Nội dung

APPLIED ECONOMETRIC TIME SERIES 4TH ED. WALTER ENDERS Chapter 4 Chapter 4 WALTER ENDERS, UNIVERSITY OF ALABAMA Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. The Random Walk Model yt = yt–1 + et (or Dyt = et). t Hence yt  y0    i i 1 Given the first t realizations of the {et} process, the conditional mean of yt+1 is Etyt+1 = Et(yt + et+1) = yt Similarly, the conditional mean of yt+s (for any s > 0) can be obtained from s Et yt s  yt  Et   t i  yt i 1 var(yt) = var(et + et–1 + ... + e1) = ts2 var(yt–s) = var(et–s + et–s–1 + ... + e1) = (t – s)s Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Random Walk Plus Drift yt = yt–1 + a0 + et Given the initial condition y0, the general solution for yt is t yt  y 0  a0 t    i i 1 t s y t  s  y0  a0 ( t  s )    i i 1 Etyt+s = yt + a0s. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. The autocorrelation coefficient E[(yt – y0)(yt–s – y0)] = E[(et + et–1+...+ e1)(et–s+ et–s–1 +...+e1)] = E[(et–s)2+(et–s–1)2+...+(e1)2] = (t – s)s2  s (t  s ) / (t  s )t = [(t – s)/t]0.5 Hence, in using sample data, the autocorrelation function for a random walk process will show a slight tendency to decay. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Panel (a): Random Walk Panel (b): Random Walk Plus Drift 12 60 10 50 8 40 6 30 4 20 2 10 0 0 10 20 30 40 50 60 70 80 90 100 10 Panel (c): T rend Stationary 20 30 40 50 60 70 80 90 100 90 100 Panel (d): Random Walk Plus Noise 60 14 12 50 10 40 8 30 6 20 4 10 2 0 0 10 20 30 40 50 60 70 80 90 100 10 20 30 40 Figure 4.2: Four Series With Trends Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. 50 60 70 80 Figure 4.3: The Business Cycle? 200 150 100 50 0 60 40 20 0 -20 -40 -60 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Table 4.1: Selected Autocorrelations From Nelson and Plosser 1 2 r(1) r(2) d(1) d(2) Real GNP .95 .90 .34 .04 .87 .66 Nominal GNP .95 .89 .44 .08 .93 .79 Industrial Production .97 .94 .03 -.11 .84 .67 Unemployment Rate .75 .47 .09 -.29 .75 .46 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Worksheet 4.1 Consider the two random walk processes yt = yt 1 + yt zt = zt 1 + zt 10 5.0 8 2.5 6 0.0 4 2 -2.5 0 -5.0 -2 -4 -7.5 20 40 60 80 100 20 40 60 80 100 Since both series are unit-root processes with uncorrelated error terms, the regression of yt on zt is spurious. Given the realizations of {yt} and {zt}, it happens that yt tends to increase as zt tends to decrease. The regression line shown in the scatter plot of yt on zt captures this tendency. The correlation coefficient between yt and zt is  0.69 and a linear regression yields yt = 1.41  0.565zt. However, the residuals from the regression equation are nonstationary. Scatter Plot of yt Against zt Regression Residuals 10 5 4 8 3 6 2 4 1 2 0 -1 0 -2 -2 -3 -4 -4 -7.5 -5.0 -2.5 0.0 2.5 5.0 10 20 30 40 50 60 70 80 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. 90 100 Worksheet 4.2 Consider the two random walk plus drift processes yt = 0.2 + yt1 + yt zt = 0.1 + zt1 + zt 25 2.5 0.0 20 -2.5 15 -5.0 10 -7.5 5 -10.0 0 -12.5 -5 -15.0 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100 Here {yt} and {zt} series are unit-root processes with uncorrelated error terms so that the regression is spurious. Although it is the deterministic drift terms that cause the sustained increase in yt and the overall decline in zt, it appears that the two series are inversely related to each other. The residuals from the regression yt = 6.38  0.10zt are nonstationary. Scatter Plot of yt Against zt Regression Residuals 25 7.5 20 5.0 15 2.5 10 0.0 5 -2.5 0 -5.0 -5 -15.0 -7.5 -12.5 -10.0 -7.5 -5.0 -2.5 0.0 2.5 10 20 30 40 50 60 70 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. 80 90 100 Panel (a): Detrended RGDP 1.00 0.75 0.50 0.25 0.00 -0.25 -0.50 0 1 2 3 4 5 6 7 8 9 10 11 12 10 11 12 Panel (b): Logarithm ic Change in RGDP 1.00 0.50 0.00 -0.50 0 1 2 3 4 5 6 Autocorrelations 7 8 PACF Figure 4.4 ACF and PACF Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. 9 3. UNIT ROOTS AND REGRESSION RESIDUALS • yt = a0 + a1zt + et • Assumptions of the classical model: – both the {yt} and {zt} sequences be stationary – the errors have a zero mean and a finite variance. – In the presence of nonstationary variables, there might be what Granger and Newbold (1974) call a spurious regression • A spurious regression has a high R2 and t-statistics that appear to be significant, but the results are without any economic meaning. • The regression output “looks good” because the leastsquares estimates are not consistent and the customary tests of statistical inference do not hold. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Four cases • CASE 1: Both {yt} and {zt} are stationary. – the classical regression model is appropriate. • CASE 2: The {yt} and {zt} sequences are integrated of different orders. – Regression equations using such variables are meaningless • CASE 3: The nonstationary {yt} and {zt} sequences are integrated of the same order and the residual sequence contains a stochastic trend. – This is the case in which the regression is spurious. – In this case, it is often recommended that the regression equation be estimated in first differences. • CASE 4: The nonstationary {yt} and {zt} sequences are integrated of the same order and the residual sequence is stationary. – In this circumstance, {yt} and {zt} are cointegrated. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. The Dickey-Fuller tests p Δyt  yt  1    i yt  i 1   t i 2 p Δyt a0   yt  1    i yt  i 1   t i 2 p Δyt a0   yt  1  a2t    i yt  i 1   t i 2 The f1, f2, and f3 statistics are constructed in exactly the same way as ordinary F-tests: i SSR ( restricted )    SSR (unrestricted )  / r SSR(unrestricted ) /(T  k ) Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Figure 4.6: The Dickey-Fuller Distribution 0.09 0.08 percentile 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 -5 -4 -3 -2 -1 0 t-statistic Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. 1 2 3 Table 4.2: Summary of the Dickey-Fuller Tests Model yt = a0 + yt-1 + a2t + t yt = a0 + yt-1 + t yt = yt-1 + t Hypothesis Test Statistic Critical values for 95% and 99% Confidence Intervals = 0  -3.45 and -4.04  = a2 = 0 3 6.49 and 8.73 a0 =  = a2 = 0 2 4.88 and 6.50 = 0  -2.89 and -3.51 a0 =  = 0 1 4.71 and 6.70 = 0  -1.95 and -2.60 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Table 4.3: Nelson and Plosser's Tests For Unit Roots Real GNP Nominal GNP Industrial Production Unemployment Rate p a0 a2  + 1 2 0.819 0.006 -0.175 0.825 (3.03) (3.03) (-2.99) 1.06 0.006 -0.101 (2.37) (2.34) (-2.32) 0.103 0.007 -0.165 (4.32) (2.44) (-2.53) 0.513 -0.000 -0.294* (2.81) (-0.23) (-3.55) 2 6 4 0.899 0.835 0.706 p is the chosen lag length. Entries in parentheses represent the t-test for the null hypothesis that a coefficient is equal to zero. Under the null of nonstationarity, it is necessary to use the Dickey-Fuller critical values. At the .05 significance level, the critical value for the t-statistic is -3.45. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Quarterly Real U.S. GDP lrgdpt = 0.1248 + 0.0001t  0.0156lrgdpt–1 + 0.3663lrgdpt–1 (1.58) (1.31) (1.49) (6.26) The t-statistic on the coefficient for lrgdpt–1 is 1.49. Table A indicates that, with 244 usable observations, the 10% and 5% critical value of  are about 3.13 and 3.43, respectively. As such, we cannot reject the null hypothesis of a unit root. The sample value of 3 for the null hypothesis a2 = g = 0 is 2.97. As Table B indicates that the 10% critical value is 5.39, we cannot reject the joint hypothesis of a unit root and no deterministic time trend. The sample value of 2 is 20.20. Since the sample value of 2 (equal to 17.61) far exceeds the 5% critical value of 4.75, we do not want to exclude the drift term. We can conclude that the growth rate of the real GDP series acts as a random walk plus drift plus the irregular term 0.3663lrgdpt–1. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Table 4.4: Real Exchange Rate Estimation H0:  = 0 Lags Mean 0.022 (0.016) t = 1.42 0 1.05 0.059 0.194 1.88 5.47 1.16 Japan 0.047 (0.074) t = 0.64 2 1.01 0.007 0.226 2.01 10.44 2.81 Germany 0.027 (0.076) t = 0.28 2 1.11 0.014 0.858 2.04 20.68 3.71 0.031 (0.019) t = 1.59 0 1.02 0.107 0.434 2.21 .014 .004 Japan 0.030 (0.028) t = 1.04 0 0.98 0.046 0.330 1.98 .017 .005 Germany 0.016 (0.012) t = 1.23 0 1.01 0.038 0.097 1.93 .026 .004  1973-1986 Canada 1960-1971 Canada / DW Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. F SD/ SEE EXTENSIONS OF THE DICKEY–FULLER TEST yt = a0 + a1yt–1 + a2yt–2 + a3yt–3 + ... + ap–2yt–p+2 + ap–1yt–p+1 + apyt–p + et add and subtract apyt–p+1 to obtain yt = a0 + a1yt–1 + a2yt–2 + ...+ ap–2yt–p+2 + (ap–1 + ap)yt–p+1 – apDyt–p+1 + et Next, add and subtract (ap–1 + ap)yt–p+2 to obtain: yt = a0 + a1yt–1 + a2yt–2 + a3yt–3 + ... – (ap–1 + ap)Dyt–p+2 – apDyt–p+1 + et Continuing in this fashion, we obtain p  y t = a0   yt  1   i yt  i 1   t i= 2     1   p  ai  and  i   i 1  p a j j i Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Rule 1: • Consider a regression equation containing a mixture of I(1) and I(0) variables such that the residuals are white noise. If the model is such that the coefficient of interest can be written as a coefficient on zero-mean stationary variables, then asymptotically, the OLS estimator converges to a normal distribution. As such, a t-test is appropriate. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. • Rule 1 indicates that you can conduct lag length tests using ttests and/or F-tests on yt = yt–1 + 2yt–1 + 3yt–2 + … + pyt–p+1 + t Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Selection of the Lag Length • general-to-specific methodology – Start using a lag length of p*. If the t-statistic on lag p* is insignificant at some specified critical value, reestimate the regression using a lag length of p*–1. Repeat the process until the last lag is significantly different from zero. – Once a tentative lag length has been determined, diagnostic checking should be conducted. • Model Selection Criteria (AIC ,SBC) • Residual-based LM tests Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. The Test with MA Components • A(L)yt = C(L)et so that A(L)/C(L)yt = et • So that D(L)yt = et – Even though D(L) will generally be an infiniteorder polynomialwe can use the same technique as used above to form the infinite-order autoregressive model – However, unit root tests generally work poorly if the error process has a strongly negative MA component. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Example of a Negative MA term yt = yt-1 + εt – β1εt-1; 0 < β1 < 1. The ACF is: γ0 = E[(yt – y0)2] = σ2 + (1 – β1)2E[(εt-1)2 + (εt-2)2 + … + (ε1)2] = [1 + (1 – β1)2(t – 1)]σ2 γs = E[(yt – y0)(yt-s – y0)] = E[(εt +(1–β1)εt-1 + … + (1–β1)ε1)(εt-s + (1–β1)εt-s-1 + … + (1–β1)ε1) = (1 – β1) [1 + (1 – β1) (t – s – 1)] σ2 The ρi approach unity as the sample size t becomes infinitely large. For the sample sizes usually found in applied work, the autocorrelations can be small. Let β1 be close to unity so that terms containing (1 – β1)2 can be safely ignored. The ACF can be approximated by ρ1 = ρ2 = … = (1 – β1)0.5. For example, if β1 = 0.95, all of the autocorrelations should be 0.22. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Multiple Roots • Consider D2yt = a0 + b1Dyt–1 + et If b1 does differ from zero, estimate D2yt = a0 + b1Dyt–1 + b2yt–1 + et If you reject the null hypothesis, b2 = 0,conclude that {yt} is stationary. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Panel (a) yt = 0.5yt−1 + t + DL 10.0 7.5 5.0 2.5 0.0 -2.5 10 20 30 40 50 60 70 80 90 100 90 100 Panel (b) yt = yt−1 + t + DP 10.0 7.5 5.0 2.5 0.0 -2.5 -5.0 10 20 30 40 50 60 70 80 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Perron’s Test • Let the null be yt = a0 + yt–1 + m1DP + m2DL + et – where DP and DL are the pulse and level dummies • Estimate the regression (the alternative): yt = a0 + a2t +m1DP + m2DL + m3DT + et – Let DT be a trend shift dummy such that DT = t – t for t > t and zero otherwise. • Now consider a regression of the residuals ˆ t a1 y ˆ t  1   1t y If the errors do not appear to be white noise, estimate the equation in the form of an augmented Dickey–Fuller test. The t-statistic for the null hypothesis a1 = 1 can be compared to the critical values calculated by Perron (1989). For l = 0.5, Perron reports the critical value of the t-statistic at the 5 percent significance level to be –3.96 for H2 and –4.24 for H3. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Table 4.6: Retesting Nelson and Plosser's Data For Structural Change T  a0 k 1 2 a2 a1 Real GNP 62 0.33 8 3.44 (5.07) -0.189 (-4.28) -0.018 (-0.30) 0.027 (5.05) 0.282 (-5.03) Nominal GNP 62 0.33 8 5.69 (5.44) -3.60 (-4.77) 0.100 (1.09) 0.036 (5.44) 0.471 (-5.42) Industrial Prod. 111 0.66 8 0.120 (4.37) -0.298 (-4.56) -0.095 (-.095) 0.032 (5.42) 0.322 (-5.47) The appropriate t-statistics are in parenthesis. For a0, 1, 2, and a2, the null is that the coefficient is equal to zero. For a1, the null hypothesis is a1 = 1. Note that all estimated values of a1 are significantly different from unity at the 1% level. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Power • Formally, the power of a test is equal to the probability of rejecting a false null hypothesis (i.e., one minus the probability of a type II error). The power for tau-mu is a1 0.80 0.90 0.95 0.99 10% 95.9 52.1 23.4 10.5 5% 87.4 33.1 12.7 5.8 1% 51.4 9.0 2.6 1.3 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Nonlinear Unit Root Tests • Enders-Granger Test yt = It1(yt–1 – ) + (1 – It)2(yt–1 – ) + t  1 if yt  1  I t   0 if yt  1   • LSTAR and ESTAR Tests • Nonlinear Breaks—Endogenous Breaks Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Schmidt and Phillips (1992) LM Test • The overly-wide confidence intervals for  means that you are less likely to reject the null hypothesis of a unit root even when the true value of  is not zero. A number of authors have devised clever methods to improve the estimates of the intercept and trend coefficients. t yt a0  a2t    t i 1 yt = a2 + t • The idea is to estimate the trend coefficient, a2, using the regression yt = a2 + t. As such, the presence of the stochastic trend i does not interfere with the estimation of a2. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. LM Test Continued • Use this estimate to form the detrended series as ytd  yt  ( y1  aˆ2 )  aˆ2t • Then use the detrended series to estimate yt a0   y d t 1 p   ci ytd i   t i 1 • Schmidt and Phillips (1992) show that it is preferable to estimate the parameters of the trend using a model without the persistent variable yt-1. • Elliott, Rothenberg and Stock (1996) show that it is possible to further enhance the power of the test by estimating the model using something close to first-differences. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. The Elliott, Rothenberg, and Stock Test Instead of creating the first difference of yt, Elliott, Rothenberg and Stock (ERS) preselect a constant close to unity, say , and subtract yt-1 from yt to obtain: y t = a0 + a2t  a0  a2(t  1) + et, for t = 2, …, = (1  )a0 + a2[(1)t + )] + et. = a0z1t + a2z2t + et z1t = (1  ) ; z2t =  + (1)t. The important point is that the estimates a0 and a2 can be used to detrend the {yt} series d t y  y d t 1 p   ci ytd i   t i 1 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Panel Unit Root Tests pi • yit = ai0 + iyit–1  y  +a t+ i2 j 1 ij it  j + it • One way to obtain a more powerful test is to pool the estimates from a number separate series and then test the pooled value. The theory underlying the test is very simple: if you have n independent and unbiased estimates of a parameter, the mean of the estimates is also unbiased. More importantly, so long as the estimates are independent, the central limit theory suggests that the sample mean will be normally distributed around the true mean. – The difficult issue is to correct for cross equation correlation • Because the lag lengths can differ across equations, you should perform separate lag length tests for each equation. Moreover, you may choose to exclude the deterministic time trend. However, if the trend is included in one equation, it should be included in all Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Table 4.8: The Panel Unit Root Tests for Real Exchange Rates Lags Estimated i t-statistic Log of the Real Rate Estimated i t-statistic Minus the Common Time Effect Australia 5 -0.049 -1.678 -0.043 -1.434 Canada 7 -0.036 -1.896 -0.035 -1.820 France 1 -0.079 -2.999 -0.102 -3.433 Germany 1 -0.068 -2.669 -0.067 -2.669 Japan 3 -0.054 -2.277 -0.048 -2.137 Netherlands 1 -0.110 -3.473 -0.137 -3.953 U.K. 1 -0.081 -2.759 -0.069 -2.504 U.S. 1 -0.037 -1.764 -0.045 -2.008 Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Limitations • The null hypothesis for the IPS test is i = 2 = … = n = 0. Rejection of the null hypothesis means that at least one of the i’s differs from zero. • At this point, there is substantial disagreement about the asymptotic theory underlying the test. Sample size can approach infinity by increasing n for a given T, increasing T for a given n, or by simultaneously increasing n and T. – For small T and large n, the critical values are dependent on the magnitudes of the various ij. • The test requires that that the error terms be serially uncorrelated and contemporaneously uncorrelated. – You can determine the values of pi to ensure that the autocorrelations of {it} are zero. Nevertheless, the errors may be contemporaneously correlated in that Eitjt  0 – The example above illustrates a common technique to correct for correlation across equations. As in the example, you can subtract a common time effect from each observation. However, there is no assurance that this correction will completely eliminate the correlation. Moreover, it is quite possible that is nonstationary. Subtracting a nonstationary component from each sequence is clearly at odds with the notion that the variables are stationary. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. The Beveridge-Nelson Decomposition • The trend is defined to be the conditional expectation of the limiting value of the forecast function. In lay terms, the trend is the “long-term” forecast. This forecast will differ at each period t as additional realizations of {et} become available. At any period t, the stationary component of the series is the difference between yt and the trend mt. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. BN 2 • Estimate the {yt} series using the Box–Jenkins technique. – After differencing the data, an appropriately identified and estimated ARMA model will yield high-quality estimates of the coefficients. • Obtain the one-step-ahead forecast errors of Etyt+s for large s. Repeating for each value of t yields the entire set of premanent components • The irregular component is yt minus the value of the trend. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. The HP Filter Let the trend of a nonstationary series be the {mt} sequence so that yt – mt the stationary component 1 T  2 ( yt   t) +  T t=1 T T1 2 [(  )  (  )     ]  t+1 t t t -1 t= 2 For a given value of l, the goal is to select the {mt} sequence so as to minimize this sum of squares. In the minimization problem l is an arbitrary constant reflecting the “cost” or penalty of incorporating fluctuations into the trend. In applications with quarterly data, including Hodrick and Prescott (1984) l is usually set equal to 1,600. Large values of l acts to “smooth out” the trend. Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. Panel (a) The BN Cycle Pane l (b) The HP Cycle 0.03 0.04 0.03 0.02 0.02 0.01 0.01 0.00 0.00 -0.01 -0.01 -0.02 -0.02 -0.03 -0.03 -0.04 -0.04 -0.05 1960 1970 1980 1990 2000 2010 1960 1970 1980 Figure 4.11: Two Decompositions of GDP Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. 1990 2000 2010 14 RGDP 12 trllions of 2005 dollars 10 8 Consumption 6 4 Investment 2 0 1950 1960 1970 1980 1990 2000 Figure 4.12: Real GDP, Consumption and Investment Copyright © 2015 John, Wiley & Sons, Inc. All rights reserved. 2010