More Backtrader way to calculate covariance matrix

cnimativ

I am trying to calculate the covariance matrix of a multiple asset portfolio at every period with N lookback period. My current implementation uses a counter to wait out the lookback period and then runs all calculations in the def next() method.

 def next(self):
        if (self.counter < self.p.rperiod):
            self.counter += 1
        else:
            # stack the closes together into a numpy array for ease of calculation
            closes = np.stack(([d.close.get(0,self.p.rperiod).tolist() for d in self.datas]))
            rets = np.diff(np.log(closes))
            # covariance matrix. will be shoved somewhere else for records
            cov = np.cov(rets)

While it 'works', what is the more backtrader way of doing the same thing?

For example, there's the PctChange indicator for calculating percentage change that is defined at __init__.

def __init__(self):
    rets = [bt.ind.PctChange(d, period=self.p.rperiod) for d in self.datas]

But how do I access the value of N periods of indicators, run calculations, and then construct cross-asset indicators? And would it be possible to have, say, an indicator that contains multidimensional matrices?

Much thanks!

backtrader

@cnimativ said in More Backtrader way to calculate covariance matrix:

My current implementation uses a counter to wait out the lookback period and then runs all calculations in the def next() method.

You should be using the addminperiod API or subclassing from the indicator PeriodN.

@cnimativ said in More Backtrader way to calculate covariance matrix:

While it 'works', what is the more backtrader way of doing the same thing?

You calculate something in next using numpy. What would you be looking for?

@cnimativ said in More Backtrader way to calculate covariance matrix:

But how do I access the value of N periods of indicators, run calculations, and then construct cross-asset indicators?

You are already using get

@cnimativ said in More Backtrader way to calculate covariance matrix:

And would it be possible to have, say, an indicator that contains multidimensional matrices?

Not a clue what you mean. Indicator have lines. You may of course store anything in a member attribute.