Saturday, April 15, 2017

For the folks writing price compression engines

Read the last post and bear with me a moment.

If we have two sequences and a black box model, then we are interest in when both sequences match in precision to the model, doing each sample set independently.

Take the loan sequence and the deposit sequence. The red/green is the bit error function.  We Huffman encode the loan sequence until it drives the bit error to some variance equal to the insignificance variance of the Huffman encoder. Do the same for the deposits, it has a difference variance.  So, pay off to the precise at the expense of the imprecise. In other words,there is a solution, by significance, tothe sequence matching problem.  The actual solution is simpler, the two sided matches are small dimensional and a convolution of the two graphs gets you the likely result.  But there is many more than one way to get the fair price. Most of it not very tricky, and all of it fair traded in that all parties have equal access to the trade book. Just set the price engine as interchangeable in the trading pit architecture. The pit boss is really just another trading bot in the pit,  with some contract requirements pinned to it.

No comments: