Iterative calculation

It does nothing clever, which is the useful thing to know about it. The workbook calculates the loop, takes the result, feeds it back in as the new starting value, and calculates again. It stops when the change between passes falls below a set threshold or when a set number of passes has run, and both of those are settings you can see and adjust.

Convergence is usually fast, because the relationship between interest and the balance it is charged on is close to linear over any sensible range. Each pass moves less than the one before, and three or four passes typically settle the answer to the precision anyone cares about.

The setting applies to the whole workbook rather than to one cell, and that is the risk. With it on, a genuine mistake that happens to be circular resolves to a stable looking number instead of raising the warning that would have prompted you to investigate. It buys convenience by removing the alarm.

It also interacts badly with sensitivity tables, which recalculate the entire workbook for every combination in the grid. On a large circular model that can be slow enough to be unusable, which is why some teams run sensitivities off a version with the loop deliberately broken.