Some merchants ask me if I use the 3D graphics optimization for my own business. Do I have a good experience with this tool? I recommend for ATS robust development? The answers are simple: NO!
I spent a lot of time with him, so I can explain in more detail (luckily I've never started using it for my development ATS).
First, we do a little summary - which is exactly the 3D graphic optimization? It is a simple visual aid that helps us to find global optimal values instead of the local.
The local optimum is a combination of two parameters that works best in a given radius, no matter how surrounding work settings.
The global optimum is a combination of 2 parameters gives us positive results in a particular combination of settings for these two variables, but also gives stable results and quality for all combinations of settings around. The more work combination around all "central" parameters, the better.
The main advantage is that the whole concept seems simple, logical and robust and it is almost easy to understand, with good visual interpretation.
On the other hand, the biggest drawback is that you can use this concept to optimize 2 parameters (there are a few tweaks how to use this concept to other parameters, but its basic simplicity disappears).
Let's take a look at a simple example - a simple model of a breakout strategy using x-bar rear closing price increased 3x ATR with a period as a starting point.
Released at the end of the day.
The simplified model may look like this:
Buy the next bar (Close [BarsBack] + 3 * ATR (ATRperiod)) Off;
Setexitonclose;
We want to see what the optimal settings for BarsBack and ATRperiod settings. To start, we can use 1-10 range for both parameters (in fact, we should use another band, but for the purposes of this example, we will use values 1-10) and create a simple graphical, easy to visually interpret 3D net income will show all the combinations.
Let's say the best value you see on the chart is 10 (ATRperiod) - 1 (Barsback) combination. We can consider that this combination of being local optimum - and, most likely, can not be the result of over-optimization.
On the other hand, that the combination 5-3 is more like the global optimum, and the combinations have similar results around 5-3 combination. We skipped the point that the global optimum should be much larger; We use this as a simple example.
Up to this point, everything is fine.
So where is the problem?
The problem is a high instability of this approach.
ATS as a trader full-time, I spend a lot of time comparing different approaches and methods. I also spent a lot of time with 3D optimization. I discovered, like other users, the simplicity very tempting concept. There are about a year, I made a long series of tests that took a month of very intensive work. The reason behind all this work was to find global optimal settings for many working group models at different time intervals using the 3D optimization table.
example:
START Model: Optimal Global Search in the period 2003-2005, 2006-2008, 2009-2011, 2012-2014.
STARTING MODEL B: Optimal Global Research in 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014.
etc.
The objective of the trial was to determine to what extent the global optimum shift during each test period.
The results are shocking. Previous less intensive testing, he knew that the parameters obtained in this way are not stable. that through the test, however, showed me that instability can be huge. For example - in many cases better global optimum for a given time interval was almost the worst possible. The world has changed incredibly optimal - if I use 3D graphics mentioned above, for a year as the global optimum recommended 9-9 to 2-2 the next year (in the opposite corner).
This instability, tested on a wide range of markets, subjects of interest models and deadlines, certainly assured me that the 3D graphics optimization method are not for me. Therefore, I do not use it for my own business at all.
Of course, for someone who can work better 3D optimization (remember, no dogma in the trade), but this concept is not for me.
http://www.annonces.immo-reve.com/
https://www.facebook.com/immo.reve.tn
I spent a lot of time with him, so I can explain in more detail (luckily I've never started using it for my development ATS).
First, we do a little summary - which is exactly the 3D graphic optimization? It is a simple visual aid that helps us to find global optimal values instead of the local.
The local optimum is a combination of two parameters that works best in a given radius, no matter how surrounding work settings.
The global optimum is a combination of 2 parameters gives us positive results in a particular combination of settings for these two variables, but also gives stable results and quality for all combinations of settings around. The more work combination around all "central" parameters, the better.
The main advantage is that the whole concept seems simple, logical and robust and it is almost easy to understand, with good visual interpretation.
On the other hand, the biggest drawback is that you can use this concept to optimize 2 parameters (there are a few tweaks how to use this concept to other parameters, but its basic simplicity disappears).
Let's take a look at a simple example - a simple model of a breakout strategy using x-bar rear closing price increased 3x ATR with a period as a starting point.
Released at the end of the day.
The simplified model may look like this:
Buy the next bar (Close [BarsBack] + 3 * ATR (ATRperiod)) Off;
Setexitonclose;
We want to see what the optimal settings for BarsBack and ATRperiod settings. To start, we can use 1-10 range for both parameters (in fact, we should use another band, but for the purposes of this example, we will use values 1-10) and create a simple graphical, easy to visually interpret 3D net income will show all the combinations.
Let's say the best value you see on the chart is 10 (ATRperiod) - 1 (Barsback) combination. We can consider that this combination of being local optimum - and, most likely, can not be the result of over-optimization.
On the other hand, that the combination 5-3 is more like the global optimum, and the combinations have similar results around 5-3 combination. We skipped the point that the global optimum should be much larger; We use this as a simple example.
Up to this point, everything is fine.
So where is the problem?
The problem is a high instability of this approach.
ATS as a trader full-time, I spend a lot of time comparing different approaches and methods. I also spent a lot of time with 3D optimization. I discovered, like other users, the simplicity very tempting concept. There are about a year, I made a long series of tests that took a month of very intensive work. The reason behind all this work was to find global optimal settings for many working group models at different time intervals using the 3D optimization table.
example:
START Model: Optimal Global Search in the period 2003-2005, 2006-2008, 2009-2011, 2012-2014.
STARTING MODEL B: Optimal Global Research in 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2013, 2014.
etc.
The objective of the trial was to determine to what extent the global optimum shift during each test period.
The results are shocking. Previous less intensive testing, he knew that the parameters obtained in this way are not stable. that through the test, however, showed me that instability can be huge. For example - in many cases better global optimum for a given time interval was almost the worst possible. The world has changed incredibly optimal - if I use 3D graphics mentioned above, for a year as the global optimum recommended 9-9 to 2-2 the next year (in the opposite corner).
This instability, tested on a wide range of markets, subjects of interest models and deadlines, certainly assured me that the 3D graphics optimization method are not for me. Therefore, I do not use it for my own business at all.
Of course, for someone who can work better 3D optimization (remember, no dogma in the trade), but this concept is not for me.
http://www.annonces.immo-reve.com/
https://www.facebook.com/immo.reve.tn
Aucun commentaire:
Enregistrer un commentaire