Journal Issue
Share
Article

Financial Regulation and the Current Account

Author(s):
Tomasz Wieladek, and Sergi Lanau
Published Date:
April 2012
Share
  • ShareShare
Show Summary Details

I. Introduction

The run-up to the global financial crisis of 2008 was characterized by large and persistent current account imbalances. One school of thought argues that ‘the origins of the crisis lie in the imbalances in the world economy that built up over a decade or more’ (King, 2009). There is both empirical and theoretical evidence to support this proposition. Theoretically, Caballero and Krishnamurthy (2009) develop a model to show how capital flows to the US triggered a sharp rise in asset prices and a decrease in risk premia and interest rates. Empirically, Warnock and Warnock (2009) estimate that, if there had been no foreign official flows into US government bonds over the course of a year, long rates would be almost 100 basis points higher. Similarly, Sa, Towbin, and Wieladek (2011) use a structural VAR model to show that, in OECD countries, capital inflows shocks have a greater effect on real house prices and residential investment than monetary policy shocks.

Others contend that current account imbalances just reflect domestic credit booms and asset price bubbles. Laibson and Mollerstrom (2010) present a model in which a current account deficit is the result of a domestic asset price bubble and credit boom, features that the authors argue are consistent with the data. Similarly, Gete (2010) shows that increases in housing demand can lead to a trade deficit via consumption smoothing across tradable and non-tradable goods. Empirically, Fratzscher, Juvenal, and Sarno (2007) find that US asset price movements can explain up to 32 percent of the US trade balance, providing support for this view.

A third, still nascent, school of thought points to financial deregulation/innovation as an explanation for global current account imbalances. In particular, Ferrero (2011) considers a shock to the collateral constraint, which he argues can be interpreted as the macroeconomic effect of financial innovation, in a two-country model. The realization of this shock in a calibrated version of his model can, to some extent, replicate the US current account deficit and house price boom observed in the run up to the crisis.1 In a similar vein, Borio and Disyatat (2011) argue that the problem is the excess ‘elasticity’ of the international monetary and financial system. They define this elasticity2 as ‘the degree to which extent monetary and financial regimes constrain the credit creation process, and external funding more generally’. They hypothesize that an increase in the elasticity of the financial system over time, as a result of financial deregulation for instance, led to greater domestic and external imbalances.3 Indeed, the data suggests that the increase in the absolute size of current account imbalances (figure 1) and their persistence (figure 2) emerged against the backdrop of financial deregulation in both OECD and emerging market countries (figure 3)4, providing informal support for the ‘excess elasticity’ hypothesis. But, to our knowledge, no previous work has rigorously tested if and to which extent, the reaction of the current account depends on the domestic financial regulatory regime. In this paper, we take a step towards filling this gap. In particular, we provide the first empirical evidence that, for a given set of net output shocks, the current account response is larger and more persistent in a country with a low, than in one with a high, degree of financial regulation.

Figure 1.Size of current account imbalances

(in percent of world GDP)

Source: IMF WEO

Figure 2.Financial deregulation index

Source: Abiad et al. (2010)

Figure 3.Current account persistence

(as a share of GDP)

Source: IMF WEO and staff calculations

In the first part of the paper, we derive robust identification restrictions and theoretical predictions from the intertemporal model of the current account (Sachs, 1981), which we augment with financial regulation. Following previous theoretical and empirical work on long-run economic growth, savings, and consumption, we model financial regulation as a liquidity constraint. The tighter financial regulation is, the fewer agents in the economy have access to the saving and borrowing technology. Recent work argues that the standard model needs be extended by either external habit formation (Bussiere, Fratzscher, and Müller, 2004), internal habit formation (Gruber, 2004) or a stochastic world real interest rate (Bergin and Sheffrin, 2000) to match the persistence of the current account typically observed in the data. We derive analytical solutions of our model under each of these assumptions (always under liquidity constraints) to show that our predictions are robust to these modeling choices. We then trace the effect of a shock to the log level (difference) of net output on the current account for different values of the liquidity constraint. This permits us to analytically show that, regardless of modeling choices and parametrization, a given net output shock will have a larger effect on the current account balance if agents are less liquidity constrained. In the case of external habit formation in consumption, the current account response also becomes more persistent as the liquidity constraint eases. In other words, the lighter financial regulation, the greater and more persistent the reaction of the current account to net output shocks. The intuition is simple: the current account reflects agents’ savings decisions in response to net output shocks. In a repressed (liberalized) financial system, few (many) agents have access to borrowing and saving and the current account will therefore show a smaller (greater) reaction to domestic shocks.

In the second part of the paper, we take our theory to the data. The theoretical model suggests that real per capita net output growth and the fiscal balance to net output ratio, together with their lags, determine the current account to net output ratio. As these variables are likely to be highly endogenous, the VAR methodology is the most suitable approach to verify the theoretical predictions. The theory also predicts that the effect of the shock should depend on the degree of liquidity constraints/financial regulation. We therefore allow the structural VAR coefficients to vary with the degree of financial regulation and capital account openness, an independent determinant of liquidity constraints (Lewis, 1997). This interacted panel VAR model is estimated with time and fixed effects on an unbalanced panel of annual data across 84 countries over the period 1973-2005, using the de jure financial regulation indices provided in Abiad, Detragiache, and Tressel (2010). Our theory also provides sign and shape identification restrictions that are robust to parametrization and specific modeling choices. In particular, a log level (difference) net output shock will lead to an increase in the level of net output and a current account surplus (deficit). Furthermore, a non-explosive net output process in log levels (differences) has the property that the response to a one-time unexpected shock does not increase (decrease) over time. Similarly, in absolute value terms, the current account response to either net output shock should not increase over time. These properties of the model naturally translate into sign and shape restrictions of the type used in Canova and Nicolo (2002), Faust and Rogers (2003), and Uhlig (2005). This allows us to identify both log level and log difference net output shocks under regimes of low and high financial regulation. We then compare the impulse responses under the two regimes to assess if the effect of these shocks on the current account balance to net output ratio varies with financial regulation. Our results confirm that the impact of both shocks on the current account differs across low- and high-regulation regimes in a statistically-significant way. The response of the current account to either type of net output shock is approximately 60 percent larger, as well as more persistent, in the average financially-liberalized country than in the average financially-repressed country. This implies that a future move toward tighter financial regulation around the globe could have a sizeable impact on current account imbalances. To our knowledge, this dimension of the financial reform debate is not usually stressed but is potentially important for the functioning of the international monetary system.

This paper contributes to the existing literature in a number of ways. We provide the first empirical investigation of the interaction between financial regulation and the current account, expanding upon existing empirical work that examines the relationship between financial regulation and savings decisions in the closed economy (Bandiera and others, 2000; Bayoumi, 1993a,b; Bayoumi and Koujianou, 1989; Jappelli and Pagano, 1994). Second, we work with a broad sample of countries for the standards of previous papers studying the effect of financial regulation on savings decisions, being the first to apply the sign restriction methodology to test the predictions of this type of model. Finally, a secondary contribution is our detailed exploration of the role of liquidity constraints in different standard versions of the intertemporal model of the current account.

The rest of this paper proceeds as follows. Section 2 lays out the theoretical model and derives theoretically robust identification restrictions and predictions. Section 3 discusses the empirical specification and the data used. Section 4 presents the empirical results and robustness checks. Section 5 concludes.

II. Theory

Our aim is to embed financial regulation in a model of the current account that is simple enough to take to the data in a neat way. We build on the Sachs (1981) intertemporal model of the current account (ICA). Conceptually, financial regulation can affect macroeconomic variables in at least two ways. First, regulation can affect the efficiency of transforming savings into productive investment (Goldsmith, 1969) by promoting or hindering competition in the banking system. Second, the volume of savings flowing into investment can either increase (McKinnon, 1973; Shaw, 1973) or decrease (Devereux and Smith, 1994; Jappelli and Pagano, 1994) following deregulation. The former is likely to affect the price, while the latter the quantity, of capital available for investment. Since current account balances reflect quantities rather than prices we focus on the latter channel. While theoretically ambiguous, empirical evidence supports the proposition that tighter financial regulation decreases savings through the liquidity constraints channel. Diaz-Alejandro (1985) points out that in developing countries, the volume of savings tends to fall following financial liberalization (deregulation). Bayoumi (1993a), Bayoumi (1993b), Sarno and Taylor (1998) find that financial deregulation in the UK in the 1980s decreased liquidity constraints, leading to a decline in aggregate savings. Bayoumi and Koujianou (1989) and Jappelli and Pagano (1994) confirm this pattern across a range of industrial countries. Bandiera and others (2000) also find empirical support for the idea that financial regulation decreases savings via the liquidity constraints channel in eight emerging market economies. Finally, using data for 72 countries, Lewis (1997) finds that consumers in countries with government restrictions on international transactions tend to act as if they are liquidity constrained. This breadth of evidence justifies our approach of introducing financial regulation via liquidity constraints in the ICA model.

Bergin and Sheffrin (2000) and Ghosh (1995) among others, do not find empirical support for the simple intertemporal current account model in most G7 countries. The literature has proposed three different modifications of the basic model to improve its empirical performance. Gruber (2004) introduced internal habit formation (where utility is a function of past individual consumption), Bussiere, Fratzscher, and Müller (2004) used external habit formation (where utility is a function of past average consumption), and Nason and Rogers (2006) argued that a time-varying stochastic world interest rate delivers a more realistic model. Recently, Kano (2009) has shown that the internal habits and time-varying world real interest rate approaches provide observationally equivalent predictions. Together with one of these three frictions, we add liquidity constraints to the standard ICA model in order to capture the effects of financial regulation. This allows us to show that our identification restrictions are robust across these modeling choices and reasonable parametrization.

We consider a small open, endowment economy populated by a large number of households who maximize utility subject to a budget constraint. Output, investment, government expenditure, and lump-sum taxes are exogenous. 5 There are two types of households in the economy. One type of household is liquidity constrained in the sense that they do not have access to any savings technology. We will refer to them as non-Ricardian households. These households consume their exogenous income each period:

where Yt is output, It is investment and Tt are taxes. The second type of household has access to incomplete international financial markets in which only non-contingent riskless bonds are traded. These households are characterized by their optimal behavior with respect to the intertemporal allocation of consumption. We will refer to this group of households as Ricardian throughout. We assume that non-Ricardian households make up a fraction of λ ε [0,1] of the population with Ricardian households as the remainder. Hence, aggregate consumption is defined as Ct=λCtNR+(1λ)CtR.

The representative Ricardian household solves the following maximization problem:

where β ε {0,1} is the discount factor and CtR consumption at time t, and Ct−1 is past consumption. Bt+1P is the net stock of international bonds held by the Ricardian household at the end of time t, which are reimbursed at the world real interest rate rt. Households choose CtR and Bt+1P to maximize discounted lifetime consumption subject to the budget constraint and a no-ponzi condition. We assume 0 ≤ h < 1, which implies that Ricardian households may be habit forming with respect to consumption. We explore both the possibility of external and internal habit formation in consumption. As a result, utility is increasing in consumption expenditure that exceeds the depreciated value of last period’s average consumption in the economy (in the case of external habits) or the depreciated value of the household’s lagged consumption (in the case of internal habits), with h being the rate of depreciation. The utility function takes a log form throughout. The first-order necessary conditions of this optimization problem comprise the budget constraint (2), the transversality condition (3) and the following Euler equation:

Iterating the budget constraint and imposing the no-ponzi condition yields the intertemporal budget constraint of the Ricardian household:

We also assume the presence of an exogenous government sector. The government budget constraint is:

where Bt+1G is the net stock of international bonds held by the government at the end of time t. Gt is government spending. Iterating the government budget constraint and combining it with (6), one can show that

where NOt = YtItGt is net output. Since only Ricardian households hold bonds, the stock of external assets, Bt, is

In order to derive an analytical solution to the present value of the current account, we take linear approximations of both the budget constraint (7) and the Euler equation (5). Following the approach in Kano (2008)), one can show after a fair amount of algebra (see appendix) that

Let c, bG, bP, μ and γ denote the unconditional means of the consumption to net output ratio, the net foreign private asset to net output ratio, the net foreign public asset to net output ratio, the log of the gross world real interest rate and the growth rate of net output and consumption, respectively. Let us define X˜t=Xt-X¯ where X¯ is the steady state value of variable X. Taking a first-order Taylor expansion of (8) around the steady state yields:

where κ = exp(γμ). This is the linearized budget constraint. At this point further assumptions have to be made to solve the model. The literature has proposed three different simplifications to solve the model. The first option is to assume that the world real interest rate is constant and that there is external habit formation (i.e., the argument of the utility function is CtRhCt−1 where C denotes average consumption in the economy). The second option is to assume a constant world real interest rate and internal habit formation (i.e., the argument of the utility function is CtRhCt1R). Finally, one can assume a stochastic world real interest rate and no habit formation (h = 0 in our model). Kano (2009) shows that these last two frictions imply observationally equivalent data generating processes. Since in practice it is difficult to know which model is underlying the data generating process, we show that our identification assumptions are robust to any of these specific modeling choices. The detailed derivation of the current account reaction functions is relegated to the appendix.

Under a constant world real interest rate and external habit formation the current account reaction function is

where ft is an expectational error, defined as ft=h(1λ)2cΣi=1κi(EtEt1){ΔlnCt+i1R˜},cat˜=CAt˜NOtandfbt¯=rBtG˜NOt+Tt˜NOtGt˜NOt. Let us develop some intuition underlying this equation. In the absence of liquidity constraints, the current account under external habit formation is a function of the lagged current account and a weighted average of current and future net output changes.6 As habits become stronger, the importance of both the lagged current account to net output ratio increases (the first term in (10)) and the weight of expected future net output growth (third term) increases. Liquidity constraints play three roles in equation (10). First, they create an interaction between the fiscal balance and the current account. This is because the way government spending is financed directly affects the income and consumption of non-Ricardian households. A fiscal deficit results in a current account deficit since it increases the income of these households.7 Second, as a larger fraction of households is liquidity constrained, the importance of net output shocks diminishes since fewer households are able to smooth such shocks by borrowing and lending internationally. Finally, the coefficient on the lagged current account term is a function of both habits and the fraction of liquidity constrained households. This follows directly from the Euler equation, since households’ desired smoothing pattern depends on aggregate average consumption and hence implicitly on the liquidity constraint. With external habits, therefore, the speed of current account adjustment also becomes a function of the liquidity constraint.

Under a constant world real interest rate and internal habitual consumption the current account reaction function is

where ft is an expectational error, defined as ft=(1λ)cκh1κh(ΔlnCtR˜Et1ΔlnCtR˜). The intuition underlying (10) and (11) is very similar. Unlike in (10), the lagged current account coefficient is not affected by liquidity constraints in (11). This is because, in the case of internal habitual consumption, only individual consumption is relevant and there is no scope for the liquidity constraint to interact with the lagged current account term.

Without habit formation, but under the assumption of a stochastic time-varying world real interest rate (rt˜=ρrrt1˜+ηt), the current account reaction function is

where ft=[((1λ)(eμ1)bGeμbG)+(1λ)ρrκ1ρrκ]ηt. Note that, as in Kano (2009), the current account reaction function under internal habits (equation (11)) is observationally equivalent to the reaction function under a stochastic world real interest rate (equation (12) ). For the remainder of this paper, we will therefore only derive results for the internal habits model, keeping in mind that its prediction are observationally equivalent to the time-varying stochastic world real interest rate model.

A. How does financial regulation affect the response of the current account to output shocks?

In this section we derive robust identification restrictions and examine how the current account reacts to net output shocks at different levels of regulation (the liquidity constraint) in our theoretical model. In order to understand how the current account reacts to net output shocks at different levels of financial regulation, we need to make further assumptions on the stochastic process driving lnN˜Ot. Net output can be subject to shocks in log differences, Δ lnNOt= ρ Δ lnNOt−1 + εt, or log levels, lnNOt= ρ lnNOt−1 + εt. A priori, it is not feasible to know which process is driving the log of net output. We will therefore consider both of them and examine the effect of an unexpected shock to either net output process on the current account under the assumption of internal and external habits individually. At this point, most previous work would solve the model numerically and show that those theoretical impulse responses, from which the sign identification restrictions are derived, are robust to many different possible parameter values.8 The advantage of our approach is that we can demonstrate the robustness of our identification restrictions and theoretical predictions analytically.

Let us consider the case of external habitual consumption first. Under the assumption that net output is subject to shocks in log differences, one can show that (see appendix) (10) becomes

It is then easy to see that:

as long as λ, ρ, h, κ ∈ [0,1) and k ≥ λ, which will be satisfied under any plausible parametrization of this model.9 Given these parameter restrictions, one can clearly see that the impact response of a net output log difference shock upon impact is negative and that greater liquidity constraints make the impact response of this shock less negative, thus smaller. In other words, financial deregulation, i.e. the removal of liquidity constraints, makes the response of the current account to a net output log difference shock larger. Since the coefficient on cat1˜ is a function of λ one can also clearly see that the effect of a past shock on the current account today declines with a greater fraction of liquidity constrained households. This means that the persistence of the current account is also decreasing in the liquidity constraint.

Under the assumption of internal habits and considering a net output log difference shock, the current account reaction function is

With regard to the size prediction, it is then easy to see that the effects of financial regulation under internal habits are qualitatively identical to those under external habits:

But unlike in the case of external habits, liquidity constraints no longer affect current account persistence. Let us turn to the case of net output shocks in log levels. In this case, the model with external habits becomes

It is then easy to show that cat˜εt>0 and cat˜εtλ<0..

For the case of shocks in log-levels, the current account under internal habits is

and

This means that the current account goes into surplus following a log level net output shock and that the reaction of the current account is smaller as liquidity constraints increase. As in the case of log difference shocks, with external habit formation, the persistence of the current account is declining in the liquidity constraints. In short, the model predicts that under lower liquidity constraints, the response of the current account becomes larger in any case, and more persistent in the presence of external habits, for a given log level/difference net output shock.

Finally note that our theory suggests that the data generating process for the current account will be a reduced form VAR model with cat˜, fbt˜, Δ lnN˜Ot as the regressors.

III. Empirical methodology and data

This section develops a methodology to test the main implications of our model, namely that the reaction of the current account to net output shocks is larger and more persistent under lighter financial regulation.

A. General methodology

In practice, all the variables required to test this model are likely to be endogenous with respect to each other. To address this concerns, we adopt a panel VAR approach where we explicitly identify net output shocks while addressing endogeneity concerns.10 We use annual data given that financial regulation indices are not available at quarterly frequency. The panel VAR includes the three variables that drive the data-generating process for the current account according to our model: the current account to net output ratio, the fiscal balance to net output ratio and the net output growth rate. We therefore estimate the following interacted panel VAR model:

where Yi, t is an n × 1 vector consisting of the current account to net output ratio, the fiscal balance to net output ratio and the net output growth rate in country i at time t. Since the theory provides predictions for variables expressed in deviations from their steady state value, we express these three variables in deviations from their country-specific means (which are our proxies for their steady state values). This removes the country fixed effect for these variables. γt is a time fixed effect.11Ak is an autoregressive matrix up to lag L. ei, t , is an n × 1 vector of residuals, assumed to be uncorrelated across countries and normally distributed with a constant covariance matrix Σ. The model is estimated with 2 lags.12

Xi, t is a vector of the variables that we interact the panel VAR coefficients with. In other words, we allow the impact and lagged VAR coefficients to vary with the interaction term. Most existing studies model time variation in the VAR coefficients as a random walk (Cogley and Sargent, 2005). In contrast to these papers, we follow Towbin and Weber (2011) and Sa, Tow-bin, and Wieladek (2011) and allow the coefficients to vary with observable deterministic variables.13 In our baseline specification, the q × 1 vector Xi, t contains Finregi, t, an index of financial regulation, and KAOpeni, t, a measure of capital account openness. Previous work by Lewis (1997) has documented that restrictions on international transactions may lead households to act as if they are liquidity constrained. Similarly, Towbin (2008) has shown that the trade balance, an important component of the current account, becomes more persistent with greater capital account openness. To avoid the omitted variable bias, capital account openness is therefore included as an independent determinant of the VAR coefficients. We use the de jure versions of these indicators because they are unlikely to be affected by net output shocks directly and can therefore be treated as exogenous with respect to the VAR system.14 In addition, Xi, t enters separately to ensure that the change in the VAR coefficients does not occur due to omission of the interaction variable in levels.

The matrix A0 is an n × n matrix with ones in the diagonal. Any element above the diagonal is 0. As a result of the lower triangular form, the residuals, ei, t, will be orthogonal to each other and the covariance matrix Σ will be diagonal. The lower triangular elements of this matrix are interacted with Xi, t and take the following form: the coefficient in row j and column h is modeled as A0,i, t(j, h) = A0(j, h)Yi, t(h) + B01(j, h)Yi, t (h)Finregi, t + B11 (j, h)Yi, t(h)KAOpeni, t, where j < h . B01(j, h) and B11(j, h) are both scalar coefficients that denote the marginal change in financial regulation and capital account openness on A0(j, h), respectively. By setting Finregi, t at a high (low) value, one can therefore obtain A0,i, t (j, h), which would correspond to the impact coefficient in a country with high (low) financial regulation. Estimating the VAR equation by equation in this recursive form provides a simple way of allowing the reduced form covariance matrix of the VAR, Σi,t˜=A0,i,t1ΣA0,i,t1, to vary with financial regulation and capital account openness. All coefficients in the lagged dependent variable coefficient matrix, Ak, i, t(j, h), are also allowed to vary with the interaction terms, Xi, t, in an identical manner, but without any zero or one restrictions.

In summary, we are allowing the coefficients in both the impact and lagged dependent variable matrix to vary with observable deterministic variables. The advantage of this approach is that we can assess to which extent the impulse responses to a given shock differ with the degree of financial regulation (the liquidity constraint). Since our theory predicts that the effect of a net output shock on the current account to net output ratio depends on the degree of the liquidity constraint, this method is better suited to test our theory than standard time-varying coefficient VARs, which typically do not provide information on the source of time variation.

B. Identification of net output shocks

Previous work by Kano (2008) used zero restrictions derived from a model similar to (11) to identify net output shocks in Structural VARs. It is however difficult to establish whether the zero restrictions proposed by either of the models presented here are valid in the data. We therefore adopt an identification procedure that does not rely on zero restrictions. As we have shown in section II.A, a log level net output shock increases the log level of net output and results in a current account surplus. Conversely, a log difference net output shock results in an increase in the log level of net output and a current account deficit. This provides mutually exclusive sign restrictions, robust across all of the theoretical models considered in this paper. We therefore use sign restrictions as pioneered by Canova and Nicolo (2002), Faust and Rogers (2003), and Uhlig (2005) to identify the shocks of interest. In particular, we use the QR decomposition approach presented in Rubio-Ramirez, Waggoner, and Zha (2010) to search across the space of all possible structural VAR decompositions. Those that do not produce impulse responses that satisfy the restrictions in table 1 are discarded and the remaining ones are kept for inference.

Table 1.Sign restrictions
Log Level Net Output shockLog Difference Net Output shock
LN NOt≥ 0≥ 0
CAtNOt≥ 0≤ 0
FBtNOt??

The avid reader will note that our model is estimated on the log difference, but that we impose identification restrictions on the log level, of net output. This is necessary to disentangle the two shocks, as the log level net output shock only implies a positive sign upon impact in the net output log difference space and we impose restrictions upon impact and 4 years there after.15 In other words, in the case of net output, we are imposing sign restrictions on the cumulative impulse responses of the log difference of this variable. Paustian (2007) points out that this methodology will only recover the structural shocks if a sufficient number of restrictions is imposed. To further ensure that this is the case, we also impose ‘shape’ restrictions on our impulse responses to rule out shocks to net output that do not correspond with the theory.16 In particular, a stationary autoregressive log level (difference) net output process has the property that the response to an unexpected one-time shock does not increase (decrease) over time. Similarly, the theory implies that the current account response to any shock does not increase over time in absolute value. We therefore also discard structural decompositions which imply impulse responses that do not satisfy these shape restrictions.

Our proposed sign restrictions also emerge robustly from more complex open economy DSGE models. Koske and Fournier (2010) use the two-country New Open Economy Macroeconomics model of Ferrero, Gertler, and Svensson (2008), albeit without sticky prices, to investigate the effect of temporary and permanent productivity shocks on the current account through the savings channel. They find that the only scenario in which a temporary (permanent) productivity shock leads to a current account deficit (surplus) is by setting by their intertemporal elasticity of substitution parameter to 10, a high and unrealistic value compared to the existing literature. 17 Otherwise, their model predicts a non-negative (non-positive) current account response following a temporary (permanent) productivity shock. Similarly, Enders and Müller (2009) study the impact of permanent technology shocks on net exports18, allowing for both the savings and investment channel, in a variant of the Backus, Kehoe, and Kydland (1994) international RBC model. Under the assumption of incomplete markets, which is strongly supported by their time series evidence, they find that a permanent productivity shock leads to a current account deficit. Finally, Enders, Müller, and Scholl (2011) derive theoretically robust sign restrictions from a very general two-country DSGE model. They assume a near-unit root process for their technology shock, meaning that it can be interpreted as permanent, and find that with a low trade-price elasticity, a feature of consistent with the data (see Enders and Müller (2009)), a very large fraction of their net exports impulse responses display a negative reaction to the technology shock. This implies that the fact that a permanent technology shock has a negative effect on the current account is theoretically robust to most parametrizations, given their very general two-country DSGE model with very wide parameter intervals.19 Clearly, a temporary (permanent) technology shock would raise the level (growth rate) of output. We therefore argue that the log level (difference) net output shock identified in our model probably reflects a temporary (permanent) productivity shock in these more complex DSGE models. If this is indeed the case, then our proposed identification restrictions are consistent both with the standard intertemporal model of the current account as well as more general open-economy DSGE models.

C. Estimation and inference

There are different ways of estimating (17). One option is to pool the coefficients across countries, assuming identical autoregressive dynamics across all units. If that assumption is violated, the resulting dynamic heterogeneity bias will typically lead to an upward bias in the VAR coefficients (Canova, 2007), meaning that it is easy to mistake the temporary effect of a shock for a permanent one. Alternatively, Pesaran and Smith (1995) propose the mean group estimator as a solution to the problem of heterogeneity in the lagged slope coefficients.20 This model is implemented by estimating the VAR model country by country and then averaging the country-specific VAR estimates to obtain the panel estimate. Rebucci (2003) points out that in small time series, as in our case, mean group panel VAR estimates may be subject to serious small-sample bias. We therefore follow his prescription and pool the autoregressive coefficients across countries. However, since we allow the VAR coefficients to depend on financial regulation, the dynamic heterogeneity bias should not be a significant problem in our case. Our theoretical model predicts that the variation in the persistence of the current account is a result of different liquidity constraints across countries and our empirical strategy controls for this explicitly.

Following Uhlig (2005), we compute Bayesian error bands using a Normal-Wishart prior.21 Since the covariance matrix in model (17) is diagonal by construction, we can proceed equation by equation. For each equation, we draw the parameters jointly from the posterior. Given the posterior draw, we then evaluate the interaction terms at the value of interest. For example, to obtain coefficients of the impact matrix, A0, for a country with high financial regulation, we use A0,FinregHigh(j, h) = A0 (j, h) + B01 (j, h)FinregHIGH + B11 (j, h)KAOpenMEDIAN, where FinregHIGH and KAOpenMEDIAN are high and median values of these indices. We use an analogous expression for each coefficient in the lagged dependent variable matrices. We then implement our identification procedure in the following manner: given A0,FinregHigh, which would correspond to the inverse of the Choleski decomposition in a reduced from VAR, we follow Rubio-Ramirez, Waggoner, and Zha (2010) and obtain the QR decomposition of a matrix of random numbers. We then verify if the structural decomposition A0,FinregHigh1Q, together with the correspondingly evaluated lagged dependent variable matrices, generate impulse responses which are consistent with our sign and shape restrictions. If this is the case, we keep this draw of A0,FinregHigh1Q and the corresponding impulse responses. We repeat this procedure until we have retained 100 draws which are used for inference.

D. Data

Our empirical strategy requires standard macroeconomic data and a good empirical measure of financial regulation. There are a few data sets that measure financial liberalization across countries (Abiad, Detragiache, and Tressel, 2010; Kaminsky and Schmukler, 2003; Williamson and Mahar, 1998). We rely on the data set by Abiad, Detragiache, and Tressel (2010) for basically two reasons. First, their country and time coverage is very wide (91 countries over the period 1973-2005) and thus appropriate for our panel VAR methodology. Second, their data set has seven graded components with special emphasis on domestic financial reform. Previous indices (eg, Kaminsky and Schmukler (2003)) put more weight on the liberalization of capital flows, which is not the central object of study in this paper. The seven components of the dataset are: credit controls, interest rate controls, entry barriers, state ownership in the banking sector, prudential regulation, securities market policy and capital account restrictions. Each component can take the values {0,1,2,3} with higher values meaning less regulation/restrictions.22 We sum all components, but capital account restrictions, to come up with the aggregate domestic financial regulation index we use in our empirical exercise. This index is normalized to one. Capital account restrictions are accounted for separately using the index by Chinn and Ito (2008)). 23

Data on the following macroeconomic variables for the period 1973-2005 is necessary to test the implications of the model: the current account balance, the fiscal balance, and a measure of net output. The IMF WEO gives us the current account balance and the fiscal balance in current US dollars. Net output is defined as GDP minus government consumption and investment. We obtain the shares of government consumption and investment in GDP from the World Bank WDI and we combine them with GDP in current US dollars from the IMF WEO to calculate net output. To obtain real net output per capita, we divide this by the US GDP deflator and total population. We remove outliers from the log difference of per capita net output, current account to net output ratio and fiscal balance to net output ratio by dropping values outside the 98th and 2nd percentile of each variable’s distribution. Our sample consists of 84 countries (table 2) and 1,756 annual observations.

Table 2.Country sample
AlbaniaEstoniaMozambique
ArgentinaEthiopiaMalaysia
AustraliaFinlandNetherland
AustriaFranceNorway
BelgiumUnited KingdomNepal
Burkina FasoGeorgiaNew Zealand
BangladeshGhanaPakistan
BulgariaGreecePeru
BelarusGuatemalaPhilippines
BoliviaHungaryPoland
BrazilIndonesiaPortugal
CanadaIndiaParaguay
SwitzerlandIrelandRomania
ChileIsraelRussian Federation
ChinaItalySenegal
Cote d’IvoireJamaicaSingapore
CameroonJordanEl Salvador
ColombiaJapanSweden
Costa RicaKazakhstanThailand
Czech RepublicKenyaTunisia
GermanyKyrgyz Rep.Turkey
DenmarkKoreaTanzania
Dominican RepublicSri LankaUganda
AlgeriaLithuaniaUkraine
EcuadorLatviaUruguay
EgyptMoroccoVenezuela, RB
SpainMadagascarVietnam
MexicoSouth Africa

IV. Empirical results

In this section we present the results from our panel VAR model and explore a number of robustness checks. Evaluating the Finregi, t term at high and low values of financial regulation allows us to obtain average VAR coefficients under regimes of high and low financial regulation.24 The remaining interaction term, KAOpeni, t, is evaluated at the median value of its distribution. We can then obtain impulse responses under both regimes and compare them to assess whether the reaction of the current account varies with the financial regulation regime as predicted by the theory. The impulse response for the log level of net output was normalized to one in order to ensure that we are comparing current account responses to log level net output shocks of identical size across both regimes. This allows us to assess whether there is a statistically significant difference between the impulse responses due to a change in the financial regulatory regime, rather than the nature of net output shocks. Figure 4 shows impulse responses to a log level net output shock with the financial regulation term evaluated at both the 90th percentile (‘low finreg’ column) and 10th percentile (‘high fin reg’ column) of the distribution of this variable. These percentiles correspond to .23 and .83, respectively. The third column shows impulse responses obtained from a distribution of the difference in the impulse responses obtained under the high and low financial regulatory regime. The red and green dashed lines indicate the 90 and 68 percent confidence bands, respectively. Figure 5 repeats this exercise but for a log difference net output shock.

Figure 4.Impulse response functions to log level net output shock — financial regulation

Figure 5.Impulse response functions to log difference net output shock — financial regulation

The median impact response of a log level net output shock on the current account is 0.57 under high financial regulation and 0.9 under low financial regulation, which represents a 58 percent increase in the impact of the shock if a country switches from high to low financial regulation (figure 4). As one can see in column three of figure 4, this difference is statistically significant. Furthermore, column three in figure 4 shows that the median of the log net output response is close to zero throughout. This suggests that the difference in the current account response cannot be attributed to changes in the nature of the net output shock. For a log difference net output shock (figure 5), the median impact response is -0.65 under high regulation and -1.05 under low regulation (61 percent increase). Again, the third column suggests that the difference in impulse responses is statistically significant and that the net output impulse responses are not statistically different from each other. In both cases, the absolute value of the current account response is larger and more persistent under low financial regulation than under high regulation. Our theory predicted, that all else equal, the change in financial regulation should affect the impulse response of the current account balance to either type of shock in a similar way. Indeed, the change in the impact response and the persistence profile is very similar across both shocks, which provides additional verification for the theory.

We also evaluate the capital account openness term (KAOpen) at the 90th percentile (‘high KA Open’ column) and 10th percentile (‘low KA Open’ column) with the financial regulation term evaluated at the median. This yields less clear-cut results than for the financial regulation term. As figure 6 shows, the response of the current account to a log level net output shock intensifies at high levels of capital account restrictions but the third column shows that the difference between low and high capital account restrictions is not statistically significant. For the case of a log difference net output shock (figure 7), we find a more statistically significant role for capital account restrictions. But the difference in impulse responses goes in the wrong direction and as shown below is not robust.

Figure 6.Impulse response functions to log level net output shock — capital account openness

Figure 7.Impulse response functions to log difference net output shock — capital account openness

A. Robustness

This subsection shows that the results presented so far are robust to the inclusion of the exchange rate regime as an additional determinant of the panel VAR coefficients and also attempts to address potential concerns regarding the dynamic heterogeneity bias.

The empirical results presented so far rely on the assumption that financial regulation and capital account restrictions are the only relevant determinants of the VAR coefficients. But there could also be other important empirical determinants of these coefficients that we are not controlling for, leaving the findings presented thus far vulnerable to omitted variable bias. An additional natural determinant of current account dynamics is the exchange rate regime. In theory, one would expect that the speed of adjustment of the current account under a fixed and floating exchange rate regime would differ (Friedman, 1953). However, Chinn and Wei (2008) document that it is difficult to find a statistically significant relationship between current account persistence and the exchange rate regime. Nevertheless, we also introduce the fine defacto exchange rate classification by Reinhart and Rogoff (2010) as an additional determinant of the VAR coefficients.25Figures 8-11 show that our previous results regarding the effect of financial regulation on the current account to net output ratio are robust to controling for the exchange rate regime. Furthermore, the difference in the current account response between high and low capital account restrictions, to either type of shock, is no longer statistically significant.

Figure 8.Impulse response functions to log level output shock — financial regulation controling for fx regime

Figure 9.Impulse response functions to log difference output shock — financial regulation controlling for fx regime

Figure 10.Impulse response functions to log level output shock — capital account openness controlling for fx regime

Figure 11.Impulse response functions to log difference output shock — capital account openness controlling for fx regime

The potential for dynamic heterogeneity bias in our panel VAR makes inference about the adjustment dynamics of the current account with absolute certainty difficult. Recently, Jarocinski (2010) introduced a new Bayesian panel VAR estimator to deal with this problem. In his approach, there is a shrinkage parameter λ which determines the degree of pooling in the panel VAR model. If λ = 0 then the VAR estimates will correspond to a pooled panel VAR model as in our case and if λ → ∞, his model will estimate country-by-country VARs. The main advantage of his approach is that λ, and therefore the degree to which pooling is supported or not, is determined by the data. The disadvantage of his methodology is that it is not possible to estimate panel VARs with different variables in each equation, as in (17). Nevertheless, to assess if the dynamic heterogeneity bias is likely to be a problem in our application, we estimated the reduced form version of (17) with the Jarocinski (2010) approach, dropping all the contemporaneous endogenous variables and the corresponding interaction terms from the model. We estimated this model with 150,000 repetitions of the Gibbs sampling chain, discarding the first 100,000 draws as burn in and keeping every 100th draw of the remainder for inference. While it is not possible to undertake similar structural VAR exercises with estimates from this model, the estimated λ is still likely to be informative about the appropriateness of pooling. Jarocinski (2010) suggest looking at the square root of λ for this purpose. Figure 12 shows the distribution of λ from this model. λ is heavily clustered around a value of .0014, which is close to zero. This suggests that dynamic heterogeneity bias is probably not a substantial problem in our application.

Figure 12.Histogram of the square root of λ

Finally, there is also an issue regarding the potential observational equivalence of the net output shock in log differences and a stationary shock to the fiscal balance, fbt˜. Following a stationary negative shock to the fiscal balance, our theory predicts a current account deficit. If this shock has a positive effect on the level of net output as well, it would not be possible to separately identify it from a log difference net output shock using only sign restrictions. But, since the government’s budget has to be balanced in the long-run, a temporary fiscal balance shock would only have a temporary effect on the level of net output. In other words, following a negative fiscal balance shock, one would expect a current account deficit, with the level of net output increasing upon impact and then decreasing over time. This could be potentially observationally equivalent to the log difference net output shock. Fortunately, our shape restriction, that a log difference net output shock leads to a non-decreasing (permanent) effect on the level of log net output, allows us to separate these two shocks explicitly. Furthermore, our theory suggests that, all else equal, the easing of the liquidity constraint should have a similar effect on the current account, regardless of the assumed process for the log net output shock. Indeed, the results do suggest that this is the case, which implies that our proposed shape restriction is probably ensuring that inference from the retained impulse responses is not contaminated by observationally equivalent fiscal policy shocks.

V. Conclusion

Global current account imbalances and financial regulation have been at the forefront of economic policy debates since the global financial crisis broke out in late 2008. A growing literature has analyzed the role of global imbalances and financial regulation in the run-up to the crisis and the debate on the future regulatory landscape is equally active. However, the idea that financial regulation might affect the size and dynamics of current account imbalances remains largely unexplored despite its potentially important implications. This paper provides the first empirical evidence supporting the existence of a link between financial regulation and the current account.

We introduce liquidity constraints, as a proxy for domestic financial regulation, into the standard intertemporal model of the current account and show that the response of the current account to net output shocks depends on the intensity of the liquidity constraints. More specifically, the response of the current account to a net output shock increases as fewer households are liquidity constrained. This conclusion is robust to various standard modeling choices, such as the introduction of a stochastic world real interest rate or habit formation in consumption. In the case of external habit formation in consumption, the persistence of the current account response increases with a smaller liquidity constraint as well. Using a sample of 84 countries over the period 1973-2005, we test these theoretical predictions in an interacted panel VAR framework, identifying net output shocks with sign and shape restrictions. The structural VAR coefficients are allowed to vary with the degree of domestic financial regulation at the individual-country level. This setting allows us to assess if financial regulation affects the size and persistence response of the current account to a net output shock as predicted by the theory. In our baseline specification, the median current account impulse response to a net output shock is 60 percent larger upon impact, as well as substantially more persistent, under a low than under a high financial regulatory regime. This is robust to allowing the coefficients to vary with capital account openness and the exchange rate regime.

The results in this paper show that domestic financial reform has an important impact on the size and dynamics of current account balances. To our knowledge, this insight is not a prominent part of the current global financial reform debate but our findings suggest that it could have important implications for the functioning of the international monetary system. Changes in financial regulation may have a meaningful impact on current account imbalances and could be a tool that helps mitigate the imperfections in the international monetary system. Most likely, we are just scratching the surface of a policy-relevant research topic. There are many interesting questions left to explore, such as for example how domestic financial regulation affects the composition of capital flows to a given country, that future research could address.

Appendix A. Appendix

A.1. Deriving the linearized budget constraint

First, to go from (7) to (8), start with (7):

then note that Rt,1 = 1 and factor out CtR and NOt on the LHS and RHS, respectively:

Then divide the whole expression by NOt

Now recall that 1/(Πt+ij=t+1(1+rj)) and take logs and exponential in the infinite summations:

Applying the log to the interest rate factorial, it is easy to see that:

Now we can add and subtract infinitely many ln CtR and lnNOt :

which is (8) in the main text:

To linearize this expression, we use the standard formula f(x) = f(a) + f′(x) (xa)

Now note that [1+Σi=1κi]=[1+κΣi=0κi]=[1+κ1κ]=[11κ] and also note that in steady state, when all the hat variables take the value of 0,c[1+Σi=1κi]=[1+Σi=1κi], then it is easy to obtain the final linearized form:

A.2. Derivation of the current account reaction function with external habits and a constant world real interest rate

We start by linearizing the definition of aggregate consumption Ct=(1λ)CtR+λCtNR to obtain

Following Obstfeld and Rogoff (1996), the current account can be expressed as as:

which, noting that Bt=(1λ)BtP+BtG and r = eln(1 + r) − 1, can be linearized as the following the current account to net output ratio:

where cat˜=CAtNOt˜. Now plugging (18) into (19), then into (20) and simplifying yields26:

The constant world real interest rate assumption implies ln (1+rt˜)=0 i. We also add and subtract (1λ)hca˜t1 and define fbt˜=[rBtGNOt˜+TtNOt˜GtNOt˜] to obtain

Furthermore, log-linearizing the Euler equation gives

which can be expressed as

Plugging (22) into (21) gives

where c=CR¯NO¯;g=G¯NO¯ and t=T¯NO¯.

At this point, it is convenient to linearize the government’s budget constraint. Using EtΣi=0Rt,iGt+i=(1+rt)(BtG)+EtΣi=0Rt,iTt+i, the government’s intertemporal budget constraint can be written as:

The linearized version of this expression is

Under the assumption that g = t, this can then be expressed as:

Plugging (24) into (23) yields

Define ft=h(1λ)2cΣi=1κi(EtEt1){ΔlnC˜t+i1R} and use Σi=1κiEt{ΔlnN˜Ot+i1}=κΣi=0κiEt{ΔlnN˜Ot+i} and Σi=1κiEt{ΔlnN˜Ot+i}+ΔlnN˜Ot=Σi=0κiEt{ΔlnN˜Ot+i}toobtain

Now using Σi=1κiEt{ΔlnN˜Ot+i1}=κΣi=0κiEt{ΔlnN˜Ot+i} and adding and subtracting hκ(1λ)2Σi=0κiEt{ΔlnN˜Ot+i}

which is (10) in the main text. To further solve this expression, it is necessary to assume a process for lnN˜Ot. Under the assumption of ΔlnNOt=ρΔlnNOt1+ɛt (10) can be expressed as:

where we replaced hκ(1λ)(1λ)i=0κi(EtEt1)ΔlnN˜Ot+i with hκ(1λ)(1λ)(1κ)ɛt and (1λ)(1hκ)i=0κiEtΔlnN˜Ot+i with (1λ)(1hκ)1ρκΔlnN˜Ot. After further simplification, one can then obtain (13) in the main text:

To derive (15) note that under the assumption that NOt=ρlnNOt1+ɛt,ΔlnN˜Ot=(ρ1)lnNOt1+ɛt and follow the steps above. In this case hκ(1λ)(1λ)i=0κi(EtEt1)ΔlnNOt+i=(1ρ)hκ(1λ)(1λ)(1κ)ɛt and (1λ)(1hκ)i=0κiEtΔlnN˜Ot+i=(1ρ)(1λ)(1hκ)(1ρκ)lnN˜Ot1.

A.3. Derivation of the current account reaction function with internal habits and a constant world real interest rate

Start with

In order to solve out for the growth rate of consumption in the current account equation, we add and subtract hca˜t1 on both sides of the equation. This yields

Now note that with internal habits in consumption, (5) can be log-linearized as EtΔlnCRt+1˜=hΔlnCRt˜. Then use i=0κiEt{ΔlnN˜Ot+i1}=κi=0κiEt{ΔlnN˜Ot+i},i=0κiEt{ΔlnN˜Ot+i}+ΔlnN˜Ot=i=0κiEt{ΔlnN˜Ot+i} and add and subtract (1λ)(hκ)i=0κiEt{ΔlnN˜Ot+i} to obtain (11) in the main text:

where ft=(1λ)cκh1κh(ΔlnCRt˜Et1ΔlnCRt˜). Solving under the assumption that ΔlnNOt=ρΔlnNOt1+ɛt yields (14)

where we replaced (1λ)(1hκ)i=0κiEt{ΔlnN˜Ot+i} with (1λ)(1hκ)ρκ1ρκΔlnN˜Ot, using 11x=Σi=0xi, and (1λ)(hκ)i=0κi(EtEt1){ΔlnNOt+i} with (1λ)(hκ)(1κ)ɛt. To derive (16) note that under the assumption that lnNOt=ρlnNOt1+εt,ΔlnN˜Ot=(ρ1)lnNOt1+εt and follow the same steps as before.

A.4. Derivation of the current account reaction function under a stochastic time-varying world real interest rate and no habitual consumption

This section derives the current account reaction function under a stochastic world interest rate and shows the observational equivalence with the model with internal habits. As a first step, linearize the definition of the current account (CAt = NOt + rtBtCt) under the assumption of a time-varying stochastic world real interest rate to obtain

Equation (5) can be written as

linearizing this expression yields: ΔlnCRt˜=ln(1+rt+i˜). Plugging this into (9) yields

Now, plugging this into (25) and simplifying27 yields:

Now, to show observational equivalence between the above and the model with internal habits, note that (1+rt+i˜)r˜t+i. Assuming an AR(1) process for the world real interest rate rt˜=ρrr˜t1+ηt, one can then rewrite the above as

where ft=[((1λ)(eμ1)bGeμbG)+(1λ)ρrκ1ρrκ]ηt. Comparing these last two expressions and using Σi=1κiEt{ΔlnN˜Ot+i}+ΔlnN˜Ot=Σi=0κiEt{ΔlnN˜Ot+i}, one can see that:

which, as in Kano (2009)), is observationally equivalent to the solution under internal habits.

References

Strategy, Policy, and Review Department, International Monetary Fund, Washington D.C., and London Business School, London, U.K.

Similarly, Favilukis and Nieuwerburgh (2009) consider a 2-sector general equilibrium model of housing and non-housing, where households have limited opportunity to insure against shocks. They show that financial liberalization, together with a rise in foreign ownership of domestic debt, can replicate the house price dynamics observed in the US during the boom period. But in their model, foreign ownership of domestic debt is taken as exogenous, therefore explicitly ruling out the link between household behavior and the current account we want to model.

The authors trace the term back to as early as Jevons (1875), who defined ‘elasticity’ to be the degree with which money grows to accommodate the demand in the economy.

Merrouche and Nier (2010) analyze quantitatively the role of the three factors discussed so far (capital flows, loose monetary policy, and financial regulation), and conclude that the effect of capital inflows on the build-up of current account imbalances is amplified where financial regulation is weak.

We define persistence as the AR(1) coefficient from an autoregressive panel data model of the current account to GDP ratio.

The exogeneity assumption is a convenient simplification that is unlikely to hold empirically. However, since our main theoretical predictions follow directly from the Euler equation, they would most likely also hold in more complex DSGE models where all macroeconomic variables are endogenous. Unlike in those models however, the fact that we can obtain an analytical solution and we do not need to resort to numerical methods, allows us to easily and transparently demonstrate that our predictions are robust to reasonable parameter values.

In this case the model becomes virtually identical to Gruber (2004).

This is consistent with empirical evidence that documents twin deficits in the data (Corsetti and Müller, 2008; Kim and Roubini, 2008).

See Enders, Müller, and Scholl (2011) for more details on this approach.

Since κ = exp(γ − μ) and γ − μ is unlikely to be large, Kano (2008) argues that κ should be fairly close to, but smaller than, one. We therefore assume that κ ∈ [0.9,1) Since λ, the fraction of the population that is liquidity constrained, is unlikely to exceed this lower bound in reality, the condition κ ≥ λ will always be satisfied for any plausible parametrization of the model.

An alternative approach would of course be the instrumental variable approach, but for annual data Chinn and Prasad (2003) have shown that it is difficult to obtain reliable instruments for the variables driving the current account.

Since we want to ensure that we identify net output shocks of domestic, rather than external, origin, we include time fixed effects in the model to account for external shocks.

Ex-ante lag length selection criteria, such as the Akaike, Hanan-Quinn, and Schwartz-Bayesian criteria suggest a lag length of one. However, one the main assumptions of the VAR model is that residuals behave like white noise. Estimated with one lag, the residuals were autocorrelated of order 1, which is obviously inconsistent with white noise behavior and suggested that in at least one of the equations 2 lags would be necessary. Since the bias from omitting a lag is typically worse than that from including an extra lag, we estimate the model with 2 lags.

Our exposition of this model closely follows Sa, Towbin, and Wieladek (2011).

These variables are not demeaned as they are already standardized.

Imposing sign restrictions upon impact only does not produce substantially different results.

Identifying shocks with only sign, but no shape, restrictions yields similar results.

In their survey of this parameter value in calibrated open-economy DSGE models, Enders, Müller, and Scholl (2011) report that the intertemporal elasticity of substitution is typically set between 0.5 and 1. Empirically, Guvenen (2006) and Gruber (2006) estimate the elasticity of substitution to be 1 and 2, respectively. On the other hand, Hall (1988) and Yogo (2004) conclude that tit is not statistically different from zero. While the empirical literature has not reached a clear conclusion, an elasticity of 10 is a order of magnitude larger than what is supported by the existing empirical and theoretical work.

Net exports, or the trade balance, are linked to the current account through the following identity: CAt = TBt + rBt.

Calibrated for annual data and with a perfect unit-root in the process for the productivity shock, it is probable that all of their net exports impulse responses would react negatively to the permanent productivity shock.

See Sa, Towbin, and Wieladek (2011) for an application of the mean group estimator in interacted panel VAR models.

Specifically, we implement algorithm 2.1 described in Del Negro and Schorfheide (2011). As in Cogley and Sargent (2005), we discard draws that imply explosive VAR dynamics.

In Abiad, Detragiache, and Tressel (2010), the prudential regulation component is an exception, as higher values reflect more prudential rules. Since we want higher values of the aggregate index to reflect greater financial regulation, we change the numbering in this category such that higher values indicate less regulation. We note that our results are robust to this change.

Using the capital account openness index provided by Abiad, Detragiache, and Tressel (2010) yields similar results.

We tried entering the banking supervision sub-indicator separately as it may affect liquidity constraints differently than the other sub-components but the results were not significant.

Results using the de jure classification from the IMF AREAER database are very similar.

Where we used 1eγ = 1 and (eμ − 1) = r as in Kano (2008).

An important result, following Kano (2008), that we use throughout our derivations to simplify the equations is(eμ1)=(eμeγ)=eμ(1eγμ)=(1κ)eμ since eγ ≈ 1.

Other Resources Citing This Publication