Thanks a lot. Could you also add one set with 10 times profit target chosen.
R, matlab, etc: as long as the charts are there, then who cares.
Optimization of 3 or more parameters
- Thread starter ehbeehefak
- Start date
Thanks a lot. Could you also add one set with 10 times profit target chosen.
R, matlab, etc: as long as the charts are there, then who cares.
What were the results? How did you plot them? Can you show some example plots?
EDIT: Again, this post was delayed due to waiting for moderator approval. Sorry.
I've optimized parameters using Solver. They weren't necessarily the same parameters you are trying to optimize.
For example, I created a sheet that included all historical data for a particular stock/etf. I added two SMA columns. I added another column to indicate the SMA crossovers. I added columns to maintain a simulated trading account balance based upon simulated trades based upon the SMA crossovers. I had Solver Maximize the Balance by changing the parameters of both SMA's (5 days, 6 days, 9 days, etc.).
This is a simple example of what you can do. You can do several parameters, not just two.
Also, there was no need to look at a plot. Solver found the best parameters to produce the greatest profit. The output was the parameter settings, not a graph.
I've optimized parameters using Solver. They weren't necessarily the same parameters you are trying to optimize.
For example, I created a sheet that included all historical data for a particular stock/etf. I added two SMA columns. I added another column to indicate the SMA crossovers. I added columns to maintain a simulated trading account balance based upon simulated trades based upon the SMA crossovers. I had Solver Maximize the Balance by changing the parameters of both SMA's (5 days, 6 days, 9 days, etc.).
This is a simple example of what you can do. You can do several parameters, not just two.
Also, there was no need to look at a plot. Solver found the best parameters to produce the greatest profit. The output was the parameter settings, not a graph.
Last edited:
EDIT: Again, this post was delayed due to waiting for moderator approval. Sorry.
I've optimized parameters using Solver. They weren't necessarily the same parameters you are trying to optimize.
For example, I created a sheet that included all historical data for a particular stock/etf. I added two SMA columns. I added another column to indicate the SMA crossovers. I added columns to maintain a simulated trading account balance based upon simulated trades based upon the SMA crossovers. I had Solver Maximize the Balance by changing the parameters of both SMA's (5 days, 6 days, 9 days, etc.).
This is a simple example of what you can do. You can do several parameters, not just two.
Also, there was no need to look at a plot. Solver found the best parameters to produce the greatest profit. The output was the parameter settings, not a graph.
That means you did 2-parameter strategy optimization (the fast-sma and the slow-sma). It is easy to find the combination that results in maximum profit, but the problem is that might just be a "peak" combination as opposed to a "plateau" combination of parameters (http://www.multicharts.com/img/2595_Exhaustive_optimization_Genetic_optimization.png). So just finding the combination that returns the maximum profit is not enough.
I've done multiple parameter optimization with Solver--not just two.
Maximum profit is ... MAXIMUM. What else would you want besides the maximum? If you are suggesting that the optimization is producing a local maximum rather than a global maximum; then that is a problem I can't solve. You have to put your trust in the optimization algorithm you use. Unless the parameter universe is such that you can perform a brute-force search via VBA, etc. for example, you are at the mercy of your search algorithm.
No, I am not suggesting that the optimization is producing a local maximum, I am suggesting that the optimization might produce a global maximum but with "nearby" combinations having profits drastically lower than the found maximum your excel solver finds. I have read a lot that those global maximums won't result in a return similar to the one produced in the backtest simply because for the real trading to produce the same results, the instrument should behave exactly the same as it did in backtest. One small change in market characteristics, you may even be looking at losses. All in all, finding the maximum, if it is enough for you, go with it. Read sensitivity analysis, and reanalyze your method. If finding just the combination that result in global maximum is enough for you, I am sorry for your losses.
If you have a trading system in which the entry strategy depends in whole or in part on the maximization or minmization of a quantitative indicator, then that optimization can be separated from the overall profit optimization. For example, let's suppose your entry depends in whole or in part on ADX being above a certain threshold. Then the ideal lookback period for ADX would be the lookback that maximizes the average ADX value. This is one less parameter that has to be considered in the profit optimization.
I see. I don't trade based upon the optimizations I've done, btw. I found that such optimizations done over many years of history seem to curve fit. As you suggest, though profitable over many, many years; a system based on the optimizations may go for over a year without profit. Optimizing over a shorter 'sliding window' of historical prices might be better.
Sorry I couldn't help with your original question.
Vow. Do you use a very large training data set ( tick data, multiple stocks/futures)?
Last edited: