Practical GARCH Choices for Volatility Filtering and Risk Modeling

# What is the preferred GARCH method in practice?


# What is the preferred GARCH method in practice?












My advance apologies, if this question is too naive or basic. Please be patient with my first experiences with SE; ask for clarification, if needed.

I recognize there are many (often-criticized) flavors in the univariate (and multivariate) GARCH methods. Univariate GARCH varies from standard GARCH to GJR, AP-GARCH, etc. Multivariate world offers natural generalization (hundreds of parameters to estimate) to GO-GARCH, DCC-GARCH, and alike simplifications.

Yet, I want to know what the industry actually uses for bond, stock, and derivative analysis.

It'd be great to see concrete references, personal employment experiences, and practical work examples that I can follow, instead of opinions and speculations.

UPDATE: I still monitor this post for answers...

## Answer by Kiwiakos (score 7)

https://quant.stackexchange.com/a/21738

I personally use the simple Garch(1,1) for volatility filtering in the risk management area.

In fact in most cases I don't even estimate the parameters, I stick 0.94 for mean reversion, 0.04 for the squared error and I get the constant by matching the series variance. My experience is that there is no point pretending to finetune parameters when vol is unobserved anyway.

EDIT

To add some more color here are some actual examples, without going into details as these are proprietary models:

- Calculate risk for a portfolio under historical simulation. Used for regulatory and economic capital calculations. Fix portfolio positions, calculate P&L based on one-day shocks over the last couple of years, then use Garch to forecast tomorrow's VaR. I have used parametric (Gaussian/ StudentT) and empirical versions, using fixed parameters as above. Although is outperforms other methods, it does not go down well with business as it gives VaR which is too volatile for limit setting etc.

- Produce vol time series when none can be sourced. Say that you want to build a historical time series for implied vols. I have used Garch, among other techniques, to do that. As Garch filters 'true vol' these have to be converted to 'implied vol'.

- Produce long term simulations (say 10 or 20 years) of volatilities for IMM/ counterparty risk calculations. As volatilities are globally integrated, we do not want them to move out of sync as horizons increase. Therefore used multivariate versions where one vol (say US) enters the propagation mechanisms of others (say UK, Germany, etc).



Hope that this helps

## Answer by owner (score 3)

https://quant.stackexchange.com/a/22033

Interesting question, as All the answers (including mine) could not be generalized unfortunately. As far as I am concerned, I use a univariate EGARCH for risk modelling purposes (Filtered Historical Simulation (FHS), etc.).

1 - EGARCH, merely because GARCH models do not take into account so-called leverage effects, which is crucial to me for skewed and leptokurtic data.

2- Univariate rather than multivariate processes, for simplicity purposes.

** I am not generally in favor of using static parameter estimates (such as the notorious `lambda = 0.94`), but it's fine to me if the people resorting to these approaches obtain reliable estimates while backtesting their models. I prefer sticking with the `mle` estimates which are nowadays quick to perform.

PS: I only use univariate GARCH in conjunction with Alpha - Stable distributions for my risk metrics, when I think the outlook is highly stressed. It's just for safety measures :).

Hope it helps