The Consequences of Sudden Stops in Gross Capital
Inflows on Industry-Level Growth
Christine Annuß
*Supervisor: Dr. Robert Inklaar
Co-assessor: Prof. Dr. Ingo Geishecker
July 6, 2012
Abstract
This paper analyses the consequences of sudden stops in capital flows on the growth of industries at different levels of dependence on external finance, with the focus on the distinction between stops in net and gross capital flows. Using data from 29 developing and developed economies with 22 net stop and 29 gross stop events during the period 1980 – 2002, the results suggest that firstly, the implications for net and gross capital flows are largely similar, and secondly, externally dependent industries suffer less from sudden stops if financial development is high. Further, the roles of growth opportunities and banking crises in this framework are examined.
Keywords: sudden stops, gross capital flows, industry characteristics
* University of Groningen, Faculty of Economics and Business, student number: 2243784, email:
1 1 Introduction
During the past three decades global capital flows have increased considerably which, as a downside, is often suspected of leaving economies more exposed to the risk of considerable and largely unexpected drops in capital inflows due to an external shock. Such drops in capital inflows are referred to as sudden stops and can have severe consequences for the real economy. While the existing literature has extensively dealt with the causes and consequences of sudden stops in net capital inflows, the effects of stops in gross capital inflows are as yet a largely uncharted terrain. However, there are a number of reasons why the two different types of sudden stops may affect economies in different ways, the main reason being that they are based on different concepts; while net capital flows are defined as the difference between the net purchases of foreign assets by domestic agents on one hand, and the net purchases of domestic assets by foreign agents on the other, gross capital flows only encompass the latter type of purchase, which is equivalent to foreign capital flowing into the economy. Hence, sudden stops in net and gross capital flows may be driven by different agents and therefore by different motives. Based on this, it is likely that the two concepts of sudden stops differ empirically in terms frequency, timing and severity, and as a consequence also with regard to their effects. However, it seems virtually impossible to predict which one affects the economy in which way and for which type the impact is likely to be more severe.
2
institutions in highly developed financial systems are most likely to allocate resources to industries with good growth opportunities. For the first time, this paper introduces this perspective in the framework of sudden stops as an alternative to the dependence on external finance. Lastly, the literature on banking crises is taken up in this study, and their interaction with sudden stops is analysed in more detail, by accounting for stops in net and gross flows.
Based on this, the empirical analysis attempts to provide evidence on three main issues, in each case distinguishing between the implications of sudden stops in net and in gross flows. Firstly, the effect of the interaction between financial development and external dependence on industry growth is analysed when a sudden stop occurs. It is assumed that externally dependent industries fare better in highly developed financial systems, as they are more likely to be capable of compensating the loss of access to foreign finance caused by the stop. Secondly, a similar analysis is applied to the interaction of financial development and the growth opportunities of individual industries. The conjecture here is similar: Financial development is expected to make sectors with good growth opportunities better off even if the sources of foreign capital dry up during a sudden stop, indicating that the interaction has a positive influence on industry growth. Lastly, this paper takes into account the possibility that sudden stops per se may not exert any influence on the link between the aforementioned interactions of country and industry characteristics on one hand, and industry growth on the other, but may just be relevant if they coincide with banking crises which weaken financial intermediation. The analysis is based on a total sample of 29 developing and developed countries with 22 net stop events and 29 gross stop events during the period of 1980 – 2002.
Methodologically, the present study chooses an approach to examine these issues which is relatively new to the analysis of sudden stops. This approach implies accounting for the consequences of sudden stops on industry-level growth rather than on the country level. This allows controlling for certain characteristics of industries, specifically the dependence on external financing and growth opportunities, which are expected to be major determinants of an industry's vulnerability to sudden stops. In order to analyse their impact on growth, this paper distinguishes between periods preceding the sudden stops and the stop periods themselves. The relationships are then estimated with the aid of OLS techniques with the difference between industry growth before and during the stop as dependent variable. In each case, the analysis also considers the implications for different levels of economic development, as well as for an alternative measure of external dependence, and most importantly, for stops in net as well as gross capital flows.
3
literature, first providing a theoretical background on the balance sheet effects associated with sudden stops. Then a variety of studies on the consequences of stops for the real economy is discussed, and the different concepts of net and gross capital flows as well as their implications are explained and compared. Based on this, two hypotheses which underlie the empirical analysis of this study are derived. Chapter 3 discusses different methodologies which are available to analyse the issues under study and then introduces the methodology applied here. Chapter 4 defines the variables, presents the data and provides a first descriptive analysis. The regression results are presented and discussed in chapter 5, which is subdivided into sections on the role of external dependence and alternatively growth opportunities, and lastly on the influence of banking crises. Chapter 6 concludes and suggests issues for future research.
2 Literature Review
The world economy is subject to constant changes; technical innovations have facilitated communication and transactions, economies have developed closer links among each other in trade and finance, and financial markets have become more globalised. This evolution of financial globalisation is praised by some and demonised by others, and economists in their attempt to identify and quantify the implications and consequences have turned their attention to various fields of study in this area, without reaching a general consent. The main focus of this work is on international capital flows, more specifically on sudden stops of capital inflows, which are likely to play an important role when capital becomes more mobile in globalised financial markets, and which have attracted special interest in particular since the 1990s, when they were a common phenomenon not only in emerging economies. This chapter provides an overview of relevant literature related to the broad topic of sudden stops, starting out from a theoretical background of balance sheet effects associated with sudden stops in the next section. Following this, several perspectives on the link between stops and real economic activity are presented, with focus on different aspects that influence this link, like trade openness, capital mobility, financial frictions, growth opportunities and the choice of level of aggregation. Then, the role of gross capital inflows is discussed and lastly, two research hypotheses are derived. First however, a brief overview of the history of sudden stops and a preliminary definition are in order.
4
capital account restrictions with the intention of preventing capital flight, whereas more recently their purpose has shifted to the restriction of large capital inflows (cf. Edwards 2007, 76). Then, in the end of the 1980s, many of those economies started opening up their capital accounts and removing restrictions on capital flows, thereby increasing international capital mobility. This trend is criticised by many, arguing that freely moving capital leads to macroeconomic instability and higher financial vulnerability towards external shocks and crises (cf. for example Stiglitz 2002). According to this, a likely consequence of removing capital restrictions has been the rather frequent occurrence of sudden stops, as for example in the early 1990s in the Mexican Tequila Crisis (1994/95), then also affecting many Asian economies (cf. for example Durdu et al. 2009, 194f., Guidotti et al. 2004, 171). The financial crises in large parts of Latin America, among them the Mexican crisis, were often blamed on low savings rates, fiscal and current account deficits. But in contrast to this, the countries affected by the Asian crisis in 1997/98 displayed high savings rates and mostly less severe current account deficits (cf. Calvo 1998, 36). Yet, sudden stops may not be perceived as an issue merely affecting the developing world as they have frequently occurred also in many developed countries, for instance in Europe. In general, capital flows have increased, particularly after the recent global financial crisis, which may imply that sudden stops are likely to occur more often and more severely in the near future (cf. Cowan and Raddatz 2011, 1). Because of this it is of such importance to study and understand the mechanisms associated with sudden stops, and this can only be done with the aid of an appropriate method to identify them.
5
(Mendoza 2002, 335). Due to these characteristics, severe damage to real economic activity is a likely consequence, and section 2.2 provides an overview of the existing literature on possible consequences. As a basis for an empirical analysis, two alternative concepts of sudden stops are available, one being based on cuts in net capital inflows, the other on drops in gross inflows, where the latter is a rather novel approach and still requires more investigation. As a contribution to existing literature, this paper takes into account both concepts, with the main objective being a comparison of the two, of their strengths and weaknesses and of the empirical findings they imply. A more detailed discussion of the alternative concepts is provided in section 2.3.
2.1 Theoretical Background: Balance Sheet Effects
In order to understand the phenomenon of sudden stops it is helpful to consider the balance sheet effects of capital inflows and their stops. The work of Calvo (1998) is one of the first studies to provide an analytical approach to sudden stops in this context. In his framework net1 capital inflows (KI), implying a capital account surplus, are associated with a current account deficit (CAD) as shown in equation (1) thus bringing the balance of payments into equilibrium (cf. Calvo 1998, 37).
KI = CAD (1)
An exogenous sudden stop corresponds to a sharp decline in capital inflows KI, which according to equation (1) is associated with a corresponding decline in CAD. How is this possible? Equation (2) allows a closer look at the components of CAD:
CAD = Z – Y – NFTA (2)
CAD equals the difference between the absorption of tradable goods (Z) and the output of tradables (Y) and what is transferred abroad (NFTA, net factor transfers abroad). If in this scenario absorption Z is higher than what is produced, the excess demand is met with imports, which leads to a CAD. In turn, the decline in CAD required to accommodate a sudden stop in capital inflows implies that demand Z needs to fall as well in order to restore equilibrium (as reducing the production obviously is not a desirable option) (cf. Calvo 1998, 38).
1 Sudden stops based on gross capital flows are seen as a reasonable alternative to net flows, but the theory on
6
If this absorption Z falls in order to accommodate a sudden stop, the composition of Z plays an important role for the subsequent effects. Assuming that a large share of this demand is made up of consumption (which is assumed to be relatively labour-intensive) rather than investment, the demand for labour declines with a fall in Z. Hence, as labour is seen as a non-tradable good, the demand for non-non-tradables decreases as well, which results in an increase in the relative price of tradables to non-tradables, and thus an increase in the real exchange rate (cf. Calvo 1998, 38f.). The reaction of the real exchange rate and therefore an economy's vulnerability to sudden stops depends on the reduction in demand needed to close the current account gap (the ratio of CAD to Z), with a decline in Z leading to an increase in the exchange rate. Moreover, the higher the need for financing from abroad (leverage) in the absorption Z, the higher the real exchange rate reaction (cf. Calvo et al. 2008, 15f.). Furthermore, this rise in the real exchange rate changes the ratio of foreign exchange denominated debt to GDP, thereby increasing the likelihood of financial imbalances (cf. Calvo et al. 2008, 4f.). This is called domestic liability dollarisation and is crucial for emerging economies, where debt is mostly denominated in terms of US dollars or other strong currencies. Therefore, in a highly dollarised financial system, the liquidity constraint associated with a sudden stop has an even stronger adverse impact on the real economy (cf. Mendoza 2002, 338). As a conclusion, theory predicts that the adverse effect of a sudden stop (in net flows) on the economy is more severe when the current account deficit and dollarisation are large.
2.2 Consequences of Sudden Stops on the Economy
7
(cf. Cavallo and Frankel 2008, 1431). Similarly, Kose et al. (2009) establish a link between sudden stop and debt default probability of an economy. They argue that for a required current account adjustment, larger real exchange rate depreciations are needed in countries with lower trade openness, entailing more severe balance sheet effects and consequences for the real economy, which increase the likelihood of default. But most importantly, trade openness can facilitate the recovery after a potential crisis, and thus an open economy may be able to service its debt rather than to default, and thereby “export its way out of a recession” (Kose et al. 2009, 24).
One example of a study documenting the mitigating effects of trade openness, among other aspects, is provided by Edwards (2007). In his rather general and comprehensive work, he combines different approaches of studying sudden stops in one analysis which is based on cross country data of 157 countries from the 1970s to 2000. In a joint estimation of external crisis probability and the decline in real GDP growth caused by such a crisis (he considers both sudden stops and current account reversals2), the results of his growth regression show that with low trade openness, the negative effect of a current account reversal on economic growth is strong, whereas less capital mobility reduces the damage done by the reversal. To illustrate this, comparing two countries with trade openness of 60 per cent shows that a country with a low degree of capital mobility with an index of 25 faces a growth decline of only 3.48 per cent, but a country with high capital mobility with an index of 90 experiences a growth decline of 5.43 per cent following a capital account reversal (cf. Edwards 2007, 105f.). Summing up, the consequences of reversals are more costly when countries are not open to trade and capital is very mobile. Yet, capital mobility per se is not found to increase the probability of external crises like sudden stops or current account reversals; but as soon as one of these events occurs, a high degree of capital mobility makes it more damaging.
Related to this, Guidotti et al. (2004) show how trade openness can impact the consequences of sudden stops in very different ways, by comparing the two sudden stop regions of Latin America and Asia. In their pooled regression analysis, they find that output in economies relatively open to trade, and especially in economies with a floating exchange rate regime, recovers relatively quickly from sudden stops: A floating regime on average improves the output performance by four to six percentage points after a stop. On the other hand, it is more difficult for economies with high domestic liability dollarisation (i.e. the ratio of foreign exchange denominated debt to GDP) to regain economic growth, as they typically suffer
2A current account reversal is defined as the reduction of the current account deficit of at least 4 percent of GDP
8
strong import contraction rather than export growth (yet, while being statistically significant, the economic effect is rather negligible) (cf. Guidotti et al. 2004, 191ff.). As an example, adjustment in Latin America tended to be slow, mostly working through import and demand contraction and being associated with exchange rate depreciations. On the other hand, several Asian economies relied on export growth to cope with the loss of access to capital markets and like this experienced quick recoveries after their sudden stops (Guidotti et al. 2005, 172, 198f.).
Note that from a theoretical point of view, sudden stops may even lead to output increases. As sudden stops imply that access to international capital markets and therefore external financing is lost, an economy affected by such an event will increasingly rely on exports in order to pay its debts. To avoid a decrease in consumption due to rising exports, consumers are likely to work more and thus increase production, indicating that a sudden stop in the absence of other frictions theoretically can have a positive impact on output (cf. Chari et al. 2005, 380ff.). This view is formalised in equilibrium models discussed by Kehoe and Ruhl (2009), where an increase in output is predicted due to a sudden stop, as the reversal of the current account balance causes a decrease in the leisure consumption (cf. Kehoe and Ruhl 2009, 236).
How can it be explained that the experience of most sudden stop economies in reality deviates considerably from this simple theory? Sudden stops are often triggered by external shocks and these can be modeled as the abrupt tightening of the collateral constraint on foreign borrowing, which depends on a country's reputation in international financial markets. If a country's budget is hence constrained, this provides the basis for the possibility of negative real economic effects of sudden stops of capital inflows (cf. Chari et al. 2005, 381). More specifically, it is the presence of certain frictions that causes sudden stops to result in output drops, the most relevant type being financial frictions. Apart from constraints on international borrowing, financial frictions can also occur on the firm level, for instance as working capital financing constraints. In combination with such frictions, sudden stops are expected to have more severe consequences for the economy, especially in countries and industries that are more vulnerable to financial frictions. Above all, this is the case when an industry is dependent on external financing (cf. Cowan and Raddatz 2011, 3).
In order to study the role of such financial frictions empirically, Kroszner et al. (2007) analyse the impact of banking crises3 (instead of sudden stops) on industry value added
3 According to Caprio and Klingebiel (2002), a banking crisis implies that the banking sector of a country
9
growth, focusing on the interaction of external dependence and financial development. This approach is based on the pioneering work of Rajan and Zingales (1998), who establish that industries which are more dependent on external finance grow faster in countries with well-developed financial markets. In such a well-developed financial system, access to external finance is less costly, as financial institutions efficiently allocate resources to their most productive use and to where there are needed most, in this case to industries which depend on external finance relatively more. Using data from 38 countries over the period from 1980 – 2000, Kroszner et al. (2007) build on these findings by introducing banking crises as external shocks into this framework; they perform an ordinary least squares (OLS) analysis to compare growth before, during and after the crises. In line with Rajan and Zingales (1998), they find that in absence of a banking crisis, financially dependent sectors on average grow faster. However, during a banking crisis, those sectors perform worse than industries with low external dependence, especially in countries with deep financial systems. The contraction in industry growth during the banking crisis experienced by an industry at the 75th percentile of external dependence in a country at the 75th percentile of financial development is 1.6 per cent larger than that for an industry where the levels of both measures correspond to the 25th percentile (cf. Kroszner et al. 2007, 201). The main message is that in countries where the ratio of private credit to GDP is high, banks play a particularly important role as financial intermediaries4 and when these intermediaries are lost due to a banking crisis, this affects above all those sectors which rely on external finance relatively heavily. These results are in line with the financial frictions view illustrated above.
Similarly, the evidence presented in Cowan and Raddatz (2011) shows that in industries depending strongly on external finance, the decline in industrial production due to a sudden stop is clearly larger than for industries with low external dependence. Using a difference-in-differences approach, they focus on the effects of sudden stops on industry growth in a sample of 45 countries. To provide an example, in an industry whose external dependence is one standard deviation above the mean, the decline in industrial production is three percentage points larger than in an industry with external dependence of one standard deviation below the mean. This effect is found to be even stronger in emerging economies. All this points to the importance of financial frictions, like micro-level financing constraints, in determining the effects of sudden stops on the industry level (cf. Cowan and Raddatz 2011, 16f.). Furthermore, especially those sectors that depend on external finance only to a small
4 For an exhaustive overview of theoretical and empirical literature on the link between finance (financial
10
extent and therefore are less sensitive to financial frictions seem to be the ones that have driven the rapid recovery of some emerging economies after sudden stops, in some cases even without considerable expansion of credit (Phoenix miracles) (cf. Cowan and Raddatz 2011, 8). Lastly and contrary to the results of Kroszner et al. (2007) for banking crises, Cowan and Raddatz (2011) conclude that the adverse effect of stops on externally dependent industries deteriorates if financial development is low.5
As an alternative to the analysis of financial frictions, other industry characteristics may be considered, as for instance an industry’s growth opportunities. Fisman and Love (2007) reconsider the study of Rajan and Zingales (1998), arguing that in a broader view of financial development, the notion of financial institutions channelling resources to their most productive use may also imply that resources are allocated to industries and firms to enable them to take advantage of their growth opportunities (cf. Fisman and Love 2007, 471). Hence, in a developed financial market, resources are allocated to industries with the best opportunities for growth, rather than only to those that are most dependent on external finance. The assumptions underlying the analysis of this hypothesis are that growth opportunities of all industries in all countries are affected by global and industry-specific shocks (for instance due to technological innovation or factor price shifts) and further, that industries in the U.S. respond perfectly to these shocks, due to the highly developed financial market. If this is the case, then U.S. industry growth is a proxy for growth opportunities and, thus, can be used as benchmark for growth opportunities of industries in other countries (cf. Fisman and Love 2007, 472). As predicted, the results show that in countries with highly developed financial markets, value added of industries with good growth opportunities grows faster.6 Yet, if external dependence and growth opportunities (both interacted with financial development) are included in the regression simultaneously, only the interaction of financial development and growth opportunities is significant. Because of this, the growth opportunities view is seen as a broader view of financial development, as it is found to encompass the external dependence approach rather than substituting it (cf. Fisman and Love 2007, 474f.).
5 However, the results for countries with high financial development are insignificant, so that it may not be
concluded that financial development has a positive effect on the link of external dependence and industry growth during sudden stops (cf. Cowan and Raddatz 2011, 17).
6 For another study on the link of financial development and resource allocation that also finds a positive link
11
2.3 Sudden Stops in Gross Capital Flows
In spite of large conceptual differences between net and gross capital flows, the majority of the empirical literature on sudden stops is based on net flows. And even if these studies usually work with data on net flows, the concept which is described in most of them, according to Rothenberg and Warnock (2011), in fact is that of sudden stops in gross capital inflows (cf. Rothenberg and Warnock 2011, 509). How do these two concepts differ? Net capital flows are “defined as the difference in gross capital flows, that is, the net purchases of domestic assets by foreign agents minus the net purchases of foreign assets by domestic agents” (Broner et al. 2011, 1). This definition shows that net stops do not distinguish between actual drops of capital inflows (true sudden stop) and alternatively increases in investment abroad (capital flight or sudden start)7, which would have the same effect on the capital account balance (cf. Broner et al. 2011, 2). Thus, this implies that a sudden stop in net flows does not necessarily mean that the country loses access to foreign capital, as put forward by theory. In contrast to this, “true sudden stops are episodes in which gross capital inflows decrease more than gross capital outflows increase” (Rothenberg and Warnock 2011, 1). To provide an idea of the relative importance of those different concepts, it has been shown that just one out of five stops in net flows is driven by increases in gross capital outflows, whereas drops in inflows are far more common (cf. Cowan et al. 2007, 2). However, the incentives behind the actions of domestic and foreign agents are different ones, which can be explained for instance by asymmetric information concerning the return of domestic assets between foreign and domestic agents, and those asymmetries are increased during financial crises (cf. Broner et al. 2011, 5).
A number of studies deal with empirical implications of net and gross capital flows. For instance, Broner et al. (2011) compare the dynamic behaviour of gross capital flows in contrast to net capital flows during the business cycle as well as during financial crises and present very different results for the two kinds of flows. Most importantly, gross capital inflows (international capital flows by foreign and domestic agents) are found to be larger and more volatile than net flows (cf. Broner et al. 2011, 12). This clarifies why the identification of gross capital flows can be of importance, also when studying the effects on the real economy. Cowan et al. (2007) apply the distinction of gross and net flows to the analysis of sudden stops and find that sudden starts (of gross capital outflows) affect output and
7 For a study on outflow induced stops, see Faucette et al. (2005). For the distinction of gross capital inflows and
12
investment less strongly than gross sudden stops (cf. Cowan et al. 2007, 5). Also Rothenberg and Warnock (2011) graphically establish that the so called true sudden stops are associated with larger declines in GDP, investment and consumption than sudden flights (cf. Rothenberg and Warnock 2011, 515). This could possibly imply that the average effect of net stops relative to gross stops is weaker, as it might be mitigated by the inclusion of the less damaging sudden starts. But what should matter empirically is which of these effects is larger. Lastly, Cowan et al. (2007) make one step in the direction of comparing gross and net sudden stops by investigating how many stops in gross flows also manifest themselves in net stops; they find that only 42 per cent of the gross stops in their sample coincide with net stops (cf. Cowan et al. 2007, 6), which implies that analyses relying on net capital flows are likely to disguise a considerable number of true sudden stop events and may disregard the information connected to them.
2.4 Contributions of this Study
Although waves of gross capital in- and outflows have been compared theoretically and empirically, literature comparing net and gross stops is scarce, not to say non-existent. This paper breaks ground by performing such a comparison, and analyses the implications for the real economy of both types of sudden stop. Of course the effect of stops on economic activity has already been studied in various ways, yet these results in most cases are based on stops in net capital flows, which might differ from gross flow stops to a certain degree, with regard for instance to causes, frequency, timing, and severity. Due to these differences, it is conceivable that the results for gross stops in general differ from those for net stops, and the empirical analysis in this paper makes an attempt to identify some of those differences. This leads to the following hypothesis:
Hypothesis 1) Sudden stops in net and gross capital flows have different impacts on
13
the quality or availability of banks as financial intermediaries, and at the same time cause a loss of access to finance, whereas sudden stops just imply a drop in capital inflows. Thus, while high financial development in the case of banking crises signifies an even worse damage to the real economy, it is likely to have a more mitigating effect when sudden stops are considered. Therefore, contrary to Kroszner et al. (2007), the following result is expected:
Hypothesis 2) The interaction of the measures of financial development and external
dependence has a positive effect on industry growth when a sudden stop occurs. Note that Hypothesis 1) is implied in the second hypothesis, as the outcomes are expected to differ for the two types of sudden stops. Moreover, Hypothesis 2) entails an important feature of this study: In order to control for industry characteristics like external dependence, the data under study is on the industry level, and literature on the consequences of sudden stops in this context is still scarce. This level of aggregation brings with it several advantages, one of them being that it overcomes the issues of reverse causality and endogeneity of the occurrence of sudden stops and financial development (chapter 3 discusses these aspects in more detail).
In section 5.2, Hypothesis 2) is modified by considering an industry’s growth opportunities instead of and in addition to its external dependence. This novel approach to sudden stops is taken in order to identify another case where outcomes for gross and net stops could potentially be different, and if in this case the same logic applies as for the external dependence view, and the intermediary quality of financial institutions is not affected (remember that sudden stops are considered here, not banking crises), similar results are expected. The next two chapters describe the methodological approach and the data used to explore the hypotheses, and chapter 5 presents the empirical results.
3 Methodology
Concerning the methodology of this paper, there are three central aspects which have to be discussed. One is the calculation of sudden stops, the next is the definition of the dependent variable to capture the consequences of stops, and the last one is the econometric method applied to analyse these relationships. This chapter reviews how these issues are dealt with in existing literature and thereby derives the methodology used in this paper.
14
Guidotti et al. (2004) and Cowan and Raddatz (2011), a sudden stop is an annual change of the capital account by one standard deviation below the average as well as five per cent of the GDP of a country in one year. Calvo et al. (2008) extend the one-standard-deviation definition and also require the capital account to finally reach a level two standard deviations below the average. Independently of those details however, the concept underlying the sudden stops is of major importance but has largely been neglected so far. Specifically, this paper is based on the two different concepts of sudden stops in net and gross capital inflows. The concept of net sudden stops used here draws on the work of Cavallo and Frankel (2008) on the link between a country’s trade openness and its vulnerability to sudden stops. They define a sudden stop as a reduction of a country’s financial account surplus and a reduction of the current account deficit, which coincide with a fall in GDP. This being disruptive for the economy ensures that the fall in the current account deficit is not just due, for instance, to a rise in exports. Analytically, this is represented by a drop in the financial account surplus of a country by at least two standard deviations below its sample mean in one year compared to the past year, and a reduction of both the current account deficit and GDP per capita in the same or the subsequent year. According to this methodology, they calculate 86 sudden stop events in 67 countries during the years 1970 – 2002 (cf. Cavallo and Frankel 2008, 1433f.). This definition combines the most important elements of calculating sudden stops as suggested in related literature, and is used here not least because the list of sudden stop events in their study covers a large number of countries and years.
15
International Monetary Fund’s International Financial Statistics (IFS) (cf. Forbes and Warnock 2011, 7ff.). Like this, they obtain a number of 221 sudden stop events in 58 countries in the period of 1980 – 2009.
As the empirical analysis of the present paper deals with annual data, the quarterly observations of Forbes and Warnock (2011) need to be assigned to sudden stop years. Therefore, a stop is required to last for at least two quarters of a year to make it a sudden stop year; moreover, the stop has to persist for more than two quarters, while two events that are only one quarter apart are counted as one event if the other conditions are met. Arguably, this is a rather arbitrary way of aggregating stop quarters to stop years; but as the net stop sample contains 22 sudden stop events and the gross stop sample 29 (see chapter 4 for detailed descriptions of samples and stop events), the difference in the number of stops in both samples is not considered large enough to distort the results, and therefore the method of aggregating stop quarters is deemed a reasonable approximation. Yet, future research is needed to assess this assumption. Another difficulty may arise from the fact that sudden stops are defined slightly different in both underlying studies, as Forbes and Warnock (2011) fail to consider the accompanying drops in current account deficit and GDP. However, as the investigation of gross sudden stops is still in the early stages, future research also needs to be devoted to this issue.
While a number of dependent variables are conceivable,8 this paper examines the effects of stops on the industry level, which entails a number of advantages. Firstly, the problem of reverse causality between economic performance and the occurrence of sudden stops is largely avoided; on the aggregated country level, anticipated output drops may cause investors to withdraw their capital from a country, possibly leading to a sudden stop. Whereas for individual industries, this seems to be less of a concern (cf. Cowan and Raddatz 2011 and Chari et al. 2005). Also concerning causality between financial development and growth, industry level data offers some benefits. Firstly, the fact that financial development and growth could be driven by household savings in an economy makes causality ambiguous. Also it is conceivable that financial development is not an actual cause of growth but just leads financial markets to anticipate growth and therefore to lend more, similar to a self-fulfilling prophesy. Yet, if financial development turns out to affect economic growth via industry growth, this adds reliability to their causal link, and this can be shown by studying effects on the industry level (cf. Rajan and Zingales 1998, 559f.). Lastly and most
8 For literature on the effect of sudden stops on GDP growth refer for example to Edwards (2004, 2007), for the
16
importantly, this level of aggregation allows controlling for industry characteristics, for instance for the dependence on external finance which reflects the sensitivity to financial frictions, and at the same time for country characteristics like financial development. As shown in the literature review above, these characteristics are empirically important and therefore the analysis in this study is performed on industry level.
One example of the effects of sudden stops on this level of aggregation is the work by Cowan and Raddatz (2011) who study the effect of sudden stops on growth in industrial output. As main regressors, they use a dummy taking the value one if a country experiences a sudden stop, and interact this dummy with a measure of external dependence and a number of other industry characteristics. They perform a difference-in-differences analysis in order to identify how sudden stops affect growth in industries with different levels of external dependence. With this approach, they only compare growth during a sudden stop year to other years in general, without defining specific periods as done for example in Kroszner et al. (2007); in addition, with this method Cowan and Raddatz (2011) do not account for the fact that the period between two sudden stops in one country can most likely not be considered normal but in many cases are characterised by financial troubles; furthermore, potential growth losses after the sudden stop year are neglected as only growth during the stop year is considered. Defining sudden stop periods to last more than one year can solve this problem. In addition, Cowan and Raddatz (2011) divide their sample into two sub-samples, one containing countries with high financial development and one with countries with low financial development. Just distinguishing between either high or low financial development does not seem to be a very sophisticated approach; instead, using values of financial development for each country allows for a more detailed analysis and does not require a separation of the data into sub-samples, thereby reducing the number of available observations in each sample.
The study by Kroszner et al. (2007) is similar in spirit to that of Cowan and Raddatz (2011) but avoids the mentioned drawbacks by using a different methodological approach.9 They run OLS regressions for three sub-periods – one before the crisis, one during crisis, and one after it – and compare the different growth rates of those periods. In order to avoid endogeneity, they use some of the main variables, share in value added of an industry and financial development, at their pre-crisis levels (cf. Kroszner et al. 2007, 193). Although the cross-country OLS estimation approach, as opposed to panel data techniques, does not fully exploit the time-series dimension of the data, it is still useful to capture growth during
9 The crises considered here are banking crises instead of sudden stops, which lead to different results, but the
17
different periods and also permits to define stop periods which are longer than just the sudden stop year. Moreover, by including country and industry dummies, it is equally appropriate to account for country- and industry-specific effects, which are of particular importance when analysing data on the industry level. Lastly, another advantage of this study over the method of Cowan and Raddatz (2011) is the treatment of financial development; rather than just distinguishing between high and low development, they apply a measure with values for each country and interact this measure with external dependence, which is facilitated by the fact that external dependence does not have to be interacted with any sudden stop dummy.
As a conclusion of the previous discussion, this paper contributes to the existing literature by analysing the consequences of sudden stops in gross capital flows on the industry level, accounting for the influences of both industry and country characteristics, relying on the methodology of Kroszner et al. (2007). So far, only a small number of empirical studies have been dedicated to the analysis of gross stops; while the determinants of sudden stop probability have already been studied in this way by Forbes and Warnock (2011), the effects of gross sudden stops on the real economy have not been exploited yet. To the best of my knowledge, doing this with industry-level data is a combination that has not been examined before.
The basic model, following Kroszner et al. (2007, 192) is an ordinary least square (OLS) cross-country estimation with country and industry fixed effects which takes the form
yij = β1 SHAREij + β2 FDi x EDj + αi + µj + εij (3)
where yij is the difference in real value added growth of industry j in country i between the pre-stop period and the stop period, that is yij = gij,stop - gij,pre-stop , where gij,pre-stop and gij,stop are real value added growth in sector j in country i during the pre-stop and stop period, respectively. SHAREij is the share of value added of industry j in total value added of the manufacturing sector in country i, FDi is the financial development of country i, EDj is the external dependence of industry j, αi is a country dummy and µj an industry dummy. This model is used to analyse the influence which the interaction of financial development and external dependence exerts on the difference between growth in the pre-stop period and growth in the stop period (for period and variable definitions, as well as data sources, see chapter 4).
18
years indicated in Cavallo and Frankel (2008) or derived from the work of Forbes and Warnock (2011). The pre-stop period, i.e. the benchmark period to which the stop period is compared, is [t – 4, t – 2] in the baseline specification, also including three years and being separated from the stop period by one year, to clearly distinguish these two periods.
4 Data
This chapter defines the variables used in the above models and gives information on the data sources. In general, all variables cover the period from 1980 – 2002 for every country, unless stated otherwise. However there are a number of exceptions where value added data is available only after 1980 or ends before 2002. Although the initial data is a panel dataset covering a time period of 22 years, the final dataset contains only cross-sectional observations as averages of time periods or at specific points in time.
19
All countries and their net and gross stop events considered in the two main samples are found in table 1.10 The numbers in brackets indicate stop events that include observations from countries which, due to data restriction, are not included in the regression but which coincide with stops according to the alternative definition. Both samples combined comprise 29 countries with 34 banking crises, the net stop sample 18 countries with 22 net stop events, and lastly, the gross stop sample includes 18 countries with a total of 29 gross stop events.
Table 1
Years of banking crises, sudden stops in net and in gross capital inflows.
Algeria 1990 1990 Argentina 1989, 1999 1989, 1995, 2001 Bolivia 1999 1986, 1994 Denmark 1987, 1989 1987 Egypt 1990 1983, 1991 Finland 1991 1986, 1991 1991 India 1990 1993 Indonesia 1997 1997 1994, 1997 Ireland 1991 Israel 1988, 1998 (1988, 1998) 1980 Japan 1991, 1998 1992 Jordan 1992, 1993 1989
Korea (Republic of) 1997 1997 1997
Malaysia 1997 1985, 1997 Malta 2000 Mexico (1995) 1995 1981, 1994 Morocco 1995 1980 New Zealand 1988 1988 1987 Norway 1988, 1991 1990 Portugal 1992 1984, 1992 Slovak Republic 1999 1993 South Africa 1985, 1990, 1998 1989 Spain 1992 1980 Sri Lanka 1998 1989 Sweden 1991 1984, 1991, 1997 1991 Syrian Arab Republic 1989
Trinidad and Tobago 1984 1982
Turkey 1991, 1994, 1998 1991, 1994 1983, 1994, 2000
Venezuela 1994 1994
10 Note that not all sudden stop events experienced by the respective countries are listed in table 1, but only those
included in the regression, i.e. events for which growth during pre-stop and stop periods could be calculated.
20
In 11 cases, an event is observed according to both the net and gross stop approach, resulting in a correlation of = 0.43. This strength of linear association seems surprisingly low, but it supports the conjecture that the calculation of sudden stops based on net capital flows yields different results than that based on gross flows. For the non-crisis stop sub-samples excluding stops that coincide with banking crises, the net stop sample consists of 7 countries with 10 stops, and the gross stop sample of 19 stops in 15 countries.
To examine the effect on industry-level growth, value-added data on the 3-digit International Standard Industry Classification (ISIC) industry level is taken from the United Nations Industrial Development Organization (UNIDO) database INDSTAT3 (2006 ISIC Rev.2) and is deflated by the consumer price index (taken from the World Bank World Development Indicators (WDI)), yielding real value added. Then growth for pre-stop and stop periods is calculated as the change in the logarithm of real value added of an industry from the first to the last year of the respective period. Each industry just has one pre-stop growth observation, namely the period before the first stop event, even if the country has experienced multiple stops. This is based on the assumption that years between two sudden stops may not be treated as ‘normal’ years but are more likely to be characterised by financial turbulences which might affect industry growth. Also concerning the stop period, each industry is assigned just one growth observation. In the case of multiple stops in one country, the average growth over all stop periods is calculated (cf. Kroszner et al. 2007, 194). The sectoral observations on growth in real value added over the two periods are winsorised at 1 per cent, meaning that outliers are dealt with by replacing one per cent of the smallest and one per cent of the largest observations by their adjacent, less extreme value, rather than just dropping them. Lastly, the crucial variable is the difference between growth of the pre-stop and the stop period for each sector in each country.
21
Table 2
Mean growth difference over all industries
Net stop -0.163
Gross stop 0.018
Gross stop
(excluding Argentina)
-0.066
Closer inspection of the data reveals that this result is strongly influenced by growth observations of Argentina, which is included only in the gross stop sample; while average growth during the stop periods in Argentina is of regular size (-0.094), pre-stop growth amounts to an average of -1.517, which is clearly more extreme than the sample average and therefore distorts the overall average growth difference. This can be explained by the high inflation as well as economic stagnation, which prevailed in Argentina during part of the 1970s and 80s, whereas part of the 1990s was characterised by economic reforms and a slight recovery of the economy. Therefore, if Argentina is excluded from the sample, the growth difference indeed has the expected negative sign, meaning that the growth rate before the stop period is larger than that during the stop. The example of Argentina suggests that heterogeneity across countries does play a role in the data, which justifies the inclusion of country (and industry) dummies in the regression to control for this heterogeneity. Due to this, excluding Argentina from the sample is not required because the country dummies control for country-specific fixed effects. The following empirical analysis will show whether these general descriptive patterns are statistically relevant.
The main variable of interest is the interaction between an industry's dependence on external finance (external dependence) and the country’s financial development. External dependence is taken from Kroszner et al. (2007, Table 12), who base their calculations on the work of Rajan and Zingales (1998). Relying on U.S. firm-level data from 1980 – 1999, the dependence on external finance is the share of capital expenditures which is not financed with cash flows generated through the industries’ operations (cf. Kroszner et al. 2007, 200). These measures for U.S. firms in the 1980s are applied to all industries in all countries during the entire sample period, as U.S. data is considered a valid benchmark. As an alternative to external dependence in the 1980s, Kroszner et al. (2007) calculate the same measure for the
22
period of 1980 – 1999. This measure may be drawn on to test the robustness of results. Next, as described above, growth opportunities can be an alternative to external dependence. This measure is taken from Fisman and Love (2007), who also base their calculations on the data from Rajan and Zingales (1998): Analogously to the calculation of external dependence, an industry’s growth opportunities are measured by real sales growth of U.S. industries in the 1980s (cf. Fisman and Love 2007, 474). Apart from these industry characteristics, financial development is of interest: As in Kroszner et al. (2007), it is calculated using data on private sector credit in lines 22d and 42d from the IMF International Financial Statistics. This comprises the claims of depository corporations on the private sector, thus excluding claims to the public sector or other financial corporations, as well as claims of the central bank. Private sector credit then is divided by GDP in local currency (obtained from the World Bank WDI database). In order to circumvent endogeneity issues, financial development is calculated at its initial vale for each country, i.e. the value of 1980 or the first year in the sample, as this value is not likely to be affected by future value added growth.
Value added data from the UNIDO database is used to calculate each industry’s share in total value added of the entire manufacturing sector in the respective country. With this variable it is possible to account for growth convergence effects of different industries. Moreover, it is calculated at its initial value (the value in 1980 or the first year of the sample) in order to avoid its endogeneity. Lastly, OECD member countries are classified as developed countries, while all others are seen as emerging or developing countries, henceforth referred to as developing economies. According to this definition, there are 8 developed countries and 10 developing countries in the net stop sample, for the gross stop sample it is 12 developed and 6 developing economies.
5 Results
23
homoskedastic data in all three cases can be rejected on the 5 per cent significance level for net stops, and on the 1 per cent level for gross stops, and therefore, White’s (1980) heteroskedasticity-consistent standard errors are used in all regressions. Secondly, multicollinearity can be an issue if the data do not provide sufficient variation and can result in large variances and poorly estimated coefficients. However, the correlations of the estimated coefficients are found to be low throughout all periods and both samples: In only very few cases, a correlation of = 0.77 is reached, while correlations below = 0.1 are a common result in the data. As a consequence, the data contain variation enough to allow for good estimation results. In the following, all estimations include country and industry dummies and are estimated with intercepts, all of which are not reported here.
5.1 The Difference of Industry Growth between Pre-Stop and Stop Periods
This section tests the importance of the interaction term of financial development and external dependence with respect to the difference of industry growth before and during the stop, and in turn compares the corresponding results for net and gross stops. Table 3 presents the results of the baseline OLS regression of the model in equation (3).
Table 3
Comparing net and gross stops
Dependent variable: growth differences between pre-stop and stop periods
(1) (2)
Developing (3)
countries only (4) Net stop Gross stop Net stop Gross stop Share in value added 0.244 0.682 -0.408 1.055
(0.865) (0.779) (1.069) (1.726) FD*ED 0.0267 0.0410 1.431 -1.700 (0.0186) (0.0357) (2.243) (3.197) Observations 370 318 160 90 R-squared 0.680 0.603 0.682 0.562 F-value 34.93 21.42 11.92 .
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
24
gross stop sample respectively, indicating that 68 or 60 per cent of the variation in the growth difference between pre-stop and stop periods are explained by variation in the explanatory variables. Comparison with the measures of other specifications will shed more light on the relative goodness-of-fit of each model specification. Turning to the coefficients, the coefficient on share in value added is positive but not significant, both for the net and the gross stop sample. In theory, a positive sign on this coefficient means that industries diverge rather than converge; that is, for industries which are larger in terms of value added, growth on average is affected relatively less during a stop period than for smaller ones. However, due to insignificance of this coefficient, growth divergence can neither be ruled out nor confirmed.11 The focus of attention is on the coefficient on the interaction term of financial development and external dependence, which has a positive sign but is not significant at any conventional level.12 Nevertheless, a positive sign in theory means that industries with high external dependence in countries with well-developed financial systems are on average better off than if development and dependence are low. Note that this is opposite to the results of Kroszner et al. (2007) for banking crises and could be said to lend some – although insignificant – support to Hypothesis 2). This may be illustrated with a comparison of an industry with low external dependence in a country with low financial development on one hand, and an industry in a country where those variables are high on the other (ignoring that the coefficients are insignificant). For example, the predicted reduction in growth between pre-stop and stop growth for the industry at the 33rd percentile of external dependence in the country at the 33rd percentile of financial development (i.e. comparably low levels) is 0.43 percentage points larger than that for the industry at the 66th percentile of external dependence in the country at the 66th percentile of financial development (i.e. comparably high levels) in the net stop sample.13 For the gross stop sample, this value even amounts to a difference of 0.54 percentage points between countries and industries at low and high percentiles. Hence, the tendency of a smaller growth contraction caused by a stop in industries with high dependence which are located in countries with high development, is what would be expected
11 The same applies for this coefficient in most of the following regressions, but since it never turns out to be
significant, it is not mentioned any further.
12 Experimenting with different combinations of time windows of three, four or five years for both pre-stop and
stop period does not yield any more significant results, except when the pre-period is set to five and the stop period to for years. But this result seems rather random and therefore is not dealt with any further. Moreover, the exclusion of Argentina, which has turned out to strongly influence the average growth difference, does not increase the significance of the coefficients, as country-specific effects are controlled for by country dummies.
25
if the coefficients on the interaction terms in regression (1) and (2) were significant. Yet, as this is not the case, these figures are not a valid interpretation of the results.
Kroszner et al. (2007) suggest an alternative measure of external dependence, encompassing the period of 1980 – 1999 rather than the 1980s only. If this measure is applied here, the signs do not change, but the significance of the coefficients on the interaction terms increase considerably, with a p-value of 0.12 for the net stop sample, and for the gross stop sample the coefficient even turns significant at the five per cent level (results are not reported). The results of Kroszner et al. (2007) are not sensitive to using the different measures and therefore they rely on the measure for the 1980s in their baseline regressions. As there is no clear reason why the longer-term measure should be able to better explain industry growth, and for the sake of consistency, also here the baseline measure remains external dependence in the 1980s, but the alternative variable will further be used to test robustness of the results.
As an additional test, the baseline regressions (1) and (2) are run on the sub-samples of developing countries for both types of stops, for it is conceivable that the link between financial development and external dependence on one hand, and the effect of sudden stops on industry growth on the other, depends on the level of economic development. As mentioned above, many emerging economies were affected by sudden stops after opening up their financial systems to international capital flows. However, institutional quality tends to be lower in these countries and financial markets face larger difficulties in coping with the sudden loss of access to capital. If institutional quality is low in the first place, high private credit to GDP can be particularly damaging to industries with high external dependence when a sudden stop occurs, as poor financial institutions are less likely to be able to compensate the loss of access to foreign capital. Hence, industries highly dependent on external finance in countries with high financial development are assumed to fare worse.
26
in gross capital inflows), net stops by definition can also be due to domestic agents increasing their investment activity abroad (rise in gross capital outflows). Concerning gross capital inflows (which are crucial for gross stops), a sudden stop means the abrupt loss of access to foreign finance and, as explained above, high levels of external dependence in combination with high financial development are painful in developing countries, where institutional quality is low. For net stops, also gross capital outflows need to be considered, which imply that domestic sources of finance are transferred abroad and therefore lost for the domestic economy. And although this could be partly compensated by gross capital inflows, which do not necessarily have to fall but could even increase during a net sudden stop, the increase in gross outflows has to exceed the cut in gross inflows, in order for a net stop to occur. Hence, the consequence of a net sudden stop for developing economies is similar to the case of gross stops, namely that part of the sources of finance, domestic or foreign or even both, are lost and that the poor quality of financial institutions makes it harder for externally dependent sectors to access capital, especially when the rate of private credit to GDP is high. The argumentation for both gross and net stops is similar, and therefore it is concluded that the different signs on the (insignificant) interaction coefficients must be due to irregularities in the data, especially since the sub-samples of developing countries are comparably small and contain only few observations.
5.2 Alternative Industry Characteristics: Growth Opportunities
Although the results above have shown that the interaction of external dependence and financial development does not exert a statistically significant influence on the difference in industry growth from pre-stop to stop periods, this does not mean that the same has to be true for all measures of industry characteristics. As an alternative, the measure of growth opportunities of industries is included into the regression, as suggested by Fisman and Love (2007) in their analysis of industry growth. Two similar specifications are considered, which contain the interaction of financial development and a measure of growth opportunities either
instead of (equation (4)) or in addition to (equation (5)) the interaction of external
dependence. Analogously to equation (3), those equations look as follows.
27
The first two columns in table 4 show the results of the model specified in equation (4), where growth opportunities are featured instead of external dependence.
Table 4
External dependence versus growth opportunities
Dependent variable: growth differences between pre-stop and stop periods
(1) (2) (3) (4)
Net stop Gross stop Net stop Gross stop Share in value added 0.246 0.467 0.251 0.480
(0.863) (0.573) (0.868) (0.578) FD*growth opportunities -0.115 -0.114 -0.421 -0.286 (0.213) (0.190) (0.283) (0.249) FD*ED 0.0423** 0.0239 (0.0201) (0.0179) Observations 370 401 370 401 R-squared 0.678 0.665 0.681 0.666 F-value 35.18 33.83 34.34 33.36
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
First of all, the fraction of variation in the dependent variable which can be explained by the regressors has not changed considerable relative to the first two columns in table 3; for the net stop sample, R-squared it fell slightly, for gross stops it has increased, but in both cases ranges above 0.6. In that respect, the model with growth opportunities does not seem to be superior to that with external dependence. Furthermore, all coefficients in columns (1) and (2) again are insignificant at all conventional levels, which also does not suggest any improvement over the first model. However, the magnitude of the coefficients on the interaction term is considerably larger in the case of growth opportunities, where in turn they are quite similar for net and gross stops (however, these relations are not significant and can at the most give a very rough impression of the true relationships). Besides the magnitude of the coefficients, their negative signs are what sets them apart from the results in table 3. Analogously to
Hypothesis 2) in section 2.4, the interaction term is expected to have a positive effect on
28
This result, even if insignificant, is somewhat surprising based on the findings of Fisman and Love (2007). According to them, in a highly developed financial system, resources should efficiently be channeled towards industries with the best growth opportunities, and thus increase their growth. If a sudden stop occurs, this link should not be disturbed, since stops do not affect the intermediary activities of financial institutions in the same way as banking crises. Then, how can the negative signs on the interaction terms theoretically be explained? Only because an industry has good growth opportunities, it does not necessarily follow that it is in need of or makes use of external finance to a larger extent. That is, there is no obvious reason why the measures of external dependence and growth opportunities should be correlated, and the data confirms this assumption: The correlation between the two measures is very low with a value of = 0.08 for the net stop sample, and = 0.11 for the gross stop sample. Therefore, an industry with good growth opportunities fares better during a sudden stop if the ratio of private credit to GDP is low; in that case, a drop in net capital flows does not affect the industry too strongly, as it may not have received much credit in the first place. Remember that this is not necessarily a bad thing, because industries with good growth opportunities are not automatically more dependent on external finance, but may as well finance their investments internally. On the other hand, if private credit to GDP is high, it is more probable that a greater fraction of credit is allocated to those industries with good opportunities, which are then hit harder if foreign funds dry up.
29
the net stop sample (with the share in value added and the level of growth opportunities at its sample mean), this predicted reduction being slightly more pronounced than without controlling for the growth opportunities interaction. The results from column (3) lend support to the assumption that the effects of both interaction terms on the growth difference in fact work into opposite directions, thus confirming Hypothesis 2) for external dependence, but contradicting it for growth opportunities. To some extent, this casts doubt on the conclusion of Fisman and Love (2007) that growth opportunities are a broader concept of finance, encompassing external dependence, at least when studying the effects of sudden stops on industry growth. Although the coefficients on the growth opportunities interaction are greater in magnitude, the opposite signs and the coefficient on the external dependence interaction turning significant at most indicate that the two concepts of finance might complement each other. However, the lack of significance for most coefficients makes it difficult to analyse these relationships in detail, and in particular does not allow any conclusion as to why the coefficient is significant for the net stop sample, but not so for the gross stop sample.
Testing the sensitivity of the results towards the alternative measure of external dependence for the period of 1980 – 1999, the estimation yields coefficients with the same signs and similar in size, and even the interaction term of financial development and external dependence for the net stop sample remains significant at the ten per cent level when controlling for growth opportunities interacted with financial development (results are not reported). Finally, running the same regressions on the developing country sub-samples does not yield any significant coefficients but it stands out that the coefficient on the interaction with growth opportunities becomes much larger in magnitude (but still negative and insignificant. Results are not reported). The explanation here is analogous to section 5.1, where the lower institutional quality in developing countries is the reason why a high ratio of private credit to GDP can have damaging consequences during a sudden stop.
5.3 Sudden stops and banking crises
