michael_bower
Newbie
- Messages
- 0
- Likes
- 0
Past performance indicative of future results?
Past performance indicative of future results?
- Thread starter michael_bower
- Start date
- Watchers 3
Hey Michael,
I am sure there are more experienced traders then myself that can reply here, but awaiting a more complete reply I will post mine. Just for the records I have worked as derivatives broker and professional trader, and I have spent abt 2.5 years working on automated strategies.
The problem or one of the problems of algorithmic trading is that it relies on a function with variables. A change in the value of these variables influence the results of the strategy. Different combination of variables value give different results.
The trader is then looking for the "optimal" results ( we can discuss then how we define optimal , in terms of draw-down, profit factor, av.win/av loss, or solidity in general), but the greater the number of variables in the equation the bigger the number of combination the could have, as the function become less and less "structural" compared to the market.
Looking for the best values to use the automated strategy the trader would "optimize" the equation ( exhaustive or genetic, with or without walkforward) finding the values that provide the best performance. Here the first "tricks" when you look at an automated strategy, even before looking at performance reports, you want to know if it takes into account commission cost , slippage and what is % of risk used and max numbers of contracts held per trade ( and consec trades ).
This will avoid bad surprises as , looking at a nice equity curve finding out that all the profits where made in just 3 trades out of 100, or finding out that it comes out from a strategy that makes 3 ticks profit each trade ( as that wont make money in real market). More important is the sample. The bigger the sample the lesser the risk to incur into "curve fitting". The optimizer will show you the best values for your equation over that sample, which are, as per your question Michael not necessarily the best for the past sample or the future. By my experience less than 400 trades in a sample is kind of meaningless, with high risk of curve fitting. A 1000 trades starts to be ok.
Two other means I use to understand if an automated strategy is solid are:
1) Change the time frame and symbol the strategy is loaded on an see if it remains profitable
2) observe the optimization process, I want a large area of profitable values. If the strategy is profitable only only 10 tests out of a 1000 it is quite weak.
Past performance indicative of future results? I would say that is not carved in stone but at least you have a good statistical approach . In the end that is a bit the basis of trading techniques and of empirical science in general : we cannot be sure the sun will rise tomorrow but we can take a good guess.
As for automated strategies a good genetic optimization over a large sample would help put you on the right track, followed then by a walk forward optimization which will test past optimization results forward in time. The idea is that recent segment test are better to evaluate the equation parameters as for future possible performances. That been said, unless the strategy is veeeeery solid, you will need it to re-optimized it every X time.
All this to say, that before buying TrainingTraders software I would like to take a look at those data at least. Not because I expect necessarily the strategy to perform equally but to know that at least it has been tested properly.
I just hope these guys would provide some info as I am interested .
In the free webinars on their webiste Mr. Baghdady shows quickly a performance report for the strategy on crude, but with only abt 70 traders and I have immediately noticed the equity curve had a big plateau and made most of the profit in the last three trades. As for my experience when I see that in a graph I wonder " if I were trading this live , would I be sitting for 40 trades without making a penny and let the strategy do its thing"?
I hope they reply.
p.s.
If some of you guys are interested and if by miracle I learn how to do it I could post a performance report of an automated strategy with a 1000 trades, to show what I am talking abt here.
Did I somehow answered the question? A bit too much?
I am sure there are more experienced traders then myself that can reply here, but awaiting a more complete reply I will post mine. Just for the records I have worked as derivatives broker and professional trader, and I have spent abt 2.5 years working on automated strategies.
The problem or one of the problems of algorithmic trading is that it relies on a function with variables. A change in the value of these variables influence the results of the strategy. Different combination of variables value give different results.
The trader is then looking for the "optimal" results ( we can discuss then how we define optimal , in terms of draw-down, profit factor, av.win/av loss, or solidity in general), but the greater the number of variables in the equation the bigger the number of combination the could have, as the function become less and less "structural" compared to the market.
Looking for the best values to use the automated strategy the trader would "optimize" the equation ( exhaustive or genetic, with or without walkforward) finding the values that provide the best performance. Here the first "tricks" when you look at an automated strategy, even before looking at performance reports, you want to know if it takes into account commission cost , slippage and what is % of risk used and max numbers of contracts held per trade ( and consec trades ).
This will avoid bad surprises as , looking at a nice equity curve finding out that all the profits where made in just 3 trades out of 100, or finding out that it comes out from a strategy that makes 3 ticks profit each trade ( as that wont make money in real market). More important is the sample. The bigger the sample the lesser the risk to incur into "curve fitting". The optimizer will show you the best values for your equation over that sample, which are, as per your question Michael not necessarily the best for the past sample or the future. By my experience less than 400 trades in a sample is kind of meaningless, with high risk of curve fitting. A 1000 trades starts to be ok.
Two other means I use to understand if an automated strategy is solid are:
1) Change the time frame and symbol the strategy is loaded on an see if it remains profitable
2) observe the optimization process, I want a large area of profitable values. If the strategy is profitable only only 10 tests out of a 1000 it is quite weak.
Past performance indicative of future results? I would say that is not carved in stone but at least you have a good statistical approach . In the end that is a bit the basis of trading techniques and of empirical science in general : we cannot be sure the sun will rise tomorrow but we can take a good guess.
As for automated strategies a good genetic optimization over a large sample would help put you on the right track, followed then by a walk forward optimization which will test past optimization results forward in time. The idea is that recent segment test are better to evaluate the equation parameters as for future possible performances. That been said, unless the strategy is veeeeery solid, you will need it to re-optimized it every X time.
All this to say, that before buying TrainingTraders software I would like to take a look at those data at least. Not because I expect necessarily the strategy to perform equally but to know that at least it has been tested properly.
I just hope these guys would provide some info as I am interested .
In the free webinars on their webiste Mr. Baghdady shows quickly a performance report for the strategy on crude, but with only abt 70 traders and I have immediately noticed the equity curve had a big plateau and made most of the profit in the last three trades. As for my experience when I see that in a graph I wonder " if I were trading this live , would I be sitting for 40 trades without making a penny and let the strategy do its thing"?
I hope they reply.
p.s.
If some of you guys are interested and if by miracle I learn how to do it I could post a performance report of an automated strategy with a 1000 trades, to show what I am talking abt here.
Did I somehow answered the question? A bit too much?
Mr. Charts
Legendary member
- Messages
- 7,370
- Likes
- 1,200
Thanks for taking the time and trouble.
Very good post, just what this site needs.
Richard
B
Black Swan
Blimey, great post, welcome to the asylum...
Nope it wasn't too much, cracking info. thanks. Please do post up the performance report and your thoughts when you get the time, perhaps put it up on its own separate thread, methods section? Please give me a nudge (pm) when you do.
