Transcript VAR
VAR Małgorzata Bednarek Maria Derezińska Magdalena Sadowska Theory About VAR Sims critique Validity of interrelated analytical exercises (common in reality where everything depends on everything) implicit incorrect analysis of applications if econometric inquiry is dependent on prior theoretical restrictions Vector autoregression (VAR model) is possible to deal with dynamic relationships between macroeconomic variables, where causality may be mutual Vector autoregression xt =Π1xt -1+...+Πkxt -k+ΦDt +εt εt ~NID(0, Ω) Applications of vector autoregressions Forecasting VAR models allow complete flexibility in specifying the correlation between future, present and past Causality tests Granger tests Sims tests Both tests are implications of the same null hypothesis. In the model joint significance of all lags except the lags of variable supposed to be a cause is tested Hypothesis-seeking Data characterization Impulse response analysis Monetary and fiscal policy analysis VAR diagnosis • Testing for number of lags • Testing for VAR stability • Testing for Granger causality • Testing for autocorrelation Empirical results Model In our model we analyze the potential for lending and consumption booms in Hungary DATA SOURCES: OECD Maxdata database, extension of papers: Brzoza-Brzezina (2004) and Susan Schadler, Zuzana Murgasova, Rachel van Elkan (2004 ) DATA: quarterly dataset for Hungary for period 1996-2004 1. lloan - logarithm of total nominal loans to the private sector 2. lrate - logarithm of Nominal interest rate 3. lgdp - logarithm of GDP at constant prices 4. lcon - logarithm of Private Consumption, Volume Vector Autoregression for Hungary with consumption (quarterly dataset October 1995 - May 2004) Sample: 1996q2 2004q2 -------------------------------------------------------------------------Equation Obs Parms RMSE R-sq chi2 P -------------------------------------------------------------------------lloan 23 12 .001624 0.9960 5721.318 0.0000 lrate 23 12 .742514 0.4584 19.46434 0.0533 lgdp 23 12 .000417 0.9980 11602.81 0.0000 lcon 23 12 .004917 0.9998 96633.75 0.0000 -------------------------------------------------------------------------Model lag order selection statistics -----------------------------------FPE AIC HQIC 1.362e-17 -27.9393 -27.34332 SBIC -25.569572 LL 369.30194 Det(Sigma_ml) 1.329e-19 -----------------------------------------------------------------------------| Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------lloan | lloan | L1 | .6797655 .1784921 3.81 0.000 .3299274 1.029604 L2 | .1464823 .2067421 0.71 0.479 -.2587247 .5516893 lrate | L1 | -.0000959 .0007495 -0.13 0.898 -.001565 .0013732 L2 | -.0012674 .0007255 -1.75 0.081 -.0026895 .0001546 lgdp | L1 | -.1985731 .7957318 -0.25 0.803 -1.758179 1.361033 L2 | .2478325 .688455 0.36 0.719 -1.101514 1.597179 lcon | L1 | .0266011 .0772088 0.34 0.730 -.1247254 .1779276 L2 | -.0134382 .0670136 -0.20 0.841 -.1447824 .117906 _q_2 | -.0015721 .001348 -1.17 0.243 -.0042141 .0010698 _q_3 | -.0016634 .00163 -1.02 0.307 -.0048582 .0015313 _q_4 | -.001651 .0007053 -2.34 0.019 -.0030334 -.0002686 _cons | .269556 1.732104 0.16 0.876 -3.125306 3.664418 -------------+---------------------------------------------------------------lrate | lloan | L1 | -6.605417 81.60332 -0.08 0.935 -166.545 153.3342 L2 | 115.7716 94.51868 1.22 0.221 -69.48158 301.0249 lrate | L1 | .3951908 .3426723 1.15 0.249 -.2764345 1.066816 L2 | -.1638084 .3317012 -0.49 0.621 -.8139309 .486314 lgdp | L1 | -145.4654 363.794 -0.40 0.689 -858.4885 567.5577 L2 | -455.4286 314.749 -1.45 0.148 -1072.325 161.4681 lcon | L1 | -28.46556 35.29845 -0.81 0.420 -97.64925 40.71812 L2 | 36.38052 30.63737 1.19 0.235 -23.66762 96.42867 _q_2 | .503419 .6162614 0.82 0.414 -.7044312 1.711269 _q_3 | .5327072 .7452087 0.71 0.475 -.927875 1.993289 _q_4 | -.472785 .3224592 -1.47 0.143 -1.104793 .1592233 _cons | 1293.563 791.8863 1.63 0.102 -258.5061 2845.631 var lloan lrate lgdp lcon, exog(_q*) Parameters of VAR model in standard form have no structural interpretations !! lgdp lloan | | L1 | .076064 .0458807 1.66 0.097 -.0138604 .1659885 L2 | -.0677189 .0531422 -1.27 0.203 -.1718757 .0364379 lrate | L1 | .0000754 .0001927 0.39 0.696 -.0003022 .000453 L2 | -.0001645 .0001865 -0.88 0.378 -.00053 .0002011 lgdp | L1 | .9302852 .2045396 4.55 0.000 .529395 1.331175 L2 | -.160274 .1769645 -0.91 0.365 -.5071181 .1865701 lcon | L1 | .0047614 .0198462 0.24 0.810 -.0341365 .0436592 L2 | .000569 .0172256 0.03 0.974 -.0331925 .0343305 _q_2 | .0001349 .0003465 0.39 0.697 -.0005442 .000814 _q_3 | .0000478 .000419 0.11 0.909 -.0007734 .000869 _q_4 | .0000322 .0001813 0.18 0.859 -.0003231 .0003876 _cons | .5750082 .4452303 1.29 0.197 -.297627 1.447643 -------------+---------------------------------------------------------------lcon | lloan | L1 | -1.00662 .5403622 -1.86 0.062 -2.065711 .0524706 L2 | 1.114147 .6258854 1.78 0.075 -.1125662 2.34086 lrate | L1 | -.0035271 .0022691 -1.55 0.120 -.0079745 .0009203 L2 | -.0021112 .0021965 -0.96 0.336 -.0064162 .0021938 lgdp | L1 | 2.890635 2.408977 1.20 0.230 -1.830874 7.612143 L2 | -4.470218 2.08421 -2.14 0.032 -8.555195 -.3852413 lcon | L1 | 1.085324 .2337399 4.64 0.000 .6272024 1.543446 L2 | -.0913518 .2028751 -0.45 0.653 -.4889797 .306276 _q_2 | .0017963 .0040808 0.44 0.660 -.0062019 .0097944 _q_3 | -.0179587 .0049346 -3.64 0.000 -.0276304 -.008287 _q_4 | -.0144019 .0021353 -6.74 0.000 -.0185869 -.0102169 _cons | 4.021655 5.243726 0.77 0.443 -6.255859 14.29917 Table 1: Testing for number of lags varsoc lloan lrate lgdp lcon, maxlag(3) Selection order criteria endogenous variables: lloan lrate lgdp lcon constant included in models Sample: 1996q3 Obs = 20 2004q2, with gaps ------------------------------------------------------------------------------lag LL LR df p FPE AIC HQIC SBIC ------------------------------------------------------------------------------0 182.019 . . . 2.18e-13 -17.8019 -17.763 -17.6028 1 287.577 211.116 16 0.000 2.94e-17 -26.7577 -26.5633 -25.762 2 303.730 32.305 16 0.009 3.66e-17 -26.773 -26.4231 -24.9806 3 362.188 116.917* 16 0.000 1.08e-18* -31.0188* -30.5134* -28.4299* Source: OECD maxdata 2004 Table 2: Testing for VAR stability varstable Eigenvalue stability condition ---------------------------------------------Eigenvalue Modulus ----------------------------------------------.3669971 + .44064736 | .57346052 -.3669971 - .44064736 | .57346052 .3679289 + .43235172 | .56771447 .3679289 - .43235172 | .56771447 .8478558 + .14019235 | .85936801 .8478558 - .14019235 | .85936801 .9212941 | .92129405 .4716963 | .47169626 Source: OECD maxdata 2004 All the eigenvalues lie inside the unit circle, which means that VAR satisfies stability condition. Table 3: Testing for autocorrelation vargranger H0: no autocorrelation at lag order j ------------------------------------j chi2 df p ------------------------------------1 26.0522 16 0.05330 2 9.4846 16 0.03212 Source: OECD maxdata 2004 In this test a 0 hypothesis tells that there is no autocorrelation at lag of order j. We test it at the 5% level so the p-values bigger than 0,05 means that there is no autocorrelation in the model. Table 4: Testing for Granger causality varlmar Granger causality Wald tests ---------------------------------------------------------------------------Equation Excluded chi2 df Prob > chi2 ---------------------------------------------------------------------------lloan lrate 7.1690 2 0.0278 lloan lgdp 0.1414 2 0.9318 lloan lcon 0.8457 2 0.6552 lloan ALL 19.0689 6 0.0040 ---------------------------------------------------------------------------lrate lloan 2.7777 2 0.2494 lrate lgdp 3.3249 2 0.1897 lrate lcon 6.3980 2 0.0408 lrate ALL 6.8007 6 0.3397 ---------------------------------------------------------------------------lgdp lloan 2.7658 2 0.2509 lgdp lrate 0.9085 2 0.6349 lgdp lcon 2.7804 2 0.2490 lgdp ALL 24.7132 6 0.0004 ---------------------------------------------------------------------------lcon lloan 3.8855 2 0.1433 lcon lrate 11.6411 2 0.0030 lcon lgdp 4.7092 2 0.0949 lcon ALL 14.4984 6 0.0245 Source: OECD maxdata 2004 Graph 1: The temporal change of consumption in Hungary (impulse- lcon, response- lloan) varirf graph irf, i(lloan) r(lcon) irf, lcon, lloan .4 .2 0 -.2 0 2 4 6 step 95% CI Graphs by irfname, impulse variable, and response variable Source: OECD maxdata 2004 irf 8 Graph 2: The permanent change of consumption in Hungary (impulse- lcon, response- lloan) varirf graph oirf, i(lloan) r(lcon) irf, lcon, lloan .001 .0005 0 -.0005 0 2 4 6 step 95% CI Graphs by irfname, impulse variable, and response variable Source: OECD maxdata 2004 orthogonalized irf 8 Graph 4: Trend line of interest rate since 1995q4 and prediction for period (2004q1 till 2008q2) 0 19 95 19 q4 97 19 q1 98 19 q2 99 20 q3 00 20 q4 02 20 q1 03 20 q2 04 20 q3 05 20 q4 07 20 q1 08 q2 -1 -2 -3 -4 -5 -6 -7 -8 Source: OECD maxdata 2004 lrate Graph 5: Trend line of a number of loans since 1995q4 and prediction for period (2004q1 till 2008q2) 2,73 2,72 2,71 2,7 2,69 2,68 2,67 2,66 2,65 2,64 2,63 Source: OECD maxdata 2004 2008q2 2007q1 2005q4 2004q3 2003q2 2002q1 2000q4 1999q3 1998q2 1997q1 1995q4 lloan Graph 6: Trend line of GDP since 1995q4 and prediction for period (2004q1 till 2008q2) 2,72 2,715 2,71 2,705 2,7 2,695 2,69 2,685 2,68 Source: OECD maxdata 2004 2008q2 2007q1 2005q4 2004q3 2003q2 2002q1 2000q4 1999q3 1998q2 1997q1 1995q4 lgdp Graph 7: Trend line of consumption since 1995q4 and prediction for period (2004q1 till 2008q2) 6 5 4 3 2 1 19 95 19 q4 97 19 q1 98 19 q2 99 20 q3 00 20 q4 02 20 q1 03 20 q2 04 20 q3 05 20 q4 07 20 q1 08 q2 0 Source: OECD maxdata 2004 lcon Graph 8: Predicted trend lines for consumption and loans for period 2004q1 till 2008q2 6 5 4 3 LOANS CONS 2 1 20 04 q 20 1 04 q 20 3 05 q 20 1 05 q 20 3 06 q 20 1 06 q 20 3 07 q 20 1 07 q 20 3 08 q1 0 Source: OECD maxdata 2004 Both trend lines seem to be very stable. While consumption is slightly increasing the predicted amount of loans stays fairly constant and with hardly noticeable the tendency to decrease. Table 5: Comparison between data on consumption and prediction (quarterly dataset January 2004 - October 2004) January 04 April 04 July 04 October 04 OECD data 5,00739 5,01837 5,02886 5,03924 Prediction 5,00739 5,01784 5,01058 5,00267 Standard Error 0 0,000106 0,003635 0,007257 Source: OECD maxdata 2004 Arguments for no lending boom There is no empirical evidence for the near future threat according a possible lending boom in Hungary. Undoubtedly , after the EURO adoption, the interest rate will have to accommodate to the one in EMU but some adjustments have already been accomplished. There is smaller deviation from the interest rate of EMU Hungary still have less developed financial market than the old member state Forecast is based on the internal outcome of the past events in Hungary. Awareness of an external shock or unexpected twist in domestic economy is the key issue for ruling elites. The end Now is time for applause!