Scanning Moving Average Crossover, 11/09/05

The best edges have to be the ones that few, if any, have found - the emerging edges - and I'm spending a fair amount of research time searching for methods of identifying ascendant edges.

Below is an inquiry that uses a scanning algorithm to create a Scanning Moving Average Crossover system. The scanner attempts to identify ascendant frequencies as the current frequency fails.

Develop a methodology for scanning for an emerging edge

Continue working with Control Charts, this time in the form of P&L shutoffs.

Create a profitable, or at least improved, moving average crossover system

A simple long/short, always-in-the-market n-period moving average crossover

If the k-period P&L is negative then search for a new frequency

Frequency search is a modified binary search:
Randomly choose a starting frequency n

If the k-period P&L is < 0 then choose a new 'n' that is either double or half of the current frequency

If the new frequency is above an upper bound or below a lower bound then choose a new random frequency

Results (SP Futures 1997 to present):
The Scanning Moving Average Crossover is consistently profitable (over 95% of trials), but results were statistically insignificant w/ an average z of around 1.0

This compares with a simple moving average with a randomly chosen period which was profitable in about 78% of trials, and an average z of less than one.

Yet another way not to build a moving average crossover system.

The scanning methodology is interesting; I'd appreciate hearing ideas on how to improve it:

Henry Carstens
Vertical Solutions