Correlation Trading -Ideas and Strategies for quant analytics

(Last Updated On: May 25, 2012)

Correlation Trading -Ideas and Strategies for quant analytics

looking to improve knowledge and best information possible on the subject . Interested mainly in Etf and S&p 500 stocks strategies.
If possible a white paper?

just google pairs trading or cointegration

Conduct me by email to send you a couple of them as zip file. I am doing similar research… Except the pairs trading or cointegration, you can look for machine learning or data mining (association rules) methods.

I have been experimenting with machine learning techniques for a few years. I am still not convinced they work but they may have potential. Genetic Programming (evolving strategies), Genetic Algorithms, Support Vector Machines and Random Forests are probably most suited to this task. I haven’t had any luck with neural networks, and they are very slow. Generally you generate a file of information on the price using a bunch of different indicators. Then you have your machine learning algorithm predict either the price tomorrow, or the direction of the move tomorrow.
Another approach I haven’t looked into but which has a lot of potential is writing a bot to react to news or to analyze sentiment on social media sites such as twitter.

I agree that the concept of using the ANNs for predicting the markets (producing trading signals) although it sounds promising, it is very difficult to implement (requires much research). But if you try / research more, you can develop very profitable ANN system. I could give some directions such as:

(i) There are a lot of different types of predictions( not only tomorrow’s prediction) such as predict future price values of x(t), classify a series into one of new classes (Price will go up, price will go down, no change), transform one time series into another (e.g. oil prices => interest rates), predict noise (if noise is greater than a threshold value then NOT TRADE), predict indicator, predict volatility, etc.

(ii) ANNs can play the role of a filtering system in a rule based strategy. For example, imagine that you have a trading system that produces 42% false signals and 58% true signals. If you find an ANN that filters and transforms your trades in two states: TRADE or NOT TRADE. Then, after combining it, your final system produces: 40% true signals (TRADE), 40% NOT TRADE and 20% false signals. Is it an improvement?

And a lot of other….

web site http://www.market-topology.com will help you find pair for trading.

market-topology is good but , for forex? Do you have some similar site?

At one point I spent considerable time applying Neural Networks to developing trading systems. It is much harder to create a PROFITABLE system than it might appear from the “technical” measures. Some things I did find helpful:

* Rather than attempting to solve the entire trading problem all at once, break it down into smaller “pieces”. For example, given a trending market (based on some other criteria), develop a filter to predict when a fast moving trend is over.

* Predicting a price at some point in the future is almost always futile. In a side-ways market, using Sine transfer functions in the hidden layer can be trained to approximate market movement.

* when you should buy/sell/exit is highly dependent on transaction costs and slippage. One useful strategy was to use a GA to develop a sent of entry / exit points on historical data then train the network to predict the “ideal” buy/sell/exit points developed by the GA.

* As useful as predicting the “price” is to predict the expected range of prices at some point in the future. This is an easier problem to solve, and can be a useful input to a more complex trading strategy.

* As an interesting experiment, I set up a series of 1,500 different MACD oscillators and used them as the input to a simple long/short (always in the market) trading strategy. I tested each of them over a window of 100 trading days, then moved the window forward 10 trading days and repeated the experiment over a period of 1,000 trading days in total. I then plotted the results. What was fascinating about this experiment was that for a given window, some of them were always very profitable. Others very unprofitable. And there were times that the market went through transitions in which performance across the board was mixed. The profitability of a particular MACD oscillator would swing from profitable to unprofitable. There was no one set of MACD parameters that was always profitable.

There are many lessons to be learned from this:
* Most obvious is that there are times when the markets go through transitions in which nothing works well.

* Different parameters work well in different market regimes. It is virtually impossible to “optimize” a system over one period of time and expect it to perform well in another.

* Even a simple strategy can be profitable under the right circumstances.

* If you can identify the market “regime”, you can pick a strategy that is profitable for that regime from a list of potential candidates.

Hope you find this useful. Casey.

very interesting you remarks. I see you have spent a lot of time in experimenting NNs. What additionally I have seen is that you can get better results if you involve other related markets as inputs.

You make a good point. However, the problem is that the correlations between markets shift. As an example, one of the pairs that I follow is the EUR/CL pair. Often, EUR leads CL (futures markets). If the E6 starts to move in one direction, oil often follows shortly afterward. My thesis is that Oil (CL) is denominated in US dollars, so a change in the USD/EUR exchange rate is shortly thereafter reflected in CL. This is particularly true for Intra-day (minute/tick) trading. However, there are also times when they decouple.

I have seen this in a number of markets.Net result is that yes, sometimes one market leads another.Unfortunately, the relationship does not always hold. There is nothing certain in life or financial markets.

p vs dj indu ~m http://markbrown.com/linkedin/pics/LI-51.png

Are you interested in BlackBox Trading? Have you ever Pair Traded?
If so you you might want to take a look at the new Iris Pairs Trading Platform.

The platform is offered at a hand full of CBSX firms, (Now Including T3 Trading). Iris is also offered at a few retail trading shops including LiveVol and Great point capital

Iris is a pairs trading platform with 100% automation capability. Using advanced algorithms, Iris is able to manage an entire portfolio by deciphering patterns in correlated instruments based upon pre-defined quantitative factors.

Iris executes in the most cost effective manner using our smart routing algorithm, which reduces the total execution costs by up to 50% (from standard route out), yet still fills both sides of the pair immediately.
Iris has a three trading modes – Manual Mode, Grey Box Mode
and Black Box Mode

Here’s a link to our website where you can watch a short video about the Iris Trading Platform.


If you are new to pairs trading and want to know more about it you might want to take a look at this video explaining the concepts behind Pairs Trading.


Please feel free to shoot me back with any questions you might have

how much is the cost of software, also, is it works with Forex? Is it based on cointegration?

I would love to have a read of your thesis. can you provide a link?

I have found the same thing across a lot of different markets and ccy pairs.

I have nothing written about this. By “thesis” I meant “my explanation” (in contrast to an academic work in pursuit of a “degree”).

Iris does not currently trade Forex.

Yes co integration is a big part of it for the entry’s and exits. The platform manages the pairs over/up to a 20 level entry & exit STD scale. scaling in and out of the pairs when co integration occurs. Iris uses a proprietary standard deviation calculation that measures the divergence from the mean price relationship between the two paired stocks, resulting in the standard deviation value. By using the standard deviation measurement the volatility for each pair is treated uniquely. It is extremely important to standardize each pair’s volatility when using a black box algorithm such as Iris, because the system is executing each trade based on predefined standard deviation levels generalized for all pairs. Another benefit to using the standard deviation measurement is the time value effect. Over time the standard deviation will fall even if the stock prices have not converged due to such effect, which will force a non-profitable pair to close. The capital can then be redeployed to potentially more profitable opportunities.

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

Subscribe For Latest Updates

Sign up to best of business news, informed analysis and opinions on what matters to you.
Invalid email address
We promise not to spam you. You can unsubscribe at any time.


Check NEW site on stock forex and ETF analysis and automation

Scroll to Top