Analysis Series request: ExitBar
Author: streak
Creation Date: 3/11/2012 3:31 AM
profile picture

streak

#1

Just upgraded to 6.3 Like & appreciate it.

Suggest adding 2 options to this tool, please.

ExitBar-1

ExitBar


Thanks Again

profile picture

Eugene

#2
Sorry, this option was dropped for several reasons and is not considered.
profile picture

streak

#3

Be interesting to know the reasons? Is it because most analysis is being done on optimising entry points?

Would have thought it a reasonable and very useful addition(?)
profile picture

Eugene

#4
The reasons why the choice of exit bar was ultimately dumped:

1. First and foremost, analyzing the indicator values at the time of exit is less useful. This is because the Analysis Series filter removes the trades in its filter analysis, but in reality you couldn't undo a trade that was already created.
2. Purposes of simplification.
3. Sort of a programming problem (it's been several years so I don't remember precisely)
profile picture

streak

#5

Thanks for responding Eugene.

re: 1 Why so? Why could trades not be more profitable on analysed exits?

profile picture

Eugene

#6
Because you can't undo a trade, so it's purposeless and confusing. That's why we decided to concentrate on entries.
profile picture

Cone

#7
streak, let us know your workflow on how you'd use the Exit Bar information. For example, here's how I use Analysis Series:

First, I develop a profitable strategy the merits further investigation. To find ways to maximize profit, generally I try to filter (i.e., keep from entering) the "worst trades". To do that, I'd create an indicator so that Analysis Series can get a handle on it, but I'd still run the strategy without applying it. In this way, you can determine the correlation between the indicator values and trade profit for all the trades in the basic strategy. If you find a high correlation in a range of values for the losers, then using the indicator to filter the entries might help the basic strategy.

The question is, how would use the knowledge of the correlation of an indicator with trade profit on the Exit Bar of the basic strategy?
profile picture

streak

#8

Eugene, Cone

Firstly my apologies for having you explain it. In short, say one takes one indicator and makes a series of peeking offsets of it, like indicatorX << 1; indicatorX << 2; indicatorX <<3; then would not Analysis Series (with bar of or before exit) be able to make hay on this? Eg with pos results one might add a filtering exit requiring some short term indicator to turn, or conversely require something like TrueRange to make an extended range before exiting.

... getting into Wld4's Evalulator tool?

I mess a bit with peeking. It all came about I guess from Larry Williams workshops and testing things like "Bailout on first (or 2nd and so on) profitable opening". There was from his workshops a function that could work 'outside' TradeStation called NBopen() and with only this much peeking one could test (and validly implement) such an exit. I can understand the mess that many new to trading/programming could make of peeking so I won't contest anything about it here.

However, about 10 years ago I started working on exits. Then came WealthLab with portfolio testing, much control in position management and so on. With 6.3 just recoded what I originally had in t.s. and then wld4. Its crappy in some ways but the exit stuff is, to me very interesting. There are 20 odd exits, mainly short term pattern stuff that I send to an array, sort at the end of each bar to spit out the best, and for the last bar was sending to file where I'd be alerted via excel ddes on the day.

Anyway don't mean to suggest I'm extremely mechanical at present.



profile picture

streak

#9
Looking at another angle on this, wondered if reversing a Bars data then testing, what would otherwise be a long system, as a short system ie the real long exits become tested as short entries. whoa how to go fuzzy ay

Anyway copied some code off the website here to make a news Bars object with data reversed (Bars.Count - 1 becomes zero etc) which started off ok until attempting to apply SetContext() which, when un-commented in either of the statements in the code below, a runtime error is thrown "Object not set to an instance of an object."

Any ideas where I've lost it, please?

CODE:
Please log in to see this code.

profile picture

Eugene

#10
I remember this having been discussed before, but the moral is that one can't SetContext to a synthetic symbol (of sorts) -- the symbol must exist in a real DataSet.
profile picture

streak

#11

OK, thanks Eugene