Does anyone have any insights into daily forecasting of exchange rates using time-varying GARCH parameters or just modelling time-varying GARCH parameters. I know it can easily be done using constant parameters, but I find that the predictive power loses strength rapidly.

I don't understand well what you meant, but I'm in the middle of working on that and my supervisor seems convinced that standadizing data dauly using a local changing variance enhaces the predictive stenght of a simple GARCH model. I don't know if this is what you meant.

If you can provide more information as to how you will use GARCH we could be more helpful, however using time varying GARCH model will require you to optimize the time frame you are using. Depending on your optimization algorithm (how and what you optimize) your results might vary substantially.

Thank you for replying. I'm trying to forecast foreign exchange rates out to 12 months on a daily basis. I have built a GARCH(1,1) model from 6 years of daily returns with the GARCH coefficients specified through maximizing the log likelihood of the Student's T-Distribution. The forecast is determined through a Monte Carlo simulation sampling (with replacement) of the historical returns. My concern is: How far into the future is the model valid if the GARCH coefficients are not varying with time, i.e., they are fixed based on the historical data? Is a daily forecast valid out for 12 months?Thank you.

You should read this paper:How Relevant is Volatility Forecasting for Financial Risk Management? Peter F. Christoffersen & Francis X. Diebold.

Why not re-estimate the GARCH parameters every night?Not necessarily suggesting this as the answer, just throwing it out there for comment.

Well for one thing it's embedded in a Monte Carlo sim, which means re-calculating the MLE each day for each iteration, and there are 10,000 iterations. I remember reading an article about re-calculating them every 10 trading days. Maybe I'll go that route if it's programmable through VBA in Excel.

Hi, I´m new so I´m not that advanced, but may be this coul be a part of an answer. Use asymetric GARCH in combination with ARCD proposrd by Hansen. you get time varying parameters that depend on the daily available information on forecasting variables. see also an interesting article:Hueng/McDonald: Journal of Empirical Finance 12, 2005, 666-685

Hi, I´m new so I´m not that advanced, but may be this could be a part of an answer. Use asymetric GARCH in combination with ARCD proposrd by Hansen. you get time varying parameters that depend on the daily available information on forecasting variables. see also an interesting article:Hueng/McDonald: Journal of Empirical Finance 12, 2005, 666-685

Thanks Diana. I'll look it up and don't worry, I'm fairly new at this too.