Operations Research and Supply Chain Management have made a lot of amazing progress over time. A few weeks ago I was thinking about this general problem and I scraped the Amazon economics job listings to see what skills they seemed to prioritize. Lots of reduced form causal inference (A happens before B, driven by timeseries), forecasting, empirical industrial organization (how to be a good oligopoly with Walmart and Alibaba) etc… generally leaning hard into the data-driven subfields as expected.
Amazon seems to be becoming something like a centrally-planned economy anyway. This is due to being able to out-compete smaller firms who do not have access to as much information. The larger it gets, the higher quality the information available, the better their predictive models become. For the things that they don’t produce themselves, they buy in massive bulk and achieve better economies of scale than smaller competitors can achieve. The more they out-compete smaller firms with less information.
Due to their causal models, they have better information about things like how certain complementary goods will be demanded based on spikes in the goods that are likely to cause future demand. If some battery-powered toy is starting to get really popular, batteries are likely to see a swell in demand etc…
They lean heavily into ML for probabilistic solutions that let them tackle logistics problems that (as mentioned in the article) are computationally intractable for their optimal solution.
My perspective is that this is all a rerun of the Gilded Age Robber Barons, but with the compute power to tackle far more verticals. Maybe all of them. It seems that Amazon is already our centrally-planned economy to a large degree. In a “normal” climate that I was trained to think within in Industrial Organization classes while studying economics in college, Amazon would have been split up by now due to the difficulty that a new firm would have in competing with them on anything. But we do not live in that climate, and we do not have the ability that Teddy Roosevelt had to “bust the trusts.”
In 1944, the major economies formed the Bretton Woods system in an attempt to control the forces of international finance that were widely viewed as contributing to the struggles of everyday people. Economic insecurity for everyday people resulted in nationalism, which resulted in genocides and the world wars. Among other things, Bretton Woods introduced capital controls that stood in the way of investors quickly pulling out of one country and reinvesting elsewhere. This is also why you used to need things like travelers checks when you traveled. But this friction had lots of nice effects that contributed to the rising of real wages. In Detroit, the unions were able to effectively maintain high wages because the auto companies could not easily shift manufacturing to another country. It meant that large corporations could be meaningfully opposed without them simply running off to another country. It meant that real wages continued to rise between 1944 and 1971 along with rising profits. Since Bretton Woods was ended by Nixon (who wanted to print money to pay for Vietnam War debts, which messed up the core structure of the system, dissolving it) real wages have stagnated due to labor losing its negotiating position. As programmers, our real wages are going down as well, for indirect but intimately related reasons.
Now, we have trade agreements that make it easier for a firm to simply shift their manufacturing and investment to another country if the workers in one place want a living wage, or if the government wants to introduce a stronger social safety net which would raise the taxes on a corporation. Countries are in competition with each other to have the crappiest social support systems, because that will attract investment that will provide some people with jobs. Countries worldwide are afraid to impose the standard antitrust measures on companies that can more credibly threaten to just go somewhere else. This is why Amazon is unopposed.
I’m quite pessimistic about this situation. I believe that because we have no credible way of preventing capital flight without a strong international agreement (which seems unlikely to me, but hopefully I’ll be disproven) there will be no real way to prevent oligopolies from destroying the abilities for everyday people to live with dignity, inequality will continue to explode (although individual wealth does not tend to move as much as investment capital in response to taxation, unless you’re a billionaire living in a particular Swiss canton who will negotiate with the canton for a lower tax rate otherwise you’ll move down the street to a canton who will give you a better deal). I see this as a significant structural risk that is likely to lead to WWIII before we get the next Bretton Woods. But for those of us who still have jobs, we’ll be enjoying our centrally planned economy of Amazon-delivered goods. This is a big reason why I’m motivated to take 3-6 months off of work every year. I have a lot of doubt about the near future, and I think it’s important to enjoy life without taking the current relative stability for granted.
I think that Yanis Varoufakis’s view that he lays out in Another Now is particularly compelling, if anyone is interested in “Economic Sci-Fi”. His ideas are very similar in some ways to an expanded Bretton Woods along with a realization of The high-growth high-social-standards Meidner Plan that became possible with the capital controls of Bretton Woods. He advocates keeping the market to avoid the inefficiencies and abuse of planned economies, while proposing an alternative to the labor market for letting people live with dignity in an increasingly fast-moving economy that will result in firms continuing to fail when they cannot compete. High-growth, low-inflation, without the social consequences that lead to genocide and war.
This is great; I’ve heard bits of this story, mostly as stories about Leonid Kantorovich, but never quite a blow-by-blow in this fashion. Economics is an optimization problem, and current market capitalism works as well as it does because it combines lots of local and not-so-local optimization processes, and incentivizes doing the work to gather information and optimize processes. Looking at it from a data-science point of view is thoroughly fascinating, and does a good job of describing the problems with capitalism or with other systems as well: computational complexity, data quality, and objective function.
This is really interesting, especially considering that if alpha-things can win at go, StarCraft, and protein folding, it seems like “100 years of Moore’s law” is an excessively conservative estimate for how long it’ll take before we have the tools to actually solve this for real life in real life.
it’s already been thoroughly solved by amazon and walmart, i suggest you read people’s republic of walmart, it provides great insight into this exact problem (large-scale economic planning). for example, the book cites the fact that walmart, internally, is a larger planned economy than the ussr ever was
It is solved for the small subset of the economy and range of products that is Wal-Mart - and then, may still founder entirely on personal preferences. The economic calculation problem is still very much unsolved at the scale of first world economies.
Like, Walmart will probably, within applicable laws, avoid hiring disabled people. What should a country/society do with that information? It’s quite easy to break out the line items “care for the disabled” and take action to minimize them - often in ways that are quite frankly horrific.
i’m talking specifically about their forecasting and logistics procedures, which could be executed by anyone. not sure what your example is meant to point out
I think what he’s trying to point out is that, while Walmart is larger in dollar terms than many national economies, it’s also a) significantly simpler in terms of number of goods and services offered, and b) unlike national economies in that there are unlikely to be catastrophic moral consequences from centrally planning Walmart.
It feels as though what we’re really debating, here, is whether central planning of national economies is possible, and if so, whether it’s moral. I would argue no to both.
Edited to clarify: What I meant by (a) is that, although Walmart has a larger value of transactions than the USSR did, it’s much much simpler in terms of the total number of relationships between components, total number of products and services, and options available. They’re not even in the same ballpark, I’d estimate many many orders of magnitude difference.
Apologies for the delayed reply, and i’m not sure it matters anymore, but the point of People’s Republic of Walmart is to point out that a perfect algorithm is not needed for large-scale allocation of resources. I don’t think morality factors in here at all, unless you mean to imply that slavery or some other iffy aspect would be involved here.
I can’t claim to be an expert on the subject, so I feel this discussion will probably go in circles, but I would argue the complexity of numbers of goods and services offered isn’t that different at all. And, if we’re counting other areas, like healthcare, for example, we have the golden era of NHS as an example to point out (the NHS is also another example mentioned in the book, and how it became much worse than it could have been after the community health councils were dismantled, and after funding was gutted).
To reiterate my first comment, I think the book makes a very strong argument that we already have all the pieces necessary (and most of them are already in play) for large-scale planning of economies is possible. I only now just noticed you said “central planning”, which is not an ideal method, and I would argue central planning would definitely be insufficient. Distributed, democratic planning would be much better at fulfilling people’s needs. (again, see the NHS’ councils as an example.)
The Soviet Union didn’t have computers like we do today, and considering the groundwork that Walmart and Amazon have already laid down, I think we could handle higher levels of complexity when planning.
I can’t claim to be an expert on the subject, so I feel this discussion will probably go in circles, but I would argue the complexity of numbers of goods and services offered isn’t that different at all.
Walmart offers 35m products, with 2.2m employees. Bearing in mind that a first world economy includes the labour of all of its citizens, plus a vastly larger range of goods and services, I think that it is in fact many many orders of magnitude different, as I already suggested.
Distributed, democratic planning would be much better at fulfilling people’s needs.
Can you explain this in a bit more detail? And how would distributed, democratic, planning handle dissenters? What say I didn’t want my labour or products to be planned in accordance with the system?
Walmart offers 35m products, with 2.2m employees. Bearing in mind that a first world economy includes the labour of all of its citizens, plus a vastly larger range of goods and services, I think that it is in fact many many orders of magnitude different, as I already suggested.
I can’t find any numbers on the different types of products a country produces, so I can’t comment on this. The intuition is that it shouldn’t matter, because if you take into consideration that Walmart/Amazon’s model is scalable, and if you consider both to have a microcosm of all the products that are planned, then it follows that a country could be planned in their mold. Without hard numbers, this is hard to prove. I’ll research it, but I don’t promise I’ll get back to you in this thread, because it has probably way overrun its course.
Can you explain this in a bit more detail? And how would distributed, democratic, planning handle dissenters? What say I didn’t want my labour or products to be planned in accordance with the system?
Dissenters is the wrong way to think about it, I guess. I think it’s useful to think of the planning system as an “information tap”. That is, There is a number of people (consumers, customers, patients, whatever, people who make use of the services and products being offered) that are in constant need of the aforementioned services and products. There is a number of producers and service providers.
The main question is how to allocate work such that the flux of necessity is satisfied. The common mainstream answer is through the market. I would argue that planning would be the more efficient answer, simply because it’s easier to produce according to forecasts of what one will use, rather than have the main mathematical factor be profit. It is not profitable to produce antibiotics, for example (and most medicines in general), and it is highly profitable to produce fossil fuels. The democratic part here would basically mean that the consumers are in charge of the flux of information, explaining their needs, and their needs would be input into some forecasting system. Producing according to current usage and forecasts is much, much more efficient at satisfying the needs with which that is not profitable.
If such a planning system would attain critical mass in the number of products it plans and consumers it serves, certainly, you’re free to not join, but not joining would mean you’re basically serving an ever-diminishing amount of people, producing ever-more-useless products and services.
Apologies if this wasn’t clear. I mean to present my current viewpoint based on my current knowledge of the world, and I don’t think I understand planning well enough that a clear explanation would flow easily yet.
This post reads great until the conclusion, where it makes claims I find odd. The first one is:
Let’s start with computational complexity. […] which according to Cosma would correspond to an optimization problem that would take a thousand years to solve on a modern desktop computer. However, if Moore’s Law holds up, it would be possible in 100 years to solve this problem reasonably quickly.
We’re already claiming Moore’s law doesn’t hold anymore, and it seems odd to rely on it to hold for the next 100 years to make planned economies a reality at scale. As @luiz pointed out below, some modern corporations run large de facto planned economies, so an answer may lie in some direction other than raw increase in computational power.
As described earlier, the second serious issue with a centrally planned economy was data quality: […] Whether a government would be able to harness [data on demand] as competently as Amazon is doubtful, and it’s obviously worth asking whether we would ever want a government to be using that type of data.
And here I was thinking the lesson of the last decade of social media was that we couldn’t trust tech companies with our data. Why is it only worth asking whether we’d want governments to use that type of data? Where does this default trust of large corporations come from?
This latter objection seems much more serious than the former, which is “just” a tech problem. Even if we assume planned economies are better than capitalist ones, how can we trust an institution with the data it needs to run such an economy?
Operations Research and Supply Chain Management have made a lot of amazing progress over time. A few weeks ago I was thinking about this general problem and I scraped the Amazon economics job listings to see what skills they seemed to prioritize. Lots of reduced form causal inference (A happens before B, driven by timeseries), forecasting, empirical industrial organization (how to be a good oligopoly with Walmart and Alibaba) etc… generally leaning hard into the data-driven subfields as expected.
There are people who claim that Amazon and Walmart could be the basis for a centrally planned economy. Tristan’s argument (with some of my rephrasing for technical, less polemic readers) is that:
My perspective is that this is all a rerun of the Gilded Age Robber Barons, but with the compute power to tackle far more verticals. Maybe all of them. It seems that Amazon is already our centrally-planned economy to a large degree. In a “normal” climate that I was trained to think within in Industrial Organization classes while studying economics in college, Amazon would have been split up by now due to the difficulty that a new firm would have in competing with them on anything. But we do not live in that climate, and we do not have the ability that Teddy Roosevelt had to “bust the trusts.”
In 1944, the major economies formed the Bretton Woods system in an attempt to control the forces of international finance that were widely viewed as contributing to the struggles of everyday people. Economic insecurity for everyday people resulted in nationalism, which resulted in genocides and the world wars. Among other things, Bretton Woods introduced capital controls that stood in the way of investors quickly pulling out of one country and reinvesting elsewhere. This is also why you used to need things like travelers checks when you traveled. But this friction had lots of nice effects that contributed to the rising of real wages. In Detroit, the unions were able to effectively maintain high wages because the auto companies could not easily shift manufacturing to another country. It meant that large corporations could be meaningfully opposed without them simply running off to another country. It meant that real wages continued to rise between 1944 and 1971 along with rising profits. Since Bretton Woods was ended by Nixon (who wanted to print money to pay for Vietnam War debts, which messed up the core structure of the system, dissolving it) real wages have stagnated due to labor losing its negotiating position. As programmers, our real wages are going down as well, for indirect but intimately related reasons.
Now, we have trade agreements that make it easier for a firm to simply shift their manufacturing and investment to another country if the workers in one place want a living wage, or if the government wants to introduce a stronger social safety net which would raise the taxes on a corporation. Countries are in competition with each other to have the crappiest social support systems, because that will attract investment that will provide some people with jobs. Countries worldwide are afraid to impose the standard antitrust measures on companies that can more credibly threaten to just go somewhere else. This is why Amazon is unopposed.
I’m quite pessimistic about this situation. I believe that because we have no credible way of preventing capital flight without a strong international agreement (which seems unlikely to me, but hopefully I’ll be disproven) there will be no real way to prevent oligopolies from destroying the abilities for everyday people to live with dignity, inequality will continue to explode (although individual wealth does not tend to move as much as investment capital in response to taxation, unless you’re a billionaire living in a particular Swiss canton who will negotiate with the canton for a lower tax rate otherwise you’ll move down the street to a canton who will give you a better deal). I see this as a significant structural risk that is likely to lead to WWIII before we get the next Bretton Woods. But for those of us who still have jobs, we’ll be enjoying our centrally planned economy of Amazon-delivered goods. This is a big reason why I’m motivated to take 3-6 months off of work every year. I have a lot of doubt about the near future, and I think it’s important to enjoy life without taking the current relative stability for granted.
I think that Yanis Varoufakis’s view that he lays out in Another Now is particularly compelling, if anyone is interested in “Economic Sci-Fi”. His ideas are very similar in some ways to an expanded Bretton Woods along with a realization of The high-growth high-social-standards Meidner Plan that became possible with the capital controls of Bretton Woods. He advocates keeping the market to avoid the inefficiencies and abuse of planned economies, while proposing an alternative to the labor market for letting people live with dignity in an increasingly fast-moving economy that will result in firms continuing to fail when they cannot compete. High-growth, low-inflation, without the social consequences that lead to genocide and war.
Cosma Shalizi’s post this references is much more detailed: https://crookedtimber.org/2012/05/30/in-soviet-union-optimization-problem-solves-you/
This is great; I’ve heard bits of this story, mostly as stories about Leonid Kantorovich, but never quite a blow-by-blow in this fashion. Economics is an optimization problem, and current market capitalism works as well as it does because it combines lots of local and not-so-local optimization processes, and incentivizes doing the work to gather information and optimize processes. Looking at it from a data-science point of view is thoroughly fascinating, and does a good job of describing the problems with capitalism or with other systems as well: computational complexity, data quality, and objective function.
This is really interesting, especially considering that if alpha-things can win at go, StarCraft, and protein folding, it seems like “100 years of Moore’s law” is an excessively conservative estimate for how long it’ll take before we have the tools to actually solve this for real life in real life.
it’s already been thoroughly solved by amazon and walmart, i suggest you read people’s republic of walmart, it provides great insight into this exact problem (large-scale economic planning). for example, the book cites the fact that walmart, internally, is a larger planned economy than the ussr ever was
It is solved for the small subset of the economy and range of products that is Wal-Mart - and then, may still founder entirely on personal preferences. The economic calculation problem is still very much unsolved at the scale of first world economies.
walmart may be a small subset of the economy worldwide, but it is still larger than most countries, and its model can be used for planning
OK.. how?
Like, Walmart will probably, within applicable laws, avoid hiring disabled people. What should a country/society do with that information? It’s quite easy to break out the line items “care for the disabled” and take action to minimize them - often in ways that are quite frankly horrific.
i’m talking specifically about their forecasting and logistics procedures, which could be executed by anyone. not sure what your example is meant to point out
I think what he’s trying to point out is that, while Walmart is larger in dollar terms than many national economies, it’s also a) significantly simpler in terms of number of goods and services offered, and b) unlike national economies in that there are unlikely to be catastrophic moral consequences from centrally planning Walmart.
It feels as though what we’re really debating, here, is whether central planning of national economies is possible, and if so, whether it’s moral. I would argue no to both.
Edited to clarify: What I meant by (a) is that, although Walmart has a larger value of transactions than the USSR did, it’s much much simpler in terms of the total number of relationships between components, total number of products and services, and options available. They’re not even in the same ballpark, I’d estimate many many orders of magnitude difference.
You’ve expressed my position more clearly than I have, thanks.
Apologies for the delayed reply, and i’m not sure it matters anymore, but the point of People’s Republic of Walmart is to point out that a perfect algorithm is not needed for large-scale allocation of resources. I don’t think morality factors in here at all, unless you mean to imply that slavery or some other iffy aspect would be involved here.
I can’t claim to be an expert on the subject, so I feel this discussion will probably go in circles, but I would argue the complexity of numbers of goods and services offered isn’t that different at all. And, if we’re counting other areas, like healthcare, for example, we have the golden era of NHS as an example to point out (the NHS is also another example mentioned in the book, and how it became much worse than it could have been after the community health councils were dismantled, and after funding was gutted).
To reiterate my first comment, I think the book makes a very strong argument that we already have all the pieces necessary (and most of them are already in play) for large-scale planning of economies is possible. I only now just noticed you said “central planning”, which is not an ideal method, and I would argue central planning would definitely be insufficient. Distributed, democratic planning would be much better at fulfilling people’s needs. (again, see the NHS’ councils as an example.)
The Soviet Union didn’t have computers like we do today, and considering the groundwork that Walmart and Amazon have already laid down, I think we could handle higher levels of complexity when planning.
Walmart offers 35m products, with 2.2m employees. Bearing in mind that a first world economy includes the labour of all of its citizens, plus a vastly larger range of goods and services, I think that it is in fact many many orders of magnitude different, as I already suggested.
Can you explain this in a bit more detail? And how would distributed, democratic, planning handle dissenters? What say I didn’t want my labour or products to be planned in accordance with the system?
I can’t find any numbers on the different types of products a country produces, so I can’t comment on this. The intuition is that it shouldn’t matter, because if you take into consideration that Walmart/Amazon’s model is scalable, and if you consider both to have a microcosm of all the products that are planned, then it follows that a country could be planned in their mold. Without hard numbers, this is hard to prove. I’ll research it, but I don’t promise I’ll get back to you in this thread, because it has probably way overrun its course.
Dissenters is the wrong way to think about it, I guess. I think it’s useful to think of the planning system as an “information tap”. That is, There is a number of people (consumers, customers, patients, whatever, people who make use of the services and products being offered) that are in constant need of the aforementioned services and products. There is a number of producers and service providers.
The main question is how to allocate work such that the flux of necessity is satisfied. The common mainstream answer is through the market. I would argue that planning would be the more efficient answer, simply because it’s easier to produce according to forecasts of what one will use, rather than have the main mathematical factor be profit. It is not profitable to produce antibiotics, for example (and most medicines in general), and it is highly profitable to produce fossil fuels. The democratic part here would basically mean that the consumers are in charge of the flux of information, explaining their needs, and their needs would be input into some forecasting system. Producing according to current usage and forecasts is much, much more efficient at satisfying the needs with which that is not profitable.
If such a planning system would attain critical mass in the number of products it plans and consumers it serves, certainly, you’re free to not join, but not joining would mean you’re basically serving an ever-diminishing amount of people, producing ever-more-useless products and services.
Apologies if this wasn’t clear. I mean to present my current viewpoint based on my current knowledge of the world, and I don’t think I understand planning well enough that a clear explanation would flow easily yet.
This post reads great until the conclusion, where it makes claims I find odd. The first one is:
We’re already claiming Moore’s law doesn’t hold anymore, and it seems odd to rely on it to hold for the next 100 years to make planned economies a reality at scale. As @luiz pointed out below, some modern corporations run large de facto planned economies, so an answer may lie in some direction other than raw increase in computational power.
And here I was thinking the lesson of the last decade of social media was that we couldn’t trust tech companies with our data. Why is it only worth asking whether we’d want governments to use that type of data? Where does this default trust of large corporations come from?
This latter objection seems much more serious than the former, which is “just” a tech problem. Even if we assume planned economies are better than capitalist ones, how can we trust an institution with the data it needs to run such an economy?
suggest
finance[Comment removed by author]