Fading a system....

Quote from TriPack:

Mr Subliminal, how do you determine the bid/ask in ES from the tick data?
More...
Let's take today's (November 17) CME data file, gx031117.iom. The format of this file can be found on the CME website :

http://www.cme.com/dta/hist/ftp_gtimeandsales3098.html

Suffice it to say that the 5 digits immediately preceding the last E represent the quantity traded (4 for first line, 115 for second...). I have programs which first transform the file from this :
Code: 
ES  0312F       0830000104325    06927200031117ES22    03111700004E  620031200N
ES  0312F       0830000104325    06927500031117ES22    03111700115E  620031200N
ES  0312F       0830000104300    06927600031117ES22    03111700001E  620031200N
ES  0312F       0830000104300    06927700031117ES22    03111700001E  620031200N
ES  0312F       0830000104325    06928000031117ES22    03111700001E  620031200N
ES  0312F       0830000104325    06928300031117ES22    03111700050E  620031200N
ES  0312F       0830000104300    06928600031117ES22    03111700004E  620031200N
ES  0312F       0830000104300    06928900031117ES22    03111700001E  620031200N
ES  0312F       0830010104325    06930000031117ES22    03111700002E  620031200N
ES  0312F       0830010104325    06930300031117ES22    03111700001E  620031200N
ES  0312F       0830010104325    06930600031117ES22    03111700005E  620031200N
ES  0312F       0830010104325    06931100031117ES22    03111700004E  620031200N
ES  0312F       0830010104325    06931600031117ES22    03111700010E  620031200N
ES  0312F       0830010104300    06931900031117ES22    03111700007E  620031200N
ES  0312F       0830020104325    06933000031117ES22    03111700007E  620031200N
ES  0312F       0830020104325    06933500031117ES22    03111700005E  620031200N
ES  0312F       0830020104325    06933800031117ES22    03111700001E  620031200N
ES  0312F       0830020104300    06933900031117ES22    03111700004E  620031200N
ES  0312F       0830020104300    06934000031117ES22    03111700008E  620031200N
ES  0312F       0830020104325    06934500031117ES22    03111700001E  620031200N
ES  0312F       0830020104325    06934800031117ES22    03111700067E  620031200N
ES  0312F       0830020104325    06935100031117ES22    03111700016E  620031200N
ES  0312F       0830030104300    06935400031117ES22    03111700080E  620031200N
ES  0312F       0830030104325    06935700031117ES22    03111700010E  620031200N
ES  0312F       0830030104300    06936000031117ES22    03111700020E  620031200N
ES  0312F       0830030104325    06936500031117ES22    03111700001E  620031200N
ES  0312F       0830030104325    06936800031117ES22    03111700002E  620031200N
ES  0312F       0830030104325    06937100031117ES22    03111700010E  620031200N
ES  0312F       0830040104325    06937600031117ES22    03111700003E  620031200N
.....
.....
.....
into this :
Code: 
083000  0104325     119
083000  0104300       2
083000  0104325      51
083000  0104300       5
083001  0104325      22
083001  0104300       7
083002  0104325      13
083002  0104300      12
083002  0104325      84
083003  0104300      80
083003  0104325      10
083003  0104300      20
083003  0104325      16
083004  0104300       3
083004  0104325      10
083004  0104300      22
083007  0104325       2
083007  0104300       7
083008  0104325     175
083008  0104300       5
083009  0104325      72
083009  0104350       2
083009  0104325       1
083010  0104350       9
083010  0104325       1
083010  0104350      28
083011  0104325      53
083012  0104350       9
083013  0104325     227
083016  0104300       1
083016  0104325      33
083017  0104300       2
083017  0104325       6
083017  0104300      15
083018  0104325      25
083018  0104300       4
083018  0104325     250
083018  0104300      12
.....
.....
.....
and finally into this :
Code: 
                            Bid       Ask
083000  0104325     119   1043.00   1043.25
083000  0104300       2   1043.00   1043.25
083000  0104325      51   1043.00   1043.25
083000  0104300       5   1043.00   1043.25
083001  0104325      22   1043.00   1043.25
083001  0104300       7   1043.00   1043.25
083002  0104325      13   1043.00   1043.25
083002  0104300      12   1043.00   1043.25
083002  0104325      84   1043.00   1043.25
083003  0104300      80   1043.00   1043.25
083003  0104325      10   1043.00   1043.25
083003  0104300      20   1043.00   1043.25
083003  0104325      16   1043.00   1043.25
083004  0104300       3   1043.00   1043.25
083004  0104325      10   1043.00   1043.25
083004  0104300      22   1043.00   1043.25
083007  0104325       2   1043.00   1043.25
083007  0104300       7   1043.00   1043.25
083008  0104325     175   1043.00   1043.25
083008  0104300       5   1043.00   1043.25
083009  0104325      72   1043.00   1043.25
083009  0104350       2   1043.25   1043.50
083009  0104325       1   1043.25   1043.50
083010  0104350       9   1043.25   1043.50
083010  0104325       1   1043.25   1043.50
083010  0104350      28   1043.25   1043.50
083011  0104325      53   1043.25   1043.50
083012  0104350       9   1043.25   1043.50
083013  0104325     227   1043.25   1043.50
083016  0104300       1   1043.00   1043.25
083016  0104325      33   1043.00   1043.25
083017  0104300       2   1043.00   1043.25
083017  0104325       6   1043.00   1043.25
083017  0104300      15   1043.00   1043.25
083018  0104325      25   1043.00   1043.25
083018  0104300       4   1043.00   1043.25
083018  0104325     250   1043.00   1043.25
083018  0104300      12   1043.00   1043.25
.....
.....
.....
 
Quote from amigasearch:

Ok. A burning question thats been on my mind. Please help!
I have been creating and designing "systems". Just simple rules, over three months data. Looking at different results, I will decide which one to trade.

Some systems preform better then others - positive equity curve, and such (passes all by the book rules).

But, there are some systems, which look absolutely horrible. They produce an equity curve that would wipe Bill Gates out in a month or two.

So, you all know where I am going (if you read the header of the thread) why not reverse the signals of this system, and trade it from the other way?

I am in the precept that there are strategy traders out there who pour over data and methods until they find the right one - is there a time that they just decide on the opposite of the worst one they have ever seen?

This makes logical sense, and I am wondering if this is valid in any way, and worth persuing.


Please help out. Thanks.
More...

The systems I have tested paradoxically had nearly the same win/loss as it's mirror.

Most of the time, your system will not work because you are on the wrong time frame, or your stops are too tight.

Regards
Oddi
 
Mr. Subliminal,

To get to the last result: turning tick prices into bid/ask I presume you make the assumption that if it is a downtick, then it occurs on the bid, and if it is an uptick, then it occurs at the ask. I wonder how accurate this method is in reality. Maybe 70-80%? It may be that the 20% where the bid/ask doesn't follow this rule, that they cancel themselves out so this method gives the accurate bid/ask picture.

Thanks for the posting Mr. Sub, very enlightening.
 
I find that this data uniquely identifies the bid/ask for my purposes. Where there is room for ambiguity, an A (ask) or B (bid) appears in column 30 (Ask/Bid Indicator) of the CME file - during RTH, there may be 3 or 4 occurrences of this per 1,000 records.

Why do you say 70-80%?
 
In response to the question of the why - yes, I understand the why (of the opposite). After clearing my head, it is me being stubborn (how can my theory not work?? How can it be opposite of what i was Convinced was true?...).

As to transaction costs and where i am in and out - something I need to work on. Current system trades off 1 minute bars, and I only have 4 months data. Now, my exits entries are based on bar open and close. The only "advanced" way I have to determine types of slippage is I determine by my Intended exit / entry price, and factor this in with what I actually got from the close for exit, open for entry.
The difference, is a positve value (means the testing is giving me points ) or a negative value (means it is taking points from me). I consider this splippage, and so far my results are not affected greatly....
An example: Signal to sell position is generated at 1000.25. Since I get out at close, if close is 1000.00, then 1000.25 - 1000.00 = .25, if I was long, I am missing .25, short, getting it.
I understand markets move fast, and I need to create a bid / ask method. This is my next plan. (I do all coding, so this is on my list!).

Finally, I reversed the buy / sell signals, and results are positve ( at least I think so!) 60 percent winners, 40 percent losers, avg trade at $12.67 (after commsisions), 450 trades over 5 months (1 contract).

Oh - and as a post was mentioned early about reversing and getting similar results - yes - this seems to be the case, except this method trades more ( i removed an indicator from the results also ) and takes less risk.
 
First, three months is way too short a timeframe to come up with definitive data. Some pretty decent systems can be flat or lose equity for months.

Second, you need to understand why the original system was no good. Did it get chopped up in a range-bound market? Did it exit winners too soon? Did a few big losers kill it? Were the stops too tight? A lot of things can undermine performance. You need to look at all aspects carefully before you just decide to flip the buy/sell rules around.

A good system seeks to capitalize on some observed phenomenon or anamoly. Although it's conceivable that you just totally misunderstood the significance of the market action, a more likely explanation is that you are seeing something that is not there, in other words, it has little statistical significance. In such a case reversing the system will not help you.
 
Agreed. Three months is WAY to short. My data provider only gives me three months. Expect a new thread from me to ask anyone if they can sell to me minute data over 1 year or more period, in a format I can transfer to meet my applications format (should be easy - I read text file - someday, if it works, i will sql it).
After I have it looking at +1 year data and more, I will take it up a notch (lots of programming to do) and test it real time simulation a minimum of 6 months.

As for advice on conditions and such, thanks for that heads up. This is all known to me, and I am making sure to implement this thinking.

All my questions, such as this one (is reversing valid) is the newbie to systems testing in me being inquisitive. I always wish to know others experience, and this was helpful....
The most important thing I learned here, on this thread, was from Hypo, and to sum it up, his description of his systems bizarrness, (having system prove a hunch is not the only way), and also, positioning my entry exits in best place to take advantage of price. This is profound concept to me, and has opened many doors, as most of my systems thinking was based on "ok, what theory do I have to see works?.". Thanks to all - although your input has lead me to more programming I need to do to put these thoughts in code!
 
...take care in what conclusions you draw from posters you do not know either personally or from long-term corrspondence. My systems are bizarre because I am bizarre. It is not necessarily a criterion for successful system development in general.

Also, with all due respect to AAA, I assert without offering proof that systems do not need to exploit an observed phenomenon or anomaly. I have one system whose entry often is a surprise to me because I cannot "see" the setup. I no longer attempt to understand the "why" of something that works, just the "how".

Re your need to estimate slippage, I have found that I get approximately theoretical slippage (one tick) for a round trip on NQ. For some peculiar reason, I actually get about 0.8 tick.

Re your net per trade, I don't recall that you identified the issue you tested on. If it's NQ, IMO $12.67 is too marginal to trade. If ES, WAY too marginal.
 
Quote from Mr Subliminal:

I find that this data uniquely identifies the bid/ask for my purposes. Where there is room for ambiguity, an A (ask) or B (bid) appears in column 30 (Ask/Bid Indicator) of the CME file - during RTH, there may be 3 or 4 occurrences of this per 1,000 records.

Why do you say 70-80%?
More...

Just a guess - not based on any research. I didn't realize the simple algorithm you described was that solid. This may change the way I do things... Thanks again.
 
Quote from TriPack:

Just a guess - not based on any research. I didn't realize the simple algorithm you described was that solid. This may change the way I do things... Thanks again.
More...
Before you thank me and "change the way" you "do things", be forewarned that, while it is simple and far from earth-shattering, this is something I didn't find in a book but which I developed for my own use. Having said all that, you're very welcome.
 
You must log in or register to reply here.
Back
Top