Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Value at Risk Various Models and Its Ap

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

CASIRJ Volume 10 Issue 4 [Year - 2019] ISSN 2319 – 9202

“Value at Risk”: Various models and its applications


Sumant Kumar
E.mail:sumantnandan@gmail.com

Abstract
“Value at Risk” has been one of the prime tools for measurement of risk since its inception.
Fundamental building blocks of VaR model have been identified and different combinations of
components have been applied in the Indian context. Along with simple VaR model, some
complicated VaR models have also been back tested in Indian Scenario.

JEL Classification: C0; C1; C2; C3; G0


Key Words: Value at Risk, Model Building Approach, GARCH, EWMA, Copulas,
Co-relations, Univariate and Bivariate Time series, Back testing of Models

1. Introduction
As a measure of Risk, “Value at Risk” has been one of the prime tools in financial
industry. Value at Risk (VaR) is a quantitative measure of risk, which exhibits that
over a given period (say in 10 days period) the loss amount (VaR) is expected to
exceed only y% (say 1 %, i.e. 100-X, where X=99%) of the time.
Value at risk was first designed by Dennis Weatherstone, the erstwhile Chairman of
J.P. Morgan & Co. This was a one page report that could measure the total risk faced
by J.P.Morgan & Co. globally.
2. Approaches to Calculate VaR
There are various approaches to calculate the VaR. Some of these are as below:
a. Monte Carlo Simulation: Monte Carlo Simulation is used to generate the
probability distribution for value change (in rupee terms) in the
investment of a portfolio in one day, one week or any specific time
intervals.
b. Historical Simulation: Historical simulations represent the simplest way of
estimating the Value at Risk for many portfolios. In this approach, the VaR
for a portfolio is estimated by creating a hypothetical time series of
returns on that portfolio, obtained by running the portfolio through
actual historical data and computing the changes that would have
occurred in each period on each market risk factor.

International Research Journal of Commerce Arts and Science


http://www.casirj.com Page 141
CASIRJ Volume 10 Issue 4 [Year - 2019] ISSN 2319 – 9202

c. Model Building Approach: This process map the risk in the individual
investments in the portfolio to more general market risks, and then
estimate the measure based on these market risk exposures.

Various Components of Model building for VaR:


1. Expected return during the period
2. Volatility of returns and its clustering
3. Time duration (1 Day, 10 Days, One Week)
4. Confidence level (99.5%, 99%, 95%)
5. Type of Distribution of return
6. Portfolio weights of assets within the portfolio
7. Dependencies and correlation among the securities in case of bi-
variate or multivariate (when more than one assets are in portfolio)
time series, Copulas (Normal copula, t-Copula)
3. Models of VaR
Models for VaR have evolved a lot since its inception. The components of the
models as enumerated above are changed according to the application of VaR. There
are three basic components viz. Volatility, Stock returns and Co dependence are
used, each with many different models as below:
1. Volatility: GARCH, EGARCH, EWMA
2. Stock returns : Normal Distribution, t-distribution
3. Co dependence: Normal Copula, t- Copula

The Combinations of tests may be:


1. GARCH-Normal Distribution-Normal Co-relation (G-ND-NC)
2. EGARCH- Normal Distribution-Normal Copula (EG-ND-NC)
3. GARCH- t Distribution- t Copula (G-tD-tC)
4. EGARCH- t Distribution-t Copula (EG-tD-tC)
In the extant study the first and common combination has been used and tested with
GARCH-Normal Distribution-Normal Co-relation (G-ND-NC)
4 . Selection of Data:

International Research Journal of Commerce Arts and Science


http://www.casirj.com Page 142
CASIRJ Volume 10 Issue 4 [Year - 2019] ISSN 2319 – 9202

Thus testing the VaR model with above components, two assets were selected from
NSE traded Stocks. Two Stocks – namely the prices of L&T and LUPIN have been
taken from 30th October, 2015 to 28th January, 2016
5 . The methods & Models used:
1. Volatility was estimated, which were based on the prices of respective stock
prices for 60 days. This estimation was done using GARCH(1,1) method.
2. For calculation of volatility, based on the GARCH (1,1) method, the mean and
variance for each day each underlying was calculated.
3. Then, using Maximum Likelihood Method, the value of a, w & b was found.
For maximization, the algorithm with SOLVER of Excel was used. It gives
localized maximization for the defined equation. After finding w,a & b, the
values of y was taken and then long run volatility, daily volatility and annual
volatility was estimated.
4. The Co-relation coefficient between the stocks was calculated with COREL
function with the exponential weight of the time series. As the time series
progresses, the older periods have also considered, thus forming a reliable
co-relation.
5. Further the Standard deviations of both the stocks and the standard
deviation for portfolio were calculated and the VAR of Portfolio calculated.

GARCH(1,1)
In order to estimate the historical volatility, GARCH(1,1) model was used. A study
of comparing volatility models done by Hansen and Lunde, using GARCH(1,1) as
benchmark, had as result that the best models do not provide a significantly better
forecast than the GARCH(1,1) model. For this reason GARCH(1,1) is preferred here
above all family of GARCH models.
At present, there have been many models developed to determine volatility, and
some of them act as alternatives or improvement from earlier models. The family of
GARCH models is an example, starting from the autoregressive conditional
heteroskedasticity (ARCH) model of Engle (1982).
The least square approach assumes that the squared errors have the same magnitude
across the entire dataset. This assumption is known as homoskedasticity. But these
data, having periods of high and low volatility, cluster together. This is known as
Heteroskedasticity. In reference to modeling fitting, this means the residuals vary

International Research Journal of Commerce Arts and Science


http://www.casirj.com Page 143
CASIRJ Volume 10 Issue 4 [Year - 2019] ISSN 2319 – 9202

in magnitude. Volatility clustering means the data is auto correlated. GARCH is a


statistical tool that helps predict the residuals in vast data
ARCH means Autoregressive Conditional Heteroskedasiticy and is closely related
to GARCH. The simplest method to predict stock volatility is an n day standard
deviation If we want to predict stock prices for the next day, the mean is usually a
safe starting point.
But the mean treats each day with the same weight. Giving the recent past more
significance is more logical, with perhaps an exponential weighted average being a
better method to predict tomorrow’s stock price (EWMA).
However, this method does not capture any data older than a year, and the weighting
is rather arbitrary. The ARCH model, however, varies weights on each residual such
that the best fit is obtained. The GARCH (General Autoregressive Conditional
Heteroscedasiticy) is similar, but gives recent data more significance.
The GARCH(p,q) model has two characteristic parameters; p is the number of
GARCH terms and q is the number of ARCH terms. GARCH(1,1) is defined by the
following equation.
n= V +u2n-1 + n-1
and where V=w
and where  ++ 
The estimated volatility then calculated from this GARCH (1,1) model. Further, for
estimating the w, & for the above GARCH (1,1) model, maximum likelihood
method was used with the formula:
(-ln(u2n/
Then for maximization of the above summation, solver from excel was used with
changing the parameters of w, & he limits applied on the solver parameters
were
w, 
+ 



International Research Journal of Commerce Arts and Science
http://www.casirj.com Page 144
CASIRJ Volume 10 Issue 4 [Year - 2019] ISSN 2319 – 9202

The Parameters achieved from the Solver are as below:


Parameters of GARCH (1,1) for Maximization of Likelyhood
Maximizatio Annual
STOCK Long Term
n of W A B Daily volatility Volatilit
S Volatility
likelyhood y
445.755244 0.000401462
L&T 2 0.0000272776 0.34718755 0.58487 97 0.0200365409 31.81%
436.234484 0.000152480
LUPIN 5 0.0000026358 0.00000082 0.98271 40 0.0123482954 19.60%
Table 1
6. Analysis:
The VaR model with components (GARCH-Normal Distribution-Normal Co-
relation (G-ND-NC)), has been back tested and the number of exceedances
and percentage of exceedances of days are as given in Table 2.
VaR Backtesting
PORTFOLIO L&T LUPIN
Confidence
Intervals No of % of Days No of % of Days No of % of Days
Exceedances Exceedances Exceedances Exceedances Exceedances Exceedances

95% 2 3.50877193 0 0 6 10.52631579


99% 0 0 0 0 1 1.754385965
99.50% 0 0 0 0 1 1.754385965
Table 2
Either in case of Bivariate data of portfolio or in case of univariate data (single assets
case), in low level of significance, the model underestimateas risk. It is evident from
the table 2, at 95% the percentage of days of exceedances are 3.50% in case of
portfolio and in case of Lupin it is 1.052%. In case of LUPIN , the model has very
unreliable risk estimation. Thus in cases of high level of significance, the
combination of models: GARCH-Normal Distribution-Normal Co-relation (G-ND-
NC) is reliable for estimation of risk i.e. at 99% and 99.5% no exceedances occurred
in portfolio. However, at individual levels it occurred in Lupin 1.75% of days. It sets
off against the profit of L&T when both stocks merged in a portfolio.
The losses in Lupin actually get setoff with L&Ts margin of profits when the
portfolio is formed. Thus portfolio formation pays off in terms of Risk management;
however it may minimize the return in short run.
This is evident from the Chart 1. The VaR of both the stocks are added directly and
plotted as ADDITIVE VAR (grey color) and the Portfolio VaR is plotted in blue

International Research Journal of Commerce Arts and Science


http://www.casirj.com Page 145
CASIRJ Volume 10 Issue 4 [Year - 2019] ISSN 2319 – 9202

color. Portfolio VaR line is significantly below the Additive VaR line at most of the
time period of 60 days. Whereas at chart 2 the individual stocks’ one day VaR line
have been plotted for reference.

One Day Portfolio VaR


135

130

125

V 120
A
ONE DAY VAR
R 115
ADDITIVE VAR
110

105

100
0 10 20 30 40 50 60
TIME

Chart 1

1 Day VaR: Individual Stocks


75.00

70.00

V 65.00
A
1 Day_VaR_L&T
R 60.00
1 Day_VaR_LUPIN

55.00

50.00
0 10 20 30 40 50 60
TIME

Chart 2

International Research Journal of Commerce Arts and Science


http://www.casirj.com Page 146
CASIRJ Volume 10 Issue 4 [Year - 2019] ISSN 2319 – 9202

7. Summary and Conclusion:


a.) In cases of high level of significance, the combination of models : GARCH-
Normal Distribution-Normal Co-relation (G-ND-NC) is reliable for
estimation of risk.
b.) A portfolio is beneficial over the single asset holdings, keeping in view the
risk management aspect.
c.) AS the time series progresses, the consideration of older periods forms a
reliable co-relation. It may further be tested on rigorous time series.
d.) Other combinations of components like EGARCH, EWAM, Complicated
copulas and t distributions are to be tested in Indian context.
e.) Further research on back testing of VaR with other assets like Forex,
Commodities, derivatives and multivariate time series are imperative in
Indian context.

8. Reference:
a. Jarrow, Robert A. & Chatterjea, Arkadev (2016), An Introduction to
Derivatives Securities, Financial Markets and Risk Management, 1st ed.,
Viva, India
b. Hull, J.C. (1993), Options, Futures, and other Derivative Securities, 2nd ed.,
Prentice Hall, NJ.
c. Skoglund, J., Erdman, Donald., Chen, Wei., November 1, Spring 2010, “The
performance of value at risk models during the crisis”, The Journal of Risk
Model Validation.
d. Bollerslev, T. (1986). “Generalized autoregressive conditional
hetroskedasticity”. Journal of Econometrics, 31, 307-327
e. Duffie, Darrell & Pan, Jun, January 21st, 1997, “An Overview of Value at
risk”
f. Hopper, P. Gregory, July/ Aug-1996, “Value at Risk: A new Methodology
for measuring portfolio Risk”, Business Review, Philadelphia FED,
g. Luenberger, David G.(2006), Investment Science, 1st ed., Oxford University
Press

International Research Journal of Commerce Arts and Science


http://www.casirj.com Page 147

You might also like