Calculate rolling (moving-window) correlation between two time series

Computes the Pearson correlation between x_var and y_var over a rolling window of n time periods within each group, using runCor. The first n - 1 observations in each group will be NA because there are insufficient data to fill the window.

Usage

rolling_corr(dat, date_var = "date", grp_var, x_var, y_var, n = 14)

Arguments

dat

A data frame containing the two time series, a date column, and a grouping column.

date_var

Character string giving the name of the date column. Must be of class Date. Default is "date". Rows are sorted by date_var within each group before the rolling correlation is computed (required because runCor assumes row order is chronological).

grp_var

Character string giving the name of the grouping column. Rolling correlation is computed separately within each group.

x_var

Character string giving the name of the primary time series column.

y_var

Character string giving the name of the secondary time series column.

n

Integer. Width of the rolling window in time periods. Default is 14.

Value

A data frame with the same columns as the input plus one additional numeric column, roll_corr, containing the rolling Pearson correlation between x_var and y_var. Values range from -1 to 1. The first n - 1 observations per group are NA. Note that rows where x_var or y_var are NA are removed before the rolling correlation is computed, so the returned frame may have fewer rows than the input.

See also

cross_corr to identify the optimal lag before computing rolling correlation; runCor for the underlying method; plot_corr to visualise the result.

Examples

if (FALSE) { # \dontrun{
data_corr <- rolling_corr(
  dat      = my_data,
  date_var = "date",
  grp_var  = "region",
  x_var    = "r_mean",
  y_var    = "movement",
  n        = 14
)
} # }