Frosty's trading bot goes live part 2
- Thread starter frostengine
- Start date
Frosty,
Here's a few assumptions from the thread replies you've given.
1. I'm guessing that your execution platform is working as it is supposed to. Due to the hypothetical (test) results closely resembling the live results. Also, this can be said with the testing platform itself.
So the problem lies in the model itself. Correct?
2. You have a specific concept, which you assume to work. This is based on your comment that you tweaked the previous trading model you used.
...
So my questions are:
1. Without giving out the specific concept and actual tweaks you've done, please explain how the models were developed and process of how you derived to the current model.
(eg. You find a specific concept that seems to work. You tested it and it get a upward equity curve. The risk-metrics for the concept is favorable in XXXX... XXXX... XXX.... reasons.
Though, the initial test is favorable, you find that YYYY... YYY... YY... flaws. So you take that concept and try to filter them out to create a more favorable curve.)
2. Also, please explain how you assessed the trading models and the results. Please post the metrics and test processes you've done.
(eg. You had the historical tester pull out the following metrics with the results: %Profitability, MaxDD, Sharpe, ZZZ... ZZ.... Z.... Then you run a walk forward and get these results.
Then you run a monte-carlo... And conclude that is the system is So-and-so.
And you run AAAAA... tests and conclude BBBB... based on those results...
And etc. etc. )
Frost, let's start there...
Just like any thread in ET, the original poster never provides enough information to provide a constructive and valuable advice. (He's already mentioned he's a newbie so it's acceptable...)
Though, it's unbelievable to see how posters try to give advice. Why can't people recognize that there's no point of posting anything in here without the required information?
Most systems don't work. Most systems that tested well don't work. Though, there's usually a reason behind why it stopped working and it's usually very simple. (Definition of what "works" is a different story... robustness, duration, etc. etc. Off topic.)
Issue now is to identify whether there is a "problem" (obviously there is). If there is one, then it's to find what it is so that the OP doesn't make the same mistake again.
...
One obvious problem is that the OP is adding "discretion" by switching his system on/off. The On/Off criteria should be based on a systematic approach (or within the range of what's been researched). The obvious reason causing this problem is that he hasn't set a contingency before running it live. Even if he did set up contingency criterias, it's not being followed...
Despite of all the critisism, Jimmy Jam is correct to a certain point. The OP has to learn how to trade and build himself up to become a trader. You may not be entering all your orders manually, but ATS is trading. Here's a hint to the picky people: What is the first thing you need to make it as a trader?
Discipline.
Here's a few assumptions from the thread replies you've given.
1. I'm guessing that your execution platform is working as it is supposed to. Due to the hypothetical (test) results closely resembling the live results. Also, this can be said with the testing platform itself.
So the problem lies in the model itself. Correct?
2. You have a specific concept, which you assume to work. This is based on your comment that you tweaked the previous trading model you used.
...
So my questions are:
1. Without giving out the specific concept and actual tweaks you've done, please explain how the models were developed and process of how you derived to the current model.
(eg. You find a specific concept that seems to work. You tested it and it get a upward equity curve. The risk-metrics for the concept is favorable in XXXX... XXXX... XXX.... reasons.
Though, the initial test is favorable, you find that YYYY... YYY... YY... flaws. So you take that concept and try to filter them out to create a more favorable curve.)
2. Also, please explain how you assessed the trading models and the results. Please post the metrics and test processes you've done.
(eg. You had the historical tester pull out the following metrics with the results: %Profitability, MaxDD, Sharpe, ZZZ... ZZ.... Z.... Then you run a walk forward and get these results.
Then you run a monte-carlo... And conclude that is the system is So-and-so.
And you run AAAAA... tests and conclude BBBB... based on those results...
And etc. etc. )
Frost, let's start there...
Just like any thread in ET, the original poster never provides enough information to provide a constructive and valuable advice. (He's already mentioned he's a newbie so it's acceptable...)
Though, it's unbelievable to see how posters try to give advice. Why can't people recognize that there's no point of posting anything in here without the required information?
Most systems don't work. Most systems that tested well don't work. Though, there's usually a reason behind why it stopped working and it's usually very simple. (Definition of what "works" is a different story... robustness, duration, etc. etc. Off topic.)
Issue now is to identify whether there is a "problem" (obviously there is). If there is one, then it's to find what it is so that the OP doesn't make the same mistake again.
...
One obvious problem is that the OP is adding "discretion" by switching his system on/off. The On/Off criteria should be based on a systematic approach (or within the range of what's been researched). The obvious reason causing this problem is that he hasn't set a contingency before running it live. Even if he did set up contingency criterias, it's not being followed...
Despite of all the critisism, Jimmy Jam is correct to a certain point. The OP has to learn how to trade and build himself up to become a trader. You may not be entering all your orders manually, but ATS is trading. Here's a hint to the picky people: What is the first thing you need to make it as a trader?
Discipline.
an ats 'custodian' needs discipline in order to maintain a profitable running system in the same way a McPloyee needs discipline not to piss in the deep fryer.
we might be on the same page here.
So you're saying Frost (and all the other newbs in ET) is worth less than a McPloyee?
:eek: :eek: :eek:
Quote from TSGannGalt:
So you're saying Frost (and all the other newbs in ET) is worth less than a McPloyee?
:eek: :eek: :eek:
well i cant speak for frost. but i know when i dont feel like doing any r&d on a given day, i spend it watching reruns of star trek series and judge judy. thus i am worth far less than a McPloyee => i cant trade by hand AND my freedom fries taste funny.
How funny? Does it taste like Fanta Orange or Mello Yellow?
I think JJ is getting unnecessarily trashed in this thread. He can work with the lack of info provided by frosty and come to a conclusion, that indicates intuition, and essentially, I don't have any trouble following JJ's arguments.
Frosty needs to get a system he is comfortable with. Frosty insists that there is nothing, AT ALL, wrong with his testing but he can't live with the drawdowns. Not too hard to conclude that he is comfortable with the backtested results and not comfortable with the real results and apparently the real results are within the parameters of the backtested results. That means that the only difference is that it is live and Frosty can't handle the results live that he did in the backtest. Obvious conclusion is that Frosty has to get another system. He should take whatever he learned from this last episode and design a new system that he will be comfortable with. He should be able to determine his comfort level from what he learned recently.
Frosty needs to get a system he is comfortable with. Frosty insists that there is nothing, AT ALL, wrong with his testing but he can't live with the drawdowns. Not too hard to conclude that he is comfortable with the backtested results and not comfortable with the real results and apparently the real results are within the parameters of the backtested results. That means that the only difference is that it is live and Frosty can't handle the results live that he did in the backtest. Obvious conclusion is that Frosty has to get another system. He should take whatever he learned from this last episode and design a new system that he will be comfortable with. He should be able to determine his comfort level from what he learned recently.
Actually I thought someone else pointed out that this drawdown is either equal to worse than the worst drawdown during the entire backtested period (if that is not correct please let me know) so it is not simply the case that he can't handle an expected drawdown, its that the system is experiencing a more severe drawdown "out of the gate" then has ever been seen before in backtesting.
If that is true then either the system was over-optimized or Frosty just has terrible luck (timing) and the system will overcome this period and go on to have an equity curve that matches the backtested results.
Wow, looks like a lot of comments. I'll try to reread through this and see if there are any I can address. I did not post Tuesday, but the bot would have made $70 that day.
Today I decided to turn it live again. Worked out fairly well, nothing crazy, but had a +$220.80 day.
Today +$220.80
Week to Date: +220.80
Since going live: -$3158.8
Total days: 11
Avg per day: -$287.16
Up Days: 5
Down Days: 6
Today I decided to turn it live again. Worked out fairly well, nothing crazy, but had a +$220.80 day.
Today +$220.80
Week to Date: +220.80
Since going live: -$3158.8
Total days: 11
Avg per day: -$287.16
Up Days: 5
Down Days: 6