Skip to content

Quick Take: Relative Strength and Donchian Channel Breakouts

August 15, 2009

Note: Quick Takes will be short articles introducing new ideas or concepts that haven’t been tested. It is designed to help inspire your own personal systems. If you wish to demonstrate the results of Quick Takes ideas, your results will be published on this site subject to verification.

The stock market essentially can be traded in two dimensions: 1) trading an asset in isolation: using technical/fundamental indicators or looking at charts to time buy and sell signals for say the S&P500 2) trading an asset in relation to other assets:  using technical/fundamental indicators or looking at charts to time buy and sell signals across assets. In this case, we are using relative performance as a timing mechanism. This could be a sector vs the S&P500 or treasury bonds versus  REITs.

Most of the work that I have done in my extensive research has been on trading assets in relation to other assets. Putting aside “cointegrated” assets which are more suitable for mean reversion strategies and pair trading, the best way to increase your returns is to look at “relative strength” across assets. Relative strength is simply a measure of how much a given asset is increasing or decreasing in relation to some benchmark. Assuming two assets are not highly cointegrated (that is they do not have a stable mean that acts as a form of gravity) the asset that is increasing the fastest in the longer term (1-12months) should outperform the weaker asset. Think of Relative Strength as a horse race, we would want to bet on the horse that is in the lead. Other things being equal, the closer the race is to finishing, the more accurate it will be to bet on the leading horse. Thus, with relative strength, longer lookbacks on average should be preferable to shorter lookbacks.

Relative strength has been categorized as a “momentum effect” and it is the most robust anomaly  in financial markets since their inception. Even the most cynical finance professor will concede that using momentum is the easiest way to beat the market. Momentum works across and within asset classes (currencies, commodities, fixed income, stocks, and real estate), sectors, and individual stocks. I have tested this effect in numerous ways to rank stocks, looking back all the way to 1950. It remains one of the most powerful factors and a critical ingredient to combine with other factors such as valuations.

Over the past year, most of my focus has been diverted to timing assets in isolation. My research in this area is still in its infancy, but so far it has yielded exceptional results. They key question, is how to combine the best of both worlds– timing an asset that is also expected to be the best performing asset.  My testing indicates that using static 4-week, 12-week or 52-week momentum in conjunction with timing is an improvement, but not a signifcant one. This is because the relative strength between assets does not vary with a stable and predictable frequency. What is therefore needed is a mechanism that is dynamic.

My previous post on using the Percent Exposure Method and Donchian Channel Breakouts https://cssanalytics.wordpress.com/2009/08/10/quick-take-per…channel-method/ ,  inspired me with the solution to this problem.  Why not take the ratio of an asset to the S&P500 (SPY) and use that chart as a basis for timing relative strength? That is, we look at Relative Strength Donchian Channel Breakouts– these are created by taking the log of the ratio between a given asset and the S&P500. You can use the same percent exposure method in that post for relative strength to create a combined exposure model.  The combined model would simply take the average of the recommended percent exposure by each individual model.

To create Relative Strength Donchian Channels you can use  www.stockcharts.com . For example, if we look at XLF versus the SPY we would type this in as XLF:SPY. Now you can select various price channels under the “overlays” section (note this is free). By default, i would select the 20, 50, 100, and 200 day price channels. There are some interesting divergences that you can take advantage of using Donchian Channels instead of convential momentum. In the 1-year chart of XLF:SPY, you can see a new 20-day breakout for XLF in mid March when the rally started. In contrast if we look at XLP:SPY there is 20-day breakdown around the same time. Looking at 52-week relative strength only, we would have chosen XLP instead of XLF. The signal to buy XLF instead of XLP using 52-week relative strength would not have come until months later after the move was finished. Even using shorter time intervals such as 12-weeks or 4-weeks would have have delayed signals. Buying XLP would have lost money relative to the S&P and buying XLF (the banks) would have made a fortune!

The added advantage to using Donchian Channels is that you wouldn’t sell a winner prematurely, versus using pure momentum-based methods. If XLF slowed down for example, you might rotate prematurely into something else using a momentum method. In contrast, using Donchian Channel relative strength, you would not sell until a breakdown occurs. The final advantage is that the first breakouts/breakdowns are most revealing with respect to sector rotation. The financial sector, XLF, broke down in mid 2007–long before the market broke down.

Creating a combined percent exposure mode for ETF sector rotation using Donchian channels would simply take the percent calculation from the method used to time the ETF in isolation, and the percent used to time the ETF  in relation to the S&P500. By allocating fixed slots within your portfolio….ie you will hold say a maximum of 4 ETFs with 25% in each, you would by default  fill your portfolio with the first ETFs to breakout or breakdown. The major advantage of this strategy is that it is low turnover, and is most likely to make correct switching decisions because it is dynamic–and does not use a fixed lookback like conventional momentum methods.

8 Comments leave one →
  1. Jeff permalink
    August 16, 2009 7:11 pm

    As always, excellent work David.

  2. August 17, 2009 10:01 am

    Simple concept, meticulous explanation, profound implications. I’m looking forward to more of your insights.

    • david varadi permalink*
      August 17, 2009 2:09 pm

      i will try to deliver!
      cheers
      dv

  3. eber terandst permalink
    August 17, 2009 3:58 pm

    Nice talk.
    Now, how about posting some trades and see how they work ? ? ?
    eb

    • david varadi permalink*
      August 17, 2009 5:08 pm

      well eber, the purpose of “quick takes” are to inspire new ideas and systems. In my other posts i provide results, because they have been tested both on the computer and in real-life trading. Nonetheless, the tone of your reply with regards to “talk” vs “posting some trades” implies that what i am discussing is mere fluff–not worth even acknowledging.

      Logical concepts and theories inspire the development of good systems that work in real-life. This is also the essence of the scientific method–create a theory and then test you hypothesis. Systems that are created because the “results look great” and a few trades happen to work…….do not tend to last. Often they have been optimized or simply “discovered” through the course of near random testing. I could present dozens of bogus systems that would have readers drooling to put them to the test with real money. Data mining is a dangerous weapon if used incorrectly. So are simplistic judgements based on the success of a few winning trades.

      I have more good theories and ideas than time to test or present on a blog ……hence the purpose of these types of articles. I want to teach people how to think about the markets so they can develop their own systems, rather than simply give them a “working” system with no further explanation that could stop working next month.

      dv

  4. scrilla_gorilla permalink
    August 20, 2009 2:55 am

    This is an extremely interesting post, but how confident are you that this method actually produces positive returns? Just trying out various ETFs vs. SPY on StockCharts, it seems like the behavior of XLF was the exception. Most of the 20d and 40d breakouts I observed signaled a top of sorts, and the net results of the trades after existing at 10d/20d breakdowns were typically even or negative.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: