Estimated Policy Rules for Capital Controls
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: gpasricha@imf.org

This paper borrows the tradition of estimating policy reaction functions from monetary policy literature to ask whether capital controls respond to macroprudential or mercantilist motivations. I explore this question using a novel, weekly dataset on capital control actions in 21 emerging economies from 2001 to 2015. I introduce a new proxy for mercantilist motivations: the weighted appreciation of an emerging-market currency against its top five trade competitors. This proxy Granger causes future net initiations of non-tariff barriers in most countries. Emerging markets systematically respond to both mercantilist and macroprudential motivations. Policymakers respond to trade competitiveness concerns by using both instruments—inflow tightening and outflow easing. They use only inflow tightening in response to macroprudential concerns. Policy is acyclical to foreign debt; however, high levels of this debt reduces countercyclicality to mercantilist concerns. Higher exchange rate pass-through to export prices, and having an inflation targeting regime with non-freely floating exchange rates, increase responsiveness to mercantilist concerns.

Abstract

This paper borrows the tradition of estimating policy reaction functions from monetary policy literature to ask whether capital controls respond to macroprudential or mercantilist motivations. I explore this question using a novel, weekly dataset on capital control actions in 21 emerging economies from 2001 to 2015. I introduce a new proxy for mercantilist motivations: the weighted appreciation of an emerging-market currency against its top five trade competitors. This proxy Granger causes future net initiations of non-tariff barriers in most countries. Emerging markets systematically respond to both mercantilist and macroprudential motivations. Policymakers respond to trade competitiveness concerns by using both instruments—inflow tightening and outflow easing. They use only inflow tightening in response to macroprudential concerns. Policy is acyclical to foreign debt; however, high levels of this debt reduces countercyclicality to mercantilist concerns. Higher exchange rate pass-through to export prices, and having an inflation targeting regime with non-freely floating exchange rates, increase responsiveness to mercantilist concerns.

I. Introduction

Capital controls are restrictions on cross-border trade in assets. The recent global financial crisis has reignited the debate on systematic use of capital controls to manage domestic economic and financial cycles. Certain capital controls have increasingly been viewed as part of a broader policy toolkit for maintaining financial stability, i.e., as ex-ante tools to prevent buildup of systemic risk by limiting the growth of credit (BIS-FSB-IMF, 2011; G20, 2011; Ostry et al., 2011; and Ostry et al., 2012).2

This new approach is backed by a growing theoretical literature that views capital controls as optimal ex-ante policies that address the consequences of pecuniary externalities in residents’ borrowing decisions (Mendoza, 2002; Korinek, 2010; Korinek and Sandri, 2016; Bianchi, 2011; Uribe, 2007). In this framework, residents face a collateral constraint that depends on the real exchange rate. Individual agents take the real exchange rate (and the value of the collateral) as given when taking their borrowing decisions, but in aggregate, the real exchange rate depends on the borrowing decisions of the individuals. This feedback loop leads to excessive foreign borrowing in good times and increases the probability of a crisis. Capital controls that limit real exchange rate appreciation in cyclical upturns also limit excessive borrowing and are therefore viewed as macroprudential tools.

While much of the recent literature and the policy discussion has focused on the macroprudential objective of capital controls, there is another potential objective of capital control policy—the mercantilist objective.3 Here, the objective is to promote exports by manipulating the terms of trade or preventing foreign control of strategic industries (Bernanke, 2015; Bhalla, 2012; Costinot et al., 2014; Heathcote and Perri, 2016; Dooley et al., 2014; and Acharya and Bengui, 2018). Proponents of this view argue that attempts to prevent the exchange rate from appreciating are in fact motivated by the objective of gaining trade advantage over export competitors. Furthermore, the imposition of capital controls by one emerging-market economy (EME) during upturns in the global financial cycle can deflect capital flows to other emerging markets and can lead to a beggar-thy-neighbour currency war.4

Are capital controls macroprudential or mercantilist? This question is not merely of academic interest but has important implications for operationalizing the new consensus on limited, disciplined use of capital controls. As Carney (2019) notes, “There are two major challenges in operationalizing capital flow management measures. The first is proving intent so as not to provoke retaliation. The second is that they can panic investors and make matters worse.” However, there is surprisingly little empirical evidence on the intent with which these tools have been used by emerging markets. A recent paper by Fernández et al. (2015b) finds that capital controls do not vary over the business cycle. On the mercantilism objective, there is only indirect evidence that certain types of inflow controls benefit the largest exporting firms (Alfaro et al., 2017).

The paper asks: With which objectives—macroprudential or mercantilist—have policymakers in emerging economies used capital controls? It takes a policy reaction function approach, clearly delineating the different motivations, and the trade-offs therein. There is some recent literature that has tried to predict capital controls (Fernández et al., 2015b; Fratzscher, 2014; Forbes et al., 2015; Aizenman and Pasricha, 2013). However, these papers focus on specific variables to which policy responds, not on identifying mutually exclusive motivations that these variables represent.5 For example, the papers assess whether policy reacts to net capital inflows (NKI) and find that it does. But the motivation behind that NKI response could be macroprudential or mercantilist. This paper estimates a descriptive, empirical policy reaction function to explore how policy reacts to these competing objectives.

The idea of asking how policy should or does react to competing objectives is well established in economics. Monetary economics has a long tradition of estimating monetary policy rules (e.g., Taylor, 1993). The premise is that well-designed policy rules can allow policymakers to overcome time-inconsistency problems with monetary policy, gain credibility and therefore make policy more effective. Policy rules can also allow policymakers to communicate policy more effectively and enhance accountability of the monetary authority. In a similar vein, recent literature has explored the time-inconsistency of domestic macroprudential policy under commitment (Bianchi and Mendoza, 2016). They find that the optimal time-consistent policy is a complex function of time and state-dependent variables, and that well-optimized financial Taylor rules, while less effective than optimal policy, can improve welfare over no-intervention.

This paper estimates a descriptive reaction function, without claiming that such reaction functions reflect optimal rules. Even without an assessment of optimality, this exercise is important as it contributes to improving the transparency of policy.6 Transparent policy reaction functions can help attract capital inflows and prevent destabilizing outflows when the controls are used, by constraining the ability to expropriate past investments (Ljungqvist and Sargent, 2004).7 They can also strengthen the accountability of the macroprudential authority, assuage concerns about the spillovers of such policy, and prevent retaliation by other countries by establishing intent.

To disentangle mercantilist from macroprudential motivations, the paper introduces a novel proxy for mercantilist concerns: the appreciation of an EME’s currency against its top five trade competitors. EMEs’ use of capital controls to prevent REER appreciation or appreciation against the U.S. dollar could in theory, reflect the desire to prevent an increase in collateral value (as envisaged in recent literature) or the desire to promote exports or protect import-competing industries.8 Both these measures therefore suffer from the shortcoming that they could reflect both macroprudential and mercantilist motivations (as most EME agents are able to borrow only in hard currencies of countries which are also main export destinations and import suppliers for these EMEs). To get cleaner identification, I propose a novel proxy for mercantilist concerns that measures the appreciation of an EME’s currency against its top five trade competitors. As these competitors are emerging or developing countries, in whose currencies the EMEs do not borrow, the movements of the EME currencies against the currencies of these countries do not reflect macroprudential concerns but capture only mercantilist concerns. This proxy is positively correlated with, and Granger causes initiations of non-tariff barriers in most countries, further bolstering its usefulness as a mercantilism proxy. I also construct a version of the proxy that is orthogonal to appreciation against the US dollar.

A final contribution of the paper is that it uses a detailed weekly dataset on capital controls policy that directly measures policy actions by 21 major emerging market economies over the period 2001–2015. I extend the Pasricha et al. (2018) dataset by three years, 2013–15, and use the announcement dates of the policy actions, rather than the effective dates used in Pasricha et al. (2018). The use of data on policy actions also closely parallels the monetary literature on modeling central bank policy rate.

The paper has several new and interesting results on the use of capital controls in emerging markets. The results provide evidence that capital controls policy in emerging economies has been systematic, and that it has responded to both macroprudential and mercantilist motivations. Moreover, I find that the choice of instruments is systematic: policymakers respond to mercantilist concerns by using both instruments—inflow tightening and outflow easing. However, they use only inflow tightening in response to macroprudential concerns. This paper provides evidence that the macroprudential motivation existed in the use of capital controls policy, even before these controls were generally acknowledged (after the global financial crisis) as valid tools of the macroprudential policy toolkit. Yet, the results in this paper underline that the concerns about a currency war are also justified—capital controls have also been systematically used to preserve competitive advantage in trade.

Further, I find that policy is not countercyclical to the specific macroprudential concerns related to external or foreign currency borrowing. Rather, policy appears acyclical to these variables. However, foreign currency debt matters in a non-linear way—countries with very high foreign currency debt have a different reaction function: countries in these states respond less countercyclically to mercantilist motivations and somewhat countercyclically to macroprudential motivations, as measured by changes in foreign currency debt.

Finally, I find that the mercantilism objective is stronger in countries with higher exchange rate pass-through (ERPT) to export prices, and in inflation targeting countries that do not have a freely floating exchange rate. Higher ERPT to export prices means that exporters do not change the prices in their domestic currency much in response to appreciation of their currency.9 As a result, the customers of these countries face much of the cost of the currency appreciation, potentially making the exports of these countries more sensitive to appreciation. I find that countries with high export price ERPT react more strongly to mercantilist motivations, particularly when the exchange rate pressures against competitors are strong. Inflation targeting countries are limited in using monetary policy to manage exchange rate, and may rely more on capital controls for this purpose.

The rest of the paper is organized as follows. Section II reviews the literature on the motivations for capital controls. Section III reviews describes the new the mercantilism proxy. Section IV describes the data on capital control policy actions. Section V describes the empirical strategy and the data on other macro financial variables. Section VI describes the results and assesses the fit of the baseline models. Section VII evaluates robustness of the main results. Section VIII concludes.

II. Literature Survey: the Motivations for Capital Controls

The literature identifies two main motivations for using inflow side capital controls: mercantilist and macroprudential.10 In this section, I survey the empirical and theoretical literature on each of these motivations.

A. Macroprudential Motivation

Macroprudential policy is defined by an objective—that of addressing systemic risks in the financial sector to ensure a stable provision of financial services to the real economy over time (BIS-FSB-IMF, 2011). Under this policy framework, capital controls could be considered tools of macroprudential policy if they specifically target the systemic risks stemming from external finance, particularly those that could be addressed using other (non-residency-based) prudential tools. To do so, these tools would be used counter-cyclically to systemic risk, as is done with other macroprudential policy tools.

A large recent literature has explored the role of capital controls as macroprudential tools, in models where agents face pecuniary externalities arising from occasionally binding collateral constraints (Bianchi, 2011; Jeanne and Korinek, 2010; Benigno et al., 2011; Korinek, 2016; Schmitt-Grohé and Uribe, 2016a). As the probability that the collateral constraint will bind increases with the level of debt, some models recommend that the capital controls be set to positive values once debt levels are high, or have crossed a threshold (Bianchi, 2011; Korinek, 2011). Agenor and Jia (2015) show that a simple countercyclical capital controls rule, in which the tax on foreign borrowing is countercyclical to the changes in foreign borrowing by banks, performs well relative to the Ramsey optimal policy.11

Assessing whether capital controls have been used as macroprudential tools would therefore necessitate the assessment of whether these tools were countercyclical to measures of systemic risk. Measures of systemic risk may include, but are not limited to, credit-to-GDP gap, levels or growth of foreign credit—in particular, foreign currency or short-term credit— and asset price booms.12 Following the literature, therefore, I use countercyclicality to various measures of systemic risk as the measure of macroprudential objective, and propose a new proxy for mercantilist objective, discussed in the next section.

B. Mercantilist Motivation

Mercantilist motivation can be understood as the strategy to promote export-led development by keeping the exchange rate undervalued, through a combination of capital controls and reserves accumulation (Dooley et al., 2003, 2014). A large empirical literature has tested the macroeconomic versus prudential motivations for foreign exchange reserves accumulation, a policy complementary to or a substitute for capital controls (Aizenman and Lee, 2007; Choi and Taylor, 2017; Ghosh et al., 2012; Cheung and Qian, 2009; Jeanne and Ranciere, 2006). In this literature, export growth rates and exchange rate undervaluation relative to fundamental purchasing power parity value are used as proxies of mercantilist motivation, with higher levels of reserves associated with greater undervaluation and greater export growth. This works because these regression specifications focus on explaining cross-country differences in levels of reserves. If the mercantilist strategy is successful, one would expect countries that ended up accumulating larger reserves hoardings to have seen higher export growth and undervalued exchange rates. Yet this does not directly translate into a policy strategy: should countries intervene more (through reserves accumulation or capital controls) when export growth is high or when it is lagging?

An alternative would be to use measures of exchange rate appreciation (nominal or real). However, as discussed above, the recent theoretical literature on macroprudential capital controls views the target of macroprudential policy as encompassing targeting the REER, or even the nominal exchange rate. It views exchange rate appreciation as the channel that facilitates over-borrowing, especially foreign currency borrowing. These models imply that simply finding that policy responds to exchange rate doesn’t imply policy is mercantilist (or macroprudential).13 This suggests a need to explore other proxies of mercantilist motivations.14

III. A New Proxy for Mercantilist Motivations

In order to isolate the mercantilist motivation in exchange rate management, I propose a new proxy for mercantilist motivations. This proxy is the weighted appreciation of the exchange rate against a country’s top five trade competitors. When the exchange rate is appreciating against trade competitors, the EME can be interpreted as losing competitiveness in the world market. The reason this proxy works is that the trade competitors of most EMEs in our sample are other EMEs, and most EMEs do not borrow in the currencies of their trade competitors.15 In the terminology of the recent theoretical literature on pecuniary externalities, the collateral constraint is not denominated in the currencies of the trade competitors, rather in the base currencies (U.S. dollar or euro). Therefore, while resisting appreciation against the base currency (U.S. dollar or euro) per se could capture either mercantilist or macroprudential concerns, resisting appreciation against trade competitors should capture only the mercantilist motivation. To illustrate, an appreciation against the U.S. dollar would not be problematic from mercantilist perspective if the competitors’ currencies are appreciating faster, while it could still be problematic from a macroprudential perspective.

To identify trade competitors, I use the merchandise trade correlation index, developed by the United Nations Conference on Trade and Development (UNCTAD).16 The trade correlation index is a simple correlation coefficient between economy A’s and economy B’s trade specialization index and can take a value from -1 to 1. A positive value indicates that the economies are competitors in the global market since both countries are net exporters of the same set of products. A negative value suggests that the economies do not specialize in the production or consumption of the same goods and are therefore natural trading partners.17 The specialization index removes bias of high export values because of significant re-export activities; thus, it is suitable to identify real producers rather than traders.18

For each EME in our sample, I identify five countries with which it has the highest trade correlation index in each year. Next, I compute the weighted exchange rate appreciation of the EME’s currency against the five trade competitors, at quarterly and annual horizons, and in real and nominal terms. That is, I compute the following proxies:

(1)WAPPRQit=400*[j=15wijt{(xitL13xit)(xjtL13xjt)}]
(2)WAPPRYit=100*[j=15wijt{(xitL52xit)(xjtL52xjt)}]

And the two real proxies are defined as:

(3)WRAPPRQit=100*[j=15wijt{4(xitL13xit)4(xjtL13xjt)+(πit1πjt1)}]
(4)WRAPPRYit=100*[j=15wijt{(xitL52xit)(xjtL52xjt)+(πit1πjt1)}]

where xit and xjt are respectively the natural logs of the nominal exchange rate against the U.S. dollar for countries i and j as of the end of week t (measured in USD per domestic currency unit), L is the lag operator and πit is the year-over-year (52-week) change in log of consumer price index (CPI) as of week t, wijt is the weight assigned to competitor j and is measured by the trade correlation index between country i and country j in week t (and is the same for all weeks in a calendar year). The set of trade competitors (j) included in the calculation of the index varies over time but is reasonably stable over five-year periods in the sample.

The nominal proxies WAPPRQit and WAPPRYit measure the weighted nominal appreciation of a country’s currency over the previous quarter (13 weeks) and over the previous year (52 weeks), respectively. The real proxies are analogously interpreted. All proxies express the appreciation at annual rates and as percentages.

Finally, I compute a country-specific proxy, which uses for each country the mercantilism proxy (from equations 14 above) that is most important for that country, i.e., most highly correlated with capital control changes. I use this in the baseline models, and generally refer to this as the “Mercantilism Proxy (country-specific),” unless otherwise specified. That is, I compute the country-specific correlation coefficient (over the full sample period) between (weighted, non-FDI) net inflow tightening measures (WHNTIit), defined in the next section, and each of the four proxies defined above. Then that country’s mercantilism proxy is the series with the highest correlation coefficient. I call this proxy WAPPRit, with the understanding that it uses a different series for each country. I also construct a version of this proxy that is orthogonal to appreciation against US dollar, as described in section VI.A.

The mercantilism proxy achieves the objective of identifying mercantilist motivations separately from macroprudential motivations. While the REER is positively correlated for most countries with domestic bank credit to GDP gap and growth for most countries in sample, the mercantilism proxy is uncorrelated or negatively correlated with these variables (Figure 1). In robustness checks, I further validate the proxy as a measure of mercantilist motivations by testing its relationship with non-tariff barriers to trade.

Figure 1:
Figure 1:

Mercantilism Proxy is Uncorrelated or Negatively Correlated with Bank Credit to GDP Gap and Growth

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations.Note: Bank credit to GDP growth is the year over year change in domestic bank credit to the private sector as percentage of GDP. REER is the real effective exchange rate. Mercantilism proxy is as defined in the text.

IV. Measuring Capital Control Actions

I assess the motivations for capital controls by using a detailed dataset on actual policy actions, at a weekly frequency. I update the Pasricha et al. (2018) indices on capital control policy actions for 21 EMEs through 2015. This dataset uses a narrative approach—reading the text of the policy changes or descriptions of such changes in other sources—and converting them into numerical measures that capture the direction of policy. An advantage of this dataset compared to other high-frequency datasets on capital control actions in the literature, is that it is granular: policy announcements often contain changes of multiple regulatory instruments. These announcements are split along six dimensions to yield a granular database of policy changes or policy actions.19 The identified policy actions are then aggregated to compute measures of policy direction.

Most of the paper focuses on explaining (weighted, non-FDI) net inflow tightening measures (WHNTIit), as much of the policy debate and theoretical literature on macroprudential capital controls focuses on inflow restrictions. This measure is computed as the number of (weighted, non-FDI) inflow tightening minus easing actions per week. However, as both inflow tightening and outflow easing can be used to respond to competitiveness pressures, I also assess the motivations for (weighted, non-FDI) net NKI restricting measures, which is the sum of (weighted, non-FDI) net inflow tightening and (weighted, non-FDI) net outflow easing measures.20

It is important to note that are three main differences in the aggregation methods between the data used in this paper and the Pasricha et al. (2018) dataset. First, in this paper, I use the announcement dates of the changes, rather than their effective dates. Second, I drop changes that were pre-announced by more than 60 days, as changes that have more than a 60-day implementation lag are likely to be more structural in nature, rather than imposed for macroeconomic and macroprudential management. Third, in this paper, I include changes that potentially affect both inflows and outflows (e.g., currency-based measures) on both the inflow and outflow sides. That is, these changes are counted twice.

The resulting data shows a high degree of variation in policy, even in countries with extensive and long-standing capital controls. Figure 2 plots the cumulated versions of weighted net inflow tightening actions and weighted net outflow easing measures for China and India, two countries with extensive and long-standing capital controls. The figure shows that on the whole, both countries have taken more liberalization actions than tightening actions since 2001 on both inflow and outflow sides, but it also shows periods of tightening of inflow restrictions (2004–05, 2007–08 and again 2010–11 for China) as well as periods of tightening of outflow restrictions (2015, also for China).

Figure 2:
Figure 2:

Pasricha et al. (2018) Indices of Capital Controls Liberalization

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations. Note: Figures include policy actions related to FDI and exclude those that were implemented more than 60 days after announcement. Actions that affect both inflows and outflows are included in both red and blue lines. Last observation: 31 December 2015

Not all emerging markets were equally active in changing capital controls policies (Figure 3). In the baseline models for net inflow tightening actions, I use the 11 most active countries, i.e., those that had at least 30 policy actions in the 15-year period, with at least one inflow tightening.21 For baseline models for net NKI restricting actions, I use an additional 2 countries—South Africa and Malaysia—who have at least 30 actions in sample, but no inflow tightening. This choice of sample is based on the nature of the exercise. Although very interesting, the question we are exploring here is not why some countries rely more on capital controls as policy tools and others not at all—the answer may depend on the institutional arrangements and policy preferences in these countries as well as their international agreements (e.g., European Union or OECD). The question we are exploring here is whether the actions of countries that do use capital controls or currency-based measures are predictable based on certain macroeconomic and macro-prudential variables.

Figure 3:
Figure 3:

Baseline Models Include the 11 Most Active Countries

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations.Note: Black dots mark countries included in the baseline models. These are countries wiht at least 30 actions in sample and at least one inflow tightening action. Red shaed bard denote the total number of tightening actions (on inflows and outflows) by the country and blue bars are the total number of easing ctions (on inflows and outflows). This chart plots unweighted policy actions, includes FDI-related actions, and counts actions not specific to inflows and outflows only once.Last observation: 31 December 2015

V. Econometric Methodology and Data

A. Econometric Methodology

The baseline model is a panel ordered logit model of the form:

(5)Pr(yit=sk|wit1)=f{Xit1MPβMP+Xit1FXβFX+XtGβG+Xit1oβo},

where yit is the number of policy actions by country i in week t, Pr(yit = sk|wit-1) is the conditional probability that country i takes sk actions in week t. Xit1MPandXit1FX are the variables representing macroprudential (MP) and mercantilist (FX) motivations, respectively. XtG controls for the global variables and Xit1o controls for the other domestic variables.

In the baseline models, yit refers to either (weighted, non-FDI) net inflow tightening actions or (weighted, non-FDI) net NKI restricting measures. The weighting scheme for capital controls makes the number of policy actions per week a continuous variable, yet there is little difference in the strength of policy actions that are measured as, for example, 0.24 vs. 0.28. In the baseline models, I consolidate the number of ordered categories, the weighted capital controls variable into five, as follows:

yito={1if yit<0.50.5if 0.5yit00if yit=00.5if 0<yit0.51if yit0.5

The baseline models estimate equation (5) for yito. This transformation does not affect the main conclusions, as shown in the robustness checks, but makes the estimations substantially faster. The models are estimated using random effects, but the results are robust to adding country-specific dummies.22 The reported coefficients are proportional odds ratios—values more than 1 indicate countercyclical use of policies.

B. Macro-Financial Data

In the baseline model, I use one of the five mercantilism proxies described in section III to capture mercantilist motivations. For the macroprudential motivation, I use the domestic bank credit-to-GDP gap. This variable is defined as the deviation from a backward-looking HP-filtered trend of the ratio of domestic bank credit to private non-financial sector to GDP. The data on bank credit is from the Bank for International Settlements (BIS). The reason for choosing this variable as the main macroprudential variable is that it is viewed as a key indicator of systemic risk in the Basel III agreement, and comprises on average over 75 percent of total credit in the active countries over the sample period.23 The recent early warning literature on financial crises—for example, Jorda, Schularick and Taylor (2012)— also highlights the importance of bank credit as a measure of systemic risk.

To capture push factors (XtG), the baseline model includes the Chicago Board of Options Exchange Volatility Index (VIX).

Other domestic variables (Xit1O) include domestic variables that may capture the additional motivations for capital controls (for example, macroeconomic management), as well as variables that capture other domestic policies that are substitutes for or complements to capital control changes. To capture macroeconomic management motivation, I include the CPI inflation rate in all specifications. This variable captures the overheating pressures in the economy. I also include a crisis dummy, which equals 1 for the global financial crisis (2008: Q4) and for three domestic crises in Argentina (2001: Q1–2003: Q4), Russia (2001: Q1– 2001: Q4) and Turkey (2001: Q1–2004: Q1).

In terms of other domestic policies, reserves accumulation is the oft-used policy to manage the exchange rate, alongside or as substitute to capital controls, and is included in all specifications. To capture monetary policy stance, I use a dummy variable that takes the value 1 if the policy rate is increased in the quarter, 0 if there is no change in policy rate between the current and the previous quarter, and -1 if monetary policy is eased in the current quarter. As an increase in interest rates can make capital inflows more attractive, policymakers concerned about the value of the currency may simultaneously tighten inflow controls to curb the resulting appreciation pressures. A dummy for fiscal stance is similarly defined as taking the value +1 if the general government structural balance (as % of potential output) increased (reflecting tightening of fiscal policy), -1 if the fiscal stance eased, and 0 otherwise.

To the extent that the domestic policies are substitutes or complements for capital controls, we may expect them to be driven by mercantilist and macroprudential motivations. To test this, I ran fixed effects regressions with each of these policies as dependent variables and the mercantilist and macroprudential proxies as explanatory variables.24 The results are in Table 1 and show that for reserves accumulation, mercantilist proxies have significant explanatory power, but for other domestic policies (fiscal and monetary policies), neither of these variables have a significant linear relationship. Therefore, in the regressions explaining capital controls, I use the residuals from regression explaining reserves accumulation, to control for that part of reserves accumulation that may be driven by factors uncorrelated with our mercantilism and macroprudential proxies, but that may also drive capital controls, for example, sudden stop risk, financial globalization, ideology of the government in power.25 In later sections, I explore the role of some of these factors individually.

Finally, as in Hamilton and Jorda (2002), other domestic variables (Xit1O) also include an indicator variable that takes the value +1 if the previous policy action (whenever it was) was a tightening and -1 if the previous policy action was an easing.26 This variable captures the cycles in capital controls policy.

Table 1.

Reserves Accumulation is Partly Driven by Mercantilist Concerns

article image
Source: Author’s calculations.Note: All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. Monetary policy stance and fiscal policy stance are variables that take the value +1 if monetary policy is tightened in the previous week (or structural balance improves), -1 for expansionary policies and 0 otherwise. Robust standard errors in parenthesis. *** p<0.01, ** p<0.05, * p<0.10

A note on the frequency of the variables is in order. Exchange rates (and other financial variables used in the second stage) are available at a weekly frequency. However, many of the macro variables are available at a quarterly or lower frequency. These are interpolated to weekly frequency using linear interpolation. An alternative would have been to use the last available value, but that could mean using observations that are no longer relevant for policy decisions. Further, policymaking is a forward-looking activity. The literature on assessing motivations for changes in monetary policy suggests that the results using only lagged variables to explain policy may be biased if policy-makers anticipate future evolution of variables and act on that information: policy-makers may not only change capital controls in response to past changes in economic variables, but also respond to their expectations of future evolution of these (Ramey, 2016). The literature on Taylor Rules addresses this by using Fed’s Greenbook forecasts (Monokroussous, 2011 and others). However, such forecasts made by EME policymakers are not available. The interpolations assume that policymakers had information about the evolution of the economy that is not reflected in the previous quarter’s data, and that their forecasts are accurate on average.27

The data are collected from IMF BOPS, IMF WDI and GEM, UNCTAD, BIS macro-financial database, Haver and national sources. A full list of variable definitions and sources is in Appendix C.

C. Model Evaluation

I evaluate the predictive ability of the baseline model and alternative models using a standard criterion: the area under receiver operating characteristic curve (AUROC). The receiver operating characteristic (ROC) curve evaluates the binary classification ability of a model and has recently been used in early warning literature (Schularick and Taylor, 2012).

Let y*^ be the linear prediction of the latent variable from a binary logit model (i.e., one with a 0/1 dependent variable). Let predicted outcome be 1 whenever y*^ crosses a threshold c. That is, the predicted outcome=I(y*^c>0), where I(.) is the indicator function. Then, for a given c, one can compute the true positive rate TP(c) (i.e., the percentage of “1” observations that are correctly predicted to be “1”) and the false positive rate, FP(c) (i.e., the percentage of 0 observations that are incorrectly predicted to be 1). The ROC plots the true positive rate, TP(c), against the false positive rate, FP(c), for all possible thresholds c on the real line. The plot is a unit square, as both TP(c) and FP(c) vary from 0 to 1. Any point in the upper left triangle of the square (formed above a 45-degree line from the left corner of the square) has a higher true positive rate than a false positive rate. Therefore, an informative model is one where the ROC curve lies above the 45-degree line, that is, TP(c)>FP(c) for all thresholds c and the model always makes better predictions than the null of a coin toss. The closer the ROC curve is to the top left corner of the square, the better the model. The area under the ROC curve is greater than 0.5 for models with predictive ability.

The ROC curve assesses binary classifier, but the ordered logit model allows for multiple outcomes (five in this paper). Therefore, I compute five logit models, each with dichotomous dependent variable, to evaluate the baseline model in the first stage. The first model estimates a panel logit model, assessing the probability of the most negative outcome (yito=1) against all others. The second model predicts a binary indicator that equals 1 when yito=0.5 and 0 otherwise, and so on. I therefore assess whether the model is able to predict better than a coin-toss for each of the five outcomes.

VI. Empirical Results

Capital controls policy in emerging markets is systematic and responds to both mercantilist and macroprudential motivations. Moreover, as expected, mercantilist motivations predict inflow controls only when the exchange rate is appreciating – depreciation of exchange rate against trade competitors does not increase the likelihood of easing of inflow controls. On the other hand, inflow controls are fully countercyclical to domestic bank credit to private non-financial sector—they are tightened during credit booms and eased during busts. However, inflow controls could be better targeted to sources of systemic risk from capital flows – these controls do not systematically respond to various measures of foreign credit or its growth. However, there is one sense in which countries do change their behavior in response to foreign credit booms—countries with very high foreign currency debt respond less countercyclically to mercantilist concerns. Inflow controls also respond more to mercantilist motivations in countries that have inflation targeting monetary policy frameworks but do not have freely floating exchange rates.

EMEs use both inflow tightening and outflow easing to respond to mercantilist concerns. However, they use only inflow tightening to respond to macroprudential concerns—net NKI restricting measures do not respond to macroprudential concerns. The response of net NKI restricting measures is stronger in countries with relatively high exchange rate pass-through to export prices, i.e., those whose exports stand to suffer more because of currency appreciation.

A. Baseline Results: Mercantilist and Macroprudential Motivations in use of Inflow Tightening Policies

Inflow controls in emerging markets have systematically responded to both mercantilist and macroprudential motivations. Table 2 presents the results of the baseline model explaining (weighted, non-FDI) net inflow tightening actions. The reported coefficients are proportional odds ratios. A one-standard-deviation increase in the country-specific mercantilism proxy, other things being equal, increases the odds of taking a strong net inflow tightening measure by 21 percent, compared with the alternatives (taking a small net inflow tightening measure, doing nothing or net easing of inflow controls). The results for other mercantilism proxies are similar—a one-standard-deviation nominal appreciation against trade competitors over the previous quarter increases the odds of taking a net inflow tightening measure by 18%, compared with the alternatives. The estimated coefficients for mercantilism proxies are significant at 5% level of significance.

Table 2.

Baseline: Inflow Controls Respond to both Mercantilist and Macroprudential Concerns

article image
Source: Author’s calculations.Notes: Reported values are proportional odds ratios. Sample period is 2001w1–2015w52. All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. Monetary policy stance and fiscal policy stance are variables that take the value +1 if monetary policy is tightened in the previous week (or structural balance improves), -1 for expansionary policies and 0 otherwise. ∆Reserves/GDP are residuals from regressions in Table 1. Robust standard errors used. *** p<0.01, ** p<0.05, * p<0.10

To further test the identification of mercantilist motivation, in column (6) of Table 2, I use the country-specific mercantilism proxy orthogonal to the appreciation against U.S. dollar. To do this, I use a fixed effects panel regression on the country-specific mercantilism proxy with the 13-week appreciation against U.S. dollar and constant as explanatory variables.28 I use the residuals from this regression as the mercantilism proxy in Table 2 column (6). This is a challenging specification for mercantilist proxy, as USD appreciation may partly capture mercantilist concerns as well. The results are nevertheless largely unchanged -the mercantilism proxy (now orthogonal to USD appreciation) is significant and has nearly the same magnitude as the baseline results.

On the macroprudential side, a one-standard-deviation increase in bank credit to GDP gap increases the odds of a net inflow tightening by about 24% relative to the odds of the alternatives, other things being equal. The estimated coefficients for the bank-credit-to-GDP gap are also significant at 1% levels in all specifications.

Like monetary policy, capital controls policy changes also come in cycles—a net inflow tightening increases the odds that the next action will be a net tightening as well—the odds ratio increases by about 30%. Net tightening of capital controls also comes with improvements in general government structural balances. Monetary policy tightening reduces the odds of net inflow tightening, once previous policy actions and reserves accumulation are controlled for. VIX is not significantly associated with the probability of net inflow tightening measures, but inflow tightening (easing) measures have lower (higher) odds of being used during crisis periods.

A more intuitive way to interpret the coefficients is to compute the average marginal effect— the average change in probability of each outcome for a change in each explanatory variable. These are shown in Figure 4 below, for mercantilist and macroprudential proxies. The responses are as expected. Greater appreciation against trade competitors significantly increases the probability of a net tightening of inflow controls. This appreciation also reduces the probability of net easing of inflow controls, although this impact is not significantly different from zero. An increase in bank credit to GDP gap, on the other hand, significantly increases the probability of net tightening of inflow controls, and significantly reduces the probability of net easing of inflow controls. While the actual size of the average marginal effects may appear small, the unconditional probability of a net tightening action of 0.5 in a given week is 0.84% and that of a net tightening action equal to 1 is only 0.71% among the active countries in sample. Further, the average marginal effects are averaged over all observations—as we will see below, the marginal effects are around 10 times larger for inflation targeting (IT) and non-freely floating (non-FF) regimes.

Figure 4:
Figure 4:

Average Marginal Effects: Baseline Model for Net Inflow Tightening

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations.Note: The figure computes the average marginal effects for the baseline model with the country-specific mercantilism proxy.

Our interest is not only in the predictive power of individual coefficients, but in the ability of the model to predict policy, or the goodness of fit of the model. I formally assess the goodness of fit using AUROC, but it is also instructive to look at the actual versus predicted values from the model. Figure 5 plots the actual policy actions versus the predicted values of the latent variable from the baseline model, for four major economies: India, China, Brazil, and Turkey. The figure shows that the latent variables co-move remarkably well with actual inflow policy actions, unlike predictions from the VIX-only model (which are the same for all countries).

Figure 5:
Figure 5:

Predicted Latent Variable has a high Degree of Co-Movement with Actual Net Inflow Tightening Actions

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s Calculations.

The AUROC for the baseline model varies between 0.66 and 0.74 for predicting policy actions, with standard errors of about 0.03 (Table 3). These AUROCs are similar to those achieved in the recent models for crisis prediction, e.g., the baseline models in Schularick and Taylor (2012). This suggests that the baseline model does reasonably well as a predictor of capital controls policy.

Table 3.

Comparing Models Predicting Inflow Controls—AUROC

article image
Source: Author’s calculations.Notes: Each model is panel logit, with the dependent variable redefined to be a dichotomous variable. For example, in the first block of models, the dependent variable takes value 1 when the ordered (weighted, non-FDI) net inflow tightening variable =-1, and 0 otherwise. The final model has fewer observations because for at least one country in the sample, the model with the crisis dummy perfectly predicts action. These observations are dropped.

The baseline model also predicts better than simpler models (Table 3). The simpler models considered are: a VIX-only model with only VIX and a crisis dummy as explanatory variables, FX-only and MP-only models, which are the baseline models, but without the macroprudential, and mercantilist proxy respectively. Both MP-only and FX-only models are better than a coin toss and better than a VIX-only model, suggesting that each of the domestic factors plays a role in policy decisions. The baseline model improves over an MP-only or FX-only model in terms of AUROC, though the extent of improvement depends on the outcome being predicted, and for MP-only models is not significant. The FX-only models have an AUROC of between 0.6 and 0.74, with the highest AUROC for predicting strong tightening of inflow controls or strong easing of controls. For the strongest tightening, the FX-only model is indistinguishable from the baseline model, suggesting that mercantilist motivations play a role when policymakers decide to act decisively to tighten inflow controls. The difference in AUROCs between the MP-only and FX-only models may also be driven by the fact that the macroprudential motivations significantly change the probability of both net tightening and net easing of inflow controls, while mercantilist motivations act only on the tightening side, as shown in Figure 4 above. As we see later, the mercantilist motivation is more important in explaining net NKI restricting measures, which captures all available capital control tools to stem appreciation pressure.

As the capital controls index is based on qualitative information, one may ask how the interpretation of results is affected if the intensity of the changes is not perfectly captured. The dataset on capital controls captures the intensity of changes in two ways: (1) the capital controls data identifies the changes at a granular level—policy announcements are not the same as policy actions. A policy action is identified by splitting announcements along six dimensions, meaning that if policymakers were making bigger, “more intense” announcements in certain periods, e.g., during crisis periods, this should result in more counted actions in these periods. This is in fact the case with the index, as seen in Figure 8 below. Second, the index weights the actions by the share of the IIP category that the action affects, thus giving more weight to actions that affect a larger share of the country’s balance sheet. Nevertheless, to the extent that the data don’t capture intensity perfectly, we may underestimate the size of the responses (if policymakers systematically tightened more intensely than they eased, and we don’t have that information). Therefore, we should interpret the results as capturing the minimum policy reaction. In this context, the finding that capital controls policy did react to mercantilist and macroprudential motivations gains even more significance, as the true coefficients may be even larger.

Figure 6:
Figure 6:

Policy Reaction Function Changes in High Foreign Currency Debt States

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations.Note: The figure computes the average marginal effects for the model with the country-specific mercantilism proxy.
Figure 7:
Figure 7:

Average Marginal Effects: Baseline Model for Net NKI Restrictions

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations.Note: The figure computes the average marginal effects for the baseline model with the country-specific mercantilism proxy.
Figure 8:
Figure 8:

Net NKI Restricting Measures Respond Strongly to Appreciation Pressures Against U.S. Dollar

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: IMF International Financial Statistics, Datastream and Author’s calculations. Note: Exchange market pressure index is the EME average. Each emerging market’s EMP is computed as the sum of standardized appreciation in nominal exchange rate against U.S. dollar and standardized percentage increase in foreign exchange reserves excluding gold. The reserves series is interpolated from quarterly data before computing percentage changes. Net NKI Restricting actions are computed as (Inflow Tightenings – Inflow Easings) + (Outflow Easings -Outflow Tightenings). The measures are weighted and exclude those related to FDI but include currency-based measures. Last Observation: 2015w52.

To summarize, the results so far suggest that both mercantilist and macroprudential motivations are important in predicting the use of inflow tightening measures. Moreover, the strongest inflow tightening actions respond more to mercantilist concerns.

B. Exploring the Macroprudential Motivations: Do Capital Controls Target Foreign Credit?

So far, the analysis has focused on a relatively simple model, with domestic bank credit to GDP gap as the only proxy for macroprudential motivations. As discussed in section II.A, recent literature specific to capital controls has recommended that capital controls be targeted to foreign borrowing, specifically foreign currency borrowing. Therefore, I tested a number of additional proxies for macroprudential motivations, sequentially adding them to the baseline model. The additional variables do not have significant average marginal effects on the predicted probabilities of net inflow tightening actions, with the exception of equity share of mutual fund inflows (Table 4).29 An increase in equity share of fund flows reduces the probabilities of taking a net inflow tightening actions and increases the probabilities of net easing actions, which is consistent with equity inflow being safer than debt inflows.

Table 4.

Most Additional Macroprudential Proxies do not have Significant Average Marginal Effects

article image
Source: Authors’ calculations.Note: Dependent variable is the ordered weighted, non-FDI net inflow tightening measures. Estimation method is panel ordered logit, assuming random effects and using robust standard errors.

An interesting result in Table 4 is that capital controls are acyclical on average to foreign or foreign currency debt while they are countercyclical to the domestic bank credit gap.30 This means that regulators prevent domestic residents from borrowing abroad when domestic banks are lending at a brisk pace, and ease restrictions on foreign borrowing when domestic banks credit is growing slowly—however they do not systematically respond to changes in foreign credit itself. The tightening of controls on foreign credit when domestic credit is booming may simply reflect that regulators find it easier to restrict foreign credit than domestic credit, because of lack of adequate domestic prudential tools or because of shortcomings of domestic institutional frameworks. For example, if regulators can do little to stem excessive lending to politically preferred sectors in economies where state banks dominate domestic lending, they may prefer to change restrictions on foreign credit to manage total credit in the economy.

While capital controls may not be countercyclical to foreign credit on average, it may be that regulators focus on external credit more when it is already high. To test this, I conduct a counterfactual experiment. I run the following specification, which uses the foreign currency debt securities stock (deviation from trend) as the macroprudential proxy and interacts both the mercantilist and macroprudential proxies by a dummy variable that takes the value 1 when the measure of foreign credit is more than two standard deviations above its country-specific mean, and zero otherwise:

(6)Pr(yit=sj|wit1)=f{D(Xit1MP2)*Xit1MPβDMP+Xit1MPβMP+D(Xit1MP2)*Xit1FXβDFX+Xit1FXβFX+XtGβG+Xit1oβo}

High stock of foreign currency debt securities relative to trend changes the reaction function. Figure 6 plots the average marginal effects of the high foreign currency debt state on the probability of a positive net inflow tightening action, for different values of mercantilism and macroprudential proxies. If all countries behaved as if they were in high foreign currency debt state, they would be less likely to tighten inflow controls to stem appreciation against trade competitors, and also less likely to tighten inflow controls when foreign currency debt gap is low. That is, the reaction function becomes less countercyclical in response to appreciation against trade competitors and more countercyclical to foreign currency debt gap. However, even in this extreme case, higher foreign currency debt does not significantly increase the probability of a tightening -the increase in countercyclicality comes only from the left end of the tail.

C. Exploring the Mercantilist Motivation

Predicting net NKI Restricting Actions

The analysis so far has examined the motivations for changing controls on capital inflows. Yet, countries have another tool to resist exchange rate appreciations: the easing of outflow restrictions (Aizenman and Pasricha, 2013). In this section, I analyze the motivations for changing Net NKI restricting actions, defined as the sum of net inflow tightening actions and net outflow easing actions.

The results show that net NKI restricting actions respond systematically only to mercantilist concerns (Table 5). The size and significance of the estimated proportional odds ratios for mercantilism proxies in Table 5 is higher than those in Table 2 for net inflow tightening actions. Increases in the credit-to-GDP gap do not significantly increase the odds of net NKI restricting actions. Further, appreciation against trade competitors on average significantly increases the predicted probability of positive net NKI restricting measures, while also significantly decreasing the probability of easing net restrictions on NKI (Figure 7). On the other hand, the average marginal effect of bank credit to GDP gap on the probability of net NKI restricting actions is not significant for any outcome. Figure 8 supports these results by showing a close correspondence between net NKI restricting actions and exchange market pressure.

Table 5.

Net NKI Restricting Actions Respond only to Mercantilist Concerns

article image
Source: Author’s calculations.Notes: Reported values are proportional odds ratios. Sample period is 2001w1–2015w52. All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. Monetary policy stance and fiscal policy stance are variables that take the value +1 if monetary policy is tightened in the previous week (or structural balance improves), -1 for expansionary policies and 0 otherwise. ∆Reserves/GDP are residuals from regressions in Table 1. Robust standard errors used. *** p<0.01, ** p<0.05, * p<0.10

The results of Table 5 and Table 2 together imply that countries use both inflow tightening and outflow easing actions to respond systematically to mercantilist concerns, but use only inflow tightening actions to respond to macroprudential concerns. This is further evidence that policy is carefully calibrated: outflow easings do not directly reduce systemic risk but can mitigate exchange rate pressures.

Additional Proxies for Mercantilist Motivation

So far, the analysis has focused on the new proxies for mercantilist motivations proposed in this paper. Here, I explore additional proxies for mercantilist motivations, suggested by the empirical and theoretical literature. These include growth rate of a country’s GDP relative to the U.S. and its relative manufacturing IIP growth rate (Costinot, Lorenzoni and Werning, 2014), as well as export volume growth, which is used in the existing empirical literature on reserves accumulation. The additional variables are not significant (Table 6).

Table 6.

Additional Mercantilism Proxies are not Significant

article image
Source: Author’s calculations.Notes: Reported values are proportional odds ratios. Sample period is 2001w1–2015w52. All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. Monetary policy stance and fiscal policy stance are variables that take the value +1 if monetary policy is tightened in the previous week (or structural balance improves), -1 for expansionary policies and 0 otherwise. ∆Reserves/GDP are residuals from regressions in Table 1. Robust standard errors used. *** p<0.01, ** p<0.05, * p<0.10

D. What Drives the Weights on the two Motivations: The Role of Exchange Rate Pass-Through and Monetary and Exchange Rate Regimes

If, from an international policy coordination perspective, it is important that capital controls should respond only to macroprudential concerns, then we need to understand if there are any structural factors that drive the relative weight of the two motivations in the policy reaction function. In this section, I explore two such factors: the degree of sensitivity of a country’s export prices to exchange rate changes, and the role of inflation targeting and non-freely floating regimes.

High exchange rate pass-through (ERPT) to export prices means that the exporter’s trading partners bear more of the cost of the exporting country’s currency appreciation. This means that the country’s exports are potentially more sensitive to that appreciation, and policymakers may in turn respond more to stem such appreciation. To test this, I use a dummy variable, which equals 1 for countries with greater than median export price ERPT and add it as an interaction term in the baseline specification for net NKI restricting measures. I use Bussière, Gaulier and Steingress’s (2016) baseline (no fixed effects) estimates of export price elasticities to construct the dummy variable. That is, I run the following specification:

(7)Pr(yit=sj|wit1)=f{Xit1MPβMP+D(HighERPT)*Xit1MPβDMP+Xit1FXβFX+D(HighERPT)*Xit1FXβDFX+XtGβG+Xit1oβo},

where the dependent variable is net NKI restricting measures and the mercantilist proxy is the country-specific proxy.

The results show that countries with high ERPT to export prices respond more strongly to competitiveness changes against trade competitors than low ERPT countries (Figure 9).31

Figure 9:
Figure 9:

Countries with high Exchange Rate Pass-Through to Export Prices Respond more to Mercantilist Concerns

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations.

I conduct a similar exercise with a dummy variable that equals 1 for countries that have an inflation target and a non-freely floating regime, as identified in the IMF AREAER. I find that these regimes are also more responsive to mercantilist motivations (Figure 10). They are also less responsive to macroprudential motivations.32

Figure 10:
Figure 10:

Inflation Targeting and Non-Freely Floating Exchange Rate Regimes Respond more to Mercantilist Concerns

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations.

VII. Robustness Checks

I conduct two types of robustness checks: first, on the mercantilist proxy as a valid measure of mercantilist motivations; and second, robustness checks on the baseline specifications.

A. Mercantilist Proxy as a Valid Measure of Mercantilist Motivations

To validate the mercantilism proxy, I test whether it predicts non-tariff barriers to trade imposed by countries. I use data on non-tariff measures (NTMs) applied on trade from the World Trade Organization’s (WTO) I-TIP portal. To download the data, I used web-scraping techniques in Python, as the I-TIP portal doesn’t allow changing both the time and country dimension at the same time. The data I use includes four types of NTMs: anti-dumping duties, quantitative restrictions, countervailing duties and safeguards. I download three weekly series for each country: the number of measures coming into force, initiated and withdrawn during the week. I also compute net initiations as the number of measures initiated, less those withdrawn during a given week.

It is important to note that the trade barriers data does not capture all the activism on trade policy, for several reasons: one, the data doesn’t cover the full universe of trade barriers countries can take, due to concerns about the timeliness and completeness of data on certain measures;33 and second, that countries could respond by raising tariffs as well. It is important to also note that there may be long time lags between competitiveness pressures arising and countries initiating trade measures or enforcing them. Even with these caveats in mind, a strong correlation or causality between mercantilism proxies and future NTMs would suggest that the proxies are indeed capturing mercantilist concerns.

The unconditional correlations between the mercantilism proxies and the future initiations, coming into force or net initiations of non-tariff barriers are positive and significant (Table 7). These contrast with their correlations with appreciation against the U.S. dollar, which are low or negative. The positive correlations between mercantilism proxies and trade barriers suggest that the proxies do capture concerns about trade competitiveness. The correlations for several individual countries are much larger than for the group, reaching about 0.5 for Argentina and Poland (Figure 11).

Table 7.

Mercantilism Proxies are Positively Correlated with Future Non-tariff Measures

article image
Source: Author’s calculations.
Figure 11:
Figure 11:

Mercantilism Proxies are Positively Correlated with Future Net Initiations of Non-tariff Barriers in all but Three Countries

Citation: IMF Working Papers 2020, 080; 10.5089/9781513546100.001.A001

Source: Author’s calculations.Note: The figure plots the correlations of past appreciation of currency against trade competitors with future initiations of non-tariff barriers. 1-y Ma refers to a 52-week backward looking moving average. Future net initiation of non-tariff barriers refer to a forward looking 52-week moving average of net initiations of non-tariff barriers.

To explore these relationships further, I test whether the mercantilism proxies Granger cause any of the measures of non-tariff barriers (Table 8). For each country and series pair, I first select the optimal lag length from VARs with up to 52 lags (i.e., up to 52 weeks, or one year), using AIC statistic. Next, I conduct the Granger causality tests for net initiations and measures coming into force, using the optimal lag length.34 I find that for 16 out of the 21 countries in sample, and in 11 out of the 13 active countries, there is evidence of Granger causality from at least one of the mercantilism proxies to one of the non-tariff barriers series. The year over year nominal appreciation against trade competitors Granger causes measures entering into force over the following year in 10 countries. In the other 6 countries, the causality is to net measures initiated.

Table 8.

Granger Causality Tests: x2 Statistics

article image
Source: Author’s calculations.Note: Degrees of freedom not shown, as they vary by country and series pair. Full results available on request. The number of observations per country vary from 623 to 773, and the number of lags vary between 2 and 52. The optimal lag length for each country was selected using AIC. *** p<0.01, ** p<0.05, * p<0.10. Non-tariff barriers are 52-week forward looking moving averages.

B. Robustness Checks on the Baseline Specifications

The results presented above are robust to a number of alterations in specifications. First, I use alternative measures of capital controls policy (Table 9, columns 1–3). I run the baseline specifications without reducing the number of ordered categories. This leads to estimation of a large number of cut-offs for the latent variable but doesn’t affect the sign or significance (or the approximate size) of the estimated coefficients. Using unweighted policy actions as the dependent variable does not change the results. Estimating the reaction functions for all policy actions, including those affecting FDI, leads to a small decline in the estimated coefficient on bank credit gap in the baseline regression explaining net inflow tightening actions, but the coefficient is still significant. The other results are robust to including FDI-related changes.

Second, I test robustness to sample. I include all countries in sample, not only the active ones. This reduces the estimated size of the coefficient on mercantilist motivations, it is still significant (Table 9, column 4). I also exclude Korea from the sample, as its top four trade competitors are countries with reserve currencies (Table 9, column 5). Excluding Korea increases the responsiveness to mercantilism proxy.

Table 9.

Robustness Checks: Alternative Measures of Inflow Controls and Broader Sample of Countries

article image
Source: Author’s calculations.Notes: Reported values are proportional odds ratios. Sample period is 2001w1–2015w52. All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. ∆Reserves/GDP are residuals from regressions in Table 1. Robust standard errors used. *** p<0.01, ** p<0.05, * p<0.10

Third, I assess motivations for net outflow easings (rather than net NKI restricting measures). The mercantilism proxy is significant in these specifications as well (Table 10).

Table 10.

Robustness Checks: Motivations Behind use of Net Outflow Easings

article image
Source: Author’s calculations.Notes: Reported values are proportional odds ratios. Sample period is 2001w1–2015w52. All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. ∆Reserves/GDP are residuals from regressions in Table 1. Monetary policy stance and fiscal policy stance are variables that take the value +1 if monetary policy is tightened in the previous week (or structural balance improves), -1 for expansionary policies and 0 otherwise. Robust standard errors used. *** p<0.01, ** p<0.05, * p<0.10

Fourth, I control for structural country-specific factors, including corruption and governance indicators, and capital account openness indicators (Table 11). The baseline results are robust to adding these variables. More open countries and countries with better regulatory quality are more likely to calibrate capital control policies. Accounting for openness, the response to mercantilist motivations is stronger.

Table 11.

Robustness Checks: Adding Structural Factors

article image
Source: Author’s calculations.Notes: Reported values are proportional odds ratios. Sample period is 2001w1–2015w52. All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. Monetary policy stance and fiscal policy stance are variables that take the value +1 if monetary policy is tightened in the previous week (or structural balance improves), -1 for expansionary policies and 0 otherwise. ∆Reserves/GDP are residuals from regressions in Table 1. Robust standard errors used. *** p<0.01, ** p<0.05, * p<0.10

Fifth, I run several alternative specifications to explore the role of the domestic policies and of the nominal exchange rate (Table 12). I control for other domestic macroprudential policies, creating a variable that is the total number of domestic macroprudential measures taken (summing up the components from Cerutti et al. (2016), excluding the foreign currency reserves requirement measures, as the latter are already included in the capital controls data). This variable is not significant. I also use actual changes in structural fiscal balance and in the policy rate, instead of variables capturing the stance of these policies. The results for mercantilist and macroprudential proxies are robust to this change. Using nominal exchange rate in the same specification with mercantilism proxy reduces the size of the coefficient on each, but the coefficients on these variables are still significant. I also add an interaction term between mercantilist and macroprudential proxies to test whether the probability of tightening inflow controls increases more when the exchange rate appreciates against trade competitors, if bank credit to GDP gap in the country is high as well. The resulting coefficient on the interaction term is not significant, while the coefficients on the levels of the variables are robust to the addition of the interaction term.

Table 12.

Robustness Checks: Domestic Policies, Nominal Exchange Rate and Interactions

article image
Source: Author’s calculations.Notes: Reported values are proportional odds ratios. Sample period is 2001w1–2015w52. All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. Monetary policy stance and fiscal policy stance are variables that take the value +1 if monetary policy is tightened in the previous week (or structural balance improves), -1 for expansionary policies and 0 otherwise. ∆Reserves/GDP are residuals from regressions in Table 1. Robust standard errors used. *** p<0.01, ** p<0.05, * p<0.10

Sixth, I use alternative measures of global liquidity instead of VIX, including global bank claims (as percentage of global GDP, from BIS), oil prices and U.S. federal funds shadow rate (Table 13). These modifications do not change the baseline results. In this table, I also use an alternative measure of bank credit to GDP gap, which uses Hamilton (2018) methodology, to overcome to issues with HP-filter.35 The results are robust to this specification as well.

Table 13.

Robustness Checks: Other Global Variables and Bank Credit Gap

article image
Source: Author’s calculations.Notes: Reported values are proportional odds ratios. Sample period is 2001w1–2015w52. All domestic control variables are one-week lagged. All continuous domestic variables are standardized but centered at 0, i.e., the variables are divided by their standard deviation but not demeaned. Monetary policy stance and fiscal policy stance are variables that take the value +1 if monetary policy is tightened in the previous week (or structural balance improves), -1 for expansionary policies and 0 otherwise. ∆Reserves/GDP are residuals from regressions in Table 1. Robust standard errors used. *** p<0.01, ** p<0.05, * p<0.10

Finally, I report the goodness of fit test of out-of-sample forecasts (Table 14). I use the last three years of the sample (2012–2015) as the out-of-sample period. The out-of-sample forecast performance of the models is still good, ranging from 0.62–0.73 for policy actions.

Table 14.

Robustness Checks: Out-of-Sample Forecasts

article image
Source: Author’s calculations.Note: In-sample period is 2001w1–2011w52, and out-of-sample period is 2012w1–2015w52. Each model is panel logit, with dependent variable redefined to be a dichotomous variable. For example, in the first block of models, the dependent variable takes value 1 when the ordered (weighted, non-FDI) net inflow tightening variable =-1, and 0 otherwise.

VIII. Conclusions

Are capital controls macroprudential or mercantilist? The results in this paper strongly suggest that they are both. The results provide clear evidence that capital controls policy in emerging markets has been systematic, and that it has responded to both macroprudential and mercantilist motivations. The use of net inflow tightening measures can be described by a function of mercantilist and macroprudential motivations. Moreover, the choice of instruments is also systematic: policymakers respond to mercantilist concerns by using both instruments—inflow tightening and outflow easing. However, they use only inflow tightening in response to macroprudential concerns. This is the first paper to provide evidence of the existence of a macroprudential motivation in the use of capital controls policy, even before these controls were generally acknowledged after the global financial crisis, as valid tools of the macroprudential policy toolkit. Yet, the analysis in this paper underlines that the concerns of those who worry about a currency war are also justified— capital controls have also been systematically used to preserve competitive advantage in trade.

These results highlight an assignment problem of using one tool (inflow controls) to meet multiple objectives (Tinbergen, 1962). They suggest a need for further debate on whether it would be globally optimal if countries used capital control actions solely as a tool of macroprudential policies, and if so, how to ensure that this is the case.

The results also suggest that capital controls have not been targeted specifically to foreign-to-foreign currency debt—inflow controls are not countercyclical to the specific macroprudential concerns related to external or foreign currency borrowing. Rather, policy appears acyclical to these variables, but is countercyclical to domestic bank credit to the private non-financial sector. The tightening of controls on foreign credit when domestic credit is booming may simply reflect that regulators find it easier to target foreign credit than domestic credit, either because of lack of adequate domestic prudential tools or because of shortcomings of domestic institutional frameworks. As capital controls become more widely used as tools of macroprudential policies, future research and policy discussions could focus on how best to ensure that these instruments are targeted directly to the vulnerabilities they seek to address.

IX. References

  • Acharya, S. and Bengui, J. (2018) “Liquidity traps, capital flowsJournal of International Economics, Volume 114, September 2018, Pages 276298

    • Search Google Scholar
    • Export Citation
  • Adrian, T. (2018) “Policy Responses to Capital FlowsRemarks at the LIII Meeting of Governors of Latin America, Spain and the Philippines at the IMF-World Bank Annual Meetings in Bali, Indonesia, 11 October 2018.

    • Search Google Scholar
    • Export Citation
  • Agénor, Pierre-Richard and Pengfei Jia (2015) “Capital Controls and Welfare with Cross-Border Bank Capital Flows,” Centre for Growth and Business Cycle Research Discussion Paper Series 212, Economics, The University of Manchester.

    • Search Google Scholar
    • Export Citation
  • Aizenman, J and J Lee (2007) “International reserves: precautionary versus mercantilist views, theory and evidence”, Open Economies Review, Vol. 18, No. 2, pp 191214.

    • Search Google Scholar
    • Export Citation
  • Aizenman, J and G Pasricha (2013) “Why do emerging markets liberalize capital outflow controls? Fiscal versus net capital flow concerns”, Journal of International Money and Finance, Vol. 39, pp 2864.

    • Search Google Scholar
    • Export Citation
  • Alfaro, L, A Chari and F Kanczuk (2017) “The real effects of capital control taxes: firm-level evidence from a policy experiment”, Journal of International Economics 108 (September 2017): 191210.

    • Search Google Scholar
    • Export Citation
  • Avdjiev, S, L Gambacorta, L Goldberg and S Schiaffi (2017) “The shifting drivers of global liquidity”, Bank for International Settlements, Working paper no. 644, June.

    • Search Google Scholar
    • Export Citation
  • Benigno, G, C Huigang, C Otrok, A Rebucci and E Young (2011) “Monetary and macro-prudential policies: an integrated analysis”, 12th Jacques Polak Annual Research Conference, International Monetary Fund, mimeo.

    • Search Google Scholar
    • Export Citation
  • Bernanke, B.S. (2015) “Federal Reserve Policy in an International Context,” Paper presented at 16th Jacques Polak Annual Research Conference, November 5–6, 2015.

    • Search Google Scholar
    • Export Citation
  • Bhalla, S.S. (2012) “Devaluing to Prosperity: Misaligned Currencies and Their Growth ConsequencesPeterson Institute for International Economics, August 2012

    • Search Google Scholar
    • Export Citation
  • Bianchi, J (2011) “Overborrowing and systemic externalities in the business cycle”, American Economic Review, Vol. 101, No. 7, pp 340026.

    • Search Google Scholar
    • Export Citation
  • Bianchi, J., and Mendoza, E.G. (2016). “Optimal Time-Consistent Macroprudential Policy,” forthcoming, Journal of Political Economy, http://www.sas.upenn.edu/~egme/wp/w19704.pdf.

    • Search Google Scholar
    • Export Citation
  • BIS-FSB-IMF (2011) “Macroprudential policy tools and frameworks”, Progress Report to G20.

  • Bussière, M, G Gaulier and W Steingress (2016) “Global trade flows: revisiting the exchange rate elasticities”, Banque de France Working Papers, no. 608, Banque de France.

    • Search Google Scholar
    • Export Citation
  • Carney, M. (2019) “Pull, Push, Pipes: Sustainable Capital Flows for a New World Order” Speech given by Mark Carney, Governor of the Bank of England to Institute of International Finance Spring Membership Meeting, Tokyo, 6 June 2019.

    • Search Google Scholar
    • Export Citation
  • Casas, C, F Diez, G Gopinath and P Gourinchas (2016) “Dominant currency paradigm”, NBER Working Paper Series, no. 22943, December.

  • Cerutti, E, R Correa, E Fiorentino and E Segalla (2016) “Changes in prudential policy instruments: a new cross-country database”, IMF Working Papers, no. 16, June.

    • Search Google Scholar
    • Export Citation
  • Cheung, Y and X Qian (2009) “Empirics of China’s outward direct investment”, HKIMR Working Papers, no. 17, May.

  • Choi, Woo Jin & Taylor,