I think the best metaphor for the robo traders is the precision metaphor. Given a set of past prices, the robo trader determines the precision of that set. I mean precision in the sense that an eight bit number describes the trades with accuracy. Then the robo trader bumps its trade precision to nine or ten bits and trades in the margin between trades. The idea is that the robo trader never gets caught, it is always well within the error band of the market.
Now, this works, but the computation is simple and all traders should use the robo trade method. If you want to exercise a position, then your robo trader should first measure the precision of the market. Make your bid in the same precision as the other bids, open or closed. So, calculate your portfolio adjustment at home to infinite precision. Then when trading, round all bids you make to twice the precision of the market. (twice so you operate at Nyquist)
A simple procedure is to take the last 400 trades or so and run the Huffman encoder, measuring the average symbol size. If you are making a trade in a tech company, then go ahead and bundle closely related stocks and compute a combined precision. The effect is to constantly quash the robo traders trying to get higher precision, the market converges to a typical precision.
Looking backward at past trades, the encoder will miss the Nyquist rate changes and aggregate them, leaving the original precision. Note, this is an aggregate measure, as opposed to constructing the optimal yield curve under perturbation. Looking at structure, we are looking for aggregates that appear offset from the optimal Fibonacci set, which we can fit to the Huffman codes coming out.
All trade software should automatically be measuring precision. If yoru software company doesn't have it, e mail me and I will get you a supplier.
No comments:
Post a Comment