But the return on risk is huge for buy-and-hold vs. a system that can enter the market a few minutes per day exit. If you look at VaR you'll see that the automated system (assuming the numbers in historical testing hold true in live trading) here can be leveraged to a much greater degree than buy & hold.
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
Frosty's auto-trading bot goes live with REAL money
- Thread starter frostengine
- Start date
frost
i imagine your situation and desperation on the whole thing. i want
to offer the help of my team, which is four people trading several
million in several automated systems. i would start with looking at
your backtested curve and define at which likelihood you are
experiencing what is currently happening. and draw a conclusion
out of that likelihood.
then i would try to identify what went "wrong", which most likely
leads to identifying an overfit. question remains why papertrading
did not bring that up.
then i would try to assisst you with developing a new strategy
maybe based on preStudies of the market you choose. i do not
want to know anything from what you currently trade or have in
your pipeline to make that clear.
i offer this because i think you do a serious effort and need help.
no interest apart of that on my side.
peace.
i imagine your situation and desperation on the whole thing. i want
to offer the help of my team, which is four people trading several
million in several automated systems. i would start with looking at
your backtested curve and define at which likelihood you are
experiencing what is currently happening. and draw a conclusion
out of that likelihood.
then i would try to identify what went "wrong", which most likely
leads to identifying an overfit. question remains why papertrading
did not bring that up.
then i would try to assisst you with developing a new strategy
maybe based on preStudies of the market you choose. i do not
want to know anything from what you currently trade or have in
your pipeline to make that clear.
i offer this because i think you do a serious effort and need help.
no interest apart of that on my side.
peace.
Quote from frostengine:
dchang0,
Good post..... makes a lot of sense, and you are probably right that keeping this play by play everyday is probably going to hurt me in the long run....
It may be a bad time to bring this up... but I may have a strategy that is FAR FAR more accurate in terms of risk/reward and win%.. only downside is it suffers from what my method #2 does... that is I dont have enough data to backtest on except what I have captured over past 4 months..
Frost,
I am a newb at this: I my partners and I have been trading full time for several years with success. About a year ago we got together with a programmer and began to work on some trading bots.
We did all the back testing there is, and then papertraded for months with nothing but success. When we finally went live we blew $60K. The sad thing is that there was not much wrong with our origional system, we did what you seem to be doing now: experienced a significant draw down and began switiching systems.
I just found this site and I have been following your thread with interest because we are looking to get back into bot trading, and I wanted to see how other people faired from the beginning.
I don't have any better programming or testing advice than what you have already recieved, however, you have to remain objective and fully test one method before you go on to others.
Jumping from one method to the next/better looking thing may turn out well, or may put you in the situation I was in: down a crapload of money with no real data to go on because we switched systems 4 times in the middle of the testing.
Edward
Quote from man:
frost
i imagine your situation and desperation on the whole thing. i want
to offer the help of my team, which is four people trading several
million in several automated systems. i would start with looking at
your backtested curve and define at which likelihood you are
experiencing what is currently happening. and draw a conclusion
out of that likelihood.
then i would try to identify what went "wrong", which most likely
leads to identifying an overfit. question remains why papertrading
did not bring that up.
then i would try to assisst you with developing a new strategy
maybe based on preStudies of the market you choose. i do not
want to know anything from what you currently trade or have in
your pipeline to make that clear.
i offer this because i think you do a serious effort and need help.
no interest apart of that on my side.
peace.
id like to meet the programmer whos smart enough to create a working system but dumb enough to give it away / fall for the above. we are a cynical bunch by definition.
Frost, you're getting real knowledge now. It's not about your backtesting results.
Here's the lowdown on successful automated trading, which I've been doing successfully since 2002. You only need to know one thing.
*What's worth testing?*
It takes a very, very long time to figure that out. When you learn to ask the right questions, you simultaneously learn to interpret the answers. I've traded dozens of systems where performance declined as soon as I went live. Some of them I threw out, but others I stuck with, and some I even INCREASED capital in. I was able to do this because I knew what drove the systems' behavior.
I'm not going to give away too much, but I'll tell you this: when you look at code you've backtested, the more numbers in it, the less reliable it's going to be going forward. This is how you avoid overfitting. If you're using an x-period moving average, you're screwed. Likewise x-day highs, x-day RSI, stochs, you name it. You can get away with one or MAYBE two numeric parameters if you want stability. But again, you need a lot of knowledge / experience to know where those can go.
Last piece of info: systems decay. Markets change. Learn to figure out how far back to test something, and when your data set is no longer relevant. Right now I'd say if you're testing equity markets with data from pre-march 2003 you're hurting yourself.
You say that's just a bull market with low volatility? Yes, it is. And that's what we have right now. When it changes, use a different data set for testing. If you REALLY study, you'll know when it's changed.
Here's the lowdown on successful automated trading, which I've been doing successfully since 2002. You only need to know one thing.
*What's worth testing?*
It takes a very, very long time to figure that out. When you learn to ask the right questions, you simultaneously learn to interpret the answers. I've traded dozens of systems where performance declined as soon as I went live. Some of them I threw out, but others I stuck with, and some I even INCREASED capital in. I was able to do this because I knew what drove the systems' behavior.
I'm not going to give away too much, but I'll tell you this: when you look at code you've backtested, the more numbers in it, the less reliable it's going to be going forward. This is how you avoid overfitting. If you're using an x-period moving average, you're screwed. Likewise x-day highs, x-day RSI, stochs, you name it. You can get away with one or MAYBE two numeric parameters if you want stability. But again, you need a lot of knowledge / experience to know where those can go.
Last piece of info: systems decay. Markets change. Learn to figure out how far back to test something, and when your data set is no longer relevant. Right now I'd say if you're testing equity markets with data from pre-march 2003 you're hurting yourself.
You say that's just a bull market with low volatility? Yes, it is. And that's what we have right now. When it changes, use a different data set for testing. If you REALLY study, you'll know when it's changed.
i don't want him to give anything away and i do not intend to giveQuote from walterjennings:
id like to meet the programmer whos smart enough to create a working system but dumb enough to give it away / fall for the above. we are a cynical bunch by definition.
anything away for myself. nevertheless i thought we could
be of help for someone in his position. and maybe you leave it
to him to judge where my offer is located on the line along
generous - fair - greedy.
i don't find your post funny. no, actually i find it completely
useless. and it makes me already feeling a little stupid to
have made my offer in the first place. sometimes i am really
tired of children like yourself ...
y e s .Quote from basis:
Frost, you're getting real knowledge now. It's not about your backtesting results.
Here's the lowdown on successful automated trading, which I've been doing successfully since 2002. You only need to know one thing.
*What's worth testing?*
It takes a very, very long time to figure that out. When you learn to ask the right questions, you simultaneously learn to interpret the answers. I've traded dozens of systems where performance declined as soon as I went live. Some of them I threw out, but others I stuck with, and some I even INCREASED capital in. I was able to do this because I knew what drove the systems' behavior.
I'm not going to give away too much, but I'll tell you this: when you look at code you've backtested, the more numbers in it, the less reliable it's going to be going forward. This is how you avoid overfitting. If you're using an x-period moving average, you're screwed. Likewise x-day highs, x-day RSI, stochs, you name it. You can get away with one or MAYBE two numeric parameters if you want stability. But again, you need a lot of knowledge / experience to know where those can go.
Last piece of info: systems decay. Markets change. Learn to figure out how far back to test something, and when your data set is no longer relevant. Right now I'd say if you're testing equity markets with data from pre-march 2003 you're hurting yourself.
You say that's just a bull market with low volatility? Yes, it is. And that's what we have right now. When it changes, use a different data set for testing. If you REALLY study, you'll know when it's changed.
Quote from basis:
Frost, you're getting real knowledge now. It's not about your backtesting results.
Here's the lowdown on successful automated trading, which I've been doing successfully since 2002. You only need to know one thing.
*What's worth testing?*
It takes a very, very long time to figure that out. When you learn to ask the right questions, you simultaneously learn to interpret the answers. I've traded dozens of systems where performance declined as soon as I went live. Some of them I threw out, but others I stuck with, and some I even INCREASED capital in. I was able to do this because I knew what drove the systems' behavior.
I'm not going to give away too much, but I'll tell you this: when you look at code you've backtested, the more numbers in it, the less reliable it's going to be going forward. This is how you avoid overfitting. If you're using an x-period moving average, you're screwed. Likewise x-day highs, x-day RSI, stochs, you name it. You can get away with one or MAYBE two numeric parameters if you want stability. But again, you need a lot of knowledge / experience to know where those can go.
Last piece of info: systems decay. Markets change. Learn to figure out how far back to test something, and when your data set is no longer relevant. Right now I'd say if you're testing equity markets with data from pre-march 2003 you're hurting yourself.
You say that's just a bull market with low volatility? Yes, it is. And that's what we have right now. When it changes, use a different data set for testing. If you REALLY study, you'll know when it's changed.
you're last statement gave me an idea. it would be interesting to see if you could analyze the past week-month of data. and then filter the past 4 years worth of data to limit ranges which do not match the current market conditions. making your system to learn to trade against other trader's mindsets (bull vs bear etc). classifying current market conditions might be a bit tricky and might take some creativity. but it is definitely possible.
Quote from man:
i don't want him to give anything away and i do not intend to give
anything away for myself. nevertheless i thought we could
be of help for someone in his position. and maybe you leave it
to him to judge where my offer is located on the line along
generous - fair - greedy.
i don't find your post funny. no, actually i find it completely
useless. and it makes me already feeling a little stupid to
have made my offer in the first place. sometimes i am really
tired of children like yourself ...
I didn't mean to be insulting. Just wanted to point out how unprofessional and shady approaching someone on a public forum thread is with an 'offer' like that. Maybe try emailing next time. IMHO someone who manages 'several automated systems trading millions' wouldn't do something like that. So obviously I and probably most people reading this thread thought 'scam' as soon as they read the first sentence in your original post.
I've been reading this thread from the beginning and I really hope to see frost's system work well. That being the case I always throw in my two cents when I think it may be of benefit to reaching that goal. So please don't try to tell me to keep my mouth shut. This is obviously only my opinion, frost etc can draw their own conclusions on your offer.
walter,
I like your idea of trying to match market conditions..... would be great If I could find something that with any degree of acuracy could tell the market conditions... Thus far no ideas really pop out at me.... I will have to think about it for a while...
Bot lost -$99... this is becomming very frustrating
I like your idea of trying to match market conditions..... would be great If I could find something that with any degree of acuracy could tell the market conditions... Thus far no ideas really pop out at me.... I will have to think about it for a while...
Bot lost -$99... this is becomming very frustrating