GARCH(1,1) Forecast Series

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

horror
Posts: 7
Joined: Sun Dec 28, 2014 8:00 pm

GARCH(1,1) Forecast Series

Postby horror » Sun Dec 28, 2014 8:14 pm

Hello everyone, I am having some major problem, which I tried to solve for several days now and read through the complete forum already.

i downloaded daily trading data of the DAX (german index), i have ~1000 oberservations from end 2010 - end 2014
i want to estimate a GARCH model based on end 2010 - end 2013, then i want to forecast end 2013 - end 2014 and compare then to the real values
- imported 750 of the data to eviews (close) unten end of 2013
- transformed them into log returns: rendite = dlog(close)
- fitted a GARCH(1,1) model, which at first didnt work, then i replaced some missing NA values in the series by the surrounding values (rendite1 is the filled up log returns series)
- i have the garch11 model now and with this i want to forecast the values (logreturn values that is) for end 2013 - end 2014 via proc -> forecast

the last part just is not working, i tried first to expand the range of the file from end 2013 to end 2014 and then tried several combinations, sample range, dynmaic, static, every combination etc, i just dont know to enter, most of the time it just delivers a series of NA
from what i read up, i want a dynamic forecast, since the forecast values are supposed to be "unknown" as of yet, i just pretend that i want to predict the future
but about the rest, i am completely lost, so please help!

if you could help me out with a small step by step guide that would be really helpful!
Attachments
dax_model.wf1
(38.48 KiB) Downloaded 698 times
Last edited by horror on Mon Dec 29, 2014 4:26 am, edited 1 time in total.

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby EViews Gareth » Sun Dec 28, 2014 8:51 pm

Resize the workfile to include the observations until end of 2014 (Proc->Resize).

Open up your equation, hit the forecast button, change the forecast sample to be 12/23/2013 12/23/2014, fill in a name for the GARCH series and hit ok.
Follow us on Twitter @IHSEViews

horror
Posts: 7
Joined: Sun Dec 28, 2014 8:00 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby horror » Sun Dec 28, 2014 9:00 pm

hey Gareth, thx a lot for ur reply!

i did just what u told me, i resized the observations

then i clicked on my garch11 equation, proc->forecast, changed the forecast sample to 12/23/2013 12/23/2014
then i might be unsure, but still tried both:
- named the forecast rendite1_forecast
- entered a name for GARCH(optinal) / did not enter a name here -> same result

checkmarks are on at:
[x] insert actuals for out-of-sample-obs
[x] dynamic forecast
[x] coef uncertainty in SE calc

and both outputs


i dont get an N/A anymore, but i get a value of "0.000481" in every estimated field from 12/23/2013 to 12/23/2014, which seems to be somewhat off or sth is missing?
so something seems to be wrong?

to help it, i did a quick screenshot, whereby the blue marked area is the one, where i tried within anything in it
in the left background are the results which i keep getting
Image

(btw if its any help, using EViews 8 Student Version)

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby EViews Gareth » Sun Dec 28, 2014 9:25 pm

open the series called GARCH_OPTIONAL
Follow us on Twitter @IHSEViews

horror
Posts: 7
Joined: Sun Dec 28, 2014 8:00 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby horror » Sun Dec 28, 2014 9:32 pm

this one seems kinda wrong aswell :(

it starts at 0.000108 and is constantly increasing 0.000001 per day?

just to state it again, i am looking for a forecast of my "rendite" series of which is fitted the GARCH(1,1) model, so at the end i can a draw a combined graph of the actual rendite series and the forecasted series and compare them

but right now i am just getting a flat line as a graph and not really any forecast which seems correct, i don't know why that is? :(

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby EViews Gareth » Sun Dec 28, 2014 9:41 pm

Sorry, I thought you wanted the GARCH series. Your first one was the forecast of the underlying series. You would expect the forecast to be a constant value.
Follow us on Twitter @IHSEViews

horror
Posts: 7
Joined: Sun Dec 28, 2014 8:00 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby horror » Sun Dec 28, 2014 9:47 pm

EViews Gareth wrote:Sorry, I thought you wanted the GARCH series. Your first one was the forecast of the underlying series. You would expect the forecast to be a constant value.


hm why is that?

i am writing a paper with a very close deadline, merely minutes, not even hours anymore

i modelled a garch modell and tested it already

last thing i need to do know is to forecast the series and compare the forecast to the real values

from my point of view a constant is not really from i am looking for, then it's a pretty bad forecast and i wouldnt even need a gargh model for that? i could just model around the mean?

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby EViews Gareth » Sun Dec 28, 2014 9:47 pm

You have nothing but a constant in your mean equation. What did you expect the forecast of the mean equation to be other than a constant?
Follow us on Twitter @IHSEViews

horror
Posts: 7
Joined: Sun Dec 28, 2014 8:00 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby horror » Sun Dec 28, 2014 9:52 pm

if i do it like this, am comparism is kinda useless

Image

top graph is real data

bottom graph is 12/23/2010 - 12/23/2013 real data, 12/23/2013 - 12/23/2014 forecast aka flatline

i dont know where my mistake is, but this is barely the forecast i was lookin for?

horror
Posts: 7
Joined: Sun Dec 28, 2014 8:00 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby horror » Sun Dec 28, 2014 9:54 pm

horror wrote:if i do it like this, am comparism is kinda useless

Image

top graph is real data

bottom graph is 12/23/2010 - 12/23/2013 real data, 12/23/2013 - 12/23/2014 forecast aka flatline

i dont know where my mistake is, but this is barely the forecast i was lookin for?



maybe i am looking at the wrong thing at the moment, maybe i am not looking at a mean forecast, i am looking for a time series forecast, but dont really know where to look for it or what to look for?
i want something that forecasts me the bottom straight line, am i looking for the wrong thing?

EViews Gareth
Fe ddaethom, fe welon, fe amcangyfrifon
Posts: 13307
Joined: Tue Sep 16, 2008 5:38 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby EViews Gareth » Sun Dec 28, 2014 10:17 pm

You're doing nothing wrong in EViews. The straight line forecast is the correct forecast for your model. If you don't like the straight line forecast, you'll have to choose a different model.
Follow us on Twitter @IHSEViews

horror
Posts: 7
Joined: Sun Dec 28, 2014 8:00 pm

Re: GARCH(1,1) Forecast Series URGENT

Postby horror » Sun Dec 28, 2014 10:25 pm

am i looking for a volatility forecast then maybe? or something combined?

Kristof1992
Posts: 1
Joined: Tue Jun 30, 2015 1:47 am

Re: GARCH(1,1) Forecast Series

Postby Kristof1992 » Wed Jul 01, 2015 1:47 am

What do you mean by choose a different model ? I for example want to use the forecast of the conditional variances to apply them to a Value at risk model. If this would mean that the forecast is a constant, my Value at risk calculations will be pretty useless.

trubador
Did you use forum search?
Posts: 1518
Joined: Thu Nov 20, 2008 12:04 pm

Re: GARCH(1,1) Forecast Series

Postby trubador » Wed Jul 01, 2015 4:22 am

Kristof1992 wrote:What do you mean by choose a different model ? I for example want to use the forecast of the conditional variances to apply them to a Value at risk model. If this would mean that the forecast is a constant, my Value at risk calculations will be pretty useless.

The discussion above is relevant to mean equation. Conditional variance estimation is the second part of a GARCH model. Since it is autoregressive, you can generate out-of-sample forecasts. But those values will (and should) eventually converge to a constant value. For your VaR model, one step ahead forecasts would be most useful as you'll run the model each day and generate a forecast for the next. If you need n-step ahead forecasts, you'll have to simulate future values of conditional variance as the random shocks of the future are unknown.

Econoforecast
Posts: 25
Joined: Sun Feb 19, 2017 1:25 pm

Re: GARCH(1,1) Forecast Series

Postby Econoforecast » Mon Feb 20, 2017 6:29 pm

trubador wrote:
Kristof1992 wrote:What do you mean by choose a different model ? I for example want to use the forecast of the conditional variances to apply them to a Value at risk model. If this would mean that the forecast is a constant, my Value at risk calculations will be pretty useless.

The discussion above is relevant to mean equation. Conditional variance estimation is the second part of a GARCH model. Since it is autoregressive, you can generate out-of-sample forecasts. But those values will (and should) eventually converge to a constant value. For your VaR model, one step ahead forecasts would be most useful as you'll run the model each day and generate a forecast for the next. If you need n-step ahead forecasts, you'll have to simulate future values of conditional variance as the random shocks of the future are unknown.

trubador wrote:
Kristof1992 wrote:What do you mean by choose a different model ? I for example want to use the forecast of the conditional variances to apply them to a Value at risk model. If this would mean that the forecast is a constant, my Value at risk calculations will be pretty useless.

The discussion above is relevant to mean equation. Conditional variance estimation is the second part of a GARCH model. Since it is autoregressive, you can generate out-of-sample forecasts. But those values will (and should) eventually converge to a constant value. For your VaR model, one step ahead forecasts would be most useful as you'll run the model each day and generate a forecast for the next. If you need n-step ahead forecasts, you'll have to simulate future values of conditional variance as the random shocks of the future are unknown.

Hi Trubador.

Hope you're well. Have been working on a very similar problem was hoping for some clarification. I have read quite a lot to try plug in some gaps in knowledge and pick up the basic fundamentals, therefore any assistance from here would be a huge help, but please do not lose patience with me as I'm still very much learning. Any comments at the different stages are much appreciated.

I wish to forecast the volatility of stock market daily returns using various different models (Historical, ARCH, GARCH, EGARCH etc) to see which can forecast volatility best (i.e. least error in terms of RMSE, MAE etc).

I downloaded daily closing pricesfrom 5/31/1996 to 6/01/2016, constructed a daily returns series as well as squared returns and absolute returns. I have a fairly reasonable understanding of the tests I need to perform on the series such as LM, normality, test for ARCH effects etc.

I have estimated a GARCH(1,1) model which I assume is correct (lagged dependent variable of itself). Attached is a screenshot below with the selections and output.I have used the period 5/31/1996 to 6/01/2006 as my in-sample estimate of the model.

Image

Image

I then attempt the final part (one-step ahead forecasting) which is worrying me a little as to whether i've done everything right. First of all based on some of the exchanges i've read on this forum by yourself and Gareth on this topic I am aware that to running a static forecast gives you the fit of your estimated model and is used for in-sample forecasting as it uses actual values.

Therefore, forecasting out-of-sample is done with dynamic forecasting. I attempted to do this for the period 6/02/2006 to 6/01/2016. I wish to forecast volatility for this period and then run a plot of squared daily returns against my forecasted out of sample values to see the visual differences. However upon hitting forecast, setting my sample period to 6/02/2006- 6/01/2016 and entering a name in the Garch(optional) field to generate the volatility series, the output is rather confusing to me. It seems to be just a straight line which seems to converge to a number with each daily observation. Attached screenshot below.

Image

What am I doing wrong? Or am I doing it right, it certainly doesn't seem right when plotted, whereas static forecasting as seen below in the screenshot is akin to what I would have expected. Happy to attach the worfile if you need to confirm anything as well.

Image

Appreciate if any one else can offer assistance as well for this question, or at the very least redirect me to any useful GARCH modelling for e-views books or videos, preferably videos.


Return to “Estimation”

Who is online

Users browsing this forum: No registered users and 35 guests