NOTE: Last week’s excerpt from World After Capital described how prices cannot exist for many of our most important attention allocation decisions. Today I describe how production functions with network effects result in power law distributions that have bad social and economic implications.

Economics is not normative when it comes to the distribution of income and wealth. Many different outcomes are possible and what is realized depends a lot on the underlying production functions. Consider first a fairly manual production function such as was common pre-industrialization. If you were a cobbler making shoes by hand, there were only so many shoes you could produce. I don’t know if such data is available, but the output of cobblers likely formed a normal distribution, with even the most productive cobbler making only a small multiple of the number of shoes of the average cobbler.

Then along came industrialization and with it economies of scale. If you made more cars you could make them more cheaply and that was true until you got to a fairly large number of cars relative to total demand. That’s why, over time, we wound up with relatively few car manufacturers around the world and the owners of the surviving largest ones wound up with large fortunes (e.g., the Ford or the Piech families). It turned out that many service businesses have relatively small economies of scale (e.g., a hair salon). That has allowed a great many service businesses to exist. The biggest exception to this has been financial services in which a few large banks, insurance companies, and brokerage firms tend to dominate.

Now, however, with digital technologies we are seeing a shift to power laws for many more situations. For instance, on YouTube the most watched video has been watched billions of times compared to the vast majority of videos which have been watched just a few times. Or in ecommerce, Amazon is an order of magnitude larger than the next biggest competitor and several orders of magnitude larger than most ecommerce companies. The same goes for apps in the appstore. The leading apps have hundreds of millions (and some even billions) of users. But the vast majority of apps has just a few users.

Digital technologies are driving these power laws because of network effects combined with zero marginal cost. As I explained in the chapter on digital technology this means that in principle we need only one medical diagnosis systems to serve the entire world (in practice we would want several). So far we have seen one social network by far dominate all others. We have one search company dominate all others. Protected markets, such as Russia and China, have their own search and social network companies, but their size also follows a power law distribution.

This shift to power laws everywhere is driving a huge increase of wealth and income inequality to levels that are now beyond the peak of the early 1900s. At that time it was undone by two world wars, which destroyed much of the accumulated existing wealth as well documented in Thomas Piketty’s book “Capital in the Twenty-First Century.” Inequality beyond a certain level is socially corrosive, as people effectively start to live in different world that is disconnected from the problems faced by large parts of the population. There is no self-corrective to this kind of excessive, power-law driven, inequality built into capitalism.

Beyond the social implications of such inequality, the largest digital companies also wield undue political and market power. A recent example of that was the dramatic drop in the market capitalization of pharmacy chains when Amazon acquired a relatively small online pharmacy, such signaling its intent to compete in that market. Historically market power was bad because it produced inefficient allocations due to excessive rents (and such artificially low quantity). In digital markets powerful companies have often pushed prices down or made products free entirely thus causing seemingly no harm to consumers. The harm here comes via reduced innovation as companies and investors stop allocating capital to trying to bring better alternative products to market.

The purported self-corrective in capitalism for market power is “creative destruction” as first described by Joseph Schumpeter. And indeed if you look at the dominant companies today they are quite different from those of the Industrial Age. But going forward this may be quite a bit harder, if not impossible. Why? During the Industrial Age machines served a specific purpose. This meant when a new product or a new manufacturing technology became available, the installed base of machines became essentially worthless and was a weight on the incumbents. Today, however, the new incumbents operate general purpose computers, which take information as one of their key inputs. They can easily implement a new product, or add a feature to an existing one, or adopt a new algorithm. Production functions with information as a key input have a property known as super-modularity: the marginal benefit of additional information is higher the more information you already have. This gives the digital incumbents tremendous sustained power as they get more marginal value from a new product or service than a new entrant.

NOTE: Last week’s excerpt from World After Capital described how prices cannot exist for many of our most important attention allocation decisions. Today I describe how production functions with network effects result in power law distributions that have bad social and economic implications.

Economics is not normative when it comes to the distribution of income and wealth. Many different outcomes are possible and what is realized depends a lot on the underlying production functions. Consider first a fairly manual production function such as was common pre-industrialization. If you were a cobbler making shoes by hand, there were only so many shoes you could produce. I don’t know if such data is available, but the output of cobblers likely formed a normal distribution, with even the most productive cobbler making only a small multiple of the number of shoes of the average cobbler.

Then along came industrialization and with it economies of scale. If you made more cars you could make them more cheaply and that was true until you got to a fairly large number of cars relative to total demand. That’s why, over time, we wound up with relatively few car manufacturers around the world and the owners of the surviving largest ones wound up with large fortunes (e.g., the Ford or the Piech families). It turned out that many service businesses have relatively small economies of scale (e.g., a hair salon). That has allowed a great many service businesses to exist. The biggest exception to this has been financial services in which a few large banks, insurance companies, and brokerage firms tend to dominate.

Now, however, with digital technologies we are seeing a shift to power laws for many more situations. For instance, on YouTube the most watched video has been watched billions of times compared to the vast majority of videos which have been watched just a few times. Or in ecommerce, Amazon is an order of magnitude larger than the next biggest competitor and several orders of magnitude larger than most ecommerce companies. The same goes for apps in the appstore. The leading apps have hundreds of millions (and some even billions) of users. But the vast majority of apps has just a few users.

Digital technologies are driving these power laws because of network effects combined with zero marginal cost. As I explained in the chapter on digital technology this means that in principle we need only one medical diagnosis systems to serve the entire world (in practice we would want several). So far we have seen one social network by far dominate all others. We have one search company dominate all others. Protected markets, such as Russia and China, have their own search and social network companies, but their size also follows a power law distribution.

This shift to power laws everywhere is driving a huge increase of wealth and income inequality to levels that are now beyond the peak of the early 1900s. At that time it was undone by two world wars, which destroyed much of the accumulated existing wealth as well documented in Thomas Piketty’s book “Capital in the Twenty-First Century.” Inequality beyond a certain level is socially corrosive, as people effectively start to live in different world that is disconnected from the problems faced by large parts of the population. There is no self-corrective to this kind of excessive, power-law driven, inequality built into capitalism.

Beyond the social implications of such inequality, the largest digital companies also wield undue political and market power. A recent example of that was the dramatic drop in the market capitalization of pharmacy chains when Amazon acquired a relatively small online pharmacy, such signaling its intent to compete in that market. Historically market power was bad because it produced inefficient allocations due to excessive rents (and such artificially low quantity). In digital markets powerful companies have often pushed prices down or made products free entirely thus causing seemingly no harm to consumers. The harm here comes via reduced innovation as companies and investors stop allocating capital to trying to bring better alternative products to market.

The purported self-corrective in capitalism for market power is “creative destruction” as first described by Joseph Schumpeter. And indeed if you look at the dominant companies today they are quite different from those of the Industrial Age. But going forward this may be quite a bit harder, if not impossible. Why? During the Industrial Age machines served a specific purpose. This meant when a new product or a new manufacturing technology became available, the installed base of machines became essentially worthless and was a weight on the incumbents. Today, however, the new incumbents operate general purpose computers, which take information as one of their key inputs. They can easily implement a new product, or add a feature to an existing one, or adopt a new algorithm. Production functions with information as a key input have a property known as super-modularity: the marginal benefit of additional information is higher the more information you already have. This gives the digital incumbents tremendous sustained power as they get more marginal value from a new product or service than a new entrant.