RSI based trading strategies

frost, before going further, I would answer the question of why your baseline percent profitable is > 50% for long and short. It may be correct, but I question it and you question it. If you get through tests of 10 indicators before figuring out there is a problem with your procedure, the results will be meaningless and you will have to start over.

A few suggestions:
If your random trades are generated as 1 lots, you might instead scale based on the previous day's volatility.

I would also look at whether it makes sense to use average and standard deviation when looking at thresholds for returns since they aren't normally distributed. Quartiles might make more sense.


Edit: On second thought, the distribution of returns in this case might actually be normally distributed. I'd have to look at the data. I think it still makes sense to look at using quartiles for your cutoffs and see how it affects your results.
 
attached monthly performance of a system based on rsi trading crude oil futures.
yeah, the indicator is lagging, a lot of big moves were not catched, sometimes its losing money... plus its super simple.
please, don't trade RSI !
 

Attachments

I use something similar to an RSI which is a little bit more reliable. I then combine that with price action and then add in risk vs reward of the trade which I am still working on. I need to decide if I should use an even risk vs reward or see if I get better results with reward being higher than risk.
 
Quote from kut2k2:

To the OP:
While I agree that those who say the RSI is lagging (or all indicators are lagging) are simply dead wrong (you don't get meaningful divergences with lagging indicators), I find the RSI itself to be a klunky indicator that is a poor substitute for better, more direct measurements of momentum. But it is popular so examination of its performance is not out of line.

To Ninna:
I don't know what good applying the RSI to a stationary process will do, because neither price nor any of its derivatives are truly stationary, although log returns have a sort of quasi-stationarity about them. Is that what you meant?
More...

No that's not what I mean.

I don't agree with you, derivatives of price can be stationary (and very close to gaussian). That's what I trade.

Ninna
 
Quote from rdg:

frost, before going further, I would answer the question of why your baseline percent profitable is > 50% for long and short. It may be correct, but I question it and you question it. If you get through tests of 10 indicators before figuring out there is a problem with your procedure, the results will be meaningless and you will have to start over.

A few suggestions:
If your random trades are generated as 1 lots, you might instead scale based on the previous day's volatility.

I would also look at whether it makes sense to use average and standard deviation when looking at thresholds for returns since they aren't normally distributed. Quartiles might make more sense.


Edit: On second thought, the distribution of returns in this case might actually be normally distributed. I'd have to look at the data. I think it still makes sense to look at using quartiles for your cutoffs and see how it affects your results.
More...

The long/short both having > 50% winning percentage had me confused for a while. I finally realized that NinjaTrader counts flat trades as wins. Compounded with the fact that for this evaluation I am looking for prediction not sustainability I am not including commission or slippage. Therefore, both strategies had a number of flat trades which caused each of them to have inflated winning %. Since all evaluations are currently being handled the same way, all strategies have an inflated winning percentage and my results remain valid.

I also did a number of tests where I ran 10,000 random simulations to see if at any point I could exceed my targets purely by chance. The target values held up. Several simulations came somewhat close, but were never able to surpass the target.
 
Yeah, that makes sense.

Does NinjaTrader allow you to generate all possible trades when entering when an indicator is below a threshold value, including trades that overlap others? Ie, if you are testing RSI < 30 with a holding length of 3 bars, in the case when the RSI is below 30 for 5 consecutive bars, do you generate 2 trades or 5?
 
I just finished composing a list of which threshold and RSI period values are optimal when predicting 3 or 5 bars into the future. The results were somewhat surprising. Apparently shorter RSI period values are more predictive for this time frame than longer.

After mapping out which values are best, I took one of the better combinations RSI(3) < 30 and tested it against the 30 DOW stocks to see how robust the method was. Keep in mind the values were optimized against ES. The method was able to pass our target percent profitable thresholds in 17 of the 30 DOW stocks and pass the target profit in 7. This leads me to believe this method is fairly robust.

There is too much data to display in this post, so if you want to see all of the values found in my research go to: http://thestrategictrader.com/?p=178

Looking forward to hearing some feedback.
 
I would like to point out that this is NOT a strategy. The intention for now is to simply identify methods of using RSI that provide an edge. Once a "toolkit" of methods has been built, we will look for ways at forming a strategy using them.

The information discovered thus far could be useful in a strategy. For instance, imagine a strategy signals its time to enter long and RSI(3) <30. This would provide further reassurance that the entry is good. On the flip side, if a strategy says to go short and RSI(3) < 30, it may not be the best time to enter. Simply because the odds are tilted out of our favor at least for the next few bars.
 
Anyone have ideas on other ways RSI may be traded? The next two ideas I will be researching are:

#1 Using RSI to find divergence
#2 Finding patterns within RSI movement.. using several AI techniques

Any more basic ideas to try before taking it to that level?
 
You must log in or register to reply here.
Back
Top