Why a lagging statistic invites the wrong correction

Stock reports arrive six weeks late. The shortage is visible in the report, so orders are doubled; by the time the order lands, demand has normalised and the warehouse is full. Next quarter the report shows a glut, and orders are cut into a recovery.

What everyone sees

A measurement is trusted as a description of now, so a strong response to a clear signal looks like competence. The signal is a description of then. When the delay between the state and its measurement is comparable to the time the system takes to change, a proportional response is systematically mistimed.

What is actually happening

Sterman ran a controlled distribution experiment and found participants generated large oscillations even with stable end demand, because they underweighted the delay between ordering and delivery and over-corrected against the information they could see — the instability produced by the decision rule rather than by the environment. Lee, Padmanabhan and Whang identified the industrial version, showing that order variability amplifies at each step away from the customer, with delayed and batched information among the principal causes. Neither result requires anybody to be careless. The oscillation is generated by responding sensibly to numbers that describe a past state, and the strength of the response determines the size of the swing.

Why it stays hidden

The mistiming hides because each decision is defensible on its own evidence. The report said shortage; ordering more was correct. Reviews examine decisions individually against the data available, which is exactly the frame in which a delay-induced oscillation looks like a series of reasonable calls and bad luck.