- wilhelmsson
**Posts:**5**Joined:**

"There isn't that many paper that extensively talk about how to analysis volatility forecasts."There has been a rather active and I think sucessful research (but not so extensive as you say) agenda in evaulating volatility forecasts the last 8-10 years. To give the major contributions Andersen, T.G. and T. Bollerslev (1998) Answering the skeptics: yes standard variance models do provide accurate forecasts International Economic Review, 39: 885-905. This paper shows why you can't (or should not) use e.g. squared daily returns to evaluate 1-day ahead volatility forecast.Hansen, P., A. Lunde (2006) Consistent Ranking of Variance Models Journal of Econometrics, 131: 97-121 and Patton, A. (2006) Volatility Forecast Valuation and Comparison Using Imperfect Volatility Proxies Quantitative Finance Research Papers 175, University of Technology, Sidney. Shows what loss functions to use and what could happen when you use the "wrong" loss functions in conjunction with replacing the true volatility with a proxy. This is something you always have to do since the true vol. is not observable. A cool way to evaluate a lot of different volatility forecasts is given in Hansen, P., A. Lunde and J. Nason (2003) Choosing the Best Volatility Models: The Model Confidence Set Approach. Oxford Bulletin Of Economics and Statistics, 65: 839-861. On the empirical side Poon, S. and C. Granger (2003) Forecasting Variance in financial markets: A review Journal of Economic Literature, 41: 478-540. review 93 different studies dealing with volatility forecasting. Another interesting empirical paper is Hansen P, Lunde A. 2005. A Forecast Comparison of Variance Models: Does Anything Beat a GARCH (1,1)?. Journal of Applied Econometrics 20: 873-889. This paper compares 330 different models to the GARCH(1,1) model and does not find any conclusive evidence for other models to be better (however none of the 330 models use intraday data).

- wilhelmsson
**Posts:**5**Joined:**

QuoteOriginally posted by: exotiqHere's a general question for the ARCH gurus: do these models stop at variance, or can they be generalized to handle higher moments, such as time varying skew and kurtosis of a time series?Yes the first generalisation was made by Bruce Hansen (1994), he proposes what he calls the conditional density model, this model is GARCH like and has conditonal variance, skewness and kurtosis. The code (in Gauss) is even available from Bruce's homepage http://www.ssc.wisc.edu/~bhansen/progs/ier_94.html. Other models with time varying conditonal skewness and/or kurtosis are Brännäs and Nordman (2003a,b), Níguez and Perote (2004) as well as Lanne and Saikkonen (2005) among others. A general problem with these models are that they require several thousand observations to estimate the parameters with any precision. On the other hand you may not have to be so worried from a practical perspective of non-stationarity in the data since you have a time-varying distribution.Complete references Brännäs, K., and N. Nordman. An alternative conditional asymmetry specification for stock returns. Applied Financial Economics, 13 (2003a), 537-541.Brännäs, K., and N. Nordman. Conditional skewness modeling for stock returns. Applied Economics Letters, 10 (2003b), 725-728.Hansen, B. Autoregressive conditional density estimation. International Economic Review, 35 (1994), 705-730.Lanne, M., and Saikkonen, P. Modeling conditional skewness in stock returns. Unpublished Working paper, European University Institute Working Paper ECO No. 2005/14. (2005).Ñíguez, T., and J. Perote. Forecasting the Density of Asset Returns. Unpubl. working paper. London School of Economics and Political Science, London. (2004).

- wilhelmsson
**Posts:**5**Joined:**

QuoteOriginally posted by: DileepHi all,Well i am a newbie to this group. And I am currently using GARCH to estimate the volatility of NIFTY (India) with intraday data. I have a few doubts regarding my work.1.What is generally good time horizon to take as sample when i have a data as fine as multiple values for each second?The answer depends on if you want to sample to estimate the GARCH model or if you are talking about what frequency to use for constructing the ex post variance. If you consider estimating GARCH on intraday data you will have to deal with the strong intrday (U-shaped) pattern of volatility during the day. It this is ignored your GARCH parameters will be rubbish. As usual Andersen and Bollerslev (97) have a paper about this. If you want to estimate on intra-daily you are probably better off using ARFIMA type models directly on the realized volatility see e.g. Andersen et al. (forthcoming). There is a very active litterature of how often to sample your process in the pressence of market microstructure noise. If you want to ignore this go with 5-15min which is more or less standard. If you want to be more fancy look at e.g. Aït-Sahalia et al. (2005), Oomen (2005) as well as Hansen and Lunde (2006)2.Is there any better criterion to evaluate the models apart from RMSE and AIC? I need to compare the models to arrive at the best frequency (i am varying time gap as 1 min,3min,5min,10min and 15 min)You may be interested in "Forecasting financial market volatility: Sample frequency vis-a-vis forecast horizon" by Torben G. Andersen, Tim Bollerslevand Steve Lange Journal of Empirical Finance, 1999, vol. 6, issue 5, pages 457-477 RefsAït-Sahalia Y, Mykland, P, Zhang L. 2005. How often to sample a continuous-time process in the presence of market microstructure noise. Review of Financial Studies 18: 351-416. Andersen and Bollerslev (1997) "Intraday Periodicity and Volatility Persistence in Financial Markets," (with Tim Bollerslev), Journal of Empirical Finance 4 (1997): 115-158. Andersen Tim Bollerslev, Francis X. Diebold and Paul Labys "Modeling and Forecasting Realized Volatility,"EconometricaHansen P, Lunde A. 2006a. Realized Variance and Market Microstructure Noise Journal of Business and Economic Statistics 24: 127-218.Oomen R. 2005. Properties of bias-corrected realized variance under alternative sampling schemes. Journal of Financial Econometrics 3: 555-577.

- wilhelmsson
**Posts:**5**Joined:**

QuoteOriginally posted by: exotiqSo what is a better way than my #5 to practically fit a GARCH model/estimate its parameters? All the material I have seen either doesn't go into that step. Is there any open C++ or Mathematica code that I could look at which estimates GARCH parameters that I could look at? The packages that do it seem to be black boxes.The reasons I am looking for a simple, open algorithm is that:1.) I'd like to be able to write a prectical GARCH lesson on a postcard2.) I'd like to extend GARCH to handle time-varying higher moments, like skew, kurtosis, and co-skew.Here is some matlab GARCH code that may help http://www.kevinsheppard.com/research/u ... garch.aspx

- wilhelmsson
**Posts:**5**Joined:**

QuoteOriginally posted by: BilboIs there any good place to find some kind of GARCH manual??One place for a soft GARCH start is http://marshallinside.usc.edu/simrohoro ... rch101.pdf . This is an article by Rob Engle (the inventor or the (G)ARCH model, called GARCH 101: The Use of ARCH/GARCH Models in Applied Econometrics

Last edited by wilhelmsson on October 4th, 2006, 10:00 pm, edited 1 time in total.

Thank u very much wilhelmsson, for the info. 1 doubt though, The U-shape that you mentioned will come into play if i use the model to forecast for lets say next 10 min or 15min rather than 1 day. Do i still have to remove this effect if i am trying to find out the frequency at which the model can foreacst best for 1 day?

What's the recommened period to look at when using Garch process??What would be the recommenedation if I'm looking at minute data? a few days? do I need to go back to the start of the year or all the data I can get?

anybody knows that how garch-types code of http://www.kevinsheppard.com/research/u ... garch.aspx can work? I did try so many times in matlab, but it always has the error. I don't what reason it does not work. if anyone has the same experience and finally get successfully to work, please let me know! thanks a lot!!!!!!!!!!

Hi

Last edited by Owais on April 27th, 2008, 10:00 pm, edited 1 time in total.

Are there any concerns using normalized returns (returns/predictedvolatility) and checking stdev of normalized return series (in and out of sample) to verify performance of volatility model (GARCH, EWMA, etc)? Most of the other tests I seen in the papers referred here use realized volatility (or squared returns), that might be biased depending on how you calculate it (ex are you going give equal weights to return history, if not your choice of weights will affect realized volatility calculation etc)

- JatinVerma
**Posts:**1**Joined:**

I am new to GARCH and have some queries about GARCH processes.How do we arrive at values for P and Q?How do we decide the amount of data points i.e e.g:10 years for long term, 1 year and then recent 3 months again ( as discussed in GARCH101)?Is Bayesian updating used for these calculations?

Can anybody help me out working with conditional variance single step prediction in Matlab?I have fitted a garch model using the garchfit function. Once done, I intend to use it to predict one observation forward and try to plot it against the real series. Can anybody suggest me a function or methodology to do this.My repeated attempts at using garchpred() leaves me with conditional variance asymptotically approaching unconditional variance.Any help would be greatly appreciated.