Tuesday, August 31, 2021

Economies Of Scale

Steve Randy Waldman is a very interesting writer. He has a fascinating short post entitled Economies of scale in which he distinguishes four different types of "economies of scale". In reverse order, they are:
  1. Insurance
  2. Market power
  3. Network effects
  4. Technical economies
The effects of economies of scale in technology markets, such as storage media, digital preservation and cryptocurrencies, is a topic on which I have written many times, drawing heavily on W. Brian Arthur's 1994 book Increasing Returns and Path Dependence in the Economy. Below the fold I discuss Waldman's classification of them.

Brian Arthur's analysis focuses on the effects of economies of scale, where Waldman's focuses on their causes.

Insurance

Waldman's last type is probably one that few have identified, the economy of scale in risk pools:
there are economies of scale in the insurance of stakeholders, which is a genuine efficiency and of tremendous social value. A large firm can provide generous sick leave or parental leave, because the absent employee is one of a large stable among whom the extra burden can be shared, and over which the financial cost can be amortized. For a small firm, even temporary loss of a skilled worker can paralyze the business. And a small firm’s finances may be too weak to pay the leave. “Mom and Pop” firms are notoriously shitty at providing flexibility and insurance benefits not because Mom and Pop are bad people, but because a big insurance pool functions better than a tiny one. This is a real economy of scale.
My go-to source on risk and insurance is Peter L. Bernstein's Against The Gods: The remarkable story of risk. He writes:
In practice, insurance is available only when the Law of Large Numbers is observed. The law requires that the risks insured must be both large in number and independent of one another, like successive deals in a game of poker.
The idea that only a big firm can provide the necessary size of risk pool to generate economies of scale is a curiously US-centric one. Waldman even acknowledges this:
However, much of this advantage of bigness would disappear if the social insurance function were sensibly provided by the state instead of our relying upon individual businesses to offer “benefits”. (The state cannot relieve businesses of the risk that a critical employee may need to step back, but this risk fades even at small-to-medium scales beyond “Mom and Pop”.)
This "socialized medicine" might work in "less efficient" economies such as the EU, but how could it possibly work in the US? Two words: Medicare, and VA. Note that both huge risk pools deliver cheaper care despite catering for populations commercial insurers consider too expensive to cover. The point being that it isn't just the size of the risk pool, but its diversity that matters.

Market Power

Waldman is again on point when he describes two forms of market power. First:
There is traditional monopoly or market power by which firms can extract rents from workers, suppliers, and consumers. Market power is a correlate of scale that looks great from any firm’s perspective, but its “efficiencies” are just transfers from other stakeholders, and are destructive in aggregate.
Unfortunately, this entrenched destructiveness, a legacy of the Chicago school, is hard to displace because, second:
There are resource and coalitional “economies of scale”, the way very large firms can engage in predatory pricing, or coordinate the activities of lawyers and lobbyists and media, and eventually politicians and regulators, in a firm’s interest. Again, these are not true “economies” at all. They may benefit incumbent firms, but are of negative social value.
The "negative social value" of this second form was demonstrated in the Global Financial Crisis, when the banks collectively arranged for the negative effects of their reckelss lending to fall everywhere but on themselves.

Network Effects

Because of the dominance in technology markets of the FAANGs, network effects are what most people think of as technology's economies of scale. As we see with markets such as social media, search and operating systems, network effects provide dominant companies both extraordinary profits and very strong defenses against rising competitors. Waldman correctly points out the need for governments to regulate "natural monopolies":
These are real economies, but as John Hussman describes them, network effects should be classified as “uninvented public goods”. Firms should be rewarded for discovering them — and indeed they have been and are rewarded, quite handsomely — but networks should not remain monopoly franchises of private entities indefinitely. They are “natural monopolies”, which competition will not regulate in the public interest. They should fall, whether through outright ownership or as “regulated utilities”, into management by the state. [1]
His footnote 1 is perceptive:
[1] Yes, states are corrupt, in that they often improperly serve particular private interests. But the only reason we don’t understand firms to be even more corrupt is that serving particular private interests is each firm’s overt function and purpose. It’s not that monopolists behave better, from a social perspective, than states, it’s that their misbehavior gets coded as legitimate competence.

Technical Economies

As regards technical economies of scale, Waldman writes:
States should not try to insist that Mom and Pop should be able to bootstrap competitors to GM out of savings from their second job. But technical economies of scale peter out at scales much smaller than megafirms. Tesla, which (in physical, rather than casino-financial terms) is not so big, can compete with GM. Technical economies of scale require the scale of a factory, producing in quantities that fully amortize fixed capital costs, but not more than that.
Waldman is just wrong about this, at least in the technology space. Its massive size in "casino-financial terms" is precisely the reason Tesla can compete with GM. The world is littered with small car companies that, lacking lavish early backing from a high-profile billionaire, could not compete with GM. And lets not lose sight of the fact that, until recently, Tesla's profits came from selling carbon credits; it lost money selling cars.

Tesla isn't even a good example. In many technology markets the investment needed to build "a factory, producing in quantities that fully amortize fixed capital costs" is immense. To compete in chip manufacturing you need a state-of-the-art 5nm fab, costing $12B. To stay competitive, you'll need to be planning the next one, at 3nm and $25B. To compete in AI you need huge data centers to train the models. Arguably, these "economies of scale" are actually financial rather than technical. The larger you are, the easier it is to finance investments at the necessary scale.

This has anti-trust implications. Would breaking up TSMC improve consumer welfare? The fragments combined wouldn't be able to afford TSMC's $200B investment program, so the world would take much longer to get to 3nm and beyond, increasing the price of chips that go into so many products. The co-evolution of dominant technology companies and their suppliers has created an investment "moat" protecting them from emerging competion. When ASML's EUV machines cost $160M each and the queue for them is several years long a new entrant is hopeless.

Conclusion

I find Waldman's classification useful. Government regulation is undoubtedly needed to counteract network effects and the abuses of market power, but as he points out these are not the only economies of scale in play. Governments certianly have a role to play in eliminating the economy of scale that the employer-based (and massively dysfunctional) US health insurance system imposes. But these still leave the technical economies of scale that Waldman downplays.

Brian Arthur's analysis of the way economies of scale drive market concentration is agnostic to the cause of the economies. It is hard to see how, at least in many technology markets, governments could push back against the very large technical economies of scale. So it seems we are doomed to live with highly concentrated markets. Perhaps we can learn from Raymond Zhong and Li Yuan's The Rise and Fall of the World’s Ride-Hailing Giant:
Under Xi Jinping, the Communist Party’s most powerful leader since Mao, China has taken a hard ideological turn against unfettered private enterprise. It has set out a series of strictures against “disorderly” corporate expansion. No longer will titans of industry be permitted to march out of step with the party’s priorities and dictates.
...
On issues like data security, privacy and worker protections, Beijing’s scrutiny is long overdue. Yet Chinese officials have moved against tech companies with a speed and ferocity that might unsettle even the most ardent Western trustbusters.

2 comments:

Blissex2 said...

As a rule, the concept of "economies of scale" is highly misleading, because there are very few if any cases where scale itself gives lower costs.

In general there are economies of specialization rather than of scale, and while scale up to a point enables specialization, that runs out pretty soon.

The "economies of scale" propaganda is usually pushed by those who want to argue in favour of massive oligopolies or monopolies, by implying that economies of scale always improve with scale.

Instead it is pretty clear that economies of specialization soon run out or become negligible.

My usual example are pressed to stamp auto body parts from steel sheets or ingots: a huge press can stamp a lot more body parts per say than a worker with a hammer can shape them, but is a lot more specialized (making and changing a mould can take days or weeks). Also bigger presses can mould parts faster than smaller presses, but that soon reaches limits, and then it becomes linear, as more capacity just requires more presses, rather than bigger ones.

For the computer technology market that is pretty similar: for example most economies of specialization happen at the component level, and as all vendors buy from the same suppliers, they all have the same economies of specialization. That happens also at the web server/data center level, where specialized builds and buildings soon run out of economies.

Blissex2 said...

«In many technology markets the investment needed to build "a factory, producing in quantities that fully amortize fixed capital costs" is immense. To compete in chip manufacturing you need a state-of-the-art 5nm fab, costing $12B. To stay competitive, you'll need to be planning the next one, at 3nm and $25B. To compete in AI you need huge data centers to train the models. Arguably, these "economies of scale" are actually financial rather than technical.»

But 12B and 25B are small sums in the global economy, which is dominated by what neoliberal Economists call a "savings glut". There is a lot of capital looking for investment opportunities, and 2008 demonstrated that to protect their donors governments like that of the USA can easily find one or two trillions to refill the bonus pools of Wall Street and the City of London.
Those 12B and 25B are small sums compared to what is invested in real estate every year too, and if private investors prefer real estate it is because it gives much better and safer profits, not because there is scarcity of capital.

As to the $160M ASML steppers etc., again those are small sums, and the machines cost that much only because ASML has no incentive to build smaller machines, as its existing customers are like Intel and TSMC already giant oligopolies, and would really hate for ASML to make entering the market easier for smaller fabs. ASML, Intel, TSMC (and perhaps to some extent SAMSUNG) are just parts of the same oligoply.