QuantConnect Community

Hi Klemen,

You can look through some of the basic portfolio construction code on GitHub and some sample framework algorithms for inspiration. We do have 5 specific examples, but please be creative and try to implement this portfolio construction model in QC! If you run into any questions about the API, please do not hesitate to reach out again.

I don't know how to take a simple portfolio creation, let's say, for the sake of example, this one discussed here Adaptive Asset Allocation: ReSolve Asset Management . As you can see it trades only at the end of month 

and has three other paramters

What I want to create (and have no idea how to do it as a beginner in QuantConnect) is many sub portfolios (tranches) each parametrized in a different way and with rebalancing every trading day of the month (not only last trading day of the month) to hopefully crete a more roboust "model" (avoid overfitting to turn of the month and other cherry picked/lucky paramters).

For example I want to have momentum_lookback in the range of 120 to 130, correlation_lookback in the range of 110 to 140 and volatility_lookback in the range of 15 to 25. This means I am simultaneously running 21(number_of_tranding_days_of_the_month) * 11(number of distinct values for momentum_lookback parameter) * 31(number of distinct values for correlation_lookback parameter) * 11(number of distinct values for volatility_lookback parameter). From all this tranches (slightly differently paramterized portfolio constructions) I need to then, every day, consolidate positions and do the trading (delta from current portfolio holdings to desired portfoolio holdings).

Please look at the links I provided in original question. Does this makes sense now? How to do this in QuantConnect? Can you maybe guide me on how to extend Adaptive Asset Allocation: ReSolve Asset Management ?

Hi Klemen,

There are, broadly speaking, two ways to approach this:

1. Create an Alpha Model (see documentation for more information on what Alpha is) that takes as inputs momentum, correlation and vol lookbacks. Create one instance of this model for each set of parameters. Aggregate the resulting insights inside a Portfolio Construction model.
   
   This is a really robust approach but can come at the cost of computational inefficiency since you're looking at such a huge number of pairs (11*31*11 = 3751), each one which would need an instance of your Alpha model.
    
2. Break down the mathematics into what the "average" covariance matrix would have been for the different lookback values and use that matrix to compute the minimum variance weights. It's not immediately clear to me how to do this or if it's even possible, but what I am certain of is that there are ways to compute these values more efficiently than creating an Alpha model for each pair.

To do the code above then at some level, you'll need an abstraction before passing it to QuantConnect's basic API to implement. You won't get that stuff "out of the box" -- but its code, so you can build whatever you need. The API provides a thousand different types of lego blocks it is up to you how you clip them together.