Dual foreign currency markets and the role of expectations: Evidence from the Pacific Basin countries

Research in International Business and Finance 21 (2007) 238–259

Dual foreign currency markets and the role of expectations: Evidence from the Pacific Basin countries Panayiotis F. Diamandis a , Georgios P. Kouretas b,∗ , Leonidas Zarangas c a

b

Department of Business Administration, Athens University of Economics and Business, GR-10434 Athens, Greece Department of Economics, University of Crete, University Campus, GR-74100 Rethymno, Greece c Department of Finance and Auditing, Technological Educational Institute of Epirus, 210 Ioanninon Ave., GR-48100 Preveza, Greece Received 15 June 2005; received in revised form 13 March 2006; accepted 16 March 2006 Available online 27 June 2006

Abstract This paper analyzes the role of expectations about the government policy in the official foreign currency market in determining the black market premium. We use data for the recent float from six emerging markets of the Pacific Basin where active black markets for foreign currency exist, namely, Indonesia, Korea, Malaysia, the Philippines, Singapore and Thailand. To test the impact of anticipated and unanticipated shocks to the official exchange rate on the black market premium, we employ the two-step procedure of Hoffman et al. [Hoffman, D.L., Low, S.A., Schlagenhauf, D.E., 1984. Tests of rationality, neutrality and market efficiency: a Monte Carlo analysis of alternative test statistics. J. Monet. Econ. 14, 339–363] which provides corrected F-statistics and allows us to draw valid inference in the presence of generated regressors. The main finding of our analysis is that anticipated and unanticipated shocks to the official exchange rate have an impact on the black market premium in all six Pacific Basin countries. These results suggest that portfolio balance models provide the suitable theoretical framework for analyzing the behaviour of the black market premium in the markets for foreign currency in the Pacific Basin countries. Furthermore, this implies that economic agents in these countries are sensitive to expected returns in foreign exchange. © 2006 Elsevier B.V. All rights reserved. JEL classification: F31; F33 Keywords: Black market premium; Generated regressors; Cointegration analysis; Anticipated shocks; Expectations; Portfolio balance models

∗

Corresponding author. Tel.: +30 2831077412; fax: +30 2831077406. E-mail address: [email protected] (G.P. Kouretas).

0275-5319/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.ribaf.2006.03.008

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

239

1. Introduction The existence of parallel or ‘black’ markets, particularly for US dollars, is a well known feature of many developing countries and countries where trade and capital control exist. Foreign exchange and trade controls are a major factor in the ‘parallel’ or ‘black’ market for foreign currency. When access to the official foreign exchange market is limited, then those who need foreign exchange in order to make international transactions of goods, services and assets will have an incentive to find an alternative source. Thus, the development of excess demand for foreign currency at the official rate gives an incentive to those who have an excess supply of foreign currency to sell it illegally at a price higher than the official rate. Thus, the existence of exchange rate controls causes a divergence between the equilibrium rate and the official rate and this leads to the emergence of a parallel or ‘black‘market for foreign exchange in the country (see Edwards, 1989; Agenor, 1992; Montiel et al., 1993; Kiguel and O’Connell, 1995; Phylaktis, 1996). The importance of studying the operation of parallel markets has far reaching implications for our understanding of the economic performance under alternative exchange rate regimes. Recently, Reinhart and Rogoff (2004) provided substantial evidence that the classification of alternative exchange rate regimes over the Bretton Woods period can be substantially overturned if we employ market-determined parallel exchange rates. This argument relies on the fact that not only in developing countries but in virtually all the European countries up until the late 1950s, and sometimes well beyond, dual exchange rates markets were very important. Thus, in most cases of dual systems, Reinhart and Rogoff (2004) show that the floating parallel exchange rate, was not only a much better indicator of monetary policy than the official exchange rate, but also that in a large number of cases the dual/parallel rates have been used as a form of “back door” floating, even if they were usually accompanied by capital controls. Furthermore, it has been documented in numerous studies that the parallel market premium often serves as a reliable guide to the direction of future official exchange rate changes.1 Cheung and Lai (2005) have linked the significance of studying the dual exchange markets to another important issue, namely nominal exchange rate flexibility and real exchange rate adjustment. They argue that while numerous studies since the adoption of flexible exchange rates have discussed the issue of real exchange rate adjustment by comparing nominal exchange rate flexibility in the fixed and flexible exchange rates period no study have examined this question across exchange rate regimes. Cheung and Lai (2005) employ parallel exchange rates from developing countries for the Bretton Woods and post-Bretton Woods periods and they argue that there is no significant evidence that greater nominal exchange flexibility tends to produce faster or slower real exchange rate adjustment in either period. Finally, Dornbusch et al. (1983), Kamin (1993), Montiel and Ostry (1994) and Pozo and Wheeler (1999) have also emphasized the important role that expectations about future official exchange rate changes play in driving the premium. A ‘parallel’ or ‘black’ market for US dollars has operated for years in several Pacific Basin countries, namely Indonesia, Korea, Malaysia, the Philippines, Singapore and Thailand. Specifically, in the Philippines, Singapore and Malaysia the black market has operated continuously since the 1940s. The persistence of the existence of black markets is the outcome of various types of foreign exchange controls in force and the administrative setting of the exchange rate by the mon-

1 Ghei and Kamin (1999) recognize that the parallel exchange rate is a good, though not entirely perfect, proxy for the free-market currency value.

240

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

etary authorities, as well as of social and political unrest, and economic turbulence. Thus, in the Philippines and Thailand exchange controls have been implemented during the period under examination, whereas in the Philippines the emergence of the black market for foreign currency was further supported by the outright theft of millions of US dollars of foreign support and assistance payments, and by the funding of capital flight which arose from fear of dictatorship, of confiscation of assets and of blocking of bank accounts. Furthermore, in Thailand the development of the parallel market was linked to narcotics-related activities (Phylaktis and Kassimatis, 1994; Kiguel and O’Connell, 1995; Moore and Phylaktis, 2000; World Currency Yearbook, various issues). In Korea black markets emerged as a result of strict foreign exchange controls, which have been partially lifted following the liberalization programs on foreign exchange markets that this country implemented in 1988. Furthermore, in Korea the operation of the black market was supported through sustained capital flight, along with the widespread corruption which dominated most sectors of political and economic life. Singapore and Malaysia, where foreign exchange controls were abolished in the late 1970s, have also experienced the presence of a limited black market. With respect to Malaysia, the market was supported by capital flight from Indonesia, while the black market economy which existed for tax-evasion purposes boosted the demand for black market dollars. In Singapore, the demand for black market dollars was created by immigrants from India (mostly Muslims) who sent collected foreign currency to their families in India though the black market channels. Furthermore, in Indonesia, although no capital controls existed, the emergence of the black market for dollars resulted from a set of import restriction measures imposed, which created incentives to smuggle goods and therefore raised the demand for black market dollars. This demand was further amplified by the imposition of an exchange tax on export proceeds which diverted foreign currency to the black market. Finally, the black market in Indonesia was further supported by significant amounts of money coming from corruption, as well as by capital flight initiated by politicians securing their own wealth (Agenor, 1992; Montiel et al., 1993; Phylaktis and Kassimatis, 1994; Kiguel et al., 1997). This paper contributes to the literature on the operation of the dual exchange rate systems by focusing explicitly on the role that agents’ expectations about the official exchange rate play in determining the black market premium. We apply our analysis to monthly data for six Pacific Basin countries, our sample covers the period January 1974 to December 1998. The theoretical framework used to provide a better understanding of the determination of the black market premium is based on the monetary and the portfolio balance models of nominal exchange rate determination. Fig. 1 shows the evolution of the official and parallel exchange rates from 1974 to 1998, while Fig. 2 shows the evolution of the parallel market premium, the percentage differential between the black market and official exchange rate, for the same period.2 As expected, the black market rate exhibited greater volatility than the official rate since it was freely determined in the market, in 2 Price data on the black market for foreign currency are often not immediately available and this is the reason for not using more recent observations. However, the aim of this paper is not the dataset itself. The focus of the paper is to model and understand the workings of the black market in the Pacific Basin countries (where the black market still exists) and to learn what policy conclusions can be drawn from this analysis which can help us to understand similar positions in countries of other regions. This is what is accomplished by the present study. Furthermore, the development of new econometric methodologies provide a better framework for the study of an already known data set used in the past, but we are now offered the likelihood of obtaining richer conclusions and better modelling of this data set, as well as of applying these methodologies to black market exchange rate series from other countries where such a market is still in operation. The use of a particular data set that in many cases is very old is very common in time series analysis. A good

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

241

response to actual and anticipated changes in economic conditions. Exchange rate arrangements were not unified in the area during the period under examination. Singapore and Malaysia placed the effective rate of their currencies on a controlled floating basis in the early 1970s. The other four economies in our sample continued to link their currencies to the US dollar after its floating in 1971, and to control their exchange rates by reducing the gold content of their currencies. Indonesia and Thailand in 1978, Korea in 1980, and the Philippines in 1984, finally abolished the fixed exchange regime with the adoption of a controlled floating effective exchange rate. We also note that there are times during which the black market rate was below the official rate (such as in mid-1985), indicating a negative black market premium. Although not the regular case, negative premiums arise during periods when commercial banks are forced not to accept foreign currency without proper identification of the seller. In such cases the seller is willing to undersell his foreign currency in the black markets (Dornbusch et al., 1983). In the case of the Pacific Basin countries as the analysis above made clear, negative premiums could have emerged because of widespread corruption, a substantial black market economy, or drug-related activities.3 The econometric analysis used to test the impact of anticipated and unanticipated shocks to the official exchange rate on the black market premium, is based on the two-step procedure developed by Hoffman et al. (1984), which provides a corrected F-test that helps us to make valid inference in the presence of generated regressors. The first step of this methodology requires the estimation of the forecasting equation for the official exchange rate for each country. In the second step we use information derived from this forecast to analyze how anticipated and unanticipated changes in the official exchange rate influence the level of the black market premium in Indonesia, Korea, Malaysia, the Philippines, Singapore and Thailand. The main finding of this analysis is that both anticipated and unanticipated shocks to the official exchange rate have an impact on the black market premium in Indonesia, Korea, Malaysia, the Philippines, Singapore and Thailand. These results suggest that portfolio balance models may be appropriate for understanding the behaviour of the black market premium in countries where dual markets for foreign currency exist. This piece of evidence is of great importance to portfolio managers as well as to individual and institutional investors whose portfolios partially include these currencies which were found to be sensitive to changes in anticipated returns. The rest of the paper is organized as follows. Section 2 outlines the theoretical and methodological issues relevant to our study. Section 3 presents the econometric methodology. Section 4 presents and discusses the empirical results. Section 5 concludes the paper. 2. Models of parallel markets for foreign currency Over the last 30 years the analysis of the workings of the dual foreign currency markets has led to the development of alternative theoretical models.4 The earliest approach to modeling the

example is the use of data from the 1920s which has been used extensively by many researchers in an attempt to better understand a flexible exchange rate system or to validate the market efficiency hypothesis. Another good example is the use of century-long data used to examine the validity of PPP. 3 The sample extends beyond the period of the Asian crisis of 1997. The main economic shock during the period under examination is the financial and currency crisis of 1997–1998. However, an examination of the relevant data does not reveal abnormal behaviour on the premium, although in the period just after these events the premium increased. Then with the passage of enough time to smooth out the currency markets in Southeast Asia the premium decreased. 4 Agenor (1992) and Montiel et al. (1993) provide an extensive analysis of the alternative theoretical specifications which study the implications of the existence of parallel currency markets.

242

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

Fig. 1. The official and black market exchange rates.

Fig. 2. The black market premium.

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

243

parallel currency markets comes from the theory of international trade (Culbertson, 1975; Sheikh, 1976, 1989; Martin and Panagariya, 1984; Pitt, 1984; de Macedo, 1987; Branson and de Macedo, 1989). This approach stresses the central role that transactions demand for foreign exchange plays in the evolution of a black or “parallel” market for foreign currency (mainly that of US dollars). In these models the black market for foreign currency represents the demand for foreign currency by agents, mainly to purchase imports outside the official market and the supply of foreign currency. Martin and Panagariya (1984), Sheikh (1976, 1989) and Pitt (1984), among others argue that smuggling and underinvoincing of exports are the main sources of supply. Culbertson (1975) emphasizes that the main supply of parallel currency comes from the resale of officially allocated foreign exchange. Real trade models abstract from portfolio considerations and only focus on the impact of trade restrictions The second class of models relies on the monetary approach to the exchange rate determination. This model was initially put forward by Blejer (1978). It assumes that the demand for foreign currency in the parallel market arises from the need by economic agents to reallocate the composition of their portfolios, while the demand for foreign currency satisfied by the official currency market is channeled to the purchase of goods. This model was extended by Agenor (1991) who focused on removing some limitations of Blejer’s model. Agenor (1991) considered explicitly stock–flow interactions and developed a model that includes elements of both the monetary and currency substitution models. The significance of the monetary factors with regard to the behaviour of the black market exchange rate has been confirmed for a number of developing countries by several studies (Blejer, 1978; Edwards, 1988; Edwards and Montiel, 1989; Van den Berg and Jayanetti, 1993; Bahmani-Oskooee, 1993; Kouretas and Zarangas, 1998). Dornbusch et al. (1983), de Macedo (1985, 1987), Frenkel (1990) and Phylaktis (1991) developed the final class of parallel currency models, namely the portfolio balance model. This model combines features of real trade models with features of monetary approach models. Empirical support for the ability of portfolio balance models to explain the determinants of black market premium has been provided by Dornbusch et al. (1983), Phylaktis (1991) and Phylaktis and Manalis (1995) for several developing countries. The theory on modelling the premium has paid a lot of attention to its behaviour but less attention to its determinants. An examination of Fig. 2 reveals that the black market premia exhibit significant variation over time for all the Pacific Basin countries under examination. It is well documented that the size of the black market premium mainly depends on the ability of authorities to detect and impose an array of penalties on individuals who have possibly engaged in illegal transactions (Edwards, 1989, 1998, 1999; Montiel et al., 1993; Phylaktis, 1996; Pozo and Wheeler, 1999). As a consequence, any changes in the probability of detection and the magnitude of sanctions will cause the premium to vary. Furthermore, Kiguel et al. (1997) notes that the official rate typically diminishes in importance when the gap between the official and market-determined rate widens.5 The portfolio balance model of exchange rate determination and its variants place a significant weight on the effect that the arrival of news has on an asset price and, in our case on the nominal exchange rate. It is mainly within this framework that expectations about government policy with respect to the level of the official exchange rate are considered to be a major determinant of

5 Ghei and Kamin (1999) classify exchange rates episodes where there are dual/parallel markets into three tiers according to the percentage levels of the parallel market premia: low (below 10%), moderate (10% and above but below 50%), and high (50% and above).

244

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

the level of the parallel market premium, (Dornbusch et al., 1983; Phylaktis, 1991). Therefore, we consider likely determinants of the premium to be the official exchange rate, the effective yield on foreign currency denominated assets and expectations of official devaluation. The main prediction of the model is that the premium increases when there are expectations of depreciation of the official exchange rate, while at the time that devaluation actually takes place, the premium approaches zero (Phylaktis and Kassimatis, 1994; Phylaktis, 1996; Moore and Phylaktis, 2000). In addition, Agenor and Taylor (1993), Kamin (1993), Edwards (1994) and Gervais and Larue (2001) assume that expectations of official devaluation due to inflation, and the consequent reduction in official exchange reserves, lead to an increase in the black market premium and again the premium approaches zero when the devaluation actually takes place. Several main characteristics emerge from the models which explain the black market premium. First, these models conclude that expectations regarding future official exchange rate devaluation result in increases in the premium. Second, these models suggest that the premium tends to zero when the devaluation actually takes place. Thus, these models show that expectations of official exchange rate devaluation result in movements in the premium since actual devaluation of the official exchange rate reduces the level of the premium, as opposed to an increase in the premium when markets anticipate a devaluation, but actual devaluation is post-poned by the government. In addition, in many instances governments have some flexibility with respect to setting the level of the premium (Phylaktis, 1996; Kouretas and Zarangas, 1998; Pozo and Wheeler, 1999 and Moore and Phylaktis, 2000). 3. Econometric methodology In the late 1970s and early 1980s the growing significance of expectations led to the estimation of regression equations which used constructed variables. The most popular approach to the issue of generated regressors was the use of residuals from an estimated equation as unanticipated values. Barro’s (1977) work was one of the first attempts to test the macro rational expectations hypothesis. He used generated regressors to examine the impact of anticipated and unanticipated demand shocks on the economy. Thus, Barro (1977) developed a test based upon a two-step estimation technique in which fitted values and residuals from a regression are used as proxies for anticipated values and unanticipated shocks, respectively. This two-step methodology was criticized initially by Minskin (1982) who argued that although it generates consistent parameter statistics, it leads to invalid statistical inference since when the off-diagonal elements of the system are non-zero, then the OLS statistics do not have correct asymptotic distributions leading to invalid statistical inference. Pagan’s (1984, 1986) seminal works formalized the issues arising from the use of generated regressors and discussed alternative possible solutions to the efficiency and valid inference problems under these circumstances.6 The FIML and Generalized Least Squares are two methodologies which have frequently been applied to overcome these problems, since they allow us for the incorporation of the correction of the structural errors in the joint estimation that these approaches conduct in one step. To test the impact of anticipated and unanticipated shocks to the official exchange rate on the black market premium, we apply the two-step procedure suggested by Hoffman et al. (1984)

6 Pagan (1984, 1986), Murphy and Topel (1985) and McAleer and McKenzie (1991) are among the numerous papers which studied the issue of generated regressors. Oxley and McAleer (1993) provide an extensive literature review on this issue.

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

245

which provides corrected F-statistics that allow us to make valid inference in the presence of generated regressors. This is an approach that provides a solution to the problem of generated regressors within the framework of the OLS estimation. This approach is computationally easier relative to the FIML or GLS methods. According to the Hoffman et al. (1984) procedure, we initially construct a forecasting equation for the official exchange rate and we assume that the predicted values from this equation are measures of anticipated shocks, while the residuals from this equation are taken as measures of unanticipated shocks. The second step involves a regression between the black market premium on a constant and current and lagged values of anticipated and unanticipated shocks. The statistical significance of anticipated and unanticipated shocks of the official exchange rate on the premium is tested using corrected F-tests constructed by Hoffman et al. (1984).7 Therefore, this methodology will help us to draw conclusions about the ability of governments to influence the level of the black market premium.8 We assume that the official exchange rate in country i depends on a constant and lagged values of the domestic price level (Pi ), the black market exchange rate (Bi ), the official exchange rate (Oi ), domestic money (Mi ), exports (XPi ), imports (MPi ), US prices (PUS ), and the US money supply (MUS ). These macroeconomic variables are typically used in monetary and asset models of exchange rate determination. Since at the time expectations are formed economic agents possess all necessary information we are allowed to include in the equation lagged values of the parallel and official exchange rates. We also include in the forecasting equation lagged values of the macroeconomic variables defined above. Thus, the forecasting equation for the official exchange rate in country i is given by: i i i i i i Oit = f (Pt−1 , Pt−2 , . . . , Bt−1 , Bt−2 , . . . , Oit−1 , Oit−2 , . . . , Mt−1 , Mt−2 ,..., US US US US , Pt−2 , . . . , Mt−1 , Mt−2 , . . .) XPit−1 , XPit−2 , . . . , MPit−1 , MPit−2 , . . . , Pt−1

(1)

We consider the predicted values from the forecasting equations to be the anticipated movements in the official exchange rate, while the residuals from this equation are considered to be the unanticipated movements in the official exchange rate. Furthermore, we define the anticipated changes in the official exchange rate in country i as Ai and the unanticipated official exchange rate changes as Ui . The second step of Hoffman et al. (1984) methodology requires the use of Ai and Ui in an equation in which we estimate the black market premium for country i (ρi ). The general formulation of this equation, which includes the generated regressors, is given by: i i + . . . + c12 Ut−12 + et ρti = a + b0 Ait + bt Ait−1 + . . . + b12 At−12 + c0 Uti + c1 Ut−1

(2)

The null hypothesis under investigation is that anticipated (unanticipated) changes in the official exchange rate do not affect the black market premium. Therefore, joint significance of the bj implies that Ai influences ρi , while joint significance of the cj implies that Ui influences ρi .

7 Using Monte Carlo experiments, Hoffman et al. (1984) examine the small size properties of the alternative test statistics. They show that the corrected two-step approach gives as accurate and efficient results as the FIML without relying on a computationally difficult non-linear algorithm. In addition the two-step tests appear to be as powerful as the FIML in most cases. 8 Pozo and Wheeler (1999) who employed this methodology to examine the impact of expectations regarding official exchange rate devaluation on the black market premium for four Latin American countries, reached conclusions similar to ours.

246

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

4. Empirical results The data used in this paper is monthly and covers the period from January 1974 to December 1998 (which is the last available observation) and the countries under examination are Indonesia, Korea, Malaysia, the Philippines, Singapore and Thailand. The parallel (black) market exchange rates were taken from the monthly series in various issues of the World Currency Yearbook (1976–1998), and its predecessor, Pick’s Currency Yearbook (1968–1975). The data for the official exchange rates were obtained from the CD-ROM edition of the IMF’s International Financial Statistics. The exchange rates are end-of-month observations and are all expressed as national currency units per US dollar. The price indices for all six Pacific Basin countries and the US are the consumer price indices. M1 is the monetary aggregate for all countries. Both the consumer prices and the monetary variables are taken from IMF’s International Financial Statistics. Finally, exports from each of these countries to the US in millions of US dollars and imports of each of these countries from the US in millions of US dollars were obtained from the IMF’s Direction of Trade Statistics. As explained in Section 3, the Hoffman et al. (1984) two-step procedure requires the estimation of a forecasting equation for the official exchange rate and then in the second step the estimation of an equation of the black market premium using as independent variables the forecasts of the official exchange rate and the residuals from the first equation. It is clear that the estimation of both equations employs data which is likely to be non-stationary. Therefore, it is necessary to test for the presence of one or possibly two unit roots in the variables under consideration, namely the official exchange rate, the black market exchange rate, the domestic consumer price, the domestic money supply, exports, and imports, as well as the US consumer price and the US money supply. To examine whether the series under consideration are stationary, we apply the Elliott et al. (1996) GLS augmented Dickey–Fuller test (DF-GLSu ) and Ng and Perron (2001) GLS versions of the modified Phillips–Perron (1988) tests (MZGLS and MZGLS ). The null hypotha t esis is that of a unit root against the alternative that the initial observation is drawn from its unconditional distribution and uses GLS-detrending as proposed by Elliott et al. (1996) and extended by Elliott (1999), to maximize power, and a modified selection criterion to select the lag truncation parameter in order to minimize size distortion. In the GLS procedure of Elliott et al. (1996), the standard unit root tests (without trend) are applied after the series are first detrended under the local alternative ρ = 1 + α/T. This was found to provide substantial power gains for the DF-GLSu test resulting in power functions that lie just under the asymptotic power envelope. Ng and Perron (2001) find similar gains for the MZGLS and MZGLS tests. They also found that a modification of the AIC criterion (MIC), a t gives rise to substantial size improvements over alternative selection rules such as BIC. For robustness, we then apply the Kwiatkowski et al. (1992) KPSS test for the null hypothesis of level or trend stationarity against the alternative of non-stationarity. The results of the unit root and stationarity tests are presented in Table 1. The results show that we are unable to reject the null hypothesis of non-stationarity with the DF-GLSu and MZGLS and a MZGLS tests, and we reject the null hypothesis of stationarity with the KPSS test for the t levels of both series. The results are reversed when we take the first difference of each exchange rate series, which leads us to the conclusion that all variables are realizations of I(1) processes. The second step in our analysis deals with the issue of detecting cointegration for the set of variables in the forecasting Eq. (1). The starting point of cointegration analysis is Johansen (1988,

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

247

Table 1 Unit roots and stationarity tests Variables

MZGLS a

DF-GLSu tμ

MZGLS t

KPSS-tests ημ

ητ

−0.17 (5) −4.59* (8) 0.69 (3) −9.55* (2) 1.67 (1) −8.32* (0) 1.92 (0) −8.41* (0) 4.42 (3) −2.65* (2) 1.33 (7) −2.45* (0) 1.85 (3) −12.49* (0) 0.22 (13) −3.10* (2)

1.405* 0.097 1.671* 0.166 1.991* 0.324 1.971* 0.118 2.076* 0.202 2.004* 0.258 2.063* 0.356 2.073* 0.303

0.291* 0.102 0.325* 0.068 0.264* 0.050 0.294* 0.053 0.287* 0.180 0.469* 0.105 0.327* 0.131 0.309* 0.103

1.07 (3) −26.78* (5) 1.78 (2) −216.8* (1) 1.11 (2) −211.8* (1) 0.91 (0) −204.7* (0) 1.07 (1) −6.47* (0) 1.00 (13) −22.87* (7)

1.27 (3) −3.65* (5) 1.08 (2) −9.96* (1) 0.95 (2) −10.23* (1) 0.71 (0) −10.11* (0) 3.69 (1) −1.70* (0) 1.44 (13) −3.25* (7)

1.999* 0.204 1.781* 0.079 1.485* 0.244 1.408* 0.097 1.954* 0.370 2.055* 0.321

0.219* 0.105 0.243* 0.020 0.285* 0.084 0.266* 0.092 0.407* 0.135 0.249* 0.063

−0.86* (13) −3.21* (12) −1.37* (5) −3.82* (6) −1.30 (1) −12.91* (0) −1.20 (0) −16.42* (0) −1.00 (1) −4.75* (0) −3.02 (13) −5.84* (4)

2.29 (13) −16.98* (9) 0.68 (5) −29.13* (4) −2.47 (1) −121.0* (0) −1.22 (0) −135.3* (0) 1.44 (6) −13.92* (3) 1.70 (13) −15.80* (4)

2.19 (13) −2.90* (9) 0.68 (5) −3.71* (4) −0.63 (1) −7.77* (0) −0.34 (1) −8.22* (0) 2.61 (6) −2.61* (3) −0.69 (13) −2.75* (4)

1.654* 0.329 1.652* 0.104 0.996* 0.279 1.015* 0.303 1.865* 0.310 1.866* 0.152

0.441* 0.058 0.401* 0.097 0.349* 0.073 0.362* 0.072 0.242* 0.211 0.264* 0.122

−0.77 (14) −6.42* (4) −1.50 (4) −14.57* (2) −1.41 (1) −11.92* (0) −1.58 (1) −18.00* (0) −2.13 (6)

3.25 (12) −133.5* (8) 0.45 (4) −458.8* (3) 1.21 (1) −114.8* (0) 1.11 (1) −133.3* (0) 0.84 (6)

9.74 (12) −8.11* (8) 0.33 (4) −15.13* (3) 1.80 (1) −7.56* (0) 1.41 (1) −8.16* (0) 0.85 (6)

1.667* 0.399 1.497* 0.089 1.805* 0.123 1.766* 0.078 1.973*

0.156* 0.449 0.472* 0.073 0.353* 0.102 0.338* 0.053 0.370*

tτ

Indonesian Rupiah (Dollar)a Xpus 0.045 (5) xpus −3.24* (8) Mpus −0.48 (3) mpus −14.74* (2) Ofex 1.62 (1) ofex −13.07* (0) Blex 1.89 (0) blex −14.89* (0) Pd 3.39 (3) pd −3.45* (1) Pus 0.90 (7) pus −6.09* (0) M1 4.02 (3 M1 −20.88* (0) M1us −0.28 (13) M1us −4.71* (1)

−2.16 (8) −18.45* (1) −3.39 (3) −6.14* (2) −2.33 (4) −13.31* (0) −1.85 (0) −15.52* (0) −1.50 (3) −3.83* (2) −0.40 (7) −6.32* (0) −0.85 (1) −21.66* (0) −2.36 (13) −8.15* (1)

−0.32 (5) −42.46* (8) 1.73 (3) −197.4* (2) 2.09 (1) −138.6* (0) 2.20 (0) −141.8* (0) −1.90 (3) −14.33 (2) 0.89 (7) −13.13* (2) 1.44 (3) −312.1* (0) 0.29 (13) −20.18* (2)

Korean Won (Dollar)b Xpus 1.088 (13) xpus −3.30* (13) Mpus 0.53 (2) mpus −16.10* (1) Ofex 0.87 (2) ofex −11.71* (1) Blex 0.69 (0) blex −14.75* (0) Pd 3.29 (1) pd −11.16* (0) M1 1.06 (13) M1 −3.16* (10)

−1.46* (14) −5.88* (8) −1.47* (4) −11.78* (3) −0.73 (1) −14.24* (1) −1.50 (0) −13.93* (0) 0.04 (1) −4.56* (0) −1.00 (13) −4.36* (10)

Malaysian Ringit (Dollar)c Xpus 1.82 (13) xpus −3.31* (12) Mpus 0.90 (5) mpus −12.36* (4) Ofex −0.61 (0) ofex −11.97* (0) Blex −0.34 (0) blex −14.47* (0) Pd 1.88 (6) pd −2.77* (2) M1 −1.97 (13) M1 −4.85* (4) Philippine Peso (Dollar)d Xpus 5.03 (12) xpus −2.40* (13) Mpus 0.61 (4) mpus −13.48* (3) Ofex 1.69 (1) ofex −11.71* (0) Blex 1.53 (1) blex −18.24* (0) Pd 0.70 (6)

248

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

Table 1 (Continued ) Variables

tμ pd M1 M1

MZGLS a

DF-GLSu

MZGLS t

tτ

−2.87 (4) 5.08 (0) −15.30* (0)

KPSS-tests ημ

ητ

−3.92 (3) −1.80 (0) −16.14* (0)

−19.10 (3) 1.70 (0) −119.2* (0)

−3.08 (3) 5.18 (0) −7.71* (0)

0.279 2.035* 0.124

0.050 0.371* 0.054

Singaporean Dollar (Dollar)e Xpus −2.15 (12) xpus −5.03* (11) Mpus 1.17 (3) mpus −15.14* (2) Ofex −0.04 (1) ofex −2.86* (4) Blex −0.12 (1) blex −2.80* (6) Pd 0.90 (6) pd −3.42* (5) M1 2.89 (1) M1 −23.46* (0)

−1.24 (13) −4.95* (11) −1.34 (3) −4.17* (4) −1.84 (1) −9.99* (0) −1.37 (1) −6.79* (2) −1.72 (6) −3.72* (5) −2.26 (1) −25.89* (0)

1.84 (13) −25.89* (6) 1.72 (3) −366.2* (2) −0.03 (1) −19.39* (5) 0.09 (1) −17.56* (3) 0.90 (6) −28.65* (3) 1.25 (1) −126.8* (0)

1.47 (13) −3.59* (6) 1.77 (3) −42.76* (2) −0.02 (1) −3.05* (5) 0.08 (1) −2.95* (3) 1.12 (6) −3.78* (3) 3.01 (1) −7.95* (0)

1.873* 0.144 1.784* 0.221 1.751* 0.123 1.738* 0.121 1.866* 0.245 1.995* 0.231

0.429* 0.092 0.425* 0.059 0.232* 0.125 0.229* 0.084 0.294* 0.111 0.239* 0.038

Thai Baht (Dollar)f Xpus −1.84 (14) xpus −2.78* (12) Mpus −0.18 (3) mpus −20.95* (1) Ofex −0.24 (1) ofex −12.34* (0) Blex 0.01 (0) blex −9.93* (1) Pd 3.49 (3) pd −3.61* (3) M1 0.94 (13) M1 −3.29* (9)

−0.63 (14) −3.47* (12) −2.19 (2) −21.09* (1) −2.66 (1) −12.39* (0) −2.58 (0) −18.28* (0) −0.56 (1) −6.93* (0) −2.35 (13) −4.51* (9)

32.12 (14) −41.47* (5) −0.39 (3) −217.3* (1) −0.67 (1) −134.0* (0) −1.51 (0) −100.4* (1) 1.34 (3) −25.39* (0) 1.27 (13) −23.11* (8)

2.89 (14) −4.54* (5) −0.24 (3) −10.42* (1) −0.24 (0) −8.18* (0) −0.45 (0) −7.08* (1) 4.29 (3) −3.49* (0) 1.48 (13) −3.30* (8)

1.903* 0.346 1.716* 0.230 1.448* 0.083 1.559* 0.087 2.004* 0.341 2.074* 0.052

0.517* 0.123 0.362* 0.052 0.226* 0.044 0.232* 0.042 0.340* 0.150 0.241* 0.044

*

*

*

*

The DF-GLSu due to Elliott et al. (1996) and Elliott (1999) is a test with an unconditional alternative hypothesis. The standard Dickey–Fuller tests are detrended (with constant or constant and trend). The critical values for the DF-GLSu test at the 5% significance level are: −2.73 (with constant) and −3.17 (with constant and trend), respectively (Elliott, 1999). MZa and MZt are the Ng and Perron (2001) GLS versions of the Phillips–Perron tests. The critical values at 5% significance level are: −8.10 and −1.98 (with constant), respectively (Ng and Perron, 2001, Table 1). ημ and ητ are the KPSS test statistics for level and trend stationarity, respectively (Kwiatkowski et al., 1992). For the computation of theses statistics a Newey and West (1994) robust kernel estimate of the “long-run” variance is used. The kernel estimator is constructed using a quadratic spectral kernel with VAR(l) pre-whitening and automatic data-dependent bandwidth selection (see, Newey and West, 1994, for details). The 5% critical values for level and trend stationarity are 0.461 and 0.148, respectively, and they are taken from (Sephton (1995), Table 2). * Significance at the 95% confidence level. a Notes: xpus, mpus, ofex, blex, pd, pus, M1 and M1us are, respectively, the price of exports to the US, the price of imports from the US, the official and parallel exchange rates, the Indonesian cpi, the US cpi, the Indonesian M1 and the US M1.  indicate first differences. b Notes: xpus, mpus, ofex, blex, pd and M1 are, respectively, the price of exports to the US, the price of imports from the US, the official and parallel exchange rates, the Korean cpi and the Korean M1.  indicate first differences. c Notes: xpus, mpus, ofex, blex, pd and M1 are, respectively, the price of exports to the US, the price of imports from the US, the official and parallel exchange rates, the Malaysian cpi and the Malaysian M1.  indicate first differences. d Notes: xpus, mpus, ofex, blex, pd and M1 are, respectively, the price of exports to the US, the price of imports from the US, the official and parallel exchange rates, the Philippines cpi and the Philippines M1.  indicate first differences. e Notes: xpus, mpus, ofex, blex, pd and M1 are, respectively, the price of exports to the US, the price of imports from the US, the official and parallel exchange rates, the Singaporean cpi and the Singaporean M1.  indicate first differences. f Notes: xpus, mpus, ofex, blex, pd and M1 are, respectively, the price of exports to the US, the price of imports from the US, the official and parallel exchange rates, the Thai cpi and the Thai M1.  indicate first differences.

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

249

Table 2 Residual misspecification tests of the model Equation

σε

L-B (36)

Indonesian Rupiah (Dollar)a,b (a) Univariate residual diagnostics xpus 0.062 37.94 mpus 0.045 19.28 ofex 0.048 59.37 blex 0.066 24.77 pd 0.007 212.00 pus 0.001 75.38 M1 0.030 130.52 M1us 0.005 68.43 L-B (4288)

ARCH (6)

η3

η4

NORM (4)

R2

2.05 16.42 15.18 12.14 3.85 9.13 13.80 4.38

0.67 0.46 3.11 2.82 0.35 −0.05 0.77 0.44

7.79* 5.13* 37.70* 26.61* 3.66 3.85 6.96* 3.67

91.57* 32.18* 224.50* 140.95* 8.01 9.71 59.82* 6.61

0.515 0.701 0.416 0.482 0.714 0.314 0.890 0.264

LM (1)

(b) Multivariate residuals diagnostics 805.44 (0.09) 90.43 (0.02) Equation

σε

L-B (36)

Korean Won (Dollar)c,d (a) Univariate residual diagnostics xpus 0.055 49.44 mpus 0.012 24.92 ofex 0.028 51.09 blex 0.022 39.32 pd 0.017 28.11 pus 0.003 54.90 M1 0.032 90.22 M1us 0.004 59.21 L-B (4544)

L-B (36)

Malaysian Ringit (Dollar)e,f (a) Univariate residual diagnostics xpus 0.056 18.33 mpus 0.067 35.41 ofex 0.001 31.79 blex 0.002 18.42 pd 0.003 55.89 pus 0.001 53.12 M1 0.021 103.22 M1us 0.005 49.00 L-B (3904)

63.62 (0.49)

564.20 (0.00)

η3

η4

NORM (4)

R2

41.67* 13.20* 4.04 6.38 2.77 5.28 7.06 3.80

−0.21 1.08 6.29* 0.77 0.47 −0.03 0.009 0.073

5.70* 10.76* 69.26* 6.92 4.36 4.22 4.06 3.58

55.62* 127.47* 1497.30* 59.83* 17.39* 17.31* 13.83* 5.43

0.498 0.360 0.406 0.191 0.528 0.685 0.549 0.868

LM (1)

σε

χ2 (16)

ARCH (3)

(b) Multivariate residuals diagnostics 796.88 (0.21) 88.11 (0.55) Equation

LM (4)

LM (4)

χ2 (16)

67.90 (0.39)

1722.60 (0.00)

ARCH (6)

η3

η4

NORM (4)

R2

32.74* 21.00 9.12 12.48 19.88 6.39 3.98 3.04

0.15 0.75 −0.81 −0.25 0.26 −0.02 0.29 −0.04

6.03* 6.83* 15.51* 6.46* 4.70* 4.77* 4.85* 3.89

62.50* 54.78* 327.32* 73.07* 24.81* 28.65* 27.69* 9.78*

0.567 0.572 0.472 0.266 0.454 0.758 0.623 0.898

LM (1)

(b) Multivariate residuals diagnostics 1003.1 (0.09) 65.07 (0.44)

LM (4)

χ2 (16)

90.08 (0.02)

716.67 (0.00)

250

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

Table 2 (Continued ) Equation

σε

L-B (36)

Philippine Peso (Dollar)g,h (a) Univariate residual diagnostics xpus 0.023 40.33 mpus 0.047 68.21 ofex 0.023 49.12 blex 0.045 60.95 pd 0.007 35.08 pus 0.001 60.12 M1 0.025 48.97 M1us 0.005 53.90 L-B (3712)

ARCH (4)

η3

η4

NORM (4)

R2

9.15 7.60 10.51* 4.70 3.20 19.12* 23.23* 4.68

0.20 −2.74 2.19 2.56 −1.01 −0.07 1.03 0.27

4.07 38.37* 16.86* 22.91* 16.74* 4.16 6.88* 4.20

11.79* 31.89* 91.08* 113.58* 305.53* 13.98* 39.68* 13.71*

0.554 0.535 0.295 0.214 0.627 0.733 0.204 0.877

LM (1)

(b) Multivariate residuals diagnostics 809.13 (0.12) 78.20 (0.11) Equation

σε

L-B (36)

Singaporean Dollar (Dollar)i,j (a) Univariate residual diagnostics xpus 0.063 42.77 mpus 0.071 80.80 ofex 0.008 47.60 blex 0.015 15.76 pd 0.003 32.89 pus 0.001 35.16 M1 0.030 25.68 M1us 0.004 44.13 L-B (3712)

L-B (36)

Thai Baht (Dollar)k,l (a) Univariate residual diagnostics xpus 0.023 30.18 mpus 0.045 26.13 ofex 0.017 59.77 blex 0.029 25.23 pd 0.004 40.45 pus 0.001 56.23 M1 0.019 77.16 M1us 0.004 39.40 L-B (3904)

81.67 (0.07)

939.14 (0.00)

η3

η4

NORM (4)

R2

47.37* 33.88 18.78 51.73* 52.45* 7.49 25.50 4.41

−0.11 0.08 −0.24 −0.17 0.33 −0.46 −1.94 −0.03

4.89 4.58* 4.42* 4.61 3.74 3.90 12.48* 2.99

32.00* 24.35* 19.49* 24.40* 8.21* 11.34* 87.30* 0.14

0.736 0.691 0.510 0.450 0.563 0.794 0.590 0.695

LM (1)

σε

χ2 (16)

ARCH (11)

(b) Multivariate residuals diagnostics 776.21 (0.05) 46.69 (0.95) Equation

LM (4)

LM (4)

χ2 (16)

70.39 (0.27)

244.91 (0.00)

ARCH (11)

η3

η4

NORM (4)

R2

45.11* 35.10 19.60 6.20 25.47* 5.27 10.11 3.69

−0.05 1.39 1.17 0.26 0.37 −0.50 0.05 0.09

3.25* 8.27* 12.74* 5.11* 3.98* 4.61* 3.66* 3.34*

1.62 58.82* 167.67* 36.42* 11.62* 20.65* 6.40 2.54

0.773 0.699 0.526 0.440 0.594 0.793 0.733 0.921

LM (1)

(b) Multivariate residuals diagnostics 803.67 (0.25) 95.56 (0.01)

LM (4)

χ2 (16)

79.24 (0.09)

233.57 (0.00)

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

251

Table 2 (Continued ) Notes: The lag structure of the model is k = 6. σ ε is the standard error of the residuals, η3 and η4 are the skewness and kurtosis statistics. L-B is the Ljung-Box test statistic for residual autocorrelation, ARCH is the test for heteroscedastic residuals, and NORM the Jarque–Bera test for normality. The ARCH and NORM statistics are distributed as χ2 with 6 and 4 degrees of freedom, respectively, and the L-B statistic is distributed as χ2 with 36 degrees of freedom. * Significance at the 5% level. b Notes: L-B is the multivariate version of the Ljung-Box test for autocorrelation based on the estimated auto- and cross-correlations of the first [T/4 = 73] lags distributed as a χ2 with 4288 degrees of freedom. LM (1) and LM (4) are the tests for first- and fourth-order autocorrelation distributed as χ2 with 64 degrees of freedom and χ2 is a normality test which is a multivariate version of the Shenton–Bowman (1977) test modified in Doornik and Hansen (1994). Numbers in parentheses refer to marginal significance levels. c Notes: The lag structure of the model is k = 3. σ is the standard error of the residuals, η and η are the skewness and 3 4 ε kurtosis statistics. L-B is the Ljung-Box test statistic for residual autocorrelation, ARCH is the test for heteroscedastic residuals, and NORM the Jarque–Bera test for normality. The ARCH and NORM statistics are distributed as χ2 with 3 and 4 degrees of freedom, respectively, and the L-B statistic is distributed as χ2 with 36 degrees of freedom. * Significance at the 5% level. d Notes: L-B is the multivariate version of the Ljung-Box test for autocorrelation based on the estimated auto- and cross-correlations of the first [T/4 = 74] lags distributed as a χ2 with 4544 degrees of freedom. LM (1) and LM (4) are the tests for first- and fourth-order autocorrelation distributed as χ2 with 64 degrees of freedom and χ2 is a normality test which is a multivariate version of the Shenton–Bowman (1977) test modified in Doornik and Hansen (1994). Numbers in parentheses refer to marginal significance levels. e Notes: The lag structure is k = 6. σ is the standard error of the residuals, η and η are the skewness and kurtosis ε 3 4 statistics. L-B is the Ljung-Box test statistic for residual autocorrelation, ARCH is the test for heteroscedastic residuals, and NORM the Jarque–Bera test for normality. The ARCH and NORM statistics are distributed as χ2 with 6 and 4 degrees of freedom, respectively, and the L-B statistic is distributed as χ2 with 36 degrees of freedom. * Significance at the 5% level. f Notes: L-B is the multivariate version of the Ljung-Box test for autocorrelation based on the estimated auto- and cross-correlations of the first [T/4 = 67] lags distributed as a χ2 with 3904 degrees of freedom. LM (1) and LM (4) are the tests for first- and fourth-order autocorrelation distributed as χ2 with 64 degrees of freedom and χ2 is a normality test which is a multivariate version of the Shenton–Bowman (1977) test modified in Doornik and Hansen (1994). Numbers in parentheses refer to marginal significance levels. g Notes: The structure lag of the model is k = 4. σ is the standard error of the residuals, η and η are the skewness and ε 3 4 kurtosis statistics. L-B is the Ljung-Box test statistic for residual autocorrelation, ARCH is the test for heteroscedastic residuals, and NORM the Jarque–Bera test for normality. The ARCH and NORM statistics are distributed as χ2 with 4 and 4 degrees of freedom, respectively, and the L-B statistic is distributed as χ2 with 36 degrees of freedom. * Significance at the 5% level. h Notes: L-B is the multivariate version of the Ljung-Box test for autocorrelation based on the estimated auto- and cross-correlations of the first [T/4 = 62] lags distributed as a χ2 with 3712 degrees of freedom. LM (1) and LM (4) are the tests for first- and fourth-order autocorrelation distributed as χ2 with 64 degrees of freedom and χ2 is a normality test which is a multivariate version of the Shenton–Bowman (1977) test modified in Doornik and Hansen (1994). Numbers in parentheses refer to marginal significance levels. i Notes: The lag structure is k = 11. σ is the standard error of the residuals, η and η are the skewness and kurtosis ε 3 4 statistics. L-B is the Ljung-Box test statistic for residual autocorrelation, ARCH is the test for heteroscedastic residuals, and NORM the Jarque–Bera test for normality. The ARCH and NORM statistics are distributed as χ2 with 11 and 4 degrees of freedom, respectively, and the L-B statistic is distributed as χ2 with 36 degrees of freedom. * Significance at the 5% level. j Notes: L-B is the multivariate version of the Ljung-Box test for autocorrelation based on the estimated auto- and cross-correlations of the first [T/4 = 69] lags distributed as a χ2 with 3712 degrees of freedom. LM (1) and LM (4) are the tests for first- and fourth-order autocorrelation distributed as χ2 with 64 degrees of freedom and χ2 is a normality test which is a multivariate version of the Shenton–Bowman (1977) test modified in Doornik and Hansen (1994). Numbers in parentheses refer to marginal significance levels. k Notes: The lag structure is k = 11. σ is the standard error of the residuals, η and η are the skewness and kurtosis ε 3 4 statistics. L-B is the Ljung-Box test statistic for residual autocorrelation, ARCH is the test for heteroscedastic residuals, and NORM the Jarque–Bera test for normality. The ARCH and NORM statistics are distributed as χ2 with 11 and 4 degrees of freedom, respectively, and the L-B statistic is distributed as χ2 with 36 degrees of freedom. * Significance at the 5% level. a

252

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

Table 2 (Footnote continued ) l Notes: L-B is the multivariate version of the Ljung-Box test for autocorrelation based on the estimated auto- and cross-correlations of the first [T/4 = 72] lags distributed as a χ2 with 3904 degrees of freedom. LM (1) and LM (4) are the tests for first- and fourth-order autocorrelation distributed as χ2 with 64 degrees of freedom and χ2 is a normality test which is a multivariate version of the Shenton–Bowman (1977) test modified in Doornik and Hansen (1994). Numbers in parentheses refer to marginal significance levels.

1991) and Johansen and Juselius (1990).9 Furthermore, Johansen (1992, 1995, 1997) developed a testing procedure, which has been extended by Paruolo (1996) and Rahbek et al. (1999), that allows us to sequentially test, according to the Pantula (1989) principle, for the presence of I(2) and I(1) components in a multivariate framework. In order for the procedure to be valid, we have to check that the assumptions underlying the model are satisfied. In particular we verify that the estimated residuals do not deviate from being Gaussian white noise errors. The lag structure of the estimated VAR model is chosen based on several univariate and multivariate misspecification tests. Specifically, the multivariate L-B test for serial correlation up to the 36th order and the multivariate LM tests for first and fourth order residual autocorrelations are not significant, whereas multivariate normality is clearly violated. Normality can be rejected as a result of skewness (third moment) or excess kurtosis (fourth moment). Since the properties of the cointegration estimators are more sensitive to deviations from normality due to skewness than to excess kurtosis, we report the univariate Jarque–Bera test statistics together with the third and fourth moment around the mean. It turns out that the rejection of normality is essentially due to excess kurtosis, and hence not so serious for the estimation results. The presence of non-normality may be attributed to the fact that the official exchange rates were administratively determined throughout the period under consideration as well as to the short-term interest rates, signifying both the high volatility of money stock in both countries. The ARCH statistical criterion tests for autoregressive conditional heteroscedasticity and the null hypothesis is rejected for most of the equations. Again cointegration estimates are not very sensitive to ARCH effects.10 The R2 measures the improvement in explanatory power relative to the random walk (with drift) hypothesis, i.e. xt = μ = εt . These measures show that with this information set we can explain quite a large proportion of the variation in the exchange rates and the money supply. Thus, we note that our conditional VAR model is well-specified. Table 2 reports the univariate and multivariate misspecification tests. Table 3 presents the results of the formal tests for the joint determination of the cointegration rank and the presence of I(2) components in our system. We have included in the estimation procedure 11 centered seasonal dummies, which are necessary to account for short-run effects which could otherwise violate the Gaussian assumption. Specifically, Table 3 reports the trace test statistics for all possible values of r and s1 = p − r − s2 , where r is the rank of the cointegration space and s1 is the number of I(1) trends under the assumption that the data contain linear but not quadratic trends. The 95% quantiles of the asymptotic test distributions are taken from (Rahbek et al. (1999), Table 1) and reproduced underneath the calculated test values. Starting from the most restricted hypothesis {r = 0, s1 = 0, s2 = 8}, where s2 is the number of I(2) components, and

9 The Johansen (1988, 1991) and Johansen and Juselius (1990) multivariate cointegration methodology has been extensively used in numerous econometric applications and therefore to save space we do not discuss it. 10 Gonzalo (1994) shows that the performance of the maximum likelihood estimator of the cointegrating vectors is little affected by non-normal errors. Lee and Tse (1996) have shown similar results when conditional heteroscedasticity is present.

Table 3 Testing the rank of the I(2) and the I(1) Model p–r

r

Q(s1 ∩ r/H0 )

Korean Won (Dollar) 8 0 1003.2 (441.5) 7 1 6 2 5 3 4 4 3 5 2 6 1 7 s2 8 Malaysian Ringgit (Dollar) 8 0 976.1 (441.5) 7 1 6 2 5 3 4 4 3 5 2 6 1 7 s2 8

7 865.4 (397.4) 732.7 (351.6)

7 844.3 (397.4) 813.7 (351.6)

7 888.9 (397.4) 695.2 (351.6)

801.0 (356.5) 722.5 (311.2) 697.2 (269.2)

6 697.5 (356.5) 553.3 (311.2) 523.1 (269.2)

6 763.2 (356.5) 718.9 (311.2) 679.2 (269.2)

6 724.7 (356.5) 544.0 (311.2)

688.3 (317.9) 591.9 (274.0) 535.5 (233.8) 508.1 (198.2)

5 577.9 (317.9) 443.4 (274.0) 405.6 (233.8) 341.7 (198.2)

5 633.0 (317.9) 575.8 (274.0) 510.2 (233.8) 473.4 (198.2)

5 599.8 (317.9) 448.9 (274.0)

577.1 (283.3) 470.8 (241.2) 411.3 (202.8) 355.8 (167.9) 325.4 (137.0)

4 490.6 (283.3) 378.8 (241.2) 322.2 (202.8) 238.1 (167.9) 207.4 (137.0)

4 568.5 (283.3) 469.0 (241.2) 402.5 (202.8) 365.5 (167.9) 304.4 (137.0)

4 469.2 (283.3) 351.7 (241.2)

488.8 (252.3) 372.2 (211.6) 308.8 (174.9) 252.4 (142.2) 211.6 (113.0) 211.7 (86.7)

3 386.6 (252.3) 284.2 (211.6) 239.6 (174.9) 169.2 (142.2) 117.8 (113.0) 125.2 (86.7)

3 486.8 (252.3) 366.0 (211.6) 305.7 (174.9) 248.2 (142.2) 206.5 (113.0) 197.6 (86.7)

3 389.8 (252.3) 283.9 (211.6)

415.5 (225.6) 311.9 (186.1) 217.1 (151.3) 164.2 (119.8) 129.9 (92.2) 110.9 (68.2) 96.3 (47.6) 2 345.8 (225.6) 235.2 (186.1) 199.1 (151.3) 130.4 (119.8) 100.2 (92.2) 74.8 (68.2) 86.8 (47.6) 2 421.4 (225.6) 312.7 (186.1) 226.2 (151.3) 159.9 (119.8) 121.2 (92.2) 117.2 (68.2) 99.4 (47.6) 2 325.7 (225.6) 223.5 (186.1)

365.2 (202.2) 255.0 (164.6) 154.9 (130.9) 110.1 (101.5) 83.9 (75.3) 59.7 (53.2) 40.4 (34.4) 43.3 (19.9) 1

340.0 (182.6) 233.6 (146.8) 145.2 (115.4) 90.5 (87.2) 69.0 (62.8) 48.3 (42.7) 26.5 (25.4) 15.5 (12.5)

289.2 (202.2) 190.2 (164.6) 145.5 (130.9) 114.5 (101.5) 88.7 (75.3) 63.9 (53.2) 43.5 (34.4) 30.5 (19.9) 1

259.3 (182.6) 171.1 (146.8) 127.0 (115.4) 96.2 (87.2) 69.4 (62.8) 55.4 (42.7) 28.8 (25.4) 19.2 (12.5)

365.4 (202.2) 256.1 (164.6) 155.6 (130.9) 105.6 (101.5) 85.3 (75.3) 59.1 (53.2) 41.2 (34.4) 44.1 (19.9) 1

344.0 (182.6) 240.2 (146.8) 149.1 (115.4) 97.4 (87.2) 70.2 (62.8) 49.3 (42.7) 31.1 (25.4) 18.3 (12.5)

281.1 (202.2) 185.2 (164.6)

238.3 (182.6) 167.6 (146.8)

253

Philippine Peso (Dollar) 8 0 1009.3 (441.5) 7 1

910.3 (397.4) 880.3 (351.6)

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

Indonesian Rupiah (Dollar) 8 0 1134.1 (441.5) 7 1 6 2 5 3 4 4 3 5 2 6 1 7 s2 8

Qr

254

Table 3 (Continued) p–r

Q(s1 ∩ r/H0 )

Qr

2 3 4 5 6 7

522.6 (269.2)

8

Singaporean Dollar (Dollar) 8 0 1022.1 (441.5) 7 1 6 2 5 3 4 4 3 5 2 6 1 7 s2 8 Thai Baht (Dollar) 8 0 993.2 (441.5) 7 1 6 2 5 3 4 4 3 5 2 6 1 7 s2 8

7 858.5 (397.4) 826.2 (351.6)

7 876.6 (397.4) 815.2 (351.6)

7

6 789.0 (356.5) 724.7 (311.2) 663.9 (269.2)

6 781.7 (356.5) 719.9 (311.2) 618.7 (269.2)

6

395.8 (233.8) 345.6 (198.2)

5 656.6 (317.9) 539.0 (274.0) 508.6 (233.8) 482.6 (198.2)

5 690.4 (317.9) 593.5 (274.0) 513.8 (233.8) 443.4 (198.2)

5

374.6 (202.8) 235.9 (167.9) 196.1 (137.0)

4 590.9 (283.3) 439.8 (241.2) 399.2 (202.8) 343.4 (167.9) 295.6 (137.0)

4 557.3 (283.3) 458.7 (241.2) 386.9 (202.8) 318.3 (167.9) 285.1 (137.0)

4

245.8 (174.9) 166.9 (142.2) 113.5 (113.0) 119.6 (86.7)

3 468.3 (252.3) 319.8 (211.6) 267.9 (174.9) 270.2 (142.2) 189.1 (113.0) 151.7 (86.7)

3 469.2 (252.3) 389.1 (211.6) 279.4 (174.9) 285.5 (142.2) 211.2 (113.0) 171.4 (86.7)

3

187.0 (151.3) 133.6 (119.8) 103.2 (92.2) 81.7 (68.2) 73.2 (47.6) 2 421.9 (225.6) 284.5 (186.1) 210.9 (151.3) 158.3 (119.8) 122.9 (92.2) 92.7 (68.2) 92.2 (47.6) 2 379.3 (225.6) 315.6 (186.1) 205.2 (151.3) 158.0 (119.8) 115.9 (92.2) 98.3 (68.2) 75.4 (47.6) 2

149.9 (130.9) 110.9 (101.5) 86.8 (75.3) 64.0 (53.2) 46.8 (34.4) 24.5 (19.9) 1

129.6 (115.4) 98.8 (87.2) 67.3 (62.8) 54.8 (42.7) 30.1 (25.4) 17.5 (12.5)

342.5 (202.2) 243.9 (164.6) 159.8 (130.9) 112.5 (101.5) 85.9 (75.3) 68.5 (53.2) 45.7 (34.4) 30.7 (19.9) 1

302.8 (182.6) 205.9 (146.8) 141.7 (115.4) 102.2 (87.2) 68.1 (62.8) 52.9 (42.7) 31.3 (25.4) 19.5 (12.5)

342.3 (202.2) 253.9 (164.6) 173.5 (130.9) 111.2 (101.5) 98.1 (75.3) 66.1 (53.2) 47.9 (34.4) 35.4 (19.9) 1

295.9 (182.6) 227.5 (146.8) 131.1 (115.4) 90.8 (87.2) 76.6 (62.8) 55.5 (42.7) 39.2 (25.4) 21.1 (12.5)

Testing the joint hypothesis H(s1 ∩ r). Notes: p is the number of variables, r is the rank of the cointegration space, s1 is the number of I(1) components and s2 is the number of I(2) components. The numbers in parentheses are the 95% critical values (Rahbek et al., 1999, Table 1). For all tests a structure of eleven lags for each forecasting equation was chosen according to a likelihood ratio test, corrected for the degrees of freedom (Sims, 1980) and the Ljung-Box Q statistic for detecting serial correlation in the residuals of the equations of the VAR. A model with an unrestricted constant in the VAR equation and a deterministic trend restricted in the cointegrating vector is estimated for all cases according to the k Johansen (1992) testing methodology. A small sample adjustment has been made in the Trace test statistics, Qr , for the I(1) analysis −2 ln Q = −(T − kp) i=r ln(1 − λˆ i ) 0+1 as suggested by Reimers (1992).

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

6 5 4 3 2 1 s2

r

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

255

Table 4 First difference models: marginal significance level of F-statistics Country

Unanticipated official exchange rate

Anticipated official exchange rate

Indonesia Korea Malaysia Philippines Singapore Thailand

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

Note: each of the reported marginal significance levels is calculated from a black market premium equation estimated with a lag length of 12 months. The null hypothesis is that unanticipated (anticipated) changes in the official exchange rate do not affect the black market premium. Corrected F-statistics have been calculated according to Hoffman et al. (1984).

testing less and less restricted hypotheses, it is shown that all I(2) hypotheses can be rejected at the 5% level for all six Pacific Basin countries. Furthermore, no evidence of cointegration was detected for each case.11 Table 4 reports the results for the growth rate models for the six Pacific Basin countries. The analysis of these results reveals that unanticipated shocks to the official exchange rate influence the black market premium in Indonesia, Korea, Malaysia, the Philippines, Singapore and Thailand. This result is consistent with asset models which suggest that news is quickly incorporated into prices, such as the black market premium. More interestingly, the results show that anticipated shocks to the official exchange rate influence the black market premium in Indonesia, Korea, Malaysia, the Philippines, Singapore and Thailand. Given the prediction of the portfolio balance model that the premium responds to expected return differentials, then this model seems to hold for all six Pacific Basin foreign currency markets. This result implies that given that fluctuations in expected returns for domestic and foreign assets are significant, economic agents reallocate their portfolios. This seems to imply that they choose a strategy that leads to the choice of assets denominated in both currencies. Therefore, we argue that in these economies anticipated depreciation of the official exchange rate has led to the determination of the parallel market premium. Therefore, economic agents’ preferences appear to be guided by changes in expected returns as measured by anticipated official exchange rate changes and lead portfolio managers to use the black market in these countries to hedge against foreign exchange risk. 5. Concluding remarks Dual exchange rate systems provide an important framework to study economic performance under alternative exchange rate regimes. Recently, Reinhart and Rogoff (2004) have used data on market-determined parallel exchange rates that led to a re-classification of historical exchange rate regimes over the Bretton Woods period. Furthermore, it is argued that the floating dual or parallel rate is not only a far better barometer of monetary policy than is the official exchange

11 It is possible that a feedback from the black market to the official exchange rate may exist. This is the issue of market efficiency and in that case the models could be misspecified. However, we only focus on the role of expectations. A misspecification could possibly occur under the existence of cointegration between the variables, in which case, the error correction term should be included in the forecasting equation. As our results show, no evidence of cointegration was detected and therefore our specification is correctly specified.

256

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

rate, but also that it is often the most economically meaningful rate. During the time of the official peg, it was very frequently the case that the parallel rates were used as a form of “back door” floating, although accompanied by exchange controls. Finally, it has been argued that expectations of future official rate changes play an important role in driving the premium. This paper has focused on the official and black market for foreign currency in six Pacific basin countries during the period 1974–1998. Our purpose was to analyze the role of agents’ expectations about the government policy on the official market for foreign currency. The portfolio balance model of the exchange rate determination provided the theoretical framework to explain the movements of the black market premium in the parallel market for US dollars for the period under investigation. The econometric analysis was based on the two-step methodology suggested by Hoffman et al. (1984) which allowed us to test the impact of unanticipated and anticipated shocks to the official exchange rate on the black market premium using the corrected F-statistics which lead to making valid statistical inference in the presence of generated regressors. The main finding of our analysis is that both the anticipated and unanticipated shocks to the official exchange rate had an impact on the black market premium in Indonesia, Korea, Malaysia, the Philippines, Singapore and Thailand. This result suggests that the monetary and portfolio balance model is the appropriate framework for analyzing the behaviour of the black market premium for these six Asian foreign currency markets. Economic agents with portfolios which included the currencies of these countries appeared to be sensitive to changes in anticipated returns of foreign exchange. Our analysis contributes to the literature of dual exchange rate regimes since it focuses on the important role expectations about the official exchange rate play in the determination of the parallel market premium. Given that similar patterns are expected to be found in most dual markets for foreign currency that operate in Latin America, Central and Eastern Europe as well as Africa, the policy conclusions drawn from our analysis, with respect to the conduct of monetary and fiscal policy and the efficiency of capital controls, can also be expected to apply to these dual markets as well. More specifically, the modelling of expectations about government policy with respect to the level of the official exchange rate for these economies will reveal substantial insights into the workings of the two markets and that will have an impact on the decisions of the policy makers in these countries, as well as of the portfolio managers that are active in emerging markets. Changes in expectations about the determination of the black market premium will make agents act in a different way, as Lucas (1976) predicts, and agents will react to monetary policy and fiscal policy changes under this new perspective. Tied to the propositions put forward by Reinhart and Rogoff (2004) and Cheung and Lai (2005), the results reported in this paper make our analysis interesting, and one that brings to the forefront new and important evidence about the workings of foreign currency markets. Although we do not discuss cases from other regions that have also experienced or still have active dual markets our arguments based on the stylized facts offered by the literature about these markets in different countries clearly imply that a similar role of expectations as a determinant of the black market premium will be present in most of the dual markets in Latin America, Central and Eastern Europe, and Africa, with similar policy implications as for the case of the Southeast Asian countries. Acknowledgements Part of this paper was written while the second author was a Visiting Fellow at the Department of Economics, European University Institute, San Domenico di Fiesole, Italy. The hospitality of EUI and the Robert Schumann Centre is gratefully acknowledged. The second author acknowledges

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

257

generous financial support from EUI and the Research Committee of the University of Crete under research grants #2016 and #2030. An earlier version was presented at the 56th European Meeting of the Econometric Society, Lausanne, 25–29 August 2001; the 9th International Conference on Macroeconomic Analysis and International Finance, University of Crete, Rethymno, 26–28 May 2005; the International Conference on Policy Modeling-EcoMod 2005, Istanbul, 29 June–2 July 2005; the 37th Money, Macro and Finance Research Group Annual Conference, University of Crete, Rethymno, 1–3 September 2005; and workshops at Athens University of Economics and Business, European University Institute, University of Crete and University of Cyprus. We would also like to thank Michael Artis, Dimitris Georgoutsos, Costas Karfakis, Soren Johansen, Katarina Juselius and Dimitris Moschos for many helpful comments and discussions. Finally, we thank an anonymous referee for helpful comments and suggestions which improved the paper substantially. The usual caveat applies.

References Agenor, P.R., 1991. A Monetary model of the parallel market for foreign exchange. J. Econ. Stud. 18, 4–18. Agenor, P.R., 1992. Parallel currency markets in developing countries: theory, evidence and policy implications. Essays in International Finance, Princeton University, no. 188. Agenor, P.R., Taylor, M.P., 1993. Analyzing credibility in high-inflation countries: a new approach. Econ. J. 103, 329–336. Bahmani-Oskooee, M., 1993. Black market exchange rates versus official exchange rates in testing purchasing power parity: an examination of the Iranian rial. Appl. Econ. 25, 465–472. Barro, R.J., 1977. Unanticipated money growth and unemployment in the United States. Am. Econ. Rev. 67, 101–115. Blejer, M.L., 1978. Exchange rate restrictions and the monetary approach to the exchange rate. In: Frankel, J.A., Johnson, H.G. (Eds.), The Economics of the Exchange Rates: Selected Studies. Addison-Wesley, Reading, pp. 117–128. Branson, W.H., de Macedo, J.B., 1989. Smuggler’s blues at the Central Bank: lessons from Sudan. In: Calvo, G., Findlay, R., Kouri, P., de Macedo, J.B. (Eds.), Debt, Stabilization and Development: Essays in Memory of Carlos Diaz-Alejandro. Blackwell, Oxford, pp. 191–204. Cheung, Y.-W., Lai, K.S., 2005. Nominal exchange rate flexibility and real exchange rate adjustment: evidence from dual exchange rates in developing countries. Working Paper no. 1512, CESifo. Culbertson, W., 1975. Purchasing power parity and black market exchange rates. Econ. Inquiry 13, 287–296. de Macedo, J.B., 1985. Macroeconomic policy under currency inconvertibility. Working Paper no. 1571, National Bureau of Economic Research, Cambridge: Mass. de Macedo, J.B., 1987. Currency inconvertibility, trade taxes and smuggling. J. Dev. Econ. 27, 109–125. Direction of Trade Statistics, various issues. Washington, DC: International Monetary Fund. Doornik, J.A., Hansen, H., 1994. A practical test for univariate and multivariate normality. Discussion paper, Nuffield College. Dornbusch, R., Dantas, D.V., Pechman, C., Rocha, R., Simoes, D., 1983. The black market for dollars in Brazil. Quart. J. Econ. 98, 25–40. Edwards, S., 1988. Real and monetary determinants of real exchange rate behaviour: theory and evidence from developing countries. J. Dev. Econ. 29, 311–341. Edwards, S., 1989. Exchange controls, devaluations, and real exchange rates: the Latin American experience. Econ. Dev. Cult. Change 37, 231–254. Edwards, S., 1994. The political economy of inflation and stabilization in developing countries. Econ. Dev. Cult. Change 42, 235–266. Edwards, S., 1998. Capital inflows into Latin America: a stop-go story. NBER Working Paper 6441. Edwards, S., 1999. How effective are capital controls? J. Econ. Perspect. 13, 65–84. Edwards, S., Montiel, P.J., 1989. Devaluation crises and the macroeconomic consequences of post-poned adjustment in developing countries. Int. Monet. Fund Staff Pap. 36, 875–904. Elliott, G., 1999. Efficient tests for a unit root when the initial observation is drawn from its unconditional distribution. Int. Econ. Rev. 44, 767–784. Elliott, G., Rothenberg, J.T., Stock, J.H., 1996. Efficient tests for an autoregressive unit root. Econometrica 64, 813–836. Frenkel, M., 1990. Exchange rate dynamics in black markets. J. Econ. 51, 159–176.

258

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

Gervais, J.-P., Larue, B., 2001. Do reductions in black market exchange rate premia cause inflation? Empirical Econ. 26, 525–551. Ghei, N., Kamin, S.B., 1999. The use of the parallel market rate as a guide to setting the official exchange rate. In: Hinkle, L.E., Montiel, P.J. (Eds.), Exchange Rate Misalignment: Concepts and Measurement for Developing Countries. Oxford University Press, Oxford, pp. 497–538. Gonzalo, J., 1994. Comparison of five alternative methods of estimating long-run equilibrium relationships. J. Econom. 60, 203–233. Hoffman, D.L., Low, S.A., Schlagenhauf, D.E., 1984. Tests of rationality, neutrality and market efficiency: a Monte Carlo analysis of alternative test statistics. J. Monet. Econ. 14, 339–363. International Financial Statistics, various issues. Washington, DC: International Monetary Fund. Johansen, S., 1988. Statistical analysis of cointegrating vectors. J. Econ. Dynam. Control 12, 231–254. Johansen, S., 1991. Estimation and hypothesis testing of cointegration vectors in gaussian vector autoregressive models. Econometrica 59, 1180–1551. Johansen, S., 1992. A representation of vector autoregressive processes integrated of order 2. Econom. Theor. 8, 188–202. Johansen, S., 1995. A statistical analysis of cointegration for I(2) variables. Econom. Theor. 11, 25–29. Johansen, S., 1997. A likelihood analysis of the I(2) model. Scand. J. Stat. 24, 433–462. Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference on cointegration—with applications to the demand for money. Oxford Bull. Econ. Stat. 52, 169–210. Kamin, S., 1993. Devaluation, exchange controls, and black markets for foreign exchange in developing countries. J. Dev. Econ. 40, 151–169. Kiguel, M., Lizordo, J.S., O’Connell, S.A., 1997. Parallel Exchange Rates in Developing Countries. Saint Martin’s Press, New York. Kiguel, M., O’Connell, S.A., 1995. Parallel foreign exchange markets in developing countries: experience and policy lessons. World Bank Res. Obser., 21–52. Kouretas, G.P., Zarangas, L.P., 1998. A cointegration analysis of the official and parallel foreign exchange markets for dollars in Greece. Int. J. Financ. Econ. 3, 261–276. Kwiatkowski, D., Phillips, P.C.B., Schmidt, P., Shin, Y., 1992. Testing the null hypothesis of stationarity against the alternative of a unit root: how sure are we that economic time series have a unit root? J. Econom. 54, 159–178. Lee, T.-H., Tse, Y., 1996. Cointegration tests with conditional heteroscedasticity. J. Econom. 73, 401–410. Lucas, R., 1976. Econometric policy evaluation: a critique. Carnegie–Rochester Conf. Public Policy 1, 19–46. Martin, L., Panagariya, A., 1984. Smuggling, trade, and price disparity: a crime-theoretic approach. J. Int. Econ. 17, 201–218. McAleer, M., McKenzie, C.R., 1991. When are two step estimators efficient? Econom. Rev. 10, 235–252. Minskin, F., 1982. Does anticipated monetary policy matter? An econometric investigation. J. Polit. Econ. 90, 22–51. Montiel, P.J., Agenor, P.R., Ul Haque, N., 1993. Informal Financial Markets in Developing Countries. Blackwell Publishers, Oxford. Montiel, P.J., Ostry, J.D., 1994. The parallel market premium. Int. Monet. Fund Staff Pap., 117–213. Moore, M.J., Phylaktis, K., 2000. Black and official exchange rates in the Pacific Basin: some tests of dynamic behaviour. Appl. Financ. Econ. 10, 361–369. Murphy, K.M., Topel, R.H., 1985. Estimation and inference in two-step econometric models. J. Bus. Econ. Stat. 3, 370–379. Newey, W., West, K., 1994. Automatic lag selection in covariance matrix estimation. Rev. Econ. Stud. 61, 631–653. Ng, S., Perron, P., 2001. Lag length selection and the construction of unit root tests with good size and power. Econometrica 69, 1519–1554. Oxley, L., McAleer, M., 1993. Econometric issues in macroeconomic models with generated regressors. J. Econ. Surv. 7, 1–40. Pagan, A., 1984. Econometric issues in the analysis of regressions with generated regressors. Int. Econ. Rev. 25, 221–247. Pagan, A., 1986. Two stage and related estimators and their applications. Rev. Econ. Stud. 53, 517–538. Pantula, S.G., 1989. Testing for unit roots in time series data. Econom. Theor. 5, 256–271. Paruolo, P., 1996. On the determination of integration indices in I(2) systems. J. Econom. 72, 313–356. Phillips, P.C.B., Perron, P., 1988. Testing for a unit root in time series regression. Biometrica 75, 335–346. Phylaktis, K., 1991. The black market for dollars in Chile. J. Dev. Econ. 37, 155–172. Phylaktis, K., 1996. Black market for foreign currency: A survey of theoretical and empirical issues. Financ. Markets Inst. Instrum. 5, 102–124. Phylaktis, K., Kassimatis, Y., 1994. Black and official exchange rates in the Pacific Basin countries: an analysis of their long-run dynamics. Appl. Econ. 26, 399–407.

P.F. Diamandis et al. / Research in International Business and Finance 21 (2007) 238–259

259

Phylaktis, K., Manalis, G., 1995. Devaluation, exchange controls, and the black market for dollars in Greece. In: Papers and Proceedings of the XLIV Conference of the Applied Econometrics Association. Pitt, M., 1984. Smuggling and the black market for foreign currency. J. Int. Econ. 16, 243–257. Pozo, S., Wheeler, M., 1999. Expectations and the black market premium. Rev. Int. Econ. 7, 245–253. Rahbek, A.H., Konsted, C., Jorgensen, C., 1999. Trend-stationarity in the I(2) cointegration model. J. Econom. 90, 265–289. Reimers, H.-E., 1992. Comparison of tests for multivariate co-integration. Stat. Pap. 33, 335–359. Reinhart, C.M., Rogoff, K., 2004. The modern history of exchange rate arrangements: a reinterpretation. Quart. J. Econ. 114, 1–48. Sephton, P.S., 1995. Response surface estimates of the KPSS stationarity test. Econ. Lett. 47, 255–261. Sheikh, M.A., 1976. Black market for foreign exchange, capital flows and smuggling. J. Dev. Econ. 16, 243–257. Sheikh, M.A., 1989. Theory of risk, smuggling and welfare. World Dev. 17,√ 1931–1944. Shenton, L.R., Bowman, K.O., 1977. A bivariate model for the distribution of b1 and b2 . J. Am. Stat. Assoc. 72, 206–211. Sims, C.A., 1980. Macroeconomics and reality. Econometrica 48, 1–48. Van den Berg, H., Jayanetti, S.C., 1993. A novel test of the monetary approach using black market exchange rates and Johansen–Juselius cointegration method. Econ. Lett. 41, 413–418. World Currency Yearbook, various issues. Brooklyn: International Currency Analysis Inc.