R/features.R

In MislavSag/alphar:

library(PerformanceAnalytics)
library(data.table)
library(roll)
library(xts)
library(future.apply)
library(RollingWindow)
library(TTR)
# library(quantmod)

# ohlc <- market_data

add_features <- function(ohlc, window_sizes = c(5, 10, 20, 40, 80, 150, 300)) {

  # add ohlc transformations
  ohlc$hi_lo <- ohlc$high - ohlc$low
  ohlc$cl_op <- ohlc$close - ohlc$open
  cl_ath <- cummax(ohlc$close)
  ohlc$cl_ath_dev <- (cl_ath - ohlc$close) / cl_ath

  # returns
  ohlc$returns <- PerformanceAnalytics::Return.calculate(ohlc$close)
  ohlc$returns_squared <- (ohlc$returns)^2

  # rolling volatility
  stds <- future.apply::future_sapply(
    X = window_sizes,
    FUN = function(x) roll::roll_sd(zoo::coredata(ohlc$returns), x)
  )
  colnames(stds) <- paste0('std_', window_sizes)
  ohlc <- cbind(ohlc, stds)

  # Close-to-Close Volatility
  n_ <- 5
  ohlc$ctc <- volatility(quantmod::OHLC(ohlc), n = n_, calc = "close")
  ohlc$parkinson <- volatility(quantmod::OHLC(ohlc), n = n_, calc = "parkinson")
  ohlc$rogers <- volatility(quantmod::OHLC(ohlc), n = n_, calc = "rogers.satchell")
  ohlc$gkyz <- volatility(quantmod::OHLC(ohlc), n = n_, calc = "gk.yz")
  ohlc$yz <- volatility(quantmod::OHLC(ohlc), n = n_, calc = "yang.zhang")

  # skewness
  skews <- future.apply::future_sapply(
    window_sizes,
    function(x) RollingWindow::RollingSkew(zoo::coredata(ohlc$returns), window = x, na_method = 'ignore')
  )
  colnames(skews) <- paste0('skewness_', window_sizes)
  ohlc <- cbind(ohlc, xts::xts(skews, order.by = zoo::index(ohlc)))

  # kurtosis
  kurts <- future.apply::future_sapply(
    window_sizes,
    function(x)  RollingWindow::RollingKurt(zoo::coredata(ohlc$returns), window = x, na_method = 'ignore')
  )
  colnames(kurts) <- paste0('kurtosis_', window_sizes)
  ohlc <- cbind(ohlc, xts::xts(kurts, order.by = zoo::index(ohlc)))

  # Skewness-Kurtosis ratio
  sk_ratio <- ohlc[, paste0('skewness_', window_sizes)] / ohlc[, paste0('kurtosis_', window_sizes)]
  colnames(sk_ratio) <- paste0('skewness_kurtosis_', window_sizes)
  ohlc <- cbind(ohlc, xts::xts(sk_ratio, order.by = zoo::index(ohlc)))

  # standardized returns
  ohlc$standardized_returns <- ohlc$returns / ohlc$std_5
  ohlc$standardized_returns_squared <- (ohlc$standardized_returns)^2

  return(ohlc)
}