Hi, I am very new to financial maths, but I have decided to do a project on it in my final year of uni. I am having problems with estimating parameters for the GARCH(1,1) model in MATLAB, my data set is closing prices of companies in the FTSE 100. When I run garchfit on the data the Garch(1) coefficient comes out as zero. But when I run garchfit on the examples given in the MATLAB help they give a non-zero Garch(1) coefficient. Can anybody help or explain why the coefficient is coming out as zero for my data ?Thankyou for the help.James

jsyardley, Maybe you should apply the toolbox to the changes in prices and not the closing prices. Second, usually when one applies a variance model, one needs to specify a mean model as well, for example, an AR(1). So the question is, what variance are you modeling?

I am using the change in prices. I thought the function garchfit had a default setting for the mean and variance models ? I have also attempted to estimate the parameters using excel and solver and that is also giving me Garch(1) or beta coefficient as zero. I cant figure out what is going wrong.Thanks.

try R, there are several packages with garch-related functions.

You should first difference your data, i.e., the closing prices, to get the log returns of the data to make them startionary and then your GARCH model coefficients are likely to be more meaningful.I have used MATLAB to estimate GARCH models previously and have found it to be quiet straight forward and reliable. The only time I got spurious results is when there were no structural breaks in the data. How long does you data set run for? Are they daily, weekly, monthly or yearly observations?

You should first difference your data, i.e., the closing prices, to get the log returns of the data to make them startionary and then your GARCH model coefficients are likely to be more meaningful.I have used MATLAB to estimate GARCH models previously and have found it to be quiet straight forward and reliable. The only time I got spurious results is when there were no structural breaks in the data which meant that GARCH was not an appropriate model to use due to a lack of heteroskedasticity in the data. How long does you data set run for? Are they daily, weekly, monthly or yearly observations?

I was using the log returns of the closing prices and my data range is three months of daily closing prices. I have now increased my data range to 5 months and it seems to be working ok now. Thankyou for your help though.James

Did you run ARCH test on the 3 months data to check for the presence of heteroskedasticity? If the test shows no heteroskedasticity then that would point to the inadequacy of using a GARCH model and an ARMA model would be more appropriate.I think by the fact the you now have 5 months of observeations you are more inclined to have heteroskedasticity in your data and thus GARCH parameter coeficients are now significant.

rtlee 100! could you please contact me at cristina.strajescu@gmail.com? i'm doing a project on DCC and volatility forecasting and i need some help. i was thinking that maybe you'll be able to give me some hints.P.S. i don't have an account on wilmott.com so please don't be surprised!