The Siege Economy, Revisited — Eight Years On

In 2018 I published a piece for the Atlantic Council's SyriaSource on the siege economy of Eastern Ghouta. The argument, in brief, was that prolonged siege does not produce a frozen market — it produces a highly structured one, with cartels, choke-point rents, and a political economy of crossing-point access that outlasts the siege itself. That framework held up well enough that several humanitarian planners adopted it for the Idlib and Dar'a operating environments that followed.

I am returning to it now, from inside post-2024 Damascus, because two things look different from here that did not look different at distance.

First, the framework underestimated how completely the siege economy's information networks would survive the lifting of the siege. The traders who specialised in knowing which checkpoint was open at 3am did not disappear after 2018; they migrated. The same actors are now structurally embedded in cross-line logistics, NGO procurement, and the small but growing reconstruction supply chain. The original piece treated them as artefacts of the siege. They are better understood as the first generation of operators in whatever comes next.

Second, the framework underestimated the time signature. I described siege rents as a wartime distortion. From here, with eight years of distance, they read as the foundational layer of a multi-decade political economy. The siege ended. The price-setting power did not. That is the kind of error that comes from analysing a system from outside it: you can see the structure but you systematically underweight its persistence.

The diagnostic upgrade I would make today: when you map a wartime market, always ask two questions of every actor. One — what is their position inside the current siege? Two — what is their position inside the post-siege economy that this same set of skills will produce? The second question is the one that survives.

Re-reading the 2018 piece this week, the analysis is still recognisable. The error was not in the model. The error was in assuming the model had an expiry date.