May 14, 2012

The idea behind this research note was that it might be interesting to show the progression one might take to improve the performance of a particular script. But then, would that not be like over-optimizing or over-fitting an existing trading strategy?

I think it would be over-fitting only if the results improved just a little. But what if the results were greatly improved?

Would it still be considered as over-fitting, over-optimizing or maybe more preferably simply a better trading methodology? If it was the case, then the improvements would have to be across the board in the sense that it should apply over the entire trading history and to most of the selected stocks. It would need to improve average performance at the portfolio level.

To show this point, I've opted to use the Ichimoku Kinko Yho trading script on the old Wealth-Lab 4 website. The reason for selecting this script is simply that it is part of market folklore and almost a legend from prior to WWII onward. Should such a trading methodology prevail, a simple backtest should be sufficient to show its merits. It would not need to have a superior stock selection process, only that a sufficient number of stocks should be selected to have a representative sample. In that department, my SD1 set of 43 stocks (which was set in April 2011) should prove a sufficient challenge to the Ichimoku Kinko trading script since the script has not seen any of the data that will be tested having been developed in 2003. A kind of walk-forward, and out-of-sample, test. This will also provide a base for comparison with other tests using the same dataset.

For those wishing to test the original script, follow the link, enter the ticker you wish to test and press: execute script. Note: the old Wealth-Lab 4 website has been taken down and replaced with new software not compatible with the old.

Updated May 8, 2013: the Wealth-Lab 4 website has been shut down without notice.

To start the process, I have done the test based on the original Ichimoku Kinko Yho script. The preliminary results show that overall performance is higher than the Buy & Hold, but only by a slight margin; nothing extraordinary, and if you look at the data closely you realize that only 3 stocks out of the 43 account for some 83% of the total performance. Removing the highest performer will have for result to underperform the Buy & Hold.

Ichimoku Kinko Yho – Original script as is: May 12, 2012.

Ichimoku Kinko Yho Test

(click to enlarge)

Some observations on the test results. The script has only 35% of profitable trades, losing trades account for 65% of the total 3,268 trades taken. These are not very good numbers, some 16 of the 43 stocks showed losses over the 1,500-day test while the Buy & Hold only showed 6 with losses. My main point would be that someone would have had to work for some 6 years to obtain those results. I don't think it was worth it. This is like some of the other original scripts that have been tested using their native versions. I think the problem lies with the trading philosophy on which these scripts were based.

I will attempt to improve performance at the portfolio level. Present some of the steps taken to raise generated profits over the 43 stocks in the portfolio over the 1,500 trading day backtest. This way, progressively, performance should rise if and only if, the trading procedures adopted are beneficial at the portfolio level and not just for a single stock.

I have not tried any improvements yet, but I am confident that I will be able to improve this script by changing its trading philosophy and the way it sees its trading environment. I will try to describe the approach taken in my next note.

Meanwhile, here are all the charts generated with summary results based on the original Ichimoku Kinko Yho script.

(click to enlarge)

AAPL ADM AGQ AMZN
BHH BIDU CCK CF
CMG CRDN CSX DBS
DDS DIT ERX GLD
GTLS IBM IMAX IPGP
JNPR NFLX NTES PAAS
PANL PCLN SCCO SINA
SLV SLW SOHU SRZ
TBL TDSC TRMB TRN
TSCO TZOO ULTA URI
WLK WTW XOP  

(click to enlarge)

... to be continued...

Created... May 14, 2012,    © Guy R. Fleury. All rights reserved