Econometrie Avansata

Dr. Adrian Codirlasu, CFA

Dr. Bogdan Moinescu
I. Serii de timp
Definitie
O secventa de valori inregistrate de o
variabila aleatoare specifica intr-o anumita
perioada de timp
Caracteristici
Frecventa
Populatie vs. esantion
Momente
Stationaritate
Sezonalitate
Frecventa
Reprezinta periodicitatea cu care este observat
variabila.
Functie de specificul seriei de timp, frecventa
poate fi zilnica (cum este cazul preturilor
activelor financiare cursurile actiunilor, ratele
de dobanda, cursul de schimb), lunara (de
exemplu, rata inflatiei, salariul mediu pe
economie, rata somajului), trimestriala (cum este
produsul intern brut) sau anuala
Momentele seriei de timp
Media
Varianta
Coeficientul de asimetrie (skewness)
Kurtosis
Stationaritate
Conditiile ce trebuie indeplinite pentru ca o
serie de timp s fie stationara sunt:
media seriei de timp sa fie constanta sau cu
alte cuvinte, observatiile trebuie sa fluctueze
in jurul mediei.
varianta seriei s fie constanta.

Din punct de vedere economic, o serie

este stationara daca un soc asupra seriei
este temporar (se absoarbe in timp) si nu
permanent.
Stationaritate
In cazul in care seria nu este stationara,
prin diferentiere, se obtine o serie
stationara.
Ordinul de integrare al seriei reprezinta
numarul de diferentieri succesive
necesare pentru obtinerea unei serii
stationare (sau numarul de radacini
unitare al seriei).
Sezonalitate
Seriile de timp cu frecventa lunara sau
trimestriala prezinta adesea evoluaii care
au o anumita ciclicitate. De exemplu
activitatea economica se incetineste in
lunile de iarna, preturile cresc mai mult n
lunile reci decat in perioada de vara etc.
In analiza econometrica, pentru a elimina
aceste evolutii sezoniere seriile de timp
sunt ajustate sezonier.
II. Teste statistice
II.1. Distributii
Distributia de probabilitate
Este reprezentarea tuturor valorilor pe
care le poate lua o variabil aleatore si a
probabilitii de apariie a acestor valori

Variabile aleatoare
Discrete
Continue
Distributii
Normala
Lognormala
t
Chi patrat
F
Distributia normala
Distributia log-normala
Distributia t
Distributia Chi-patrat
Distributia F
II.2. Testarea ipotezelor
Testarea ipotezelor
Definirea ipotezei;
Identificarea testului statistic ce va fi utilizat i a
distribuiei de probabilitate a acestuia;
Specificarea nivelului de relevan al testului;
Specificarea regulii de decizie;
Colectarea datelor i estimarea parametrului;
Luarea deciziei statistice;
Luarea deciziei economice.
Definirea ipotezei
Specificarea ipotezei nule i a ipotezei
alternative
Ipoteza nul, notat cu , reprezint ipoteza ce
este testat, iar ipoteza alternativ, notat cu ,
este ipoteza acceptat n cazul n care ipoteza
nul este respins
Testarea mediei

Esantion mare Esantion mic

(n > 30) (n < 30)

Populatia are o distributie normala Testul t sau testul z Testul t

Populatia nu are o distributie normala Testul t sau testul z Nu se poate testa

Testarea variantei

cu n 1 grade de libertate

varianta esantionului de date utilizat

III. Analiza seriilor de
timp in EViews
Crearea unui fisier de lucru
Definirea seriilor
Introducerea datelor
Prelucrea seriilor
Stationaritatea seriilor de timp
Teste statistice
Augmented Dickey-Fuller (ADF)
Phillips-Perron
Testarea stationaritatii seriei
Interpretarea rezultatului
statistic
Null Hypothesis: L_EUR has a unit root
Exogenous: Constant, Linear Trend
Lag Length: 3 (Automatic based on SIC, MAXLAG=25)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic -0.981155 0.9448

Test critical values: 1% level -3.962327
5% level -3.411905
10% level -3.127850
Analiza distributiei seriei
Serie normal distribuita
Coeficientul de asimetrie (skewness) este zero
distributia normala este simetrica.
Kurtotica (kurtosis) este 3. Dac acest indicator
are o valoare mai mare dect 3, atunci distributia
se numete leptokurtotica, iar daca acesta este
mai mic dect 3 atunci distributia se numeste
platikurtotica.
Distributie EURRON
Theoretical Quantile-Quantile Kernel Density (Epanechnikov, h = 0.0018)

12 100

8 80
Normal Quantile

4
60

0
40
-4
20
-8

-12 0
-.08 -.04 .00 .04 .08 -.04 -.02 .00 .02 .04 .06

DL_EUR DL_EUR
Functia de autocorelatie
Functia de autocorelatie
Trendul seriilor de timp
Filtrul Hodrick-Prescott
Ajustarea sezoniera
a seriilor de timp
Investigarea sezonalitatii
Proceduri de desezonalizare
Census X12
Census X11
Tramo/Seats
Proceduri de desezonalizare
Serie desezonalizata
IV. Regresia liniara
multipla
Utilizare
Cu ajutorul regresiei liniare multiple, se poate
determina impactul pe care il au mai multe
variabile independente asupra unei anumite
variabile (numita variabila dependenta)
Ecuatia de regresie
Determinarea elasticitatilor
Daca variabila dependenta si variabilele
independente sunt specificate in logaritmi
naturali, atunci coeficientii variabilelor
independente pot fi interpretati ca
elasticitati
Astfel, acesti coeficienti vor arata cu cat la
suta se modifica variabila dependenta
daca variabila independenta se modifica
cu 1 la suta
Ipotezele regresiei liniare
Legtura dintre variabila dependent i
variabilele independente este liniar
Variabilele independente sunt aleatoare.
Intre variabilele independente incluse intr-
o regresie nu exista nici o relatie liniara.
Valoarea ateptat a termenului de eroare
este 0
Ipotezele regresiei liniare
Varianta termenului de eroare este
aceeasi pentru toate observaiile (erori
homoskedastice).
Termenul de eroare este necorelat intre
observatii.
Termenul de eroare este normal distribuit.
Impactul incalcarii ipotezelor

Heteroskedasticitate Erorile standard ale regresiei sunt incorecte

Corelaie serial a erorilor Erorile standard ale regresiei sunt incorecte

Multicoliniaritate Valori mari ale lui R-patrat si valori mici ale valorilor
t-statistic ale coeficientilor variabilelor independente
Teste statitistice
pentru regresia liniara

R-patrat
R-patrat ajustat (cu numarul de variabile
independente incluse in regresie)
Criterii informationale
Durbin-Watson
Teste statitistice
pentru regresia liniara
Teste pentru coeficientii obtinuti din
ecuatia de regresie
Testul t pentru testarea individuala a
coeficietilor
Testul F pentru testarea tuturor coeficientilor
Testul Wald
Teste statitistice
pentru regresia liniara
Teste pentru erorile ecuatiei de regresie
Corlograma erorilor
Corelograma erorilor patratice
Testarea distributiei erorilor (testul Jarque-
Berra)
Regresii cu variabile calitative
Variabile dummy
Acestea iau valoarea 1 dac o anumita conditie este
adevarata si valoarea 0 in caz contrar
Numarul de variabile dummy este cu 1 mai mic decat
numarul de conditii, in caz contrat existand
multicoliniaritate
Variabilele dummy pot fi utilizate si pentru captarea
impactului sezonier asupra variabilei independente,
introducand cel mult 11 variabile dummy pentru
datele cu frecven lunara sau cel mult 3 variabile
dummy pentru datele cu frecventa trimestriala, in
cazul in care datele nu au fost ajustate sezonier in
prealabil
IV.4. Regresii cu serii de
timp in Eviews
Estimarea functiei de reactie
Perioada analizata trim. I 1999 trim. I 2011
Serii de date utilizate:
r_eu rata de politic monetar a BCE;
infl_eu inflaiei, msurat prin indicele
armonizat al preurilor, in Uniunea Monetar;
gap_eu output-gap-ul, calculat pe baza unui
filtru Hodrick-Prescott pentru zona euro;
dummy variabil dummy pentru perioada de
criza financiara (ia valoarea 1 incepand cu trim. I
2009)
Parametrii regresiei
Ecuatia estimata
Indicatori ai regresiei
t-Statistic si probabilitatea asociata,
calculat pentru constanta si coeficientul
fiecarei variabile independente
R-Squared, Adjusted R-Squared
F-Statistic si probabilitatea asociata
Criteriile informationale (Akaike info
criterion, Schwarz criterion, Hannan-Quinn
criter.)
Durbin-Watson stat
Variabila dependenta evectiva
vs estimata
.05

.04

.03

.02

.01
.015
.00
.010
.005
.000
-.005
-.010
-.015
99 00 01 02 03 04 05 06 07 08 09 10 11

Residual Actual Fitted

Teste asupra termenilor de eroare
Corelograma erorilor
Corelograma erorilor patratice
Testarea tipului de distributie a erorilor
Serial Correlation LM Test
ARCH LM Test
White Heteroskedasticity Test
Selectare teste termeni de eroare
Corelograma erorilor
Corelograma erorilor patratice
 Testarea distributiei normale a
erorilor regresiei
12
Series: Residuals
Sample 1999Q1 2011Q1
10 Observations 49

8 Mean -6.09e-18
Median -0.000688
Maximum 0.013773
6 Minimum -0.010947
Std. Dev. 0.005747
Skewness 0.694016
4
Kurtosis 2.892419

2 Jarque-Bera 3.957170
Probability 0.138265
0
-0.010 -0.005 0.000 0.005 0.010 0.015
Testarea corelatiei seriale
Testarea termenilor ARCH
Teste de stabilitate

CUSUM Test
CUSUM of Squares Test
Recursive Coeficients
Teste de stabilitate
CUSUM Test
10

-5

-10

-15

-20

-25
II III IV I II III IV I
2009 2010 2011

CUSUM 5% Significance
CUSUM of Squares Test
1.6

1.2

0.8

0.4

0.0

-0.4
II III IV I II III IV I
2009 2010 2011

CUSUM of Squares 5% Significance

Recursive coeficients
.034 .4

.033 .2

.032
.0
.031
-.2
.030
-.4
.029

.028 -.6

.027 -.8
II III IV I II III IV I II III IV I II III IV I
2009 2010 2011 2009 2010 2011

Recursive C(1) Estimates Recursive C(2) Estimates

2 S.E. 2 S.E.

.8 .010

.005
.7

.000
.6
-.005
.5
-.010

.4
-.015

.3 -.020
II III IV I II III IV I II III IV I II III IV I
2009 2010 2011 2009 2010 2011

Recursive C(3) Estimates Recursive C(4) Estimates

2 S.E. 2 S.E.
V. Modele ARMA
Modele ARMA
Modele autoregresive (AR);
Modele cu medii mobile (MA);
Modele ARMA care combina cele dou
tipuri de procese.
Estimare modele ARMA
1. Testarea stationaritatii seriei
2. Stationarizarea seriei
3. Pe baza coeficienilor de autocorelaie
(funciei de autocorelaie) i a coeficienilor
de corelaie parial (funciei de autocorelaie
parial) se determin modelele
autoregresive de start pentru analiza seriei
de date.
Estimare modele ARMA
4. Se estimeaza parametri modelelor
ARMA.
5. Se testeaza caracteristicile modelelor
autoregresive ce au fost estimate n etapa
anterioara.
6. Se alege cel mai potrivit model folosind
diverse criterii de analiza.
7. Pe baza modelului selectat se fac
diverse analize si prognoze
V.4. Estimarea modelelor
ARMA in Eviews
Seria de date
250

200

150

100

50

0
97 98 99 00 01 02 03 04 05 06 07

BUBOR
Testul de stationaritate ADF
Null Hypothesis: BUBOR has a unit root
Exogenous: Constant, Linear Trend
Lag Length: 0 (Automatic based on SIC, MAXLAG=12)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic -5.024900 0.0003

Test critical values: 1% level -4.031899
5% level -3.445590
10% level -3.147710

*MacKinnon (1996) one-sided p-values.

Testul de stationaritate Philips-
Perron
Null Hypothesis: BUBOR has a unit root
Exogenous: Constant, Linear Trend
Bandwidth: 3 (Newey-West using Bartlett kernel)

Adj. t-Stat Prob.*

Phillips-Perron test statistic -5.437216 0.0001

Test critical values: 1% level -4.031899
5% level -3.445590
10% level -3.147710

*MacKinnon (1996) one-sided p-values.

Residual variance (no correction) 364.8682

HAC corrected variance (Bartlett kernel) 462.5140
Functia de autocorelatie
Specificare ecuatie
Estimare model MA(4)
Dependent Variable: BUBOR
Method: Least Squares
Sample (adjusted): 1997M01 2007M08
Included observations: 128 after adjustments
Convergence achieved after 15 iterations
Backcast: 1996M09 1996M12

Variable Coefficient Std. Error t-Statistic Prob.

C 41.54023 5.901126 7.039374 0.0000

MA(1) 0.734234 0.051566 14.23862 0.0000
MA(2) 0.495279 0.026993 18.34853 0.0000
MA(3) 0.863535 0.025672 33.63763 0.0000
MA(4) 0.804961 0.050194 16.03709 0.0000

R-squared 0.838087 Mean dependent var 43.94414

Adjusted R-squared 0.832822 S.D. dependent var 42.18206
S.E. of regression 17.24715 Akaike info criterion 8.571450
Sum squared resid 36588.11 Schwarz criterion 8.682858
Log likelihood -543.5728 F-statistic 159.1672
Durbin-Watson stat 1.582016 Prob(F-statistic) 0.000000

Inverted MA Roots .42-.90i .42+.90i -.78+.45i -.78-.45i

Analiza radacini ecuatie
Radacinile polinomului caracteristic
Inverse Roots of AR/MA Polynomial(s)
1.5
Inverse Roots of AR/MA Polynomial(s)
Specification: BUBOR C MA(1) MA(2) MA(3) MA(4)
1.0 Sample: 1997M01 2007M12
Included observations: 128

0.5
MA Root(s) Modulus Cycle

0.0
0.416806 0.900818i 0.992572 5.524002
-0.783923 0.450021i 0.903910 2.397739
-0.5
No root lies outside the unit circle.
-1.0 ARMA model is invertible.

-1.5
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5

MA roots
Corelograma erorilor
Valoarea efectiva vs estimata
250

200

150

100
80
50
40 0

-40

-80
97 98 99 00 01 02 03 04 05 06 07

Residual Actual Fitted

Estimare model AR(1)
Dependent Variable: BUBOR
Method: Least Squares
Sample (adjusted): 1997M02 2007M08
Included observations: 127 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

C 4.985180 2.639138 1.888943 0.0612

BUBOR(-1) 0.878633 0.043240 20.32007 0.0000

R-squared 0.767617 Mean dependent var 43.85480

Adjusted R-squared 0.765758 S.D. dependent var 42.33696
S.E. of regression 20.49047 Akaike info criterion 8.893420
Sum squared resid 52482.45 Schwarz criterion 8.938210
Log likelihood -562.7322 F-statistic 412.9052
Durbin-Watson stat 1.709848 Prob(F-statistic) 0.000000
Corelograma erorilor
Valoarea efectiva vs estimata
250
200
150
100

150 50

100 0

50
0
-50
-100
97 98 99 00 01 02 03 04 05 06 07

Residual Actual Fitted

Estimare model ARMA(1,10)
Dependent Variable: BUBOR
Method: Least Squares
Sample (adjusted): 1997M02 2007M08
Included observations: 127 after adjustments
Convergence achieved after 21 iterations
Backcast: 1996M04 1997M01

Variable Coefficient Std. Error t-Statistic Prob.

BUBOR(-1) 0.974210 0.017718 54.98466 0.0000

MA(5) -0.243541 0.056540 -4.307386 0.0000
MA(6) -0.226437 0.055472 -4.081987 0.0001
MA(7) -0.332302 0.055472 -5.990446 0.0000
MA(10) 0.476373 0.060775 7.838351 0.0000

R-squared 0.867500 Mean dependent var 43.85480

Adjusted R-squared 0.863156 S.D. dependent var 42.33696
S.E. of regression 15.66150 Akaike info criterion 8.378862
Sum squared resid 29924.46 Schwarz criterion 8.490837
Log likelihood -527.0577 Durbin-Watson stat 2.297258

Inverted MA Roots .88+.16i .88-.16i .56+.83i .56-.83i

-.02-.90i -.02+.90i -.53+.73i -.53-.73i
-.89+.33i -.89-.33i
 Conditia de stabilitate
Inverse Roots of AR/MA Polynomial(s)
Inverse Roots of AR/MA Polynomial(s)
1.5 Specification: BUBOR BUBOR(-1) MA(5) MA(6)
MA(7) MA(10)
Sample: 1997M01 2007M12
1.0
Included observations: 127

0.5
MA Root(s) Modulus Cycle
0.0
0.558806 0.826275i 0.997494 6.436651
-0.5 -0.890402 0.332511i 0.950463 2.256737
-0.528267 0.734115i 0.904428 2.863081
-1.0 -0.022438 0.897521i 0.897802 3.937348
0.882301 0.159193i 0.896548 35.19821

-1.5
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 No root lies outside the unit circle.
ARMA model is invertible.
MA roots
Corelograma erorilor
Valori efective vs estimate
250

200

150
120
100
80
50
40
0
0

-40

-80
97 98 99 00 01 02 03 04 05 06 07

Residual Actual Fitted

Selectarea specificatiei

MA(4) AR(1) ARMA(1,10)

Adjusted R-squared 0.832822 0.765758 0.863156

Akaike info criterion

8.571450 8.893420 8.378862
Schwarz criterion
8.682858 8.938210 8.490837
Prognoze
Dynamic forecast prognozeaza valoarea
in perioada t + 1 pe baza datelor efective
pana an momentul t, apoi pentru toate
perioadele urmatoare foloseste datele deja
prognozate incepand din momentul t + 1.
Static forecast prognozeaza o observatie
inainte numai pe baza datelor efective.
Realizarea de prognoze
Prognoza dinamica a seriei
80

60

40

20

-20

-40

-60
2007M09 2007M10 2007M11 2007M12

BUBORF
VI. Modele cu date panel
Modele cu date panel

Constau in estimarea de ecuatii de

regresie in care sunt folosite date care sunt
in acelasi timp atat serii de timp ct si date
crosssectionale
Utilizari
Rezumarea printr-un singur coeficient al impactului unei
variabile asupra unui grup de serii de timp variabile
dependente (grup de companii, de tari, etc.).
Estimarea de coeficienti specifici (constanta sau
coeficienti ai variabilelor independente) pentru fiecare
serie de timp considerata ca variabila dependenta
efecte fixe.
Gruparea variabilelor dependente in categorii si estimarea
impactului categoriei din care face parte variabila
dependenta asupra evolutiei acesteia
Definirea modelului
Definirea indentificatorilor
Definirea seriilor
Ecuatia de regresie
Rezultate regresie
Dependent Variable: DLOG(HICP?)
Method: Pooled EGLS (Cross-section SUR)
Included observations: 188 after adjustments
Cross-sections included: 4
Total pool (balanced) observations: 752
Linear estimation after one-step weighting matrix

Variable Coefficient Std. Error t-Statistic Prob.

C 0.000698 0.000445 1.566998 0.1175

DLOG(HICP_EU(-1)) 0.612692 0.104768 5.848058 0.0000
DLOG(HICP?(-1)) 0.553629 0.030088 18.40025 0.0000
DLOG(ER?(-1)) 0.083851 0.012757 6.572889 0.0000

Weighted Statistics

R-squared 0.447658 Mean dependent var 0.640863

Adjusted R-squared 0.445443 S.D. dependent var 1.207505
S.E. of regression 0.928331 Sum squared resid 644.6258
F-statistic 202.0780 Durbin-Watson stat 2.104933
Prob(F-statistic) 0.000000

Unweighted Statistics

R-squared 0.602410 Mean dependent var 0.007916

Sum squared resid 0.080251 Durbin-Watson stat 1.864487
Coeficienti individuali
Rezultate regresie
Dependent Variable: DLOG(HICP?)
Method: Pooled EGLS (Cross-section SUR)
Sample (adjusted): 1996M03 2011M10
Included observations: 188 after adjustments
Cross-sections included: 4
Total pool (balanced) observations: 752
Linear estimation after one-step weighting matrix

Variable Coefficient Std. Error t-Statistic Prob.

C 0.002113 0.000471 4.489158 0.0000

DLOG(HICP_EU(-1)) 0.631560 0.102724 6.148098 0.0000
DLOG(HICP?(-1)) 0.492847 0.031776 15.50997 0.0000
DLOG(ER?(-1)) 0.077977 0.012530 6.223189 0.0000
Fixed Effects (Cross)
_CZ--C -0.002185
_HU--C -0.000839
_PO--C -0.001681
_RO--C 0.004706

Effects Specification

Cross-section fixed (dummy variables)

Weighted Statistics

R-squared 0.466068 Mean dependent var 0.648656

Adjusted R-squared 0.461768 S.D. dependent var 1.221242
S.E. of regression 0.928200 Sum squared resid 641.8587
F-statistic 108.3850 Durbin-Watson stat 2.054515
Prob(F-statistic) 0.000000

Unweighted Statistics

R-squared 0.614957 Mean dependent var 0.007916

Sum squared resid 0.077719 Durbin-Watson stat 1.802623
Impactul anticipatiilor inflationiste
Rezultate regresie
Dependent Variable: DLOG(HICP?)
Method: Pooled EGLS (Cross-section SUR)
Sample (adjusted): 1996M03 2011M10
Included observations: 188 after adjustments
Cross-sections included: 4
Total pool (balanced) observations: 752
Linear estimation after one-step weighting matrix

Variable Coefficient Std. Error t-Statistic Prob.

C 0.001447 0.000443 3.264956 0.0011

DLOG(HICP_EU(-1)) 0.572149 0.102100 5.603786 0.0000
DLOG(ER?(-1)) 0.065592 0.012463 5.262752 0.0000
_CZ--DLOG(HICP_CZ(-1)) 0.193372 0.066868 2.891863 0.0039

_HU--DLOG(HICP_HU(-1)) 0.406401 0.052893 7.683466 0.0000

_PO--DLOG(HICP_PO(-1)) 0.326507 0.063476 5.143756 0.0000

_RO--DLOG(HICP_RO(-1)) 0.686025 0.035300 19.43402 0.0000

Weighted Statistics

R-squared 0.472755 Mean dependent var 0.644691

Adjusted R-squared 0.468508 S.D. dependent var 1.220327
S.E. of regression 0.924597 Sum squared resid 636.8850
F-statistic 111.3341 Durbin-Watson stat 2.058521
Prob(F-statistic) 0.000000

Unweighted Statistics

R-squared 0.623347 Mean dependent var 0.007916

Sum squared resid 0.076025 Durbin-Watson stat 2.116026
Impactul cursului de schimb
Rezultate regresie
Dependent Variable: DLOG(HICP?)
Method: Pooled EGLS (Cross-section SUR)
Sample (adjusted): 1996M03 2011M10
Included observations: 188 after adjustments
Cross-sections included: 4
Total pool (balanced) observations: 752
Linear estimation after one-step weighting matrix

Variable Coefficient Std. Error t-Statistic Prob.

C 0.001175 0.000383 3.064805 0.0023

DLOG(HICP_EU(-1)) 0.513687 0.090404 5.682137 0.0000
DLOG(HICP?(-1)) 0.405206 0.028875 14.03333 0.0000
_CZ--DLOG(ER_CZ(-1)) 0.053506 0.026337 2.031585 0.0426

_HU--DLOG(ER_HU(-1)) 0.006168 0.017179 0.359047 0.7197

_PO--DLOG(ER_PO(-1)) 0.020524 0.011008 1.864565 0.0626

_RO--DLOG(ER_RO(-1)) 0.417841 0.030630 13.64161 0.0000

Weighted Statistics

R-squared 0.506046 Mean dependent var 0.689472

Adjusted R-squared 0.502068 S.D. dependent var 1.354914
S.E. of regression 0.995840 Sum squared resid 738.8147
F-statistic 127.2065 Durbin-Watson stat 2.027024
Prob(F-statistic) 0.000000

Unweighted Statistics

R-squared 0.708343 Mean dependent var 0.007916

Sum squared resid 0.058869 Durbin-Watson stat 1.943093
VII. Modele GARCH
Tipuri de volatilitate
Istorica calculata pe baza preturilor
istorice ale activelor
Exponentialy weighted moving average -
EWMA
Estimata prin modele econometrice
(Generalised Autoregressive Conditional
Heteroskedasticity - GARCH)
Implicita calculata din preturile optiunilor
Calcul volatilitate istorica
1. Observatii curs spot S0, S1, . . . , Sn la
intervale de ani
2. Calcul randament in timp continuu:
Si
ui = ln
S i 1

3. Calculul deviatiei standard, s ,

pentru randamentele ui
s
4. Estimarea volatilitatii istorice ca: =

Modelul EWMA
Conform acestui model, volatilitatea din
ziua n este o medie ponderata intre
volatilitatea din ziua anterioara si
randamentul la patrat u2 din ziua
anterioara
=
2
n
2
n 1 + (1 )u 2
n 1
RiskMetrics (JP Morgan, Reuters)
foloseste = 0.94 pentru calculul
volatilitatii zilnice
Modele GARCH
In modelul GARCH, volatilitatea depinde
de volatilitatile anterioare si de
randamentele patratice anterioare ale
activului
Coeficientii variabielor sunt extimati prin
diverse proceduri econometrice
p q
= + ku
2
n
2
nk + i 2
n i
k =1 i =1
Tipuri de modele GARCH

ARCH
GARCH
GARCH in Mean
Treshold ARCH - TARCH
Exponential GARCH - EGARCH
Integrated GARCH - IGARCH
Specificare model
Conditii model GARCH
Coeficientii ecuatiei variantei sa fie pozitivi;

Suma coeficientilor ecuatiei variantei s fie mai

mica decat 1. In caz contrar, modelul este
GARCH integrat (I-GARCH), iar volatilitatea este
exploziva.
Estimare GARCH(1,1)
Dependent Variable: DL_EUR
Method: ML - ARCH (Marquardt) - Normal distribution
Sample (adjusted): 2 2148
Included observations: 2147 after adjustments
Convergence achieved after 19 iterations
Variance backcast: ON
GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*GARCH(-1)

Coefficient Std. Error z-Statistic Prob.

C 0.000198 8.77E-05 2.260467 0.0238

Variance Equation

C 2.50E-07 5.08E-08 4.926018 0.0000

RESID(-1)^2 0.138819 0.009128 15.20872 0.0000
GARCH(-1) 0.868095 0.008286 104.7609 0.0000

R-squared -0.001355 Mean dependent var 0.000427

Adjusted R-squared -0.002757 S.D. dependent var 0.006208
S.E. of regression 0.006216 Akaike info criterion -7.718359
Sum squared resid 0.082811 Schwarz criterion -7.707792
Log likelihood 8289.659 Durbin-Watson stat 1.848831
Estimare EGARCH(2,1,1)
Dependent Variable: DL_EUR
Method: ML - ARCH (Marquardt) - Generalized error distribution (GED)
Sample (adjusted): 2 2148
Included observations: 2147 after adjustments
Convergence achieved after 25 iterations
Variance backcast: ON
LOG(GARCH) = C(3) + C(4)*ABS(RESID(-1)/@SQRT(GARCH(-1))) +
C(5)*ABS(RESID(-2)/@SQRT(GARCH(-2))) + C(6)*RESID(-1)
/@SQRT(GARCH(-1)) + C(7)*LOG(GARCH(-1))

Coefficient Std. Error z-Statistic Prob.

@SQRT(GARCH) 0.112635 0.039554 2.847620 0.0044

C -0.000457 0.000156 -2.932430 0.0034

Variance Equation

C(3) -0.284286 0.056336 -5.046247 0.0000

C(4) 0.399824 0.049811 8.026844 0.0000
C(5) -0.170567 0.049394 -3.453202 0.0006
C(6) -0.026267 0.013602 -1.931131 0.0535
C(7) 0.989348 0.004282 231.0737 0.0000

GED PARAMETER 1.293394 0.047198 27.40373 0.0000

R-squared 0.002000 Mean dependent var 0.000427

Adjusted R-squared -0.001266 S.D. dependent var 0.006208
S.E. of regression 0.006212 Akaike info criterion -7.788111
Sum squared resid 0.082534 Schwarz criterion -7.766977
Log likelihood 8368.537 F-statistic 0.612409
Durbin-Watson stat 1.855328 Prob(F-statistic) 0.746122
Corelograma erorilor patratice
Volatilitatea conditionata
.028

.024

.020

.016

.012

.008

.004

.000
250 500 750 1000 1250 1500 1750 2000

Conditional standard deviation

VIII. Modele cu vectori
autoregresivi (VAR)
Definitie si utilizare

Un model VAR (Vector Autoregression) permite

tratarea simetrica a tuturor variabilelor din model,
in sensul ca nu presupune implicit exogeneitatea
unei anumite variabile (cum se intampla in cazul
OLS).
Construire model VAR
Alegere numar de lag-uri
Alegere numar lag-uri
Stabilitate model VAR
Stationaritate model VAR
Stationaritate model VAR
Autocorrelations with 2 Std.Err. Bounds
Cor(DLOG(HICP_RO),DLOG(HICP_RO)(-i)) Cor(DLOG(HICP_RO),DLOG(ER_RO)(-i)) Cor(DLOG(HICP_RO),DLOG(HICP_EU)(-i))
.6 .6 .6

.4 .4 .4

.2 .2 .2

.0 .0 .0

-.2 -.2 -.2

-.4 -.4 -.4

-.6 -.6 -.6

1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12

Cor(DLOG(ER_RO),DLOG(HICP_RO)(-i)) Cor(DLOG(ER_RO),DLOG(ER_RO)(-i)) Cor(DLOG(ER_RO),DLOG(HICP_EU)(-i))

.6 .6 .6

.4 .4 .4

.2 .2 .2

.0 .0 .0

-.2 -.2 -.2

-.4 -.4 -.4

-.6 -.6 -.6

1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12

Cor(DLOG(HICP_EU),DLOG(HICP_RO)(-i)) Cor(DLOG(HICP_EU),DLOG(ER_RO)(-i)) Cor(DLOG(HICP_EU),DLOG(HICP_EU)(-i))

.6 .6 .6

.4 .4 .4

.2 .2 .2

.0 .0 .0

-.2 -.2 -.2

-.4 -.4 -.4

-.6 -.6 -.6

1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Testarea autocorelatiei
VAR Residual Portmanteau Tests for Autocorrelations
Null Hypothesis: no residual autocorrelations up to lag h
Sample: 1996M01 2011M12
Included observations: 186

Lags Q-Stat Prob. Adj Q-Stat Prob. df

1 0.497749 NA* 0.500439 NA* NA*

2 1.487910 NA* 1.501363 NA* NA*
3 4.163052 NA* 4.220360 NA* NA*
4 22.25030 0.1353 22.70513 0.1218 16
5 35.55546 0.0786 36.37783 0.0661 25
6 52.90164 0.0204 54.30223 0.0150 34
7 64.09821 0.0201 65.93665 0.0138 43
8 81.99966 0.0050 84.64266 0.0028 52
9 89.82131 0.0096 92.86202 0.0053 61
10 95.09797 0.0247 98.43849 0.0141 70
11 99.91440 0.0561 103.5577 0.0334 79
12 154.0077 0.0000 161.3815 0.0000 88

*The test is valid only for lags larger than the VAR lag order.
df is degrees of freedom for (approximate) chi-square distribution
Testarea autocorelatiei
VAR Residual Serial Correlation LM Tests

Null Hypothesis: no serial correlation at lag order h

Sample: 1996M01 2011M12
Included observations: 186

Lags LM-Stat Prob

1 9.676269 0.3773
2 6.781290 0.6599
3 17.34069 0.0436
4 21.62957 0.0101
5 14.02825 0.1213
6 18.97320 0.0254
7 12.64418 0.1794
8 20.40055 0.0156
9 8.294159 0.5048
10 5.512769 0.7875
11 5.240046 0.8129
12 72.10294 0.0000

Probs from chi-square with 9 df.

Testarea distributiei normale
VAR Residual Normality Tests
Orthogonalization: Cholesky (Lutkepohl)
Null Hypothesis: residuals are multivariate normal
Sample: 1996M01 2011M12
Included observations: 186

Component Skewness Chi-sq df Prob.

1 1.088627 36.73835 1 0.0000

2 0.763380 18.06523 1 0.0000
3 -0.043528 0.058734 1 0.8085

Joint 54.86231 3 0.0000

Component Kurtosis Chi-sq df Prob.

1 5.840031 62.50978 1 0.0000

2 12.63516 719.4818 1 0.0000
3 4.804536 25.23672 1 0.0000

Joint 807.2283 3 0.0000

Component Jarque-Bera df Prob.

1 99.24813 2 0.0000
2 737.5470 2 0.0000
3 25.29545 2 0.0000

Joint 862.0906 6 0.0000

Testarea heteroskedasticitatii
VAR Residual Heteroskedasticity Tests: No Cross Terms (only levels and squares)
Sample: 1996M01 2011M12
Included observations: 186

Joint test:

Chi-sq df Prob.

420.2497 108 0.0000

Individual components:

Dependent R-squared F(18,167) Prob. Chi-sq(18) Prob.

res1*res1 0.359985 5.218402 0.0000 66.95714 0.0000

res2*res2 0.458122 7.843734 0.0000 85.21061 0.0000
res3*res3 0.410657 6.464790 0.0000 76.38214 0.0000
res2*res1 0.414220 6.560566 0.0000 77.04500 0.0000
res3*res1 0.137523 1.479350 0.1029 25.57924 0.1098
res3*res2 0.715462 23.32867 0.0000 133.0759 0.0000
Definitia impulsului
Definitia impulsurilor conteaza deoarece:
in cazul descompunerii Cholesky conteaza
ordonarea variabilelor daca acestea sunt
corelate intre ele;
in cazul descompunerii generalizate nu
conteaza ordonarea variabilelor;
in cazul descompunerii structurale, aceasta
poate fi utilizata doar daca a fost specificat
anterior un model structural cu restrictiile
necesare.
Functii de impuls-raspuns
Functii de impuls-raspuns
Response to Cholesky One S.D. Innovations 2 S.E. Response to Cholesky One S.D. Innovations 2 S.E.
Response of DLOG(HICP_RO) to DLOG(HICP_RO) Response of DLOG(ER_RO) to DLOG(HICP_RO)
.020 .04

.015 .03

.010 .02

.005 .01

.000 .00

-.005 -.01

-.010 -.02
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12

Response of DLOG(HICP_RO) to DLOG(ER_RO) Response of DLOG(ER_RO) to DLOG(ER_RO)

.020 .04

.015 .03

.010 .02

.005 .01

.000 .00

-.005 -.01

-.010 -.02
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12

Response of DLOG(HICP_RO) to DLOG(HICP_EU) Response of DLOG(ER_RO) to DLOG(HICP_EU)

.020 .04

.015 .03

.010 .02

.005 .01

.000 .00

-.005 -.01

-.010
-.02
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Descompunerea variantei
Variance Decomposition Variance Decomposition
Percent DLOG(HICP_RO) variance due to DLOG(HICP_RO) Percent DLOG(ER_RO) variance due to DLOG(HICP_RO)
100 100

80 80

60 60

40 40

20 20

0 0
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12

Percent DLOG(HICP_RO) variance due to DLOG(ER_RO) Percent DLOG(ER_RO) variance due to DLOG(ER_RO)
100 100

80 80

60 60

40 40

20 20

0 0
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12

Percent DLOG(HICP_RO) variance due to DLOG(HICP_EU) Percent DLOG(ER_RO) variance due to DLOG(HICP_EU)
100 100

80 80

60 60

40 40

20 20

0 0
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12