thing: set a price exactly equal to her reservation value, ensuring that exactly the buyers willing to pay more than her reservation value will end up taking the asset. Full allocative efficiency is achieved: every asset passes to the hands of the person best able to use it and invest in it.

For any tax rate below the turnover rate, the possessor will always set a price above the amount she is willing to accept. When the tax rate is zero, the possessor is free to set any price she wishes at no cost and thus would set the monopoly price. When the tax rate equals the turnover rate, she has to reveal her true value. For intermediate tax rates, she will still be discouraged by the tax from setting a very high price, but she will not have a full incentive to report her exact value. Instead, she will set a price intermediate between her true value and the monopoly price that she expects a buyer to be willing to pay. As the tax rises from zero to the turnover rate, the price she quotes will gradually fall from the monopoly price to her true value.

Even more technical detail on the Harberger tax (investment efficiency vs allocative efficiency):

What of investment efficiency? Remember that George’s original proposal failed because of the concern that people would not invest in their property if they must pay confiscatory taxes on the rents it produces. At first glance, Harberger’s tax also seems vulnerable to this problem. Suppose that the asset (which, for the sake of simplicity, let us suppose lasts only this year, like a machine that wears down from use) is currently worth $100,000 to its possessor, and that by investing $75,000 she can increase its value for her to $200,000 and also increase the value that any potential future buyer will place on it by $100,000 as well. Assuming the same turnover rate of 30%, our logic above indicates that she may as well, after the investment, declare the property to have a value of $200,000. However, this increases her tax bill by $30,000 (30% tax on the increased value of $100,000). The investment is not worth it. Although the value she gains from the asset regardless of whether the buyer acquires it has now increased by $100,000, she is forced to pay $30,000 of this amount to the government, which comes on top of the $75,000 investment. She loses $5,000 rather than making a profit.

But the investment can be improved by adjusting the tax. If a lower tax rate were charged, say 10%, then the possessor would still be able to capture $90,000 of the benefit from the investment ($100,000 minus the 10% tax payment). Now the possessor will profit from a $75,000 investment, or even a larger investment.

But if we lower the tax to improve incentive efficiency, then won’t we also harm allocative efficiency? At a 10% tax rate, by increasing the price beginning at her reservation value, the possessor could still capture .3ΔP of value from a potential buyer, but she would now be forced to pay only .1ΔP to the tax authority. She would thus have an incentive to raise the price, which would block transactions with buyers who value the property only a little more than the possessor does.

One might assume that the loss in allocative efficiency would offset the gain in investment efficiency. However — and this is a key point — the opposite happens. When the tax is reduced incrementally to improve investment efficiency, the loss in allocative efficiency is less than the gain in investment efficiency. The reason is that the most valuable sales are ones where the buyer is willing to pay significantly more than the seller is willing to accept. These transactions are the first ones enabled by a reduction in the price as even a small price reduction will avoid blocking these most valuable transactions. In fact, it can be shown that the size of the social loss from monopoly power grows quadratically to the extent of this power. Thus, reducing the markup by a third eliminates close to 5/9 = (32–22)/(32) of the allocative harm from private ownership. Furthermore, in this example the distortion to investment is eliminated.

More generally, if we considered all scenarios in which an investment could raise the value of the asset to $100,000, the only investments that would be deterred by a 10% tax are those that cost more than $90,000 to make. These investments are both rare and not terribly valuable, as the net value they create is small. By the same reasoning as above, it can be shown that only roughly one-ninth of the total distortion to investment from the 30% tax is caused by a 10% tax. Such a policy achieves five-ninths of the allocative benefit of the 30% tax at only one-ninth of its cost to investment.44 Furthermore, because different possessors often differ in their willingness and ability to invest to improve land (like the feudal lord and peasant farmer), allowing land to flow into the hands of the person best able to use it may also encourage investment.

Because of this quadratic structure, it is always optimal to have at least a very small tax. For example, a 1% tax will hardly distort investment at all but can still significantly improve allocative incentives. The owner will self-assess with reasonable accuracy to minimize her tax bill, but she will not be deterred from making valuable investments in the property. It is typically optimal to set a moderate tax rate, below turnover rate, that balances these two forces.

We refer to this tax as a “common ownership self-assessed tax” (COST) on wealth. The COST on wealth is also the cost of (holding) wealth. “Common ownership” refers to the way in which the tax modifies traditional private property. The two most important “sticks” in the bundle of rights that compose private property are the “right to use” and the “right to exclude.” With a COST, both rights are partly transferred from the possessor to the public at large.

Quadratic Voting

Weyl and Posner propose quadratic voting as a method for reaching group consensus that captures the strength of preferences more accurately than 1-person-one-vote, avoids deadlock, and escapes tyranny of the majority:

Suppose that Japan holds periodic referenda on important issues, like gun control or immigration reform. Every citizen is given a budget of “voice credits” every year, which he may spend on referenda that year or save for the future, as Kentaro did. To convert voice credits to votes, a voter can dip into his budget and spend as much of the balance as he wants to buy votes — but the cost of a number of votes is its square in voice credits. Thus, we call this system Quadratic Voting (QV). One vote costs one voice credit, which from now on we’ll denote as 1. 4 buys you 2 votes (the square root of four), 9 buys you 3, and so on. The square root is also known as the “radical” (another word for root), hence “radical democracy” — which is a kind of Radical Market, except it is one in which the goods are public rather than private. A referendum is approved if the votes in favor exceed the votes against.

Consider the gun control referendum Kentaro voted in. Every Japanese has the right to vote for or against the proposal. Rural voters like Kentaro strongly favor the proposal. In the story, he spends 400 for 20 votes; other rural voters spend 81 for 9 votes, 121 for 11 votes, etc. Most Japanese people live in cities and disapprove of guns. However, most of them have other priorities than gun control, given how low crime is in Japan and how the proposed reform exempts urban areas; they buy one vote against for 1 or two votes for 4. The government counts the votes: if the number of votes in favor of gun rights exceeds the number of votes against, the reform is adopted. In the story, the intensity of the support of Kentaro and other rural voters is enough to win the day and outweigh the mild opposition of city-dwellers.

This system enables people to cast votes that reflect the strength of their preferences. The key defect of the current system — that one can effectively register only three preferences: yes, no, indifferent — is eliminated. This makes two important things possible. First, a passionate minority can outvote an indifferent majority, solving the problem of the tyranny of the majority. Second, the outcome of the vote should maximize the well-being of the entire group, not the well-being of one subset at the expense of that of another.

Tyler Cowen posted a short list of concerns about the tendencies towards which policies might emerge from quadratic voting vs 1p1v on Marginal Revolution here.

Data as Labor

Building on analysis of Jaron Lanier in Who Owns the Future?, the authors propose a scheme where large internet corporations pay users for data exhaust generated that serves as input to machine learning algorithms:

Just as with classical statistics, there is a second critical question that determines the marginal value of data: how important it is to solve each of the problems data allow ML to tackle. If simple, early problems have much greater value than later, more complex ones, data will have diminishing value. However, if later, harder problems are more valuable than earlier, easier ones, then data’s marginal value may increase as more data become available. A classic example of this is speech recognition. Early ML systems for speech recognition achieved gains in accuracy more quickly than did later systems. However, a speech recognition system with all but very high accuracy is mostly useless, as it takes so much time for the user to correct the errors it makes. This means that the last few percentage points of accuracy may make a bigger difference for the value of a system than the first 90% does. The marginal value grows to the extent that it allows this last gap to be filled.

And:

Finally, realizing Lanier’s vision for data as labor would require building a variety of sophisticated technical systems. The architecture of many digital systems would have to be adjusted to keep track of the origin of and uses of data, so users could be rewarded at least for the average value their data create but ideally to some extent for the unique value their data may occasionally end up yielding. ML systems would have to be designed to determine particularly valuable data to them; then their requests for data would need to be channeled to consumer-facing products; and finally, these products would need to be designed to query the users for extra data in a minimally intrusive way.

/ht Namrata Patel who helped generate a bunch of these critiques of data as labor:

As the authors note, treating data as labor is not an instant win for the labor class generating data — the exact same class struggle between capital and labor would play out in this domain as it has with all other forms of work. The authors propose unions as the solution to this problem, but this feels a bit hand-wavy, as historically unions have had varying degrees of success and to me feel like more of a bandaid solution rather than cure for a root cause.

Particularly challenging with data as labor is the fact that this form of labor is perhaps the most mobile, and easiest to outsource to nations with the lowest wages and least workers’ rights. This already happens to an extent with content moderation, and you can imagine a global economy where a large percentage of it is driven by data work creates even worse outcomes for low skilled labor in relatively wealthier countries than the outsourcing of manufacturing.

Another challenge with data as labor is asymmetric value — data is not very valuable / usable to each individual worker — it becomes valuable to only those with a huge data set and lots of compute power. There will, therefore, be only a few large corporations that will be in the market looking to purchase data, and there are very likely to be monopsony pressures driving down the market price for data.

Finally, implementation of tracking provenance of data would require massive computational power, and even if that existed, there would be really tricky problems around “data laundering” — perhaps I can bootstrap my ML algorithm with some data generated by humans, but if I can then start simulating new scenarios for future training, I might be able to discard all of the data I originally collected from humans (and my need to pay them wages).

Harberger Tax on Intellectual Capital

This may be the most radical idea in the book (and one that I find the most interesting). In my experience, the intellectual left is very quick to support taxes on financial capital / income / wealth, which incidentally would shift the balance of power in favor of the very intelligent / well educated. It is really refreshing to see an acknowledgement of the inequities and power dynamics of knowledge and education in the form of Weyl and Posner’s proposal for a tax on human capital:

Consider a very radical extension of the COST: to human capital. Human capital refers to a person’s education and training; it’s a bit like physical capital (land, factories, etc.) because it enables the holder to obtain additional profits on a given investment of effort. But it’s also fundamentally different, for reasons that will become clear.

To understand how a COST on human capital might work, imagine that individuals were to self-assess a value of their time, pay a tax on this self-assessed value, and stand ready to work for any employer willing to pay this wage. Consider a surgeon, for example, who announced that she would perform a gallbladder surgery for $2,000. She would pay a tax based on that amount, and be required to perform an operation on anyone who offered that amount. The tax would discourage her from overvaluing her time and thus denying her talents to a needy community, while the need to be on-call at that wage would avoid her setting too low a wage.

In principle, a COST on human capital would be immensely valuable. Indeed, it would address the single biggest threat to equality and productivity that we have not addressed — the ability of the most talented people (the top scientists, lawyers, accountants, entertainers, financial whizzes) to withhold their services unless paid a monopoly price. A COST on physical capital simply does not address this problem, which is one of the major sources of the rise in inequality over the last half century. Once most other sources of inequality are addressed by our other proposals, this could well become a leading source of social tension, especially as genetic engineering and cybernetics redefine the idea of investing in human capacities.

Furthermore, a COST on human capital would eliminate the need for one of the largest deterrents to work at present: taxes on income. By replacing taxes on income with taxes on the underlying human capital that generates income, a COST would encourage rather than discourage work. It would also be fairer and more legitimate. Those without great talents would still have a lower potential income than the most talented people would, but would never risk falling into poverty as they would receive a large social dividend based on the taxes levied on the talented. Talented people would have greater opportunities to be rich than the less talented, but at the cost of bearing the risk of falling into poverty (through taxation of their social dividend) if they refuse to make use of these talents. A COST on human capital might turn out to be politically popular because it penalizes the highly resented educated class and lazy people of all types, while rewarding ordinary workers for their labor.

Despite these potential benefits, a COST on human capital is premature. There are two major problems. First, technology is simply not ready. A COST on human capital would need to take account of all the ways in which people enjoy or do not enjoy their work. People care about how much they work, where they work, who they work with, the conditions under which they perform their work, and much else — none of which could be captured by a COST on human capital unless accompanied by technological means for measuring all these factors. Conceivably, a COST on human capital associated with a technologically integrated form of labor — like data input, which is constantly monitored by computers — would work, but it is hard to say.

Second, a COST on human capital might be perceived as a kind of slavery — incorrectly in our view, at least if the COST were properly designed. Still, we can see the problem. Imagine that a surgeon simply decides one day that he has had enough of surgery. Under a COST, he would self-assess at a high level so that no one would purchase his services — paying a large tax in return for retirement. But people could find themselves in a position where it is not practical to do so, or they simply do not want to work anymore, whatever commitment they have made in the past. While design tweaks could avoid the coercive elements of the system, society is not yet ready for such a Radical revision in its understanding of labor.

It would be a mistake, however, to think that the current system is not coercive. In our current system, there is a wide gulf between educated elites whose native or acquired talents are highly marketable and those who have been left behind by changes sweeping the economy. The talented enjoy a kind of freedom, as they can select from among a variety of appealing jobs. These jobs allow them to quickly accumulate capital that they can depend on as they age, if they do not like the jobs that are available, or pick and choose among different levels of labor (part-time, enjoyable or rewarding but low-paying jobs in the nonprofit sector, etc.). Those with fewer marketable skills are given a stark choice: undergo harsh labor conditions for low pay, starve, or submit to the many indignities of life on welfare. Yet the waste of social resources when a talented person fails to realize her potential are far greater, and arguably their failure to work should be punished more harshly.

A COST on human capital would ameliorate this form of unequal freedom by requiring the talented people to pay a tax if they do not want to work in a job that is most efficient for society. A reasonable tax will not reduce them to starvation or a welfare-style existence, but will put them under more pressure to work for the benefit of society, just as the poor must at present, while alleviating the corresponding pressure on those with less talent in our present society. Perhaps a society more accustomed to a COST on wealth and other checks on market power, and where the talented use genetic engineering to give their children clearly unfair advantages, would come to see monopolies over talents as pernicious. Thought experiments in stories like the Handmaid’s Tale and X-Men suggest that societies that mismanage unique talents either by enslaving them (in the first case) or allowing complete self-ownership (in the latter) are likely to end poorly.

I was discussing the above tax with Sam, and while there are many merits in principal, one challenge he raised was how to prevent Person A from setting a reservation wage of $1/hr and Person B hiring A as a pool boy, effectively allowing A a vacation. You might need some sort of ‘you must accept the highest bid’ mechanic, but there are challenges with this as well (which Posner and Weyl note but do not propose a work-around for).

To conclude, it is incredibly refreshing to see so much math / game theory / practical proposals around incentive design from the left (vs only a critique of existing power structures and proposals for a new and idealistic regime).

I highly recommend Radical Markets.