Has anybody ever had a consistently profitable algorithm automated?

P

pmg

Hi guys,

I have been studying now about rule based automated trading for about 1 year aside to my job. I consider myself somewhat familiar with algorithm design and programming.

I am now using tradestation and have been designing all kinds of simple trading systems in Easylanguage, which are all based on indicators, some very simple, others more complex. I have tried trend following, break-outs and purely indicator based systems.

However, with input optimization it was easy to have positive backtest results, but I never managed to come up with an algorithm, that can consistently show even small profits on more than 1 symbol.

Now I would like to ask you, if there are any people here, who made it happen and have such algorithms running. I am not asking to reveal anything, just let me know, if it's possible or not.

My expectation is that it should be possible, or else I would probably say that the market is really random, which I do not beblieve so far. Also since most trades today are automated trades based on Algos, so there must be working Algos.

Also I would be happy, if anybody knows some goo literature about rule based/algo trading.

Looking forward to your input!
 
Quote from pmg:


Also I would be happy, if anybody knows some goo literature about rule based/algo trading.
More...

yes, definitely know some goo literature. Unfortunately not much good though.
 
Quote from pmg:

Hi guys,

I have been studying now about rule based automated trading for about 1 year aside to my job. I consider myself somewhat familiar with algorithm design and programming.

I am now using tradestation and have been designing all kinds of simple trading systems in Easylanguage, which are all based on indicators, some very simple, others more complex. I have tried trend following, break-outs and purely indicator based systems.

However, with input optimization it was easy to have positive backtest results, but I never managed to come up with an algorithm, that can consistently show even small profits on more than 1 symbol.

Now I would like to ask you, if there are any people here, who made it happen and have such algorithms running. I am not asking to reveal anything, just let me know, if it's possible or not.

My expectation is that it should be possible, or else I would probably say that the market is really random, which I do not beblieve so far. Also since most trades today are automated trades based on Algos, so there must be working Algos.

Also I would be happy, if anybody knows some goo literature about rule based/algo trading.

Looking forward to your input!
More...

You mean like this?

Take heart, child. It only requires a half a decade of random research, but start with active trader magazine and see if there aren't shell scripts for your work. They can be transferrable if you have a consistent algorithm.

Or you can hire a Topology Math PhD quant and he should be able to code something within two weeks that also works on everything in addition to the ones you already have that are universally applicable.
 

Attachments

Quote from bwolinsky:

Wrong, whred, the first is only for ES and the second is only for Gold Comex.

Neither of those are universal algorithms.
More...

You are correct, unless he uses the crossover and breakout methods on other securities. Such as Copper and Crude Oil.
 
Quote from pmg:

Hi guys,

I have been studying now about rule based automated trading for about 1 year aside to my job. I consider myself somewhat familiar with algorithm design and programming.

I am now using tradestation and have been designing all kinds of simple trading systems in Easylanguage, which are all based on indicators, some very simple, others more complex. I have tried trend following, break-outs and purely indicator based systems.

However, with input optimization it was easy to have positive backtest results, but I never managed to come up with an algorithm, that can consistently show even small profits on more than 1 symbol.

Now I would like to ask you, if there are any people here, who made it happen and have such algorithms running. I am not asking to reveal anything, just let me know, if it's possible or not.

My expectation is that it should be possible, or else I would probably say that the market is really random, which I do not beblieve so far. Also since most trades today are automated trades based on Algos, so there must be working Algos.

Also I would be happy, if anybody knows some goo literature about rule based/algo trading.

Looking forward to your input!
More...

I am pretty sure it's possible now. Previously I said I don't believe in finding the holy grail. At that time I had found algorithms that worked only within certain regimes (e.g., high volatility). However, in the last few months, I have actually found a few algorithms that seem consistent for several years. It is not as profitable as the ones that worked conditionally but it is much more reliable in the long run. I am still tweaking with them because I am pretty sure I can improve them further.

PS: I cannot stress the importance of making sure your back test results are not overestimated due to poor modeling of slippage, commission, and clock skew. It is very easy to fool oneself into thinking that you found an algorithm but once you go live, it bears no resemblance to your back test. If you have taken all the precautions and verified that forward tests match up very close to your back tests then and only then are you able to find algorithms that will work.
 
Quote from ssrrkk:

I am pretty sure it's possible now. Previously I said I don't believe in finding the holy grail. At that time I had found algorithms that worked only within certain regimes (e.g., high volatility). However, in the last few months, I have actually found a few algorithms that seem consistent for several years. It is not as profitable as the ones that worked conditionally but it is much more reliable in the long run. I am still tweaking with them because I am pretty sure I can improve them further.

PS: I cannot stress the importance of making sure your back test results are not overestimated due to poor modeling of slippage, commission, and clock skew. It is very easy to fool oneself into thinking that you found an algorithm but once you go live, it bears no resemblance to your back test. If you have taken all the precautions and verified that forward tests match up very close to your back tests then and only then are you able to find algorithms that will work.
More...

If the algorithm's adaptive, and many aren't, then adding 2 ticks of slippage should give the model a reality check. Adaptive does not mean that in the process of optimizing the results get better or seem to "wait" for more precisely defined quantitative patterns. Adaptive should only be used to referr to genetically optimizable algorithms, and those robots should be able to look at the data and logically trade them. Their parameters may change but what the op is asking for is a universally applicable method, and I have one. That I have one, the most important thing I've found is how to properly optimize your parameters, and having up to 7 inputs with 5 primary parameters of those inputs and 2 basic stops and targets you should be able to build bots that are profitable, and you'll know from your optimizations whether they are or not. The addition of 2 ticks of slippage removes unrealistically high frequency algorithms for ones more consistent and more profitable with slightly lower win percentages because they're reaching for more profit, and this exposes them to tail risk if they're holding any longer than 14 days.
 
Impressive, but how do you deal with 30% drawdowns? Especially when that means watching 150 G go bye-bye? Is that your actual ?

I'm an interested party to this discussion for sure. I'm about to go live with mine.

I know the answer is yes, but I'm at the point now that it doesn't matter until I do it myself. I expect to. And without 30% drawdowns I hope !


Quote from bwolinsky:

You mean like this?

Take heart, child. It only requires a half a decade of random research, but start with active trader magazine and see if there aren't shell scripts for your work. They can be transferrable if you have a consistent algorithm.

Or you can hire a Topology Math PhD quant and he should be able to code something within two weeks that also works on everything in addition to the ones you already have that are universally applicable.
More...
 
You must log in or register to reply here.
Back
Top