In m-g-h/R.MFIV: Calculate Model Free Implied Volatility, i.e. the CBOE VIX from Option Quotes

knitr::opts_chunk$set(echo = TRUE)

Tests for detemining which functions / packages are fastest

Setup

library(R.MFIV)
library(purrr)
library(future.apply)
library(furrr)
plan(multisession, workers = 4) ## Parallelize using four cores
library(lubridate)
library(microbenchmark)

## This function converts the mapply results below to a data.table
multicols <- function(matrix){
  data.table::as.data.table(t(matrix))[, lapply(.SD, unlist)]
}

data <- data.table::copy(R.MFIV::option_dataset)

data[, `:=`(maturity = time_length((exp + hours(16) - t),unit = "years"),
            R = interpolate_rfr(date = as_date(t),
                                exp = exp))]
data

`CBOE_F_0()`

Compare base vs. purrr vs the future versions

## base
base_F_0 <- function(){
  data[, .(F_0 = mapply(CBOE_F_0,
                        option_quotes,
                        R,
                        maturity))]
}
## future.mapply
future_F_0 <- function(){
  data[, .(F_0 = future_mapply(CBOE_F_0,
                               option_quotes,
                               R,
                               maturity))]
}
## purrr
purrr_F_0 <- function(){
  data[, .(F_0 = pmap_dbl(.l = list(option_quotes,
                                    R,
                                    maturity),
                          .f = CBOE_F_0))]
}

## furrr
furrr_F_0 <- function(){
  data[, .(F_0 = future_pmap_dbl(.l = list(option_quotes,
                                           R,
                                           maturity),
                                 .f = CBOE_F_0))]
}

microbenchmark(base_F_0(),
               future_F_0(),
               purrr_F_0(),
               furrr_F_0(),
               times = )

`CBOE_VIX_vars()`

Compare base vs. purrr vs the future versions

## base
base_VIX <- function(){
  data[, .(sigma_sq = mapply(CBOE_VIX_vars, option_quotes, R, maturity))]
}
## future.mapply
future_VIX <- function(){
  data[, .(sigma_sq = future_mapply(CBOE_VIX_vars, option_quotes, R, maturity))]
}
## purrr
purrr_VIX <- function(){
  data[, .(F_0 = pmap_dbl(.l = list(option_quotes, R, maturity),
                          .f = CBOE_VIX_vars))]
}

## furrr
furrr_VIX <- function(){
  data[, .(F_0 = future_pmap_dbl(.l = list(option_quotes, R, maturity),
                                 .f = CBOE_VIX_vars))]
}

microbenchmark(base_VIX(),
               future_VIX(),
               purrr_VIX(),
               furrr_VIX(),
               times = 5)

Short Fazit

future.apply and furrr show the best performance, and the former is slightly faster.

`CBOE_VIX_vars()` with multiple column results

Compare furrr vs the future.apply versions

## future.mapply

setupfun <- function(){
  print("setup")
  data2 <- data.table::copy(data)
}

data2 <- data.table::copy(data)
data3 <- data.table::copy(data)

future_VIX_full <- function(){
  data2[, c("F_0", "K_0", "n_put_raw", "n_call_raw", "n_put", "n_call", "sigma_sq") := 
          multicols(future_mapply(CBOE_VIX_vars,
                                  option_quotes, R, maturity,
                                  MoreArgs = list(ret_vars = T))
          )]


}

## furrr
furrr_VIX_full <- function(){
  data3[, c("F_0", "K_0", "n_put_raw", "n_call_raw", "n_put", "n_call", "sigma_sq") := future_pmap_dfr(.l = list(option_quotes, R, maturity),
                                        .f = CBOE_VIX_vars,
                                        ret_vars = T)][]
}

microbenchmark(future_VIX_full(),
               furrr_VIX_full(),
               times = 1)