GARCH forecasting

[SevenSeas]

Hi there

I am using Eviews 6 "student version". I am working on GARCH forecasting. In forecasting, Eviews is supossed to produce two things:

1- The forecasted time series "conditional mean and conditional variance"
2- The evaluations and the graphs.

My problem is that Eviews sometime only gives the forecasted time series and does not produce graphs, but instead it says "square root of negative number". I do not know how to go about this problem.

Thanks

[EViews Gareth]

Could you post an example?

[SevenSeas]

Ok

After estimating the GARCH model for the Dow Jones Index return having the oil return included in the mean and the variance, I click FORECAST. Then, I write the series names such as the forecast name, S.E. and the GARCH using Static Forecast. Before clicking ok, I tick the box of Forecast Graph as an option for the output because I want the graphs of the forecast of variance and and the forecast of the mean with +/- 2 S.E. After clicking OK, Eviews gives me an error saying " square root of negative number" and it does not produce the graphs as a result. However, it worked fine with when using different set of data.

I hope this explanation helps clarifying the problem

Regards

[EViews Gareth]

I meant, could you provide an example workfile?

[SevenSeas]

Hi

I attached the workfile here. The first 300 observations are for estimation, the last 55 is for forecasting

Regards

[trubador]

The problem results from the fact that your garch equation generates negative variances in the forecast period. This is mainly due to change in oil prices, which you use it as an exogenous variable in your variance equation. Although the estimated coefficent value of the "oil" is 0.000767 and is significant for the sample period between 1 and 300, you would estimate an insignificant coefficent value of 0.000322 if you extended the sample to cover the period between 1 and 355. Therefore, either you should consider dropping the oil variable from your variance equation (you already use it in your mean equation) or find a way to ensure the positivity of conditional variance over the forecasting period.

[SevenSeas]

That's true, thanks very much for your clarification

[Lyubov]

The problem results from the fact that your garch equation generates negative variances in the forecast period. This is mainly due to change in oil prices, which you use it as an exogenous variable in your variance equation. Although the estimated coefficent value of the "oil" is 0.000767 and is significant for the sample period between 1 and 300, you would estimate an insignificant coefficent value of 0.000322 if you extended the sample to cover the period between 1 and 355. Therefore, either you should consider dropping the oil variable from your variance equation (you already use it in your mean equation) or find a way to ensure the positivity of conditional variance over the forecasting period.

Hello! I have the same problem. But there is no an exogenous variable in variance equation in my specification of model. And when i try to do static forecast i get the error "square root of negative number".

[EViews Glenn]

With standard GARCH (not EGARCH), this can happen even if you don't have exogenous variables in the variance equation since the specification doesn't impose positivity. It's analogous to doing a linear least squares regression using in-sample positive data, which doesn't guarantee positivity of the out-of-sample forecasts.

[sagga]

I am running into the same issue when running GARCH-M, TARCH-M, CGARCH-M models using the interest rate differential (gbpjpy_fx) as the dependent with a constant and the interest rate differential (gbpjpy_int) in the mean equation with the standard deviation in the mean equation. Can anyone give me advice on how to ensure the positivity of conditional variance for my data? This also happens for the specification (gbpaud_fx c gbpaud_int)

I would really appreciate some guidance. Thanks.

[econworker]

Hi, I have a question regarding the forecasting procedure using a GARCH model. In eviews, If I use a GARCH estimation for the sample size for instance 01/01/2008 to 01/01/2014, and estimate a GARCH model, then I perform a forecasting within this model, for the sample from 01/01/2010 to 01/01/2011...I would like to know that if eviews perform an in sample or out of sample forecast for the series within a GARCH model? as it doesn't require to justify the sample size while changing the forecasting date...

I look forward to hear from you
Thanks

[EViews Gareth]

EViews forecasts over whatever sample you tell it to, whether that is in sample or not.

[econworker]

EViews forecasts over whatever sample you tell it to, whether that is in sample or not.

Thanks, sorry if my question is basic, but would you please tell me what is the difference between:

1. If I my estimation sub-sample is 01/01/2008-01/01/2014 for GARCH estimation, then I perform a forecasting within this model, for the sample from 01/01/2010-01/01/2011,

or

2. If I my estimation sub-sample is 01/01/2008-01/30/2009 for GARCH estimation, and forecast the sample from 01/01/2010-01/01/2011.

the second case is obvious as the sub-sample is divided to estimation sub-sample and forecasting sub-sample, but the first case is not clear for me, as I can not understand how Eviews apply the whole sample data and then forecast a period within that sample. Would you please guide me if in forecasting literature, is this a specific kind of forecasting that I can read about it?

Thanks

[trubador]

You should study the difference between static (in-sample) and dynamic (out-of-sample) forecasting. In-sample forecasting is actually the "fit" of your model. Forecasting of an unknown future (hence the name out-of-sample) is something you deal with dynamic forecasting.

[econworker]

You should study the difference between static (in-sample) and dynamic (out-of-sample) forecasting. In-sample forecasting is actually the "fit" of your model. Forecasting of an unknown future (hence the name out-of-sample) is something you deal with dynamic forecasting.

Thanks, I know about in-sample and out of sample forecasting by definition. Out of sample forecasting is obvious, however, I cant find any source that shows how to perform an in sample technically, for example, in my above example, if I use the whole sample to estimate a GARCH equation and then I forecast for a specific date within that equation, is it an in sample forecasting?

Thank you