Professional Documents
Culture Documents
Volatility Modelling-Revised
Volatility Modelling-Revised
0 0.05
0
This suggests residuals are highly
-4 4-Mar-08 volatile
4-May-08 towards
4-Jul-08 the end, but less4-Nov-08
4-Sep-08
-0.05
-8 volatile in the middle and
-0.1
Stationary
beginning of Series
the sample.
-12
-0.15
I II III IV
2008
Daily Exchange Rate
Exchage Rate (Daily)
60
Non-stationary Series
50
1.2
Residual series of
40
ARIMA(1,0,0) SARIMA(0,1,1)
Exchange Rate Changes (Daily) 5
30 0.8 3.0%
20 2.0%
0.4
10
1.0%
0.0
0
0.0%
3-Mar-08 3-May-08 3-Jul-08 3-Sep-08 3-Nov-08
4-Mar-08 4-May-08 4-Jul-08 4-Sep-08 4-Nov-08
-0.4
-1.0%
-0.8 -2.0%
This suggests residuals are volatile
-1.2
-3.0%
in the middle and HighofVolatile
end the Period
-4.0%
-1.6
I
sample.II But volatility towards
III
the IV
end is higher than2008the middle.
D(EX,0,5) Residuals
Monthly Peak Gold Price
2000
Peak-month
1800
Peak-month
Non-stationary Series
1600
1400
1200
1000 250
800
200 Residual series of ARIMA(1,0,1)
600
150 0.2 SARIMA(1,1,1)12
400
100 0.15 Seems…there
Monthly is
Gold Price Changes
200
0 50
0.1 variability, but
October-…
46 October-…
Septemb…
February…
21 Septemb…
26 February…
Decemb…
36 Decemb…
Novemb…
11 Novemb…
April-04
16 April-09
July-05
31 July-10
June-03
6 June-08
51 March-12
March-07
May-06
1January-08
61January-13
January-03
August-07
41 May-11
56 August-12
0
0.05 variability is constant
0
over the sample?
91
66
71
76
81
86
96
101
116
106
111
121
-50
-0.05 This suggests variability in
-100 -0.1 residuals is constant over the
-150 -0.15
sample period unlike ARIMA
-200 -0.2
2005
models
2006
of ‘SENSEX’,
2007 2008 2009
OIL2010
PRICE’ 2011
and2012 2013
-0.25
EXCHANGE RATE’ series.
D(PEAK,0,12) Residuals
What do we observe?
• Residuals of MSARIMA models of SENSEX returns, Oil
price changes and Exchange Rate changes are volatile
series.
• Volatility changes over time or over the sample. At
some points in time, it is highly volatile, but at some
points it is less volatile and at some points it is not
volatile at all.
• So, it seems residual volatility/variability has a
pattern. For example, high volatile period follows high
volatility and remains so for sometime. Similar is the
case with low volatile period.
• Residual series of ARIMA model of Gold Price changes
seems to be constant over time.
Some Obvious Questions…
• Graphically it seems variability in residual series
shows a pattern, but it is indicative.
• How can we confirm it? Is there any statistical test
which confirms that the variability in residual series is
not constant?
• If the residual series of ARIMA models exhibit some
pattern, is the ARIMA model forecast accurate?
• If not, what should be done to correct it?
What ARIMA Model does…
GARCH family of models estimate and predict
• ARIMA/SARIMA
both mean and estimates
risk returnconditional
for mean.
• those
This meanstime series dataand
it estimates where volatility
predicts changeswhat
on average
overbe
would time
theand follows
sensex a pattern.
return or oil price return or
exchange rate return in future assuming that the
How isinit residual
variation modelled? Like AR models,
or prediction error GARCH
is constant.
• So,family models
it predicts meanvolatility
returnas a function
without takingof into
its past
havingthe
account different
risk or weights.
variabilityBut
or how should
volatility thatwemean
decide
return -how
might farininfuture.
face the past the model should
• delvealong
Hence, into ?with
Just the
a few periods in
prediction ofthe
mean past or long
return,
periods?of variability or risk in return is essential to
prediction
In other
forecast thewords,
futurerisk return
return should be modelled
accurately.
as a short memory or long memory process?
Problem with ARIMA Modeling
• SENSEX returns and Oil price changes are stationary
but volatile series.
• Univariate ARIMA modelling to forecast the daily
‘SENSEX returns’ OR ‘Oil Price changes’ using OLS
estimation will predict the ‘conditional mean’ of
these series based on past information with the
assumption that the residual variance is constant over
time.
• But these series exhibit high variability suggesting
that conclusion/forecast based on OLS is likely to be
inaccurate, hence a correction is required to reduce
this variability or volatility or risk or heteroscedastic
error variance.
Need for Volatility Modeling
• The basic version of least squares model assumes
that the expected value of all error terms, when
squared, is same at any given point. The variances of
error term is constant. Known as “HOMOSCEDASTIC”
assumption.
• So,
rt = E (rt / rt-1, rt-2, … rt-s) + ut
= E (rt / rt-1, rt-2, … rt-s) + SE (rt / rt-1, rt-2, …, rt-s) * Ԑt ;
= mt + ht1/2 Ԑt ;
where
wt ut2 ht t2 ht ht ( t2 1)ht ; t ~ N (0,1)
q
ut j q
ut j p
log( ht ) 0 j j i log( ht i )
j 1 h t j j 1 h t j i 1
i 1
i
j 1
j 1
p q
i 1
i
j 1
j 1
I-GARCH (p, q) Models
• I-GARCH processes are either non-stationary or if they
are stationary they have infinite variance. Infinite
variance means heavy tails! A distribution can be heavy-
tailed with a finite variance.
Select
ARCH test
& lag 10
2,000
2,000,000 700 702 704 706 708
MAPE has improved marginally.,
710 712 714 716 718 720
PRICEF ± 2 S.E.
1,000,000 by 10 basis point.
0
700 705 710 715 720
Forecast of Variance
CASE
Modeling and forecasting of day-ahead
Exchange Rate
ARCH effect is present at higher lags.
Estimate ARCH model
Higher order lags are significant.
Positive and less than 1
ARCH (2) .
Volatility is still present
52
51
Forecast of Forecast: EXF
Actual: EX
50
ARIMA (1,0,0) SARIMA(0,1,1) 5
ARCH
Forecast sample: (2) 11/28/...
11/03/2008 Model
Included observations: 19
49
Root Mean Squared Error 0.594261
48 Mean Absolute Error 0.461180
Mean Abs. Percent Error 0.945881
47 Theil Inequality Coefficient 0.006056
Bias Proportion 0.042747
46
Variance Proportion 0.000080
45 Covariance Proportion 0.957173
11/03
11/13
11/04
11/05
11/06
11/07
11/10
11/11
11/12
11/14
11/17
11/18
11/19
11/20
11/21
11/24
11/25
11/26
11/28
Suggests …ARCH(2) is not
52
EXF ± 2 S.E. appropriate,Forecast:
other
Actual: EX
EXF volatility
2.0
51
models mayForecast
be appropriate.
sample: 11/03/2008 11/28/...
Included observations: 19
50
Root Mean Squared Error 0.563939
1.6 Mean Absolute Error 0.422610
49 Mean Abs. Percent Error 0.867142
Theil Inequality Coefficient 0.005752
1.2 Bias Proportion 0.002819
48
Variance Proportion 0.005971
Covariance Proportion 0.991210
0.8 47
0.4 46
Forecast of
3 5 7 1 3 7 9 1 5 7
1 1/0
1 1/0
11/0
11 ARIMA (1,0,0) SARIMA(0,1,1)5 Model
/1
11/1
11/1
11/1
11/2
11/2
11/2
0.0
EXF ± 2 S.E.
11/03
11/13
11/04
11/05
11/06
11/07
11/10
11/11
11/12
11/14
11/17
11/18
11/19
11/20
11/21
11/24
11/25
11/26
11/28
Forecast of Variance
CASE
Modelling and forecasting of day-ahead
Oil Price
ARCH Test
Published in
Applied Energy (Elsevier Publication): Vol -88 : 2011
Author : Dr. Sajal Ghosh, Faculty at MDI, Gurgaon
OLS regression of
Changes in exchanges rate
ARCH as effect
a is present at higher
function of changes in oil price
lags….6,20,21, 30. Estimate ARCH
model of higher lags. ARCH (6).
Test ARCH-LM test after running
ARCH(6) model
GARCH Model
CASES
CASE
Modeling and forecasting of day-ahead
electricity price in Indian Energy Exchange
IEX Energy Price CASE GARCH(1,1)
Not an GARCH(1,1)
appropriate model
CASE
Modeling and forecasting of day-ahead
Exchange Rate
Exchange Rate Case
Volatility is still present
Variance of error has become homoscedastic
CASE
Modelling and forecasting of day-ahead
Oil Price
Volatility is present
Error variance has become homoscedastic
CASE
Examining crude oil price-exchange rate nexus for
India during the period of extreme oil price volatility
Published in
Applied Energy (Elsevier Publication): Vol -88 : 2011
Author : Dr. Sajal Ghosh, Faculty at MDI, Gurgaon
An 1% increase in oil price return, will result 0.05% or 5 BP decline
in exchange rate of 5 BP depreciation
ARCH effect is present, so requires volatility modelling
GARCH (1,1) Model
GARCH-M (1,1) Model. Exchange
Rate Return is not a function of its
volatility or its own risk
Select Variance
TGARCH (1,1) Model. There is no
asymmetric effect in exchange rate return
when there is a shock in the market or the
effect of positive or negative shock has
similar effect in magnitude in exchange rate
volatility in India