[Archive!] FOREX - Trends, Forecasts and Consequences (Episode 6: August 2011) - page 175

New comment
 
marketeer:
What is wrong with such a criterion? If the adjacent forecasts overlap quite poorly, then any of them seem to be unlikely. From the algorithmic point of view it seems simple enough to sum the proposed trajectories, and if they do not agree with each other - we get a horizontally directed cloud (confidence interval will be wider than the length of the "origin-target" movement). If there are other ways of determining stability, please tell me.


Here's what came to mind. Make a forecast not from the 1st closed bar, but from the 5th (for example). The last 5 real bars should be used as a control test for the forecast - count % of coincidence with the forecast for them (let's call it a control deviation). If it is low - change the sample size (PostBars) until the control deviation becomes acceptable. An analogue of the forward test. This is one.

Two - changing PostBars plus or minus 1 should not affect the control deviation too much. If it does, then we have found a random match.

Three - the forecast cloud(dashed lines) should not diverge evenly in both directions. It must necessarily be asymmetric (more to the south or north). This is an additional signal.

These are just ideas for now. I will experiment this weekend.

 
wmlab:


Here's what came to mind.

Yes, all variants - one, two and three - work. You can complicate the indicator, or you can implement them already in the EA.
 
marketeer:
Yes, all variants - one, two and three - work. You can complicate the indicator, or you can implement them already in the EA.
But who prevents you from getting signals through icustom and creating an Expert Advisor? It seems to me it is the easiest option.
 
it is not clear the theoretical justification for the indicator prediction, why should the prediction come true?
 
trinitron: How about using confidence intervals? It seems to me that the probability on the graph is too high.
The most important thing is where to get them, these confidence intervals. The statistics are hardly too extensive to have an idea of the distribution of the values we want. And without that distribution it's hard to talk about confidence intervals.
marketeer:
What's wrong with that criterion? If the adjacent forecasts overlap rather poorly, then any one of them seems unlikely. From the algorithmic point of view it seems to be simple enough to sum the proposed trajectories, and if they do not match each other well - we get a horizontally directed cloud (the confidence interval will be wider than the length of the "start-target" movement). If there are other ways of determining stability, please tell me.
I don't have any options. But the problems are similar (also related to prediction and estimation of its reliability).
 
sever31:
it is not clear the theoretical justification for the indicator prediction, why should the prediction come true?

There is no theory. There is an underlying assumption that typical movements repeat themselves. Of course, the news introduce chaos, but no one has promised to feed on the road there is a hope to improve the forecast from 50/50 to at least 60/40. If it works - the original assumption was correct.
 
sever31:
it is not clear the theoretical justification for the indicator prediction, why should the prediction come true?
There is a lot of theory, but there is not much practice.
 
sever31:
I do not understand the theoretical basis for the indicator forecast, why should the forecast come true?

the theoretical underpinning of the indicator is the third postulate of the Dow theory (and of TA in general):
"history repeats itself", taken to the absolute

In general, this postulate denotes the consistency of the psychology of market participants, i.e. you can expect repetitive patterns and investigate such patterns... But what if we take as a pattern just a few recent bars that are not pretty, not very spectacular, but psychology should not play only where it seems "pretty" to you, and look for coincidences in the past. If you manage to find a few similar realisations, you can see their development, and if the development is similar - expect the same psychology, the same development this time.

Everything seems to be very logical.

 

a movie I saw here about a colleague))))

http://rutracker.org/forum/viewtopic.php?t=3699204

 
rigc:

a movie I saw here about a colleague))))

http://rutracker.org/forum/viewtopic.php?t=3699204


Is there a Russian translation available/not available??? :-))) I'm not that good with English yet... :-)))

Guys, can you tell me - in (-on) what it can be watched with subtitles in Russian?

Thank you.

1...168169170171172173174175176177178179180181182...202
New comment