Why VaR method in PerformanceAnalytics R Package return error "VaR calculation produces unreliable result"

Question

MY R CODE:

library(PerformanceAnalytics)

prices <- c(10.4, 11, 10.11, 9.19, 10.63, 9.68, 12.89, 9.8, 12.57, 8.23, 9.27,
            9.51, 10.51, 9.66, 9.52, 10.78, 9.47, 11.87, 11.33, 11.38, 11.16, 
            8.94)

returns <- diff(prices)

print(returns)

VaR(returns, p=.95, method="historical")
VaR(returns, p=.95, method="gaussian")
VaR(returns, p=.95, method="modified")

---

THE OUTPUT:

print(returns)
 [1]  0.60 -0.89 -0.92  1.44 -0.95  3.21 -3.09  2.77 -4.34  1.04  0.24  1.00 -0.85 -0.14  1.26
[16] -1.31  2.40 -0.54  0.05 -0.22 -2.22

VaR(returns, p=.95, method="historical")
VaR calculation produces unreliable result (risk over 100%) for column: 1 : 3.09
    [,1]
VaR   -1

VaR(returns, p=.95, method="gaussian")
VaR calculation produces unreliable result (risk over 100%) for column: 1 : 3.03916083148501
    [,1]
VaR   -1

VaR(returns, p=.95, method="modified")
VaR calculation produces unreliable result (risk over 100%) for column: 1 : 3.1926697487747
    [,1]
VaR   -1

---

The first argument is described in help as "an xts, VECTOR, matrix, data frame, timeSeries or zoo object of asset RETURNS"

What is the problem? Where is the error?

Answer 1

Try calculating log returns and then VaR

returns <- diff(log(prices))
VaR(returns, p = 0.95, method = "historical")

Also check the ES (Expected Shortfall) function from PerformanceAnalytics lib. From my personal opinion is way better than VaR.

Answer 2

It also worked when considering percentage returns.

returns <- rep(NA, length(prices) - 1)
for (i in 1:length(returns))
  returns[i] <- (prices[i+1]-prices[i])/prices[i]