Closing the feedback loop in auction market simulations

Why the first auction stayed incomplete

The earlier random-number market model let volume move price through a simulated specialist, but price did not feed back into volume until a price-responsive trader object was added.

Editorial reading: a Monte Carlo simulation of an auction is incomplete if it can print a new price and nobody is allowed to change size because of that print. The missing object is not a richer chart. It is a participant whose next order depends on the last quote.

How the trader was allowed to act

The simulated trader used only a day's high, low, and close, then acted in time for the next opening rather than as an intraday participant.

If a closing quote signaled a buy while the trader was short, the model first tried to cover the entire position at the next opening and waited for another close before adding a new long.

A specialist-limit market order was always filled, though not always at the current price, while the trader's order stayed within a tick or two of target but could be only partly filled.

A 20-percent band still changed the auction

The first trader sold or shorted when price was more than 20 percent above a base and bought or covered when price was 20 percent below that base.

The trader began able to buy 100 lots, or 10,000 shares, against a 12.50 base priced in eighths, for 125,000 of starting capital. That size was about 60 percent of that simulated stock's normal daily activity and twice the largest random order.

In the 20-percent countertrend run the trader placed nine orders, finished short 17,500 shares at 16.125, and made just under 100 percent on capital. The chart became less volatile, and the specialist's profit rose by almost 30 percent versus the specialist-only run.

A 10-percent band produced more trades and about 20,000 less trader profit. A 12,500-capital trader earned a 238 percent return, and a 250,000-capital trader earned 136 percent, versus 95 percent for the 125,000 account.

The same moving-average crossover under two fill rules

Under assumed ideal fills, 0.80 and 0.20 exponential-average constants produced 12.875 points in 27 reversals. A 500-share trader using those moving-average crossover constants earned only 5.6 points in 23 paired trades.

That small flow still left the specialist about 38.4 percent, or 31,000, richer.

Editorial reading: the chart condition can stay the same while the auction changes the result. Points counted under assumed ideal fills are not the same object as points earned by an order that may be only partly filled.

On-balance volume as a map of inventory

Daily on-balance volume, shares bought minus shares sold, tracked the specialist's inventory. Negative on-balance volume meant the specialist was long, and a late cluster of buy orders let that inventory be reduced.

Editorial reading: once a price-responsive trader is inside the loop, on-balance volume is a way to watch who is left holding the residual, not a separate signal sitting beside the chart.