Saturday, April 3, 2010

Seeking Linearity - Part 1

I have been thinking a lot lately about self-attenuation of the linearity of return series.  As a vehicle for this exploration in this first of a series of posts, I will generally proscribe the formula for a very simple indicator below:
  • A = Average_Price(Days) - Average_Price(Weeks)
  • B =  A/ Last_Price
  • C =  Percent_Rank(B, Years)
The trading rules will start with a "level-one" attenuation using the 200-day moving average as a filter proxy for volatility of the underlying, as is common practice, holding long only when the price is above this average and C is less than 70%, else going to cash.  In contrast, when price is below the 200-day moving average, long positions will be held when C is below 50%, and short when C is above 50%.

Naturally, much smoother indicators and methods exist, but I am choosing this as a starting point because it appears so linear in its own right from the 20,000-foot level -- It's that close up turbulence that will be the focus of future articles in this series, particularly when we see what happens when these returns are compounded.

The hypothetical additive, frictionless return series shown below uses the SPY back through April 1993, and excludes returns on cash.  I highly doubt the proscribed indicator would have performed as well on other indexes, but that isn't the point of this piece.  Also, as a preemptive strike -- with respect -- please don't write asking for the specific parameters.  Everything needed to replicate this very simple model is found above, and I have increasingly come to the conclusion that one cannot "own" an indicator without conducting individual research.


Anonymous said...

I enjoy your blog. Thanks!

Question: Your algorithm calls for going to cash under certain conditions, but I don't see any flat periods in your equity curve.

Am I missing something?
Thanks, Bill

jgpietsch said...

Thanks B. They are there, but are brief and get lost in the seventeen year history. Best, J

Anonymous said...

Hi and Thanks a lot for your very interesting blog !
I have a question about how you calculate A in this. Would that be a difference between an average on Daily Prices and Weekly Prices (taken on a given day), or does that include rolling/moving averages ?
Many Thanks for your insights,