Symmetric and asymmetric effects of exchange rates on money demand: empirical evidence from Vietnam

ABSTRACT

This empirical investigation aims at exploring the determinants of money demand in Vietnam by using both linear and nonlinear autoregressive distributed lag models over the period spanning from the third quarter of 2000 to the first quarter of 2018. Our findings can be summarized as follows: firstly, when the shock is symmetric (i.e. a permanent nominal appreciation of 1%), the money demand increases by 3.7% in the long term. Secondly, when the shock is asymmetric, for a permanent nominal appreciation of 1%, we observe an increase of 15.6% in the money demand. Whereas for a permanent nominal depreciation of 1%, we observe a decrease of 7.4% in the money demand. These results are consistent with symmetry tests and lead us to think that asymmetries occur mainly in the short run and are transmitted to the long run.

KEYWORDS:

• Money demand
• exchange rate
• ARDL models
• NARDL models
• dollarization

JEL CLASSIFICATION:

• C22
• E41
• F31
• F33
• F41

Acknowledgments

We would like to thank Fredj Jawadi and Naveen Kumar for their very relevant comments and remarks during the 6th International Symposium in Computational Economics and Finance which took place on the 29th-30 October 2020 in Paris. We also thank the organizers and all the participants of this conference. This work has benefitted from very valuable comments and remarks by Bertrand Koebel and from proofreading by Katharina Priedl. We are grateful to anonymous referees for useful suggestions on the literature. Sy-Hoa Ho gratefully acknowledges the financial support of the National Foundation for Science and Technology Development of Vietnam (NAFOSTED). All errors or omissions are ours.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ These measures are described in two directives issued by the SBV: the directive 11/NQ-CP of 24 February 2011, and the directive 01/CT-NHNN of 15 January 2014.

² We suppose that, at equilibrium, the money demand is equal to money supply, since the demand in excess will not be satisfied.

³ The well-known Impossible Trinity problem (i.e. an impossible combination of an autonomous monetary policy, a perfect capital mobility and a floating exchange rate) is also referred to as the Trilemma, a more neutral term.

⁴ For expected inflation, her calculations were based on Gerlach and Svensson (2003).

⁵ An alternative specification could use the trade-weighted nominal exchange rate as the explained variable. Nevertheless, we think that behaviours of economic agents are mainly driven by the nominal exchange rate vis-à-vis the U.S. in the case of the Vietnamese dong in a context of dollarization.

⁶ An increase in $NER$ indicates a depreciation of the Vietnamese dong.

⁷ The error correction form for the nonlinear model can be found in equation (2.7) of Shin, Yu, and Greenwood-Nimmo (2014). The explanatory variables are split into positive partial sums and negative partial sums in order to capture asymmetric effects.

⁸ This approach has been employed in Delatte and López-Villavicencio (2012) to investigate the asymmetric effects of exchange rate on prices in four major industrial countries. In addition, Konopczak (2019) argues that the incompleteness of exchange rate pass-through is a statistical artefact thanks to a threshold NARDL model.

⁹ There are other possibilities for the nonlinear specification as underlined by Karamelikli and Karimi (2020). In Table 4, the symmetry tests indicate that the asymmetrical effects occur in the short run and in the long run. Thus, the specifications for (i) short run symmetry/long run asymmetry and (i) short run asymmetry/long run symmetry are not presented.

¹⁰ Pal and Mitra (2016) implement an augmented version of the NARDL model with both symmetric and asymmetric regressors in an empirical investigation about the transmission from crude to oil product prices in India. Besides, they found that the use of a multiple threshold NARDL model (MTNARDL) improves precision in estimating asymmetric effects.

¹¹ Even though asymmetric error correction is not explicitly modelled by allowing regime dependency for ${\beta }_1$, we may still observe asymmetric adjustment trajectories thanks to the asymmetric dynamic multipliers. Alternatively, we could use a time-varying regression model based on the Kalman filter as in Jelassi, Trabelsi, and Turki (2017) and Sidiropoulos, Trabelsi, and Karfakis (2005).

¹² We do not include dynamic lags for the other variables for two reasons: first, our variables of interest are the positive and negative partial sums of the nominal exchange rate and, second, to preserve degrees of freedom. Nevertheless, when we introduce dynamic lags for the gross domestic product and the interest rate, these last variables are not statistically significant and do not improve the AIC or the RMSE.

¹³ Except for the first dynamic lag that has a negative sign.

¹⁴ To test auto-correlation, we use the more versatile framework for the tests of Cumby and Huizinga (1992) implemented by Baum, Schaffer, and Stillman (2007).

¹⁵ As suggested by Johansen (1995), with a simple example of a bivariate cointegrated VAR(1), a stability condition for an error-correction model is that the error-correction term must be strictly inferior to zero and strictly superior to −2. Indeed, the error-correction model can be written as an AR(1) process for the disequilibrium error.