Search code examples
pandasautocorrelation

Calculating auto covariance in pandas

Following on the answer provided by @pltrdy, in this threat:

https://stackoverflow.com/a/27164416/14744492

How do you convert the pandas.Series.autocorr(), which calculates lag-N (default=1) autocorrelation on Series, into autocovariances?

Sadly the command pandas.Series.autocov()is not implemented in pandas.

Solution

  • What .autocorr(k) calculates is the (Pearson) correlation coefficient for lag k. But we know that, for a series x, that coefficient for lag k is:

    \rho_k = \frac{Cov(x_{t}, x_{t-k})}{Var(x)}

    Then, to get autocovariance, you multiply autocorrelation by the variance:

    def autocov_series(x, lag=1):
    return x.autocorr(x, lag=lag) * x.var()

    Note that Series.var uses ddof of 1 by default so N - 1 divides the sample variance where N == s.size (and you'd get an unbiased estimate for the population variance).