Interpreting Nick Rowe's good post

He reviews and expands a bit on Uncle Milt's thermostat analogy.  Economists should read that post.

He starts with a control system, a device watching on variable and  using measurements to set a control that brings the measured variable back to its setting, a thermostat.  His point is that the changes in the setting of the control will inevitably represent forecast errors of the control system.  In any system this will be true, and these changes are called the residual, the amount of 'acceptable' error remaining in any designed control system.

In an economic setting all plans will reveal forecast errors as they are implemented and the observed errors undergo continual correction leaving a residual of noise over time. But here is the catch.  In an economy the measured variable will invariably have a noise term created by the prior delivery system, itself a controlled plan.  So, absent adjustment, errors generally cascade making specialization impossible. 

How does the economy solve the specialization problem so that the pencil maker can manufacture pencils with the same planning error as the carbon graphite manufacturer and the paint manufacturer and the wood manufacturer?  Simple, each step in the manufacture process is set to be equally imprecise, all parties agree to generate the same control noise power. So the final manufacturer of pencils can plan, apriori, for the known noise in its input deliveries.  Thus the pencil manufacturer can keep input inventories high enough to compensate for known delivery noise of the inputs.  The systematic end result is that all inventories in the economy converge to a known variance/mean level.  The mean is the plan, the variance is the apriori agreement. The implied interpretation is that all agents agree on the acceptable noise level for inventories.

How does the original control system work such that it insures a fixed residual power?  When it notices that the noise level is low,  the controller will wait a bit longer before updating the variable, it lets the input variable build up a bit before the adjustment.  It manages inventory noise by adjusting the sample period.  So, the controller is now competing for spectral space on a generalized yield curve.