Friday, May 04, 2007

Re-Testing BOB

We've been discussing BOB a lot on this site and on Bullish Jim's and Jim has even provided some modifications which we've adopted.

At any rate every so often I re-test my filters against the most recent period and compare them to the previous period tested. This is done to ensure that I’m using the most efficient filters in my toolbox for the current market conditions. Here are the results of that re-testing for the stable of BOB’s we have managed to collect. As you will see and what I often find is that last month's darling is this month's dog.


From a reward/risk standpoint BOB v32x with 1.89 is far and away the best filter for use in the current time period. Here is that code.

show stocks where close is between 15 and 35
and average volume(90) > 500000
and close 2 days ago < ema(8) 2 days ago
and close 1 day ago < open 1 day ago
and close 1 day ago is less than 1% < open 1 day ago
and close 2 days ago is less than .5% < open 2 days ago
and low 1 day ago < low 2 days ago
and volume 1 day ago is more than 20% > volume 2 days ago
and close > open
and low > low 1 day ago
and close 2 days ago < ema(8)
and close 2 days ago < open 2 days ago


You should also note in your trading notebook that a change of date can cause filter results to vary significantly - the important thing to take away from that is if you are using some ready-made screen provided by one or another of the various sites - are you sure it is still working?

You see as much as we want to believe that CANSLIM or the "way of the turtle" or even that "Thursday's have the highest percentage of gap fills" are the solutions to all of our problems because learned books have been written about them - the fact is - nothing works all the time.

As the Gipper said - Trust - but verify.

10 comments:

Anonymous said...

Marlyn, I'm still hanging with you and trying to make this filter work on Telechart. I've modified my criteria to match what you have shown and today I still get 0 matches. Was wondering if your scan showed any matches today? Also, was it intentional to have two lines showing "close 2 days ago < ema(8)" One stipulates the ema(8) two days ago, the other does not. Thanks.

Marlyn Trades said...

Re "close 2 days ago" - my error - delete the second and leave the first.

The last time this filter output anything was, believe it or not, 2 days ago on May 1st - WPI and NSR and both are up 4 and 3.67% respectively.

It isn't prolific - just efficient.

Marlyn Trades said...

Please let me know what you find.

Thanks

Jeff said...

Marlyn- I have to note that the "way of the turtle," followed correctly, has the weight of decades of data behind it. You are absolutely right that it does not work all the time, but that is where research into avg. length of drawdown and avg. drawdown come into play. Knowing how long on avg. it quits working and how severely it impacts the equity curve is key, as typically what happens is traders quit using a filter just as soon as it is getting ready to work again.

The problems, as I see them (and I am just scratching the surface here) with Stockfetcher is that it does not allow adequate time samples for backtesting. These filters do not have decades of data to test. What may be happening is that we switch out of a filter that is not currently working just as it gets ready to work again. We cannot know for sure what the pattern is without more data.

What you are doing is very valuable-testing different start and end dates- as one good or bad trade near the end or the beginning of the sample period can seriously skew the results. "Way of the Turtle" discusses another measure which does not overweight a good or bad trade at the end or beginning of the sample- I think it is called the CAGR. I could be wrong, and I am at work, without my book.

I guess my point is that this is awesome work you all are doing with Stockfetcher, but these strategies need more analysis with more powerful tools so that we can begin to say with some level of certainty what the optimum parameters are.

I hope I do not come across as raining on your parade, as that is not my intent, at all. I am terribly excited by all this and just want to take it to the next logical step.

Marlyn Trades said...

Not a problem - personally I think the BOB is a valid approach - you stick your stop under candle 2 and you win or you lose. If memory serves and it has been many years so forgive me if I miss it a bit but the "way of the turtle" was not so much about winning as how and when to take a loss. I believe it was take losses quick and let your winners run - and if that is their way - it is mine too.

And should be everyone's who trades.

Please feel free to stick your points in whenever you want - I run an open site here.

L.J. said...

I've been using two years of data at a time on Stockfetcher and I'm of the opinion that it is a sufficient amount of data. Keep in mind there is nothing preventing you from testing 2001 through 2002 and then 2003 through 2004 and then 2005 through 2006. And then test smaller periods within these larger two year periods to see how the screen did during times when the market was rising versus times when the market was falling.

I love data as much as the next guy but I believe that being able to test something over decades at a time can be overrated. What you end up with are average return stats over a long period of time but with big standard deviations within smaller time periods. Unless you plan to implement a system and stick with it for ten years come hell or high water, you're not ever going to give it a chance to "average out".

All that said, I AM a big fan of having a long term approach to short term investing. Anyone who is prudent and realistic is just trying to regularly beat the benchmarks fractionally. To that end I personally think that recent years' data is more useful in doing that than much older data. Just my opinion.

Woodshedder- I'm thrilled that you're getting involved in this discussion. Marlyn and I have developed a very collaborative approach to analyzing things and I think you have a lot to add to the discussion.

Jeff said...

Marlyn-the book Way of the Turtle. While it does detail how the Turtle Traders made money, it is more about how to develop a robust trading system, regardless of whether it is short or long term in nature. There are measures referenced in the book that I feel strongly will help you and Jim develop robust systems. I don't mean to sell your knowledge short. Its just that the book contained measures that I have never seen referenced in the trading books that typically circulate. I will extend the offer to you that if you buy the book and do not find it valuable that I will buy it from you.

I'm getting ready to host a huge party, so this likely will be my last visit to the nets til Sunday. However, at this point I'd rather scratch the party and continue system development and our conversations. So it goes.

Anonymous said...

Marlyn, Just got back to my computer today and ran the scan with today's market results. Still no hits (on any of the BOB versions). I'll keep trying and will let you know if anything turns up. Like you, I'm not looking for "a prolific filter-just efficient". Like your approach and will continue to try to apply your methods. Thanks.

Marlyn Trades said...

den0 - BOB v1x produced TRID and v32x produced WOR off today's finals. I don't particularly care for WOR but I like TRID - here is that code -

show stocks where close is between 15 and 35
and average volume(90) > 500000
and ema(8) < ema(21)
and close 2 days ago < ema(8)
and close 2 days ago < open 2 days ago
and close 1 day ago < open 1 day ago
and low 1 day ago < low 2 days ago
and volume 1 day ago is more than 20% > volume 2 days ago
and close > open
and low > low 1 day ago

It is the origin BOB

Spider said...

I was wondering if anyone noticed if any of the BOB filters work better on a weekly, daily or intra-day timeframe?

Thanks.