How do you validate that an EA is truly profitable? - page 2

New comment
 
Alex Holloway #:
I personally do this:
New EA released on the marketplace ~ I test it on demo account for at least 3 - 6 months and in meantime I make sure to keep an eye on the EA live signal page ~ compare the results on my demo account ~ now if the results was kinda good for the first 3 months I switch to live account with real funds, if it's not I keep it on demo account for another 3 months. So 6 months in total for the validation.
That's how I mak my final decision whether investing in specific EA or not. 

If the EA has 6 months - 2 years live record then I go with live account, real funds immediately

Do you try using same broker as the EA provider?
 
Ryan L Johnson #:

For starters, the graph of the Tester is not the end-all be-all of the Tester Report. The Tester generates a boatload of statistics in that Report (Testing Report - Algorithmic Trading, Trading Robots - MetaTrader 5 Help). Additionally, there are several more expansive reporting utilities in the CodeBase and Articles. Familiarize yourself with each statistic and determine which ones are most important for your particular strategy and its weak spots.

If you have demo price data and live price data, be sure to test on live price data. You need to find a way to calculate your average live spread, slippage, and/or commissions. Some broker-dealers publish/advertise that information on their websites (excluding horrible numbers around the daily close/open). If you're already live trading a strategy, there are several utilities in the CodeBase and Articles for collecting spread and slippage statistics. You can then input those numbers into the Tester (a commission slot may or may not appear depending on your specific broker-dealer).

If you're testing on tick data, I recommend searching the Forum for how to accurately evaluate ticks in your code and backtest on tick data. If your strategy only uses OHLC data, this is more straightforward.

Regarding the building process, this is basically "the elephant in the room." I find that the standard timeframe charts generally present utter chaos. I'm a big fan of transferring price data from a standard timeframe chart into a smoothed custom chart, e.g., Renko, range bars, etc. I experiment with various brick/bar sizes until I see readily apparent swings in the price data. A "shortcut" to finding the right brick/bar size is Stanislav Korotky's Uniformity Factor indicator:

If I don't get that "bingo!" moment where swings jump out at me on the custom charts that I've tried, then I simply move on to another instrument/symbol.

Only after that price smoothing process is successful and complete, do I begin to apply calculations and indicators to the custom chart. Remember, the underlying data is smoothed so the calculations/indicators applied thereto are inherently smoothed as well. My last building step prior to testing is writing an EA that analyzes the custom chart data and trades the standard symbol. There are several custom chart EA coding tutorials on the web.

I tought I knew how to build EAs, but this is very new perspective into building them.

Could you elaborate more the part about the renko charts as well as the Stanislav's indicator? Or do you have an example?

Im gonna pin this topic, very interesting answers.

 

For me, I usually backtest → demo → then small live account.

Demo already filters out a lot. If it can’t even hold on demo, no point going further. Small live is where you really see the truth, especially with execution and spreads.

I don’t rely too much on metrics alone. You can make the numbers look good with optimisation.

12
New comment