THE ARCH and GARCH

MODEL
• US stock market in the wake of escalating oil prices in the first half of
2008.

• On June 6, 2008, Dow Jones Index dropped by almost 400 points,

when oil price jumped to $139 a barrel

• Toward the end of October, oil price dropped to $67 a barrel.

• September 29, 2008, Dow Jones Index fell by 777.7 points because of
the financial crisis

• October 3, 2008, stock market again fell by almost 800 points before
recovering and closing down by 369 points
• Such gyrations in oil prices have led to wide swings in stock prices

• Such Swings in oil prices and credit crises have serious effects on real
economy and financial markets

• It is important to measure asset return volatility

 0
0.1
0.2
0.3
0.4
0.5
0.6

-0.3
-0.2
-0.1
0
500
1000
lnex

1500
2000
2500
0.01
0.02
0.03

-0.03
-0.02
-0.01
0

1
62
123
184
245
306
367
428
489
550
611
672
733
794
855
916
977
1038
1099
1160
dlnex

1221
1282
1343
1404
1465
1526
1587
1648
1709
1770
1831
1892
1953
2014
2075
2136
2197
2258
2319
 0
0.0001
0.0002
0.0003
0.0004
0.0005
0.0006
0.0007
0.0008

2
56
110
164
218
272
326
380
434
488
542
596
650
704
758
812
866
920
974
1028
1082
1136
ressqr

1190
1244
1298
1352
1406
1460
1514
1568
1622
1676
1730
1784
1838
1892
1946
2000
2054
2108
2162
2216
2270
2324
 Clusters of periods when volatility
• Simple way of looking at volatility: is high and clusters of period when
volatility is low-indication of autocorrelation
D log(ex) = c + u t
ressqr
0.0008

0.0007

0.0006

0.0005

0.0004

0.0003

0.0002

0.0001

87

1022
1107
1192
1277
1362
1447
1532
1617
1702
1787
1872
1957
2042
2127
2212
2297
2

172
257
342
427
512
597
682
767
852
937
Predict residuals Wide swings in squared residuals-indication of
Obtain residuals squared underlying volatility in the foreign exchange returns
Plot residuals squared against time
 How to measure volatility?
VARIANCE
• The variance of a random variable is a measure of the variability in the
values of the random variable. In our case it is the variance of daily
exchange rate returns. The value turns out to be 0.0000351
• Above measure is a single number for a given sample. Doesn’t take into
account the variance of past history of returns. Does not capture volatility
clustering

• Volatility clustering: Time varying volatility in asset returns-Periods of

turbulence in which prices show wide swings and periods of tranquility in
which there is relative calm.

• This results in correlation in error variance over time.

• The measure that takes into account the volatility of past history of returns
is the Autoregressive Conditional Heteroscedasticity or ARCH
 H 0 : l1 = l 2 = ...... = l p 2  2  2  2
u t = l 0 + l1u t -1 + l 2 u t - 2 + ........ + l p u t - p + e t

Compute (n-p)R2~χp2 r is the number of coefficients estimated

 THE ARCH MODEL
s t 2 = l 0 + l1u t -1 2 + l 2 u t - 2 2 + ..... + l8 u t -8 2

First three coefficients are not

significant but the last five are
There is an ARCH effect in the
dollar/euro exchange rate
That means, the error variances are
auto correlated
For example, L4 being significant
means that volatility 4 period back
is affecting current period.
That means, you may expect to
observe similar degree of volatility
in every 4 time period. Same
interpretation goes for L5, L6, L7
and L8
THE GARCH MODEL
• The GARCH model is just an extension of ARCH model to improve
upon some problems with ARCH model
• The main problem with ARCH model is that the ARCH model requires
estimation of p autoregressive terms. Determining the number of lags
and estimating large number of parameters is a problem.
• GARCH model reduces the p autoregressive terms in ARCH model into
one conditional variance. Essentially the conditional variance in the
GARCH model captures the influence of p autoregressive terms in the
ARCH model. Therefore it indicates the combined influence of p
autoregressive terms.
ARCH MODEL

GARCH MODEL

s t 2 = l 0 + l1u t -1 2 + l 2s t -1 2

Sigmat-1^2 captures the combined influence of ut-2^2, ut-3^2,………..

Residual Square
THE GARCH MODEL
s t 2 = l 0 + l1u t -1 2 + l 2s t -1 2

U2t-1 is the residuals square from the

previous period- ARCH effect. Lambda1
gives the magnitude of the arch effect.
Effect of previous period volatility on
current volatility.
Sigma2t-1 is the conditional variance at
time t-1, conditioned on future variances-
GARCH effect. Lambda 2 gives the
magnitude of the Garch effect. Cumulative
impact of historical shocks. It captures the
persistence of the volatility.
Lambda0 is interpreted as long term
volatility.

A large sum of these coefficients will imply

that a large positive or large negative
return/shock will lead future forecasts of
the variance to be high for a protracted
period.