Formulas and Tables for Time Series Econometrics / Econometrics II July, 2022 © Prof. Dr. Kai Carstensen Institute for Statistics and Econometrics, CAU Kiel Formulas and Tables for Time Series Econometrics / Econometrics II Part A. Stationary Time Series Processes 1 1.1 Stationary Time Series Regression Properties of Stochastic Time-Series Processes Autocovariance of order k: Cov(yt , yt−k ) = E{[yt − E(yt )][yt−k − E(yt−k )]} Autocorrelation of order k: Corr(yt , yt−k ) = E{[yt − E(yt )][yt−k − E(yt−k )]} p p . Var(yt ) Var(yt−k ) Cross covariance of order k: Cov(xt , yt−k ) = E{[xt − E(xt )][yt−k − E(yt−k )]}. Cross correlation of order k: Corr(xt , yt−k ) = E{[xt − E(xt )][yt−k − E(yt−k )]} p p . Var(xt ) Var(yt−k ) Weak stationarity: A time series process {yt } is called weakly stationary if the first and second moments are time-invariant and finite, i.e., E(yt ) = µ < ∞ ∀t Var(yt ) = σ 2 < ∞ ∀t Cov(yt , yt−k ) = γk < ∞ ∀t Strong stationarity: A time series process {yt } is called strongly stationary if the joint probability distribution of any set of k observations in the sequence [yt , yt+1 , . . . , yt+k−1 ] is the same regardless of the origin, t, in the time scale. Ergodicity: A strongly stationary time-series process, {yt }, is ergodic if for any two bounded functions that map vectors in the a and b dimensional real vector spaces to real scalars, f : Ra → R1 and g : Rb → R1 , lim |E[f (yt , . . . , yt+a−1 )g(yt+k , . . . , yt+k+b−1 )]| k→∞ = |E[f (yt , . . . , yt+a−1 )]| |E[g(yt+k , . . . , yt+k+b−1 )]| 2 Formulas and Tables for Time Series Econometrics / Econometrics II Ergodic Theorems for scalar processes: If {yt } is a time-series process that is strongly stationary and ergodic and E[yt ] = µ exists and is a finite constant, then ȳT = T −1 T X a.s. yt −→ µ. t=1 If in addition E[yt2 ] = m exists and is finite then T −1 T X a.s. yt2 −→ m. t=1 Ergodic Theorems for vector processes: If {yt } is a K × 1 vector of strongly stationary and ergodic processes with existing and finite moments E[yt ] = µ and E[yt yt0 ] = M, then T −1 T X a.s. yt −→ µ t=1 and T −1 T X a.s. yt yt0 −→ M. t=1 Martingale sequence: A vector sequence zt is a martingale sequence if E[zt |zt−1 , zt−2 , . . .] = zt−1 . Martingale difference sequence: A vector sequence zt is a martingale difference sequence if E[zt |zt−1 , zt−2 , . . .] = 0. Stationary linear process: A stationary linear process with mean zero is defined as ut = ψ0 εt + ψ1 εt−1 + ψ2 εt−2 + · · · = ψ(L)εt , where εt is iid white noise with mean zero and variance σ 2 > 0, if (C1) ∞ X j|ψj | < ∞ j=0 (C2) ψ(1) = ψ0 + ψ1 + ψ2 + · · · = 6 0 Then ut has mean E(ut ) = 0, variance Var(ut ) = σ 2 (ψ02 + ψ12 + ψ22 + · · · ) and long-run variance √ λ2 ≡ lim Var( T ūT ) = σ 2 [ψ(1)]2 > 0. T →∞ 3 Formulas and Tables for Time Series Econometrics / Econometrics II 1.2 Asymptotic Properties of Time Series Regressions Population model and OLS assuptions: The population model with K regressors is y t = xt β + u t , t = 1, . . . , T, where {[x, u]} is a jointly stationary and ergodic process with finite first and second moments and the usual OLS assumptions E(x0t ut ) = 0 and rank[E(x0t xt )] = K hold. Consistency: The OLS estimator of the population model is consistent, p β̂ −→ β. Martingale Difference Central Limit Theorem: If zt is a vector valued stationary and ergodic martingale difference sequence, with E[zt z0t ] = Σ, where Σ is a finite positive definite matrix, then √ T z̄T = T −1/2 T X d zt −→ Normal(0, Σ). t=1 Asymptotic distribution of the OLS estimator when ut is white noise: By the martingale difference CLT, T −1/2 T X d x0t ut −→ Normal(0, B), t=1 where B ≡ E(u2t x0t xt ). From this follows √ d T (β̂ − β) = −→ Normal(0, V), where V = A−1 BA−1 and A ≡ E(x0t xt ). Estimation of the variance of the OLS estimator when ut is white noise: Using the OLS residuals ût , consistent estimators are  = T −1 T X x0t xt t=1 and B̂ = T −1 T X û2t x0t xt t=1 such that V̂ = Â−1 B̂Â−1 . 4 Formulas and Tables for Time Series Econometrics / Econometrics II Gordin’s Central Limit Theorem: If {zt } is a stationary and ergodic stochastic process of dimension K × 1 that satisfies the following conditions: (1) Asymptotic uncorrelatedness: E[zt |zt−k , zt−k−1 , . . .] converges in mean square to zero as k → ∞. (2) Summability of autocovariances: the asymptotic variance Γ∗ is finite, where √ lim Var( T z̄T ) = T →∞ ∞ X ∞ X Cov(zt , zs ) = t=0 s=0 ∞ X Cov(zt , zt−k ) = Γ∗ . k=−∞ (3) Asymptotic negligibility of innovations: information eventually becomes negligible as it fades far back in time from the current observation. Then √ T z̄T = T −1/2 T X d zt −→ Normal(0, Γ∗ ). t=1 Asymptotic distribution of the OLS estimator when ut is temporally dependent: If x0t ut satisfies the conditions of Gordin’s CLT, T −1/2 T X d x0t ut −→ Normal(0, B), t=1 where B ≡ √ P∞ k=−∞ Cov(x0t ut , x0t−k ut−k ). From this follows d T (β̂ − β) = −→ Normal(0, V), where V = A−1 BA−1 and A ≡ E(x0t xt ). Estimation of the variance of the OLS estimator when ut is temporally dependent: A consistent estimate is V̂ = Â−1 B̂Â−1 , where  = T −1 T X p x0t xt −→ A. t=1 and q X 0 B̂ = Γ̂0 + (Γ̂k + Γ̂k ), k=1 with Γ̂k = T −1 T X ût ût−k x0t xt−k . t=k+1 5 Formulas and Tables for Time Series Econometrics / Econometrics II Newey-West estimator: A positive semidefinite estimator of B is q X 0 k (Γ̂k + Γ̂k ). B̂ = Γ̂0 + 1− q+1 k=1 2 2.1 Autoregressive Models Properties of the Autoregressive Model of Order 1 Autoregressive model of order 1 without constant: iid et ∼ (0, σ 2 ) ut = ρut−1 + et , Moving average representation: 2 ut = et + ρet−1 + ρ et−2 + · · · = ∞ X ρi et−i i=0 Conditional moments: E(ut |ut−1 ) = ρut−1 Var(ut |ut−1 ) = σ 2 Unconditional moments: The mean is E(ut ) = 0 and, if |ρ| < 1, the second moments are 1 Var(ut ) = σ 2 1 − ρ2 ρk Cov(ut , ut−k ) = σ 2 1 − ρ2 Corr(ut , ut−k ) = ρk . 2.2 Estimation of the Autoregressive Model of Order 1 Consistency of OLS when the disturbance is white noise: The OLS estimator of the AR(1) model without constant, !−1 T T X X ρ̂ = u2t−1 ut−1 ut , t=2 t=2 is consistent if et is white noise and thus E(ut−1 et ) = 0. 6 Formulas and Tables for Time Series Econometrics / Econometrics II Asymptotic normality of OLS when the disturbance is white noise: By the martingale difference CLT, √ d T (ρ̂ − ρ) −→ Normal(0, V). If the disturbance et is homoscedastic, the asymptotic covariance can be estimated as !−1 !−1 T T X X 2 2 [ yt−1 V̂ = σ̂e2 T −1 ⇒ Avar(ρ̂) = σ̂e2 yt−1 . t=2 t=2 Inconsistency of OLS when the disturbance is autocorrelated: If the disturbance et is autocorrelated, the OLS estimator of the AR(1) model is inconsistent. Asymptotic bias of OLS when the disturbance follows an AR(1) process: If et follows itself an AR(1) process, et = φet−1 + εt , where εt is iid white noise, the asymptotic bias of the OLS estimator of ρ is ρ+φ . plim ρ̂ = 1 + ρφ 2.3 The Autoregressive Model of Order p Lag operator: The lag operator is defined as Lyt = yt−1 and thus Lp yt = yt−p , p ∈ {. . . , −2, −1, 0, 1, 2, . . .}. Lag polynomial: A polynomial in L is called a lag polynomial. The lag polynomial a(L) ≡ a0 + a1 L + . . . + ap Lp can be applied to a time series variable yt to yield a(L)yt = a0 yt + a1 yt−1 + . . . + ap yt−p . Rules for lag polynomials: Lag polynomials can be multiplied. For example, define a(L) = a0 + a1 L and b(L) = b0 + b1 L + b2 L2 , then a(L)b(L) = a0 b0 + (a0 b1 + a1 b0 )L + (a0 b2 + a1 b1 )L2 + a1 b2 L3 . Lag polynomials can be evaluated at 1, a(1) = a0 + a1 + . . . + ap = p X ai . i=0 A lag polynomial applied to a time invariant quantity µ yields a(L)µ = a0 µ + a1 Lµ + . . . + ap Lp µ = a0 µ + a1 µ + . . . + ap µ = a(1)µ. 7 Formulas and Tables for Time Series Econometrics / Econometrics II The AR(p) model: The autoregressive model of order p is yt = µ + a1 yt−1 + . . . + ap yt−p + et , or a(L)yt = µ + et , where the disturbance process et is iid white noise with mean zero. OLS Estimation of the AR(p) model: The OLS estimator µ̂ â1 .. = (Z0 Z)−1 Z0 Y. . âp is based on the observation matrices yp+1 1 yp .. .. .. Z = . Y = . , . yT ··· 1 yT −1 · · · y1 .. . . yT −p If the disturbance is white noise, the OLS estimator is consistent and asymptotically normal. 3 3.1 Dynamic Regression Autoregressive Distributed Lag Model Autoregressive distributed lag (ADL) model: yt = µ + a1 yt−1 + a2 yt−2 + . . . + ap yt−p + b0 xt + b1 xt−1 + . . . + bq xt−q + εt or, using lag polynomials, a(L)yt = µ + b(L)xt + εt . OLS estimation of the ADL model: If E[εt |yt−1 , yt−2 , . . . , xt , xt−1 , xt−2 , . . .] = 0, the joint process {(yt , xt )} is stationary and ergodic, and there is no perfect multicollinearity among the regressors 1, yt−1 , . . . , yt−p , xt , . . . , xt−q , then the OLS estimator of the ADL model is consistent and asymptotically normally distributed. Interpretation of the ADL coefficients as partial effects: ∂ E(yt |1, yt−1 , . . . , yt−p , xt , . . . , xt−q ) = ai ∂yt−i ∂ E(yt |1, yt−1 , . . . , yt−p , xt , . . . , xt−q ) = bi ∂xt−i i ∈ [1, . . . , p] i ∈ [0, . . . , q] 8 Formulas and Tables for Time Series Econometrics / Econometrics II Long-run effect of a permanent shift in x on y: ∂ E(y|x) b(1) b0 + b1 + . . . + bq = = ∂x a(1) 1 − a1 − . . . − ap 3.2 Error-Correction Model Error-correction model (ECM): The ECM is a reparameterization of the ADL model: ∆yt = µ − α(yt−1 − βxt−1 ) + ā1 ∆yt−1 + . . . + āp−1 ∆yt−p+1 + b̄0 ∆xt + . . . + b̄q−1 ∆xt−q+1 + εt where ∆ = 1 − L and α = a(1) β = b(1)/a(1) p X ai , āi = − for i = 1, . . . , p − 1 k=i+1 b̄0 = b0 b̄i = − q X bi , for i = 1, . . . , q − 1 k=i+1 Estimation of the ECM by OLS: Writing the ECM as ∆yt = µ + (−α) yt−1 + αβ xt−1 + lagged differences + εt , |{z} | {z } γ1 γ2 it can be estimated by OLS. The OLS estimator is, under the conditions stated for the ADL model, consistent and asymptotically normal. Estimation of long-run parameters of the ECM: Calculate β̂ = − γ̂2 . γ̂1 The asymptotic standard errors of β̂ can be obtained using the delta method. 4 Tests of Autocorrelation and Lag Order Selection Durbin-Watson test: Test of the null hypothesis that the regression disturbance εt is not autocorrelated against the alternative that it is AR(1). Test statistic based on an estimation sample t = 1, . . . , T : PT (ε̂t − ε̂t−1 )2 DW = t=2PT 2 . ε̂ t=1 t 9 Formulas and Tables for Time Series Econometrics / Econometrics II LM test for autocorrelation (Breusch-Godfrey test): In the model y t = xt γ + u t , ut = ρ1 ut−1 + . . . + ρr ut−r + εt , it tests H0 : ρ1 = . . . = ρr = 0 LM = T against εt ∼ iid H1 : ¬H0 . The test statistic is ε̃0 PZ ε̃ , ε̃0 ε̃ where ε̃ = Y − Xγ̂ are the residuals obtained under H0 , Z = [X, ε̃−1 , . . . , ε̃−r ] and PZ = d Z(Z0 Z)−1 Z0 . Under H0 , LM −→ χ2r . Information criteria: Let k be the number of parameters in the ADL model. The following information criteria can be used: 2k , T 2k 2k 2 + 2k AICc(k) = log(σ̂ 2 ) + + , T T −k−1 2k log log T HQ(k) = log(σ̂ 2 ) + , T k log T BIC(k) = log(σ̂ 2 ) + . T AIC(k) = log(σ̂ 2 ) + 10 Formulas and Tables for Time Series Econometrics / Econometrics II Part B. Econometrics for Integrated I(1) Time Series Processes 5 5.1 Stochastic Trends Integrated Processes A process {yt } is called integrated of order d (I(d)) if it is nonstationary and differencing it at least d times is necessary to obtain a stationary process. In particular, a nonstationary process is I(1) if its first difference, ∆yt = (1 − L)yt = yt − yt−1 is I(0). 5.2 Stochastic Trends with iid Increments (Random Walks) Stochastic trend without drift: Let {εt } be a iid white noise process with mean zero and finite variance σ 2 . A driftless stochastic trend with iid increments is defined as s t = s 0 + ε1 + · · · + εt = s 0 + t X εt . t=1 Assuming s0 = 0, it has mean E(st ) = 0 and variance Var(st ) = tσ 2 . Stochastic trend with drift: Let {εt } be a iid white noise process with mean zero and finite variance σ 2 . A stochastic trend with drift and iid increments is defined as st = s0 + µt + ε1 + · · · + εt = s0 + t X (µ + εt ). t=1 Assuming s0 = 0, it has mean E(st ) = µt and variance Var(st ) = tσ 2 . 5.3 Stochastic Trends with Autocorrelated Increments Stochastic trend without drift: Let {ut } be a stationary linear process with mean zero and long-run variance λ2 . A driftless stochastic trend with autocorrelated increments is defined as s̃t = s̃0 + u1 + · · · + ut = s̃0 + t X ut . t=1 Assuming s̃0 =√0 and normalizing by limt→∞ Var(s̃t / t) = λ2 . √ √ t, it has mean E(s̃t / t) = 0 and asymptotic variance 11 Formulas and Tables for Time Series Econometrics / Econometrics II Stochastic trend with drift: Let {ut } be a stationary linear process with mean zero and long-run variance λ2 . A stochastic trend with drift and autocorrelated increments is defined as s̃t = s̃0 + µt + u1 + · · · + ut = s̃0 + t X (µ + ut ). t=1 Assuming s̃0 = √ 0 and normalizing by limt→∞ Var(s̃t / t) = λ2 . 5.4 √ √ √ t, it has mean E(s̃t / t) = µ t and asymptotic variance Beveridge-Nelson Decomposition Let s̃t be I(1) with drift so that ∆s̃t is a linear I(0) process with long-run variance λ2 , s̃t = s̃0 + µt + t X ⇒ ui ∆s̃t = µ + ut = µ + ψ(L)εt , i=1 where Var(εt ) = σ 2 . Then s̃t can be linearly decomposed into (a) a linear deterministic trend, (b) a random walk, (c) a stationary process η, and (d) an initial condition z0 = s̃0 − η0 : s̃t = µt |{z} divergence at rate t 6 + (λ/σ) s | {z }t divergence at rate + √ t ηt |{z} bounded in probability + z0 |{z} . bounded in probability Unit Root Asymptotics Standard Brownian motion (Wiener process): The standard Brownian motion W (t), t ∈ [0, 1] is defined as a stochastic process with the following properties. • Initialization: W (0) = 0. • Normality: W (t) is distributed as N (0, t). • Independent increments: For given points in time 0 ≤ t1 < t2 < . . . tk ≤ 1, the increments (W (t2 ) − W (t1 )), (W (t3 ) − W (t2 )), . . . , (W (tk ) − W (tk−1 )) are stochastically independent. Any increment W (s) − W (t), s > t, is normally distributed with mean zero and variance s − t. • The process is continuous in t. 12 Formulas and Tables for Time Series Econometrics / Econometrics II Functional central limit theorem: Let ε1 , . . . , εT be iid white noise random variables with mean zero and and variance σ 2 . Furthermore define [T r], 0 ≤ r ≤ 1, as the largest integer smaller or equal to T r. Then the step-function [T r] 1 X εt , 0 ≤ r ≤ 1 XT (r) = √ σ T t=1 with XT (r) = 0 for [T r] < 1 converges weakly towards a standard Brownian motion, XT (r) ⇒ W (r), as T → ∞. Continuous mapping theorem: By the continuous mapping theorem, if XT (r) ⇒ W (r) then f (XT (r)) ⇒ f (W (r)) for continuous functions f . Convergence of sample moments of random walks without drift: Let ε1 , . . . , εT be iid white noise random variables with mean zero and and variance σ 2 and define the random walk without drift yt = yt−1 + εt , y0 = 0. Then the following convergence results hold: R1 P ⇒ σ 0 W (r)dr (1) T −3/2 Tt=1 yt−1 (2) T −2 PT (3) T −1 PT 2 t=1 yt−1 t=1 ⇒ σ2 R1 0 (W (r))2 dr ∆yt yt−1 ⇒ 0.5σ 2 (W (1)2 − 1) Convergence of sample moments of a stochastic trend with correlated increments: Let {ut } be a linear I(0) process with mean zero, finite variance γ0 and finite long-run variance λ2 > 0 and define the driftless stochastic trend ỹ t = ỹ t−1 + ut , ut = ψ(L)εt , ỹ 0 = 0. Then the following convergence results hold: R1 P (1) T −3/2 Tt=1 ỹ t−1 ⇒ λ 0 W (r)dr 7 7.1 (2) T −2 PT (3) T −1 PT t=1 t=1 ỹ 2t−1 ⇒ λ2 R1 0 (W (r))2 dr ∆ỹ t ỹ t−1 ⇒ 0.5λ2 W (1)2 − 0.5γ0 Unit Root Tests Dickey-Fuller (DF) Tests Dickey-Fuller test without intercept and trend: Estimate by OLS the population model yt = ρyt−1 + εt , 13 Formulas and Tables for Time Series Econometrics / Econometrics II where εt is iid white noise with variance σ 2 and compute the test statistics ! PT y y t t−1 −1 DF-ρ = T (ρ̂ − 1) = T Pt=1 T 2 y t=1 t−1 ρ̂ − 1 ρ̂ − 1 DF-t = = PT 2 . SE(ρ̂) σ̂( t=1 yt−1 )−1/2 Under the null hypothesis ρ = 1, the test statistics have limiting distribution 0.5(W (1)2 − 1) DF-ρ ⇒ DFρ ≡ R 1 (W (r))2 dr 0 0.5(W (1)2 − 1) DF-t ⇒ DFt ≡ qR . 1 2 (W (r)) dr 0 Dickey-Fuller test with intercept: Estimate by OLS the population model yt = α + ρyt−1 + εt , where εt is iid white noise with variance σ 2 and compute the test statistics DF-ρµ = T (ρ̂µ − 1) ρ̂µ − 1 DF-tµ = . SE(ρ̂µ ) Under the null hypothesis ρ = 1, the test statistics have limiting distribution 0.5([W µ (1)]2 − W µ (0)2 − 1) R1 (W µ (r))2 dr 0 0.5([W µ (1)]2 − W µ (0)2 − 1) qR DF-tµ ⇒ DFµt ≡ . 1 µ 2 (W (r)) dr 0 DF-ρµ ⇒ DFµρ ≡ Dickey-Fuller test with intercept and trend: Estimate by OLS the population model yt = α0 + α1 t + ρyt−1 + εt , where εt is iid white noise with variance σ 2 and compute the test statistics DF-ρτ = T (ρ̂τ − 1) ρ̂τ − 1 DF-tτ = . SE(ρ̂τ ) 14 Formulas and Tables for Time Series Econometrics / Econometrics II Under the null hypothesis ρ = 1, the test statistics have limiting distribution 0.5([W τ (1)]2 − W τ (0)2 − 1) R1 (W τ (r))2 dr 0 0.5([W τ (1)]2 − W τ (0)2 − 1) qR DF-tτ ⇒ DFτt ≡ . 1 τ (r))2 dr (W 0 DF-ρτ ⇒ DFτρ ≡ 7.2 Phillips-Perron (PP) Test Phillips-Perron test without intercept and trend: Estimate the DF-regression yt = ρyt−1 + ut , where ut is potentially autocorrelated. In addition, estimate the autocovariances of ut , γk , up to order q, from which an estimate of the long-run variance, λ2 , is computed. Then, under the null of ρ = 1, PP-ρ ≡ T (ρ̂ − 1) − 0.5 T 2 (SE(ρ̂))2 /σ̂u2 (λ̂2 − σ̂u2 ) converges to DFρ and s λ̂2 − σ̂u2 σ̂u2 ρ̂ − 1 − 0.5 PP-t ≡ (T × SE(ρ̂)/σ̂u ) λ̂2 SE(ρ̂) λ̂ converges to DFt . 7.3 Augmented Dickey-Fuller (ADF) Tests Augmented Dickey-Fuller test without intercept and trend: Estimate by OLS the augmented model yt = ρyt−1 + φ1 ∆ỹ t−1 + · · · + φp ∆ỹ t−p + vt and compute the t-statistic ADF-t = ρ̂ − 1 . SE(ρ̂) Under the null hypothesis ρ = 1, the ADF-t statistic has limiting distribution ADF-t ⇒ DFt . 15 Formulas and Tables for Time Series Econometrics / Econometrics II Augmented Dickey-Fuller test with intercept: Estimate by OLS the augmented model yt = α + ρyt−1 + φ1 ∆ỹ t−1 + · · · + φp ∆ỹ t−p + vt and compute the t-statistic ADF-tµ = ρ̂µ − 1 . SE(ρ̂µ ) Under the null hypothesis ρ = 1, the ADF-tµ statistic has limiting distribution ADF-tµ ⇒ DFµt . Augmented Dickey-Fuller test with intercept and trend: Estimate by OLS the augmented model yt = α0 + α1 t + ρyt−1 + φ1 ∆ỹ t−1 + · · · + φp ∆ỹ t−p + vt and compute the t-statistic ADF-tτ = ρ̂τ − 1 . SE(ρ̂τ ) Under the null hypothesis ρ = 1, the ADF-tτ statistic has limiting distribution ADF-tτ ⇒ DFτt . Choice of the lag order: Information criteria such as the AIC or BIC can be used to determine the lag order p. The maximum lag order pmax in this search can be chosen as the integer part of 12 × (T /100)1/4 . 8 8.1 Cointegration Multivariate Beveridge-Nelson decomposition Linear vector I(0) process: A linear n dimensional zero-mean vector I(0) process is ut = εt + Ψ1 εt−1 + Ψ2 εt−2 + . . . = Ψ(L)εt , Ψ0 = I, where εt is iid with mean zero and positive definite variance matrix Ω, if (C1) (C2) the sequence I, Ψ1 , Ψ2 , . . . is one-summable so that Ψ(1) = I + Ψ1 + Ψ2 + · · · is finite, at least one element of Ψ(1) is nonzero. Long-run variance matrix: The long-run variance matrix of a linear vector I(0) process is √ Λ ≡ lim Var( T ūT ) = Ψ(1)ΩΨ(1)0 . T →∞ 16 Formulas and Tables for Time Series Econometrics / Econometrics II Vector I(1) process: A vector I(1) process of dimension n is ∆yt = δ + ut = δ + Ψ(L)εt , where ut = Ψ(L)εt is a linear zero-mean vector I(0) process. Multivariate Beveridge-Nelson decomposition: A vector I(1) process of dimension n can be decomposed into yt = δ · t + Ψ(1) t X εi + η t + z 0 , i=1 where δ · t is a deterministic trend, Ψ(1)[ε1 + · · · + εt ] is a stochastic trend, η t is a linear vector I(0) process, and z0 is an initial condition. 8.2 Cointegration and common stochastic trends Cointegration: Let yt be an n-dimensional vector I(1) process with stochastic increment ut that is a linear vector I(0) process. Then yt is cointegrated with cointegration vector β 1 if β 1 6= 0 and β 01 yt is stationary. (In a strict sense, stationarity of β 01 yt requires a suitable choice of initial conditions.) Cointegration rank: The cointegration rank is the number, r, of linearly independent cointegration vectors β 1 , . . . , β r , and the cointegration space is the space spanned by the cointegration vectors. Cointegration matrix: The (n × r) cointegration matrix β = [β 1 , . . . , β r ] is the matrix of all cointegration vectors. The cointegration matrix has rank r. Rank of Ψ(1): The matrix Ψ(1) of the multivariate Beveridge-Nelson decomposition has rank k ≡ n − r. Common stochastic trends: A cointegrated n-dimensional vector I(1) process with cointegration rank r is driven by k = n − r common stochastic trends. 8.3 Engle-Granger test of the null of no cointegration Spurious regression: Suppose yt is an n-dimensional vector I(1) process and not cointegrated. Then the regression y1,t = µ + γ2 y2,t + · · · + γn yn,t + zt∗ is called spurious. The disturbance zt∗ is I(1) and the OLS estimator has a non-standard asymptotic distribution. 17 Formulas and Tables for Time Series Econometrics / Econometrics II Cointegrating regression: Suppose yt is an n-dimensional vector I(1) process and cointegrated with cointegration rank r = 1. Then the regression y1,t = µ + γ2 y2,t + · · · + γn yn,t + zt∗ , is called cointegrating regression. The disturbance zt∗ is I(0). The OLS estimator is superconsistent but has a non-standard asymptotic distribution. Engle-Granger test: Suppose yt is an n-dimensional vector I(1) and either cointegrated with cointegration rank r = 1 or not cointegrated. The hypotheses H0 : r = 0 versus H1 : r = 1 can be tested as follows. Stage 1. Estimate by OLS the regression y1,t = µ̂ + γ̂2 y2,t + · · · + γ̂n yn,t + ẑt∗ . Stage 2. Estimate the ADF regression without intercept and trend on the first-stage residuals ∗ ∗ ∗ ẑt∗ = ρẑt−1 + φ1 ∆ẑt−1 + · · · + φp ∆ẑt−p + vt . Stage 3. Construct the ADF statistic ADF-t = ρ̂ − 1 SE(ρ̂) and compare it with the appropriate critical values. 18 Formulas and Tables for Time Series Econometrics / Econometrics II Part C. The Vectorautoregressive (VAR) Model 9 The VAR(1) model (reduced form) Model equation: yt = B1 yt−1 + ut , where • yt is a K × 1 vector of economic variables • A1 is a K × K parameter matrix • ut is a K × 1 vector of reduced form white noise disturbances • Sample runs from t = 0, 1, . . . , T . Moving average representation: yt = ut + B1 ut−1 + B21 ut−2 + B31 ut−3 + . . . ∞ ∞ X X i = B1 ut−i = Ci ut−i , Ci ≡ Bi1 i=0 i=0 Stability condition: Stability requires that Bi1 → 0 at i → ∞. Hence, a sufficient condition is that the eigenvalues of B1 are all absolutely smaller than 1. 10 The VAR(p) model (reduced form) Model equation: yt = B1 yt−1 + · · · + Bp yt−p + ut or, compactly, B(L)yt = ut , where B(L) = I − B1 L1 − · · · − Bp Lp is the lag polynomial in lag order p. 19 Formulas and Tables for Time Series Econometrics / Econometrics II Moving average representation: The result is qualitatively the same as for VAR(1): yt = ∞ X Ci ut−i = C(L)ut−i with C(L) = i=0 ∞ X Ci Li i=0 Computation of the parameter matrices: i p = 1: C0 = IK , Ci = B i = 1, 2, . . . P1i, p > 1: C0 = IK , Ci = j=1 Ci−j Bj , i = 1, 2, . . . Wold’s decomposition theorem: Every stationary process xt can be written as the sum of two uncorrelated processes zt and yt , where zt is a deterministic process and yt is a process with MA representation, i.e., yt = C(L)ut , 11 ut ∼ white noise. Estimation and Inference 11.1 System OLS A system OLS estimator B̂ yields the same results as equation-wise OLS but can be written more efficiently: yt−1 yt = B1 yt−1 + · · · + Bp yt−p + ut = [B1 , . . . , Bp ] ... + ut = BXt−1 + ut yt−p where • B = [B1 , . . . , Bp ] is a K × Kp coefficient matrix 0 0 ]0 is a Kp × 1 observation vector • Xt−1 = [yt−1 , . . . , yt−p Now stack observations horizontally: • y = [y1 , . . . , yT ] is a K × T observation matrix • X = [X0 , . . . , XT −1 ] is a Kp × T observation matrix • u = [u1 , . . . , uT ] is a K × T error matrix OLS estimator: B̂ = yX0 (XX0 )−1 = T X t=1 ! yt X0t−1 T X !−1 Xt−1 X0t−1 t=1 To estimate Σ, construct the residuals û = y − B̂X and compute Σ̂ = ûû0 /T, Note that OLS is consistent because the regressors are lags of yt and the error term ut is white noise and thus not correlated with these lags. 20 Formulas and Tables for Time Series Econometrics / Econometrics II 11.2 Maximum likelihood estimation Assuming normality of the disturbances, ut ∼ N (0, Σ) , t = 1, . . . , T, the parameters of a VAR model can also be estimated using a maximum likelihood estimator. Define the set of past observations as It−1 = {yt−1 , yt−2 , yt−3 , . . .}. Conditional on It−1 , yt is a linear function of ut and thus normally distributed with E(yt |It−1 ) = B1 yt−1 + . . . + Bp yt−p Var(yt |It−1 ) = Var(ut ) = Σ Hence, conditional on the past, we have the normal distribution yt |It−1 ∼ N (B1 yt−1 + . . . + Bp yt−p , Σ) , t = 1, . . . , T The conditional likelihood of observation t given the past is thus 1 1 log |Σ| − (yt − B1 yt−1 − . . . − Bp yt−p )0 × 2 2 −1 Σ × (yt − B1 yt−1 − . . . − Bp yt−p ) , `t (θ) ≡ log f (yt |It−1 , θ) = const − where θ is a vector of all parameters in B1 , . . . , Bp , Σ. 11.3 Inference Stationary variables: • Recall that the OLS estimator of a time series regression with white noise disturbances is asymptotically normally distributed by a martingale difference CLT. • Hence, we can use standard inference for the VAR parameters. • For functions of these parameters, e.g., impulse response coefficients, we can use the delta method to find the asymptotic normal distribution of their estimators. Nonstationary variables: If the VAR includes I(1) variables, fortunately not much changes as can be seen from reparameterizing a VAR(2) to a multivariate ADF-type representation: yt = B1 yt−1 + B2 yt−2 + ut = (B1 + B2 ) yt−1 + (−B2 ) ∆yt−1 + ut | {z } | {z } C Γ1 Estimating this equation by OLS yields Ĉ that may be superconsistent (convergence rate of T ) because the regressors are I(1) and Γ̂1 that is root-T consistent because the regressors are differenced and thus I(0). 21 Formulas and Tables for Time Series Econometrics / Econometrics II 12 Specification Information Criteria: Akaike criterion (AIC), Hannan-Quinn criterion (HQC), Schwarz criterion (SC): 2 × (# freely estimated parameters) T 2mK 2 = ln |Σ̂| + T AIC(m) = ln |Σ̂| + 2 ln ln T × (# freely estimated parameters) T 2 ln ln T = ln |Σ̂| + mK 2 T HQC(m) = ln |Σ̂| + ln T × (# freely estimated parameters) T ln T = ln |Σ̂| + mK 2 T SC(m) = ln |Σ̂| + where you try all m = 1, . . . , pmax . Asymptotics: HQC and SC are consistent, hence pick the right lag order as T → ∞. AIC is not: it has a positive probability to choose a lag order that is too high. 13 Forecasting Granger causality: A variable y1t is Granger causal for a variable y2t if its knowledge helps to reduce, on average, the prediction error for y2t . If b121 = b221 = 0, then y1t is not Granger causal for y2t . h-step forecast: yT +h|T = E(yT +h |IT ) = B1 E(yT +h−1 |IT ) + · · · + Bp E(yT +h−p |IT ) Forecast error: eT +h|T = yT +h − yT +h|T = yT +h − E(yT +h |IT ) The forecast error covariance matrix: the forecast MSE converges, as h increases, toward the unconditional variance matrix of yt from below: MSE(eT +h|T ) = Var(eT +h|T ) = h−1 X Cj ΣC0j → Var(yt ). j=0 22 Formulas and Tables for Time Series Econometrics / Econometrics II Sources of uncertainty: future shocks and estimation error: yT +h − ŷT +h|T = (yT +h − yT +h|T ) + (yT +h|T − ŷT +h|T ) . | {z } | {z } eT +h|T Estimation error Since the first part depends on future shocks and the second part on the shocks in the sample, the two are uncorrelated. Thus the forecast MSE is MSE(yT +h − ŷT +h|T ) = MSE(eT +h|T ) + MSE(yT +h|T − ŷT +h|T ). 14 Structural VAR Model equation: A0 yt = A1 yt−1 + · · · + Ap yt−p + Dεt . Leaving aside the lags which account for the dynamics, the contemporaneous relationship between variables and structural shocks is A0 yt = Dεt . Central assumption: the structural shocks are mutually uncorrelated so that a ceteris paribus interpretation is possible. Since we do not interpret the scale of the shocks, we normalize the variances to 1: Var(εt ) = I. Finding the reduced form VAR from the structural one: multiply by A−1 0 from the left, −1 yt = A0−1 A1 yt−1 + · · · + A−1 0 Ap yt−p + A0 Dεt . From there, define the reduced form parameters as Bi = A−1 0 Ai and the reduced form disturbances ut = A−1 Dε : t 0 yt = B1 yt−1 + · · · + Bp yt−p + ut . 14.1 Estimating the parameters of the structural form 1) Taking the variance of ut = A−1 0 Dεt we find 0 0 0 0 −1 −1 0 0 −1 0 −1 = A−1 Σ ≡ E(ut u0t ) = E(A−1 0 DD A0 0 Dεt εt D A0 ) = A0 D E(εt εt )D A0 0 We can first estimate Σ from the reduced form and then solve 0 −1 0 Σ̂ = Â−1 0 D̂D̂ (Â0 ) for the structural parameters in Â0 and D̂. 2) From Bi = A−1 0 Ai we find Ai = A0 Bi , i = 1, . . . , p. Hence, once we know Â0 from step 1, we can compute Âi = Â0 B̂i , i = 1, . . . , p. 23 Formulas and Tables for Time Series Econometrics / Econometrics II 14.2 Identification of the structural form 0 0 −1 Due to symmetry, Σ has K(K + 1)/2 distinct parameters, hence Σ = A−1 imposes 0 DD A0 2 K(K + 1)/2 on the 2K parameters in A0 and D. The contemporaneous relationship between variables and structural shocks, A0 yt = Dεt , needs to impose enough restrictions such that the parameters in A0 and D can be recovered from Σ. Two cases commonly used to achieve identification: • A0 = I and D is lower triangular, or • D = I and A0 is lower triangular. • Both imply that A−1 0 D and its inverse are lower triangular. 15 Impulse response analysis To find the impulse responses, we start from the MA representation yt = ∞ X Ci ut−i i=0 and substitute ut = A−1 0 Dεt to get the structural shocks: yt = ∞ X −1 −1 −1 Ci A−1 0 Dεt−i = A0 Dεt + C1 A0 Dεt−1 + C2 A0 Dεt−2 + . . . i=0 Hence, ∂yt = Cτ A−1 0 D. ∂ε0t−τ Therefore, the impulse response function of variable i, yit , w.r.t. structural shock j collects elements i, j from −1 −1 −1 A0 D, C1 A−1 0 D, C2 A0 D, C3 A0 D, . . . . 24 Formulas and Tables for Time Series Econometrics / Econometrics II Part D. Tables Percentiles of the χ2–distribution Fχ 2 r=1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 50 60 70 80 90 100 0.0100 0.0002 0.0201 0.1148 0.2971 0.5543 0.8721 1.2390 1.6465 2.0879 2.5582 3.0535 3.5706 4.1069 4.6604 5.2293 5.8122 6.4078 7.0149 7.6327 8.2604 8.8972 9.5425 10.1957 10.8564 11.5240 12.1981 12.8785 13.5647 14.2565 14.9535 22.1643 29.7067 37.4849 45.4417 53.5401 61.7541 70.0649 0.0250 0.0010 0.0506 0.2158 0.4844 0.8312 1.2373 1.6899 2.1797 2.7004 3.2470 3.8157 4.4038 5.0088 5.6287 6.2621 6.9077 7.5642 8.2307 8.9065 9.5908 10.2829 10.9823 11.6886 12.4012 13.1197 13.8439 14.5734 15.3079 16.0471 16.7908 24.4330 32.3574 40.4817 48.7576 57.1532 65.6466 74.2219 0.0500 0.0039 0.1026 0.3518 0.7107 1.1455 1.6354 2.1673 2.7326 3.3251 3.9403 4.5748 5.2260 5.8919 6.5706 7.2609 7.9616 8.6718 9.3905 10.1170 10.8508 11.5913 12.3380 13.0905 13.8484 14.6114 15.3792 16.1514 16.9279 17.7084 18.4927 26.5093 34.7643 43.1880 51.7393 60.3915 69.1260 77.9295 0.1000 0.0158 0.2107 0.5844 1.0636 1.6103 2.2041 2.8331 3.4895 4.1682 4.8652 5.5778 6.3038 7.0415 7.7895 8.5468 9.3122 10.0852 10.8649 11.6509 12.4426 13.2396 14.0415 14.8480 15.6587 16.4734 17.2919 18.1139 18.9392 19.7677 20.5992 29.0505 37.6886 46.4589 55.3289 64.2778 73.2911 82.3581 0.9000 2.7055 4.6052 6.2514 7.7794 9.2364 10.6446 12.0170 13.3616 14.6837 15.9872 17.2750 18.5493 19.8119 21.0641 22.3071 23.5418 24.7690 25.9894 27.2036 28.4120 29.6151 30.8133 32.0069 33.1962 34.3816 35.5632 36.7412 37.9159 39.0875 40.2560 51.8051 63.1671 74.3970 85.5270 96.5782 107.5650 118.4980 0.9500 3.8415 5.9915 7.8147 9.4877 11.0705 12.5916 14.0671 15.5073 16.9190 18.3070 19.6751 21.0261 22.3620 23.6848 24.9958 26.2962 27.5871 28.8693 30.1435 31.4104 32.6706 33.9244 35.1725 36.4150 37.6525 38.8851 40.1133 41.3371 42.5570 43.7730 55.7585 67.5048 79.0819 90.5312 101.8795 113.1453 124.3421 0.9750 5.0239 7.3778 9.3484 11.1433 12.8325 14.4494 16.0128 17.5345 19.0228 20.4832 21.9200 23.3367 24.7356 26.1189 27.4884 28.8454 30.1910 31.5264 32.8523 34.1696 35.4789 36.7807 38.0756 39.3641 40.6465 41.9232 43.1945 44.4608 45.7223 46.9792 59.3417 71.4202 83.2977 95.0232 106.6286 118.1359 129.5612 0.9900 6.6349 9.2103 11.3449 13.2767 15.0863 16.8119 18.4753 20.0902 21.6660 23.2093 24.7250 26.2170 27.6882 29.1412 30.5779 31.9999 33.4087 34.8053 36.1909 37.5662 38.9322 40.2894 41.6384 42.9798 44.3141 45.6417 46.9629 48.2782 49.5879 50.8922 63.6907 76.1539 88.3794 100.4252 112.3288 124.1163 135.8067 25 Formulas and Tables for Time Series Econometrics / Econometrics II Percentiles of the t–distribution Ft r=1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 ∞ 0.9000 3.0777 1.8856 1.6377 1.5332 1.4759 1.4398 1.4149 1.3968 1.3830 1.3722 1.3634 1.3562 1.3502 1.3450 1.3406 1.3368 1.3334 1.3304 1.3277 1.3253 1.3232 1.3212 1.3195 1.3178 1.3163 1.3150 1.3137 1.3125 1.3114 1.3104 1.2816 0.9500 6.3138 2.9200 2.3534 2.1318 2.0150 1.9432 1.8946 1.8595 1.8331 1.8125 1.7959 1.7823 1.7709 1.7613 1.7531 1.7459 1.7396 1.7341 1.7291 1.7247 1.7207 1.7171 1.7139 1.7109 1.7081 1.7056 1.7033 1.7011 1.6991 1.6973 1.6449 0.9750 12.7062 4.3027 3.1824 2.7764 2.5706 2.4469 2.3646 2.3060 2.2622 2.2281 2.2010 2.1788 2.1604 2.1448 2.1314 2.1199 2.1098 2.1009 2.0930 2.0860 2.0796 2.0739 2.0687 2.0639 2.0595 2.0555 2.0518 2.0484 2.0452 2.0423 1.9600 0.9900 31.8205 6.9646 4.5407 3.7469 3.3649 3.1427 2.9980 2.8965 2.8214 2.7638 2.7181 2.6810 2.6503 2.6245 2.6025 2.5835 2.5669 2.5524 2.5395 2.5280 2.5176 2.5083 2.4999 2.4922 2.4851 2.4786 2.4727 2.4671 2.4620 2.4573 2.3263 0.9950 63.6567 9.9248 5.8409 4.6041 4.0321 3.7074 3.4995 3.3554 3.2498 3.1693 3.1058 3.0545 3.0123 2.9768 2.9467 2.9208 2.8982 2.8784 2.8609 2.8453 2.8314 2.8188 2.8073 2.7969 2.7874 2.7787 2.7707 2.7633 2.7564 2.7500 2.5758 26 Percentiles of the F –distribution The Table shows the values k, for which P (v ≤ k) = F (k) = 0.95 holds. (r1 = degrees of freedom of the nominator,r2 = degrees of freedom of the denominator) r1 = 1 161.4476 18.5128 10.1280 7.7086 6.6079 5.9874 5.5914 5.3177 5.1174 4.9646 4.8443 4.7472 4.6672 4.6001 4.5431 4.4940 4.4513 4.4139 4.3807 4.3512 4.3248 4.3009 4.2793 4.2597 4.2417 4.2252 4.2100 4.1960 4.1830 4.1709 4.0847 4.0012 3.9201 3.8416 2 199.5000 19.0000 9.5521 6.9443 5.7861 5.1433 4.7374 4.4590 4.2565 4.1028 3.9823 3.8853 3.8056 3.7389 3.6823 3.6337 3.5915 3.5546 3.5219 3.4928 3.4668 3.4434 3.4221 3.4028 3.3852 3.3690 3.3541 3.3404 3.3277 3.3158 3.2317 3.1504 3.0718 2.9958 3 215.7073 19.1643 9.2766 6.5914 5.4095 4.7571 4.3468 4.0662 3.8625 3.7083 3.5874 3.4903 3.4105 3.3439 3.2874 3.2389 3.1968 3.1599 3.1274 3.0984 3.0725 3.0491 3.0280 3.0088 2.9912 2.9752 2.9604 2.9467 2.9340 2.9223 2.8387 2.7581 2.6802 2.6050 4 224.5832 19.2468 9.1172 6.3882 5.1922 4.5337 4.1203 3.8379 3.6331 3.4780 3.3567 3.2592 3.1791 3.1122 3.0556 3.0069 2.9647 2.9277 2.8951 2.8661 2.8401 2.8167 2.7955 2.7763 2.7587 2.7426 2.7278 2.7141 2.7014 2.6896 2.6060 2.5252 2.4472 2.3720 5 230.1619 19.2964 9.0135 6.2561 5.0503 4.3874 3.9715 3.6875 3.4817 3.3258 3.2039 3.1059 3.0254 2.9582 2.9013 2.8524 2.8100 2.7729 2.7401 2.7109 2.6848 2.6613 2.6400 2.6207 2.6030 2.5868 2.5719 2.5581 2.5454 2.5336 2.4495 2.3683 2.2899 2.2142 6 233.9860 19.3295 8.9406 6.1631 4.9503 4.2839 3.8660 3.5806 3.3738 3.2172 3.0946 2.9961 2.9153 2.8477 2.7905 2.7413 2.6987 2.6613 2.6283 2.5990 2.5727 2.5491 2.5277 2.5082 2.4904 2.4741 2.4591 2.4453 2.4324 2.4205 2.3359 2.2541 2.1750 2.0987 7 236.7684 19.3532 8.8867 6.0942 4.8759 4.2067 3.7870 3.5005 3.2927 3.1355 3.0123 2.9134 2.8321 2.7642 2.7066 2.6572 2.6143 2.5767 2.5435 2.5140 2.4876 2.4638 2.4422 2.4226 2.4047 2.3883 2.3732 2.3593 2.3463 2.3343 2.2490 2.1665 2.0868 2.0097 8 238.8827 19.3710 8.8452 6.0410 4.8183 4.1468 3.7257 3.4381 3.2296 3.0717 2.9480 2.8486 2.7669 2.6987 2.6408 2.5911 2.5480 2.5102 2.4768 2.4471 2.4205 2.3965 2.3748 2.3551 2.3371 2.3205 2.3053 2.2913 2.2783 2.2662 2.1802 2.0970 2.0164 1.9385 9 240.5433 19.3848 8.8123 5.9988 4.7725 4.0990 3.6767 3.3881 3.1789 3.0204 2.8962 2.7964 2.7144 2.6458 2.5876 2.5377 2.4943 2.4563 2.4227 2.3928 2.3660 2.3419 2.3201 2.3002 2.2821 2.2655 2.2501 2.2360 2.2229 2.2107 2.1240 2.0401 1.9588 1.8800 10 241.8817 19.3959 8.7855 5.9644 4.7351 4.0600 3.6365 3.3472 3.1373 2.9782 2.8536 2.7534 2.6710 2.6022 2.5437 2.4935 2.4499 2.4117 2.3779 2.3479 2.3210 2.2967 2.2747 2.2547 2.2365 2.2197 2.2043 2.1900 2.1768 2.1646 2.0772 1.9926 1.9105 1.8308 20 248.0131 19.4458 8.6602 5.8025 4.5581 3.8742 3.4445 3.1503 2.9365 2.7740 2.6464 2.5436 2.4589 2.3879 2.3275 2.2756 2.2304 2.1906 2.1555 2.1242 2.0960 2.0707 2.0476 2.0267 2.0075 1.9898 1.9736 1.9586 1.9446 1.9317 1.8389 1.7480 1.6587 1.5706 120 253.2529 19.4874 8.5494 5.6581 4.3985 3.7047 3.2674 2.9669 2.7475 2.5801 2.4480 2.3410 2.2524 2.1778 2.1141 2.0589 2.0107 1.9681 1.9302 1.8963 1.8657 1.8380 1.8128 1.7896 1.7684 1.7488 1.7306 1.7138 1.6981 1.6835 1.5766 1.4673 1.3519 1.2216 Formulas and Tables for Time Series Econometrics / Econometrics II F (k) = 0.95 r2 = 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 60 120 ∞ 27 Formulas and Tables for Time Series Econometrics / Econometrics II Critical values for the DF tests Table 1: Empirical cumulative distribution of the DF-ρ statistic under H0 : ρ = 1 Probability that the statistic is less than entry Sample size (T) 0.01 0.025 0.05 0.10 0.90 0.95 0.975 0.99 (a) No intercept, no trend 25 -11.8 50 -12.8 100 -13.3 250 -13.6 500 -13.7 ∞ -13.8 -9.3 -9.9 -10.2 -10.4 -10.4 -10.5 -7.3 -7.7 -7.9 -8.0 -8.0 -8.1 -5.3 -5.5 -5.6 -5.7 -5.7 -5.7 1.01 0.97 0.95 0.94 0.93 0.93 1.41 1.34 1.31 1.29 1.28 1.28 1.78 1.69 1.65 1.62 1.61 1.60 2.28 2.16 2.09 2.05 2.04 2.03 (b) Intercept, no trend 25 -17.2 50 -18.9 100 -19.8 250 -20.3 500 -20.5 ∞ -20.7 -14.6 -15.7 -16.3 -16.7 -16.8 -16.9 -12.5 -13.3 -13.7 -13.9 -14.0 -14.1 -10.2 -10.7 -11.0 -11.1 -11.2 -11.3 -0.76 -0.81 -0.83 -0.84 -0.85 -0.85 0.00 -0.07 -0.11 -0.13 -0.14 -0.14 0.65 0.53 0.47 0.44 0.42 0.41 1.39 1.22 1.14 1.08 1.07 1.05 (c) Intercept and trend 25 -22.5 50 -25.8 100 -27.4 250 -28.5 500 -28.9 ∞ -29.4 -20.0 -22.4 -23.7 -24.4 -24.7 -25.0 -17.9 -19.7 -20.6 -21.3 -21.5 -21.7 -15.6 -16.8 -17.5 -17.9 -18.1 -18.3 -3.65 -3.71 -3.74 -3.76 -3.76 -3.77 -2.51 -2.60 -2.63 -2.65 -2.66 -2.67 -1.53 -1.67 -1.74 -1.79 -1.80 -1.81 -0.46 -0.67 -0.76 -0.83 -0.86 -0.88 Source: Hayashi (2000), p. 576. 28 Formulas and Tables for Time Series Econometrics / Econometrics II Table 2: Empirical cumulative distribution of the DF-t statistic under H0 : ρ = 1 Probability that the statistic is less than entry Sample size (T) 0.01 0.025 0.05 0.10 0.90 0.95 0.975 0.99 (a) No intercept, no trend 25 -2.65 50 -2.62 100 -2.60 250 -2.58 500 -2.58 ∞ -2.58 -2.26 -2.25 -2.24 -2.24 -2.23 -2.23 -1.95 -1.95 -1.95 -1.95 -1.95 -1.95 -1.60 -1.61 -1.61 -1.62 -1.62 -1.62 0.92 0.91 0.90 0.89 0.89 0.89 1.33 1.31 1.29 1.28 1.28 1.28 1.70 1.66 1.64 1.63 1.62 1.62 2.15 2.08 2.04 2.02 2.01 2.01 (b) Intercept, no trend 25 -3.75 50 -3.59 100 -3.50 250 -3.45 500 -3.44 ∞ -3.42 -3.33 -3.23 -3.17 -3.14 -3.13 -3.12 -2.99 -2.93 -2.90 -2.88 -2.87 -2.86 -2.64 -2.60 -2.59 -2.58 -2.57 -2.57 -0.37 -0.41 -0.42 -0.42 -0.44 -0.44 0.00 -0.04 -0.05 -0.06 -0.07 -0.08 0.34 0.28 0.26 0.24 0.24 0.23 0.72 0.66 0.63 0.62 0.61 0.60 (c) Intercept and trend 25 -4.38 50 -4.15 100 -4.05 250 -3.98 500 -3.97 ∞ -3.96 -3.95 -3.80 -3.73 -3.69 -3.67 -3.66 -3.60 -3.50 -3.45 -3.42 -3.42 -3.41 -3.24 -3.18 -3.15 -3.13 -3.13 -3.12 -1.14 -1.19 -1.22 -1.23 -1.24 -1.25 -0.81 -0.87 -0.90 -0.92 -0.93 -0.94 -0.50 -0.58 -0.62 -0.64 -0.65 -0.66 -0.15 -0.24 -0.28 -0.31 -0.32 -0.33 Source: Hayashi (2000), p. 578. 29 Formulas and Tables for Time Series Econometrics / Econometrics II Critical values for the ADF test for cointegration (Engle-Granger test) Number of regressors, excluding constant 1% 2.5% 5% 10% A. Regressors have no drift 1 -3.96 2 -4.31 3 -4.73 4 -5.07 5 -5.28 -3.64 -4.02 -4.37 -4.71 -4.98 -3.37 -3.77 -4.11 -4.45 -4.71 -3.07 -3.45 -3.83 -4.16 -4.43 B. Some regressors have drift 1 -3.96 2 -4.36 3 -4.65 4 -5.04 5 -5.36 -3.67 -4.07 -4.39 -4.77 -5.02 -3.41 -3.80 -4.16 -4.49 -4.74 -3.13 -3.52 -3.84 -4.20 -4.46 Source: Hayashi (2000), Table 10.1, p. 646. 30