Question

I am try to fit an eGARCH model on an expanding basis using the rugarch package. I have 6 columns of data, and I am trying refit parameters ~6000 for each column. If I run the following code, I get an error in windows on the 2nd column (this means I am getting all the way through the first inner loop succesfully). By using gc() within the loop and removing the fitted object I have extended the length of time it takes to hit the memory error. Also, this process takes a very long time in general and I am wondering if there is anyway to improve it on my end. The package itself seems to be written pretty efficiently with most of the filtering being done in low level C. I could probably refit the model every 30-60 days, but I would really prefer to do it this way. I am running R 2.13.2 on 32-Bit windows. Thanks in advance. Edit: The error is: “The instruction at “0x6c732a07” referenced memory at “0x00000008.” The memory could not be “read””.

library(rugarch)
library(xts)
e.spec <- ugarchspec(variance.model = list(model = "eGARCH", garchOrder = c(1,1)),   mean.model = list(armaOrder = c(1,0), include.mean = TRUE)) 
dly.xts <- xts(matrix(rnorm(8000*6), nrow = 8000, ncol = 6), as.Date(1:8000))
tst.xts <- tail(dly.xts, 6000)
names(tst.xts) <- 1:6
tst.idx <- index(tst.xts)
dly.idx <- index(dly.xts)
for(j in 1:ncol(tst.xts)){
     sig.est <- rep(NA, nrow(tst.xts))
    for(i in 1:nrow(tst.xts)){
        print(i)
        dat <- dly.xts[dly.idx <= tst.idx[i], j]
        fit <- try(ugarchfit(e.spec, data = dat[-nrow(dat), ], solver = "solnp", solver.control = list(trace = FALSE)))
        if(class(fit) != "try-error"){
            spec.new <- ugarchspec(variance.model = list(model = "eGARCH", garchOrder = c(1,1)), mean.model = list(armaOrder = c(1,0), include.mean = TRUE), fixed.pars = coef(fit))
             sig.est[i] <- as.numeric(tail(sigma(ugarchfilter(spec = spec.new, data = dat)),1))
            rm(spec.new)
            rm(fit)
            gc()
        }else{
            sig.est[i] <- NA
        }
    }
    save(sig.est, file = paste("egarch", names(tst.xts)[j], ".RData", sep = ""))
}

Answer 1

When changing the data types from xts to numeric, the problem went away and the speed of processing increased dramatically. (seems obvious in hindsight)