Econometría II

ARMA I

Santiago Albarracin (EMJMD)

UPB Pregrado

2024

1 / 10
Key requirements for large sample analysis
Violations
• Strict exogeneity E(a(X ; ) + 0

• Serial correlation Muche electric

Imposed assumptions +
+ In hi
00 lox(X; ) =0

• Weak dependence: Not independence, but limit dependence E(o , (X

; ) +0 Estacioneida

1014
• Stationarity: Joint distribution of (xt1 , ..., xtm ) identical to (xt1+h , ..., xtm+h )
– Future to be like the past in a probabilistic sense 1 tiapo

These requirements allow the implementation of:

-

• Law of Large Numbers (LLN):

neimme
!T zu

– plimYT = plim 1T t=1

Yt =µ
• Central Limit Theorem (CLT):
YT −µ
– √ → N(0, 1)
σ/ T

2 / 10
Stationary and weakly dependence
Covariance stationarity: t = 1, 2, ..., T
• E(xt ), constant and finite
• Var(xt ), constant and finite
• Cov(xt , xt+h ), depends only on the distance in time h, not the location in time t.
– Cov(XY) = E(XY) − E(X)E(Y)
• If independent Cov(XY) =0
– Var(X) = E ( X ) − E ( X) 2
2

Weakly dependent time series

• Asymptotic independence
• Corr(xt , xt+h ) → 0 as h → ∞

3 / 10
MA(1) process

yt = εt + εεt−1
E(ye) E(@y 04+ s)
• MA, Moving average
= + -

=
E OY +

• This model only has error terms = 0

– !t , i.i.d. N(0, σ 2 ) VAR(y1) = VAR (4) + Oct -)

O xAR(x + )
• Outcome is formed of only random shocks in t and t − 1 VAR(47)
=
+ -

= T + or

• Characteristics: =
(s + 027

– Mean: E(yt ) = 0
– Variance: Var(yt ) = (1 + θ 2 )σ 2 2x(y + ye + n)

– Covariance: Cov(yt , yt+h ) = θσ 2 if h = 1 · Cox(y +, y +) =

E((x( + 04 + 3)(x ++ + 0x +
)]
• Corr(yt , yt+1 ) = θ E((x ) (4+ 0xf) (0x - (024 +>4 +
))
=
+ ey + + + x ++ )) +

1+θ 2
– Covariance: Cov(yt , yt+h ) = 0 if h > 1
=

Ele + Welt) +
On t o
OE(x +2)
• Corr(yt , yt+h ) = 0 for h > 1
=

= Of

corr(ye Yats) (ye yes

2

of
10
Top
Lov
:or
, = , -
:

"At

·
Cov(y + , y+ + 2) =
E((x +
+ 0x 3)(x + + 2 + 0x + + )

=
El +
Gel to el +O
Elpits)
= O

Cor(ye +2)
,

=to

4 / 10
AR(1) process

yt = φyt−1 + εt
% = $ % +
2)
• AR, Autoregresive Y =
DY + 32

• $t , i.i.d. N(0, σ 2 ) "Y

!
• By repeated substitution: :
YoD +d
t t−1
• yt = φ y0 + φ ε1 + ... + φεt−1 + εt
• To ensure the variance of yt exists, we need to assume |φ| < 1 (implosive)
101 +
– AR(1) to be covariance stationary
S

• Characteristics:
VAR (yf) = VSR (PYes +4 )
E(y + ) E(by + - )
+

= + 4 +

(2 ye(y + 1)
EX) van(x )
=
+
) DE(y + - )
– Mean: E(yt )
E( +
-
+

=0 +
=

DE(y s) VAR(yH) PvAR(ye) &

σ2
E(yi) +
=

0
-

– Variance: Var(yt )
+ - =

= 1−φ2 E(y + ) = E(y y) +

Y8R(41) -

EVER(y -1)
+ =
↑2

– Correlation: Corr(yt , yt+h ) = φh where |φ| < 1 E(y + ) PE(y )

.v
-

+ = 0

– Correlation: Corr(yt , yt+h ) = 0 if h → ∞

Elyr)(SPO
Vaz(y)

Yor(yH)(5 -

b) =
Th

=
ver(H)

5 / 10
Impulse response function
IRF

menne
E

6 / 10
Stochastic difference equation
Example: Federal Reserve’s money supply
• mt = 1.03mt−1
– mt = 1.03t m0
• ∆mt = ρ[mt − mt−1] + εt
• mt = ρ(1.03)t m0 + (1 − ρ)mt−1 + εt
• Equidistant intervals
• Stochastic process εt
• Forecasting by updating periods
White noise process:
• Mean of zero, a constant variance, and is uncorrelated with all other realizations
• MA(q)
!q
• xt = i=0
βi ε t − i
Sist
+ =
ByEtyt ........
+ B, 2+
-
+ So2 +

Y+ =

03+ -

y
+.. . .
.

- ,
2++ Dox

P> ......... Oy
↓
Ye = by+ - +
27

7 / 10
ARMA (p,q) models

p q
" "
yt = a0 + ai yt−i + β i ε t− i
i=1 i=0

·
• q = 0 → AR(p)
• p = 0 → MA(q)
• All characteristic roots are within the unit circle < 1
• If greater than 1 ARIMA

8 / 10
AR(1) with constant

yt = a0 + a1 yt−1 + εt
it
• Solution:
t−1 t−1
" "
i
yt = a0 a1 + at1 y0 + i
a1 εt−1
i=0 i=0

• Expected value for t:

t−1
"
i
E ( yt ) = a0 a1 + at1 y0
i=0

• Means are time dependent, E(yt ) $= E(yt+h )

• If |a1 | < 1 and t → ∞ then ati y0 →0
a0
• Then a0 (1 + a1 + a2 + ...) → 1−a1

9 / 10
AR(1) with constant

yt = a0 + a1 yt−1 + εt
a0
• Mean: E(yt ) = E ( yt + h ) = 1−a1
=µ
2
σ
• Variance: E(yt − µ)2 = 1−(a1 )2
Conditions:
• The homogeneous solution must be zero
• The characteristic root a1 must be less than unity in absolute value.
All generalize to all ARMA(p,q) processes

10 / 10