#' Function to download the current components of the DAX index from Wikipedia
#'
#' This function scrapes the stocks that constitute the DAX index from the wikipedia page at https://en.wikipedia.org/wiki/FTSE_100_Index#List_of_FTSE_100_companies.
#'
#' @inheritParams BatchGetSymbols
#'
#' @return A dataframe that includes a column with the list of tickers of companies that belong to the DAX index
#' @export
#' @import rvest
#' @examples
#' \dontrun{
#' df.DAX <- get_DAX()
#' print(df.DAX$tickers)
#' }
get_DAX <- function(do.cache = TRUE,
cache.folder = 'BGS_Cache'){
if (do.cache) {
# check if cache folder exists, create if not
ifelse(!dir.exists(cache.folder), dir.create(file.path(cache.folder)),'')
cache.file <- file.path(cache.folder, paste0('DAX_Composition_', Sys.Date(), '.rds') )
# check if file exists
flag <- file.exists(cache.file)
if (flag) {
df.DAX <- readRDS(cache.file)
return(df.DAX)
}
}
my.url <- 'https://en.wikipedia.org/wiki/DAX'
read_html <- 0 # fix for global variable nagging from BUILD
my.xpath <- '//*[@id="constituents"]'
df.DAX <- my.url %>%
read_html() %>%
html_nodes(xpath = my.xpath) %>%
html_table()
df.DAX <- df.DAX[[1]]
df.DAX[1] = NULL
colnames(df.DAX) <- c('company','prime_standard_industry','ticker','index_weight','employees','founded')
if (do.cache) {
saveRDS(df.DAX, cache.file)
}
return(df.DAX)
}
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.