Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the...

43
Univariate Time Series Fall 2008 Environmental Econometrics (GR03) TSI Fall 2008 1 / 16

Transcript of Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the...

Page 1: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Univariate Time Series

Fall 2008

Environmental Econometrics (GR03) TSI Fall 2008 1 / 16

Page 2: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series I

A time series Yt is a process (or data) observed in sequence overtime, t = 1, ...,T .

macroeconomic series: e.g., in�ation rate, GDP, unemployment,....temperature over time in global warmingmonthly or annual returns of �nancial assets

The key issue in time series is temporal dependence.In order to have a meaningful regression analysis, we need somerestrictions on temporal dependence.

Environmental Econometrics (GR03) TSI Fall 2008 2 / 16

Page 3: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series I

A time series Yt is a process (or data) observed in sequence overtime, t = 1, ...,T .

macroeconomic series: e.g., in�ation rate, GDP, unemployment,....

temperature over time in global warmingmonthly or annual returns of �nancial assets

The key issue in time series is temporal dependence.In order to have a meaningful regression analysis, we need somerestrictions on temporal dependence.

Environmental Econometrics (GR03) TSI Fall 2008 2 / 16

Page 4: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series I

A time series Yt is a process (or data) observed in sequence overtime, t = 1, ...,T .

macroeconomic series: e.g., in�ation rate, GDP, unemployment,....temperature over time in global warming

monthly or annual returns of �nancial assets

The key issue in time series is temporal dependence.In order to have a meaningful regression analysis, we need somerestrictions on temporal dependence.

Environmental Econometrics (GR03) TSI Fall 2008 2 / 16

Page 5: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series I

A time series Yt is a process (or data) observed in sequence overtime, t = 1, ...,T .

macroeconomic series: e.g., in�ation rate, GDP, unemployment,....temperature over time in global warmingmonthly or annual returns of �nancial assets

The key issue in time series is temporal dependence.In order to have a meaningful regression analysis, we need somerestrictions on temporal dependence.

Environmental Econometrics (GR03) TSI Fall 2008 2 / 16

Page 6: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series I

A time series Yt is a process (or data) observed in sequence overtime, t = 1, ...,T .

macroeconomic series: e.g., in�ation rate, GDP, unemployment,....temperature over time in global warmingmonthly or annual returns of �nancial assets

The key issue in time series is temporal dependence.

In order to have a meaningful regression analysis, we need somerestrictions on temporal dependence.

Environmental Econometrics (GR03) TSI Fall 2008 2 / 16

Page 7: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series I

A time series Yt is a process (or data) observed in sequence overtime, t = 1, ...,T .

macroeconomic series: e.g., in�ation rate, GDP, unemployment,....temperature over time in global warmingmonthly or annual returns of �nancial assets

The key issue in time series is temporal dependence.In order to have a meaningful regression analysis, we need somerestrictions on temporal dependence.

Environmental Econometrics (GR03) TSI Fall 2008 2 / 16

Page 8: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Example: Disposal income and consumption in USA

8000

1000

012

000

1400

016

000

1800

0pe

r cap

ita re

al d

isp.

 inc.

/per

 cap

ita re

al c

ons.

1960 1970 1980 1990 20001959­1995

per capita real disp. inc. per capita real cons.

Environmental Econometrics (GR03) TSI Fall 2008 3 / 16

Page 9: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Example: Average temperature in Central London

78

910

11te

mp

1700 1800 1900 2000year

Environmental Econometrics (GR03) TSI Fall 2008 4 / 16

Page 10: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Example: Return on three-month T-bills in USA

05

1015

3 m

o. T

­bill

 rate

1960 1970 1980 1990 20001959­1995

Environmental Econometrics (GR03) TSI Fall 2008 5 / 16

Page 11: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series II

The two main economic problems from time series are dynamiccausal e¤ects and economic forecasting.

(Dynamic causal e¤ect) What is the e¤ect of X on Y over time?

the short-run/long-run e¤ect of a change in an interest rate (by centralbank) on in�ationthe e¤ect of a decrease of carbon dioxide (by regulation) on globalwarming in 1/10/100 years later

(Economic forecasting) What are predicted future values of Y , givenavailable information?

average temperature in 2050?the forecasting is not necessarily based on the causality relationship.

We can separate time series into two categories: univariate (Yt isscalar) and multivariate (Yt is vector-valued).

Environmental Econometrics (GR03) TSI Fall 2008 6 / 16

Page 12: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series II

The two main economic problems from time series are dynamiccausal e¤ects and economic forecasting.(Dynamic causal e¤ect) What is the e¤ect of X on Y over time?

the short-run/long-run e¤ect of a change in an interest rate (by centralbank) on in�ationthe e¤ect of a decrease of carbon dioxide (by regulation) on globalwarming in 1/10/100 years later

(Economic forecasting) What are predicted future values of Y , givenavailable information?

average temperature in 2050?the forecasting is not necessarily based on the causality relationship.

We can separate time series into two categories: univariate (Yt isscalar) and multivariate (Yt is vector-valued).

Environmental Econometrics (GR03) TSI Fall 2008 6 / 16

Page 13: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series II

The two main economic problems from time series are dynamiccausal e¤ects and economic forecasting.(Dynamic causal e¤ect) What is the e¤ect of X on Y over time?

the short-run/long-run e¤ect of a change in an interest rate (by centralbank) on in�ation

the e¤ect of a decrease of carbon dioxide (by regulation) on globalwarming in 1/10/100 years later

(Economic forecasting) What are predicted future values of Y , givenavailable information?

average temperature in 2050?the forecasting is not necessarily based on the causality relationship.

We can separate time series into two categories: univariate (Yt isscalar) and multivariate (Yt is vector-valued).

Environmental Econometrics (GR03) TSI Fall 2008 6 / 16

Page 14: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series II

The two main economic problems from time series are dynamiccausal e¤ects and economic forecasting.(Dynamic causal e¤ect) What is the e¤ect of X on Y over time?

the short-run/long-run e¤ect of a change in an interest rate (by centralbank) on in�ationthe e¤ect of a decrease of carbon dioxide (by regulation) on globalwarming in 1/10/100 years later

(Economic forecasting) What are predicted future values of Y , givenavailable information?

average temperature in 2050?the forecasting is not necessarily based on the causality relationship.

We can separate time series into two categories: univariate (Yt isscalar) and multivariate (Yt is vector-valued).

Environmental Econometrics (GR03) TSI Fall 2008 6 / 16

Page 15: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series II

The two main economic problems from time series are dynamiccausal e¤ects and economic forecasting.(Dynamic causal e¤ect) What is the e¤ect of X on Y over time?

the short-run/long-run e¤ect of a change in an interest rate (by centralbank) on in�ationthe e¤ect of a decrease of carbon dioxide (by regulation) on globalwarming in 1/10/100 years later

(Economic forecasting) What are predicted future values of Y , givenavailable information?

average temperature in 2050?the forecasting is not necessarily based on the causality relationship.

We can separate time series into two categories: univariate (Yt isscalar) and multivariate (Yt is vector-valued).

Environmental Econometrics (GR03) TSI Fall 2008 6 / 16

Page 16: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series II

The two main economic problems from time series are dynamiccausal e¤ects and economic forecasting.(Dynamic causal e¤ect) What is the e¤ect of X on Y over time?

the short-run/long-run e¤ect of a change in an interest rate (by centralbank) on in�ationthe e¤ect of a decrease of carbon dioxide (by regulation) on globalwarming in 1/10/100 years later

(Economic forecasting) What are predicted future values of Y , givenavailable information?

average temperature in 2050?

the forecasting is not necessarily based on the causality relationship.

We can separate time series into two categories: univariate (Yt isscalar) and multivariate (Yt is vector-valued).

Environmental Econometrics (GR03) TSI Fall 2008 6 / 16

Page 17: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series II

The two main economic problems from time series are dynamiccausal e¤ects and economic forecasting.(Dynamic causal e¤ect) What is the e¤ect of X on Y over time?

the short-run/long-run e¤ect of a change in an interest rate (by centralbank) on in�ationthe e¤ect of a decrease of carbon dioxide (by regulation) on globalwarming in 1/10/100 years later

(Economic forecasting) What are predicted future values of Y , givenavailable information?

average temperature in 2050?the forecasting is not necessarily based on the causality relationship.

We can separate time series into two categories: univariate (Yt isscalar) and multivariate (Yt is vector-valued).

Environmental Econometrics (GR03) TSI Fall 2008 6 / 16

Page 18: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Time Series II

The two main economic problems from time series are dynamiccausal e¤ects and economic forecasting.(Dynamic causal e¤ect) What is the e¤ect of X on Y over time?

the short-run/long-run e¤ect of a change in an interest rate (by centralbank) on in�ationthe e¤ect of a decrease of carbon dioxide (by regulation) on globalwarming in 1/10/100 years later

(Economic forecasting) What are predicted future values of Y , givenavailable information?

average temperature in 2050?the forecasting is not necessarily based on the causality relationship.

We can separate time series into two categories: univariate (Yt isscalar) and multivariate (Yt is vector-valued).

Environmental Econometrics (GR03) TSI Fall 2008 6 / 16

Page 19: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Stationarity: future will be like past

A time series fYtg is covariance stationary if its mean and(co-)variances are constant across time periods:

E (Yt ) = µ, Var (Yt ) = σ2 for all t

Cov (Yt ,Yt+k ) = γ (k) for all t and k

γ (k) is called the autocovariance function and ρ (k) = γ (k) /γ (0)is the autocorrelation function.

fYtg is said to be strictly stationary if the joint distribution of(Yt , ...,Yt+k ) is independent of all t and k.

Thus, a stationary time series is one whose probability distributionsare stable over time.

Environmental Econometrics (GR03) TSI Fall 2008 7 / 16

Page 20: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Stationarity: future will be like past

A time series fYtg is covariance stationary if its mean and(co-)variances are constant across time periods:

E (Yt ) = µ, Var (Yt ) = σ2 for all t

Cov (Yt ,Yt+k ) = γ (k) for all t and k

γ (k) is called the autocovariance function and ρ (k) = γ (k) /γ (0)is the autocorrelation function.

fYtg is said to be strictly stationary if the joint distribution of(Yt , ...,Yt+k ) is independent of all t and k.

Thus, a stationary time series is one whose probability distributionsare stable over time.

Environmental Econometrics (GR03) TSI Fall 2008 7 / 16

Page 21: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Stationarity: future will be like past

A time series fYtg is covariance stationary if its mean and(co-)variances are constant across time periods:

E (Yt ) = µ, Var (Yt ) = σ2 for all t

Cov (Yt ,Yt+k ) = γ (k) for all t and k

γ (k) is called the autocovariance function and ρ (k) = γ (k) /γ (0)is the autocorrelation function.

fYtg is said to be strictly stationary if the joint distribution of(Yt , ...,Yt+k ) is independent of all t and k.

Thus, a stationary time series is one whose probability distributionsare stable over time.

Environmental Econometrics (GR03) TSI Fall 2008 7 / 16

Page 22: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Nonstationarity

A time series process that is not stationary is called a nonstationaryprocess.

changing mean:Yt = β0 + β1t + ut

changing variance:Yt = Yt�1 + ut

This process is called the random walk.changing mean and variance:

Yt = β0 + Yt�1 + ut

This process is called the random walk with drift.

It is easier to spot certain nonstationary series rather than stationaryones.

Environmental Econometrics (GR03) TSI Fall 2008 8 / 16

Page 23: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Nonstationarity

A time series process that is not stationary is called a nonstationaryprocess.

changing mean:Yt = β0 + β1t + ut

changing variance:Yt = Yt�1 + ut

This process is called the random walk.changing mean and variance:

Yt = β0 + Yt�1 + ut

This process is called the random walk with drift.

It is easier to spot certain nonstationary series rather than stationaryones.

Environmental Econometrics (GR03) TSI Fall 2008 8 / 16

Page 24: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Nonstationarity

A time series process that is not stationary is called a nonstationaryprocess.

changing mean:Yt = β0 + β1t + ut

changing variance:Yt = Yt�1 + ut

This process is called the random walk.

changing mean and variance:

Yt = β0 + Yt�1 + ut

This process is called the random walk with drift.

It is easier to spot certain nonstationary series rather than stationaryones.

Environmental Econometrics (GR03) TSI Fall 2008 8 / 16

Page 25: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Nonstationarity

A time series process that is not stationary is called a nonstationaryprocess.

changing mean:Yt = β0 + β1t + ut

changing variance:Yt = Yt�1 + ut

This process is called the random walk.changing mean and variance:

Yt = β0 + Yt�1 + ut

This process is called the random walk with drift.

It is easier to spot certain nonstationary series rather than stationaryones.

Environmental Econometrics (GR03) TSI Fall 2008 8 / 16

Page 26: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Nonstationarity

A time series process that is not stationary is called a nonstationaryprocess.

changing mean:Yt = β0 + β1t + ut

changing variance:Yt = Yt�1 + ut

This process is called the random walk.changing mean and variance:

Yt = β0 + Yt�1 + ut

This process is called the random walk with drift.

It is easier to spot certain nonstationary series rather than stationaryones.

Environmental Econometrics (GR03) TSI Fall 2008 8 / 16

Page 27: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Example 1: Changing mean

Yt = β0 + β1t + ut , where β0 = 0, β1 = 1 and ut � iid N (0, 1)

020

4060

8010

0y/

mea

n_y

0 20 40 60 80 100time

y mean_y

Environmental Econometrics (GR03) TSI Fall 2008 9 / 16

Page 28: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Example 2: Random walk

Yt = Yt�1 + ut , where Y0 = 0 and ut � iid N (0, 1)

­50

510

y

0 20 40 60 80 100time

Environmental Econometrics (GR03) TSI Fall 2008 10 / 16

Page 29: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Example 3: Random walk with drift

Yt = β0 + Yt�1 + ut , where β0 = 1, Y0 = 0 and ut � iid N (0, 1)

050

100

y/m

ean_

y

0 20 40 60 80 100time

y mean_y

Environmental Econometrics (GR03) TSI Fall 2008 11 / 16

Page 30: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Ergodicity

In most of stationary time series analysis, we will still use the OLSestimation method.

In order to have properties of consistency and asymptotic normality ofestimators, we need another property of stationary time series.

A stationary time series is (loosely) said to be ergodic if γ (k)! 0as k ! ∞. Sometime a similar property is called weakly dependent.(Ergodic Theorem) If Yt is strictly stationary and ergodic andE jYt j < ∞, then as T ! ∞,

1T

T

∑t=1Yt !p E (Yt )

Using the Ergodic theorem, we can establish the consistency andasymptotic normality of OLS estimatiors in the regression withstationary and ergodic time series.

Environmental Econometrics (GR03) TSI Fall 2008 12 / 16

Page 31: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Ergodicity

In most of stationary time series analysis, we will still use the OLSestimation method.

In order to have properties of consistency and asymptotic normality ofestimators, we need another property of stationary time series.

A stationary time series is (loosely) said to be ergodic if γ (k)! 0as k ! ∞. Sometime a similar property is called weakly dependent.(Ergodic Theorem) If Yt is strictly stationary and ergodic andE jYt j < ∞, then as T ! ∞,

1T

T

∑t=1Yt !p E (Yt )

Using the Ergodic theorem, we can establish the consistency andasymptotic normality of OLS estimatiors in the regression withstationary and ergodic time series.

Environmental Econometrics (GR03) TSI Fall 2008 12 / 16

Page 32: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Ergodicity

In most of stationary time series analysis, we will still use the OLSestimation method.

In order to have properties of consistency and asymptotic normality ofestimators, we need another property of stationary time series.

A stationary time series is (loosely) said to be ergodic if γ (k)! 0as k ! ∞. Sometime a similar property is called weakly dependent.

(Ergodic Theorem) If Yt is strictly stationary and ergodic andE jYt j < ∞, then as T ! ∞,

1T

T

∑t=1Yt !p E (Yt )

Using the Ergodic theorem, we can establish the consistency andasymptotic normality of OLS estimatiors in the regression withstationary and ergodic time series.

Environmental Econometrics (GR03) TSI Fall 2008 12 / 16

Page 33: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Ergodicity

In most of stationary time series analysis, we will still use the OLSestimation method.

In order to have properties of consistency and asymptotic normality ofestimators, we need another property of stationary time series.

A stationary time series is (loosely) said to be ergodic if γ (k)! 0as k ! ∞. Sometime a similar property is called weakly dependent.(Ergodic Theorem) If Yt is strictly stationary and ergodic andE jYt j < ∞, then as T ! ∞,

1T

T

∑t=1Yt !p E (Yt )

Using the Ergodic theorem, we can establish the consistency andasymptotic normality of OLS estimatiors in the regression withstationary and ergodic time series.

Environmental Econometrics (GR03) TSI Fall 2008 12 / 16

Page 34: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Ergodicity

In most of stationary time series analysis, we will still use the OLSestimation method.

In order to have properties of consistency and asymptotic normality ofestimators, we need another property of stationary time series.

A stationary time series is (loosely) said to be ergodic if γ (k)! 0as k ! ∞. Sometime a similar property is called weakly dependent.(Ergodic Theorem) If Yt is strictly stationary and ergodic andE jYt j < ∞, then as T ! ∞,

1T

T

∑t=1Yt !p E (Yt )

Using the Ergodic theorem, we can establish the consistency andasymptotic normality of OLS estimatiors in the regression withstationary and ergodic time series.

Environmental Econometrics (GR03) TSI Fall 2008 12 / 16

Page 35: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Autoregressions

We �rst focus on the case with univariate time series, fYtgTt=1. LetIt�1 = fYt�1,Yt�2, ...g denote the past history of the series.

The primary example in the univariate time series is a model ofautoregression specifying that only a �nite number of past lagsmatter:

E (Yt jIt�1) = E (Yt jYt�1, ...,Yt�k ) .An autoregressive process of order k, called AR (k), is

Yt = µ+ ρ1Yt�1 + ...+ ρkYt�k + ut ,

whereE (ut jIt�1) = 0.

Environmental Econometrics (GR03) TSI Fall 2008 13 / 16

Page 36: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Autoregressions

We �rst focus on the case with univariate time series, fYtgTt=1. LetIt�1 = fYt�1,Yt�2, ...g denote the past history of the series.The primary example in the univariate time series is a model ofautoregression specifying that only a �nite number of past lagsmatter:

E (Yt jIt�1) = E (Yt jYt�1, ...,Yt�k ) .

An autoregressive process of order k, called AR (k), is

Yt = µ+ ρ1Yt�1 + ...+ ρkYt�k + ut ,

whereE (ut jIt�1) = 0.

Environmental Econometrics (GR03) TSI Fall 2008 13 / 16

Page 37: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Autoregressions

We �rst focus on the case with univariate time series, fYtgTt=1. LetIt�1 = fYt�1,Yt�2, ...g denote the past history of the series.The primary example in the univariate time series is a model ofautoregression specifying that only a �nite number of past lagsmatter:

E (Yt jIt�1) = E (Yt jYt�1, ...,Yt�k ) .An autoregressive process of order k, called AR (k), is

Yt = µ+ ρ1Yt�1 + ...+ ρkYt�k + ut ,

whereE (ut jIt�1) = 0.

Environmental Econometrics (GR03) TSI Fall 2008 13 / 16

Page 38: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

AR(1) Model

AR(1) Model:Yt = µ+ ρYt�1 + ut

if jρj < 1, thenE (Yt ) =

µ

1� ρ

Var (Yt ) =σ2

1� ρ2

Cov (Yt ,Yt�k ) = σ2ρk

1� ρ2

If jρj < 1, the process fYtg is stationary and ergodic.(Estimation) In order to estimate µ and ρ, we just need to run theOLS regression of Yt�1 on Yt .

Environmental Econometrics (GR03) TSI Fall 2008 14 / 16

Page 39: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Example: Consumption growth in USA

We model the consumption growth series in USA as AR(1):

gct = µ+ ρgct�1 + ut

Coe¤. Std. Err.

gc_1 0.45 0.156constant 0.01 0.003

­.01

0.0

1.0

2.0

3.0

4lc

 ­ lc

[_n­

1]/L

inea

r pre

dict

ion

­.01 0 .01 .02 .03 .04gc[_n­1]

lc ­ lc[_n­1] Linear predictionEnvironmental Econometrics (GR03) TSI Fall 2008 15 / 16

Page 40: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Moving Average

Another example is a model of moving average process of order q,fYtg, that is a weighted sum of lagged i.i.d. shocks:

Yt = µ+ ut + λ1ut�1 + ...+ λqut�q

The simplest case is a moving average process of order 1, calledMA (1):

Yt = µ+ ut + λut�1

E (Yt ) = µ, Var (Yt ) =�1+ λ2

�σ2

Cov (Yt ,Yt�1) = γ(1) = λσ2

Cov (Yt ,Yt�k ) = γ(k) = 0 for k � 2

Thus, an MA(1) process is stationary and ergodic.

Environmental Econometrics (GR03) TSI Fall 2008 16 / 16

Page 41: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Moving Average

Another example is a model of moving average process of order q,fYtg, that is a weighted sum of lagged i.i.d. shocks:

Yt = µ+ ut + λ1ut�1 + ...+ λqut�q

The simplest case is a moving average process of order 1, calledMA (1):

Yt = µ+ ut + λut�1

E (Yt ) = µ, Var (Yt ) =�1+ λ2

�σ2

Cov (Yt ,Yt�1) = γ(1) = λσ2

Cov (Yt ,Yt�k ) = γ(k) = 0 for k � 2

Thus, an MA(1) process is stationary and ergodic.

Environmental Econometrics (GR03) TSI Fall 2008 16 / 16

Page 42: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Moving Average

Another example is a model of moving average process of order q,fYtg, that is a weighted sum of lagged i.i.d. shocks:

Yt = µ+ ut + λ1ut�1 + ...+ λqut�q

The simplest case is a moving average process of order 1, calledMA (1):

Yt = µ+ ut + λut�1

E (Yt ) = µ, Var (Yt ) =�1+ λ2

�σ2

Cov (Yt ,Yt�1) = γ(1) = λσ2

Cov (Yt ,Yt�k ) = γ(k) = 0 for k � 2

Thus, an MA(1) process is stationary and ergodic.

Environmental Econometrics (GR03) TSI Fall 2008 16 / 16

Page 43: Univariate Time Series - UCLuctpsc0/Teaching/GR03/TS1.pdf · (Dynamic causal e⁄ect) What is the e⁄ect of X on Y over time? the short-run/long-run e⁄ect of a change in an interest

Moving Average

Another example is a model of moving average process of order q,fYtg, that is a weighted sum of lagged i.i.d. shocks:

Yt = µ+ ut + λ1ut�1 + ...+ λqut�q

The simplest case is a moving average process of order 1, calledMA (1):

Yt = µ+ ut + λut�1

E (Yt ) = µ, Var (Yt ) =�1+ λ2

�σ2

Cov (Yt ,Yt�1) = γ(1) = λσ2

Cov (Yt ,Yt�k ) = γ(k) = 0 for k � 2

Thus, an MA(1) process is stationary and ergodic.

Environmental Econometrics (GR03) TSI Fall 2008 16 / 16