Does PPP underadjust?
Epistemic status: I am not an economist, but I’m trying to make sense of some confusion here. There’s definitely some spots where I think economists have more relevant things to say than I’m aware of.
A while ago, I wrote this note on GDP metrics, trying to make sense of why GDP (or even PPP) metrics tend to be pretty different from how rich a place actually feels. Today, I’ll try to take another shot at it, this time just via PPP.
The concept of PPP is to try to adjust for the local cost of goods and services. Imagine a country where there’s just one job that everyone can do (say, janitor). It’s going to be a poor, low-GDP place.
Now imagine they get a new kind of job - say, manager. This job pays twice as much, but only the smartest 20% of people can do it. so 20% of people will be managers, the GDP will go up, but so will the price of janitorial work (there’s still just as much demand and 20% less supply), which means the PPP doesn’t go up by as much (someone making $X as a manager in the new country can afford less janitorial services than someone making the same $X as a manager in the old country would have, so he’s not actually as rich). Also, since the smartest 20% of the population aren’t working as janitors anymore, the quality of janitorial work goes down.
Now let’s imagine there’s a third job (engineer). This job pays even better than management, but only 2% of people can do it. So now the smartest 2% are engineers, the next smartest 18% are managers, and the rest are janitors. Again, GDP goes up, PPP goes up less (management is more expensive now too), and the quality of management declines.
So if I understand PPP metrics correctly, they adjust for the increase in price of management and cleaning services in the third country. But there’s two effects they may be dropping: First, the quality decline part mentioned earlier. It’s definitely politically incorrect, probably hard to measure, and I’m not sure economists adjust for it1.
A different way to see this effect: talented people are generally overpaid. There’s a stereotype of a 10x engineer, who’s ten times as productive as his coworkers2. This person never gets paid ten times as much as his coworkers (and often doesn’t get paid more at all). In our earlier example, a typical engineer was in the smartest 10% of managers, so probably he was some kind of 10x manager. Now he’s just a typical engineer. But before he was getting paid typical manager salaries (or maybe a bit more), and now he’s getting paid a typical engineer salary (which is much higher, since it’s full of people like him). Even if an engineer of a given ability level isn’t any more productive than a manager of the same ability level (so in total the country isn’t actually any richer), we’re measuring it as having higher GDP now (since the engineers are getting paid more3.
This is the biggest problem with government jobs, which are often high-impact (government employees manage some massive budgets) out of proportion to how well-paid they are (the chief of the Federal Reserve only makes as much as a typical banking associate). And that’s with jobs that are known to be senior and important - imagine how underpaid that one junior secretary at city hall who actually knows what’s going on with everyone’s planning applications is. If a country’s GDP goes up, all these people have higher-paying jobs to jump to - but since the jobs they’re leaving (presumably to be replaced by less competent people) actually are important, the country could actually end up becoming poorer, even if its GDP looks like it’s going up. This is consistent with how the places that feel richer than the GDP metrics say they should be are usually the ones with competent government.
What are the real world consequences? For one thing, it means we should be a lot more worried about losing a potential conflict between the west and the Iran/China/Russia axis. If we look at the graphs on this page, the west does much better on GDP, somewhat better on PPP, and then about even on manufacturing output. If PPP underadjusts, then the last graph is probably the most correct one - it probably wouldn’t be affected by cost of labour adjustments, and it’s what we would expect the underlying “true measure” to look like if it’s a further directional adjustment from PPP in the GDP→PPP direction.
(Note that this is just in terms of manufacturing - if there’s ever a military conflict, we have to worry about whether China does a better job recruiting talented people to military roles. The US military is a somewhat more prestigious career than other government jobs, so it’s not as certain, but it is a concern.)
On the other side, it could mean we overestimate the convexity of benefits of near-term AI. If AI gradually gets better at replacing more roles, the later roles are probably higher-value and have more productive benefits (and would also free more increasingly talented people up for jobs that are harder for AI but equally hard for humans - although that one only works if we assume human retraining is faster than the pace of AI improvement for those jobs).
Conversely, if a lot of the potential benefits of AI are in management or government roles, and those roles are just too safe from competition to be automated even when they should be, we’re likely to end up leaving a lot of the potential economic benefits of AI unused.
As mentioned at the top, I’m not actually an economist. If someone knows more about this, please say more.
I think this is wrong in a couple of ways - I doubt the competence distribution is bimodal (although it probably is lognormal), and at least some of the productivity gap comes from the fact that once one person takes ownership of a project, other people tend to step back - but there definitely is a productivity curve, and in my experience it’s both sharper and not super well-correlated with actual salary.
This one is very contingent on how exactly GDP is measured, which is another thing I’m a bit confused by. But I assume it mostly assumes people’s value add, for things that are hard to directly measure like software or services, can be figured out by their pay.