From theory to practice - page 438

New comment
 
The only person with the lowest degree of education who can make a kolkhoz speech here is bas. He's not a bad speaker sometimes. Apparently, when he's asleep. In his sleep he has an epiphany. Sometimes it's interesting to read.
 
khorosh:

No, price overshooting is not all it's cracked up to be. For the arrow to appear, it requires a change in the direction of the last two bars formed and the value of the last bar is higher than the threshold value. Not all signals are used for trading, as they are also filtered by the trend indicator.

OK

I will finish filtering

 
Alexander_K2:

I suspect that, when calculating in the time window, you only use the increments that are above/below a certain value. Right?

No, I take all the increments of the transformed price.
 
Andrei:


Let's talk without emotion. For, after all, my goal is not good/bad relations with forum members, but the Grail.

Starting from some initial point in time, the integral of all the increments + the initial price = the current value of the price.

So, the sum of the increments is the price in the sliding time window of observation, with starting point =0

I have completely ruled out any average price itself. They piss me off. The only robust estimate of expectation in the moving window is median.

But even in relation to it, if we look at the probability distribution, we'll see Laplace distribution at best.

There is absolutely no formalised reason to "go back to the mean".

Once again - using ANY moving averages fails to reduce the process to an Ornstein-Uhlenbeck process.

And you need it like air!!!

So, if we consider the sum of the increments in a moving window, i.e. the price relative to 0, and look at its probability distribution, we see a VERY normal distribution, and the moving distribution of the increments of such a process is strictly symmetric.

We have, in this case, grounds to assert that if in the current moving window, the transformed price exceeds the confidence level, and the current probability distribution --> to normal, plus one more function, which you and bas know about, then the price will definitely return to the expectation, to 0.

You just need to keep a close eye on the above parameters.

In tests all my trades are profitable.

But practice... Yes, we will see next week.

 
Renat Akhtyamov:

ok

finish filtering

Well, the remaining erroneous arrows are weeded out by the stoploss).

 
Evgeniy Chumakov:
No, I take all the increments of the transformed price.


Not only that, I removed the multiplier from the variance formula. Just Abs(Return)/Sqrt(TimePeriod), but when I looked at the multiplier, I saw that the increments are not outside the channel, so I decided to remove it.

 
Evgeniy Chumakov:


Not only that, I removed the multiplier from the variance formula. Just Abs(Return)/Sqrt(TimePeriod), but when I did it with a multiplier and saw that the increments don't go out of the channel, so I decided to remove it.

I see. Is that on the minutes?

 
Alexander_K2:

Once again - using ANY moving averages does not reduce the process to an Ornstein-Uhlenbeck process.

Why can't you use moving averages?

Here is a screenshot from a post on indicator using this theory.

Forum on trading, automated trading systems and strategy testing

From Theory to Practice

khorosh, 2018.07.07 21:33

There are no penties for June 14. On M15 it is:


Here's my screenshot, used moving average (increments), vertical lines show matching signals, no filter screening out false signals.


 
Alexander_K2:

I see. Is it on the minutes?


Yes.

 
Alexander_K2:
The only person with the lowest level of education who can make a kolhozny speech here is bas.


Not true. I'm the least educated person here, so it's hard for me to digest everything that's written here; I don't understand 99 percent of it at all.


It's like if UFO already gave me the blueprints for a flying saucer, but I keep riding around in a horse-drawn cart.
1...431432433434435436437438439440441442443444445...1981
New comment