Modeling the bankers channel, with corrections

There are still a few bugs in this model, so expect me to return now and then.

In this post I try to expand the simple idea of using a Huffman encoder as an entropy tester. 
I want to fill the Huffman encoding tree with actual bankers, and assign dynamics that fit the Shannon channel model.  The I want to talk the model through the events leading up to the crash.  I am assuming the simple model that the bankers receive and process a sequence of bond trade events. The  series of trade events that arrive have some redundancy because of forecasting error.  So bankers reconfigure the encoding tree adaptively to reduce forecasting error.  The actual trades come from market trades that reflect all other dynamics in the economy.

Starting with the channel model I make the following assignments,

channel noise = inflation.
channel capacity = the number of bankers, and is fixed.
The way the Shannon theory reads, the channel has a fixed capacity and high noise reduces the available channel rate.  High inflation implies high noise implies lower channel rate.  The yield is a quantized value measured above the noise level.  The more digits assigned to a particular yield level, the higher the quantization level, the longer the recoded word length.

So we can think of this bankers network as operating a dual generator that would generate a higher entropy encoding of the incoming stream.  The Huffman tree is the output of the system, the banking structure, and really generates no data.  The tree tells us the internal structure of the network that should result from the redundancy detected in the sequence.  The economy derives and publishes the yield curve from the Huffman tree. 

Example.  When the number of home loans drop, and become rare, the Huffman tree will configure a shorter chain of bankers to encode it  into a higher word length, a higher yield.  Hence more bankers are available to process the more frequent sequences which will have longer chains. Both adjustments would obey some hidden law of supply and demand.  Fewer customers for home loans means the queue for home loans has shortened below normal.  So increasing yields, and reducing the number of bankers ends up matching the arrival rate of home loans to bankers. Meanwhile, some of those customers have been pushed to lower yielding more frequent loans, and more bankers balance that side of the tree.   The better insight into the process comes from using queue length rather than supply and demand curves.

What happens when inflation rises?   Bankers want higher yields to get above the noise, that means longer word lengths encoded in the tree.  So bankers increase the rank of the tree, but reduce the number of branches.  We get fewer active terms but with higher yields.  Customers have fewer choices between term and yield.

What happen when some part of the sequence has lower yields, but more frequent yields. Bankers will lengthen that branch of the tree, gaining higher word lengths and reduce the other branches/  Yields rise for the frequent, but lower yields.  Some customers are pushed away from that term and onto adjacent terms.  This is the case of a dip in the yield curve, an inversion, and the Huffman tree will flatten out the dip.  And finally, when the incoming sequence is all one yield at the channel rate, then the bankers will create one long branch of the highest rank they can make,  exposing a vertical yield curve.

Notice that this model always generates something very close to an upward sloping yield curve even though it has no concept of time, it only deals with frequency of arrival.

Does this model have a Taylor rule equivalent?  Yes, look at the Huffman tree above.  As it is organized, we always want to label the farthest node to the right left right Ben.  Thus Ben will always serve the most frequently arriving yields,  he runs the 15 items or less checkout counter.

Anyway, this is the idea, more later,