### A New Twist On The Coppock Guide By Jay Kaeppel

Published March 11th, 2010 in TradingYou may be familiar with Jay Kaeppel from his K-ratio which measures the ratio of the price of gold relative to the share price of gold stocks. In a recent piece he uses the Coppock Guide which we’ve looked at there on the blog several times to create a long term timing model for the stock market.

The result is very impressive. In the chart below, the blue line is the equity line for the signals provided by Jay’s own version of the Coppock curve and the red line the Dow Jones Industrial index:

“Jay’s Coppock Model” has two inputs: the Coppock Guide and the Dow Jones (both levels and momentum). More specifically:

I use three measures in one to generate a green, yellow or red light, using end of month Dow readings. One measure uses just the Coppock Guide, another uses the Coppock Guide plus a moving average of monthly Dow closing prices and another uses two moving averages of Dow monthly closing prices (do I know how to have a good time or what?). Each measure is graded as a +1 or a zero at the end of each month. The three are then totaled together to arrive at “Jay’s Coppock Model.”

Measure #1:The first measure simply compares this month’s CG reading with the reading of two months ago. If this month’s reading is greater than the reading two months ago then the JCM gets one point added to it.

Measure #2:Yes this involves a little “double dipping.” For Measure #2, if the CG is above its reading of two months ago AND the Dow is above its 12-month moving average (calculated using the last twelve monthly closes for the Dow) then the JCM gets another point added to it. If the Dow is below its 12-month moving average then Measure #2 is a zero regardless of the action of the Coppock Guide itself.

Measure #3:This requires - you guessed it - a few more calculations.

- A = (Last month’s 5-month exponential moving average of Dow closes * .6666)
- B = (This month’s Dow close * .3333)
- C = (Last month’s 10-month exponential moving average of Dow closes * .82)
- D = (This month’s Dow close * .18)
- E = (A + B) or the new 5-month exponential moving average
- F = (C + D) or the new 10-month exponential moving average
- G = (E - F) or 5-month exponential MA minus 10-month exponential MA
So to put it into plain (albeit admittedly fairly unintelligible) English, Measure #3 goes like this:

- If this month’s value for G is less than zero AND this month’s value for G is below last month’s value for G, then no point is added to the JCM.
- Under any other circumstance - i.e. if the 5-month average minus the 10-month average is positive or if the 5-month average minus the 10-month average at the end of this month is greater than the value for last month (like I said, thank God for the spreadsheet), then Measure #3 adds one point to the JCM.

Based on these 3 inputs, Jay is either long/short or in short term T-Bills. Personally, I prefer the S&P 500 index but it has a much shorter history so I can understand why Jay used the Dow Jones (with a +100 year history). More importantly, Jay provides a model for how a real market master approaches an indicator like the Coppock Guide.

Most will learn about it and take it at face value - checking in once in a while to see where it is. But he uses it as an ingredient for further research. Having just attended Jake Bernstein’s DSI webinar, this is very familiar territory by the way.

You can read Jay’s complete piece here: Forecasting the Stock Market in 5 Minutes a Month

**Enjoyed this? Don't miss the next one, grab the feed or **

Sounds like overoptimization to me.

burt, are you saying that because of the ‘complicated’ formula he’s using to come up with his 3 point system? If so, based on that, you could argue that the Coppock curve itself is also ‘overoptimized’, know what I mean?

I have found that the systems that give the best guidance are those that:

1. Make sense and

2. Work well within a wide range of parameters

That’s why I like the 200-day moving average and the stocks/vix relationship.

The price of gold relative to stocks makes a lot of sense. It gives you a worthwhile indicator regardless of how you manipulate it. I’m not sure what the Coppock curve is, although it seems to be something like the MACD. Am I right? If so, that is a somewhat robust indicator, especially at turning points after large moves. I think it’s a mistake to optimize in effect 5 or 6 parameters. In doing so, the laws of statistics assure that you will get a good backtest, but so what?

For more on this, see my sporadic blog and especially my first post

burt, thanks for sharing your work with us. re 50/200 your simple moving average system, would you have had the fortitude to stay with it for the 9 years it went sideways? that was just before the fireworks came and personally, if I would have been running that system, I would have lost patience.

Mmm … I am a little bit skeptical about the results of the Jay’s Coppock Model.

I have written down the code in Wealth lab Pro (check out here).

Alex, skeptical in the sense that it is curve fitting or that you come up with different results?

The Coppock is a reliable long term indicator, i would not confuse things by introducing some one elses interpretation of this indicator, Jay is a unknown, the formulea looks overly complicated, why complicate things?

Coppock is bullish, the long term trend is up-just leave it at that surely?

jezza, some people just want to guild the lilly

Burt, I would be interested to hear how laws of statistics can insure a positive backtest. An honest question. I don’t see how the outcome of 0/1/2/3 on a monthly basis, inserted into previous months, would not show the actual outcome that would have occurred.

Becky: the way it works is that you get a different system for every different value of the parameters. For example, if you are using a moving average, you could have a length of anywhere from 1 to 200 days. That gives you 200 different possible systems. If you have 5 or 6 variables, like this system does (3 for the Coppock indicator and I think another 3 for the additional manipulation), you will have a huge number of potential systems. The way some traders work it that they try them all and pick the best. The problem is that with all these possible systems, it is almost statistically certain that some will backtest well just by pure chance. That doesn’t mean they have real predictive value, and will lose money going forward if you run them.

Hope this helps…

Babak, I come up with different results. So I decided to better check my script. Also the Coppock GUide Formula cited by Jay is not the original one (or see this).

Very nicely described Burt. I haven’t had a statistics class in a long time, but the professors used to reference something called ‘degrees of freedom’. Each time you added a parameter, you lost a degree of freedom and you had to weigh whether the enhanced predictability for adding an additional variable outweighed the loss in degrees of freedom.

The jest of which was, you would rather have a system that had a 10% profitability based on one variable, than say a system that would yield a 11% return based on two variables, or a 15% return based on 10 variables. As Burt alluded, you can keep adding variables and angles until you get the return you desire, often referred to as curve fitting. If you are trying to get published or write a book, it is tempting to do such, if your trying to make money as a trader, you want to shy away from such practices.

The systems should always pass sanity checks as well. if the system has predictive ability on a 100 day moving average, but absolutely none when measured on a 75 or 125 day moving average, you are probably modeling what is referred to as noise or randomness, such as the results obtained when flipping a coin 10 times to determine if you are in a bull or bear market.

When I did my moving average studies, I was quick to point out that although the 200 day ma was the best N day system when backtested over the last 40 years, if the characteristics of the market change, so will the results going forward. It is almost a guarantee, that some other N day system will yield the best results going forward, especially since the 200 day average is not the best system when tested against other commodities that have different characteristics (sanity check).

For readers with little knowledge of statistical theory,

Just remember ’simpler is better’. The less variables in the system, the more likely you are to get the same returns going forward that you get in the backtest. The objective is not to see who can come up with the best backtested system, but who can come up with the best forward tested sytem.

While on the subject, the Stock market has a bullish bias since inflation is a factor. It is as easy to find bullish sytems that have a 20% return than bearish sytems that return -5%. Good reliable bearish systems are almost priceless, when dealing with SPs. That is why, Babak’s put call study this week was worth the space. Any bearish system that produces negative returns over a sample space of 20 or more, is worth acknowleging.

Useful, thanks guys.

I wonder if the system Jay uses overcomes some of these objections by eliminating certain possibilities in the calculation? That is, he compares (subtracts) MA#1 vs MA#2 and if it does not reach a certain point, it is not reflected in the final number? Just thinking out loud.

Sure seems like an awful lot of colinearity to me…

But I have little faith in the original Coppock (it is just a a bunch of optimized MA’s after all; averaging anything long enough will inevitably produce a trend), so maybe this will be an improvement.

Buying the S&P 500, starting with $10K, no commissions or slippage deducted, 0% interest earned when in cash, no dividends included. Buy on 3, Sell on 0.

Enter Date Exit Date % Profit Cum. Profit MAE MFE

6/29/1951 11/29/1957 88.27% 8816.4 -2.42% 131.10%

5/29/1958 7/31/1962 28.66% 14207.76 -0.96% 66.84%

2/28/1963 9/30/1966 18.38% 18657.11 -2.35% 42.84%

3/31/1967 2/27/1970 -1.81% 18138.71 -2.30% 23.78%

12/31/1970 12/31/1973 8.15% 20427.45 0.00% 37.39%

3/31/1975 2/28/1978 8.12% 22895.37 -2.82% 30.27%

5/31/1978 10/30/1981 19.69% 29361.61 -4.85% 44.42%

9/30/1982 12/31/1987 94.94% 66629.8 0.00% 181.46%

11/30/1988 11/30/1990 9.91% 74219.6 -4.99% 32.27%

3/28/1991 3/30/2001 235.08% 271911.63 -0.61% 319.16%

5/30/2003 4/30/2008 48.98% 409764.63 0.00% 71.37%

8/31/2009 3/19/2010 16.50% 478981.25 -1.24% 16.56%

Annual Return 6.68%

Exposure 81.66%

Winners 11 (91.67 %)

Max. system % drawdown -30.40%