ECONOMETRICS
ECONOMETRICS
Sponsored by a Grant TÁMOP-4.1.2-08/2/A/KMR-2009-0041 Course Material Developed by Department of Economics,
Faculty of Social Sciences, Eötvös Loránd University Budapest (ELTE) Department of Economics, Eötvös Loránd University Budapest
Institute of Economics, Hungarian Academy of Sciences Balassi Kiadó, Budapest
ECONOMETRICS
Authors: Péter Elek, Anikó Bíró Supervised by: Péter Elek
June 2010
ELTE Faculty of Social Sciences, Department of Economics
ECONOMETRICS
Week 9.
Univariate time series analysis I.
Péter Elek, Anikó Bíró
Plan
Time series models
Autocovariance, autocorrelation Random walk, white noise
Stationary and nonstationary time series
Trend stationarity and difference stationarity
To read: M 13.1, 13.3, first part of 13.4
Time series models
Time series: X
tis a sequence of random
variables (i.e. a stochastic process)
Continuous
e.g. EKG data
Discrete: annual, monthly, or even in every minute
e.g. inflation, GDP, stock prices
Seasonality, time trend etc.
Time series regression
60000 80000 100000 120000 140000 160000 180000 200000
00 01 02 03 04 05 06 07
közszféra bérei (Ft/hó)
Properties of time series
Joint distribution: distribution of (X(t1), X(t2),…,X(tn)) Expectation: t = E(Xt)
Variance: t2 = Var(Xt)
Autocovariance: t1,t2 = cov(Xt1, Xt2)
t,t = t2, t2,t1 = t1,t2
Autocorrelation: t1,t2 = corr(Xt1, Xt2) = t1,t2/(t1t2)
t,t = 1
Partial autocovariance (t1,t2): covariance between Xt1 and Xt2 after controlling for the the observations
between t1 and t2
Stationary time series
“constant temporal relationship”
Strict stationarity: (X
t1, X
t2,…,X
tn) ~
(X
k+t1, X
k+t2,…,X
k+tn) for every (t
1,t
2,…,t
n) and k Weak stationarity:
= t = E(Xt) expectation does not depend on t
k = t,t-k = cov(Xt, Xt-k) autocovariance does not depend on t
k = corr(Xt, Xt-k) = k/0 autocorrelation
-k = k, 0 = 2
-k = k, 0 = 1
Strict stationarity → weak stationarity
(if moments exist), the reverse is also true
for Gaussian processes
White noise
(purely random process)
Xt = t ,
t is i.i.d.
k = 0 for every k ≠ 0
sometimes this is assumed, not equivalent for non-Gaussian processes
If Yt is a random walk then
∆Yt = Xt – Xt–1 is a white noise.
-3 -2 -1 0 1 2 3
25 50 75 100
Nonstationary time series
Random walk is nonstationary (even for = 0)
Two common causes of nonstationarity
Trend
Seasonaliy
Random walk
X
0= 0
X
t= X
t-1+ +
t,
t~ IN(0,
2)
(independent normal)
X
t= t +
1+
2+…+
t
t= E(X
t) = t
t2= Var(X
t) = t
2-24 -20 -16 -12 -8 -4 0
50 100 150 200 250 300
Trend stationary vs. difference stationary time series
Deterministic trend (trend stationarity)
X
t= t + Y
t, where Y
tis stationary, E(Y
t) = 0
The effect of shocks is temporary, the process eventually returns to its long run trend
∆Xt = + Yt – Yt–1 is stationary, E(∆Xt) = These are trend + I(0) processes.
Stochastic trend (difference stationarity)
X
t= X
t–1+ +Y
t, where Y
tis stationary, E(Y
t) = 0
The effect of shocks is persistent, the process does not have a deterministic trend
∆Xt = + Yt is stationary, E(∆Xt) =
These are first order integrated (I(1)) processes.
Deterministic vs. stochastic trend
-50 0 50 100 150 200 250
25 50 75 100 125 150 175 200
X(t) = t+IN(0,100) X(t) = 1+X(t-1)+IN(0,100)
Motivation: why is stationarity important?
Spurious trend in time series
Spurious trend in time series II.
Spurious regression in time series
Two independent random walks
Xt = Xt-1 + 1t Yt = Yt–1 + 2t
Regression: Yt = c + βXt + ut β = 0 because of
independence, but the t-test is significant!
The t-statistic does not have a limit distribution!
Reason: ut is nonstationary