main.R

# attach libraries to search path
library(httr)

# read external functions
source("resources/sparql.R")

# define the SPARQL query (to run this on a federal computer, set proxy server's address)
data = sparql("resources/query.rq", proxy = NULL)

# remove duplicates
data = unique(data)

# subset the data after 2010
data = subset(data, subset = year >= 2010)

# convert units for the producer price from 0.01 CHF/kg to 0.01 CHF/L
data$producerPrice <- data$producerPrice * 1.03

# convert `producerPrice` unit
data$producerPrice = data$producerPrice * 0.01

# create a new variabel of `Date` type
data$date <- as.Date(paste(data$year, data$month, "1", sep = "-"))

# sort data by `date`
data <- data[order(data$date),]

# compute the producer's share
data$producerShare <- 100 * data$producerPrice / data$consumerPrice

# compute smoothed versions of all shares (using Seasonal Decomposition of Time Series by Loess)
for (i in c("consumerPrice","producerPrice","producerShare")) {
  data[,paste0(i,"Smoothed")] <- data[,i] |>
    ts(start = 2001, frequency = 12) |>
    stl(s.window = 11) |> # s.window is the window size used to estimate the seasonal component. If it's not "periodic", the seasonal component is allowed to change over time
    getElement("time.series") |>
    as.data.frame() |>
    subset(select = "trend")
}

# write producer and consumer prices as machine-readable data
write.csv(x = data[,c("date","producerPrice","consumerPrice","producerPriceSmoothed","consumerPriceSmoothed")],
          file = "results/prices.csv",
          row.names = FALSE)

# write producer's share as machine-readable data
write.csv(x = data[,c("date","producerShare","producerShareSmoothed")],
          file = "results/producers-share.csv",
          row.names = FALSE)