VaR/ES with portfolio_method='single' crashes when mu/sigma/m3/m4 are NULL (no auto-computation of moments)

[braverock]

Summary

VaR() and ES() crash with an unhelpful error when called with portfolio_method='single' and either method='modified' or method='gaussian' if the caller does not explicitly supply moment parameters (mu, sigma, m3, m4). The multi-asset branches of the same functions auto-compute missing moments from R, but the single portfolio branch does not.

Affected versions

Reproduced with PerformanceAnalytics 2.0.9 (current CRAN).

Steps to reproduce

library(PerformanceAnalytics)
data(edhec)
R <- edhec[, 1:4]
w <- rep(0.25, 4)

# Crash 1: method='modified', moments not supplied
VaR(R = R, weights = w, portfolio_method = "single", method = "modified", invert = FALSE)
# Error in skewness.MM(w, sigma, M3) : M3 must be a matrix

# Crash 2: method='gaussian', moments not supplied
VaR(R = R, weights = w, portfolio_method = "single", method = "gaussian", invert = FALSE)
# Error in t(w) %*% mu : requires numeric/complex matrix/vector arguments

# Same crashes with ES():
ES(R = R, weights = w, portfolio_method = "single", method = "modified", invert = FALSE)
# Error in skewness.MM(w, sigma, M3) : M3 must be a matrix

ES(R = R, weights = w, portfolio_method = "single", method = "gaussian", invert = FALSE)
# Error in t(w) %*% mu : requires numeric/complex matrix/vector arguments

Workaround — supply moments explicitly:

VaR(R = R, weights = w, portfolio_method = "single", method = "modified",
    mu     = apply(R, 2, mean),
    sigma  = cov(R),
    m3     = M3.MM(R, as.mat = FALSE),
    m4     = M4.MM(R, as.mat = FALSE),
    invert = FALSE)
# [1] 0.02442032   <-- works

Root cause

Inside VaR() / ES(), the switch(portfolio_method, ...) has two main branches:

1. Multi-asset branch (portfolio_method != "single") — explicitly auto-computes missing moments before use:

   if (is.null(m3)) m3 = M3.MM(R, as.mat = FALSE)
   if (is.null(m4)) m4 = M4.MM(R, as.mat = FALSE)

2. Single-portfolio branch (portfolio_method == "single") — does not auto-compute moments, yet still passes mu, sigma, m3, m4 (which remain NULL) to mVaR.MM() / GVaR.MM():

   # 'single' branch, method='modified':
   rVaR = mVaR.MM(w = weights, mu = mu, sigma = sigma, M3 = m3, M4 = m4, p = p)
   # mu=NULL, sigma=NULL, m3=NULL, m4=NULL  -->  skewness.MM(w, NULL, NULL)
   #   --> M3.mat2vec(NULL) --> "M3 must be a matrix"
   
   # 'single' branch, method='gaussian':
   rVaR = GVaR.MM(w = weights, mu = mu, sigma = sigma, p = p)
   # mu=NULL  -->  multivariate_mean(w, NULL) = t(w) %*% NULL
   #   --> "requires numeric/complex matrix/vector arguments"

Downstream impact in PortfolioAnalytics

This bug surfaces whenever applyFUN() (in PortfolioAnalytics) calls a function like SharpeRatio() that internally invokes VaR() or ES() with portfolio_method='single' but without passing explicit moments. SharpeRatio has no m3/m4 formals, so even if applyFUN computes higher-order moments and puts them in nargs, pmatch() cannot match them to SharpeRatio's formals, and they never reach the inner VaR/ES call:

# This is what chart.Concentration does when risk.col/return.col don't match extractStats:
applyFUN(R = R, weights = wts, FUN = "SharpeRatio", arguments = NULL)
# -> SharpeRatio(R, weights=...) -> VaR(R, weights=..., portfolio_method='single')
# -> crash: "M3 must be a matrix"

Proposed fix

In the portfolio_method == "single" branch of both VaR() and ES(), add the same moment-auto-computation guard that already exists in the multi-asset branch:

# Before the switch(method, ...) block in the single+weights path:
if (is.null(mu))    mu    <- apply(R, 2, mean)
if (is.null(sigma)) sigma <- cov(R)
if (is.null(m3))    m3    <- M3.MM(R, as.mat = FALSE)
if (is.null(m4))    m4    <- M4.MM(R, as.mat = FALSE)

This mirrors the existing behavior in the multi-asset branch and makes the API consistent: callers should not need to pre-compute moments just to call a single-portfolio VaR/ES.

Additional note: SharpeRatio cannot accept m3/m4

Even if PortfolioAnalytics passes pre-computed moments via applyFUN, SharpeRatio() has no m3/m4 formals and no documented way to forward them to its internal VaR/ES calls. Once the above fix is applied to VaR/ES, this secondary issue becomes moot, but it would still be good to accept (and pass through) m3/m4 in the SharpeRatio ... argument when FUN %in% c("VaR", "ES").