Exchange Rates and Domestic Credit-Can Macroprudential Policy Reduce the Link?
Author: Erlend Nier, Mr. Thorvardur Tjoervi Olafssonhttps://orcid.org/0000-0001-7478-02981, and Yuan Gao Rollinson2
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund
  • | 2 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: ENier@IMF.org, TOlafsson@IMF.org, monicagr@uw.edu
Publication Date:
11 Sep 2020
eISBN:
9781513556550
Language:
English
Keywords:
WP; exchange rate; monetary policy; currency appreciation; real GDP; Macroprudential Policies; Capital Flows; Systemic Risk; credit gap; credit development; exchange rate movement; Macroprudential policy; Credit gaps; Domestic credit; Exchange rates; Capital controls; Global; credit demand and supply; credit demand; household-sector credit; aggregate credit series
Download PDF (4.5 MB)

This paper examines empirically the role of macroprudential policy in addressing the effects of external shocks on financial stability. In a sample of 62 economies over the period of 2000: Q1–2016: Q4, our dynamic panel regressions show that an appreciation of the local exchange rate is associated with a subsequent increase in the domestic credit gap, while a prior tightening of macroprudential policies dampens this effect. These results are strong for small open economies, and robust when we explicitly account for potential simultaneity and reverse causality biases. We also examine a feedback effect where strong domestic credit pulls in additional cross-border funding, potentially further increasing systemic risk, and find that targeted capital controls can play a complementary role in alleviating this effect.

Abstract

This paper examines empirically the role of macroprudential policy in addressing the effects of external shocks on financial stability. In a sample of 62 economies over the period of 2000: Q1–2016: Q4, our dynamic panel regressions show that an appreciation of the local exchange rate is associated with a subsequent increase in the domestic credit gap, while a prior tightening of macroprudential policies dampens this effect. These results are strong for small open economies, and robust when we explicitly account for potential simultaneity and reverse causality biases. We also examine a feedback effect where strong domestic credit pulls in additional cross-border funding, potentially further increasing systemic risk, and find that targeted capital controls can play a complementary role in alleviating this effect.

I. Introduction

As is well-known, exchange rate movements are difficult to predict or even to explain ex post, and their interactions with other economic and financial variables continue to be subject to debate. A recently growing literature explores how movements in exchange rates not only affect macroeconomic outcomes, but can also affect financial conditions and credit developments (Blanchard et al., 2015, Shin 2018, Ghosh et al. 2018, Hofmann et al. 2019, Bank for International Settlements (BIS), 2019, Carstens, 2019), which may in turn feed back into the macroeconomic outlook. The main idea is that a currency appreciation would tend to ease domestic financial conditions, and this would boost the demand and supply of domestic credit. In this way, an appreciation may potentially be expansionary, in contrast to the standard notion in the earlier literature where an appreciation is held to be contractionary, by reducing net exports. Moreover, when an appreciation leads the domestic provision of credit to expand, this can contribute to systemic risk, and potentially require a policy response on the part of the macroprudential policymaker.

An appreciation of the local exchange rate can drive up domestic credit through a number of channels that may be at work at the same time and reinforce each other (see, e.g., Carstens, 2019). An exchange rate appreciation raises collateral values and net worth of domestic market participants and can then both increase borrowers’ capacity to accumulate debt and ease lenders’ constraints to provide it (Krugman, 1999; Céspedes et al., 2004; and Bruno and Shin, 2015b). Currency appreciation can also lead to a lower perception of risk on the part of lenders, and an enhanced sense of prosperity on the part of borrowers (e.g., due to cheaper imported goods and services), thereby again increasing the demand and supply of credit. When increased credit in turn affects local asset prices, or is funded through borrowing from across the border, this can strengthen the ultimate effects of increases in exchange rates, potentially giving rise to a build-up of systemic risk (Gertler et al. 2007; Borio, 2014; Bruno and Shin 2015a,b; IMF 2017; and Baskaya et al. 2017).

In this paper we study the link between exchange rate movements and domestic credit in a panel of 62 countries over the period 2000 to 2016, and ask to what extent macroprudential policy can attenuate the effects of currency movements on domestic credit cycles (left-hand side of Figure 1). We also evaluate a complementary role of targeted controls on inflows, when strong developments in credit in turn lead to increases in cross-border borrowing by banks and corporate firms (right-hand side of Figure 1).

Figure 1.
Figure 1.

Exchange Rates, Credit, and Capital Flows

Citation: IMF Working Papers 2020, 187; 10.5089/9781513556550.001.A001

Source: Author’s descriptions.

In this context, this paper makes three main contributions to the literature. First, we contribute to the recent empirical literature on the link between currency appreciation and domestic credit developments (Hofman and others 2019, Bruno and Shin, 2015a,b; Hahm et al., 2013; and Shin, 2018), by using the credit-to-GDP gap as a continuous indicator of the build-up of systemic risk.

Second, we expand the literature on the effectiveness of macroprudential policy (Cerutti et al., 2017; Galati and Moessner, 2018, Alam et al., 2019), by examining the interaction effect of macroprudential policy in mitigating the impact of the exchange rate on domestic credit. This is important both from a policy perspective and from the point of view of improving the identification of the causal effects of macroprudential measures.

Finally, with regard to the literature on the relationship between capital flows and credit (Caballero, 2016, Mendoza and Terrones, 2012; Igan and Tan, 2017, Merrouche and Nier, 2017), and the effectiveness of capital controls (Ostry et al., 2011; Ghosh et al., 2018), we examine the complementary role of targeted capital controls when strong domestic credit pulls in cross-border funding and macroprudential policy faces leakages.

The empirical analysis in the main part of the paper examines the relationship between changes in real exchange rates and movements in the credit-to-GDP gap for 62 advanced and emerging market economies (AEs and EMEs, respectively) over the period 2000: Q1– 2016: Q4. The main results are the following:

  • First, exchange rate movements are associated with subsequent changes in domestic credit relative to GDP. In particular, an appreciation of the local exchange rate vis-à-vis the United States (U.S.) dollar is followed by an increase in the credit-to-GDP gap in the next quarter.

  • Second, macroprudential policy is found to have a direct effect on domestic credit developments. Where macroprudential policy is tightened, it leads to a reduction in the credit-to-GDP gap in the next quarter.

  • Third, macroprudential policy weakens the extent to which exchange rate movements affect credit developments. When we interact changes to the exchange rate with changes in the macroprudential policy stance, we find that for a given appreciation of the real exchange rate, the subsequent increase in the credit-to-GDP gap is weaker where macroprudential policies had been tightened in the previous quarter.

We find that our results are robust to a number of changes in the specification. First, we address potential simultaneity concerns that arise when some economic fundamentals might be driving both the exchange rate and credit developments. For this we apply a two-step procedure that involves “purging” the impact of domestic fundamentals on the real exchange rate and distilling more “exogenous” exchange rate shocks for use in our analysis. We find that our results continue to hold. Second, in order to address potential endogeneity of the macroprudential policy variable, we construct macropurdential policy shocks as the difference between the actual reading on the macroprudential indicator variable, and its expected value, based on prior developments in the credit gap and the exchange rate. Third, we find that results continue to hold when we move away from the credit gap and use an alternative and simpler measure of domestic credit developments.

In an extension, we proceed to examine a feedback effect from domestic credit developments to “other investment flows”, which mainly capture cross-border loans and deposits received by financial institutions and the nonfinancial corporate sector. Controlling for global push factors, as well as other domestic pull factors, we find that increases in domestic credit are associated with increases in such flows. Moreover, where macroprudetial policy is tightened this leads to further increases in cross-border flows. We interpret this as evidence of policy leakage, where domestic corporates respond to macroprudential tightening by directly borrowing from abroad. On the other hand, targeted capital controls that aim to limit these types of flows appear to be effective.We find that where these controls are in place, this reduces the effect of credit developments in stimulating other investment flows, thereby limiting the further build-up of systmic risk from direct cross-border borrowing.

The remainder of this paper is organized as follows: Section II discusses how this research is related to the existing literature. Section III presents the empirical framework and the details of variables and data. Section IV discusses the baseline empirical findings. Section V reports on various further analyses and robustness checks. Section VI presents the extension on how domestic credit may fuel capital inflows. Section VII concludes and discusses some policy implications.

II. Related Literature

This paper links up four main strands of the literature: The first analyzes the linkage between currency appreciations and systemic risk, the second the effectiveness of macroprudential policies, the third the relation between capital flows and credit, and the fourth the effectiveness of capital controls.

An emerging literature has documented the link between currency appreciation and domestic credit (Blanchard et al, 2015, Bruno and Shin 2015a,b; Hahm et al., 2013; IMF 2017, Baskaya et al., 2017, Shin, 2018, Hofmann et al., 2019; and Carstens 2019). Building on a considerable body of earlier literature that had examined sudden stops (e.g., among many, Caballero and Krishnamurthy, 2004), these studies examine how a currency appreciation in the run-up to such events raises collateral values and net worth, and is also associated with a reduction in credit spreads, thereby encouraging market participants to take greater risks and allowing for an expansion in credit volumes (see, e.g., Hofmann et al., 2019). Indeed, parts of the literature have referred to these mechanisms as the “risk-taking channel” of currency appreciation in the context of cross-border spillovers of monetary policy (Bruno and Shin 2015a; Borio and Zhu 2012; and Hofmann, Shim, and Shin, 2019). Moreover, Gourinchas and Obstfeld (2012) find that a rapid increase in leverage and a sharp real appreciation of currency emerge consistently as the two most robust and significant predictors of financial crises. Our paper builds on this literature by documenting empirically a link between exchange rate movements and the credit-to-GDP gap, which is widely understood to be an early-warning indicator of a future financial crisis, in a large cross-country panel of 62 advanced and emerging economies.1

A second strand of literature that this paper relates to examines the effectiveness of macroprudential policies. These papers typically consider the effect of macroprudential policies on financial indicators that measure “excessive” financial risk, such as the growth in credit or asset prices. Lim et al. (2011) find that macroprudential policies are effective in reducing the procyclicality of credit and leverage in the banking sector. Vandenbussche et al. (2015) report a significant impact of macroprudential policies in reducing housing price inflation in 16 countries in the Central, Eastern, and Southeastern Europe region. A large number of studies find that macroprudential policy reduces the growth of credit (Claessens et al., 2013; Akinci and Olmstead-Rumsey 2018; Cerutti et al. 2017, and Alam et al 2019), while some studies also find evidence that such policies can reduce the incidence of credit booms (Mendoza and Terrones 2012), and lower the credit-to-GDP gap (Fendoğlu 2017, Lang and Welz 2018). Most studies find evidence supporting the notion that macroprudential policies can be effective in reducing systemic risks, with the evidence overall strongest for borrower-based macroprudential tools (such as caps on loan-to-value or debt-service to income) (Cerutti et al., 2017; Claessens et al., 2013; Fendoğlu 2017, Alam et al. 2019).

We build on this literature by examining the effectiveness of macroprudential policy in a large cross-country panel and focusing on the effects on the credit-to-GDP gap, which we take to be continuous indicator of the build-up of systemic risk. Our analysis of effectiveness deploys a novel database of macroprudential policy actions, the iMaPP database compiled by IMF staff from the Monetary and Capital Markets Department (Alam et al. 2019), which integrates a number of existing databases and is the most comprehensive of such databases to date, covering 17 instruments for a total of 138 countries since 1999. Moreover, our study differs from the existing literature in that we focus on the coefficients of the interaction terms of macroprudential policies and real exchange rate movement, so as to evaluate the effects of macroprudential policies in mitigating the impact of real exchange rate on domestic credit developments.2

Third, this paper relates to the literature that has examined the relation between capital flows and credit. A large number of papers has found a positive association between “surges” of capital inflows and credit booms (e.g., Mendoza and Terrones, 2012, Elekdag and Wu, 2011), as well as between surges of capital inflows and subsequent banking crises (Caballero, 2016). Some studies highlight a tighter relation between particular types of inflows and credit developments (Bruno and Shin 2015b; Bruno and Shin 2015a; Hahm et al., 2013; IMF 2017; Igan and Tan, 2017, and Baskaya et al., 2017), and commonly find that “other investment inflows” (which is mainly deposits and loans received by banks and corporates) have the most robust positive correlation with domestic credit growth (IMF, 2017; Igan and Tan, 2017).

Most discussions focus on a causal link that runs from capital inflows to credit, where capital inflows lead to an increase in loanable funds for the domestic banking system, and thereby “push up” the supply of domestic credit. However, many studies acknowledge that there is likely to be a two-way relationship, where strong domestic credit can also “pull-in” additional capital from abroad (Igan and Tan, 2017, Amri, Richey and Willet, 2016, Lane and McQuade, 2014). For instance, a domestic (demand or supply) shock may generate rapid credit growth, which in turn fuels sentiment, boosts asset prices and pulls in international capital (Igan and Tan, 2017, Caballero, 2016).

A feedback effect running from credit demand to capital inflows may be strong in particular for “other investment flows,” including direct-cross border borrowing by corporates and cross-border funding of the banking system (Borio, McCauley and McGuire, 2011, Avdjiev, Mc Cauley and McGuire, 2012, and Hahm et al., 2013). When the growth in credit outruns the growth in domestic deposits, this can lead corporates to borrow directly from abroad (increasing direct cross-border credit), and lead banks to tap international markets to complement domestic deposit funding by wholesale funding (indirect cross-border credit).

This feedback effect is closely linked to the phenomenon of the cross-border leakage of macroprudential policy, where if policymakers put constraints on domestic lending, this may lead to increased provision of credit from abroad (e.g., Ahnert et al. 2018, Reinhardt and Sowerbutts 2015). We therefore embed our analysis of leakages of macroprudential measures in an empirical framework that focuses on the feedback effect from credit to increases in “other investment inflows.”

Our paper relates finally to a growing literature that investigates the effectiveness of capital flow management measures and other policies in managing risks from capital inflows (e.g., Cerutti and Zhou, 2018, Ostry et al. 2011, Forbes et al. 2014). Bruno et al. (2017) find that capital flow management policies in the banking sector and bond market are effective in slowing down banking inflows and bond inflows, respectively. Klein (2012) examines the association of capital inflows controls on financial variables and distinguishes between the effect of long-standing controls and episodic controls, reporting findings that suggest longstanding controls (“walls”) may be more effective than episodic controls (“gates”). In our extension we examine the effect of both macroprudential policy and targeted capital controls on “other investment flows,” and explore whether these policies can affect the degree to which strong domestic credit pulls in cross-border funding.

III. Empirical Framework and Data

3.1 Baseline Setup and Methodology

In the main part of the paper we use a dynamic panel framework to investigate the determinants of the credit gap. Our baseline set-up, which we expand on in further analysis, relates the credit gap (denoted Y) to changes in the real exchange rate, macroprudential policy actions, their interactions, as well as controls:

Yi,t=ρYi,t−1+β1Δ4RERi,t−1+β2MaPPi,t−1+β3MaPPi,t−1×Δ4RERi,t−1+θZi,t−1+μi+vi,twhere E[μi]=E[vi,t]=E[μivi,t]=0control varialbles Zi,t−1=[MPSi,t−1,Δ4F−RGDPi,t−1]

The subscripts i and t represent country and time (quarter) respectively; μi is a fixed effect that captures time-invariant country characteristics, and vi,t is the error term.3 Δ4RERi,t-1 is the year-over-year log change of real exchange rate, which is lagged by one quarter. MaPPi,t-1 is an ordinal indicator variable representing the number of macroprudential policy actions by direction (tightening actions net of loosening actions) that are taken during period t-1 in country i. In addition to this measure, we analyze the effects of tightening (T_MaPP) and loosening (L_MaPP) actions separately.

Zi,t-1 is a vector of control variables, which includes the monetary policy stance (MPS) and forecasted year-over-year real GDP growth (A4F_RGDP), both again lagged by one quarter.

In contrast to the approach in the existing literature (e.g., Claessens et al., 2013; Akinci and Olmstead-Rumsey 2018; Cerutti et al. 2017, and Alam et al 2019), we use the forecasted GDP growth rates rather than the more widely used actual GDP growth rates to mitigate a potential simultaneity concerns stemming from both credit and the exchange rate being driven by good news about the economy. By including the growth forecast, we measure the effect of the residual variation in the exchange rate that is orthogonal to these effects.

The aggregate measure MaPPi,t-1 is a lagged ordinal indicator variable representing the number of macroprudential policy actions by direction (tightening actions net of loosening actions) which are taken during period t – 1 in country i. For example, a value of +3 (-3) represents three policy actions being taken to tighten (loosen) the macroprudential policy stance within the quarter, and a value of 0 represents no action is taken within the quarter.

If macroprudential action has the effect of reducing the credit gap, we would expect β2 < 0 in the equation above. Moreover, by construction, a negative change in Δ4RER represents a real exchange rate appreciation. If an appreciation is associated with an increase in the credit gap, we would therefore expect β1 μ 0.

A key focus of our investigation is the interaction between macroprudential action and the change in the real exchange rate MaPPi,t-1 x Δ4RERi,t-1. If macroprudential action is effective in containing the impact of the real exchange rate appreciation on the credit gap, we expect β > 0, with this effect thereby attenuating the negative coefficient on the real exchange rate.

We estimate the above equation using the Generalized Method of Moments (GMM) estimator developed by Arellano and Bond (1991), to address endogeneity concerns and avoid the Nickell bias4 arising in the presence of the lagged dependent variable. In addition, we verify that the Arellano-Bond approach is suitable for our purposes since all further conditions on its use are found to hold.5

We lag the MaPP variable by one-quarter, which is consistent with the approach in the previous literature (Akinci and Olmstead-Rumsey 2018; Cerutti et al., 2017; Fendoğlu 2017). We finally lag the real exchange rate, as well as all other independent and control variables to mitigate endogeneity concerns.

In our estimation, the first-differenced lagged dependent variable is instrumented with its 1–3 lags of its level. All independent and control variables are treated as predetermined rather than strictly exogenous.6 We use the forward orthogonal deviation transformation (Arellano and Bover 1995) to mitigate data gap issues in unbalanced panels. We also use two-step covariance estimates to obtain robust standard errors and to correct their downward bias (Windmeijer 2005).

3.2 Data and Variables

Our sample includes 62 economies (35 AEs plus 27 EMEs) as shown in Figure 2. These economies have sufficiently good data at quarterly frequency, not just on macroprudential policy measures, but also on credit and GDP. The sample covers economies that have taken frequent macroprudential policy actions (e.g., India, Korea, and Russia) and those that have rarely done so (e.g., Chile, Germany, and the United States). The sample period spans from 2000: Q1 to 2016: Q4, as macroprudential policy has been increasingly used across countries since the early 2000. To remove the effect of outliers, we winsorize the top and bottom 1 percent observations of each variable except the dependent variable and the ordinal variable MaPP. See Table 1 for description of variables, and Table 2 for summary of statistics.

Figure 2.
Figure 2.

Credit-to-GDP Gap and Usage of iMaPP Actions

(2000: Q1–2016: Q4, by Economy)

Citation: IMF Working Papers 2020, 187; 10.5089/9781513556550.001.A001

Source: Authors’ calculations.Note: Country classification is based on the latest WEO.
Table 1.

Description of Variables

article image
Note: (a) Initially take the first 24 quarters, then compute the trend and cyclical components recursively (add one quarter at a time).(b) We use the financial corporations’ domestic claims on private sector from IMF IFS where available, otherwise, we use depository corporations (or monetary) domestic claims on private sector from IMF IFS. The credit data of Iceland and Taiwan POC are from the central banks.(c) The weighted average of variable X is calculated as X_WA= 0.4*X + 0.3*L1.X + 0.2*L2.X + 0.1*L4.X
Table 2.

Summary of Statistics

article image
Source: Authors’ calculations.Note: We winsorize the top and bottom 1 percent observations of each variable except the dependent variable Y and the ordinal variable MaPP.

Dependent variable (Y): Our domestic credit measure is the credit-to-GDP gap, which is the quarterly credit-to-GDP ratio relative to its long-run trend. Following the method proposed by the BCBS (2010), we calculate the gap from a one-sided HP filter using a long-run smoothing parameter λ=400,0007. Credit is broadly defined in this paper as total claims on the private non-financial sector from both banks and non-bank financial institutions to capture all domestic sources of debt funds for the private sector.8 We use the financial corporations’ domestic claims on private sector from the IMF’s International Financial Statistics (IFS) database where available, otherwise, we use depository corporations’ (or monetary) domestic claims on the private sector from the same source. For brevity, we refer the credit-to-GDP gap as “credit gap” in the rest of this paper. We consider the simpler 4-quarter change of the credit-to-GDP ratio in a robustness check.

Forecasted YoY growth of real GDP (Δ4F_RGDP): We include the consensus forecast of future GDP growth as a control, since this serves to mitigate a potential simultaneity problem, when “good news” about the economy leads the exchange rate to appreciate and at the same time stimulates credit. We construct the forecasted quarterly year-over-year real GDP growth by taking a weighted average of the current year’s and next year’s forecasted growth rates from the Consensus Forecast.9 For five countries (Iceland, Lebanon, Luxembourg, Malta, and Mongolia) for which consensus forecasts are not available, we apply the same weighted average method but use the realized forward growth rates instead as a “perfect foresight” measure. Optimism with respect to short-run economic outcomes is expected to drive up both credit demand and supply, so a positive coefficient is expected. We use the actual GDP growth rates in a robustness check.

Monetary policy stance (MPS): We use (lagged) central bank policy rates to capture the monetary policy stance. For countries that have implemented unconventional monetary policies during the sample period (Euro Area, U.S., U.K., and Japan), we use the so-called shadow policy rates estimated by Krippner (2016). As a monetary policy tightening is generally found to reduce aggregate demand and increase the cost of borrowing, we expect a negative coefficient.

YoY change of real exchange rate (Δ4RER): We use the (lagged) year-over-year log change of the weighted average of the bilateral nominal exchange rate prevailing over the past four quarters10, which is denoted in national currency relative to the USD, and deflated by the U.S. consumer price index (CPI) against domestic CPI. An appreciating real exchange rate can fuel the build-up in credit through multiple channels as described in section I. By convention, a negative change in Δ4RER represents a real exchange rate appreciation, so we expect a negative coefficient.

Macroprudential policy stance (MaPP): The data source for macroprudential policy actions is the IMF’s iMaPP database (Alam et al., 2019), which is, to the best of our knowledge, the most comprehensive database of macroprudential policies to date (covering 17 instruments for a total of 138 counties over the period 1999–2016 at a monthly frequency). We consider an aggregate measure of the macroprudential policy stance (iMaPP) as well as two subgroups: borrower-based tools (MaPP_Br) and financial institutions-based tools (MaPPJl). Moreover, the iMaPP database allows us to distinguish policy adjustments that amount to a tightening and those that lead to a loosening of macroprudential constraints. Details of the individual macroprudential instruments covered in each subgroup are available in Table 3 and Figure 3.

Table 3.

Definition of Macroprudential Policy Instruments

article image
Source: Alam et al. (2019). We exclude SIFI from the original database to construct the aggregated macroprudential policy measure because this paper focuses on the time dimension of systemic risk.
Figure 3.
Figure 3.

Number of Tightening and Loosening Macroprudential Policy Actions

(2000: Q1–2016: Q5, All Sample Economies)

Citation: IMF Working Papers 2020, 187; 10.5089/9781513556550.001.A001

Source: IMF iMaPP Database, Alam et al. (2019), authors’ calculations. Excluding SIFI from the original database.Note: Country classification is based on the latest WEO.

3.3 Endogeneity of Macroprudential Policy

A common challenge faced by the literature on the effects of macroprudential policy is the problem of endogeneity, and more specifically that of potential reverse causality (Galati and Moessner, 2018; Alam et al., 2019). Macroprudential policy actions are not taken in a vacuum, but may be taken in response to macroeconomic and financial developments, which may be the same variables used to assess their effects. In our context, when we want to assess the effect of the policy action on the credit gap, but the credit gap is used as a signal for policy actions by policymakers, this can result in reverse causality. When high values of the credit gap are likely to result in tightening, this could induce a positive correlation that would bias the estimates on the impact, which are expected negative, up towards zero (attenuation bias).

We mitigate the risk of biased estimates due to endogeneity in four ways (with the first two ways being commonly applied in the literature (e.g., Claessens et al. 2013, Cerutti et al., 2017)):

  • In our baseline set up, we lag the macroprudential indicator and control variables by one-quarter and also include the lagged dependent variable;

  • We use the Arellano-Bond difference GMM methodology, which is suitable for independent variables that are not strictly exogenous;

  • We focus on the interaction term of MaPPi,t-1 x Δ4RERi,t-1. This should suffer less from an endogeneity bias, on the assumption that changes to exchange rates are not commonly taken into consideration when setting macroprudential policy. The change in the exchange rate then functions as exogenous shifter of the effect of prior macroprudential action, reducing the potential endogeneity problem;11 and

  • Finally, in further analysis (presented in section V), we construct macroprudential policy shocks, by taking the difference between the actual macroprudential indicator and its expectation, conditional on development in the credit gap, as well as the exchange rate. The advantage of this approach is that, by construction, these shocks are orthogonal both to the credit gap and the change in the exchange rate.

IV. Empirical Findings of the Baseline

Table 4 presents the baseline regression results on the effects of the exchange rate and macroprudential policy on the credit gap.

Table 4.

Baseline Results

article image
Source: Authors’ calculations.Note: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.

We show results for the aggregated index of macroprudential policy action, which includes all measures (iMaPP) and two subgroups of macroprudential policy tools, the borrower-based tools, MaPP_Br, and the financial institutions-based tools, MaPP_FI. The first two columns shown for each group are based on a measure of net tightening (tightening actions net of loosening actions); and the last two columns of each group separate the tightening (T) and loosening actions (L). We emphasize four results:

First, exchange rate movements have a measurable effect on domestic credit developments (column 1). In particular, a 10 percent real exchange rate appreciation is associated with a subsequent increase in the credit gap of 0.5 percentage points of GDP. This is on the same order of magnitude as the median credit gap in the sample (0.33 percent of GDP) and therefore economically meaningful. The size of the effect turns out to be robust across different specifications (with the size of the effect measured as between 0.5–0.6 percentage points) and statistically highly significant (typically at the 1 per cent level).

Second, macroprudential policy has a direct effect on domestic credit developments. In particular, a net tightening of macroprudential policy is estimated to decrease the credit gap by 0.875 percentage points of GDP in the next quarter, which again exceeds the median and is roughly 0.08 of the standard deviation of the credit gap. Looking at different groups of macroprudential policy, the effect is stronger for borrower-based than financial institutions-based tools (columns 5 and 9), in line with prior literature. Considering tightening and loosening actions separately (column 3), the effect is strong and significant for tightening actions (1.168 percentage points of GDP) but insignificant for loosing actions.

Third, in addition to having a direct impact on the credit gap, macroprudential policy has the effect of weakening the extent to which exchange rate movements impact credit developments. This effect is reflected in the coefficient of the interactions term MaPPi,t-1 x Δ4RERi,t-1 which shows a strong and statistically significant effect of macroprudential policy in mitigating the effect of the exchange rate on the credit gap. This mitigating effect holds for both the aggregated and the two subgroups of macroprudential policy, that is for both borrower-based and financial institutions-based policies. In economic terms, a one standard deviation increase in iMaPP is estimated to reduce the sensitivity of the credit gap to real exchange rate movements by 0.82 percentage points of GDP. This effect is again stronger for the borrower-based tools, with a one standard deviation increase in such tools being estimated to reduce the sensitivity of the credit gap to the real exchange rate by 1.4 percentage points of GDP, compared with 0.9 percentage points of GDP for the financial institutions-based tools.

Fourth, we find evidence that a prior relaxation of macroprudential policy can have a strong and significant effect in increasing the extent to which an exchange rate shock affects the credit gap. This is evident from looking at the interaction terms for tightening and loosing actions separately (column 4). Columns 8 and 12 reveal that this effects appears to be driven by loosening of financial institutions-based tools rather than borrower-based ones.

As for the control variables, the estimated coefficients have the expected sign and are highly significant: A one percentage point tightening of monetary policy is estimated to reduce the credit gap by about 0.24–0.29 percentage points of GDP, while expectations of improved macroeconomic conditions increase the credit gap, with a one percent improvement in annual forecasted real GDP growth leading to an increase of the credit gap by about 0.45–0.5 percentage points of GDP.

The coefficients of the lagged dependent variable appear close to unity, but the Fisher-type panel unit root test confirms that the credit gap variable is stationary (in at least one panel).12 Also, the large autoregressive coefficient is partially due to the setting of forward orthogonal deviation transformation in the GMM estimation.

Furthermore, the instrument lag choice is validated by the p-value of AR(1) and AR(2) at the bottom of Table 4. Result yields small p-values of AR(1) about 2 percent, so the null hypothesis of no first order autocorrelation in first differences is rejected as expected. The instrument lag choice yields all AR(2) p-values above the 10 percent threshold (ranging from 23–37 percent ), so the null hypothesis of no first order autocorrelation in levels (AR(2)) is not rejected, suggesting the second lags are appropriate instruments for their current values.

V. Further Analysis and Robustness Checks

We perform a number of further analysis and tests in order to examine more deeply the relationships identified in the main results and to assess their robustness. These findings are presented in Tables 5–12.

Table 5.

Robustness—Results with Actual GDP Growth Rates

article image
Source: Authors’ calculations.Note: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Table 6.

“Purging” the Real Exchange Rate

article image
Source: Authors’ calculations.Note: (a) Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1(b) FE Regression (1): Δ4RERi,t = β1 Δ4Inflationi,t + β2 Δ4F_RGDPi,t + ηi + εi,t(c) FE Regression (2): Δ4RERi,t = β1 Δ4Inflationi,t + β2 Δ4F_RGDPi,t + β3Δ4CA_Deficiti,t + ηi + ei,t
Table 7.

Robustness—Results with “Purged” Exchange Rate Shocks

article image
Source: Authors’ calculations.Note: (a) Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1(b) FE Regression (1): Δ4RERi,t = β1 Δ4Inflationi,t + β2 Δ4F_RGDPi,t + ηi + εi,t(c) FE Regression (2): Δ4RERi,t = β1 Δ4Inflationi,t + β2 Δ4F_RGDPi,t + β3Δ4CA_Deficiti,t + ηi + ei,t
Table 8.

Results from Ordered Probit Regression

article image
Source: Authors’ calculations.Note: (1) Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1(2) The constant cut points represent the thresholds of predicted cumulative normal distribution of the dependent variable corresponding to its different categorical values.
Table 9.

Robustness—Results with Macroprudential Policy Shocks

article image
Source: Authors’ calculations.Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Table 10.

Effectiveness of Macroprudential Policies by Country Characteristics

article image
Source: Authors’ calculations.Note: (a) Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1(b) We use the de facto financial account openness index developed by Lane and Milesi-Ferretti (2017). We take the average index for each country over the sample period (we take 2015 value for the index in 2016). The more financially open (closed) non-G7 economies are categorized as those having an average index above (below or equal to) the sample median
Table 11.

Robustness—Alternative and Simpler Measure of Credit Developments

article image
Source: Authors’ calculations.Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Table 12.

Effectiveness of Macroprudential Policies by Individual Instrument

article image
Source: Authors’ calculations.Note: (a) Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1(b) The coefficient of loosening action of “other” tools is omitted due to its insufficient action number, as indicated in the table.

5.1 Using Actual Rather than Forecasted GDP Growth

In the baseline result of Table 4, we control for the impact of the forecasted real GDP growth on domestic credit development. In this section, we replace the forecasted growth rates with the realized growth rates, i.e., the (lagged) year-over-year log change of quarterly real GDP from the IMF World Economic Outlook (WEO) database (Δ4RGDP). Actual GDP growth rates are more commonly used in the existing literature (Fendoğlu 2017; Cerutti, Claessens, and Laeven 2017). However, they may, in the context of our exercises, be inferior as a control for potential simultaneity issues that arise when both credit and the exchange rate changes are being driven by economic expectations, e.g., related to good news about economic prospects.

Overall, results presented in Table 5 are consistent with Table 4. However, they also suggest that GDP forecasts are a much stronger control than past realizations of GDP. While the coefficient of Δ4GDP is positive and statistically significant, as was found in other studies, the size of the coefficient is only about a half of that for the GDP forecast: a one percent increase in the actual real GDP growth rates leads to an increase of the credit gap by about 0.24–0.29 percentage points of GDP, compared with about 0.45–0.5 percentage points of GDP for the forecasted real GDP growth. This suggests that the evolution of credit gaps is more closely related to the expectation of future growth than to past growth in GDP. Moreover, and by contrast, the coefficient on the real exchange rate movement is somewhat larger when we use actual GDP compared to when we use the growth forecast, consistent with the idea that the larger coefficient is capturing some of the “good” news effects we control for when using forecast GDP. The coefficient of the interactions term MaPPi,t-1 X Δ4RERi,t-1, on the other hand, roughly remain with the same magnitudes and levels of significance.

5.2 Distilling Exchange Rate Shocks

We take our examination of potential simultaneity one step further by accounting more fully for economic fundamentals that may simultaneously be driving both the real exchange rate and credit developments. For instance, it is conceivable that, in addition to growth expectations, inflation and current account developments affect the exchange rate, and these might also affect the credit gap. This would result in a simultaneity bias in that the coefficient on the exchange rate would reflect both the causal effect and the correlation induced through this simultaneity, thereby overstating the former. While this need not be the case, this bias could conceivably then also affect the interaction term between the exchange rate and macroprudential policy.

To address this concern, we attempt to distill more “exogenous” exchange rate shocks for use in our main regressions. The proposed two-stage procedure we use here is similar in spirit to those used in Auerbach and Gorodnichenko (2013) for fiscal policy, Furceri et al. (2016) for monetary policy, and Ahnert et al. (2018) for macroprudential policy: First, we “purge” the impact of domestic fundamental factors on the real exchange rate by running a fixed-effect regression of the exchange rate on those domestic fundamentals13, and in the second stage we use the residuals from this regression to replace Δ4RERi,t-1 in our baseline dynamic panel regression with those (lagged) “purged shocks”. The first stage regressions have the following forms:

Δ4RERi,t=β1Δ4Inflationi,t+β2Δ4F−RGDPi,t+β3Δ4CA−Deficiti,t+ηi+ei,t

Where Δ4inflation is the year-over-year change of the CPI, Δ4F_RGDP is the forecasted real GDP growth, and Δ4CA_Deficit is the year-over-year change in the current account deficit (positive values entail a greater deficit while negative values a move towards surplus). Data availability means that the size of the sample is slightly reduced from 62 to 60 economies when the current account variable is included in the first-stage regression.

Table 6 shows the results of the first stage. Unsurprisingly, while individual coefficients on some of the fundamentals come out significant, the overall explanatory power of the first-stage regressions is low (R-square is about 0.04), in line with the well-known notion that exchange rate movements are difficult to explain, and are therefore to a considerable extent “random” (Rossi, 2013). On the other hand we find that both growth forecasts and inflation developments individually are significant in explaining movements in exchange rates, with the effect of the GDP forecast overall the strongest.

Table 7 shows the results of the second stage, using the residuals obtained from the first-stage regressions. This shows that, while there is a very slight drop in the size of the coefficient on the exchange rate, relative to the baseline, the results on this effect continue to hold strongly. Moreover, the ability of macroprudential policy to affect the credit gap directly and indirectly, by reducing the impact of the exchange rate shocks on credit, continues to hold. Finally, we when we include inflation and changes in the current account in the main regressions these variables are not significant in affecting the credit gap, and the results on other variables are mainly unchanged (results not shown).

In addition, we have attempted to instrument the domestic real exchange rate fluctuations in the baseline regression using a standard instrumental variable approach, with potential instruments in the first stage regression including time fixed effects, average changes in real exchange rates against the dollar in countries within the same region, as well as changes in commodity prices from Gruss and Kebhaj (2019). However, the results throughout suffer from a “weak instruments” problem, in that the correlation of the predicted values with the actual change in real exchange rates is low, then reducing also the correlation with credit developments that we investigate in the second stage. We take away from these exercises that idiosyncratic variation in exchange rate movements account for an important part of the co-movement between exchange rates and the credit gap.

Overall, we conclude that it is important to control for potential simultaneity in our context, and that including GDP forecasts in the specification is already a quite powerful way of doing so. Constructing more explicit exchange rate shocks can also be useful, but does not lead to major changes in the estimates when comparted to a baseline that already controls for growth expectations. Finally, we find that idiosyncratic variation in exchange rate movements account for an important part of the co-movement between exchange rates and the credit gap.

5.3 Using Macroprudential Policy Shocks

As set out above, a common concern when estimating the effects of macroprudential policy on credit is the reverse causality that arises when macroprudential policy is not random, but reacts to credit developments. When we want to measure the effects of policy action on credit developments, this can lead the estimated coefficients to be biased towards zero -the so-called attenuation bias (see also Alam and others, 2019). While our baseline regression already attempts to mitigate this bias, by using the lag of the macroprudential indicator, we here go a step further, by using macroprudential policy “shocks.”

Specifically, the identification of macroprudential policy shocks follows a three-step method closely following Brandao-Marques and others (2020), and related again also to the literature that computes policy shocks for monetary (Furceri et al., 2016) and fiscal policy (Auerbach and Gorodnichenko, 2013):

  • Step 1: we estimate an ordered probit model of the macroprudential policy indicator variable14 conditional on observables. As independent variables we use the year-over-year change of the quarterly credit-to-GDP gap, the change of the real exchange rate (Δ4RER as used in the baseline), the quarterly change of net capital inflows (as percent of GDP), an indicator of lagged policy actions (the sum of lags one to four of the quarterly macroprudential policy indicator), and a country indicator to capture cross-sectional heterogeneity. The ordered-probit (first stage regression) is shown in Table 8.

  • Step 2: we compute the “expected” macroprudential policy stance using the probabilities obtained from the ordered probit regression conditional on the independent variables.

  • Step 3: we compute the macroprudential policy shocks as the actual macroprudential indicators minus their expected values. Thus, positive values represent tightening shocks and negative values represent loosening shocks.

By construction, the shocks are orthogonal to credit developments, as measured by the past changes in the credit gap, helping to reduce endogeneity. In addition, importantly, they are orthogonal to exchange rate changes, thereby mitigating concerns that macroprudential policy might respond to exchange rate movements.

When we replace the variable MaPPi,t-1in the baseline regression with the macroprudential policy shocks identified above (MaPP_Shocki,t-1), the estimation result, shown in Table 9, is roughly consistent with the baseline. In particular, all interaction terms are essentially the same. However, at the margin we find that the base effects of macroprudential action are measured larger and more significant compared to the baseline. This holds in particular for the coefficients on the aggregate indicator and the financial institutions-based indicator, suggesting that these coefficients are measured with less bias when using the shocks.

When inspecting the first-step regressions (Table 8), consistent with this, we find that the coefficient on the change in the credit gap is sizable and statistically significant at the five and one per cent levels in the regressions explaining the overall and financial-institutions-based indicators, respectively, while the credit gap is not significant in the regressions explaining the use of borrower-based tools. This suggests that the former actions respond more strongly to credit developments, relative to the borrower-based tools, creating a greater potential for attenuation bias in the measurement of policy effects. As regards the first-step results, it is worth noting also that the change in the exchange rate does not enter statistically significant in any of the regressions, in line with our prior that macroprudential policy does not tend to react to movements in the exchange rate.

Overall, we find that when using macroprudential policy shocks, the attenuation bias from reserve causality is reduced. This affects mainly the base effect of macroprudential policy, and in particular the financial-institutions-based tools while interactions of all types of tools with the change in the exchange rate are not affected substantially when using macroprudential policy shocks in place of the indicator we have in the baseline.

5.4 Effects of Macroprudential Policies by Country Characteristics

We next examine whether the effects vary by country characteristics (Table 10). We initially divide the country sample into two groups: the G7 countries, which are considered to be the source countries of global capital flows, and the non-G7 group, which tend to receive these flows.

We first present results for specifications that use both macroprudential policy shocks and exchange rate shocks, as introduced separately in the sections just above, for the full sample of countries. The results, shown in columns 1–4, document once again that macroprudential policy shocks have significant effects in reducing credit, and that there is an interaction, where in the presence of a shock to the domestic real exchange rate, a macroprudential tightening reduces the expansionary effect of the appreciation on the credit gap.

The results for different sample splits are shown in columns 5–12. Overall, the effects are strong for the non-G7 group, but not for the G7. In particular, the base effect of the exchange rate on credit developments is significant and strong for the non-G7 group while it is insignificant for G7 countries (as indicated by the coefficients of Δ4RERi,t-1 in columns 9–12). The same holds for the extent to which tightening macroprudential policy can weaken the interactions between currency and credit movements, being strong and significant for non-G7 but insignificant for G7 countries.

We further divide the non-G7 group into more and less financially open economies15 and find that the interaction effects between macroprudential policy and exchange rate shocks remain strong and significant in the more financially open group, despite the reduction in the sample size to 27 countries, while both base and interaction effects are measured statistically insignificant for the group of relative more financially closed economies (columns 13–21). This leads us to conclude that the effects are strongest for small open economies, and less relevant statistically for either advanced economies or relatively closed EMDEs.

5.5 Alternative Measure of Domestic Credit Developments

We finally consider an alternative and simpler measure of domestic credit developments. Although the credit gap is a well-established broad-based indicator of systemic risk in the time dimension (Drehmann, 2013), it relies on a statistical filtering of the aggregate credit series that is subject to well-known issues regarding end-date biases, structural breaks, and the parameters driving the filtering. To side-step these criticisms, we employ an alternative and simpler measure, which is the four-quarter change of the credit-to-GDP ratio. As does the credit gap, this measure continues to relate credit aggregates to the size of the economy. It does not however, rely on any filtering method.

When using this alternative outcome variable, our specification continues to employ exchange rate shocks, macroprudential shocks, as well as monetary policy and GDP forecasts as controls. The results, shown in Table 11, document that currency appreciations are associated with increases in credit also using this alternative measure, with results for increases in the ratio of credit to GDP as statistically significant as those for the credit gap. The direct effects of macroprudential policy on the alternative measure of credit appear somewhat weaker, even as the effect remains statistically highly significant for the borrower-based tools. The strength of the interaction effects, on the other hand, is quite similar to the results using the credit gap, and economically if anything somewhat larger. For example, the coefficient ofMaPPi,t-1 x Δ4RERi,t-1 in column 2 is 0.446 in the bottom panel, compared with 0.256 when using the credit gap (upper panel).

Looking at different groups and specifications of macroprudential policies, the detailed results line up slightly differently. For instance, relaxations of macroprudential policy continue to strengthen the effect of appreciation on credit and this now applies to both borrower-based and financial institution-based tools. Overall, however, the headline results on the link between exchange rates and credit and the interaction effects of macroprudential policy in reducing this link carry over to this alternative measure of credit.

We finally return to the credit gap as the outcome variable and report the results for individual policy tools. Table 12 indicates strong and significant interaction effects for caps on loan-to-value (LTVs) in particular. Those results should be interpreted with caution, however, since when it comes to individual tools, the number of macroprudential actions recorded in the database is in general quite small.

VI. Extension: The Feedback Effect from Credit to Capital Inflows

Up until now, we have focused on the role of exchange rate shocks in driving domestic credit developments and the ability of macroprudential policy to attenuate these effects. In this section, we extend our analysis to examine feedback effects from credit developments to specific types of capital inflows and the extent to which these are influenced by domestic policy settings (right-hand side of Figure 1).

More specifically, we examine whether strong domestic credit (as measured by a high lagged credit-to-GDP gap) leads to increases in the so-called “gross other investment inflows” that capture cross-border borrowing by domestic financial institutions and non-financial corporates, and how different types of policies (including monetary policy, macroprudential policy, and capital controls) affect these flows.

As noted already, there is likely to be a two-way causation between capital flows and credit. On the one hand, strong credit demand may cause domestic financial institutions and corporates to “pull in” capital from abroad, as domestic funding sources are exhausted or constrained (Avdjiev, McCauley and McGuire, 2012, Hahm et al., 2013). On the other hand, where capital inflows are strong for other reasons, this may reduce the cost of domestic credit and thereby “push up” domestic credit supply (Mendoza and Terrones, 2012). In general equilibrium, both these effects would lead us to observe a positive correlation between measures of credit and measures of capital inflows.

In our analysis we regress gross other inflows on the lagged credit gap as one way of controlling for reserve causality. In addition, and importantly, we include quarter time-fixed effects to account separately for all global push factors. Inclusion of these quarter fixed effects should render the variation in “other capital flows” that is left to be explained by the credit gap (and other domestic variables) orthogonal to global push factors that might be driving an impact running from capital flows to increases in domestic credit.16 Despite this, it is difficult to rule out that the coefficient on the credit gap still captures the “push” effect that runs from inflows to credit to some extent.17 We therefore prefer to interpret the coefficient on the credit gap in the analysis below as measuring a conditional correlation that is consistent with a feedback relationship, rather than measuring a causal effect.

This analysis is embedded in otherwise standard “push-and-pull” capital flow regressions with domestic factors that have been identified in the existing literature as driving capital inflows, including monetary policy and growth expectations, as well as the ‘catch-all’ push factor discussed above. We consider three policy levers, macroprudential and monetary policy, as well as capital controls. The main variables we use are the following:

Capital inflows (CFLOW): As our dependent variable we consider gross other investment inflows18 because it is these types of flows that have been found to exhibit the most robust positive association with domestic credit (IMF 2017; Igan and Tan 2017). The capital inflows data are expressed in terms of gross other investment inflows within quarter t as a percent of GDP in the previous quarter t-1. The data source is the Financial Flows Analytics (FFA) database consolidated by the IMF Research Department.

Push factors: As discussed above, we include quarterly time-fixed effects (μt), following Ahnert et al. (2018), in order to account for global “push” factors that affect all countries equally to the fullest possible extent. The benefit of these time fixed effects is that they control for all global factors that are common across countries in each period. This can include changes in global risk aversion, the monetary policy stance in advanced economies, and other variables that may be difficult to observe. Time-fixed effects can help control for all these push factors in a parsimonious way.

Pull factors: In line with the existing literature, we include monetary policy and (forecasted) real GDP growth as domestic pull factors (with variable definitions and data sources as before). These variables are meant to capture to what extent tight monetary policy and a positive GDP growth outlook attract cross-border flows.

Capital controls (FARI): Different from the majority of previous studies that consider broad-based measures of capital controls and apply indicators on in- and outflow restrictions of all types of flows, we use an index of controls that are targeted at the type of “other investment (banking and corporate) inflows” we examine in the analysis. The relevant “Financial Accounts Restrictiveness Index” (FARI) is compiled by the IMF’s Monetary and Capital Markets Department, based on source data from the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER). Following Klein (2012), we distinguish between long-standing capital controls (“walls”), measured as the level of the index on “other investment inflows” for each country in a given quarter, and temporary adjustments (“gates”), measured as incremental changes (+1 for net tightening and -1 for net loosening actions) in such controls.

Our regressions are based on the two equations given below, where we regress gross other capital flows on the credit gap (Y), lagged by one quarter, and we also include lagged policy variables, with separate regressions for long-standing controls and episodic controls. As explained further just below, our main interest is on the interaction effect between measures of capital controls (i.e., the walls and the gates), and the credit gap: do such controls reduce the extent to which high domestic credit stimulates direct borrowing from abroad?

  • The “Walls” Effect of Capital Controls (levels, FARI) + MaPP
    CFLOWi,t=ρCFLOWi,t−1+β1Yi,t−1+β2FARIi,t−1+β3Yi,t−1×FARIi,t−1+β4MaPPi,t−1+β5Yi,t−1×MaPPi,t−1++β6MPSi,t−1+β7Yi,t−1×MPSi,t−1+θiΔ4F−RGDPi,t−1+θtμt+θiαi+vi,t
  • The “Gates” Effect of Capital Controls (1-quarter change, AFARI) + MaPP
    CFLOWi,t=ρCFLOWi,t−1+β1Yi,t−1+β2ΔFARIi,t−1+β3Yi,t−1×ΔFARIi,t−1+β4MaPPi,t−1+β5Yi,t−1×MaPPi,t−1++β6MPSi,t−1+β7Yi,t−1×MPSi,t−1+θiΔ4F−RGDPi,t−1+θtμt+θiαi+vi,t

Cerdeiro and Komaromi (2019) point out the difficulty of identifying the effects of capital controls. On the one hand, these controls do not vary much over time, reducing the power of standard fixed effects regressions. On the other hand, their level could be correlated with a number of country-specific factors, exposing a random effects regression to potential omitted variables. They argue in favor of an identification through interaction effects, by showing in a simple model that capital controls not only affect the unconditional mean of flows, but importantly the sensitivity of flows to various push and pull factors; a point that is usually not exploited in regressions that neglect interactions and assume an additive linear effect of capital controls on the level of capital flows.

Following Cerdeiro and Komaromi (2019), we include country-fixed effects, in addition to the time-fixed effects, in order to control as tightly as possible for time-invariant omitted variables at the country level.19 In addition, and in order to achieve identification of the effects of policies, including capital controls, we include interaction terms to analyze how policies interact with the credit gap (the domestic pull factor of interest) in affecting the level of other investment flows.

The estimation here does not use the previously applied dynamic panel (GMM) model, in the main since the assumption of autocorrelation within individual panel’s error terms vi,t is not satisfied in the estimated equation of the feedback effect. The p-value of AR(1) test ranges from 0.62 to 0.8, implying strong evidence against the first order serial correlation of residuals in differences, and thus would invalidate the moment conditions used in the dynamic panel estimation (Roodman, 2009). 20

Estimation results for the “Walls” effects of capital controls are presented in Table 13. Throughout, we include macroprudential and monetary policy variables in addition. Panel (a) reports the findings when we use the macroprudential indicators; panel (b) shows results for macroprudential policy shocks that account for reverse causality (as above in section 5.3); and panel (c) reports the finding of a robustness check applying the random effects approach. There are four noteworthy findings:

Table 13.

Leakages and the “Walls” Effect of Capital Controls

article image
article image
article image
Source: Authors’ calculations.Note: (a) Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1(b) We drop Taiwan POC in the sample due to missing data of financial account restriction index.(c) The blue italic numbers are standardized coefficients, representing the change of standard deviation in the dependent variable by one standard deviation change in corresponding independent variables.(d) Both of the capital inflows and capital control measures are based on the gross other investment inflows.