Wednesday, January 17, 2007

RSI 8 Vs RSI 7 - Shootout at Filter Corral

Stephen asked an interesting question - how much different from 8 20 was 7 20? I didn't know so I went to see. First - here is the filter as it was originally written:

show stocks where close is between 15 and 35
and average volume(90) > 500000
and RSI(8) < 20.00

They don't get much simpler than that.

This filter produced 69 completed trades with 45 winners for a 65% win rate with a 121% ROI. Now the ROI is not that important except for comparison purposes with other filters - you probably won't get that exact value because you will be playing different priced stocks with different amounts of money for different periods of time. What does matter though is the next item of comparison the net over days held.

What I selected was
1 day chg +++ 4 day chg+++10 day chg++++20 day chg+++30 day chg
.66% ++++++++ 1.70% +++++ 2.84% ++++++ 4.69% +++++++ 5.30%

Obviously the longer you hold the better off you are. Then I changed the last line above to

and RSI(7) < 20.00

Ran a back test with all of the same conditions as I had for the above set of results and this is what I came up with:

72 completed trades with 42 winners for 58.33% with a net gain and a 70.89% ROI.

1 day chg +++ 4 day chg+++10 day chg++++20 day chg+++30 day chg
.31% ++++++++ 1.01% +++++ 1.70% ++++++ 2.71% +++++++ 4.49%

What a difference a day makes.

Now it might make a difference if you use 7 periods over another time period rather than the one I selected. It is always dangerous to assume that because a filter tests so well that you have found the Holy Grail - or the Rosetta Stone for all TA. So I tried a different time period.

I switched to 6/01 - 9/28 which overlapped into the first month of the last period.

This time the RSI 8 tested at 76 completed trades with 58% win percentage and a 28% ROI.

The net change over time numbers looked like this:

1 day chg +++ 4 day chg+++10 day chg++++20 day chg+++30 day chg
.36% ++++++++ .62% +++++ .84% ++++++ 4.47% +++++++ 7.48%

Which suggests that the longer you hold the better off you are. Still we are just using these numbers for comparison purposes.

The RSI 7 tested at 78 completed trades with 51% win rate and 13% ROI (which happens to be 2 points less than the SPX ROI for the same period).

But the net change over time tells another story.

1 day chg +++ 4 day chg+++10 day chg++++20 day chg+++30 day chg
.09% ++++++++ .20% +++++ .82% +++++++ 3.90% +++++++ 7.72%

Short term the RSI 8 is a lot better than the RSI 7. As you get out into the longer net change numbers they are about the same - but that is explained by the fact that the period after mid-July was hot for just about anything.

And something even more subtle takes place with this kind of testing - the starting values in the later months of last year are higher - i.e. the dips are higher than they were earlier in the year. So stocks entered in Augues for example will show better appreciation than stocks entered in October even though both periods were characterized by dips.

Which is why when you have someone bragging about their great portfolio returns for the past year you should congratulate them but don't think that they are something special (even if they think it themselves) - everybody had great returns last year - including me. And if your portfolio manager didn't - then you might think about finding another manager.

Just a thought - never advice - I know nothing.


Steven said...

Great Stuff!

Here is another question:

1. Since we are looking at < 20 RSI to buy with RSI(8), how about using an overbought number as the sell instead of a certain number of days.

Like sell on a 70 or 80 RSI cross?

NO DooDahs said...

Have you checked adding a filter for RSI on a longer timeframe? Something like buying a low RSI on a daily scale when the RSI on a weekly scale is "overbought?"

Anonymous said...

Your blog keeps getting better and better! Your older articles are not as good as newer ones you have a lot more creativity and originality now keep it up!


Thats the way to play it.