November 14, 2011

My intent was to use a script that uses a lot of technical indicators, modify its trading philosophy to more reflect my mental view of price movements. I've extensively modified the Hershey Equity Version 5.0 script (over 1,000 lines of code) as mentioned in a previous post. My primary objective was to have the portfolio exceed 100M; a 20-fold appreciation over the testing period. I program mostly blind, seeing only about 220 of the 1,500  trading days of a test.

It often gives an advantage, you can not curve fit what you don't see. It also forces me to have a mental picture of the Schachermayer pay-off matrix Σ(Q.*ΔP) as it evolves.

All the trading procedures are based on technical indicators, and as such, this new script should be considered as purely technical. No fundamental or statistical information is used. There are no random processes as all trading decisions surrogates are based on the value of technical indicators only.

I must say that I greatly exceeded my objectives. I used AAPL as my prototype, (the need to debug), and then ran once the 43 stocks in data set 2 which had not seen the new script; making it an out of sample test. I knew from the very first stock tested that I would exceed my goals by a wide margin. I should have used a less optimistic stock than AAPL in the prototyping and debugging phase.

Well, here are the results:

New Hershey Script

New Hershey Script Table

(click to enlarge)

The bottom line is 550 times the original portfolio value over the 1,500 trading days testing period. This is phenomenal. Almost 215,000 trades. The average holding period for longs is over a year, while stop losses have an average duration of about 9 months. The hit rate is an impressive 81%. The system ends up mostly in cash (over 75% of portfolio value). It means that over 1.5B is sitting idle in the account doing nothing. This suggests that I should improve upon the excess equity utilization rate. This was just the first draft, and there is so much room for improvements considering that it is also scalable. I started wondering how far it could go.

So I for one, must conclude that technical analysis is not dead. It is just another decision surrogate that can be used in designing automated trading systems. This test makes more than an exception, it makes a statement: technical analysis do not necessarily fail. But it goes even further than that. It implies that alpha generation can be exponential which in turn should force the rewriting of some academic papers.

(click to enlarge) 

AAU AKAM ARUN ASYS
ATML BIDU CAM CAT
COOL ETN FFIV FIRE
GMCR HK HNL IDCC
IGTE LTXC LULU MELI
MENT MFL MGH MSN
NDSN PFCB PNRA PTI
QCOR QLTY REDF RVBD
SCSS SF SFLY SHS
SPRD SVVS TLEO TPX
UA UTEK VSEA  

( click to enlarge) 


Modified ... November 14, 2011,    © Guy R. Fleury. All rights reserved.