A Price Shock Is Not a Preference

Between February and early April, WTI went from $59.07 to $112.56 a barrel — a 90.55% run in ten weeks. That is the kind of thing usually read as a macro event, but it is also, incidentally, one of the largest natural experiments anybody will get on how cities move. A new paper treats it that way, running a hierarchical panel regression discontinuity over roughly 1.7 trillion point-of-interest visits — about a trillion across 80,000 US census block groups, 650 billion across 40,000 Chinese jiedao. The headline contrast looks like a story about how much people care: US mobility range fell 19.0%, China’s fell 2.33%. Even after correcting for the fact that the shock landed harder at the pump in the US (retail rose 52.56% there against 32.7% in China — pass-through of 0.58 versus 0.36), a 5.1x gap survives. The placebo, the same calendar weeks a year earlier, is essentially flat in both countries. Something real happened. The temptation is to call it a difference in preference.

It isn’t, and the paper’s best move is showing why. Contraction was near-universal — 73.5% of US neighbourhoods and 72.1% of Chinese ones cut their range — but a cut means two opposite things depending on who makes it. Split neighbourhoods by whether they could reorganize, and you get an absorbed group (paid the higher price, kept moving) and a constrained group (locked into trips it had no way to restructure). The mixes are close to inverted: 33.4% absorbed and 24.1% constrained in the US, against 11.0% and 37.2% in China. In the Chinese constrained group, 82.7% of neighbourhoods contracted, and they contracted by 3.63% against 0.86% among the absorbed — more than four times as much. So the same observed number, “range fell,” encodes a household that shrugged and one that was cornered, and averaging them produces an aggregate elasticity (−0.50 US, −0.08 China here) that describes neither. It’s the forced car ownership problem generalized: what looks like revealed preference is frequently revealed absence of alternatives.

The finding I did not expect is in the mode-shift decomposition. If fuel gets expensive, the intuitive substitution is car → transit. In the Chinese data it went the other way: driving share fell 0.89 percentage points, subway 0.83 — and bus fell 2.65, the largest drop of anything. Walking and cycling absorbed it, gaining 1.23. The arithmetic is embarrassingly simple once stated: the marginal fuel cost of a one-to-three-kilometre drive is somewhere around ¥0.6–1.9, and a bus fare is ¥2. For short trips, the bus was never the cheap option; it was the option you took for comfort, and comfort is the first thing a squeezed household drops. Which means the mode people actually abandoned was the one policy usually treats as the refuge. Roughly 1.64 of China’s 2.33 point decline is this compositional reshuffling rather than trips foregone at all — same destinations, worse means of getting there, and nothing in a trip-count statistic that would tell you so.

I want to be careful about how far this travels. Mobility range — distance to the places you visit — is not gasoline consumption, so the −0.50 here is not the same object as the classic short-run gasoline elasticities (Hughes, Knittel and Sperling put those at −0.034 to −0.077 for 2001–06 against −0.21 to −0.34 for 1975–80), and the paper says so rather than glossing it. But the structural claim holds independent of the units. A price shock is not an opinion poll. It is a stress test, and what it measures is slack: how much room a household has to rearrange its life when the cost of one input moves. Two places can post identical contractions and be in opposite welfare states, and a single elasticity averages the comfortable and the trapped into one number that then gets used to calibrate policy for both. My bet is that within a few years, fuel and carbon price evaluations stop reporting one coefficient and start reporting two — magnitude and mechanism — because for the first time the data exists at a resolution fine enough to separate them.