|
In this issue: - Invisible Efficiencies—If you're looking for the impact of AI in macroeconomic data, keep in mind that some of its effects will show up as an increase in the supply of labor, because better job matching means more people working. This is still an efficiency gain, but can easily lead to lower measured output per hour, at least in the short term.
- Making a Market—GameStop finds a new trade to offer to its favorite counterparty.
- Outsourcing—Sometimes a superficially expensive product is actually the cheapest option.
- Multicloud—What happens when a cloud provider accidentally deletes a customer?
- Can Frackers Collude?—Frackers' economics have improved since they stopped drilling so much. If they wanted to form a cartel, though, could they pull it off?
- Infrastructure—China revises its economic model.
- Area Man Tweets—RoaringKitty is back.
Invisible Efficiencies
The other day, my wife and I bought a new washer/dryer, which, because it's a consumer product in the year 2024, means that it will manage to incorporate AI somehow. It does a pretty good job so far, and for all I know it's using some incredibly sophisticated model of dampness-dispersal to minimize the energy consumption needed to dry clothes. The main impact of this purchase is the slight efficiency gain of having a washer and dryer in the same device instead of two of them, offset by the inconvenience of a longer wash/dry cycle. The world has come a long way from when laundry took up to a day of manual labor once a week, but the good people at LG correctly suspected that consumers would, in time, forget this incredible bounty of labor-saving and start wondering if things could be even faster.
And that's the basic story of a surprising share of year-to-year economic growth: big discoveries matter a lot, but the way they show up in GDP is when they're scaled to a large population. And once those discoveries are operating at full scale, they create a platform for incremental improvements. Those improvements can happen over the course of product cycles, where usage data and customer feedback from version 1.0 makes for a better version 2.0 (or, along some dimension consumers tolerate, worse in a way that reduces its cost). Or, for products that are either pure software or are physical goods controlled through software, through over-the-air updates (our washer/dryer has already gotten its first software update).
Two recent stories highlight this trend, both as a source of economic growth and as a form of invisible-by-default growth: The Economist has a fascinating piece about how the grid is getting smarter ($): sensors can monitor weather to get a better sense of the absolute maximum capacity for a given power line, and, longer-term, smarter arrangements of batteries, also informed by usage data. Meanwhile, the NYT covered how airlines are getting better at predicting cancellations and rerouting customers, improving flight plans based on weather to reduce fuel consumption, and sending people more detailed explanations of why their flights have been delayed or canceled.
These stories help answer the current version of what's been a surprisingly durable question: today, people ask why, if AI is such a big deal, its economic footprint is so small. OpenAI is worth $80bn, but that's about 50x revenue. Or, put another way, OpenAI may be running an influential business, but it's about as big a business as The Children's Place. The question of “where's the impact?” is older than the AI wave, though; in 1987, Robert Solow was already saying "You can see the computer age everywhere but in the productivity statistics."
Measuring productivity runs into some tricky paradoxes, though. This paper has a fun example:
Let’s say someone invents a useful painkiller, and that makes it easier for many people to show up to work and be productive. Output will rise, yet that advance will show up as an increase in labor supply, rather than as an increase in technology or scientific knowledge. Similarly, a new method for discovering oil may boost output, but that will be classified as an increase in oil supply, even though it does properly represent a form of scientific progress.
There are some companies whose economic impact is that they've unlocked a previously inaccessible cohort of workers. Uber and DoorDash have made it so that someone who can work in increments of a few hours at a time, with an unusual and inconsistent schedule, can still earn money. If they were trying to put in their fifteen hours a week at some business that operated on pen and paper, it would require saintly patience from management to get the maximum value out of them. But if they're driving for Uber or delivering for DoorDash, all they have to do is open an app, or decide not to. A cycle or two earlier, companies buying online content at piecework rates were basically monetizing the labor slack of stay-at-home moms who had enough time to earn some extra money but couldn't spend forty hours a week at an office.
When these companies get more efficient, some of what they're doing is inducing more demand: Uber increases the amount of traveling people will do by making some otherwise-inconvenient trips possible. But DoorDash isn't doing much to increase aggregate food consumption, just changing the mix away from restaurants and home-cooked meals and towards delivery and takeout. To the extent that they impact GDP, it will show up as an increase in the labor supply, an increase in employment, and some combination of an increase in the price of food and an increase in the consumption of transportation. To the extent that users of these products are using some of their extra time to work, and are earning enough to justify it, there might be a marginal productivity impact. But again, it's showing up in the wrong place: higher labor input gets measured, but less time spent driving to and from Chipotle in your own car is not part of measured economic output, so it won't show up in productivity per hour.
None of this is truly a problem with GDP, since GDP does a surprisingly good job of what it's trying to measure. But what it's really trying to measure is the scope of the taxable economy; it's impractical to assess a sales tax on the markup in value between a home-cooked meal and its ingredients. There are efforts to impute the economic value of a stay-at-home mom, but it's not exactly popular to argue that the family should owe payroll taxes on this—cooking at home, watching your own kids, and picking up after yourself instead of hiring a maid are the country's most popular forms of tax avoidance.
Meanwhile, these kinds of productivity gains have a subtler effect on the discourse. There's a lot of demand for either more renewables, more nuclear, or both, because these are discrete investments that get made in a public way and in regulated industries. There isn't much lobbying for the logistics system to continue getting incrementally more efficient, even though that would amount to a universal tax cut weighted to the consumption of physical goods. And that lobbying would be pointless, because the companies involved all have direct incentives to do exactly that. Uber knows that the less time drivers spend between gigs, the less those drivers need to get paid per minute they're actually driving; McDonald's knows that the more jobs it can make literacy-optional, the larger the labor pool it can hire from; DoorDash is perfectly aware that better predictions of demand spikes mean more satisfied customers, more optimally-busy drivers, and thus better contribution margins from its existing cohorts. DoorDash's revenues are a cut of the efficiency gain, so the result is visible there, but the aggregate impact on productivity is weak because of how the result gets measured. So if you're looking for broad-based productivity gains from the deployment of efficiency-improving technologies, focus on anecdotes over data: if the product is convenient, and someone's making money from it, then there are two possibilities: either it's unsustainable because one or more participants are getting a bad deal, or it's a productivity gain in disguise.
Diff JobsCompanies in the Diff network are actively looking for talent. See a sampling of current open roles below: - A seed-stage startup is using blockchains to enforce commitments and is in need of a fullstack developer with Solidity experience. (Remote)
- A top prop trading firm is looking for an intellectually curious, mathy generalist to work on projects spanning business strategy, technology, and markets. (NYC)
- A concentrated crossover fund is looking for an experienced full stack software engineer to help develop and maintain internal applications to improve investment decision-making and external applications to enable portfolio companies. (SF)
- A seed-stage startup is helping homebuyers assume the homeseller’s low-rate mortgage, and is in need of an experienced product manager. (NYC)
- A well-funded startup is building a platform to identify compliance risks associated with both human- and AI-generated outputs. They are looking for a cloud infrastructure engineer to join their team of world-class researchers. (NYC)
Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up. If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority. Elsewhere
Making a Market
Market-makers pay brokers for the privilege of trading with investors who aren't going to have much of a short-term edge in predicting prices. This is an old phenomenon—it's mentioned not just in E-Trade's S-1, but even in Understanding Media, where Marshall McLuhan talks about brokers tracking trades in smaller-than-100-share units as a reliable indicator of what to bet against. They're all looking for the right counterparty, ideally trying to find someone who is desperate to buy GameStop at $400, a trade they'll happily make.
GameStop, too, is on the hunt for counterparties; it's apparently trying to get into the business of trading rare, but not too rare, Pokémon cards. There are some reasons this makes sense, like the overlap between different gamers and GameStop's own model of accepting trade-ins for store credit. What they're really trying to do is provide a particular kind of liquidity to a specific kind of counterparty: someone who's selling some collectibles, but in an amount that's small enough that they might well spend the windfall on some new products and also that's not so large that there's an incentive to rip GameStop off with a forgery.
Outsourcing
The bull case for consulting firms is that no company has a comparative advantage at everything, and sometimes it's advantageous to lease that advantage from McKinsey or BCG. The bear case is that consulting firms are in the business of spending an institution's money to give leadership in that institution legitimacy to do whatever it is that they'd planned to do all along. And somewhere in the middle of that is this wonderful story of Saudi Arabia's intense and lucrative relationship with the biggest consulting firms, to whom they've outsourced many of the details of the Saudi government's effort to transition their economy away from depending on oil ($, FT). Petrostates have a weird set of comparative advantages relative to other governments; they don't have to be great at economic management, beyond not killing their only competitive industry like Venezuela did. That risk aversion extends to being cautious about non-oil industries: since oil exports have such a distorting effect on trade balances, and on where talented people go, these countries' efforts to diversify often turn into an expensive grift. Especially in cases where there are dense social networks and limited state capacity, industrial policy in commodity exporters tends to be a nexus for bribes. Which might, in the end, be the best defense of using the big consulting companies: they're risk-averse about outright corruption, and will probably produce some nice operating models and rules that make other kinds of bribes and kickbacks harder to get away with. Perhaps some of the money they're getting is excessive, but at least in this case the government knows where it's going and what they're getting. (And, better yet, the money that gets wasted goes outside the country, where it can't fund a competing power base to the House of Saud.)
Multicloud
620,000 Australian investors spent a week being unable to access their accounts because—yes, this buries the lede a bit—Google Cloud accidentally deleted their account and their backup account in a different region. The company had a third backup with another cloud provider, so the data's not lost forever. Annoyingly, these are always a matter of probability. It's unclear exactly what happened, but it does seem to be a unique case, if only because GCP has been around since 2008 and this appears to be the first such incident of this scale. There are always mitigation options, but they involve tradeoffs: if Google quietly backs up customer information in case they or a customer erroneously delete it, they also make it harder for customers to get rid of the data they actually want to get rid of, in addition to taking up lots of extra space. Meanwhile, a system of multiple backups across different cloud providers has its own redundancy cost, and is also the kind of process that will get run more often than it's audited—second only to sad stories about deleting everything and not having backups are stories about deleting everything, restoring from backups, and only then discovering that the backup process has been broken for months and the data really was gone.
(Disclosure: Long AMZN and MSFT, one of which was very likely to be the unnamed third backup provider.)
Can Frackers Collude?
One of the world's most popular conspiracy theories is that companies constantly collude to raise prices for consumers and lower wages for workers. One thing that helps this theory out is that it is always, to some degree, true: saying that a company exists to maximize profits for shareholders and saying it's a conspiracy to supress wages and maximize prices is a distinction in terminology, not in underlying claims.
What matters is whether or not they're able to pull this off. Oil companies, particularly frackers, have recently faced lawsuits claiming that they coordinated to reduce production ($, FT), which raises prices and has the handy side effect of lowering workers' wages. (In a cyclical industry where labor is a complement to sometimes-idle capital equipment, there are always too many workers or far too few.) Shale companies used to be notorious for consuming capital upfront and promising that they'd be more efficient later on, but since the harrowing 2020 experience of negative oil prices, they've actually been quite disciplined, have raised their hurdle rates on investments, and are returning cash to shareholders. One question is: how well can they really coordinate? Oil is a fragmented business, where the biggest US companies have single-digit global market share. It might be possible to get them on the same page, but if oil is artificially expensive and rigs are artificially cheap, both because a group of companies have directly or indirectly agreed to hold back production, that presents an enormous opportunity to anyone who doesn't care to play along. And energy has, at least historically, been full of people with exactly that attitude. It might still make sense to legally discourage companies from trying to withhold production, but if so, that's because of the chilling effect it will have on industries where this can be pulled off.
Infrastructure
For a long time, a key element of China's economic model was that a skimpy welfare state forced workers to save, and those savings got channeled into badly-needed infrastructure. China has since gone well past the point where additional roads, railroads, and ports are automatically high-ROI, and some of their showier projects have ended up being the most expensive: [China's high-speed rail network has been raising prices and exploring shorter routes with slower trains, while other state-backed entities are also raising their prices. When there was an unlimited supply of capital for this kind of thing, it made sense to run some of these businesses at low profitability to benefit from spillover effects, but if enough of China's capital stock is run that way, there just isn't enough of the rest of the economy to benefit from whatever those spillovers turn out to be.
China's capital allocation model is far from perfect, but "perfect" does not exist, and the US model has flaws of its own. By what is no doubt popular demand: RoaringKitty, originator of the GameStop short squeeze and patron saint of meme stocks, tweeted for the first time in almost three years. It's an ambiguous meme, but GameStop traded up 30%+ pre-open and is currently up 110% from yesterday's close. It's not a great bet that the meme stock boom will return to its 2021 levels, but it's a good reminder that shorting is a risky proposition, and that some assets are more sensitive to ambiguity than certainty.
|
