Arch Garch Assignment
Arch Garch Assignment
Arch Garch Assignment
9/30/2010
ASSOCIATE PROFESSOR IIM K
Dr. SSS Kumar
GROUP MEMBERS
To examine the efficacy of the different volatility forecasting models in the Indian market.
This paper is about volatility forecasting. It attempts to evaluate the ability of ten different statistical and
econometric volatility forecasting models in the context of Indian stock and forex markets. These competing
models are evaluated on the basis of two categories of evaluation measures- symmetric and asymmetric
error statistics. A lot of papers have been published in the past that have tested volatility forecasting models
in the developed markets. This paper is the first of its kind to find out the relative ability of different
volatility forecasting models in the Indian stock and forex market.
Stock market: closing Nifty index values from 3rd June 1990 to 31st December 2005.
Exchange rate: Indian rupee /USD exchange rate over the period 3rd January 1994 to 31st December 2005.
To make the series stationary the actual closing NIFTY index values are converted into continuous
compounded returns in the standard method as the log differences:
Under Gaussian assumption of homoscedasticity it is assumed that the variance of the error terms
are constant. But in the context of the data that is used for estimating volatility i.e., NIFTY index for
1990-2005, it is very unlikely that the variance of the error term of nifty returns would be a
constant. Hence, it makes sense to consider a model that does not assume that the variance is
constant.
2. Also, as shown in the figure2 there are volatility clustering .Large positive and large negative returns
were observed in 1993 leading to high volatility where as after 1995 onwards we can see that there
is prolonged period of tranquility in the market. This implies that volatility is autocorrelated. Hence,
the concept of conditional variance comes in which takes care of the past volatility.
The above requirements could have been met by applying ARCH also.
But the researcher has used Generalized ARCH (GARCH) model. The reason could be-
The value of the conditional variance so obtained must always be strictly positive. A negative value at any
point in time would be meaningless. That implies a should be >=0 for every i=1,2,3….q but this may not be
true always. Hence, ARCH model suffers from non negativity constraint.
But using the GARCH model it is possible to avoid overfitting and the model is less likely to breach the non
negativity constraint.
Stock market
EWMA 0.27224587 2 0.00182689 2 0.00375781 1 0.43634152 1
GARCH (1,1) 0.36603526 3 0.00205138 3 0.00453581 3 0.66207931 3
GARCH (4,1) 0.18421407 1 0.00181562 1 0.00442270 2 0.44342788 2
Hist mean 0.67727683 8 0.00447397 10 0.00601602 7 1.90841812 10
MA0.25 0.70598849 9 0.00299082 4 0.006142212 8 0.787414084 5
MA1 0.590602663 4 0.00314696 6 0.00561790 4 1.12509497 7
MA3 0.62189710 6 0.00347056 7 0.00576481 6 1.35809469 8
MA5 0.61518752 5 0.00372339 8 0.00573363 5 1.53291022 9
Random Walk 1.00000000 10 0.00314237 5 0.00729752 10 0.73871373 4
Simple reg 0.65787628 7 0.00380932 9 0.00637964 9 1.09204205 6
The volatility is simple terms, for Nifty index, can be defined as a statistical measure of the
dispersion of returns for index. RMSE (Root Mean Square Error) is a good measure of precision or
fitment. A higher value of RMSE indicates the wider difference in outcome of model (GARCH) &
actual values of Nifty index. The RMSE value for GARCH(5,1) is minimum indicating a reduction
in forecast error compared to ARCH & other models used by researcher.
As the variables considered in paper are continuous, MAE(Mean Adjusted Error) can be said as a
good estimator to measure accuracy of forecast. It captures the average forecast error without
considering their signs..MAE and RMSE can range from 0 to ∞. They are negatively-oriented scores
which implies that lower values are better. Outcome of GARCH model has least MAE.
Hence,GARCH model is given rank1 by the researcher to forecast volatility in case of stock and
forex markets.
How Is The GARCH Model Useful In Forecasting Volatility?
GARCH is a model to describe movements in the conditional variance of an error term ,u t,which
may not appear particularly useful. But it is possible to show that
Var(yt yt-1,yt-2…..) = Var(ut ut-1,ut-2,…..)
So the conditional variance of y given its previous values, is the same as the conditional variance of
u, given its previous values. Hence, modeling σt2 will give models and forecast of the variance of yt
as well. Thus if the dependent variable in the regression ,yt is an asset return series,forecasts of σt2
will be forecasts of the future variance of yt. .So, one primary usage of GARCH model is in
forecasting volatility.
CONCLUSION-
The paper concluded that GARCH model is superior to the other mentioned models in forecasting
the volatility in the forex market. The finding is contrary to the findings of Brailsford and Faff(1996)
who found no single method as superior.