Manufacturers' response to 'supercycles' in the market for computer memory

Back in May, David Oks wrote a fascinating Substack article about the economics of dynamic random access memory (DRAM) - the memory that is used in computers, smartphones, and gaming consoles, but also importantly in servers and data centres, such as those used to train generative AI models. I'm not going to focus on the effect of generative AI in this post, but instead on this particular part of Oks's article, which caught my attention:

And that combination—capital-intensive manufacturing plus fungibility—is a punishing combination. Because memory is fungible, the industry is intensely cyclical: the entire history of the DRAM industry is a history of boom-and-bust supercycles. First, strong demand from one sector or another—like Windows PC adoption in the 1990s—drives surging prices and a wave of investment from every player; cumulative overinvestment in an undifferentiated good produces oversupply; and then oversupply leads to collapsing prices.

And because production is so expensive, those down-cycles turn out to be existential: the memory industry is marked by constant wreckage. Intel dominated the memory game in the early 1970s but left in the 1980s, opting to focus on processors. Texas Instruments and IBM, also once major players, left in the 1990s. Germany’s Qimonda collapsed in 2009; Japan’s Elpida, once the world’s third-largest DRAM manufacturer, declared bankruptcy in 2012.

In my ECONS101 class, we teach a model of dynamic supply and demand that can be used to explain the boom-and-bust 'supercycles' that Oks describes. Consider the market for memory, and assume that it is perfectly competitive - most importantly, there are no barriers to entry into the market or barriers to exit from the market.[*] The market for memory is shown in the diagram on the left below. The diagram on the right will track changes in memory manufacturers' profits over time. Initially (at Time 0) the market is at equilibrium (where demand D0 meets supply S0) with price P0, and memory manufacturers are making profits π0. Now say there is a permanent increase in demand at Time 1, to D1. This increase in demand may be because of Windows PC adoption, or some other positive demand shock. Prices increase to P1, and memory manufacturers' profits also increase (to π1). There are no barriers to entry (this is a perfectly competitive market), so the higher profits encourage new manufacturers to enter this market (or more realistically, they encourage the existing manufacturers to increase capacity). However, new manufacturing capacity takes time to bring online, and firms make their investment decisions independently. By the time all of that new capacity becomes available, supply may have overshot the level required to simply meet the higher demand. Supply increases to S2 (more producers) at Time 2. Price falls to P2, and memory manufacturers' profits also fall (to π2).

Next, at Time 2 profits are low and some memory manufacturers will choose to exit the market (no barriers to exit because this is a perfectly competitive market), or more realistically it encourages the manufacturers to reduce their capacity. This helps explain the pattern that Oks describes, with Intel, Texas Instruments, IBM, and others exiting the market. Supply will decrease to S3 (fewer producers) at Time 3. Price will increase to P3, and memory manufacturers' profits will increase to π3. So, these 'supercycles' arise as memory manufacturers enter and exit the market in response to an initial increase in demand.

Now, memory manufacturers are not stupid. The manufacturers that remained after previous busts realised that they needed to change their approach in order to avoid these problems. Oks notes that:

And decades of collapse and consolidation left only a few players standing. In the 1990s, there were perhaps 20 meaningful producers of DRAM around the world; today there are three that account for more than 90 percent of global production. South Korea has two, SK Hynix and Samsung; and the United States has one, Micron.

And these memory makers have learned a very particular lesson from the unforgiving history of their industry: always leave demand unmet.

So, rather than rapidly increasing production in response to higher demand, the memory manufacturers have instead become much more cautious about adding capacity, deliberately allowing some demand to remain unmet. That reduces the risk that increasing supply causes a decline in prices and profits. However, the downside is, as Oks notes, that we now have a global shortage of memory as current production levels (and manufacturing capacity) are unable to keep up. That may open opportunities for new manufacturers to enter the market, even if they are less efficient (higher cost) than the incumbents. And therein lies the problem - by deliberately restricting supply tight, the incumbent manufacturers keep prices and profits high and that may eventually set off the next supercycle.

*****

[*] The market for computer memory isn't actually perfectly competitive, as there are likely to be large economies of scale in memory manufacturing. That means that a small firm entering the market would be at a large cost disadvantage to the large incumbent firms, and so small firms would be deterred from entering the market. However, there is another way of looking at this situation. Instead of considering new firms entering and exiting this market, think about the incumbent firms adding and subtracting additional manufacturing capacity. This has the same effect of increasing and decreasing supply, and leads to the same dynamic pattern in prices and profits.

Europe faces a flood of cocaine

The Financial Times reported back in July:

Cocaine production in Latin America has quadrupled in the past decade, with criminal networks in Colombia, Peru and Bolivia exploiting global trade routes to shift vast quantities of the drug. Cocaine is also increasingly shipped from Brazil into Europe via west Africa.

The UN warns that supply could soon exceed demand, increasing traffickers’ incentives to dump even more product on to Europe’s streets...

The result is that more cocaine is available in Europe today than in the 1980s — often seen as the drug’s heyday — according to the UN. Last year alone, its residue in city wastewater rose by more than a fifth, according to the EU’s drugs agency...

Lower prices and frictionless dealing on popular encrypted messaging apps have made cocaine more attainable.

My ECONS101 class covered the model of supply and demand last week, and this seems like a good example. The European market for cocaine is shown in the diagram below. The market was initially in equilibrium, where supply S0 meets demand D0. The equilibrium price of cocaine was P0, and the equilibrium quantity traded was Q0. The supply of cocaine has increased (a shift to the right, or down, of the supply curve for cocaine), to S1, due to increased production in Latin America being shipped to Europe. This lowers the equilibrium price of cocaine to P1, and increases the quantity traded to Q1.

However, the increase in cocaine supply hasn't had the same effect everywhere. The FT article also reports that:

One country where the retail price has risen rather than fallen, according to UN data, is the UK, which one former senior European police officer says may be linked either to stronger demand or to traffickers’ perception that it is riskier to smuggle cocaine into the country than elsewhere.

If cocaine traffickers believe that it is riskier to smuggle cocaine into the UK, that would decrease the supply of cocaine, and increase the price (see the diagram above, but with the change in supply reversed). That might also contribute to the increase in supply to Europe, if shipments that previously would have gone to the UK go to continental Europe instead. As for 'stronger demand', that effect is shown in the diagram below. The market was initially in equilibrium, where supply S0 meets demand D0. The equilibrium price of cocaine was P0, and the equilibrium quantity traded was Q0. If supply remained constant, and demand increased (as noted in the FT article), the demand curve shifts to the right (to D1). This increases the equilibrium price of cocaine to P1, and the quantity of cocaine traded to Q1.

The increase in demand could even lead to a price increase even with supply increasing as well. If the supply increased to S1, then the equilibrium price goes back to P0, with the quantity of cocaine traded increasing to Q1. 

Combining an increase in demand with an increase in supply is certain to lead to an increase in the equilibrium quantity. However, the change in the equilibrium price is ambiguous. Any smaller increase in supply than shown in the diagram would lead to a net increase in the price of cocaine in the UK, as would any decrease in supply. Any larger increase in supply would lead the equilibrium price to decrease.

Europe is, of course, not the only country dealing with an influx of cocaine. It has also been in the news in New Zealand recently as well. The underlying drivers are very similar, and are a recurring issue (see this post from 2017, related to the US cocaine market).

Read more:

• Cocaine makes a comeback in the U.S.

This week in research #143

Here's what caught my eye in research over the past week:

• Magnuson and Ruggles (open access) explore the creation of the first census microdata sample at the US Census Bureau in the early 1960s (I had no idea it was so long ago!)
• Coelli and Borland (open access) find that the decline in the estimated return to a bachelor's degree in Australia over the period from 2001 to 2016 is potentially explained by strong earnings growth in low‐skill occupations, due to increases in real minimum and award wages, as well as possibly the mining boom
• Bonilla-Mejía et al. (with ungated earlier version here) find that rising violence increases intentions, plans, and preparations to emigrate from Central America, particularly to the United States
• Delavande et al. (with ungated version here) find that student attendance at a public UK university is highly dependent on timetable structure, that students compensate for marginal non-attendance at some events with increased attendance at others within the same module, and that timetable features and attendance choices do not have systematic impacts on academic achievement
• Dang (with ungated earlier version here) finds that one additional international student per thousand working-age residents of a US local labour market increases the employment-to-population ratio by 0.19 percentage points and average hourly wages by 0.48 percent
• Frank and Vickery (open access) look at how LGB students differ from their heterosexual counterparts using data from the UK from 2012 to 2020, and find that LGB men and women both have large shifts towards Humanities subjects, where they perform better, and away from law, economics, management, and STEM, and that they have lower early-career salaries than heterosexual students

The political language of economists, and the associated principal-agent problem for government

Governments often employ economists as policy analysts or consultants, to provide advice on policy. That relationship between the government and the economist is a principal-agent relationship (regardless of whether the economist is a government employee or a consultant), as my ECONS102 class covered this week. 

In a principal-agent interaction, one person or group (the agent) is given the power to decide how to use resources that ‘belong’ to someone else (the principal). In the case of economists working for government, the government is the principal and the economist is the agent. The 'resource' that the agent is using is their work time, which has been paid for by the government.

Now, the government wants the economist to use that work time to provide unbiased economic or policy advice. [*] However, the economist has their own goals and motivations. They might use that work time to scroll TikTok, and then use ChatGPT to write the advice. Or, they might use the time to provide biased advice, based on their own political preferences. Either way, this is detrimental to the interests of the government. This is the essence of the principal-agent problem (or the agency problem).

Do economists act that way? Many economists (including, at times, myself) argue that we are non-partisan, and unbiased in our policy advice, even if we hold particular political views. But is that actually the case?

That is the question addressed in this 2024 article by Zubin Jelveh (University of Maryland), Bruce Kogut, and Suresh Naidu (both Columbia University), published in The Economic Journal (ungated earlier version here). They first link over 53,000 economists from the American Economic Association member registry in 1993, 1997, 2002, and 2009 to two measures of political activity: (1) their campaign contributions (between 1979 and 2012) drawn from the Federal Election Commission's website; and (2) their signing of one or more of 35 petitions aligned with the political left or right.

Next, Jelveh et al. obtain the full text of over 62,000 academic articles and over 17,000 NBER Working Papers published by a subset of 2471 of the economists from the larger sample, between 1973 and 2011. They then rank each of the articles and papers in terms of 'ideological valence', based on the phrases that appear in it, and use that to derive several different 'ideology scores' for the writing of each economist.

Jelveh et al. then start to do some analysis of the ideology scores, finding first that:

...the fields of finance, macroeconomics and industrial organisation are more conservative, while labour is considerably more liberal than the average. Other fields, such as history and international trade, show less political valence. We further see that faculty at business schools are more conservative, as are professors affiliated with ‘freshwater’ schools, while ‘saltwater’ schools have a left-wing bent. Professors of European origin also seem to be somewhat more conservative, and there seems to be no association with Latin American origin, full professor rank or top five department ranking.

The 'saltwater schools' are predominantly those on the east coast of the US, like Harvard or MIT, while the 'freshwater schools' are predominantly those in the Midwest, such as Chicago or Minnesota. Those results, and the results by field of economics, will not surprise many people who know those schools or those fields.

The more interesting results involve the next stage of the paper, where Jelveh et al. look at several economics debates, where there is a clear left-right divide in terms of expected effects. For example, conservative economists may be more likely to believe that the minimum wage reduces employment, while liberal economists may be more likely to believe that it doesn't. Jelveh et al. take several meta-analyses on the minimum wage, and similarly political topics, and look at the relationship between the estimated elasticity in each paper in the meta-analysis, and the estimated political ideology of the authors. They find a statistically significant relationship - elasticities reported by more conservative economists tend to be consistent with more conservative policy prescriptions, while elasticities reported by more liberal economists tend to be consistent with more liberal policy prescriptions.

Economists have political leanings (as does everyone else), Jelveh et al. show that those political leanings are correlated with the results that economists report in their research. This isn't to say that economists are engaged in falsifying data to support their political beliefs. Jelveh et al. have no evidence of that. However, their results could arise if economists choose to apply methods or investigate datasets that are more likely to lead to results that are consistent with their beliefs. Or, economists may simply choose not to publish results that are inconsistent with their beliefs. Either way, this research provides some evidence that knowing the political preferences of economists may be important in interpreting the results from their research.

In my ECONS102 class, we discuss various ways that the principal can act to reduce the principal-agent problem. Those options include stricter monitoring of the agent, paying efficiency wages, or performance-based pay (or delayed payment). In this case, the government might ask the economist agent to fully document their research, or subject it to careful peer review. This would constitute stricter monitoring. However, the relationship between the government and the peer reviewer opens up an additional layer of potential principal-agent problems (what are the political preferences of the peer reviewer?). The government might pay a wage to the economist that is much higher than the equilibrium wage, hoping that would motivate them to produce higher-quality and less biased work (because if they didn't, they would lose their job and have to work elsewhere for less). Government might also use performance-based pay, or may hold back payment until after a peer review, or a replication of any analyses. However, for all of those solutions, some form of monitoring is still required in order to determine the quality of the work. There is an additional problem, though. Efficiency wages and performance-based pay work best when the agency problem involves the effort the agent puts into their work. Those solutions may be less effective when the problem arises from sincerely held beliefs about which models, methods, or evidence are most appropriate.

It seems that the principal-agent problem for governments employing economists (and, possibly, other consultants) is challenging to solve. Perhaps the best that governments can do is insist on transparency around economists' assumptions, methods, and evidence, and subject their analyses to replication and peer review. Political preferences may still matter, but good institutions can make it harder for those preferences to determine the advice that governments receive.

[HT: Marginal Revolution, back in 2024]

*****

[*] Or, maybe the government wants the economist to provide economic or policy advice that accords with the preferred policy platform of the government, rather than independent or unbiased advice. In that case, the nature of the principal-agent problem changes. Nevertheless, in both cases, what really matters is whether the economist's goals and motivations are aligned with those of the government.