I am looking for a function, or package, that will help me with this goal. I've looked through several packages but can't find what I am looking for:
Lets say I have an xts object with 10 columns and 250 rows.
What I want to do is run a simulation, such that I get a robust calculation of my performance metric over the period.
So, lets say that I have 250 data points, I want to run x number of simulations over random samples of the data computing the Sharpe Ratio using the function (PerformanceAnalytics::SharpeRatio) varying the random samples to be lengths 30-240, and then find the average. Keep in mind I want to do this for every column and I'd rather not have to use apply if possible. I'd also like to find something that processes the information rather quickly.
What package or functions would best serve this purpose?
Thank you!
Subsetting xts objects for the rows you want to randomly sample should be good enough, performance wise, if that is your main concern. If you want some other concrete examples, you may find it useful to look at the monte carlo simulation functions recently added to the R blotter package:
https://github.com/braverock/blotter/blob/master/R/mcsim.R
Your requirements are quite detailed and a little tricky to follow, but I think this example may be what you're after?
This solution does use apply functions though! Because it just makes life easier. If you don't use lapply, the code will expand quickly and distract from achieving the goal quickly (and you risk introducing bugs with longer, messier code; one reason to use apply family functions where you can).
library(quantmod)
library(PerformanceAnalytics)
# Set up the data:
syms <- c("GOOG", "FB", "TSLA", "SNAP", "MU")
getSymbols(syms)
z <- do.call(merge, lapply(syms, function(s) {
x <- get(s)
dailyReturn(Cl(x))
}))
# Here we have 250 rows, 5 columns:
z <- tail(z, 250)
colnames(z) <- paste0(syms, ".rets")
subSample <- function(x, n.sub = 40) {
# Assuming subsampling by row, preserving all returns and cross symbol dependence structure at a given timestamp
ii <- sample(1:NROW(x), size = n.sub, replace = FALSE)
# sort in order to preserve time ordering?
ii <- sort(ii)
xs <- x[ii, ]
xs
}
set.seed(5)
# test:
z2 <- subSample(z, n.sub = 40)
zShrp <- SharpeRatio(z2)[1, ]
# now run simulation:
nSteps <- seq(30, 240, by = 30)
sharpeSimulation <- function(x, n.sub) {
x <- subSample(x, n.sub)
SharpeRatio(x)[1, ]
}
res <- lapply(nSteps, FUN = sharpeSimulation, x = z)
res <- do.call(rbind, res)
resMean <- colMeans(res)
resMean
# GOOG.rets FB.rets TSLA.rets SNAP.rets MU.rets
# 0.085353854 0.059577882 0.009783841 0.026328660 0.080846592
Do you realise that SharpeRatio uses sapply? And it's likely other performance metrics you want to use will as well. Since you seem to have something against apply (possibly all apply functions in R), this might be worth noting.