Back Testing Indicators For Optimal Parameters

[GiantBull]

There are plenty of different kinds of indicators that have default values inputted in. For example the MACD default settings are 12,26,9 which have been proven to not be profitable in multiple studies. Those parameters worked in the 1980s when the indicator was invented but during this era using those settings will ultimately be a losing strategy. In order to find the optimal parameters for MACD (or any other indicator) backtesting needs to be done to find the most profitable parameters. I want to test numbers 1-100 for each setting on the MACD to find the best combo. Does anyone know of a program where this could be done? Do you need to have extensive knowledge on coding to do this by yourself? I have heard of genetic algorithms that do this kind of testing but have no idea on how it is done. If someone on here has any sort of knowledge in this area I could use the help!

https://medium.com/swlh/how-to-opti...rameters-with-genetic-algorithms-23671d7b446c
Here is an interesting article explaining what genetic algorithms are!

 

[mashume]

The answer is probably in python. The problem is in limiting the number of variables you want to work with. for even a simple strategy involving, lets say the MACD, your variables are:

• candle timeframe
• 3 MACD lengths

Getting loops set up to go over the MACD variables isn't difficult:

for x in range (10, 15):
    for y in range (20, 30);
        for z in range(5, 11):
            #do your test here and store returns in a table you can evaluate later for profit/loss drawdown etc...

nested loops are slow but should finish for this reasonably, if you're not testing on 2 years of 1 minute data.

which brings us to the crux of the problem. There is a python package called yfinance which allows you to grab data from yahoo. It's good enough data, but is limited in the length of what you can retrieve to the last 60 days for intraday.

There are a few good packages for indicators which will run on top of a pandas data frame (which is what yfinance will return).

If all of this is too much for you to undertake I understand. It's a lot of work to backtest so thoroughly. Backtesting is probably an art unto itself and deserving of a grander thread and some code... someday I may work on something for the community when I have time

-mashume