Software for strategy optimization

[J_C_Anderson]

Dear forum community,

Please, share your opinion regarding strategy optimization tools. Does anyone use such tools to figure out optimal parameters for indicators-based strategies?

I have a chance to be engage in the development process of one tool intended to assist traders seeking for improvement of their strategies. The idea is to utilize AI-based approach to reduce the number of combinations checked – this will help to save time. Actually, as in many other spheres, we do not need to check each possible option – it is enough to check a random set of options, make a quick search “nearby” each of them, and then choose the best option among them. We use the same approach when making any choice in our day-to-day life, so it was obvious to follow the same way in trading.

So, the idea itself is quite simple: we create a list of all possible combination, and then the AI-based algorithm chooses a few of them to backtest using historical data. Depending on the backtesting data, the algorithm will tighten the circle searching for the best combination.

Kindly share your opinion regarding this. Would you like to have a tool like this? If yes, what features it should have?

 

[trendie]

I would be willing to use an optimisation tool.
Do you recommend any?
The tool would only be as good as the quality of the data it is run on?

One concept I have been mulling over is robustness testing:
1: assume any set of parameters: eg, Stochastics, Moving Averages, etc.
2: Determine whatever outcome, and this data-set becomes your baseline.
3: now, re-run the tests against each parameter 10% or 20% bigger or smaller, eg change the MA by 10% or 20%, then see if the outcomes change disproportionately.
I have a vague idea this could be used to test the robustness of core principles, so you don't screw yourself over thinking you've found the Holy Grail.

My reasoning is, the change in the settings can be used as a proxy for change in volatility or ATR, or change in the behaviour of the market.

Does any tool do this?
How much would it cost?

 

[Pat494]

I bought a tool once from Biocomp years ago that optimised various indicators like MAs etc.
It looked great and showed profits but when run in real time invariably made losses. Past profits are no indication of future profits it seems.

 

[correct23]

Dear forum community,

Please, share your opinion regarding strategy optimization tools. Does anyone use such tools to figure out optimal parameters for indicators-based strategies?

I have a chance to be engage in the development process of one tool intended to assist traders seeking for improvement of their strategies. The idea is to utilize AI-based approach to reduce the number of combinations checked – this will help to save time. Actually, as in many other spheres, we do not need to check each possible option – it is enough to check a random set of options, make a quick search “nearby” each of them, and then choose the best option among them. We use the same approach when making any choice in our day-to-day life, so it was obvious to follow the same way in trading.

So, the idea itself is quite simple: we create a list of all possible combination, and then the AI-based algorithm chooses a few of them to backtest using historical data. Depending on the backtesting data, the algorithm will tighten the circle searching for the best combination.

Kindly share your opinion regarding this. Would you like to have a tool like this? If yes, what features it should have?

I took it to simplicity- the more simple my analyze is the better I focus in trading. that's why my go to tool for a more edge in trading is Bookmap heatmap. graphically display on the real pure battle between buyers and sellers.

 

[rulox]

I am also looking for an easy to use backtesting tool, ....
Anyway I am trying now an open source python tool called "Backtrader" (backtrader.com), and some time ago I tried Matlab also. For these two tools you need to have some skills with software coding.

 

[iq200]

No tool will do what you want the way you want it. The only way is to write it yourself. I tried quite a few commercial solutions and eventually got fed up with limitations and wrote a tool myself.

The tool started off as a scanner to scan the current market using daily data that could get candidates to trade for the next trading day. Over the last 4-5 years, this has turned into a scanner, backtesting and optimization tool using multiple threading.
I can now test portfolios of my required size which can be created internally based on prefiltering criteria. I can split the data and backtest and forward test over my required intervals. A run over the last 50 years using all data including delisted equities takes approx 20-30 mins. The testing can also be done using multiple strategies simultaneously.

The optimizer allows me optimize the strategy data using a genetic optimizer. Obviously the time taken will be dependent on the number of optimized strategy parameters as well as the genetic optimizer parameters.

Future development includes plug and play entry and exit conditions to various multiple strategies and finding the top candidates.

 

[Pat494]

No tool will do what you want the way you want it. The only way is to write it yourself. I tried quite a few commercial solutions and eventually got fed up with limitations and wrote a tool myself.

The tool started off as a scanner to scan the current market using daily data that could get candidates to trade for the next trading day. Over the last 4-5 years, this has turned into a scanner, backtesting and optimization tool using multiple threading.
I can now test portfolios of my required size which can be created internally based on prefiltering criteria. I can split the data and backtest and forward test over my required intervals. A run over the last 50 years using all data including delisted equities takes approx 20-30 mins. The testing can also be done using multiple strategies simultaneously.

The optimizer allows me optimize the strategy data using a genetic optimizer. Obviously the time taken will be dependent on the number of optimized strategy parameters as well as the genetic optimizer parameters.

Future development includes plug and play entry and exit conditions to various multiple strategies and finding the top candidates.

You might find that you have a profitable market selling copies of your testing system ??
The guys selling stores to the goldfields made the steady profits. Trump's granddaddy made his fortune there but it wasn't for selling shovels.