Prevented Sudden Stops

Transcription

1 Prevented Sudden Stops Eduardo Cavallo Inter-American Development Bank Alejandro Izquierdo Inter-American Development Bank John J. Leon-Diaz University of Maryland This Version: September 15, 2016 Preliminary Version Abstract This paper explores the determinants behind the decision of domestic investor to adjust their asset position in response to a variation in gross capital inflows and avoid episodes of net sudden stops. We present evidence that while sudden stops in gross inflows are associated with global conditions, domestic factors such as the degree of domestic liability dollarization, economic growth and institutional background are important to prevent these episodes in becoming net sudden stops. 1 Introduction Sudden stops in net capital flows, also referred as net suddent stops, entail large costs in terms of output and investment when they materialize. An important fraction of these net sudden stops are triggered by capital account reversals originated in transactions attributable to nonresidents. However, due to the potential offsetting effect of changes in gross flows from residents, not all reversals from nonresidents lead to a sudden stop in net flows. This suggests that the behavior of residents can prevent economies from experiencing the pervasive effects of the discontinuity in international capital flows. We denote the episode where a sudden stop in gross flows is not accompanied by a sudden stop in net flows as a prevented sudden stop. This paper explores the conditions under which net sudden stops that are preceded by sudden stops in gross inflows do not occur; i.e. the circumstances under which a net sudden stop is prevented even though a sudden stop in gross inflows does occur. This problem can be considered 1

2 as the sequence of two transitions: first, the transition from normal times to periods of sudden stops in gross inflows. And second, after experiencing a sudden stop in gross inflows, the transition or not to a net sudden stop. We study the determinants behind each transition and conclude that while global conditions are important to explain the incidence of gross sudden stops, domestic conditions are relevant to understand why these episodes become also a net sudden stop. As a corollary we find that, in periods of global distress, the ability of a country to build resilience against capital outflows relies heavily on the soundness of their domestic conditions. By evaluating a comprehensive set of variables, we conclude that episodes of prevented sudden stops are not associated with changes in global conditions, unlike gross sudden stops, but they are positively related to domestic growth and institutional background, and negatively related to the country s level of foreign liabilities. Until the mid-1990s the role of domestic capital outflows in emerging markets was negligible, and thus making the distinction between gross and net inflows was irrelevant. As a consequence, the discussion on capital flows was focused exclusively on net flows and the disruptive effects they might infringe in emerging markets when they stop abruptly. But to the extent that domestic investors in emerging markets started playing a more sizable role, the discussion shifted towards understanding inflows and outflows separately, as well as the role of domestic and foreign investors in shaping them. The distinction between gross and net flows makes it possible to analyze sudden stop episodes from different perspectives. On the one hand, sudden stops in net flows can be the consequence of a sharp decline in gross inflows by foreigners; on the other hand, they can be a consequence of a sharp increase in gross outflows by domestic agents. In this regard, Cavallo et al. (2013) extend the scope of the term sudden stop to reflect also an abrupt change in gross capital inflows. The authors schematize a taxonomy for sudden stops based on the possibility that a sudden stop in gross inflows does not translate into a sudden stop in net flows. 1 This is possible when domestic residents adjust their asset positions to offset the decline in inflows from foreigners, and suggest that domestic agents have the ability to prevent episodes of net sudden stops. In the same way, Broner et al. (2013) find that both gross capital inflows and outflows increase 1 They present seven potential sudden stops: a sudden stop in inflows that does not imply a sudden stop in net flows (SSI); a sudden stop in inflows that translates into a sudden stop in net flows (SSIN); a sudden stop in net flows that is not a sudden stop in inflows or outflows (SSN); a sudden stop in inflows and sudden stop in outflows that is not a sudden stop in net flows (SSIO); a sudden stop in inflows and a sudden surge in outflows that is also a net sudden stop (SSION); a surge in outflows that is not a sudden stop (SSO); and, a sudden surge in outflows that is a sudden stop in net flows (SSON). 2

3 during expansions and decrease during downturns. More importantly, the authors find that in periods of crisis, total gross capital flows collapse due to the retrenchment of outflows from foreign markets, and that this phenomenon is particularly stronger if the crisis is global. This fact suggests, in the same line as IMF (2013), that during the recent global financial crisis domestic residents compensated the fall in international capital inflows with larger capital retrenchments. Our view that global conditions are relevant to understand episodes of decline in gross inflows is consistent with the findings of Forbes and Warnock (2012). They highlight that during the last financial crisis there was an unprecedented number of countries experiencing both stops and retrenchment episodes. The authors indicate that global factors, especially global risk through changes in economic uncertainty, as well as changes in risk aversion and global growth, are key drivers of sudden stops in gross inflows. Developing a deeper understanding of the determinants of net sudden stops and of the necessary conditions that must be met in order to prevent them is of utmost importance. As stated by Cavallo et al. (2013), episodes of net suddenstops areeven more pervasivethan suddenstops ingross inflows in terms of GDP losses. Therefore, finding a way to prevent them can have a positive impact on countries that are more prone to experience such episodes. Countries need to build resilience to capital inflows volatility and avoid real adjustments through the current account. And our paper sheds light on the role of domestic investors in doing so and the conditions that characterize this process. Related Literature. This paper is closely related to Adler et al. (2014). The authors quantify the dynamic impact of global financial shocks on both net and gross capital flows to emerging markets, and analyze the role played by local investors in offsetting the behavior of foreign investors. Making use of a panel vector autoregression and the analysis of the resulting impulse responses, they find that, when facing global uncertainty and shocks to long-term interest rates, local investors can neutralize the decline in inflows from foreign investors. Our paper differs in some relevant dimensions from theirs. First, we consider only periods in which sudden stops occur, i.e. times in which a given country is more vulnerable because of the reduction in external financing. This allows us to control for any bias stemming from asymmetries in the behavior of domestic agents during normal and bad times. Second, the methodology used in Adler et al. (2014) does not allow to exploit the cross-sectional variation in capital outflows. Countries display heterogenous patterns in their capital outflows dynamics, and the ability of domestic agents to neutralize a sudden stop in gross flows depends on specific characteristics of their home country, as we document in this paper. 3

4 This paper is also related to Cifuentes and Jara (2014). These authors stress the role of assets held abroad and of the exchange rate flexibility in shaping the probability that a retrenchment episode occurs when the economy is facing a sudden stop in gross inflows. Even though their approach is in line with our view on prevented sudden stops, there are important differences between the two papers that are worth mentioning. First, the set of events under study in both papers do not overlap. This is because the occurrence of a retrenchment is neither a necessary nor a sufficient condition to prevent a sudden stop in net flows. Second, we use a more comprehensive dataset, which includes a broader set of countries as well as more explanatory variables, e.g. foreign liabilities, institutional quality or contagion effects. Our paper is part of the literature on the determinants of sudden stops, which includes papers such as Calvo et al. (2004) and Calvo et al. (2008). This literature initially explored the determinants of the abrupt decline in net flows, particularly in emerging economies, as they were the most vulnerable to these type of episodes. This approach was consistent with the fact that, at the time, the magnitude of gross capital outflows were not as large as the ones observed in the last 20 years. However, when looking at the dynamics behind gross outflows, some of the episodes that were initially associated to a reduction in flows from foreign investors, instead were driven by a change in the positions of local investors. Thus, this paper is related to the strand of the literature that departs from the net flow approach, and that studies inflows and outflows separately in the effort to understand the role that local and foreign investors have in shaping each of them. This strand includes Calderon and Kubota (2013) and Forbes and Warnock (2012). Similarly, Schmidt and Zwick (2013), using data for the Euro area, conclude that domestic volatility (i.e. uncertainty about the evolution of the economy and the economic policy implemented) has played an important role in determining the dynamics of gross flows and the increase in home bias observed in the data. Thehypothesistested inourpaperis alsoinlinewiththefindingsofghoshet al. (2014). Despite the fact that the authors only focus on episodes of capital surges in emerging markets, there are important lessons to be extracted from this paper. In particular, Ghosh et al. (2014) postulate that global factors, such as US interest rates and global risk, are important elements associated with capital surges. However, the attractiveness of a country as an investment destination is largely a function of domestic factors. This does not necessarily imply that foreign investors do not react to local conditions. On the contrary, foreign investors consider local conditions as much as domestic investors do, but they are more sensitive to changes in global conditions. Using micro-level data from investment funds, Fratzscher (2011) finds additional evidence on the role of global factors driving 4

5 flows during the global financial crisis of and its aftermath. According to this author the rise in risk was the culprit of the reallocation of flows from many emerging to some advanced economies during the crisis. This is in contrast with the pre- and post-crisis periods, in which they had the opposite effect. Domestic factors are instead related to the observed cross-country heterogeneity. The existence of home bias has been also part of the recent literature on capital flows. In particular, Milesi-Ferretti and Tille (2011) have pointed out a generalized but heterogenous collapse in international capital flows during the financial crisis. In this line, Giannetti and Laeven (2012) show that, during periods of crisis that involve higher uncertainty, investors become more risk averse and revert to domestic investments that can be evaluated at lower costs due to lower asymmetric information. Alternatively, Jochem and Volz (2011) argue that the home bias in the Euro zone is associated to changes in the portfolio structure in favor of domestic assets mainly by financial institutions in an effort to deleverage due to the inherent risk in their balance sheets. This paper is structured as follows. Section 2 presents definitions and determinants of sudden stops. Section 3 provides a brief description of the methodology, and presents the baseline results as well as the sensitivity and robustness checks. Section 4 concludes. 2 Definitions, Measurement and Data 2.1 Sudden Stops in Capital Flows A gross (net) sudden stop in capital flows is defined as an event in which the year-on-year change in foreign capital inflows (net flows) falls below two standard deviations from its historical mean. In terms of measuring its length in time, the sudden stop episode starts from the moment in which the series falls one standard deviation below its historical mean, but conditional on the fact that it will eventually cross the two-standard-deviations threshold. The episode ends when the series goes back to one standard deviation below the historical mean. Using quarterly data on gross capital flows obtained from the Balance of Payment Statistics (BOPS) developed and reported by the International Monetary Fund (IMF), we denote net flows of country j in period t as N jt = I jt +O jt, where I jt and O jt represents gross inflows and outflows respectively. Inflows are defined as the negative of total asset transactions in the Financial Account of the BOP. Outflows correspond to the total liability transactions in the Financial Account of the BOP. It should be noted that in 2009 there was a change in the methodology to compute BOP 5

6 statistics, from BPM5 to BPM6. This methodological change had a particular impact in the way Foreign Direct Investment (FDI) is computed. 2 The BPM5 series are only available up to 2008 while the new BPM6 series start in As a consequence, we had to combine both datasets to get long and consistent series of capital flows. For the period we just use the aggregate series of total asset and total liability transactions as reported in BPM5. For the period we construct the aggregate series of assets (liabilities) by adding the asset (liability) transactions associated to direct investment, portfolio investment, financial derivatives and other investment (see Appendix A for further details). We apply a moving average filter to reduce the effects of seasonality in net and gross flows. In particular, for quarterly series we define C n jt = 3 t=0 N jt and C i jt = 3 t=0 I jt for t = 4,5,...,T. The year-on-year change in net financial flows is defined as C x jt = C jt C j,t 4 with x = {n,i}. Therefore, a sudden stop in net and gross flows can be defined as an episode in which the variable C x jt falls below two standard deviations from of its historical mean. A more detailed description of the series used to compute sudden stops is presented in Table 7 in Appendix D. 2.2 Definition and Evidence on Prevented Sudden Stops Based on the above concepts, this paper defines a prevented sudden stop in economy j during period t as the event in which a gross sudden stop does not translate into a net sudden stop due to the offsetting variation in capital outflows from domestic agents. Under this definition a prevented episode is conceived as one or more consecutive periods (quarters) in which a sudden stop in net flows does not coexist with a sudden stop in gross inflows. This implies that, within the period in which a gross sudden stop episode occurs, there can be more than one prevented sudden stop episode. Table 1 summarizes gross and net sudden stops in terms of the number of episodes, their average duration and the total number of quarters in which countries have experienced these events. An analysis of the data indicates that not all sudden stops in gross inflows become sudden stops in net flows, suggesting that capital outflows have an important role in preventing the occurrence of net sudden stops. By comparing columns (3) and (8) we can observe that around 46 percent of 2 In BPM5 FDI was presented on a directional basis, i.e. direct investment in the reporting economy included assets and liabilities between a resident direct investment enterprise and its nonresident direct investor, while direct investment abroad included assets and liabilities between a resident direct investor and its nonresident direct investment enterprises. In BPM6 FDI is presented on a gross assets and liabilities basis, which is more in line with the way the other components of the Financial Account are reported. 6

7 total potential net sudden stops periods are prevented. This proportion is even bigger if we only consider advanced economies, in which case around 63 percent of total gross sudden stop periods are not sudden stops. However, the fraction of prevented periods diminishes if we focus in emerging and frontier economies. 3 In these groups of countries the fraction of prevented periods decreases considerably to 32 and 17 percent, respectively. It is important to emphasize that prevented sudden stops are not implicitly related with a pronounced decrease in gross capital outflows(or retrenchments) as defined in(forbes and Warnock, 2012). In other words, extreme events of capital outflows are neither necessary nor sufficient to avoid sudden stops in net flows. On the one hand, they are not sufficient: columns (6) and (7) in Table 1 show that 17 percent of total net sudden stop periods were accompanied by retrenchments; this percentage is around 22 percent for emerging economies. This suggests that even a sharp variation in capital outflows would not suffice to prevent a fall in net flows. On the other hand, they are not necessary: columns (8) and (9) in Table 1 shows that 22 percent of the total periods of prevented sudden stops were not accompanied by periods of retrenchment in capital outflows. In the case of emerging economies this fraction is around 34 percent, which rises to 38 percent and 53 percent for emerging economies in Latin America and Eastern Europe, respectively. 4 To get a better understanding of the incidence of capital outflows adjustments in preventing net sudden stops, we consider the particular dynamics of flows for a set of countries. Figures 1 and 2 display the dynamics of the smoothed series of capital inflows and outflows changes for advanced and non-advanced economies, respectively. The dashed line in the graphs corresponds to the threshold that defines a sudden stops in gross inflows. More specifically, when the solid black line falls below the dashed line we define this as a gross sudden stop. The shaded areas correspond to prevented sudden stop episodes. For the case of Germany and Canada, figure 1 shows that changes in capital flows exhibit a diamond pattern. This implies that periods of large declines in capital inflows coincide also with a decline in capital outflows of approximately the same magnitude. Notice that with the exception of the episode in 2013, the behavior of German investors was able to compensate the fall in foreign inflows and avoid net sudden stops. This pattern is also present in Canada where the only episode of a gross sudden stop did not translate into a net sudden stop. 3 For a detailed description on country classification, see Appendix B. 4 In fact as described in section (4.2.4) episodes of retrenchment and prevented sudden stops are not necessarily driven by the same domestic conditions either. 7

8 Table 1: Sudden Stop Episodes SS Inflows SS Net Prevented SS Average # Quarters Average # Quarters # Quarters # Quarters # Quarters # Episodes # Episodes Duration (Total) Duration (Total) (with Retrench) (Total) (No Retrench) (1) (2) (3) (4) (5) (6) (7) (8) (9) All countries Advanced Frontier Latin America East Asia and Pacific Eastern Europe Emerging Latin America East Asia and Pacific Eastern Europe

9 Figure 1: Inflows and Outflows during episodes of Gross Sudden Stops (Advanced Economies) Germany Y-o-Y Change in Filtered Series (Thousands) 1, Inflows Outflows Inflows - 2 Std. Dev. -1, Y-o-Y Change in Filtered Series (Thousands) Inflows Outflows Inflows - 2 Std. Dev. Canada Source Author s own calculations based on data from IMF-IFS. Shaded areas indicate episodes which are catalogued as sudden stops in capital inflows but not net sudden stops. 9

10 Figure 2: Inflows and Outflows during episodes of Gross Sudden Stops (Non Advanced Economies) Turkey Y-o-Y Change in Filtered Series (Thousands) Inflows Outflows Inflows - 2 Std. Dev Thailand Y-o-Y Change in Filtered Series (Thousands) Inflows Outflows Inflows - 2 Std. Dev Source Author s own calculations based on data from IMF-IFS. Shaded areas indicate episodes which are catalogued as sudden stops in capital inflows but not net sudden stops. 10

11 The behavior of capital flows in non-advanced economies is less consistent. To exemplify the behavior of capital flows in non-advanced economies we consider the cases of Turkey and Thailand. The series for these countries are shown in Figure 2. The case of Turkey presents a radical scenario as all the sudden stops in gross inflows were also sudden stops in net flows. This is evident as the variation in capital inflows displays a relatively higher volatility, when compared to the volatility in capital outflows. In this particular case, the behavior of domestic agents appears to be not enough to compensate the variation in inflows. When we consider the case of Thailand, we observe a slightly different result. In general, this country does not display the diamond pattern, particularly in the episodes after Very few episodes in the last 20 years were prevented. In some cases it was because because the variations in outflows were not enough to compensate the fall in inflows. However, it calls the attention the episodes in 2011, when outflows remained relatively constant, despite the massive collapse in gross inflows. Behind this phenomenon might be the fact that Thailand was going through a rough time in 2011, not only because of the long-lasting and contagious effects of the financial crisis, but also because Thailand face particular unfavorable domestic conditions 5 that could have jeopardized domestic investor s incentives. The graphic evidence presented in this section suggests that even in the presence of periods of global distress, the final allocation of resources is not merely driven by the home bias motive, but also by additional factors related to the domestic stability of the economy. These factors are defined, and their effects are quantitatively analyzed, in the next sections. 2.3 Determinants of Sudden Stops To definetheset of determinants of suddenstops (gross and prevented) to beused for our econometric exercise, we have taken into account all the explanatory variables that have been considered in the empirical literature analyzing net sudden stops(calvo et al., 2008), gross sudden stops(calderon and Kubota, 2013; Alberola et al., 2012), currency crisis(frankel and Rose, 1996; Milesi-Ferretti and Razin, 1998), current account reversals (Edwards, 2007) and retrenchments (Forbes and Warnock, 2012). We first define a set of baseline explanatory variables used in the benchmark regression, and then consider additional variables for the sensitivity analysis. In all cases we distinguish between global and domestic determinants. A brief description of each variable is provided below; for further details refer to Table 6 in Appendix D. 5 During this period two natural disasters affected the region: the earthquake in Japan and the Thai floods. 11

12 Baseline regressors Regarding the global factors, we consider four variables: global risk, global liquidity growth, global interest rates and global growth. We proxy global risk by the US stock market volatility, measured as the VXO the implied volatility index calculated by the Chicago Board Options Exchange for the period , extended back to 1980 based on Bloom (2009). Growth in global liquidity is quantified by the yearly growth rate of global money supply; this measure is computed as the average of the growth rate of M2 in the United States, Eurozone and Japan and the growth rate of M4 for the UK. Global interest rates are calculated as the average of the interest rates on long-term government bonds in the United States, core Euro Area and Japan. And finally, global growth corresponds to the year-on-year growth rate in the World s real GDP. The source of the last three variables is International Financial Statistics (IFS) from IMF. We use a more comprehensive set of domestic factors relative to those considered in previous literature. The data series were obtained mostly from IFS complemented with Datastream and local sources whenever not available, unless otherwise stated. Economic performance is measured by the year-on-year growth rate of real GDP. As it was previously stated, a more countercyclical policy and better monetary policies can enhance the resilience of emerging markets to the vulnerabilities associated to sudden stop episodes. The proxy for soundness of the macroeconomic policy used in the baseline scenario is monetary stability measured by CPI inflation. We account for the soundness of the financial sector by incorporating to the regression bank credit to the private sector as percentage of GDP, obtained from Beck et al. (2009) 6. In addition, we introduce a measure of the degree of domestic liability dollarization, defined as bank foreign borrowing from IFS and Bank of International Settlements (BIS) as a share of GDP. As trade openess can increase the vulnerability of a given country to the international goods and asset markets, the ratio of real exports plus imports to GDP is considered. Following Calvo et al. (2008), we also include the current account deficit as a share of the absorption of tradable goods, which is computed as imports plus tradable output domestically consumed. The latter is calculated as the sum of agricultural and industrial output obtained from the World Development Indicators 6 Alternative measures considered are: our own measure of private credit to GDP constructed based on IFS data, credit to the private sector by financial institutions as percentage of total deposits in financial institutions also constructed from IFS data, and bank credit to the private sector as a percentage of total deposits in banks obtained from Beck et al. (2009). 12

13 (WDI) constructed by the World Bank minus exports. Contagion episodes are accounted for by including a dummy variable that takes the value of 1 if a country reports a sudden stop in t and there is at least one country with geographic proximity 7 with a sudden stop in t 1 8. The innovation of this paper, in terms of the determinants of sudden stops considered, is to include a variable that accounts for each country s institutional background. One of the important challenges of this variable is the lack of measures at high frequency. A way to overcome this issue is by making use of the composite risk rating index produced by the Political Risk Services Group. This index is composed of 12 components: government stability, socio-economic conditions, investment-profile, internal conflict, external conflict, corruption, military and politics, religious tensions, law and order, ethnic tensions, democratic accountability and bureaucracy quality. Since the individual indexes are also reported, we not only consider the overall index but also construct our own sub-index just with the categories that are relevant for this study: rule of law, investment profile, government stability, bureaucracy quality, and corruption. The last measure is the one used in the baseline regression, and the overall index, denoted as Political Risk is considered for the robustness analysis. Additional regressors Additional regressors are included in the sensitivity analysis of Section 4. Exchange rate flexibility is measured by the fine classification of exchange rate regimes constructed by Reinhart and Rogoff (2004) and updated by Iltzezky et al. (2009); higher values of this indicator is associated to a more flexible exchange rate regime. The degree of openness of the capital account is proxied by the capital controls index developed by Chinn and Ito (2006); the higher this index the lower the capital controls. A more detailed description of the series used to compute sudden stops and the variables involved in the regressions are presented in Table 7 in Appendix D. 7 The set of countries with geographic proximity to a country j is defined by all the countries that share a land border with country j. 8 An alternative measure of contagion considered in the robustness analysis is a dummy variable that takes the value of 1 if a country reports a sudden stop in t and at least one of the top 10 trading partners experienced a sudden stop in t 1. 13

14 3 Estimation Strategy and Results We construct a comprehensive dataset at quarterly frequency, from 1980 through 2014, which comprises in the baseline scenario 48 countries and includes all the variables detailed in the previous section. All country-level variables, except for the index of capital controls and the contagion variables, are winsorized at the 1 percent level to reduce the impact of extreme outliers. 3.1 Methodology In order to study the factors that prevent gross sudden stops to become fully-fledged net sudden stops, we use an estimation strategy that exploits the sequential nature of the problem. The problem addressed in this paper can be decomposed into two stages. First, the economy either experiences a sudden stop in gross inflows or it does not. If it does, then it can transition either into a prevented sudden stop or into a sudden stop in net flows. Therefore, the transition into a prevented sudden stop only occurs when the economy has experienced a sudden stop in gross inflows, and this should be taken into account at the time of estimation. The model we resort to in order to capture this phenomenon is the sequential logit model, initially proposed by Mare(1981) to describe the process of educational attainment and then applied to many other problems in the orbit of empirical microeconomics. The sequential logit model entails the estimation of separate logit regressions for each step, restricting the sample only to those countries at risk of making the transition. In other words, in the first stage, that we denote as inflows, we estimate a logit with the full sample of countries, while in the second state, that we call prevented, we restrict the sample only to those countries that in the previous stage experienced a sudden stop in gross inflows. The assumption made in this model is that, beside the temporal precedence, the decision in the first stage is independent from the one in the second stage, and this is the reason why it is valid to run separate regressions for each transition. 3.2 Baseline Results In column (1) of Table 2 we present our baseline regressions for gross sudden stops. Our results suggest that global conditions, in particular global risk and economic growth, are significant in predicting the transition of economies into periods of sudden stop in gross inflows. On the one hand, periods of higher global volatility are more prone to experience sudden stops in gross inflows; 14

15 on the other hand, periods of high economic growth reduce the incidence of such type of episodes. In addition, the incidence of sudden stops in inflows are also tight to domestic conditions. A higher levels of foreign liabilities, private credit, or the exposure of a border neighbor to a sudden stop increase countries vulnerability to stops in gross inflows; while economic growth reduces this vulnerability. These results are robust to the exclusion of the global financial crisis in 2007, as presented in column (9), although the relevance of global risk is undermined. Some interesting scenarios appear when we compare the previous analysis by groups of countries. Advanced economies appear more susceptible in periods of global volatility than non-advanced economies as seen by comparing columns (3) and (5). When we distinguish emerging markets from the rest of non-advanced economies, we observe a higher degree of vulnerability in periods of lower economic growth. Also, compared to Advanced economies, Emerging countries face higher difficulties with rises in inflation levels or in private credit. Once the economy has experienced a sudden stop in gross inflows, then it can transition either to a prevented or to a net sudden stop. Column (2) in table 2 presents the baseline regression that corresponds to the situation in which the economy transitions to a prevented sudden stop. In this case, global conditions do not influence the likelihood of preventing a net sudden stop; only domestic conditions matter. We find that countries with a higher GDP growth, a better institutional background and lower foreign liabilities have a higher likelihood of offsetting the foreign investor dynamics and the occurrence of a net sudden stop. As noted previously, a high degree of domestic liability dollarization (DLD) makes countries more prone to experience gross sudden stops. In addition, higher DLD reduces the likelihood for net sudden stop episodes to be prevented. This implies that DLD is important to understand the transitions into net sudden stops, which is consistent with the findings of Calvo et al. (2008). Column (2) presents also evidence of a positive effect of regional contagion in preventing sudden stops. Notice that we have considered land borders as source of contagion in our benchmark, and the significant positive coefficient can be accounting the systemic nature of these episodes or simply a catching up with the Joneses effect in the behavior of domestic investors vis-à-vis their neighbors. In order to validate this statement, we evaluate an alternative measure of contagion based on trade linkages in Table (3) and discussed in more detail in section (4.1). For now, it suffices to say that there is no longer a relationship between a sudden stop in a trading country partner and prevented episodes. Columns (9) and (10) show that the results previously discussed are not exclusively driven by 15

16 the global financial crisis of When excluding this period of global distress, we find that prospectives of economic growth, foreign liabilities and, to a lesser extent, institutional background are important in explaining the likelihood of preventing a net sudden stop. When the episodes from the financial crisis are included, the institutional background acquires more relevance in determining of prevented sudden stops. 4 Robustness and Sensitivity Analysis We conduct an extensive series of robustness and sensitivity tests including additional control variables, alternative measures of the variables presented in the baseline regression and different definitions of sudden stops. 4.1 Alternative Measures for Variables in the Baseline Regression We consider alternative measures of the explanatory variables included in the baseline regression in order to assess the validity of the chosen regressors. We show that the results are robust to variations in the number of periods necessary to adjust the current account, credit measures, institutional quality definition and to trade rather than regional contagion. In columns (1) and (2) of Table 3 we present the baseline model after adding up to 4 lags of the current account deficit (CAD) as a share of tradable absorption. As pointed out by Calvo et al. (2008), the reason for doing this is that the possibility of experiencing a sudden stop in the near future generates a change in relative prices that is not captured by the contemporaneous CAD over tradable absorption ratio, since by that time the current account gap would have closed and the relative prices would have adjusted. The inclusion of additional lags does not affect our main results. We evaluate different measures of credit conditions. Notice that in our baseline regression, we include the measure of private credit by banks as a percentage of GDP developed by Beck et al. (2009). In columns (5) and (6) we introduce an alternative measure also based on these authors and find no sizable distinctions compared to the benchmark. Once a broader measure of credit (total credit as % of GDP) is introduced in columns (3) and (4) the transitions in the second stage are clearly affected. In this scenario, global conditions such as risk and liquidity growth are now significant. We see this result as evidence that private credit conditions in each specific country account for a fraction of global risk and liquidity growth and consider this channel worth exploring 16

17 in future research. As the measure of institutional background is relevant in several of our regressions to reduce the probability of a net sudden stops. We evaluate the significance of this variable by introducing the overall index of political risk produced by the Political Risk Services Group, and not just specific subcomponents as in the baseline regression. The results reported in columns (7) and (8) are robust to the inclusion of this alternative measure of institutions and are not quantitatively different from our baseline results. Finally, in columns (9) and (10) we replace our measure of regional contagion by one of trade contagion. In this new scenario, if a trading partner of any given country experiences an episode of gross sudden stop, then the likelihood of a sudden stop increases for this country. Nevertheless, trading contagion does not appear to be relevant in explaining how a sudden stop in net flows can be prevented. 4.2 Alternative Definition of Episodes Systemic and Bonanza Related Episodes We assess in this section the validity of our results in light of alternative definitions of sudden stops, that account for potential mitigating or reinforcing triggers. We introduce two additional extensions to the standard definition of sudden stops: systemic sudden stops and bonanza-filtered sudden stops. Although some of these definitions are not totally comparable with the ones used in our baseline regression, they shed light on the relevance of specific country factors depending whether the economy is going through periods of distress or not. Systemic sudden stops, as introduced by Calvo et al. (2008), are associated to large and unexpected capital account movements. They are of particular importance given the nature of the recent global crisis. A systemic sudden stop is defined as a gross sudden stop that coincides with a sharp increase in aggregate spreads. More specifically, it is a dummy variable that takes the value 1 if a gross sudden stop episode coincides with an aggregate-spread window, and 0 otherwise. An aggregate-spread window is defined as a dummy that takes de value 1 if the sovereign bond spread rises beyond two standard deviations from its historical mean. In terms of measuring its length in time, this window starts from the moment in which the series rises one standard deviation above its historical mean, but conditional on the fact that it will eventually cross the two-standarddeviations threshold. The window ends when the series goes back to one standard deviation above the historical mean. 17

18 We also study the determinants of bonanza-filtered sudden stops. Favorable terms of trade shocks can add sources of financing directly from the current account; this fact can potentially dismiss the need of domestic agents in offsetting the contraction in capital inflows. As there is no specific metric to catalogue bonanza episodes, we characterize two alternative measures for this purpose. First, we define our bonanza measure (bonanza1) as a period in which the year-on-year growth of the terms of trade exceeds 4%, similarly to Cavallo et al. (2013). Alternatively, a bonanza is defined as a terms of trade window, i.e. a dummy variable that takes de value 1 if the seasonally adjusted terms of trade rise beyond two standard deviations from their historical mean (bonanza2). The length of bonanza episodes is determined from the moment in which the series increases one standard deviation above its historical mean to the moment in which returns to be within the one standard deviation threshold. Based on these distinction, columns (1) and (2) in Table 4 show the determinants of the transitions to systemic gross and net sudden stops, respectively. By only considering systemic episodes, we are focusing on periods in which the economy exhibits large and unexpected cuts in capital inflows that occur simultaneously with sharp increases in aggregate interest-rate spreads. Since the determinants of the sudden stops in this case are mostly external and of a financial nature, then we expect the results to differ from those in our baseline regression. It is not surprising that higher global interest rates increase the likelihood of a systemic sudden stop since, as by construction, systemic episodes are defined as periods of high interest rates. In terms of domestic factors, a gross systemic sudden stop is highly influenced by the perspectives of growth in the economy. Once the countries have experienced a systemic sudden stop in gross inflows and transition to the second stage, the likelihood of preventing a systemic net sudden stop is affected by global conditions. In fact, lower interest rate increase the chances of preventing a systemic net sudden stop. Instead, the results suggest that the ability to prevent a systemic episodes is undermined by its global nature. In fact, we find little evidence that particular domestic conditions boost the countries resilience. Now we move to analyze episodes of sudden stops accompanied by a bonanza in the terms of trade. An improvement in the terms of trade can reduce the reliance on the compensating behavior in domestic agents after the disruption of capital flows. Thus, we evaluate how sudden stops are more likely to be prevented in absence of a funding mechanism stemming directly from the current account. Columns (3) to (6) in Table 4 depict the results when bonanza episodes are excluded. Our first measure bonanza1 which is less restrictive in terms of the episodes classified as bonanzas is 18

19 presented in columns (3) and (4). Note that the estimates for the transitions to prevented sudden stops are unaffected by global conditions; prevented episodes are exclusively determined by the degree of DLD. A more restrictive definition of bonanza based on deviations from two times the standard deviation of historical terms of trade is presented in columns (5) and (6). In this scenario, there are few variation from our baseline scenario; sudden stops in gross inflows are predominantly influenced by global conditions while the likelihood for these episodes to be prevented is only determined by domestic characteristics Preventable Episodes Theability of a country to prevent a sudden stop can be influenced by the size of its assets available; in other words, if existing assets are enough to offset the change in capital inflows. These assets can be the result of either capital repatriation or simply a reduction in planned outgoing flows; unfortunately given data availability is impossible to disentangle the origin of outflows; in particular, the fraction corresponding to the reduction of investment abroad from domestic investors. Based on these restriction we can only proxy this capacity to respond in magnitude to a sudden stop in net inflows, making use of the existing stock of assets abroad. We test the baseline results, excluding episodes in which the ratio between the stock of assets (during previous quarter) and the change in capital inflows duringasudden stop is two standard deviations below its historical mean. 9 Results are presented in columns (7) and (8) in Table 4. There is no significant variation compared to our baseline scenario. Although we acknowledge the challenge behind a true understanding of the capacity to prevent a sudden stop Domestic Private Agents Hitherto, the analysis has considered both private and public outflows and their role offsetting changes in capital inflows. In this section we specifically address how private agents can mitigate the incidence of a sudden stop. For that purpose we construct a series of capital outflows considering only private flows and reconstruct the episodes of net sudden stops based on this newly created series. 10 Column (9) in Table 4 presents the results of the second transition when only private assets are considered. The results from our baseline framework are also consistent even when we 9 This measure accounts for the possibility that not all assets are susceptible to repatriation. The results reported in this section are robust to the period in which assets are measured. 10 For a detailed description on how private flows are constructed, please refer to appendix A 19

20 exclude the public sector. Domestic private investors are more responsive to domestic conditions on the verge of a sudden stop in capital inflows. And this responsiveness is tied to lower levels of foreign liabilities, better institutional quality and influenced by country neighbors conditions Sudden Stops and Retrenchments Finally, we study the determinants of episodes of sudden stops in net inflows that are accompanied by retrenchments in net outflows. Notice that as stated previously capital retrenchments are neither necessary nor sufficient to prevent a net sudden stops. In fact, the results reported in column (10) of Table 4 suggest that prevented sudden stops and retrenchments are driven in certain degree by different domestic conditions. The simultaneity of a sudden stop and a retrenchment appear to be determined by credit conditions and in a less extent to the degree of domestic liability dollarization as is the case of prevented episodes. Since the resilience of a country to a sudden stop builds upon the capacity of domestic agents to offset the fall in foreign inflows, it is necessary to understand the distinction between retrenchment and prevented episodes and the condition driving each one of these episodes. 4.3 Exchange Rates and Capital Controls Some authors have emphasized how exchange rate flexibility is an important tool to reduce capital flows volatility. In this section we evaluate the role of exchange rate flexibility in preventing sudden stop episodes. We account for the effect of different exchange rates regimes using the index constructed by Reinhart and Rogoff (2004) and updated by Iltzezky et al. (2009). The results are reported in columns (1) and (2) of Table 5. We find that the degree of exchange rate flexibility is not relevant in explaining neither the transitions to gross sudden stops nor the transitions to prevented sudden stops after controlling for several global and domestic conditions. This result is robust to the use of course classification in Iltzezky et al. (2009) or differences with respect to mean as in Cifuentes and Jara (2014) (results not tabulated). However the analysis of exchange rate in isolation can be misleading. In fact, the consolidation of macro policies enhancing a higher degree of flexibility in exchange rates have been accompanied by inflation targeting as the monetary anchor in several emerging economies. We add as additional regressors a dummy variable (IT) that takes the value of 1 if the country has adopted inflation targeting. We also combine categories in Iltzezky et al. (2009) to construct a dummy variable 20

21 (Flex) that takes the value of 1 if the country has a flexible exchange rate scheme. And finally, we add the interaction between these two indicator variables. How this combination of policies are relevant for prevented episodes is presented in columns (3) and (4) in Table 5. We found that higher flexibility can increase the resilience to a sudden stops in net inflows if it also involves a de-jure commitment to stabilize the price level in the economy. Finally we address if capital controls have any influence in preventing or not a net sudden stop. Columns (5) and (6) of Table 5 show the estimation results after introducing the measure of capital controls constructed by Chinn and Ito (2006). These capital controls are not significant in explaining the transition to gross sudden stops, which are still mainly determined by global risk and growth. However, the degree of capital account openness increases the likelihood of preventing a net sudden stop. The estimation results of the second stage regression indicate that global conditions are insignificant in preventing net sudden stops; the main determinants of prevented sudden stops are the degree of DLD, the prospectives of economic growth, the institutional background and regional contagion. 5 Final Remarks This paper presents evidence that the capacity to build resilience and offset the volatility of capital inflows, is tied mainly to sound domestic conditions. Overall a lower level of domestic liability dollarization, better prospects of economic growth and institutional quality are important to determine the extent at which capital outflows can accommodate a sudden stop in capital inflows. We have emphasized that these changes in capital outflows are neither necessarily nor sufficiently related with capital retrenchments. In other words, sharp variations in capital outflows cannot suffice to prevent a fall in net flows, or small variations can be enough to prevent a net sudden stop. This makes the problem of prevented sudden stops a different one from the analysis of simultaneous sudden stop in capital inflows and retrenchment in capital outflows. 21