This chapter estimates money demand under two alternative specifications, one using the domestic interest rate, and the other using the interest rate differential between the domestic interest rate and the United States interest rate. In the period 1992:1-1999:1, there is cointegration between the real monetary aggregates Ml and M2, real output, and either the domestic interest rate or the interest rate differential The long-run income elasticity is not statistically different from one. The long-run interest rate semi-elasticity, or the long-run interest rate differential semi-elasticity, is only significant for real M2. The two semi-elasticities are not statistically different.

Abstract

This chapter estimates money demand under two alternative specifications, one using the domestic interest rate, and the other using the interest rate differential between the domestic interest rate and the United States interest rate. In the period 1992:1-1999:1, there is cointegration between the real monetary aggregates Ml and M2, real output, and either the domestic interest rate or the interest rate differential The long-run income elasticity is not statistically different from one. The long-run interest rate semi-elasticity, or the long-run interest rate differential semi-elasticity, is only significant for real M2. The two semi-elasticities are not statistically different.

V. Money Demand in a Small Open Economy: The Case of the Dominican Republic56

A. Introduction

95. The objective of this chapter is to estimate a money demand equation for the Dominican Republic. The formulation of monetary policy in the Dominican Republic is centered around an annual monetary program prepared by the central bank (BCRD) and discussed with the government.57 The theoretical framework of the program is the monetary approach to the balance of payments. Given expected annual real output growth, together with the inflation and exchange rate/foreign reserve objectives of the monetary authorities, an estimated money demand establishes a constraint on the assets and liabilities of the BCRD’s balance sheet. The program also specifies quarterly objectives for the intermediate targets (currency) and monetary policy instruments (e.g., central bank paper). The quarterly objectives serve as guidelines for the Monetary and Exchange Affairs Committee as it monitors higher frequency indicators of the demand for money. Deviations from the projected path trigger a consultation with the governor of the BCRD and the Monetary Board, which ultimately decides the course of action to take.

96. The main instrument used for the implementation of monetary policy is central bank paper called certificados de participatión. However, the BCRD also manages liquidity in the system using direct measures such as credit controls and the freezing of excess reserves held by financial institutions at the BCRD. The BCRD intervenes in the free (commercial bank) foreign exchange market mostly with the objective of smoothing the irregular and seasonal components of exchange rate behavior.58 Since late 1991, interest rates have been freely determined by market forces.

97. The motivation for this study is threefold. The first motivation for estimating a money demand equation for the Dominican Republic is to test whether there is a long-run (cointegrating) relationship between real monetary aggregates and real income. The key role that money demand plays in the formulation and implementation of monetary policy in the Dominican Republic contrasts with the doubt that there is a long-run relationship (cointegration) between real money aggregates and real income both in academia and among policy makers (Leiderman and Svensson, 1995, Blinder, 1998).59

98. The second motivation for this study is to test the degree of independence that the BCRD has in setting monetary policy. According to the Mundell-Fleming model, which underlies the estimation of money demand in this chapter, a policy induced increase in interest rates encourages capital inflows (Mundell, 1963). In the absence of central bank intervention in the foreign exchange market, the exchange rate appreciates as a result.60 Those capital inflows are normally intermediated by the banking system, which may buy the foreign currency with cash or lend the foreign currency domestically. If private agents demanding the foreign funds do not run down their bank deposits, it is possible for M2 (and credit to the private sector) to grow, and for interest rates to fall. The main point is that, depending on the degree of capital mobility and asset substitutability, the final effect of the monetary policy tightening may be smaller than its initial effect, both on M2 and on the interest rate.

99. The third motivation, a corollary of the last point, is to be able to assess the stance of monetary policy (see Christiano et al. 1998). The literature normally finds that a contractionary monetary policy increases domestic interest rates and appreciates the domestic currency. This highlights the role of capital flows in open economies discussed above. In the recent Asian crisis, this has been at the heart of much debate. The press argued that high interest rates in Asia indicated a “tight” monetary policy. Based on the growth of monetary aggregates, Corsetti et al. (1998) characterized the monetary policy stance in Asia as “loose.” This debate suggests that the “monetary policy stance” may not be well measured by interest rates alone, or by the growth of monetary aggregates alone, whenever there is feedback between monetary aggregates and interest rates, as is the case in small open economies. Interest rates contain both policy- and market-determined elements and it is important to consider the evolution over time of financial variables in accurately assessing the stance of monetary policy (Tanner, 1999).

100. Econometric estimators of money demand equations should be able to deal with the suggested endogeneity of interest rates. This paper uses a Phillips-Loretan (1991) non-linear dynamic least squares estimator to estimate two versions of a standard money demand equation, one that uses as a regressor a domestic interest rate, and another one that uses the interest rate differential between the Dominican Republic and the United States. The next section describes the model and the econometric technique used. Section B discusses the results. Section C concludes and discusses some policy implications. The appendix discusses unit root and cointegration results in detail.

B. The Model and the Estimation Technique

101. Given the theoretical framework of the Dominican Republic monetary program, this paper investigates whether two measures of real money aggregates, Ml and M2, are cointegrated with real output and nominal interest rates. Two sets of interest rates are used, a domestic interest rate, and an interest rate differential between the country and the rest of the world. Because the interest rate differential can be viewed as a measure of the degree of capital mobility (Cuddington, 1983, or Siklos, 1996), a statistical comparison between the estimates using the domestic interest rate and the estimates using the interest rate differential will be used to assess whether it is valid to ignore the openness of the capital account in money demand estimation in the Dominican Republic.

102. With all variables except interest rates expressed in logs, the money demand equation is:

(MP)t=a+byt+crt+εt,(1)

where M is a nominal monetary aggregate, P is the consumer price index, y is real output, r is an interest rate, and e is a normally distributed disturbance with zero mean and variance σ2ϵ. Similarly, the money demand equation with the interest rate differential is:

(MP)t=a+byt+c(rtrt*)+ηt(2)

where r* is the foreign interest rate, and η is a normally distributed disturbance with zero mean and variance σ2η.

103. If there is a long-run relationship among real monetary aggregates, real income, and interest rates, then there will be feedback between that long-run equilibrium relationship and the errors that drive the regressors (i.e., real output and interest rates). OLS, single equation error correction methods, and unrestricted VARs will lead to estimators that are asymptotically biased and inefficient. Therefore, equations (1) and (2) were estimated using the non-linear dynamic least squares estimator of Phillips and Loretan. The authors show that this single-equation technique is asymptotically equivalent to a maximum likelihood estimator on a full system of equations under Gaussian assumptions. The technique provides estimators that are statistically efficient, and whose t-ratios can be used for inference in the usual way. Most importantly, the method takes into account both the serial correlation of the errors and the endogeneity of the regressors that are present when there is a cointegration relationship. The two regressions estimated are given by equations (3) and (4),

(MP)t=a+byt+crt+Σi=kk[diΔyti+eiΔrti]+ρ[(MP)t1abyt1crt1]+εt,(3)
(MP)t=a+byt+c(rr*)t+Σi=kk[diΔyti+eiΔ(rr*)ti]+ρ[(MP)t1abyt1c(rr*)t1]+ηt.(4)

104. Note that equations (3) and (4) include leads and not just lags. Phillips and Loretan (1991) show that leads are required to produce valid conditioning (i.e., to make the residuals ϵt and ηt orthogonal to the entire history of the regressors). Similarly, the estimator includes the lagged equilibrium relationship as well as lags of changes in the left-hand side variable (MP)t1. The reason is that lags of (MP)t1 are not good proxies for the past history of ϵt and ηt because of the persistence in effects of innovations from the unit roots in equations (1) and (2). This requires the use of a non-linear technique.

105. In this study, we are mostly interested in the estimated values of the coefficients “b” and “c” because they are the parameter estimates of the long-run relationship between money aggregates, real output, and interest rates. It is expected that b = 1, and that c < 0 in the case of M1. As indicated earlier, the openness of the capital account and the frequent foreign exchange interventions of the BCRD may have resulted in a positive correlation between real M2 and the domestic interest rate, suggesting that c>0. Perhaps equally important, in the relatively underdeveloped state of the Dominican financial markets, time deposits (quasi-money) serve as the main savings instrument, also suggesting that c>0 for M2.

106. Equations (3) and (4) were estimated using quarterly data from 1992:1 to 1999:1. There are no indices of real activity available at a higher frequency in the Dominican Republic. Extending the sample back in time would imply going into a period when interest rates were not market-determined and important structural reforms, documented in other chapters of this report, had not yet taken place. The domestic interest rate used was the 90-day deposit rate and the foreign interest rate used was the 90-day U.S. treasury bill rate.

C. Unit Roots, Cointegration, and Long-Run Elasticities

Unit root and cointegration tests61

107. Table 1 reports the results for unit root tests. In general, the two sets of tests considered tend to indicate that real M1, real M2, the domestic interest rate, and the interest rate differential are unit root processes. However, in the case of real output, only one of the two tests confirmed it was a unit root process, but it was the more powerful of the two. Other econometric tests were consistent with real output being a unit root process (see Appendix).

Table 1.

Dominican Republic: Unit Root Tests at 5 Percent Level

ΔXt=α+βt+γXt1+Σi=1p1φiΔXti+εt

(1992:1–1999:1)

article image
Sources: Central Bank of the Dominican Republic; and Fund staff estimates.Note: IDOM=90-day lending rate in the Dominican Republic.

  • IRD=interest rate differential; i.e., 90-day lending rate in the Dominican Republic minus 90-day T-bill rate in the United States.

  • M1D=M1 first differenced.

  • M2D=M2 first differenced.

  • GDPD=GDP first differenced.

  • IDOMD=90-day lending rate in the Dominican Republic first differenced.

  • IRDD=90-day lending rate in the Dominican Republic minus 90-day T-bill rate in the United States first differenced.

Lags were chosen according to the Akaike Information Criterion and for white noise of the residuals.

The power of ρμ (only constant) and ρτ (constant and time trend) is higher than the power of Tμ (only constant) and Tτ (constant and time trend) when the alternative is stationary.

The Newey-West weighting scheme was used for estimating the variances of Sμ2 and Sτ2.

108. Tests for cointegration were based on the Johansen-Juselius (1990) method with critical values corrected for small sample bias using Cheung and Lai’s (1999) approach (Table 2). Tests of the residuals indicated that they were not serially correlated. Overall, there is strong statistical evidence of a long-run cointegration relationship between real monetary aggregates, real output, and interest rates in the Dominican Republic during the sample period. For more information on the tests see the Appendix.

Table 2.

Dominican Republic: The Johansen-Juselius Maximum Likelihood Test for Cointegration

1992:1–1999:1

article image
Sources: Central Bank of the Dominican Republic; and Fund staff estimates.r is the number of cointegrated vectors.p is the number of variables.The 99 percent (denoted with *) and 95 percent (denoted with **) critical values corrected for small samples using Cheung and Lai (1993) are also used to evaluate the results.The models include a drift term in the variables but not in the cointegration space. The normality test is a multivariate version of the Shenton-Bowman test for normality for individual time series. The LM1 and LM4 are the Langrange multiplier tests. p values are in parentheses.

The long-run elasticities of the model

109. Although the main objective of the paper is the testing of the existence of a long-run relationship between real money aggregates, real output, and interest rates, it was also thought important to look into the dynamics of the short-run disequilibrium. As a result, Table 3 reports not only the long-run parameters of the models, but also the parameters of the short-run dynamics from the Phillips-Loretan non-linear dynamic least squares estimator.62

Table 3.

Dominican Republic: The Phillips-Loretan Nonlinear Dynamic Least Squares Estimator

(1992–1999:1)

article image
Table 3.

Dominican Republic: The Phillips-Loretan Nonlinear Dynamic Least Squares Estimator

(1992–1999:1)

article image
Sources: Central Bank of the Dominican Republic; and Fund staff estimates.t ratios are in parentheses. Barlett-Kolmogorov-Smirnov (B-K-S) 10 percent critical value is 0.305.

110. Analysis of the residuals indicated that they were white noise; there is agreement between the non-parametric test (Bartlett-Kolmogorov-Smirnov) at the 10 percent level and the visual observation of the residuals in Figures 1-4.63 The residuals are also homoskedastic according to two chi-square tests using one and four lags.64

Figure 1.
Figure 1.

Dominican Republic: Residual from Model 1–M1

Citation: IMF Staff Country Reports 1999, 117; 10.5089/9781451811285.002.A005

Figure 2.
Figure 2.

Dominican Republic: Residual from Model 1–M2

Citation: IMF Staff Country Reports 1999, 117; 10.5089/9781451811285.002.A005

Figure 3.
Figure 3.

Dominican Republic: Residual from Model 2–M1

Citation: IMF Staff Country Reports 1999, 117; 10.5089/9781451811285.002.A005

Figure 4.
Figure 4.

Dominican Republic: Residual from Model 2–M2

Citation: IMF Staff Country Reports 1999, 117; 10.5089/9781451811285.002.A005

111. The constant and the long-run output elasticity are significant at the 99 percent level. As expected, the long-run output elasticity is not statistically different from one in any case at the 90 percent level (and above) as denoted by the χ2 statistic.

112. Importantly, the long-run interest rate semi-elasticity of real M1, using either the domestic interest rate or the interest rate differential, is not statistically different from zero at conventional confidence levels. In contrast, the long-run interest rate semi-elasticity of real M2 is positive, and strongly significant. Its value, however, is small (0.05). It should be noted that the coefficient on the domestic interest rate is not statistically different from the coefficient on the interest rate differential. Given the econometric technique used, this is consistent with predictions of the Mundell-Fleming model, suggesting that it is the interest rate differential that matters for money demand in a small open economy such as the Dominican Republic. In other words, it is the ability of the central bank to affect the interest rate differential that influences money demand, not its ability to set a domestic interest rate independent of foreign interest rates.

113. The positive long-run interest rate semi-elasticity of real M2 is consistent with the open economy paradigm of Mundell-Fleming suggesting a reduction over time in the efficacy of monetary policy in creating a wedge between domestic and foreign interest rates. A tightening of monetary policy, for instance, increases the domestic interest rate, encouraging capital inflows and an appreciation of the exchange rate.65 If the authorities let the currency appreciate, real M2 may increase as the banking system intermediates the capital inflow and domestic currency denominated deposits rise. If the authorities intervene to prevent the appreciation of the exchange rate, the monetary base may increase (increasing real M2, ceteris paribus) as long as they do not sterilize their intervention. Sterilization, to offset the increase in the monetary base, would put upward pressure on interest rates, attracting further capital inflows.

114. It is noteworthy that all coefficients of changes either in the domestic interest rate or in the interest rate differential (lagged, contemporaneous, led) that are significant, are also negative. Moreover, they are not statistically different across models. The coefficients on the interest rate changes reflect the short-run dynamics of the model.

115. The varying results for short- and long-run coefficients illustrate that in assessing the stance of monetary policy in open economies, it is important to distinguish the long-run equilibrium from the short-run dynamics. Otherwise, the identification of the effects of policy shocks (and nonpolicy shocks) on interest rates is likely to be difficult. For example, a policy induced monetary tightening will increase interest rates and reduce money demand in the short run. However, because of the feedback between interest rates and capital flows, over time, we may find that the initial monetary tightening produces an increase in monetary aggregates. In addition, if inflation expectations decline (as they should with a monetary policy tightening), interest rates will eventually decline, and this should not be interpreted as a loosening of monetary policy.

116. Finally, the highly significant values of the coefficients measuring the previous period deviation from long-run equilibrium indicate that adjustment takes place between three and six quarters. This is consistent with most accounts of the lags with which monetary policy normally operates.

D. Conclusions and Policy Implications

117. This study reports the estimation of two money demand models for two real monetary aggregates (M1 and M2), one using the domestic interest rate and the other using the interest rate differential between the 90-day domestic deposit rate and the 90-day U.S. treasury bill rate. The results suggest that in the sample period 1992:1–1999:1 there is cointegration between the real monetary aggregates M1 and M2, real output, and either the domestic interest rate or the interest rate differential.

118. The long-run income elasticity is not statistically different from one in any of the cases studied. The long-run interest rate semi-elasticity, or the long-run interest rate differential semi-elasticity, is significant for real M2 demand, but not for real M1 demand. Moreover, the long-run interest rate semi-elasticity of the domestic interest rate is not statistically different from the semi-elasticity of the interest rate differential. The long-run semi-elasticities always assume a low value.

119. Overall, the results of the paper lend support to the central role that money demand has in the monetary program of the BCRD. However, the results also indicate that the long-run efficacy of monetary policy in the Dominican Republic is reduced when it is measured by its efficacy in affecting the domestic interest rate in a lasting manner.

120. Despite the robustness of the results of the paper, it should be kept in mind that the short sample available prevented any meaningful stability test. Similarly, an out-of-sample simulation could not be performed. Finally, the use of the terms “long-run cointegration relationships” between real monetary aggregates, real output, and interest rates in this paper should be put in the context of the seven-year length of the sample available.66

List of References

  • Blinder, A. S., 1998, Central Banking in Theory and Practice (Cambridge, Massachusetts: The MIT Press).

  • Cheung, Y.W. and K.S. Lai, 1993, “Finite Sample Sizes of Johansen’s Likelihood Ratio Test for Cointegration,” Oxford Bulletin of Economics and Statistics, 55: 313328.

    • Search Google Scholar
    • Export Citation
  • Christiano, L. J., M. Eichenbaum, and C. L. Evans, “Monetary Policy Shocks: What Have We Learned and to What End?” National Bureau of Economic Research, Working Paper No. 6400.

    • Search Google Scholar
    • Export Citation
  • Corsetti, G., P. Pesenti, and N. Roubini, (1998), “What Caused the Asian Currency and Financial Crisis? Part II: The Policy Debate,” National Bureau of Economic Research, Working Paper No. 6834.

    • Search Google Scholar
    • Export Citation
  • Cuddington, J. T., 1983, “Currency Substitution, Capital Mobility, and Money Demand,” Journal of International Money and Finance, 2: 111133.

    • Search Google Scholar
    • Export Citation
  • Dickey, D. and W.A. Fuller, 1979, “Distribution of Estimates for Autoregressive Time Series with Unit Root,” Journal of American Statistical Association, 74: 427431.

    • Search Google Scholar
    • Export Citation
  • Dickey, D. and W. Fuller, 1981, “The Likelihood Ratio Statistic for Autoregressive Time Series with a Unit Root,” Econometrica 49: 10571072.

    • Search Google Scholar
    • Export Citation
  • Gonzalo, J., 1994, “Five Alternative Methods of Estimating Long-Run Equilibrium Relationships,” Journal of Econometrics, 60: 203233.

    • Search Google Scholar
    • Export Citation
  • Johansen S. and K. Juselius, 1990, “The Full Information Maximum Likelihood Procedure for Inference on Cointegration—With Application to the Demand for Money,” Oxford Bulletin of Economics and Statistics, 52: 169210.

    • Search Google Scholar
    • Export Citation
  • Leiderman, L. and L. O. Svensson (eds), 1995, Inflation Targets, Centre for Economic Policy Research (London).

  • Mundell, R. A., 1963, “Capital Mobility and Stabilization Policy Under Fixed and Flexible Exchange Rates,” Canadian Journal of Economics and Political Science 29, 475485.

    • Search Google Scholar
    • Export Citation
  • Nadal-De Simone, F. and W. Razzak, 1999, “Nominal Exchange Rates and Nominal Interest Rate Differentials,” (Washington DC: International Monetary Fund) Working Paper (forthcoming).

    • Search Google Scholar
    • Export Citation
  • Perron, P., 1988, “Trends and random Walks in Macroeconomic Time Series,” Journal of Economic Dynamics and Control, 12: 297332.

  • Phillips, P. C. B., 1987, “Time Series Regressions with a Unit Root,” Econometrica, 55: 277301.

  • Phillips, P. C. B. and M. Loretan, 1991, “Estimating Long-Run Equilibria,” The Review of Economic Studies, 58: 407436.

  • Said, S. and D. A. Dickey, 1984, “Testing for Unit Roots in Autoregressive-Moving Average Models of Unknown Order, Biometrika, 71: 599608.

    • Search Google Scholar
    • Export Citation
  • Siklos, P., 1996, “Searching for an Improved Monetary Indicator for New Zealand,” (Wellington: Reserve Bank of New Zealand), Discussion Papers Series G96/1.

    • Search Google Scholar
    • Export Citation
  • Tanner, E., 1999, “Exchange Market Pressure and Monetary Policy: Asia and Latin America in the 1990s” (Washington DC, International Monetary Fund, unpublished).

    • Search Google Scholar
    • Export Citation

ANNEX Unit Roots and Cointegration

121. The literature on unit roots and cointegration is vast and will not be reviewed here. Suffice it to say that there is a valid concern among economists about the appropriateness of the tests for unit roots and their power against stationary alternatives. The choice of a particular testing methodology is not straightforward. Ultimately, one may not be able to determine whether there is a unit root in a given time series. Inevitably, however, a choice has to be made. In testing for unit roots and cointegration, the strategy followed in this study is to use different tests. A decision is then made based on whether the results of these various tests converge or not. Two popular methods to test for unit roots were used: the ADF test (Dickey and Fuller, 1979, 1981, and Said and Dickey, 1984), and the Phillips-Perron test (Phillips (1987) and Perron (1988)).

122. Table 1 reports the results for unit root tests.67 In general, the two sets of tests considered tend to agree that real M1, real M2, the domestic interest rate, and the interest rate differential are unit root processes. However, while the Tμ and the Tτ versions of the ADF and the Phillips-Perron tests reject the null of a unit root in real output, the ρμ and the ρτ versions of those tests accept the null. As the latter versions of the tests are more powerful against a stationary alternative, it was decided that real output may contain a unit root. That decision was also based on the observation that the spectrum of the first difference of real output has Granger’s typical spectral shape.68

123. Gonzalo (1994) compared five different residual-based tests for cointegration. Among them, he recommends using the Johansen-Juselius (1990) method. Although very popular in the literature, this test has been highly criticized for its lack of power in finite samples, and—among other problems—by its sensitivity to the choice of the lag length. This test was used and the results are reported in Table 2. Given the relatively short sample period available, the critical values were corrected for small sample bias using Cheung and Lai’s (1993) approach. Lag length was evaluated as follows. A general lag model was fit. Then, unnecessary lags were eliminated by testing backward using the Schwarz Criterion. The residuals of the models were checked for white noise each time using LM(1) and LM(4) tests, and for normality using a multivariate version of the Shenton-Bowman test. In all cases the tests accepted the null of Gaussian residuals. The LM(1) tests for real M2 indicated some serial correlation. The LM(4) tests—and LM tests with longer lags—however, accepted the null of white noise for the residuals at reasonable confidence levels.

124. At the 99 percent level and in all cases analyzed in this study, the λtrace statistic strongly rejected the null hypothesis of no cointegrating vectors against the alternative of one or more cointegrating vectors (r>0). Similarly, at the 99 percent level and in all cases, the λmax statistic rejected the null hypothesis of no cointegrating vectors against the alternative of one cointegrating vector (r=1). For M2, the λtrace rejected the null hypothesis of r≤1 cointegrating vectors against the alternative of two or more cointegrating vectors at the 95 percent level. For M1, that rejection occurred at the 90 percent level. However, the λtrace statistics rejected the null of one cointegrating vector (r=1) against the specific alternative of two cointegrating vectors (r=2) at the 99 percent level in all cases except M1 and the interest rate differential where the rejection occurred at the 95 percent level.69

125. Finally, if real output were stationary, one more cointegrating vector would be required. As a result, multivariate tests of non-stationarity were performed. They did not reject the null of a unit root for real output when the aggregate is real M1 but they did reject the null of a unit root for real output in the case of real M2. This confirmed the decision taken based on Dickey-Fuller and Phillips-Perron tests for the case of M1. As a result, real GDP was detrended assuming a deterministic trend and all the cointegration tests were run again. At the 99 percent level, the λmax statistic rejected the null hypothesis of two cointegrating vectors against the alternative of three cointegrating vectors for real M2 and the domestic interest rate. That rejection occurred at the 95 percent level for real M2 and the interest rate differential.

56

This chapter was prepared by F. Nadal-De Simone. I am grateful to P. Brenner, J. Chan-Lau, D. Dunn, M. Kaufman, S. Lizondo, R. Rennhack, E. Tanner, and P. Young for their comments.

57

The Monetary and Financial Code currently being discussed in congress would require that the monetary program be submitted to congress.

58

There is a dual foreign exchange market in the Dominican Republic: all traditional exports, credit card, and telecommunication transactions are subject to surrender requirements (about 15 percent of the total volume of foreign exchange transactions) and the remainder goes through the free market. The BCRD is responsible for providing foreign exchange for the payment of the petroleum import bill and the servicing of the public sector’s foreign debt.

59

Nonetheless, many economists believe that there may be a role for monetary policy in the short run.

60

If the central bank intervenes in the foreign exchange market to moderate the change in the exchange rate, M2 will increase unless the intervention is fully sterilized.

61

The Appendix elaborates on the unit root and cointegration tests performed.

62

The tests were started using a lag (and lead) structure similar to that in Johansen. The lag-lead structure necessary to eliminate serial correlation varied across models; one lag and one lead were preferred for all cases except for real M2 and the domestic interest rate where two leads and two lags were preferred. In all cases, however, conscious about Phillips and Loretan warning of over-fitting, the number of leads was reduced by one first. Lags were reduced then if necessary. Every time, the parameter estimates and their significance as well as whether the residuals were white noise, was checked.

63

The figures have an upper and a lower confidence interval calculated as a Bartlett’s test which is normally distributed. The confidence intervals are wide due to the relatively short sample period. However, note that 100 observations would give a value of ±0.20 for the 95 percent level and about ±0.16 for the 90 percent level.

64

The R2 is reported although in a cointegrated system estimated with valid conditioning it is not meaningful as a measure of fit.

65

Nadal-De Simone and Razzak (1999) found that increases in the interest rate differential between the United States and Germany, and between the United States and the United Kingdom, appreciated the U.S. dollar during the floating period.

66

Admittedly, seven years is already “the long-run” for monetary policy.

67

The choice of the lag structure always has been an issue. The objective of the lags is to remove serial correlation. With this objective in mind, the lag order was set as the highest significant lag order—using an approximate 95 percent confidence interval—from either the autocorrelation function or the partial autocorrelation function of the first-differenced series. Also the Akaike Information Criterion (AIC) was used. Every time a lag was eliminated, serial correlation was checked using the Ljung-Box test for white noise. The approach followed in selecting the lags was also followed in testing non-stationarity in the individual series.

68

Results are available upon request.

69

The λmax statistic has a sharper alternative hypothesis than the λtrace statistic. In case of conflict, the former is to be preferred to the latter.

Dominican Republic: Selected Issues
Author: International Monetary Fund