Monday, June 1, 2009

Question: What Does The Bullish Signal From The Coppock Guide Mean? Answer: Absolutely Nothing!

I am seeing several articles in the main stream press and blogosphere regarding the "bullish signal" given by a technical indicator known as the Coppock Guide or Coppock Curve. As we can see in figure 1, a monthly chart of the Dow Jones Industrial Average (symbol: $DJIA), the Coppock Guide has turned up from a very low level. Thus according to the above referenced sources, this indicates a new bull market.

I am not so sure about that interpretation.

Figure 1. $DJIA/ monthly

Let's take a look at the indicator which was developed by E.S.C. Coppock and presented in Barron's in 1962 as a very long term buying guide. Coppock advised buying stocks when the indicator was below the zero and then turned up. The formula that I use to calculate the Coppock guide is:

1) use monthly data
2) use closing prices
3) calculation A: 14 month rate of change
4) calculation B: 11 month rate of change
5) summation of: (calculation A + calculation B)
6) take the 10 month weighted moving of the value in line 5

In TradeStation Easy Language code this looks like:

(WAverage((RateofChange(C,14) + RateofChange(C,11)),10))

My Coppock indicator, as presented in figure 1, is then wrapped in trading bands with a 36 bar look back period (or 3 year) to assess for extremes in the data.

In the one article that analyzes the current buy signal the author writes:

"Valid signals are those that turn up from under the zero line. And historically, the deeper the level at which the signal arrives, the more strength the following bull market has. This most recent signal is coming from a deeply oversold level - the most since 1938 (-417 to -400) and even further, 1932 (-643 to -616)."

This is a true statement. However, in the article the author is utilizing the S&P500 Index, which was not in existence until 1957, so one must assume that the data from 1920 to 1957 is the synthetic index that linked the S&P90 (pre 1957) with the S&P500 (post 1957).

More importantly, if we use a differernt (but similar) data set, like the Dow Jones Industrial Average, we get very different results. For example, in 1931 there were two deeply oversold signals (using DJIA data), when the Coppock indicator value was actually at or below the current and 1938 levels of the indicator. The first turn up of the Coppock Curve was in February, 1931, and the Dow closed at 190.30. The second turn up of the Coppock Curve was in August, 1931 with the Dow at 139.40. In both cases, the indicator turned down 1 month later, and the ultimate low was at Dow 40.60. Ouch!

Furthermore, the deeper the oversold level doesn't necessarily equate to a strong bull market once the indicator turns. Following the 1938 signal, the Dow only went about 15% higher before rolling over.

The author does acknowledge that "the Coppock Curve has given its share of false signals", but I am not sure what he means when he states that "we haven’t seen any (false signals) occur when the metric has curled up from such a deeply negative level." I know I have only shown you these two very, very oversold signals from the Dow in 1931, but very, very oversold didn't work for the Nikkei in the 1990's or for gold in the 1980's and 1990's either.

I have "sliced and diced" the Coppock Curve many ways and found that it does have some merit of identifying trend momentum. For example, if you buy the DJIA when the Coppock Guide is rising and sell when it is falling you get an equity curve that looks like figure 2.

Figure 2. Equity Curve
Since 1924, such a strategy yielded 6830 DJIA points versus buy and hold of 8600 DJIA points. There would have been 72 trades of which 50% were profitable; your time in the market was 47%. The strategy draw down is about 40%, which is not much of an improvement over buy and hold. From this study and by following the Coppock Guide, you can make 80% of buy and hold with approximately 50% market exposure, but it doesn't improve draw down over buy and hold.

My impression is that the Coppock indicator functions like most oscillators. They are great in a range bound market (like the 1960's); I don't believe they are very useful in an oversold or overbought market. Across multiple markets and in the Dow, there appears to be nothing unique about the Coppock Curve to indicate the onset of a new bull market.

6 comments:

biscosc said...

I'm speechless. Fantastic analysis!

Jérôme said...

As usual... amazing

Anonymous said...

You mentioned the failures prior to 1950, but not the excellent buy signals, 1904 and 1924 among them. Also, its most recent record since being published in Barron's is almost impeccable if being analyzed on the time scale for which it was meant by the creator.

Guy M. Lerner said...

Anonymous: see the articles by Hulbert...if we follow the rules for the indicator that Mr. Coppock set forth -buy when the indicator turns up in negative territory--there would have been 3 buy signals from 2001 to 2002 using the Dow.

Oscillators work great when they do; but when you take the enormity of the data (in this case the Dow) or apply across other markets (to gold and nikkei) the indicator is fair at best (especially when you take all signals)

Trendsurfer said...

I agree with most of the article - with one strong caveat. Normally I dismiss oscillators altogether, but they're generally a bit better than usual in oversold conditions on stock indexes for two reasons: (a)The indexes are biased long, and (b) stock indexes are mean reverting by nature.

The coppock is no different (and no more magical) than other smoothed oscillators. You can get better (or worse) results by changing the moving average lengths, and you can get similar and often better results from using a long-term RSI, Stochastic, etc.

Now, if only someone would develop an indicator that was valid for deeply overbought conditions! God knows I've tried.

Guy M. Lerner said...

Trendsurfer:

thanks for the input; I always tinker and try to develop an indicator for deeply oversold conditions as well; some of the things that I have looked at are:

1) # of divergences over a period of time occurring at a bottom

2) the distance from a pivot point high both in time and distance

all this is stuff you won't find anywhere else; nothing is completely perfect but a lot of these elements have gone into my "next big thing" indicator which isn't a timing indicator as so much as an indicator that assesses the likelihood of a trend reversal