MetaStrategy-Developing-an-evolutionary-approach-to-strategy-selection

So are you saying the Strategy Ranking tool (which WL already supports) should be "automatically called" by a MetaStrategy to determine which child strategy to use on a bar-by-bar bases?

(I may need to add more memory to my computer for this. It's the "bar-by-bar bases" part that requires the extra memory.)

superticker,

I would imagine that Glitch would run all of the proposed strategies once, store that data of trades, and then use it in various combinations to do the assessment. I may be mistaken, but I believe that would result in a fast, memory-efficient assessment of any proposed combination of strategies.

QUOTE:
I would imagine that Glitch would run all of the proposed strategies once, store that data of trades, and then use it in various combinations to do the assessment.

So a bar-by-bar analysis evaluated as a "batch process" where the trade history (List<Position> data) is "temporarily saved" for batch comparison. Yes, that would save memory because the strategies would be run sequentially to establish the temporary trade history for later comparison.

My only comment is that you can't expect WL users to install 128GB of memory on a laptop. But your temporary-trade-history approach should avoid that need. Good thinking.

If you think that this is a worthwhile addition, please vote for it!

QUOTE:
If you think that this is a worthwhile ...

Personally, in practice, I don't use metastrategies. This is because most of my entry and exit trading conditions employ exactly the same indicators, so it makes more sense for me to create a

CODE:
      public override void Execute(BarHistory bars, int idx)
      {
         if (HasOpenPosition(bars, PositionType.Long))
         {
            //sell conditions below
            if (...)
               PlaceTrade(...,"Signal sell 1 msg");
            else if (...)
               PlaceTrade(...,"Signal sell 2 msg");
            else if (...)
               PlaceTrade(...,"Signal sell 3 msg");
         }
         else
         {
            //buy conditions below
            if (...)
               PlaceTrade(..., "Signal buy 1 msg");
            else if (...)
               PlaceTrade(..., "Signal buy 2 msg");
            else if (...)
               PlaceTrade(..., "Signal buy 3 msg");
         }
      }

then I use the contribution performance visualizer to break down the performance of each entry and exit arm into profit statistics expressed as a pie chart by the performance visualizer. Yes, that means the only break down will be of profit and APR stats (which may require a ScottPlot pie chart of the List<Postion> data), but that's good enough for me. Beyond that is an over complication.

Now some users may want to see individual ScoreCard metrics (e.g. Sharpe Ratios) for each individual methodology, and they should be using metastrategies instead to get that. But I don't need that.

Now I do have a buy-high and a buy-low production strategy that use very different indicators, and those might be candidates for a metastrategy. But the truth is there's only an 8% overlap of stocks that trade both of these strategies, so the need for a metastrategy in this case is relatively small.

The Contribution Performance Visualizer will give you a break down of the entry and exit arms out of the box (without any ScottPlot pie charts of the List<Position> data). Here's what it looks like for my buy-high strategy running over 300 stocks.