Hi all,
Thank you for responses. If any one need the data, I can email it to you. I
don't think I can attach it to R-help. It is only S&P 500 monthly returns I
downloaded from Yahoo finance, with only date and adj. close kept.
Thank you,
Tian
On 11/22/06, T Mu <[EMAIL PROTECTED]> wrote:
>
> Hi all,
>
> I post it on both r-help and r-finance since I don't know where is most
> appropriate for this topic. Sorry if it bothers you.
>
> I did garch fitting on S&P500 monthly returns with garchFit from fSeries.
> I got same coefficients from all cond.dist except normal. I thought that
> is probabaly usual for the data. But when I play with it, I got another
> question.
>
> I plot skew normal with skew = 1 and a standard normal, they overlap
> eachother, so I think they are the same. Skew = 1 means no skewness (I can
> not find the paper defining the distribution).
>
> library(fSeries)
> curve(dsnorm(x, 0, 1, 1), -2, 2, add = F, col = 'red') #skew normal with
> skew 1
> curve(dnorm(x, 0, 1), -2, 2, add = T, col = 'blue') #normal
>
> Then I try them as innovations,
>
> #normal innovation
> garch_norm <- garchFit(series = logr, include.mean = F)
>
> #skew normal innovation
>
> #this line do not include skew, so it got same result as normal
> garch_snorm <- garchFit(series = logr, cond.dist = 'dsnorm', include.mean= F,
> include.skew = F)
>
> #this line includes skew, but use default skew = 1, and it got results
> different from normal, which I don't understand
>
> garch_snorm <- garchFit(series = logr, cond.dist = 'dsnorm', include.mean= F,
> include.skew = T)
>
> Have I done something wrong? I am attaching the code, thank you.
>
> Tian
>
> #GARCH analysis of monthly return
> rm(list=ls(all=TRUE))
> sp500 <- read.csv('s&p_m90.csv', header=TRUE)
> sp500 <- sp500[,2] #only adjusted close
> n <- length(sp500)
>
> logr <- log(sp500[1:n-1] / sp500[2:n])
> acf(logr)
> ar5 <- arima(logr, order = c(5, 0, 0), include.mean = T)
> logr<- ar5$res
> acf(logr)
>
> #fit GARCH distribution
> hist(logr, freq = F, ylim = c(0, 12), breaks = 'FD')
>
> norm_fit <- normFit(logr)
> curve(dnorm(x, norm_fit$est[1], norm_fit$est[2]), -.15, .15, add = TRUE,
> col=2)
>
> t_fit <- stdFit(logr)
> curve(dstd(x, t_fit$est[1], t_fit$est[2], t_fit$est[3]), -.15, .15, add =
> TRUE, col=6)
>
> snorm_fit <- snormFit(logr)
> curve(dsnorm(x, snorm_fit$est[1], snorm_fit$est[2], snorm_fit$est[3]),
> -.25, .15, add = TRUE, col=4)
>
> st_fit <- sstdFit(logr)
> curve(dsstd(x, st_fit$est[1], st_fit$est[2], st_fit$est[3],
> st_fit$est[4]), -.25, .15, add = TRUE, col=3)
>
> library(fSeries)
>
> #normal innovation
> garch_norm <- garchFit(series = logr, include.mean = F)
>
> #t inovation
> garch_t <- garchFit(series = logr, cond.dist = 'dstd', include.mean = F,
> include.shape = T)
> garch_t1 <- garchFit(series = logr, cond.dist = 'dstd', include.mean = F,
> shape = t_fit$est[3], include.shape = T)
>
> #skew normal innovation
> garch_snorm <- garchFit(series = logr, cond.dist = 'dsnorm', include.mean= F,
> include.skew = T)
> garch_snorm1 <- garchFit(series = logr, cond.dist = 'dsnorm', include.mean=
> F, skew = snorm_fit$est[3],
> include.skew = T)
>
> #skew t innovation
> garch_st <- garchFit(series = logr, cond.dist = 'dsstd', include.mean = F,
> include.skew = T, include.shape = T)
> garch_st1 <- garchFit(series = logr, cond.dist = 'dsstd', include.mean =
> F, skew = st_fit$est[4], shape = st_fit$est[3], include.skew = T,
> include.shape = T)
>
> vix <- read.csv('D:/Documents and Settings/Mu Tian/Desktop/8780/8780
> project/vix_m.csv', header=TRUE)
> vix <- (vix[,2]/100) / (12^.5)
>
> plot_sd <- function(x, ylim = null, col = null, ...) {
> xcsd = [EMAIL PROTECTED]
> plot(xcsd, type = "l", col = col, ylab = "x",
> main = "Conditional SD", ylim = ylim)
> abline(h = 0, col = "grey", lty = 3)
> grid()
> }
>
> plot_sd(garch_norm, ylim = c(0.02, 0.13), col = 2)
> xcsd = [EMAIL PROTECTED]
> lines(xcsd, col = 3)
> lines(1:n, vix)
>
> #predict
> predict(garch_norm)
> predict(garch_t)
>
> #demonstration of skew distributions
> #skew normal
> curve(dsnorm(x, 0, 1, .1), -2, 2, add = F, col = 'green')
> curve(dsnorm(x, 0, 1, snorm_fit$est[3]), type = 'l', col = 'blue', add =
> T)
> curve(dsnorm(x, 0, 1, 1), -2, 2, add = T, col = 'red') #normal
>
> #skew t
> curve(dsstd(x, 0, 1, 4, 1), -2, 2, add = F, col = 'red')
> curve(dsstd(x, 0, 1, st_fit$est[3], st_fit$est[4]), type = 'l', col =
> 'blue', add = T)
> curve(dsstd(x, 0, 1, 100, .5), -2, 2, add = T, col = 'green')
>
> #t
> curve(dstd(x, 0, 1, 4), -2, 2, add = T, col = 'red')
> curve(dstd(x, 0, 1, t_fit$est[3]), type = 'l', col = 'blue', add = T)
> curve(dstd(x, 0, 1, 100), -2, 2, add = T, col = 'green')
>
> curve(dsnorm(x, 0, 1, 1), -2, 2, add = F, col = 'red') #normal
> curve(dnorm(x, 0, 1), -2, 2, add = T, col = 'blue') #normal
> curve(dsnorm(x, 0, 1, .1), -2, 2, add = T, col = 'green') #normal
>
[[alternative HTML version deleted]]
______________________________________________
R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
