QuantConnect Community

Hi Nevo,

1. In QC, each algo can only access one account. And one account with existing positions generally gets “loaded” into the algo so it knows where it is at. And since you do not want multiple algos to mix in one account, it sounds to me you'll need multiple accounts so you can trade the same assets OR different assets in multiple ways. Two algos? Two accounts.
2. If multiple algos connect to the same account it will be difficult to control that. All algos will want to see the portfolio in its entirety. Moreover, with IBKR, you cannot make multiple connections to the same account.

Fred

Hey Nevo, what Fred said is true. 

Interactive brokers might help you, however. I think it is easy to create subaccounts in IB with unique account numbers that you can deploy your algorithms to. If you do decide to compromise and use one account, the alpha framework actually allows you to add multiple alphas together into one super alpha, and you can pass in the weights you desire manually if you set your alpha model to consume arguments that way.

Ex:

# set alpha models

self.AddAlpha(ABCAlphaModel(0.4))

self.AddAlpha(XYZAlphaModel(0.4))

I had a similar dilemma and ending up using a single account. There are actually benefits to using a single account instead of separate accounts for each algorithm you have. 

Ex:

You have 3 different accounts that you have unique alphas deployed to. In 1 year, each account makes 2 trades that double the balance. So pretending the starting value for each account was X, the starting value for all accounts was 3X and the ending value is (2 x 2 x X) x 3 = 12X. But if you had everything in a single account, your starting capital is 3X, but in the same account, you make 6 trades that each double the balance, so your ending balance is 3X x 2 x 2 x 2 x 2 x 2 x 2 = 192X. An extreme example, but the same thing applies if your returns are 15%, 30%, etc.

There are some other assumptions there too, like maybe that's now how your alpha works, but hopefully this example can inform your decision. 

Nevo Raz

-> What is the best way to do it if I'm using more than one algorithm? 

The best way to do it is to create a composite algorithm that combine several strategies. 

Attached “RSI and Ema Cross Composite Alpha Model” is a QCAlgorithmFramework that combines the built-in QC RsiAlphaModel and EmaCrossAlphaModel. 

If you are satisfied with my answer, please accept it.