ARDL Approach to Cointegration

  • 格式:pdf
  • 大小:180.38 KB
  • 文档页数:6

The ARDL procedure involves 2 stages:1.Testing the existence of the long-run relation between the variables by computing theF-statistic for testing the significance of the lagged levels of the variables in the error correction form of the underlying ARDL model. The asymptotic distribution of this F-statistic is non-standard, irrespective of whether the regressors are I(0) or I(1).-Pesaran, Shin, and Smith (1996) have tabulated the appropriate critical values for different numbers of regressors (k), and determined whether the ARDL modelcontains an intercept and/or trend. They give two sets of critical values: one setassuming that all the variables in the ARDL model are I(1), and another computedassuming all the variables are I(0). For each application, this provides a bandcovering all the possible classifications of the variables into I(0) and I(1), or evenfractionally integrated ones.•If the computed F-statistic falls outside this band a conclusive decision can bemade without needing to know whether the underlying variables are I(0) or I(1), orfractionally integrated.•If the computed statistic falls within the critical value band the result of theinference is inconclusive and depends on whether the underlying variables areI(0) or I(1). It is at this stage in the analysis that the investigator may have to carryout unit roots tests on the variables.2.Estimating the coefficients of the long-run relations and making inferences abouttheir values using the ARDL option. Note that it is only appropriate to embark on this stage if you are satisfied that the long-run relationship between the variables to be estimated is not in fact spurious.STEPS for ARDL:1.Run Unit root tests (ADF, PP, KPSS) to make sure that the variables are integrated either at I(0) orI(1) to successfully run the ARDL technique, but not I(2).1.1 Turn ON Simulation of Critical Values for Unit Root Tests,1.2 Run unit root tests:A. ADF test (Null: Nonstationarity):ADF LDM(5); ADF LDP(5); ADF LES(5); ADF LFA(5);ADF LKI(5); ADF LHG(5); ADF LFE(5); ADF LDA(5);ADF DDM(5); ADF DDP(5); ADF DES(5); ADF DFA(5);ADF DKI(5); ADF DHG(5); ADF DFE(5); ADF DDA(5);B. Phillips-Perron test (Null: Nonstationarity):DF_PP LDM; DF_PP LDP; DF_PP LES; DF_PP LFA;DF_PP LKI; DF_PP LHG; DF_PP LFE; DF_PP LDA;DF_PP DDM; DF_PP DDP; DF_PP DES; DF_PP DFA;DF_PP DKI; DF_PP DHG; DF_PP DFE; DF_PP DDA;C. KPSS test (Null: Stationarity):KPSS LDM; KPSS LDP; KPSS LES; KPSS LFA;KPSS LKI; KPSS LHG; KPSS LFE; KPSS LDA;KPSS DDM; KPSS DDP; KPSS DES; KPSS DFA;KPSS DKI; KPSS DHG; KPSS DFE; KPSS DDA;2.Test for long-run relationship between the variables:2.1 Go to Univariate -> Linear Regression Menu -> Ordinary Least Squares2.2 Type in the command area:DDA INPT DDA{1-4} DDM{1-4} DDP{1-4} DES{1-4} DFA{1-4} DFE{1-4} DHG{1-4} DKI{1-4}2.3 Run and close the output window since we are not interested in these results.2.4 In the Post Regression Menu, choose Option 2 (Move to Hypothesis Testing Menu). OK.2.5 In the Hypothesis Testing Menu, select Option 6 (Variable Addition Test). OK.(NOTE: This procedure is used as an alternative to Wald test).2.6 In the Command Area, type and run the following:LDA(-1) LDM(-1) LDP(-1) LES(-1) LFA(-1) LHG(-1) LKI(-1)2.7 The F-statistic for testing the joint null hypothesis that the coefficients of these level variables are zero (namely there exists no long-run relationship between them) is given in the last row of the result table.Compare the F-statistic from the output with the values from F Table of Pesaran below (B.1). If it is lesser than the lower bound, then we cannot reject the null of no long-run relationship among the variables. If greater than the upper bound, then we reject the null hypothesis. However, if the value falls in between the lower and upper bounds, then the result is inconclusive.2.8 Repeat the analysis replacing the dependent variable as:DDM INPT DDA{1-4} DDM{1-4} DDP{1-4} DES{1-4} DFA{1-4} DFE{1-4} DHG{1-4} DKI{1-4} DDP INPT DDA{1-4} DDM{1-4} DDP{1-4} DES{1-4} DFA{1-4} DFE{1-4} DHG{1-4} DKI{1-4} DES INPT DDA{1-4} DDM{1-4} DDP{1-4} DES{1-4} DFA{1-4} DFE{1-4} DHG{1-4} DKI{1-4} DFA INPT DDA{1-4} DDM{1-4} DDP{1-4} DES{1-4} DFA{1-4} DFE{1-4} DHG{1-4} DKI{1-4}DFE INPT DDA{1-4} DDM{1-4} DDP{1-4} DES{1-4} DFA{1-4} DFE{1-4} DHG{1-4} DKI{1-4}DHG INPT DDA{1-4} DDM{1-4} DDP{1-4} DES{1-4} DFA{1-4} DFE{1-4} DHG{1-4} DKI{1-4} DKI INPT DDA{1-4} DDM{1-4} DDP{1-4} DES{1-4} DFA{1-4} DFE{1-4} DHG{1-4} DKI{1-4}Similarly, run addition tests with each of above equation (same as step 2.6):LDA(-1) LDM(-1) LDP(-1) LES(-1) LFA(-1) LHG(-1) LKI(-1)Here are the F-statistics for each equation:1. F (LDA | LDM, LDP, LES, LFA, LFE, LHG, LKI) = 4.522. F (LDM | LDA, LDP, LES, LFA, LFE, LHG, LKI) = 6.123. F (LDP | LDA, LDM, LES, LFA, LFE, LHG, LKI) = 5.634. F (LES | LDA, LDM, LDP, LFA, LFE, LHG, LKI) = 4.165. F (LFA | LDA, LDM, LDP, LES, LFE, LHG, LKI) = 3.716. F (LFE | LDA, LDM, LDP, LES, LFA, LHG, LKI) = 5.377. F (LHG | LDA, LDM, LDP, LES, LFA, LFE, LKI) = 4.318. F (LKI | LDA, LDM, LDP, LES, LFA, LFE, LHG) = 4.04 All these statistics fall well above the upper bound of the critical value band (which is 2.604 - 3.746), and hence we reject the null hypothesis that the level variables do not enter significantly into the equations, except for F(LFA | LDA, LDM, LDP, LES, LFE, LHG, LKI) = 3.71, which falls within the bounds, hence inconclusive.2.9 Close -> Cancel -> Cancel. Done.3. ARDL Test:3.1 Go to Univariate -> ARDL Approach to Cointegration and define order of lags as 4 (weeks, 1 month) and type in the Command Area:LDA LDM LDP LES LFA LHG LKI & INPTRun. Click OK when the message box pops up.3.2 First, in ARDL Order Selection Menu, select Option 3 (Akaike Information Criterion). OK.3.3 Next, in ARDL Model Selection (AIC) Menu, select Option 1 (Display the Estimates of the Selected ARDL Regression). OK.[Here you can find information about the lower and upper bounds for testing the level relationship among the variables, as well as diagnostics for autocorrelation, heteroscedasticity, normality, and functional form below the table.]Close -> Cancel on the next window to arrive at ARDL Model Selection Menu.3.4 Select Option 2 (Display Long-Run Coefficients and their Asymptotic S.E.). OK.Close -> Cancel on the next window to arrive at ARDL Model Selection Menu.3.5 Select Option 3 (Display Error Correction Model). OK.Here check for:A. Significance of ECM(-1) [p-value],B. Coefficient of ECM(-1). If it is close to zero (in absolute terms), then there is slow speed of convergence to equilibrium, otherwise - moderate or fast speed.Notes:•If the value of speed of adjustment (coefficient of ECM[-1] ) is zero it means that there exist nolong-run relationships.•If it falls between -1 and 0, there exists partial adjustment.•A value smaller than -1 indicates that the model over adjusts in the current period;•A positive value implies that the system moves away from equilibrium in the long-run.3.6 To run SBC test, follow the same steps as explained above starting from 3.2.In ARDL Order Selection Menu, select Option 4(SBC) -> OK, and so on.4. Impulse Response analysis4.1 Go to Multivariate -> Unrestricted VAR4.2 Define number of lags -> 64.3 Type in the Command Area and run:DDA DDM DDP DES DFA DFE DHG DKI & INPT4.4 Select Option 4 (Hypotheses testing and lag order selection in the VAR) from UnrestrictedVAR Post Estimation Menu -> OK4.5 Select Option 1 (Testing and selection criteria for order (lag length) of the VAR) from VARHypotheses Testing Menu -> OK[NOTE: Identify the VAR order based on highest values of AIC and SBC. For example, 4].Close -> Cancel.4.6 Define number of lags -> 44.7 Select Option 3 (Impulse Response and Forecast Error Variance Decomposition) fromUnrestricted VAR Post Estimation Menu -> OK4.8 For Impulse-Response: Select Option 1 (Orthogonalized IR of variables to shocks inequations) or Option 2 (Generalized IR of variables to shocks in equations) from Unrestricted VAR Dynamic Response Analysis Menu -> OK.4.9 For VDC: Select Option 3 (Orthogonalized forecast error variance decomposition) orOption 4 (Generalized forecast error variance decomposition) from Unrestricted VAR Dynamic Response Analysis Menu - OK.DONE!Alhamdulillah。