Intermarket Divergence – A Robust Method for Signal Generation

May 2, 2016 5:00 am21 commentsViews: 4060

Many markets are interrelated. These interrelationships can offer predictive capabilities for many markets. The study of these interrelationships is called intermarket analysis. In this article I will briefly explain a robust method for generating robust signals for a wide range of markets. I will also offer a free TradeStation tool to help you explore intermarket relationships.

Standard correlations between markets are not useful if our goal is to either predict future prices or generate profitable signals because current correlation does not tell us anything about future prices. A methodology we originally developed in the mid 1990s called intermarket divergence allows us to gauge the predictive power of an intermarket relationship and produce 100% objective signals. During the past 17 years we have used this methodology to develop trading systems which have produced robust and reliable trading signals even 17 years after the models were originally developed without any re-optimization. Other methodologies of processing intermarket relationships to develop trading signals might perform as well during in-sample periods, but do not perform as well during walk forward periods and during real trading.

A widely known intermarket relationship is the one between the S&P 500 and the 30 year Treasury bond. Bond prices generally are positively correlated with the S&P 500 (while yields are negatively correlated), although this is not always true, bonds should generally lead stocks at turning points. Another important fact is that one of the best trades you can make in the S&P 500 is when 30 year Treasury bond diverge from the S&P 500; for example, when (a) bonds are rising and the S&P 500 is falling, buy the S&P 500 and (b) bonds are falling and the S&P 500 is rising, sell S&P 500. Although this relationship has broken down over the past few years, its long term existence is of historical importance to the science of intermarket analysis.

Simple But Powerful Method of Market Predictions

We will use classic mechanical methods for trading intermarket relationships, applying them to 30 year Treasury bond using a concept called “intermarket divergence,” (first coined in 1998) which is when a traded market moves in an opposite direction to what is expected. For example, if we trade the S&P 500, 30 year Treasury bond rising and the S&P 500 falling would be divergence since these are positively correlated. If we were trading the 30 year Treasury bond, both bonds and gold rising would classify as divergence since they are negatively correlated. We will define an uptrend as when prices are above a moving average and a downtrend as when they are below the moving average. Now we can predict with some reliability the future direction of bonds, stocks, gold, crude and even currencies using this simple intermarket divergence model. Pseudo code for this basic model is as follows:

Price relative to a simple moving average

Let InterInd = Close of Intermarket - Average (Close of Intermarket,N)
Let MarkInd = Close Traded Market - Average (Close of Traded Market,M)

Positive correlation

If InterInd > 0 and MarkInd < 0 then buy at next bars open
If InterInd < 0 and MarkInd > 0 then sell at next bars open

Negative correlation

If InterInd < 0 and MarkInd < 0 then at buy at next bars open
If InterInd > 0 and MarkInd > 0 then sell at next bars open

This simple concept represented above has proven to be a robust methodology for predicting future price action using intermarket analysis. In 1998, I published a simple intermarket based system for trading 30 year Treasury bond futures. This model used ‘The NYSE Utility Average (NNA)’, which was a basket of Utility stocks. The NNA was discontinued in 2004. Another utility index which also worked fairly well was the Philadelphia Electrical Utility index which was used as a replacement for NNA in our research. Back in 1998, when I did the original research and article, both indexes worked similarly, but NNA had a longer price history than UTY did. The original analysis using NNA was done as follows. We used a positive correlated intermarket divergence model and a moving average of eight days for 30 year Treasury bond and 18 days for NNA. We tested over the period Jan 1, 1988 to Dec 31, 1997. We did not deduct anything for slippage and commission. My original published results were as follows:

Net profit: $111,293.00
Trades: 126
Win %: 60%
Average trade: $883.38
Drawdown: $-8,582.00
Profit factor: 2.83

Now let us see how UTY worked during this same period using the original set of parameters used with NNA. This set of parameters was suboptimal for UTY but we are using the NNA set of parameters for consistency to show the robustness of our model.

Total Net Profit: $83,557.98
Total # of trades: 141
Percent Profitable: 58.87%
Avg. Trade (win & loss): $592.61
Max intraday drawdown: ($11,722.50)
Profit Factor: 2.03

Out-of-Sample Results

Let us study just the out-of-sample period with a first trade after 01/01/1998 to 10/25/2011.

Total Net Profit: $129,166.32
Total # of trades: 257
Percent Profitable: 61.87%
Avg. Trade (win & loss): $502.59
Max intraday drawdown: ($26,133.36)
Profit Factor: 1.67

We can see these out-of-sample results are very similar to the results over the whole period and the average trade differs by less than 20% between the in-sample and out-of-sample period. Let us look at the year by year out-of-sample results (see Table I).


We have seen that intermarket divergence is a powerful concept. When an intermarket divergence occurs we stay in that position until an opposite divergence occurs. One question is, “Why does this divergence concept work?” Also, what is interesting is that my research shows that the zero crossing is significant, we cannot improve the results of intermarket divergence by using a non-zero threshold. It is my belief that this concept works as an arbitrage play. Since we do not know the relative equilibrium between the traded market and underlying market, for example in the case of Treasury bonds and UTY, divergence is the only confirmed mispricing; we have in terms of a reliable arbitrage play. We know that this cannot be the most efficient signal. We can see by studying our Treasury bond trades that some trades are early; for others we give back large percentage of open profits and sometime large winning trades can become losers, even though intermarket divergence still produces outstanding results.


Here, we have a very profitable trade but we gave back almost all of the profit and then the market moved back in the direction of the trade. This shows a problem with intermarket divergence namely “Reversal Strategy” which is always in the market. There are other cases including (a) a winning trade ending up as a losing one and (b) trades which never become profitable. Despite these issues our results are amazing. One solution to this problem is to build a finite state machine which covers all possible states of the intermarket relationship during the process of going from ‘long to short’ or ‘short to long’. My research has shown that this state map of all possibilities is the key in greatly improving the performance of these simple divergence models. We can also create a state map which will allow us to combine multiple intermarkets against a market we are trading. Correlation and forward correlations analysis between markets can also be used to filter and improve these models. Sometimes correlation analysis can make the long term out-of-sample performance less robust if it is not integrated carefully. Hence, it is important to do the surface analysis discussed earlier to make sure that the correlation relationships we are looking at are robust and stationary.


Intermarket divergence is not something which just works on the bond market. It works on a broad range of markets from bonds, to stock groups, to currencies; even markets like gold, crude, live cattle and copper.

Intermarket analysis is an exciting area of market production. New methodologies of representing these relationships will help not only classic trading system development but also using advance technologies as for example using a finite state model can allow machine learning methods to easily see patterns which can be used to build more reliable models.

Build robust and profitable systems that predict market turning points with this tool.

Intermarket Divergence Pro

— by Murray Ruggiero of Using EasyLanguage.

This article was created based upon the paper, Intermarket Divergence, which was published within Computational Intelligence for Financial Engineering & Economics (CIFEr), 2012 IEEE Conference on March of 2012.

Murray Ruggiero is the chief systems designer, and market analyst at TTM. He is one of the world’s foremost experts on the use of intermarket and trend analysis in locating and confirming developing price moves in the markets. Murray is often referred to in the industry as the Einstein of Wall Street.

He is a sought-after speaker at IEEE engineering conventions and symposiums on artificial intelligence. IEEE, the Institute of Electrical and Electronics Engineers, is the largest professional association in the world advancing innovation and technological excellence for the benefit of humanity. Due to his work on mechanical trading systems, Murray has also has been featured on John Murphy’s CNBC show Tech Talk, proving John’s chart-based trading theories by applying backtested mechanical strategies. (Murphy is known as the father of inter-market analysis.)

After earning his degree in astrophysics, Murray pioneered work on neural net and artificial intelligence (AI) systems for applications in the investment arena. He was subsequently awarded a patent for the process of embedding a neural network into a spreadsheet.

Murray’s first book, Cybernetic Trading, revealed details of his market analysis and systems testing to a degree seldom seen in the investment world. Reviewers were universal in their praise of the book, and it became a best seller among systems traders, analysts and money managers. He has also co-written the book Traders Secrets, interviewing relatively unknown but successful traders and analyzing their trading methodologies. Murray has been a contributing editor to Futures magazine since 1994, and has written over 160 articles.

As chief systems designer, Murray digs into the depths of niche and sub-markets, developing very specialized programs to take advantage of opportunities that often escape the public eye, and even experienced high level money managers.



  • What about between 2012 and now? Does it work? Because many systems no longer work. S&P 500 and bonds have long negative correlation durinf previous years.

  • Intermarket Divergence , because it requires divergence is more robust and volatility of equity will increase when correlation a weak, but it will not fall apart completely. You also could add a correlation filter to a model and deal with this issue. I’ve attached the performance report (through 3/30/2015) at the very bottom of the article.

  • Alex Argyros

    Thanks, Jeff. Great work.

    I was wondering why you chose 60 minute charts for ES and YM. Would other time frames work with the moving averages you used?

    • Thanks. Murray created a great free tool. I doubt other timeframes would work with the periods I used for the 60-minute chart. I picked a 60-minute chart because 1) I wanted an intra-day chart 2) I know the smaller the timeframes, the more noise there is thus, it’s more difficult to build a system. So, I just picked 60-minutes. If you do move to other timeframes, it would be best to optimize the look-back periods on both SMAs.

  • Just watched the video. I see that the 2 moving averages have different lengths. Could you not just put a second moving average(the one on YM) on the ES get the same results? You would get moving average divergence doing that instead of using a second symbol to do that.

    • This would be best answered by Murray, as he is the creator of both the concept and code. Here is my $.02 however. You could use two moving averages with one instrument, but all your information for taking that signal is coming from on market – your tradable. In this case we are actually using another market to signal trades. In other words, we are using data not directly related to the current price of the traded instrument. This might be analogous to using volume or $TRIN to generate signals instead of just relying on the price of the tradable.

    • No , This is truly controlled by the intermarket relationship. I have tested many different intermarket relationships for example Tiffany stock on intra-day data leads S&P500. I normally use 45 minute bars for the S&P500. Some currency pairs also lead the S&P500 on a daily as well as a Intra-day basis. Another classic example is the one in the manual trading US versus UTY. If we just use two different lengths for US (thirty year treasury daily data) even selecting the most profitable combinations will still make about 120K less than using the intermarket UTY. That is about a 40% decrease in profit and the results are a lot less robust.
      I have a professional version which allows you to filter out times when the correlations decouple and generate 100% mechanical signals.

  • “Bond prices generally are positively correlated with the S&P 500 ” – Did I miss anything here? Shouldn’t they be negatively correlated in last twenty years? Which time scale was your correlation based on? Thanks.

    • At times they are negatively correlated but the truth is divergence between bonds and stocks since 1971 when we completely went off the gold standard have been positively correlated.

      Negative correlation occur during black swan events, in those cases bond prices go up and stocks go down.

  • Hi Jeff

    Great blog – Thx for sharing.

    A thought. Would the same/similar results be achieved by using 31MA on S&P & 11MA on S&P? In other words leave out the DOW completely. My thought is that the results are more linked to the difference in MA’s rather than any divergence between S&P & DOW. Does that make sense?

    • If you don’t pick a good intermarket a reliable one you could see the intermarket not outperform by much.
      In my Intermarket Divergence Professional Product, you can select the markets you are trading and up to 50 intermarkets and use the optimizer to optimize not only the moving average lengths but also the markets. You can run an experiment and select 20 different positively correlated markets limit in TradeStation is 50, include Bonds,Currencies,Sectors,Other stock indexes and find which ones are more predictive. The way I do this is I will look for how many times a given market appears in the top 20 sets of parameters. Often times you will see 17 or 18 of the top 20 sets of parameters will be made up of only 1 or 2 markets.

      • Good, but do you consider data-mining issues? if you have 50 intermarkets it is highly likely that the correlation you end up with and the system as a result are spurious.

        I suggest something you are not of course obliged to do: Post here the results of your best intermarry divergence system for in sample and out of sample. Then we can have a base for more serious talk. I strongly believe these are spurious systems. No offense intended.

        • See the answer is no, If you have 50 intermarkets for example and 15 out of 20 of the top sets of parameters are the same market,you pass any statistical test for significance you need. In addition for this best 1-2 markets , you also run out of sample testing, 3D analysis.
          I am not sure that you understand the significance of the paper Jeff linked to of mine. That was given at IEEE, Cifer in 2012, it covers out of sample testing, analysis of 3D space ect. It shows two different markets for trading bonds one reliable and one which should be reliable but is not and we predicted this issues for that market in 1998.
          We also show a system which was published in 1998 and has traded real walk forward with the same parameters since. This is peer review academic conference associated with IEEE.
          See Intermarket analysis is even less data mining because I would only put in markets which I can logically think would be predictive. If we were trading the Canadian dollar, we could use Canadian stock indexes and various commodities but for example I would not use Coffee.

          • You paper is fine in terms of demonstrating a method for optimizing parameters but you have not proved that your intermarket method can generate excess returns over and above a method that uses only one of the correlated securities. Our analysis shows that your method cannot generate the excess returns required to discount redundancy. Can you provide proof that the method generated the excess returns? If not, it is not really useful and largely redundant.

  • First , can you share your research. Has it been published peer review. I been using intermarket analysis for over 20 years and trading systems which I developed in the mid 1990’s still work today. I totally disagree with you, bonds have been in a great bull market for 20 years and this system still makes money on the short side. In addition , this is not the only intermarket relationship that works. For example Silver,Chemical Stocks are negatively correlated to bonds and produce good results also. Silver performs better than gold. There are many different intermarket relationships for many different markets. I will cover these in future articles. This is not data mining, I don’t use relationships I can’t explain. For example , even though it works reasonably good , I won’t trade silver using soybeans, because there is not a good premise there.

    • You have not even provided evidence that your method can generate higher returns as compared to trading only one of the correlated securities. More importantly you did not provide standard measures of returns such as the CAGR that allow making comparisons. The burden of proof is on you to prove that your method is sound, not on your readers. Asking your readers to present data to prove that you are correct is not the best way of doing things. Until you provide the data that show that your intermarket method can do better than a simple crossover I can assume that it is not working as advertized. This is what our analysis shows and we do not care about peer review. Pleas note that nobody who is serious in the industry compares absolute profit numbers. You have to present accepted measures such as the CAGR and MAR. Thank you.

  • Thanks Ruggerio and Jeff for publishing this. I have made a couple of valid systems out of this. I couldnt get the Indexes to work but found a solid edge in a couple of Bond markets.

  • I don’t think he’s asking you to share research to show he is correct, Sam: he’s asking you to share your research claiming he is incorrect. I think he has every right to ask for that.

    By the way, my gut says intermarket divergence is bunk and the fact that Murray is selling a product here doesn’t help. However, I think there would be a great deal of value in a peer-reviewed article. I am going to take a read through the IEEE paper.

Leave a Reply

Time limit is exhausted. Please reload CAPTCHA.


Download these four simple steps to help ensure you make money on the live market!