Roger Farmer is a stock picker

Beyer and Farmer (2007) find evidence that the money interest rate is cointegrated with the unemployment rate and (Farmer 2012, 2015) shows that the stock market is cointegrated with, and Granger causes, the unemployment rate. 

I saw his chart, and corporations do indeed provide the hiring/firing impulse.  Economists acknowledge this publicly when they record mass layoffs, generally from big corporations (and big government).

The corporation is a standard measure, the SEC regulates the spread sheet layout so the various corporations can have computable price/earnings ratios.  Corporate spreadsheets are indeed good instrument variables for unemployment.

But Roger still needs to select an index, and keep his index updated.  His theory only works to the extent his stock index is a representative sample of  supply/demand balances across the monetary zone.  So in his model he has the implied pit boss, the market maker. That market maker has a market pricing error account it maintains and publishes. That account is Roger's third color he introduces in his ISLM chart.  That third color is managed inside the large primary dealers, where wealth make believes about the market.

Still a bit incomplete, it does not capture the complete structure of chaotic budgeting in government and excludes the cost of picking stock index. To get that, the math needs to extract the structure in the stock market maker, and abstract that back to S/L only, that compression leave the bit error a small random white noise. That is the whole idea, completely simplify it down to the two color.

The essential math simply has deposits arriving asynchronously and loans arriving asynchronously.  The structure is discovered in run time as these flows demonstrate a willingness to wait in line in the mathematically defined free trade pits.  My model says the agents have to pick the stocks, currency banker is not doing it.  Government expresses its willingness to wait in line by government agency, not the aggregate. The economists measure the over all accuracy (about 3% per year), split according to a sparse spectrum and reconstructs how they all must have acted with their free traded S/L in the recent past.

But we need not go through all that, we can use technology to compress the currency bankers to its theoretical form in run time, auto traded pits. Just write a software version of Roger done right, and pay the Coasian fee for right to coin.   Just make sure to get a representative sample of accounts in the Fed pits, especially large government and pension funds.  Understand risk equalized, and congestion priced entry and exit works.

Dynamics

The market maker notices a shift in S/L. The market maker applies an interest rate swap such that S/L and delta S/L minimize the bit error.  Within the bounds of the bit error contract, the market maker has commutative property over the bid/ask. This is all fundamental two color pit.

Pits become mathematically tractable only when the transaction costs are trivial, they are not counted, bankers make their money on services, S/L is a not profit, an open source download.  Hence, the assumption of auto pricing is the assumption that bots are always fast enough to keep up with humans. The miracle here, that it is actually fact, Moore's law made it so.

Simple example

A sudden surge in borrowing by Treasury, maybe it is due to a pension stampede, short term.  The Fed has no clue, it has no mechanism to deal with that account. But the SS agency gets first word, and it would like to shift S/L based on its peculiar and accurate knowledge.  It cannot.  The Congressional insurance function is sticky.  Under the new system, large public organizations can make their anticipated shift, adapt and operate under a smaller interest charge regime. Congress gets an accurate velocity equation in return.

Central banking design is about variance budgeting.  If you use gold, pit operations, shoes, or Congress, the gains and benefits can be separated out into variance allocations, which sectors suffer what uncertainty. Then predict a stable Coasian result once the NPV can be computed. Pay the one time cost, and move on; always accounting for the fact someone, just like you, will do it again.