Data Mine Strategy - What's Been Done?

[goodgoing]

I have researched RSI on a fair number of instruments and as far as I can tell it has the same predictive capacity as tea leaves or chicken entrails.

I have an RSI system that has been profitable every year for the last 15 years. Maybe you are mixing RSI and oranges

 

[Jerry030]

I have an RSI system that has been profitable every year for the last 15 years. Maybe you are mixing RSI and oranges

Is your system automated, that is 100% rule based so that it could be programmed in a computer and operate without your or any human intervention? Does it include any measures, factors or metrics in addition to the RSI? My analysis of the RSI was as an independent variable in a very sparse trading system shell and was evaluated on about 20 years of data for several dozen instruments. The mathematics indicated that the RSI had almost zero predictive value (on a par with chicken entrails). Since I don't know what your system involves I can't speak to why it works except to say that if it is based only on RSI you are probably just lucky and mistaking that luck for predictive functionality.

 

[Sergio77]

RSI is one of the best indicators if you know how to use it but using it properly is an edge nobody will talk about.

 

[thefuturestrader]

Jerry030, Orange looks really cool. Have you used it at all? I am learning how to do some basic stuff in R, but Orange seems intriguing if I were to ever want to try a data mining strategy (which is way above my head at this point).

 

[Jerry030]

RSI is one of the best indicators if you know how to use it but using it properly is an edge nobody will talk about.

Of course it is. It is just that when subjected to rigorous statistical evaluation over decades of data it exhibits virtually zero predictive ability......on the level of chicken entrails. Now we should not knock chicken entrails. The Romans used them to predict the out come of battles and did well for 600 years.....until the damn Germans destroyed their civilization. Perhaps they ran out of chickens and started using duck entrails which modern scientific analysis has shown to always predict the reverse of what will actually happen.

While I will not reveal the secrets of Duck Entrail Trading here, you can attend my seminar next month for $4,000. I promise a surprise visit by J. Wells Wielder's ghost.

 

[WeToddDid2]

Please take this question seriously.

My questions / topics of interest to discover trading strategies:
- Have you come across any technical analysis studies that leverage modern day computational power to combine N variable combinations over various holding periods to ascertain return maximization? A lot of ideas there so let me give a few examples:
o Simplest idea would be using e-Mini future contract ES1 while combining RSI and MACD indicators. If each variable is split into deciles and a buy signal is generated for each permutation (ie, when RSI decile 1 occurs with MACD decile 1 then buy, when RSI decile 1 occurs with MACD decile 2 then buy, etc etc.). And if the holding period is 30 minutes, what is the max return across all scenarios? The study would also determine if there is a discernable pattern or a fluke result.
o Then the study would change holding period from 5 minutes to 10 minutes to more periods to 5 days.
o There are a few other parameters in this exercise but you get the idea- program a market setup and see what the computer finds.

- Similar to the above study, have you come across any research that data mines all price scenarios for a maximized return? For example, using OHLC data on a X minute interval, one could create the following six scenarios:
Open (prev bar) < High(prev bar) Open (prev bar) < Low(prev bar) Open (prev bar) < Close(prev bar)
Open (prev bar) > High(prev bar) Open (prev bar) > Low(prev bar) Open (prev bar) > Close(prev bar)

Then varying the holding period, is there any statistically significant return?
This approach could also be expanded to the candlestick patterns such as a doji, hammer, etc.
This could also be combined with the above approach for conditional returns. Ie, perhaps RSI has little predictive ability but when given situation ABC, it works very well.

- I’m wondering if this analysis is old news or a potentially meaningful endeavor. Tracking down studies is quite difficult (at least for me) and I’d be curious to know the major technical analysis enhancements over the past 10 years and what is cutting-edge today (excluding high frequency trading at a milli or nanosecond level). That answer would help me get a sense if I’m on the right track for trading strategies.

Yeah, I have seen several platforms capable of the majority of your list above.

 

[Jerry030]

Jerry030, Orange looks really cool. Have you used it at all? I am learning how to do some basic stuff in R, but Orange seems intriguing if I were to ever want to try a data mining strategy (which is way above my head at this point).

I have just started using it. The appeal to me as a software engineer is there is no programming! I'm so tired of endless coding to build data flow and process infrastructure.

 

[Sergio77]

Of course it is. It is just that when subjected to rigorous statistical evaluation over decades of data it exhibits virtually zero predictive ability......on the level of chicken entrails. Now we should not knock chicken entrails. The Romans used them to predict the out come of battles and did well for 600 years.....until the damn Germans destroyed their civilization. Perhaps they ran out of chickens and started using duck entrails which modern scientific analysis has shown to always predict the reverse of what will actually happen.

While I will not reveal the secrets of Duck Entrail Trading here, you can attend my seminar next month for $4,000. I promise a surprise visit by J. Wells Wielder's ghost.

RSI is one of the best indicators if you forget about backtesting and learn how to trade.

 

[Jerry030]

RSI is one of the best indicators if you forget about backtesting and learn how to trade.

Sadly I'm not much into religious trading. I approach it from a scientific perspective. You conduct an experiment: if x then what is y?, if very similar to x what is y?. You repeat this over and over again. It's called the scientific method. It is used to test everything from new drugs to alloy composition of auto power train components.

In functional trading systems it is called backtesting. If it did not work at any time in the past the probability that it will magically start working just when I place a trade is pretty close to the null set. However if you like this kind of thing I would suggest adding Duck Entrails to the RSI calculation after consultation with J. Wells Wilders ghost. Jesus be with you.

 

[WeToddDid2]

Jerry030, Orange looks really cool. Have you used it at all? I am learning how to do some basic stuff in R, but Orange seems intriguing if I were to ever want to try a data mining strategy (which is way above my head at this point).

What is Orange?