Co-Integration, Causality and Export-Led Growth in Portugal - Article Example

Co-integration, causality and Export-led growth in Portugal, 1835 to1985 by Les Oxley This study focuses on exploring the relationship between real GDP and real Exports by using bivariate co-integration and Granger causality. Recent researchers have widely accepted the export lead economic growth hypotheses. However author of the current study advocates co-integration has not been estimated in most of the recent empirical analyses. Alternatively simple Granger-type tests and arbitrary lag lengths are employed by these researchers for testing the hypothesis which states the export growth causes economic growth in countries. Statistical reliability of accepting Export Lead Economic Growth hypothesis in these studies is questioned by the current author. In the current study firstly, the Augmented Dicky Fuller (ADF) test is used for finding the order of integration between the two data series. Secondly, the Johansen maximum likelihood estimates are used for testing co-integration. Thirdly, standard Granger-type test is adapted by using lag residual of the co-integrating regression model. Lag length for Granger causality test is determined by minimizing the Akaike’s Final Prediction Error (FPE). The data used in this study comprise annual secondary data of GDP and Exports values in Portugal between 1835 to1985 time period. The base year has been selected as 1914 for calculating the real prices. Statistical estimates of ADF test showed that log GDP and log Exports are 1(1) while first difference of the level variables is 1(0). The trace statistics of the Johansen maximum likelihood is used to conclude that real GDP and real Exports values of Portugal are co-integrated and causally related. Four lags of the dependent variable are used in this model. Based on FPE criteria Granger causality test structure is determined as m=3, n=2, q=3 and r=4. Accordingly the Wald test statistics of the Granger causality test rejected the null hypothesis in favor of reverse causality. Thus economic growth has caused export growth in Portugal during 1835 to1985 time period. Serial Correlation As A Convenient Simplification, Not A Nuisance: A Comment On A Study Of The Demand For Money By The Bank Of England by David F. Hendry and Grayham E. Mizon The first section of the paper describes the concept of autocorrelation in relation to conventional research studies of applied economics. Autocorrelation or serial correlation is a common condition found in time series data. In OLS estimation residual is assumed to be independently distributed and does not contain any long run correlations. Thus in the presence of autocorrelation OLS estimates are not associated with minimum variance. Nevertheless autocorrelation is not considered as a fatal statistical issue in econometrics analyses. In contrary statistical software use parameterized example: moving averages and non- parameterized example: spectral methods for processing residuals with serially correlating residuals. Durbin Watson d and h statistics are useful for detecting autocorrelation in empirical data set. Computer software enables researchers conveniently processing of serially correlated residuals in single and simultaneous equation systems. Cochrane-Orcutt technique is used to estimate common root parameters conveniently. Authors of the current study suggest autocorrelation is wrongly perceived as “undesirable” and nuisance by the conventional social scientists. Although it is viewed as an error in residual processing such long run correlations can increase the accuracy of predicting and forecasting of economics data. Moreover such linkages can substantially increase the estimation efficiency of econometric coefficients without loss of consistency. It also allows testing of hypotheses which are often time considered indifferently valid by economics theory. Thus authors of the current study suggest the possibility of exploring autocorrelation and formulating them properly in econometrics models as a versatile extension in econometrics. The second section is devoted for describing Common Factor approach which is useful for exploring and modeling of autocorrelation as adapted by Sargan (1975). By using mathematical models the convenience of having a common root in first order autoregressive residual is described. Accordingly when a common root is present the number of parameters needed to be estimated reduce from four (conventional processing of first order autoregressive) to three (common factor approach). Increased estimation efficiency of the proposed approach is thus attributed to the reduced number of parameters. Furthermore the proposed approach allow redundant dynamics in the presence of spurious fit of moving average models and seasonal data i.e. error process with negative unity. The third section presents the application of COMFAC algorithm to the money demand function adopted by Hacche (1974). Accordingly authors have provided evidence of feasibility of the proposed computer software i.e. COMFAC developed by Sargan and Sylwestrowicz (1976), in application to empirical analyses of autoregressive residual. The five statistical issues associated with Hacche, 1947 study are stationarity and differencing, dynamic specification, serial correlation and seasonal adjustment of the data. In this study seasonal adjustment of interest rates and money stocks does not incorporate common filters thus seasonal bias problem occurs. COFAC approach is used for overcoming these issues. Authors conclude by stating serial correlation as perfectly feasible to test and if properly formulated such models are superior to the conventional time series descriptions which ignores the long-run correlations present in the empirical data. An Empirical Analysis of Alternative Parametric ARCH Models by Geoffrey F. Loudon, A Wing H. Wattb And Pradeep K. Yadav This study describes the usefulness of ARCH models for forecasting the systemic conditional volatility of stock market returns. Authors advocate the occurring of sequential variations in stock market data such as large price changes following by large price changes and small price changes by small price changes and the capability of ARCH models in predicting the conditional variance component present in stock market data. Modeling conditional variance is important for portfolio investment decisions and pricing of assets. Advocates of systematic volatility describe so called conditional variance at time t as a function of time, exogenous variables and lag endogenous variables example: firm specific and economy wide indicators, residual at time t-1and vector of estimated parameters. New parameters included in rather recent ARCH and GRACH models are useful for ex post descriptions such as return-generating process as opposed to the linear models. Current authors indicate the ex ante usefulness of these models as the most important feature for portfolio selection and asset pricing decisions by the stock marketers. Out-of-sample predictive ability, outperformance consistency across different sub-periods and incorporating market micro-structural factors are listed as the determinants of ARCH and GRACH models ex ante usefulness. The first section of the report consists of the relative effectiveness of commonly adopted following models in the literature; (1) Linear GARCH model, (2) Multiplicative GARCH model, (3) Exponential GARCH model, (4) Glosten, Jagganathan and Runkle GARCH model (5) Non-linear asymmetric GARCH model, (6) VGARCH model (7) TS-GARCH model and (8) Threshold GARCH model. Reported negative correlation between current returns and future volatility and incorporating asymmetry in the impact of news on volatility are the notable issues associated with linear GRACH (1, 1) model. Thus alternative versions as listed above are commonly adopted by the researchers. The second section of the paper consists of evaluating the ex ante usefulness of these models for portfolio selection and asset pricing decisions. Out-of-sample predictive ability is used as the evaluation criteria. Time series secondary data is used for the empirical analysis. The studied 27 years period (January 1971 to October 1997) is divided into three sub periods as January 1971 to December 1980, January 1981 to December 1990 and January 1991 to October 1997. The daily values of FT All Share Index of the London Stock Exchange obtained from and Datastream and their log values are employed in the analysis. To evaluate the out-of-sample predictive ability an earlier sub-period is used for modeling the volatility of a given period. Accordingly parameters estimated from 1971 to 1980 data is used for predicting the conditional volatility in 1981 to 1990 time period while the parameters estimated from 1981 to 1990 data is used for predicting the conditional volatility in 1991 to 1997 period. Maximum likelihood method adopted by algorithm of Berndt is used for analyzing the data. Then to evaluate the performances of the eight types of ARCH and GRACH models listed above the following tests are adopted; Root mean squared error in conditional variance forecasts, equality of the conditional variance forecast error, Skewness and kurtosis in standardized residuals, LM tests for serial correlation up to ten lags in the levels of the standardized residuals, LM tests for serial correlation up to ten lags in the squares of the standardized residuals, news impact curve tests of the standardized residuals, consistency tests of the squared residuals (PS) and consistency tests of the squared residuals using the logarithmic form. Results show that conditional variance is statistically significantly related to its previous level and to past surprises in returns. Asymmetric behavior is tested by incorporating linearly filtered raw returns (residuals of previously estimated models) as an explanatory variable. This was found statistically significant in all the models which permit asymmetry across the three time periods. Furthermore sign and significance of all ARCH parameters are found consistent across the three time periods while the magnitude of the parameter estimates varied in each model. The likelihood ratio for MGARCH model is found more negative compared to all the models. Alternatively the maximum log-likelihood value of asymmetry model is found inconsistence inter-temporally. TGARCH and NGARCH models incorporate the best, maximum likelihood ratio in first two and final time periods respectively. VGARCH model is associated with the worst likelihood across the time periods. Work Cited Hacche, G 1974. The Demand for Money in the United Kingdom: Experience since 1971. Bank of England QuarterlyBulletin. Vol. 14, no. 3, pp. 284-305. Hendry, DF and Mizon, GE 1978. Serial Correlation As A Convenient Simplification, Not A Nuisance: A Comment On A Study Of The Demand For Money By The Bank Of England. The Economic Journal. Vol. 88, no. 351, pp. 549-563 Loudon, GF, Wing AH, Wattb and Yadav PK 2000. An Empirical Analysis of Alternative Parametric ARCH Models. Journal of Applied Econometrics. Vol. 15, pp. 117-136 Oxley L 1993. Co-integration, causality and Export-led growth in Portugal, 1835 to1985. Economics Letters. Vol.4, pp. 163-166 Sargan, J. D. and Sylwestrowicz, J. D. (I 976)." COMFAC: Algorithm for Wald Tests of Common Factors in Lag Polynomials." Users Manual, London School of Economics Sargan, JD I975. A Suggested Technique for Computing Approximations to Wald Criteria with Appli-cation to Testing Dynamic Specfications. Discussion Paper A2, London School of Economics Read More