RSI And How To Profit From It

[BlueHorseshoe 10]

  Adrian said:

@BlueHorseshoe A statistician will tell you, you need 30 trades to be statistically significant. When it comes to trading though, they haven't a clue what they are talking about. Some traders will tell you they wish to see hundreds or even thousands of trades to validate a model. They also do not know what they are talking about. I have absolutely no problems at all with the sample of 60 trades or so over the large time period. Why? Because when you understand markets and what is important, then you simply don't need a sample of hundreds of trades. Do you think Warren Buffet needs to see hundreds of sample trades to know that buying a business at a low price relative to its earning potential will work or not? Obviously not. Those with experience 'just know' because its obvious. When it comes to trading models, what's important is testing over different market environments, and understanding how and if the model is designed to deal with them adequately. Certainly you would have to be extremely patient, and be trading other models as well beside the RSI model, as the trades are few and far between, but that is irrelevant as to its validity or not. Now if a short term model was presented with 60 trades over the last month was presented, then I would laregly dismiss it until it was tested over a much larger database. Why? Because 1 month of intraday data is NOT indicative at all, of the type of market environments that lie ahead in coming years.

 

The robustness of the variables is also of clear importance, but even there you need to be very careful. You can't simply test an RSI from 1 to 10 and see if the model keeps working. Markets are more complex than that. A 10 period RSI won't reach the same extremes a 2 period will. So you might try adjusting the oversold level with it, but even there it isn't that simple. A 2 period RSI will show situations that simply cannot occur with larger period RSI's. This is why creating a successful and robust model is far more difficult than most think. It requires a great understandsing of markets in the first place.

 

I agree with all but your very first statement. A statician will not tell you that 30 data events are significant; they will tell you that 30 data events are significant for certain types of data. The exotic data produced by the financial markets not does not fall into this category.

 

Other than the above, I am in total agreement with you. My original post was just pointing out, regardless, that the strategy does hold up well even across greater sample sizes than the 60 trades the thread starter shows.

[Phil-n-Texas 10]

  grose said:

Hi Phil-n-Texas,

 

Thanks for your post. I also have OT Pro. I can probably figure out how to filter on hit rate and select based on profit per trade, but if you have any suggestions, they would be great. Do you combine the strategies into a single strategy in OT Pro and require some n number of them to fire, or just allow all of them to fire signals and then filter the way you described?

 

Last question - promise - is the equity curve you mention from live trading or from OT reports? I've never fully trusted the OT backtest results.

 

The separate discussion about stops is a good one. I would want to have some kind of disaster stop in place - the flash crash would be a good example of why. But in general, selecting the "right" stop is very difficult and I have also seen system results actually go down thanks to stops!

 

Thanks,

 

Gordon

 

Gordon,

 

i posted port-sim equity curve in this thread...

 

as for real trade the curve is actually better as OT trade-plans are hosed on scale-in/out at mkt open/close...

 

yes, i have a luv/hate relationship with OT and the trade-plans issue in this regard is really a major screw-up on their part... they know about it and have yet to address it... there is quite a lengthy thread on the OT Pro forum and a bunch of us have piled on but it has not been fixed.... I'm hoping some of them make it to the trader expo here in Dallas so I can hit Ed up about it directly...

 

anywho... so basically i end up manually doing the scale in based on actual rules of buying and selling @ mkt close... so whenever i'm testing i can figure that the results in real trading will be as good or better. YMMV

 

the vote-line config is all done in "strategy voting"...

 

 

 

[fxtrader12 10]

  ValueTrader said:

The best way to profit from RSI is by steering clear of it. Basing your trading on two moving averages! There's no such thing as oversold / overbought. When will people realize the markets ain't what they were and the tried n tested methods are old hat.

 

Both RSI and MA's are valuable as momentum indicators. Oversold and overbought levels concept is working very well, and can be used as signals.

Maybe try to analyse the issue at low settings.

[onesmith 10]

  Quote

A statistician will tell you, you need 30 trades to be statistically significant. When it comes to trading though, they haven't a clue what they are talking about. Some traders will tell you they wish to see hundreds or even thousands of trades to validate a model. They also do not know what they are talking about. I have absolutely no problems at all with the sample of 60 trades or so over the large time period. Why? Because when you understand markets and what is important, then you simply don't need a sample of hundreds of trades. Do you think Warren Buffet needs to see hundreds of sample trades to know that buying a business at a low price relative to its earning potential will work or not? Obviously not. Those with experience 'just know' because its obvious. When it comes to trading models, what's important is testing over different market environments, and understanding how and if the model is designed to deal with them adequately.

 

When coders begin to question what bots are seeing at any given point of interest, and think in terms of the rule based lines of code that could be active at that moment in time then you will be able to write statistically significant strats based on a sample size of one. Your interest in backtesting your models will become secondary to reverse engineering what's happening right before your eyes because you intuitively know when you see something that adds value to your models.

Edited April 15, 2012 by onesmith

[BlueHorseshoe 10]

  onesmith said:

When coders begin to question what bots are seeing at any given point of interest, and think in terms of the rule based lines of code that could be active at that moment in time then you will be able to write statistically significant strats based on a sample size of one. Your interest in backtesting your models will become secondary to reverse engineering what's happening right before your eyes because you intuitively know when you see something that adds value to your models.

 

Your references to 'bots' (here and on another thread) . . . Do you mean the kind of algorithmic strategies a retail trader such as myself could produce, or do you mean the super-sophisticated, zero-latency, servers in the CME, programmed by fifteen PhDs, octagonal-arbitrage type thing that multibillion dollar HFT firms unleash? Could anyone hope to gain an inkling of what the latter are doing by looking at a price chart?

 

Thanks

 

Bluehorseshoe

[onesmith 10]

Everyone who looks can find James Simmons's b o b bots. Don't try to understand them within the context of his reasons, his processes, or his charts. Build your understanding upon whats in your chart, and base what's in your chart on whether it helps you see bots.