Saturday, April 5, 2008

Does an ADX filter work?

This week I set out to determine if there was a way that an ADX filter could be incorporated into a breakout trading strategy that would increase expectancy in an unambiguous way. It seems logical that a system developed to take advantage of a trending market would benefit enormously from an indicator designed to tell us if we indeed had one on our hands.

I’ve also decided to change the standard entry criteria I use as the base for the tests on this blog to something I am comfortable revealing in its entirety. This will hopefully make everything more transparent for everyone. The goal for this was not to create a tradable entry system as much as a simple, generic breakout entry that would (a) take advantage of our volume test from last week, and (b) be simple enough to layer up and test additional criteria with.

The “standard” entry criteria are as follows:

Average Volume >= 100,000 shares
Today’s Volume >= AverageVolume * 2 (100% increase)
Today’s Close > Yesterday’s Close

Average volume is a 10-day average not including today - Average(Vol,10,1).

This is a simple volume breakout. If these 3 conditions are met, buy $1000 worth of shares on the following open with a market order. In all examples, I used a 14-period ADX as it is generally the “default” on most websites and charting software. There are several common ways of using an ADX filter and I tested five layered over the above entry criteria:

ADX > 20
ADX > 30
ADX > ADX 10 days ago
ADX > ADX 10 days ago * 1.5
ADX > 30 and ADX greater than ADX 10 days ago.

These represent the most common uses for the ADX and the basic concepts behind them is to A) buy if the ADX is above a certain level, B) buy is the ADX is rising, C) buy if there is a strong jump in the ADX, D) buy if the ADX is above a certain level and rising.

I also decided to test these with two exits. Confidence in our volume tests from last week was increased with similar behavior over multiple exit strategies so I decided to continue that theme. One exit was the fixed 8% protective stop with a 20% profit target we have used before. The second exit was a simple 1-month timed exit (21-days). This is an unclamped exit and should be more effective at revealing an increase in the average strength or magnitude of breakouts associated with use of an ADX filter, if they exists.

As usual, I will run our test on three separate portfolios of stocks in the hope that any results we see are confirmed over multiple portfolios as I am looking to add only tools with systemic advantages to my trading toolbox.

Last but not least, there are now three evaluation ratios we will now look at instead of the previous one. They are as follows:

EDR=Expectancy per dollar risked. We’ve covered this plenty already but it is calculated by Average Trade/Average Loss.

EER=Efficient Expectancy Ratio. This is our EDR/Avg. Days in Winning trades. This number is meant to reveal the most efficient use if capital as it reveals expectancy per dollar risk per day in trade.

DDR= Drawdown ratio. This number is calculated by Net Profit/Maximum Drawdown. A higher number is better.

Alright let’s get to it.

8% Protective + 20% Profit Target

21-day Timed Exit


I started this section off by attempting to write a very detailed analysis of what was going on. The fact of the matter is that the results are the results and they are there for everyone to see and analyze for themselves. Anything I write is my interpretation of what I am seeing and others will undoubtedly see something else. Unlike many of the previous tests run on this blog, the advantages here are much more subtle and much less consistent.

My own conclusion is that unfortunately, the ADX is not something I personally feel comfortable using as a filter for significantly improving the reliability or expectancy of a breakout system. There are countless people out there who use the ADX, many of whom probably with some degree of success, but the test for me do not reveal a consistency or magnitude of improvement I’m comfortable with.


cam said...

It's interesting that a sell signal using a random number of days (21) would yield about the same return as using loss/gain percentages as a sell signal, but in much fewer days, at least for the winning trades.

From looking at the data, it seems that the longer the stock is held you can get bigger gains on the winning trades, but then you give back most of these gains when other winning trades turn to losing trades.

bhh said...

Sorry I missed your comment. That is an interesting obervation. Timed stops often end up having the best results but they are impractical to trade without a disaster stop in place and position sizing becomes an issue without a fixed risk amount. I will be looking at implementing some various forms of timed exits in the future, primarily as a way of minimizing average loss.