On the Exchange Rates Volatility and Economic Policy Uncertainty Nexus: A Panel VAR Approach for Emerging Markets

Abstract

We examine the Economic Policy Uncertainty (EPU) transmission over the Exchange Rate Volatility (ERV) for 8 Emerging Market Economies (EME) using the recent panel VAR methodology of Abrigo and Love (Stata Journal 16:778–804, 2016). The econometric investigation reveals that: (a) both domestic and US-EPU shocks exert positive effects on the ERV, (b) the contribution of the US-EPU to the ERV fluctuations overcomes the own EPU’s share, (c) the ERV acts as a possible transmission channel of the US-EPU to the domestic economic activity, (d) the domestic EPU increases in response to a higher US-EPU and vice versa and (e) the latter is surprisingly and markedly sensitive to EME macroeconomic conditions. Our findings are robust to different sensitivity analyses, provide novel insights into EPU international spillovers, and have interesting policy implications for EME decisions makers and investors.

Introduction

The predictive power of fundamentals on the exchange rate fluctuations remains one of the most discussed puzzles in international economics. While their long-run effects on currency movements are confirmed (Beckmann and Czudaj 2017), understanding their short-run impacts have much less empirical success (Chen and Chou 2015).¹ Nonetheless, the short-run exchange rate dynamics are particularly affected by transitory shocks including the economic policy uncertainty (EPU hereafter) (Bartsch 2019). At a general level, the applied literature provides evidence in support of EPU’s harmful effects on economic activity (Baker et al. 2016; Bloom 2009) and asset prices (Pástor and Veronesi 2012; Brogaard and Detzel 2015; Arouri et al. 2016).² Within this growing literature, the foreign exchange market (FOREX) exposure to EPU shocks is, to date, explored only by a handful of studies. Firstly, they predict that a surge in the EPU coincides with a national currency depreciation (Abid 2020; Nilavongse et al. 2020) as well as a rise of the exchange rate volatility (ERV hereafter) (Chen et al. 2020; Bartsch 2019; Christou et al. 2018; Balcilar et al. 2016). Secondly, exchange rate movements are found to be more sensitive to the domestic rather than the foreign EPU (Nilavongse et al. 2020). From a technical point of view, it is clear that this branch of the literature has sustainable interest to time series econometric modelling mainly applied to advanced economies (Cf. Bartsch 2019; Nilavongse et al. 2020) or some major emerging markets such as China (Chen et al. 2020). Notable few exceptions are Balcilar et al. (2016) and Christou et al. (2018) who consider a mixture of developed and developing countries.

To complement this literature, our study primarily builds on the body of the research pioneered by Bloom (2009) in investigating the macroeconomic effects of uncertainty. Our main purpose is to examine how the ERV responds to both domestic and foreign EPU in Emerging Market Economies (EME). We contribute to the current debate in several ways. The first novelty belongs to the exclusive focus on EME as a case study.³ These countries (i) display a higher ERV especially those having adopted floating exchange regimes (ii) face higher levels of uncertainty comparing to advanced economies (Bloom 2009) and (iii) are extremely vulnerable to shocks spillovers stemming from developed countries (Bhattarai et al. 2020).

In addition, this paper aims at revealing the extent to which considering cross sectional dependence between countries can affect the sensitivity of the ERV to the EPU. Using the recent Panel VAR (PVAR hereafter) methodology of Abrigo and Love (2016), our paper is not only the first to use a panel framework (to the best of our knowledge) but also one of the rare studies relying on VAR models to deal with ERV-EPU nexus.⁴ Numerous reasons make this methodology a well suited tool for that purpose. First, it allows to account for the variables’ interactions and to capture market interdependences. So, why such issues could be of interest? On the one hand, in a globalized economy, a shock from one country can easily spill to other countries/markets. As it is also the case of the EPU international transmission (Jiang et al. 2019 among others), a PVAR set might help better understand such spillovers given its ability to account for international transmission of shocks (Canova and Ciccarelli 2012). On the other hand, beyond the impacts running from the EPU to the ERV, the literature provides significant evidence on the ability of the EPU in a given country to predict the path of EPUs abroad (Gupta and Sun 2020; Jiang et al. 2019).⁵ Furthermore, we assume that higher currency fluctuations lead governments to undertake macroeconomic policy actions. This policy decision-making process is expected to be a source of policy uncertainty. Accordingly, we conjuncture that a VAR setting might potentially (i) provide a novel insight regarding the endogenous linkage among these variables and (ii) be more relevant than single-equation models which could suffer from serious endogeneity biases resulting in possible misleading results.⁶ Second, our procedure interestingly allows simulating the Impulse Response Functions (IRFs) conventionally used to analyze the size and the sign of shocks propagation. Third, this methodology offers the possibility of appreciating the contribution of each shock to the variance of the ERV using the Forecast Error Variance Decomposition (FEVD).

In what follows, "Data and Econometric Framework" section describes the data and the econometric framework. "Findings and Discussion" section reports the main findings. "Robustness Checks" deals with robustness issues. Finally, "Conclusion" offers some concluding remarks.

Data and Econometric Framework

Our study is conducted on a monthly basis for 8 EME: Singapore, China, Chile, India, Korea, Russia, Brazil and Mexico and includes 4 variables: the ERV, the domestic and foreign EPU, and the domestic output. While the sample covers the period from January 2003 through December 2018, the starting date is dictated by the data availability in order to ensure a balanced panel structure. Our sample period interestingly englobes turmoil phases that are likely to increase the EPU: the collapse of Lehman Brothers in 2008, the global financial crisis (2008–2009), the US loss of the AAA sovereign rating for the first time of its history (summer 2011), the mid-2014 oil price decline and the Sino-American trade conflict (2018).

The exchange rates examined are the daily spot rates of the domestic currencies against the US dollar and are retrieved from DataStream.⁷ To account for the ERV for each country, the monthly standard deviation of the corresponding exchange rate is computed. The domestic and foreign EPU’s are the Baker et al. (2016)’s News-based indexes obtained from www.policyuncertainty.com. This index is a monthly coverage of Newspapers text search of articles including at least one term related to three categories (i) economy/economic, (ii) uncertainty/uncertain and (iii) policy (e.g. legislation, deficit, regulation etc.). As a proxy for the foreign EPU, we use the US-EPU since it is a benchmark of the international policy uncertainty (Das and Kumar 2018).⁸ Furthermore, as a measure of output, we use the Industrial Production Index (IPI) drawn either from DataStream or from the OCDE database. This choice is mainly due to the unavailability of the Gross Domestic Product (GDP) on a monthly basis. At a general level, the use of the IPI is not a challenging issue as it remains “a very good proxy” of real GDP (Fontaine et al. 2018). Finally, all our series are transformed in logarithm form.

In Eq. 1, we describe the estimated PVAR that associates the panel data approach allowing unobservable individual heterogeneity to VAR models assuming the endogeneity of all the variables (Love and Zicchino 2006).

$$ Z_{{it}} = U_{{i~}} + \mathop \sum \limits_{{j = 1}}^{p} A_{j} Z_{{it - j}} + ~\mu _{{it~}} $$

(1)

where ${Z}_{it}$ is a 4*1 vector of endogenous variables discussed above and A is a 4*4 coefficient matrix. The optimal lag $p$ is chosen to minimize information criteria. ${U}_{i}$ are vectors of country fixed effects that account for individual heterogeneity in the levels of the variables, and ${\mu }_{it}$ are independently and identically distributed disturbance terms. i and t denote temporal (month) and individual (country) dimensions, respectively.

One of the advantages of using the PVAR methodology is to explicitly introduce fixed effects which means that (i) each country in the sample can have a specific level of each variable used and (ii) the fixed effect can potentially capture other invariant factors (e.g. exchange rate regimes, country size, financial regulation etc.). However, estimation biases may occur since the fixed effects are likely to be correlated with regressors that contain lagged dependent variables (Love and Zicchino 2006). In such a circumstance, it is necessary to eliminate the fixed effects based on the “Helmert” method, a commonly used tool for that purpose.⁹ In doing so, we preserve the orthogonality between the transformed variables and the lagged regressors that are used to instrument the VAR dynamics. Finally, the estimation is conducted using the Generalized Method of Moments (GMM). Then, we check the stability condition via modules of each eigenvalues of the estimated models. The results reported in Fig. 3 in the Appendices confirm that all eigenvalues are inside the unit circle. Consequently, the estimated PVAR specification fully satisfies the stability condition.

Moreover, an important advantage of the PVAR models is to allow producing the orthogonalized IRFs. These functions capture the effects of one variable to another endogenous variable when keeping the other variables constant (this is what is known as an orthogonal shock). Once all the coefficients of the PVAR model have been estimated, the confidence bands of the orthogonalized IRFs are computed via Monte Carlo simulations as initially developed by Love and Zicchino (2006) and extended in Abrigo and Love (2016). To this end, we use the variance–covariance matrix and the estimated coefficients from the PVAR to randomly build a draw of the VAR coefficients. Hence, the 5th and 95th percentiles are generated by replicating this procedure 1000 times in such a way the confidence intervals of the IRFs are obtained.

Nevertheless, the variance–covariance matrix is unlikely to be diagonal (Love and Zicchino 2006). Accordingly, preserving the orthogonality of the residuals is essential in order to isolate shocks to one of the variables in the VAR system. In this case, a usual convention is to adopt a Cholesky decomposition of the variance–covariance matrix of the residuals i.e. transforming the VAR in a recursive form for an identification purpose. The Cholesky identification assumes that the variables enter the VAR system according to an ascending degree of endogeneity. Specifically, when a variable x appears earlier in the system and a variable y appears later, this means that x is exogenous with respect to y in the short run (Love and Zicchino 2006).

Our identification assumption of structural shocks is based on a Cholesky decomposition that closely follows the bulk of the literature.¹⁰ We assume that the bloc of domestic variables can be affected by the US-EPU both contemporaneously and with a one-period delay. However, the US-EPU is able to react to domestic variables only with a lag. Within the bloc of domestic variables, the output is on the one hand assumed to exert an instantaneous effect on the ERV and the EPU as well as with a lag. On the other hand, it reacts to both variables only with a one-period delay. In order to constraint the exposure of the domestic EPU to the macroeconomic innovations, it is ordered last in the VAR (Colombo 2013). To briefly sum up, the following ordering of the variables is retained (US-EPU, domestic output, ERV, and domestic EPU).

Findings and Discussion

Prior to the estimation we have undertaken a preliminary analysis. The corresponding results are reported in the Table 1. Firstly, we conduct the Levin et al. (2002) 1st generation unit root test. As can be seen in Table 1, the results indicate the stationarity of all the variables used here. Second, we implement the Pesaran (2004) CD test in order to check for the cross-sectional dependence. The latter means that the countries forming the panel are expected to respond not only to their idiosyncratic shocks but also to common shocks across other panel members. We particularly account for this property given that panel data specifications are likely to present such a dependence that arises when error terms contain unobservable components. Our results clearly support the rejection of the null hypothesis of no cross-sectional dependence. This finding interestingly implies that panel members exhibit a strong exposition to common shocks. Hence, they might probably display similar macroeconomic policies features. Finally, our conclusion supports the adequacy of using a panel data framework in our study.

Table 1 Preliminary analysis

The previous conclusion conducted us to check for the presence of unit roots relying on the 2nd generation panel unit root test of Pesaran (2007). Assuming cross-sectional dependence, this test highlights again the stationarity of all the considered variables.¹¹

Moving now to the PVAR framework, two lags are selected as they were found to be optimal according to the selected information criteria. We report the corresponding results in Table 2.¹²

Table 2 Optimal lag selection

Once the estimated coefficients of the PVAR model obtained, the orthogonalized IRFs are generated. The Fig. 1 displays the IRFs that particularly plot the US-EPU spillover to EME. Our findings highlight that a positive US-EPU shock foreshadows domestic EPU increase.

Fig. 1

US-EPU shock spillover to EME. This figure plots the orthogonalized impulse response of all the endogenous variables to a one standard deviation shock of the US-EPU. The red line presents the response of each variable. The bluish-gray boxes are the 95% confidence intervals constructed via Monte-Carlo simulations with 1000 replications. Periods are months (Color figure online)

It is argued that higher EPU induces more revision in expectations of the macroeconomic fundamentals which leads exchange rates to be subject to strong volatility (Krol 2014). In the particular context of EMEs, Bhattarai et al. (2020) explain that the US uncertainty is particularly perceived by EME policy makers as a major source of economic forecasts revisions and the volatility increase of the international capital flows. Consequently, exchange rate volatility in EMEs may naturally increase.¹³

Furthermore, we find that an unanticipated increase of the US-EPU has no significant effect on the domestic output. Nevertheless, our evidence interestingly supports significant indirect effects. Firstly, we find that an ERV’s increase significantly translates into an output contraction.¹⁴ Secondly, combining this result with the observed positive US-EPU effects on the ERV, we emphasize the ERV role as a transmission mechanism of the negative US-EPU spillovers to the EME economic outlook. Accordingly, we conjecture that countercyclical current accounts and exports of EME (Bhattarai et al. 2020) could be more complicated with higher ERV induced by the foreign EPU. Such an effect leads to an aggregate supply decrease i.e. an output contraction. Overall, we support the role of (i) the US economy as a net transmitter of policy changes (Jiang et al. 2019) and (ii) the US-EPU as negative foreign aggregate demand disturbances. Finally, the results we obtained show that the domestic EPU is positively associated to the US-EPU positive shock. Hence, our evidence gives supports to the empirical literature on the international EPUs interdependence.

Let us now discuss the domestic EPU shock effects on all variables (Cf. Fig. 2). Our findings show a significant rise of the ERV following an unanticipated positive change of the domestic EPU, which is consistent with previous works (Chen et al. 2020; Bartsch 2019; Zhou et al. 2020). As explained above, it is recognized that currencies movements are altered through expectations about fundamentals (Beckmann and Czudaj 2017) and economic policies changes (Alesina et al. 1997). If the latter are significantly exposed to the EPU, currencies are expected to be highly volatile. While the effects running from EPU to ERV are confirmed in the empirical literature, the extent to which ERV can predict the path of the EPU is not clear enough. In this regard, we interestingly provide novelty by showing that an unanticipated increase of the ERV generates a positive change in domestic EPU (Cf. Fig. 4 in Appendices).¹⁵ A candidate explanation is that higher exchange rate volatility is one of the sources of the macroeconomic instability especially in small open economies. Then, a better monetary and fiscal policies’ coordination can help attain policy goals of macroeconomic stability. This is known as the monetary and fiscal policy mix which means that central banks and fiscal policy makers are expected to synergize in the conception, the implementation, and the monitoring of the macroeconomic policies. Such collaboration is also expected to revise the macroeconomic policies and to continuously share in public the policy orientations. Nevertheless, higher economic uncertainty prevents authorities from making suitable decisions. In such unpredictable environment, the macroeconomic policy decision making process slowed down until uncertainty is dispelled.¹⁶ This misalignment of the macroeconomic policies can, in turn, translate into a high EPU as it is a natural product of such government policy making process (Pástor and Veronesi 2012). To sum up, we suggest here that domestic EPU and ERV relationship is bidirectional in EME.

Fig. 2

Effects of the local EPU shock in EME. This figure plots the orthogonalized impulse response of all the endogenous variables to one standard deviation shock of the domestic EPU. The red line presents the response of each variable. The bluish-gray boxes are the 95% confidence intervals constructed via Monte-Carlo simulations with 1000 replications. Periods are months (Color figure online)

Moreover, we find no evidence of a significant output response to the domestic EPU shock. First, we conjecture that such puzzling result may be due to the selected lag structure.¹⁷ Second, the sample size might play a role in this observation since the period of the study is constrained by the availability of the EPU index. Finally, another possible explanation relates to the composition of the industrial production. Given that we rely on monthly data, we use the industrial production index as a proxy of the output instead of GDP, not available at this frequency. Nevertheless, our countries might feature different economic compositions where industrial production does not have the same proportion of the economy in each country.

Our evidence is consistent with the recent literature.¹⁸ For instance, Baker et al. (2016) suggest that the short-run impact of the EPU on output is expected to be smaller. On the one hand, they show a negative but not statistically significant impact of the domestic EPU on output. On the other hand, they highlight that firms in policy exposed sectors experience drops in investment and employment growth rates in response to higher EPU. However, the authors conclude that the association of the EPU to the growth rate of these firms is muted in the short run.

Interestingly and perhaps surprisingly, we show that the US-EPU is positively responsive to the EME-EPU. Thus, we provide novel insights into EPU international spillovers and take a step further from studies supporting the prominent influence of the US on EME (Bhattarai et al. 2020).

Hence, we suggest that the US economy is not immune to a negative international policy shock. Even if the US is generally perceived as a net exporter of EPU (Klößner and Sekkel 2014), we show that it also appears as an importer of such an uncertainty. Therefore, we suggest an important role of the EME as an international policy uncertainty exporter. This can particularly make sense given the Sino-American trade tensions, for instance. Finally, we find that a favorable economic outlook in EME (an increase in output) results in a higher US-EPU.

On one side, it is observed, during the recent years, that the emerging markets’ influence on global economic growth, global trade and advanced economies is continuously growing. For example, China and India weight heavily on the world trade and they became increasingly an important player in the global economy. Furthermore, given the economic weight of the BRICS, these countries are gaining greater power to influence the World economic policy.

On another side, it is widely believed that the economic growth is positively correlated to stock returns. If economic growth in emerging markets is good for stockholders, investors can gain through holding assets in countries where economic activity is stable, and the growth prospects are strong. In these circumstances, emerging markets might be a privileged destination of foreign investments. Such an economic success is likely to influence the global competitiveness of developed countries and particularly the US. Hence, US decision makers may regard favorable EME conjuncture as valuable information that may alter their economic decisions. Consequently, the US-EPU would naturally increase.¹⁹

From the FEVD analysis in Table 3, we clearly observe that domestic and US-EPUs shocks explain nearly 10% of the ERV variance and their shares remain unchanged along all forecast horizons.

Table 3 Forecast error variance decomposition results

Our result highlights the relevance of the US-EPU over the bloc of domestic variables in explaining the ERV. The US-EPU contribution is twice higher than the share of the domestic EPU across all horizons. Moreover, the share of US-EPU innovations is greater than the contribution of the domestic output (6.9% against 0.4%) in the short run. Such a result strengthens our evidence that alongside domestic variables, the US-EPU is an important influential factor of EME currencies.²⁰

Contrary to previous papers but in line with our IRFs, we show that the US-EPU contribution to the output fluctuation is surprisingly weak. Furthermore, our FEVD highlights that the short-run domestic EPU fluctuation is driven to a great extent (nearly 25%) by US-EPU (18.9%) and output shocks (5.9%). The contribution of US-EPU (output) shocks slightly (considerably) decreases (increases) with the forecast horizon.²¹ Specifically, the role of the US-EPU (output) shocks is more pronounced in the short (long) run.

What can explain the prominent role of the US-EPU especially in the short run? As previously discussed, the US uncertainty plays an important role in revising economic expectations by EME policy makers. Because of the ubiquity of television and social media, natural cycles of the EPU are short. Reasonably economic expectations are quickly influenced since EPU information is extracted from daily newspapers (Bartsch 2019). Second, even though policy uncertainty can also have a long run effect on economic activity, political events-driven uncertainty (elections, for example) affects fundamentals’ expectations particularly in the short-run (Beckmann and Czudaj 2017). Naturally, it is expected that the US-EPU’ impacts on EME are greater in the short-run.

Interestingly, we also observe that the fraction of the EME output in explaining the US-EPU fluctuations rises along forecast horizons to achieve one third in the long-run. This not only echoes again our IRFs simulations but also gives support to our conclusion regarding the EME economic outlook as a source of US economic changes.

There is a possible explanation of the increasing effect of the output on the US-EPU along the forecast horizon. When the economic environment is unpredictable, governments undertake policy actions to reduce uncertainty. However, the complexity of the policy decision-making process raises concerns regarding the efficiency of policy actions as to achieve the expected outcomes. Because the latter, in the best-case, scenario, take time to be attained, this effect further amplifies the EPU.

Robustness Checks

To investigate the robustness of our findings, we have undertaken several sensitivity analyses. Specifically, we check if our results are altered if we (i) re-estimate IRFs based on an alternative lag structure, (ii) use the Global EPU (GEPU) index as an alternative proxy of the foreign EPU²² and (iii) cut Singapore and China out of our sample since they have adopted different exchange regimes compared to the other countries.²³ Our sensitivity findings confirm that the overall dynamics of our PVAR are immune to all modifications.²⁴ Therefore, we interestingly provide strong and robust evidence.

Conclusion

In this paper we use the recent panel VAR methodology of Abrigo and Love (2016) that accounts in particular for variables’ interaction, and cross-country lagged interdependencies to investigate the Exchange Rate Volatility (ERV) response to Economic Policy Uncertainty (EPU) shocks, which is, to our best knowledge the first attempt in this context. Our results suggest that the ERV in EME is sensitive to both domestic and foreign EPU. Such findings support the negative US uncertainty spillovers to macroeconomic activities abroad (e.g. Colombo 2013) and complement this by showing that the ERV acts as a possible channel of such transmission. By showing that the US-EPU accounts more than the domestic EPU in explaining the ERV fluctuation, we provide novel insights into FOREX-EPU nexus which generally conclude that domestic activity (exchange rates) is more reactive to the global (domestic) EPU (Nilavongse et al. 2020). Assuming that such a difference is probably due to our procedure and specific context, we support the key role of markets cross-sectional dependence when dealing with EPU implications. Then, it seems not suitable to generalize time series results and advanced economies or some major EME (e.g. China) findings to the whole emerging economies context.

Beyond the transmission of the policy uncertainty over the ERV, two important results emerge from our investigation. First, we provide evidence of a positive effect of the ERV increase on the domestic EPU. Second, another important finding is the responsiveness of the US-EPU to EME outlook. Hence, even though the US is perceived as a net exporter of EPU, it also imports such an uncertainty from EME.

Our findings have interesting implications to decisions makers and investors in EME. Picking sound economic policies and reducing the exposure to the US economic policy fluctuations is relevant for their future monitoring of exchange rate fluctuations. In addition, EME investors should incorporate domestic and abroad policy information in order to properly manage their exchange exposure and optimize portfolios allocation. On the US side, policy decisions makers’ should probably monitor the EME macroeconomic conditions in order to predict their own EPU changes.

Finally, a nonlinear assessment of the ERV-EPU nexus is worth mentioning as a potential avenue for future researches.

Appendices

Appendix A: The PVAR Estimation

See Appendix Figs. 3 and 4.

Fig. 3

Stability of the PVAR estimates

Fig. 4

IRFs for PVAR model. This figure plots the orthogonalized IRFs of all the variables to a positive one standard deviation shock of each variable in the PVAR system along with 95% error bands. These plots show the response of the column variable to an innovation on the row variable based on PVAR estimates with two lags. The bluish-gray boxes are confidence intervals constructed via Monte-Carlo simulations with 1000 replications. Periods are months. Non-zero effects i.e. significant shocks impacts are observed when the zero line is outside of the corridor (Color figure online)

Appendix B: Robustness Check

Estimating the PVAR Model with an Alternative Lag Structure

See Appendix Fig. 5.

Fig. 5

Stability of the PVAR model augmented with a lag

The results illustrated in Fig. 5 indicate the stability of the PVAR model with three lags.

See Appendix Fig. 6.

Fig. 6

Orthogonalized IRFs. Robustness Check (PVAR model augmented with a lag). This figure plots the orthogonalized IRFs of all the variables to a positive one standard deviation shock of each variable in the PVAR system along with 95% error bands. These plots show the response of the column variable to an innovation on the row variable based on PVAR estimates with three lags. The bluish-gray boxes are confidence intervals constructed via Monte-Carlo simulations with 1000 replications. Periods are months. Non-zero effects i.e. significant shocks impacts are observed when the zero line is outside of the corridor (Color figure online)

Estimating the PVAR Model with an Alternative Proxy of the Foreign EPU

See Appendix Fig. 7.

Fig. 7

Stability of the PVAR model with Global EPU (GEPU)

Our PVAR estimates using the GEPU are stable as highlighted in the Fig. 7

See Appendix Fig. 8.

Fig. 8

Orthogonalized IRFs. Robustness Check (PVAR model with GEPU). This figure plots the orthogonalized IRFs of all the variables to a positive one standard deviation shock of each variable in the PVAR system along with 95% error bands. These plots show the response of the column variable to an innovation on the row variable based on PVAR estimates when replacing the US-EPU by the GEPU. The bluish-gray boxes are confidence intervals constructed via Monte-Carlo simulations with 1000 replications. Periods are months. Non-zero effects i.e. significant shocks impacts are observed when the zero line is outside of the corridor (Color figure online)

Estimating the PVAR Model with a Different Sample

See Appendix Fig. 9.

Fig. 9

Stability of the PVAR model excluding Singapore and China from the sample

We drop both Singapore and China out of the sample and re-estimate our PVAR model. The optimal lag selection based on the three information criteria retained previously also provides indicative evidence of two optimal lags. Corresponding results are not reported here but are available on request. Regarding the stability condition, we can observe on Fig. 9 that eigenvalues are inside the unit circle indicating a strong stability of this alternative PVAR specification.

See Appendix Fig. 10.

Fig. 10

Orthogonalized IRFs. Robustness Check (PVAR model excluding Singapore and China). This figure plots the orthogonalized IRFs of all the variables to a positive one standard deviation shock of each variable in the PVAR system along with 95% error bands. These plots show the response of the column variable to an innovation on the row variable based on PVAR estimates when excluding china and Singapore out of the sample. The bluish-gray boxes are confidence intervals constructed via Monte-Carlo simulations with 1000 replications. Periods are months. Non-zero effects i.e. significant shocks impacts are observed when the zero line is outside of the corridor (Color figure online)