Abstract
This paper describes early contributions of Staff Papers to international economics. The paper highlights that Staff Papers has, since its inception in 1950, been an important vehicle for the dissemination of research done by the IMF staff. This paper discusses three areas in which articles published in Staff Papers up until the 1970s made major contributions to the literature in international economics. The areas covered are: the absorption approach and the monetary theory of the balance of payments; the Mundell-Fleming model; and foreign trade modeling.
IN AN INTERESTING ARTICLE in Staff Papers, Edward Gardner and William R.M. Perraudin (1993)—henceforth GP—examined whether the shock associated with German unification had any impact on the dominant position that Germany may hold within the European exchange rate mechanism (ERM). The authors tackled the question by running vector autoregressions (VARs) on daily changes in the one-month French, German, and American onshore and offshore interest rates and by looking at the derived impulse-response functions and long-run multipliers. This corresponds to an analysis in terms of the Sims (1980) type of causal linkages.
In both the preunification (October 1987-December 1989) and postunification periods (January 1990-August 1992), GP rejected asymmetry, defined as unidirectional causality, since “[t]he effect of French innovations on Germany is significant, albeit smaller than the German effect on France.” Moreover, the examination of their year-by-year estimation results suggests that Germany handed over its predominant position1 to France during 1990, never completely regaining it thereafter. This structural break is also corroborated by Lagrange multiplier tests, which indicate that a big change did occur at the end of 1989.
In this note we attempt to show while there has indeed been a structural break, it appears that the major asymmetric shock that happened with German unification has not weakened Germany’s position in the ERM. On the contrary, since German unification the ERM has moved toward an even more asymmetric functioning, namely, German dominance. In addition, we believe that our results are also more in line with the recent European monetary system (EMS) crises. The fact that Germany’s leadership role in the ERM has not eroded with German unification, in conjunction with the asymmetric shock of the latter, is definitely a central point in explaining the so-called “collapse” of the European monetary system in July 1993.
Using both Granger (1969) and Sims (1980) causality concepts, we confirm that German unification caused a structural break in a system consisting of three interest rates. We use data similar to those of GP, that is, daily one-month offshore rates for the French franc, the U.S. dollar, and the deutsche mark, but we cover a somewhat extended sample, between April 1983 and December 1993.2 Contrary to GP, we derive this result by considering the levels of interest rates in a cointegrated VAR, which we believe statistically and economically to be more suitable than a model in first differences. It follows from the Granger representation theorem that, when the variables are nonstationary and cointegrated, an unrestricted VAR in first differences is misspecified because the lagged equilibrium errors are not accounted for. In other words, we allow for cointegration relations when analyzing Granger and Sims causalities. This model leaves room for mean-reversion properties in the system, and at the same time allows an analysis in terms of common trends. In addition, we not only compute the long-run impulse response function but also test for zero restrictions. We thus perform the “neutrality” analysis in a cointegrated VAR (see Stock and Watson (1989) in the bivariate case and its extension to the trivariate case in Bruneau and Nicolaï (1992)). Put differently, we test whether a persistent shock, for instance on the level of the German rate, has a long-lasting impact on the level of the French rate.
I. The Changes in Long-Run Relations
Following Hansen and Johansen (1992), we first perform a recursive analysis in a multivariate setting to test for a structural break in the long-run linkages, namely, the number of cointegration relations. We run a backward recursive estimation of Johansen’s (1988 and 1991) Trace and λmax statistics.3 Figure 1 shows that the normalized Trace4 test for the null of one cointegration relation drops sharply when observations prior to the third quarter of 1989 are included in the sample, thus bearing out GP’s result. Nevertheless, the nature of our structural break differs from the one found by GP as it is equivalent to a shift in the system’s degrees of freedom. There is only one cointegration relation before the structural break, whereas thereafter we find two long-run relations.5 One would expect that this also has an impact on the causal linkages at work.
Normalized Trace
Citation: IMF Staff Papers 1995, 004; 10.5089/9781451930900.024.A008
Notes: Let r* stand for the number of coiniegralion relations. The normalized Trace is defined as the Trace statistic divided by its critical value. The dates along the x-axis are the starling dales for regressions ending in December 1992.Normalized Trace
Citation: IMF Staff Papers 1995, 004; 10.5089/9781451930900.024.A008
Notes: Let r* stand for the number of coiniegralion relations. The normalized Trace is defined as the Trace statistic divided by its critical value. The dates along the x-axis are the starling dales for regressions ending in December 1992.Normalized Trace
Citation: IMF Staff Papers 1995, 004; 10.5089/9781451930900.024.A008
Notes: Let r* stand for the number of coiniegralion relations. The normalized Trace is defined as the Trace statistic divided by its critical value. The dates along the x-axis are the starling dales for regressions ending in December 1992.In light of the previous findings, we split the sample into two sub-periods: April 4, 1983 to November 29, 1990, and December 3, 1990 to December 22, 1993—the break point corresponding to the most obvious shock to the German rate, that is, a leap occurring in late November 1990. Regarding the second subperiod, the two stationary components of the system can be identified: they are the U.S. rate and the French-German interest rate differential. See Tables 1 and 2 for details. Indeed the p-value of the likelihood ratio test statistic for this joint overidentifying restriction is only 11 percent, but much higher values are found when both restrictions are tested separately. Both the stationarity of the U.S. rate in a univariate framework and the stationarity of the French-German interest rate differential in a bivariate model are easily accepted. Given the straightforward interpretation of these two restrictions, we consider them relevant to our analysis.
Johansen Teste /or Cointegration in the Gaussian VAR
Johansen Teste /or Cointegration in the Gaussian VAR
| Statistics and H0 | p = 0 | p = 1 | p = 2 |
|---|---|---|---|
| λmax (r* = p) | 29.0*** | 17.7** | 4.4 |
| Trace (r* ≤ p) | 51.1*** | 22.1** | 4.4 |
Johansen Teste /or Cointegration in the Gaussian VAR
| Statistics and H0 | p = 0 | p = 1 | p = 2 |
|---|---|---|---|
| λmax (r* = p) | 29.0*** | 17.7** | 4.4 |
| Trace (r* ≤ p) | 51.1*** | 22.1** | 4.4 |
Estimation Results with an Identified Normalized Long-Run Model
Estimation Results with an Identified Normalized Long-Run Model
| ECM coefficients (Percent) | |||
|---|---|---|---|
| Variable | α1 | α2 | Cointegration vectors for (rFRF, rDEM, rUSD, constant) |
| rFRF | 5.1 | 0.39 | β1 = (l -1 0 0.55) |
| rDEM | 0.17 | 0.15 | β2 = (0 0 1 -3.77) |
| rUSD | 0.33 | -1.2 | |
Estimation Results with an Identified Normalized Long-Run Model
| ECM coefficients (Percent) | |||
|---|---|---|---|
| Variable | α1 | α2 | Cointegration vectors for (rFRF, rDEM, rUSD, constant) |
| rFRF | 5.1 | 0.39 | β1 = (l -1 0 0.55) |
| rDEM | 0.17 | 0.15 | β2 = (0 0 1 -3.77) |
| rUSD | 0.33 | -1.2 | |
II. The Changes in Long-Run Causal Linkages
We now check whether the changes in the system’s long-run linkages coincide with some parallel modification in its causal linkages. The causal linkages can be examined either exclusively—through the error-correction terms in the cointegrated VAR or through the long-run multipliers in the moving average representation—or together with the effects of changes in the variables of interest (as in Granger (1988), or Toda and Phillips (1993)). Here, we focus on the former approach, and thus on the long-run connections between the levels of Eurorates (see Henry and Weidmann (1994a) for further analyses including the changes in the interest rates). The first significant finding relates to the weak exogeneity properties, that is, the long-run component of the Granger noncausality. According to Johansen’s likelihood ratio tests, the U.S. rate is weakly exogenous prior to German unification, but the German rate becomes the sole weakly exogenous variable in the system thereafter (see Table 3 for the results of the weak exogeneity and neutrality tests). The current change in the German rate depends neither on the past level of the U.S. rate nor on the lagged French-German interest rate differential, indicating that from 1990 on German monetary policy has become more independent. This holds for the transitory dynamics as well, since the changes in the German rate are not driven by previous changes in either of the other two rates. (The p-value for the joint hypothesis that the other two rates do not Granger-cause the German one is 35 percent using the Mosconi and Giannini (1992) procedure.)
Weak Exogeneity and Neutrality Tests
Weak Exogeneity and Neutrality Tests
| Weak exogeneity tests | Neutrality tests vis-à-vis rFRF | |||||
|---|---|---|---|---|---|---|
| 1983-90 | 1991-93 | |||||
| Variables | 1983-90 | 1991-93 | caused by | causing | caused by | causing |
| rDEM | 1.5 | 64 | 7 | 0 | 6 8 | 0.9 |
| rUSD | 52 | 0 | 49 | 0.1 | irrelevant | irrelevant |
Weak Exogeneity and Neutrality Tests
| Weak exogeneity tests | Neutrality tests vis-à-vis rFRF | |||||
|---|---|---|---|---|---|---|
| 1983-90 | 1991-93 | |||||
| Variables | 1983-90 | 1991-93 | caused by | causing | caused by | causing |
| rDEM | 1.5 | 64 | 7 | 0 | 6 8 | 0.9 |
| rUSD | 52 | 0 | 49 | 0.1 | irrelevant | irrelevant |
Quite similar results are obtained when a Wald test for neutrality, that is, long-run Sims noncausality, is applied (see Figure 2 for an assessment of the results for both subperiods).6 For the first subperiod, we reject German dominance but asymmetry seems to prevail, since the innovations in the German rate and the U.S. rate have persistent effects on the French rate.7 For the second subperiod, the neutrality test in the trivariate system boils down to a simpler test that can be implemented in a bivariate system involving only the two European rates. Since the U.S. rate appears to contain no stochastic trend, it cannot contribute to the nonstationary component of the other two rates, with respect to which it is therefore neutral. This brings us back to the Stock and Watson (1989) analysis of money-output causality. Once the system is orthogonalized, meaning that the instantaneous correlation between the Eurorates is removed, the moving-average representation of the vector error correction model (VECM) yields a zero long-run response of the German rate to an impulse from the French one. In contrast, the French Eurorate is influenced by the orthogonal innovation of the German rate. The same neutrality results are obtained without the 1993 data when resorting a trivariate system.
Non-Zero Long-Run Impulse Responses (Non-Neutrality)
Citation: IMF Staff Papers 1995, 004; 10.5089/9781451930900.024.A008
Notes: The continous (dashed) lines represent rejection of neutrality at 5 percent (10 percent). rDEM, rFRF, and rUSD denote the German, French, and American rate, respectively.Non-Zero Long-Run Impulse Responses (Non-Neutrality)
Citation: IMF Staff Papers 1995, 004; 10.5089/9781451930900.024.A008
Notes: The continous (dashed) lines represent rejection of neutrality at 5 percent (10 percent). rDEM, rFRF, and rUSD denote the German, French, and American rate, respectively.Non-Zero Long-Run Impulse Responses (Non-Neutrality)
Citation: IMF Staff Papers 1995, 004; 10.5089/9781451930900.024.A008
Notes: The continous (dashed) lines represent rejection of neutrality at 5 percent (10 percent). rDEM, rFRF, and rUSD denote the German, French, and American rate, respectively.III. Concluding Remarks
Regarding the effect of German unification, we confirm the presence of a structural break occurring around the end of 1989 as already put forward by GP, but with a somewhat different methodology and implications. As for the causality linkages, the picture drawn by the empirical evidence presented above differs from GP’s conclusion.
For the preunification period, our findings of an asymmetric functioning of the ERM resemble those of GP, though we use a rather ‘stronger’ definition of asymmetry, since it involves the multipliers between the levels of interest rates. For the postunification period, however, our results indicate that the German leadership was not attenuated by German unification as suggested by GP. On the contrary, Germany’s dominant position seems to have become even stronger, as the German rate appears quite independent from any other interest rate, including the U.S. one.
We could interpret the conjunction of this autonomous German monetary policy after German unification with German dominance in the ERM as one possible reason for the 1992-1993 EMS crises. Moreover, it is worth noting that the establishment of the new ERM does not affect our results, and more particularly, does not result in a weaker German position.
References
Bruneau, Catherine, and Jean-Paul Nicolaï, “Persistent Causality in a Multivariate Non Stationary System,” CDC Working Paper No. 1992-24/T (Paris: Caisse des Dépôts et Consignations, August 1992).
Gardner, Edward H., and William R.M. Perraudin, “Asymmetry in the ERM: A Case Study of French and German Interest Rates Before and After German Unification,” Staff Papers, International Monetary Fund, Vol. 40 (June 1993), pp. 427–50.
Gonzalo, Jesus, “Five Alternative Methods of Estimating Long-Run Equilibrium Relationships,” Journal of Econometrics, Vol. 60 (January-February 1994), pp. 203–33.
Granger, C.W.J., “Investigating Causal Relations by Econometric Models and Cross-Spectral Methods,” Econometrica, Vol. 37 (July 1969), pp. 424–38.
Granger, C.W.J., “Some Recent Developments in a Concept of Causality,” Journal of Econometrics, Vol. 39 (September-October 1988), pp. 199–211.
Hansen, Henrik, and Søren Johansen, “Recursive Estimation in Cointegrated VAR Models,” Institute of Economics Discussion Paper No. 92-13 (Copenhagen: University of Copenhagen, 1992).
Henry, Jérôme, and Jens Weidmann (1994a), “Asymmetry in the EMS Revisited: Evidence from the Causality Analysis of Daily Eurorates,” forthcoming in Annales d’Economie et de Statistique; also Banque de France, Note d’Etude et de Recherche, No. 29 (Paris: Banque de France, October 1994).
Henry, Jérôme, and Jens Weidmann (1994b), “The French-German Interest Rate Differential since German Unification: The Impact of the 1992-1993 EMS Crises,” University of Bonn, Sonderforschungsbereich 303 Discussion Paper No. B-295 (Bonn: University of Bonn, 1994).
Johansen, Søren, “Statistical Analysis of Cointegration Vectors,” Journal of Economic Dynamics and Control, Vol. 12 (June/September 1988), pp. 231–54.
Johansen, Søren, “Estimation and Hypothesis Testing of Cointegrating Vectors in Gaussian Vector Autoregressive Models,” Econometrica, Vol. 59 (November 1991), pp. 1551–80.
Mosconi, Rocco, and Carlo Giannini, “Non-Causality in Cointegrated Systems: Representation, Estimation and Testing,” Oxford Bulletin of Economics and Statistics, Vol. 54 (August 1992), pp. 399–417.
Sims, Christopher, “Macroeconomics and Reality,” Econometrica, Vol. 48 (January 1980), pp. 1–48.
Stock, James H., and Mark W. Watson, “Interpreting the Evidence on Money-Income Causality,” Journal of Econometrics, Vol. 40 (January 1989), pp. 161–81.
Toda, Hiro Y., and Peter C.B. Phillips, “Vector Autoregressions and Causality,” Econometrica, Vol. 61 (November 1993), pp. 1367–93.
At the time the paper was written, Jérôme Henry was an Economist at the Macroeconomic Research Unit of the Bank of France. Jens Weidmann is a Research Fellow at the Institute for International Economics, University of Bonn. The authors would like to thank M.J.M. Neumann and the reviewers for helpful comments. The views presented in this paper are strictly those of the authors and should not be interpreted as reflecting the views of the Bank of France.
In this context it is essential not to confuse “predominant position” with “asymmetry.” To be semantically rigorous, we define predominant position (similar to GP) as bilateral causality with the German influence on France being more important than the other way around. Asymmetry designates a unilateral causality from Germany to France, whereas German dominance means that Germany’s influence on France is “monopolistic,” that is, we face asymmetry and the absence of causal linkages between France and the rest of the world (here incorporated by the United States).
The sample period in GP is October 1987-August 1992. The inclusion of the 1993 observations will help us to assess whether Germany still exerts a leadership role in the “new” ERM, allowing for fluctuation bands of ± 15 percent, instead of ±2.25 percent, around the central parities. The widened fluctuation bands could have left more room for maneuver to the other member states’ monetary policy. In any case, it should be emphasized that our results on the structural breaks in both cointegration relations and causal linkages are not qualitatively altered when the sample period is shortened so as to correspond to the one analyzed by GP.
For the univariate unit-root tests, see Henry and Weidmann (1994a and 1994b), where the following results are discussed more extensively and detailed results on all of the causality tests—that is, for both long- and short-run considerations—are presented. Once the existence of some long-run relations is admitted, it is advisable to use full-information maximum likelihood (FIML) estimates of the Johansen (1988 and 1991) type, even though the variables were affected by non-normality (see Gonzalo (1994)).
The normalized Trace statistic is defined as the Trace statistic divided by its critical value and should increase with the sample size under the null hypothesis, that is, from right to left in Figure 1. Hence, we reject the null hypothesis if the normalized Trace statistic exceeds unity.
This result stands up whatever lag length between 5 and 25—the limits corresponding to the “optimal” lag length found by the Schwarz and Akaike criteria, respectively—is chosen. The conclusions are not affected by the exclusion of the data before April 1986 (the date of the last major realignment) or after January 1993 (the date of the EMS crisis and “new” ERM) from the sample. Our detailed results suggest a transition between 1989 and 1990, rather than a sudden one-time break, which would be in line with GP. According to the latter, a less pronounced change in the system’s structure occurred at the beginning of 1991, probably concurring with the break point we selected.
Non-neutrality means nonzero terms in the impulse response function derived from the moving-average representation of a stationary process. In the bivariate case, without any instantaneous correlation, it is equivalent to weak exogeneity. One advantage of the neutrality concept, based on the long-run effects, is that it is less likely to be affected by the misspecification of volatility, for example, omission of ARCH effects, than, for instance, Granger causality, which passes through the variables’ changes, that is, the short-run linkages.
The neutrality of the French rate to the German rate, however, depends on the lag length. When few lags, namely ten, are used, the French rate seems to be non-neutral to the German rate at the 10 percent significance level. This property disappears when the lag length is increased. Moreover, no instantaneous correlation is found between the French rate’s innovations, on the one hand, and the other two rates’ innovations, on the other hand. On balance, it is very likely that the two stochastic trends come from the German and U.S. monetary policies. Figure 2 presents the findings for ten lags.
