Do Sustained International Gross Capital Inflows Influence The Probability Of Banking Crises?

(1)

1

Do Sustained International Gross Capital Inflows Influence The Probability

Of Banking Crises?

University of Groningen Faculty of Economics and Business

Master Thesis International Economics and Business

Name: Joseph Tichband

Student ID number: S3453898

Student email: j.tichband@student.rug.nl

Date Paper: 3/7/18

(2)

2

Over the past four decades increased global financial liberalization has nurtured an environment of increasingly turbulent financial conditions. The incidences of financial crises have become far more frequent when compared to the decades preceding the 1970’s. Since that time four distinct periods of crises stand out amongst the rest: firstly there were multiple crises in developing nations during the beginning of the 1980s1. This was followed by Finland, Sweden and Japan in the early 1990s. Next came the Asian Financial Crisis initiating in 1997 involving Thailand, Indonesia amongst others. Finally the infamous global financial crisis originating in the U.S sub-prime mortgage market during 2007, impacting many developed countries including the United Kingdom and the Netherlands. This study joins a growing literature within the field investigating the occurrence of banking crises and their potential determinants2.

Time and time again the same conventional reasons consisting of stupidity, greed and reckless lending on the side of banks and other lenders are used as the explanation for banking crises3; resulting in a proverbial witch hunt of perceived villains such as Lehman’s Dick Fuld. Some even see the occurrence of major crises, particularly 2008, as an act of God, yet it is posited had Mother Teresa been appointed Lehman’s CEO in 2007 even she could not have prevented what was coming. When one looks deeper at the key characteristics displayed in the majority of banking crises occurring since the 1970’s, similarities are striking (Reinhart and Rogoff, 2008). Every country witnessed an increase in its capital account surplus, or in the case of Japan a decrease in its capital account deficit, during the years leading up to the crisis. Figures 1 and 2 display the build up of the net capital account surplus seen in Iceland and Colombia before the occurrence of a banking crisis in 2008 and 1982 respectively4. Once seen as a bastion of economic growth, windfalls of capital are receving increasing skepticism from researchers and policymakers alike (IMF, 2011a).

1_{Distinct examples include those crises witnessed in Mexico and Brazil}

2_{This paper does not seek to determine exact causes of banking crises however.}

3_{This is the opinion of former president of the Federal reserve bank of New York Timothy Geithner (Thomson}

Reuters, 2012).

(4)

4 Figure 1: Development of the Net Capital Account/GDP for Iceland 1975 - 2012

Figure 2: Development of the Net Capital Account/GDP for Columbia 1975 - 2012

The aim of the paper is to investigate the influence sustained high gross capital inflows has upon financial instability through their impact on the probability of a banking crisis occurring in the following year. Therefore this paper seeks to answer one overarching question with two sub questions, as follows:

(i) Whether sustained high gross capital inflows increases the likelihood of a systemic banking crisis

(5)

5

(b) If an asset boom coinciding with the high gross capital inflows alters the impact sustained high gross capital inflows has upon the likelihood of a banking crisis

The study contributes to the literature through concentrating upon the relative5_{long term} buildup of high gross capital inflows over a sustained period of time as opposed to the impact of a one period bonanza, defined as one standard deviation from trend (Caballero, 2012). Along with providing initial investigation into the leverage mechanism.

This paper constructs an unbalanced panel data set for 153 countries over the period 1973-2012 upon gross capital inflows along with other potential determinants of banking crises in order to carry out a multivariate regression analysis upon a binary outcome model. The results from the study indicate that the presence of sustained high gross capital inflows increases the likelihood of a systemic banking crisis starting in the next year. This impact is independent to that found previously upon bonanzas of capital inflows. The implication of this result is that sustained high gross capital inflows leads to the buildup of instability risk in the domestic banking system. This impact is present even in the absence of high leverage in the domestic banking system. Furthermore it is found that the increase in likelihood of a banking crisis starting the following year is further exacerbated when these sustained high gross capital inflows coincide with a domestic asset boom.

The remainder of the paper is organized as follows: section 2 discusses relevant literature upon the subject. Section 3 provides information upon how the variables of interest have been defined, along with detail upon the data set as a whole and the control variables included. Section 4 discusses the choice of methodology used throughout the study, with the results from the analysis are provided in section 5. Following that in section 6 conclusions are drawn upon the results along with suggestions for further research.

5_{Relative is key here as even though the high gross inflows are defined as sustained this does not imply they}

(6)

6

2. Literature Review

2.1 Global Imbalances

The process of financial liberalization witnessed over the past few decades has allowed the increasingly free movement of capital between developed and emerging economies alike (Turner, 2008)6. During this time period, there has been an increasing presence of large ‘global imbalances’; whereby countries have been able to sustain large current account deficits7. These imbalances had been viewed by many as relatively harmless (Backus et al., 2009)8, not everyone was of this opinion however9. At the root of these imbalances is the large increases in the capital account witnessed by those countries running current account deficits10. It had been long argued that increasing capital inflows directed toward a country, particular a developing nation, aids in the process of growth; for example through the provision of funds necessary for increased domestic investment and consumption, inducing higher growth rates (Calvo et al., 1996)11.

Windfalls in capital, however have been greeted with far more caution in recipient countries, with their increased association to both macroeconomic and financial risks (Caballero, 2012). These potential risks are of particular importance during the current global financial climate of low-interest rates and output growth in advanced economies inducing a greater potential for capital flows to be directed towards emerging economies. The crucial risk believed by many to be associated with windfalls of capital is that of their impact upon the onset of a systemic banking crisis. Aliber (2018) argues that the DNA all banking crises since the 1980’s “was embedded in the surge in cross-border investment inflows”.

It is of no wonder why policy makers are concerned with the potential onset of a banking crisis. Due to a variety of reasons12 economic downturns that are preceded by the occurrence of a

6_{For a discussion upon the push and pull factors influencing the flow of capital see Calvo, Leiderman &}

Reinhart (1996), along with Ghosh et al. (2014) who discuss push and pull factors with regards to large “surges” in net capital flows to emerging market economies.

7_{The U.S recorded its largest current account deficit position to date at the end of 2007 (Mendoza et al., 2009).} 8_{See also Kamin, Leduc & Croke (2005), Hausmann & Sturzenegger (2006), Engel & Rogers (2006). Others}

believed that the rebalancing of these deficits would be smooth (Cavallo and Tille, 2006).

9_{See for example Roach (2004) and Summers (2004)}

10_{An autonomous increase in the capital account surplus of a country must be met by a counterpart increase in}

its current account deficit. This is the transfer problem identified by (Keynes, 1920)

11_{For a further discussion on the potential benefits of capital inflows see Bailliu (2000).}

12_{A discussion upon the reasons why banking crises reduce output far more than a regular downturn in the}

(7)

7

banking crisis result in larger GDP losses than downturns which are not (Reinhart & Rogoff, 2008; IMF, 2008)13_{, and these losses in output last longer (Reinhart & Rogoff, 2009). Banking} crises are also associated with a marked and lasting increase in government debt (Furceri & Zdzienicka, 2012) and rising unemployment levels that are maintained (Reinhart & Rogoff, 2009). Therefore identifying potential determinants that induce an increased likelihood of a banking crisis is crucial to aid policy makers in their decision when it is appropriate to carry out policy responses to reduce the chance of this bank crisis beginning or to reduce the consequences of one14.

2.2 The Association Of Capital Inflows and Banking Crises

Surprisingly, the statistical association between high capital inflows and the onset of banking crises, has not been found to be consistently robust. Earlier studies by Fernandez-Arias & Hausmann (2001) and Eichengreen & Arteta (2002) do not find there to be an association between the two. More recent studies, such as that of Furceri et al., (2011) however, find that a capital inflow “bonanza” episode does increases the conditional probability of a banking crisis originating during the next year. Caballero, 2012 through utilizing a multivariate regression analysis finds that a net capital inflow bonanza increases the probability of a banking crisis in the following year by a magnitude of three.

The majority of the current literature upon the association focuses upon a defined “bonanza”15 episode in capital inflows occurring and the impact this has upon the probability of a banking crisis occurring the following year16_{. An issue with this methodology is that it ignores the} impact of these high capital inflows being sustained; persistent exposure to foreign liabilities has been identified as a key reason for financial turmoil (Bordo, Cavallo and Meissner, 2010). Furthermore this method also ignores the sustained buildup of risk that would accompany high capital inflows yet do not necessarily cross this bonanza threshold.

13_{For example Haugh, Ollivaud & Turner, 2009 find that output losses were 3 times greater following the “big}

five” (Reinhart and Rogoff, 2008) banking crises than during a regular downturn period

14_{This paper does not discuss potential policy responses to reduce the impact of banking crises, however for}

analysis upon policy responses see Rosas(2006) and for a discussion upon what went wrong with regards to the policy responses during the 2008 global financial crisis consult Taylor (2009).

15_{Defined as one standard deviation above trend}

16_{See for example Caballero, (2012) and Furceri et al., (2011). These classifications of a bonanza; following on}

(8)

8

A common feature since the 1970’s has been that capital inflow episodes begin numerous years before the onset of a systemic banking crisis and are usually maintained at a comparatively high level for multiple consecutive years17_{. Some posited that the likelihood of an Icelandic banking} crisis was 99.44 percent predictable by no later than the end of 2005 (Aliber, 2018), implying that it was the occurrence of capital inflows many years before 2007 that induced this increased probability and not necessarily the bonanza in capital inflows during 2007 itself. Figure 3 displays the development of gross capital inflows for Iceland over the period 1975-2012, with the red line identifying 2005; this highlights the extended period in which capital inflows were sustained at a comparatively high level for Iceland preceding the crisis of 2008.

Figure 3: Development of Gross Capital Inflows/GDP for Iceland 1975 - 2012

The majority of studies within the literature upon the association of capital inflows and banking crises do so for net capital flows. This paper however draws upon recent findings which support the view that gross capital inflows are a better sign of financial fragility than net inflows (Gourinchas, Truempler and Rey, 2012; Obstfeld, 2012a) particularly during recent decades whereby gross capital flows dwarf that of net flows (Obstfeld 2012a, 2012b). Countries running a balanced current account, which when concentrating on net inflows would not posit an imbalance issue, are still susceptible to increased domestic financial instability risks stemming from gross inflows. It is these gross flows that capture a countries role with regards to international borrowing, lending and intermediation (Borio and Disyatat, 2011); thus reveal

17_{This finding is also backed up by the work of Calderon & Kubota (2012) with regards to lending booms. In}

(9)

9

how net flows are funded. This is not to say the monitoring of imbalances in the current account is a fruitless exercise, as they do warrant caution. However through focusing upon gross capital inflows, it is hoped this will better capture the impact sustained international capital inflows has upon domestic financial stability (Borio and Disyatat, 2011).

2.3 Capital Inflows and Leverage

The commonly researched mechanism through which capital inflows induce an increased probability of a banking crisis is that of a lending boom18. However a boom in credit does not necessarily represent a failure within the banking system inducing instability as it may reflect a healthy market response to expected future productivity gains (IMF, 2011b)19. This paper instead contributes to the literature through investigating the channel of increased leverage within the domestic banking system enabled by the sustained high gross capital inflows.

The correlation between bank deposits in the domestic financial system and private credit will break down as a countries banks increasingly resort to utilizing wholesale cross border funding (Hoggarth et al., 2010) allowing for higher leverage. Sustained gross capital inflows induces an increase in the ‘elasticity’ of the domestic banking system; defined by Borio & Disyatat (2011) as “the degree to which the monetary and financial regimes constrain the credit creation process”, thus an increase in the ‘elasticity’ reflects a reduction in constraints.

Furthermore, capital inflows are associated with a real exchange rate appreciation (Aliber, 2018). This results in a strengthening of domestic borrowers balance sheets as the funds lent to them are in domestic currency, yet the capital inflows behind this lending are denominated in foreign currency20. Measured risk upon the domestic banks’ balance sheets who have loaned money to these borrowers will reduce, incentivizing further lending by these banks; raising the leverage ratio higher thus creating a self-sustaining cycle due to its pro-cyclical nature (Bruno & Shin, 2015). High leverage has been found previously to be a robust and significant predictor of financial crises (Gourinchas & Obstfeld, 2012).

Overtime the system becomes increasingly reliant upon these cross border inflows as in order to increase the size of their balance sheets. It is highly likely that these domestic borrowers can

18_{Caballero, 2012 reports that capital inflows do not impact the likelihood of a banking crisis necessarily}

through the over lending channel.

19_{In addition to this the results upon the association between capital inflows and lending booms are mixed; ,}

Sachs, Tornell & Valasco (1996) find no association but Mendoza & Terrones (2008) do.

20_{Pre-dominantly this will be in dollars. This can be dangerous as Domestic Liability Dollarization has been}

(10)

10

be characterized as having a “speculative” financing profile; requiring the issuance of new liabilities in order to service their debt principle (Minsky, 1992). When these capital inflows subside however, the real exchange rate shall depreciate, given that a currency mismatch21_will have developed, the cost of debt servicing rises leading to increased defaults, of which the domestic banking system is heavily exposed to, inducing a banking crisis.

2.4 Capital Inflows and Asset Values:

As has been commonly determined in the literature, capital inflows lead to the appreciation in value of domestic assets (Tillman, 2013; Reinhart & Reinhart, 2008)22, particularly real estate (Aizenman & Jinjarak, 2009). Calvo (2011) proposes a model in which an assets price is determined by the combination of the intrinsic value and liquidity value. Capital inflows increases the liquidity value of the asset as they become more salable23. These rise in asset prices do not require a lending boom in order to occur, however a positive feedback loop is created whereby capital inflows increases the liquidity of the assets and expected value, inducing more gross capital flows directed towards these assets. Unsurprisingly Episodes of high capital inflows have been identified as a key contributing factor in the formation of asset bubbles (Calvo, 2011)24_{and these do not require the capital flows to be intermediated by} domestic banks. High capital inflows association with asset bubbles is unsurprising given that uncertainty surrounding credit expansion is a key determinant of bubbles (Allen & Gale, 2000), which high gross capital inflows results in due to their highly volatile nature. When these asset bubbles are bust by a decrease in capital inflows the banking system is likely to be impacted heavily though falling collateral values. Furthermore as asset values are rising, the sustained high gross capital inflows flowing into the country will be directed to the non-tradeable goods sectors in order to receive the higher returns25. These higher asset returns allow for increased credit creation by banks

21_{This will be more likely for those emerging economies whom suffer from “Original Sin” Eichengreen,}

Hausmann & Panizza (2005)

22_{The context of net capital flows, this rise in domestic values is required to allow the current account deficit to}

adjust and match the capital account surplus (Aliber, 2018).

23_{Referred to as market liquidity by Brunnermeier & Pedersen (2008)} 24_{Going against the traditional view that bubbles attract capital inflows}

25_{The reallocation of productive resources to the non-tradable sector from the tradable sector is documented}

(11)

11

3. Definition of Crises, High Gross Capital Inflows

and Data

3.1 Banking Crises

Similar to previous work undertaken on the determinants of banking crises, the definition of a banking crisis adopted in this paper is taken from Laeven and Valencia (2013), as this database has been found to be the most accurate upon systemic banking crises (Chaudron and De Haan, 2014). In the Laeven and Valencia (2013) database, a (systemic) banking crisis is defined when two conditions are met:

(i) Significant signs of financial distress in the banking system (as indicated by bank runs, losses in the banking system and/or bank liquidations); and

(ii) Significant banking policy intervention measures were undertaken in response to losses in the banking system.

According to Laeven and Valencia (2013) the starting year of this (systemic) banking crisis is identified when, in addition to both these conditions being met, at least three of the following policy interventions have been undertaken in the country to deal with the banking crisis:

a) deposit freezes and bank holidays; b) significant bank nationalizations;

c) bank restructuring gross costs (at least 3% of GDP);

d) extensive liquidity support (5% of deposit and liabilities to non-residents); e) significant guarantees put in place; or

f) significant asset purchases (at least 5% of GDP)

(12)

12

3.2 High Gross Capital Inflows

As has been eluded to unlike the majority of previous studies in the area this study utilizes gross capital inflows, as opposed to net. These gross capital inflows have been normalized by GDP26. In order to identify high levels of gross capital inflows the threshold methodology laid out in Mendoza and Terrones (2008) is utilized27. This method proposes a country-specific threshold; determined by the deviation from the long-run trend in gross capital inflows, represented here by 𝛺_𝑖,𝑡. In order to determine the long-run trend the Hodrick-Prescott filter is used with the smoothness parameter set to 100 as is common for annual data (Mendoza & Terrones, 2008). Gross capital inflows are characterized to be high when they are above this trend value, thus 𝛺𝑖,𝑡 > 0 . This identification method relies upon the assumption that gross capital inflows below this trend level are able to be absorbed by the domestic financial system without inducing a significant increase in instability that would induce an increased likelihood of a systemic banking crisis. Furthermore a non-negativity constraint is applied to all variables constructed involving gross capital inflows. Through utilizing the method to define periods in which gross capital inflows are above their corresponding trend value, this overcomes the weakness of relying upon an arbitrary threshold set for all countries. This methodology, is able to capture well known periods of heightened capital inflows, such as in many advanced economies during the early 2000’s (Benigno, Converse & Fornaro, 2015).

The classification of a sustained episode of high gross capital inflows is of course open to interpretation. Therefore multiple variables have been created with the baseline being a one time period of high gross capital inflows: 𝛺_𝑖,𝑡 > 0. This is then extended to two consecutive time periods in which the gross capital inflows are identified as high: 𝛺_𝑖,𝑡 > 0 , 𝛺_{𝑖,𝑡−1} > 0. The number of consecutive years where these inflows are high is then extended to three (𝛺_𝑖,𝑡 > 0 , 𝛺_{𝑖,𝑡−1} > 0, 𝛺_{𝑖,𝑡−2} > 0) four (𝛺_𝑖,𝑡 > 0 , 𝛺_{𝑖,𝑡−1} > 0, 𝛺_{𝑖,𝑡−2} > 0, 𝛺_{𝑖,𝑡−3}> 0) and five years (𝛺𝑖,𝑡 > 0 , 𝛺𝑖,𝑡−1 > 0, 𝛺𝑖,𝑡−2 > 0, 𝛺𝑖,𝑡−3 > 0, 𝛺𝑖,𝑡−4 > 0) respectively. Figure 4 displays the gross capital inflows against the trend values for Indonesia. As shown, the gross capital inflows

26_{Normalizing by GDP is appropriate as the changes in gross capital inflows are far more pronounced than that}

of GDP growth and therefore scenarios in which gross capital inflows would increase at a lower rate than GDP are unlikely.

27_{This is a common method used and adapted by many with regards to capital inflows e.g Forbes and Warnock}

(13)

13

remain at a high level from 1990 until 1997. This would represent a high gross capital inflow episode sustained for 8 consecutive years.

Figure 4: The relation of Gross Capital Inflows/GDP VS Trend for Indonesia 1980-2012

3.4 Leverage

The definition of domestic banking sector leverage follows on from that adopted by Lund-Jensen (2012) whereby:

𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 = 𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝐶𝑟𝑒𝑑𝑖𝑡 𝐵𝑦 𝐷𝑒𝑝𝑜𝑠𝑖𝑡 𝑀𝑜𝑛𝑒𝑦 𝐵𝑎𝑛𝑘𝑠 % 𝐺𝐷𝑃 𝐵𝑎𝑛𝑘 𝑑𝑒𝑝𝑜𝑠𝑖𝑡𝑠 𝑡𝑜 𝐺𝐷𝑃%

Here private credit by deposit money banks reflects the financial resources provided to the private sector by domestic money banks28_{. Bank deposits include the total value of demand,} time and savings deposits at domestic deposit money banks29_{. Therefore if bank deposits to} GDP remains at the same percentage, yet private credit extended by domestic banks as a percentage of GDP increases, this would rightly be reflected by a higher leverage ratio value.

As leverage has shown to be very pro-cyclical the Hodrick – Prescott filter30_{has been utilized} in order to de trend the leverage ratio data. The deviation from the long run trend 𝜑_𝑖,𝑡 is

28_{According to the definition provided by the World Bank with the data deposit money banks are comprised of}

commercial banks and other financial institutions which accept transferable deposits

(14)

14

calculated by subtracting the trend value from that of the actual leverage ratio at time 𝑡. The leverage ratio is characterized as high when 𝜑𝑖,𝑡 > 0. Therefore a two year sustained period in which the leverage ratio is high is expressed as 𝜑_𝑖,𝑡 > 0 , 𝜑_{𝑖,𝑡−1} > 0. Figure 5 represents the development of the leverage ratio for Iceland from 1975 till 2012. As can be seen from the graph there is a five year sustained period of high leverage ratios beginning in 2004 and ending in 2008.

The control variables present in all models utilized in this paper reflect the results of previous studies into the determinants of banking crises. This set of controls consists of domestic credit extension as this has been found to be associated with sudden stops (Eichengreen, Gupta & Mody, 2008). There is also a measure of inflation in order to provide an indication of macroeconomic mismanagement (Furceri et al., 2012). The natural log of a countries reserves, as reserves act as a buffer against banking crises and a measure of GDP growth as Furceri and Mourougane (2009) find that banking crises tend to follow excessive and sustained growth. Another explanatory variable included is the existence of an explicit deposit insurance scheme, as this has previously been found to increase the likelihood of a banking crisis (Demirgüç-Kunt & Detragiache, 2002). All of these explanatory variables are lagged one period in order to reduce the potential issue of reverse causality.

A further control variable is the existence of a contemporaneous currency crisis, which has been identified in the literature as having a strong association with banking crises (Glick & Hutchinson 2000; Kaminsky & Reinhart, 1999) and has had a significant impact upon the likelihood of a banking crisis in previous empirical studies (Caballero, 2012). In order to identify a currency crisis within the data set, the criteria adopted is that defined by Laeven and Valencia (2013).The onset year of a currency crisis is defined by fulfilling two criteria:

i) a nominal depreciation of the currency vis-a`-vis the U.S. dollar of at least 30 percent; and

ii) that the depreciation is also at least 10 percentage points higher than the rate of depreciation in the year before

On top of this, when a country which fulfills these criteria for continuous years, it is the first year of a five year window that is identified as the onset of the currency crisis. This means that a different currency crisis shall not be identified in the four years following a country year pairing identified as the onset of a currency crisis. A total of 223 currency crises according to this criteria are identified in the data set. The data upon exchange rates used is the end of period31 nominal exchange rate against the dollar, provided by the IMF.

The final control variable included is a dummy variable to indicate the presence of a net capital inflow bonanza in the previous year32. The method to identify a bonanza episode in net capital

31_{As this is an annual study, end of period here represents the end of year exchange rate.}

32_{Both Baseline and Intense surges, with the latter used for sensitivity analysis. Furthermore surges in gross}

(16)

16

inflows is similar to that of high gross capital inflows. 𝐷_𝑖,𝑡represents the deviation of net capital inflows from the long-run trend in country 𝑖 at time 𝑡33_{. The country-specific standard deviation} of this difference from trend is calculated;

𝜎(𝐷

_𝑖

)

. A baseline bonanza is one in which the condition of 𝐷𝑖,𝑡 > 𝜎(𝐷𝑖) is fulfilled and an intense bonanza is characterized by 𝐷𝑖,𝑡 > 2𝜎(𝐷𝑖). Following the baseline indicator in Reinhart & Reinhart (2008), the net capital account is calculated as the inverse of the current account normalized by GDP. The data for the current account is provided by the IMF’s International Financial Statistics.

3.5 Data Sources

The data for all these control variables comes from common sources upon macroeconomic data. The World Bank provides the requisite data for total reserves minus gold (FI.RES.XGLD.CD), GDP growth in annual percentage terms (NY.GDP.MKTP.KD.ZG), Inflation measured by GDP deflator in annual percentage terms (NY.GDP.DEFL.KD.ZG) as well as Domestic credit to private sector measured in terms of % of GDP (FS.AST.PRVT.GD.ZS). Furthermore the data upon GDP utilized in normalizing both gross and net capital inflows is also taken from the World Bank (NY.GDP.MKTP.CD). The leverage ratio data is calculated by the Author utilizing data upon bank deposits measured as a percentage of GDP (GFDD.OI.02) and private credit by deposit money banks as a percentage of GDP (GFDD.DI.01) both taken from the World Bank’s Global Financial development Database. Finally, the World Bank also provides the data upon total value stocks traded as a percentage of GDP (CM.MKT.TRAD.GD.ZS).

Data upon Gross Capital Inflows is calculated by the author from the IMF’s balance of payments database. This database also provides the data for the current account utilized when determining net capital inflows. Exchange rate data utilized to determine currency crises is measured as the end of period nominal exchange rate against the dollar, is taken from the IMF’s International Financial statistics database. Finally The data for the existence of an explicit deposit insurance scheme is taken from Demirgüç-Kunt, Kane and Laeven (2015) whom provide an extensive database covering which countries possess an explicit deposit insurance scheme and the year this was enacted.

33_{A Hodrick-Prescott filter with a smoothness parameter of 100 is applied to the data upon net capital inflows}

(17)

17

4. Empirical Strategy

4.1 Non-Parametric Analysis

The non-parametric analysis based on conditional and unconditional probabilities provides initial analysis upon the impact of the defined conditions upon the likelihood of facing a systemic banking crisis in the next year. These probabilities are reported along with the results of corresponding independence tests: likelihood-ratio, Fishers exact test and Pearson chi squared34. The null hypothesis for all the independence tests is such that there exists statistical independence between banking crises and sustained high gross capital inflows.

4.2 The multivariate models

Although Non-parametric analysis provides a simple and intuitive analysis of the association it has multiple limitations. It is unable to analyze the association between the two variables while controlling for other potential determinants upon the likelihood of a banking crisis starting. Furthermore it is unable to inform us about causality as simply shows correlation. Finally as is shown by the results of the various models, the non-parametric analysis distorts the true magnitude the varying conditions have upon the likelihood of a banking crisis the following year.

To overcome these issues, multivariate analysis is undertaken. The methodology in this study is based upon a regression analysis of a binary outcome model. In this model the dependent variable is a binary response outcome 𝑦_𝑖,𝑡 which takes a value of one if country 𝑖 experiences the onset of a systemic banking crisis in time period 𝑡, or takes a value equal to zero if the country does not. The likelihood of a crisis starting in the year t is therefore measured by an underlying continuous latent variable 𝑦_𝑖,𝑡∗ _{. The likelihood of a crisis beginning is approximated} by y_{i t}_, =1 [y >0], t = 1,…,T. This likelihood is hypothesized to be a function of a vector of _{i t}*_, control variables  and the variables of interest; a dummy variable for sustained high gross _{i t}_, capital inflows Ks*y 35 _{, the interaction of this variable with high leverage and sustained high} gross capital inflows interacting with a domestic asset boom. A linear regression model is assumed for the latent responsey . Due to the suspected presence of country fixed effects, *_{i t}_, further confirmed through testing, the model is estimated controlling for these. The benefit of

34_{These independence tests are utilized for all non-parametric analysis undertaken in this study.}

35_{This dummy variable changes per the defined length an episode of high gross capital inflows is sustained –}

(18)

18

this fixed effect estimator is that it provides consistent estimates that control for the presence of both country-specific and time-invariant characteristics that may impact the likelihood of experiencing a banking crisis or sustained high gross capital inflows.

4.2.1 Model 1:

The fixed effect estimator for the first model which utilized to answer the overarching research question is represented by:

𝑦

_𝑖,𝑡∗

= 𝛼 + 𝛿𝐾𝑠

_𝑖,𝑡∗𝑦

+ 𝛽𝑋

_𝑖,𝑡

+ 𝜐

_𝑖

+ 𝜀

_𝑖,𝑡 (1)

Where

𝐾𝑠

_𝑖,𝑡∗𝑦 can take the following definitions:

𝐾𝑠

_𝑖,𝑡1𝑦 =

{1 𝑖𝑓 𝛺

In equation (1) 𝜐_𝑖 represents the unobserved time-invariant country effect and 𝜀_𝑖,𝑡 the error term. The probability of the onset of a systemic banking crisis in country 𝑖 during year 𝑡 for all the fixed effect estimators present in this study assumes a logistic distribution, thus a logistic regression is utilized for all these estimators.

(19)

19

Utilizing a fixed effects estimator however, presents potential issues. Importantly those countries which do not experience a banking crisis during the time period shall be dropped from the estimator as they experience no variance within the dependent variable over the period. Consequently the results of the study shall be conditional on a crisis occurring. This is not desirable for the study as the aim is to assess the impact of the conditions upon the probability of a crisis occurring the next year. This reliance upon a fixed-effect estimator has been reported to be a weakness of previous studies (Caballero, 2012).

To overcome this issue all the models shall additionally be estimated with the use of a random effects estimator. A random effect estimator, however is likely to generate biased estimates as it is very improbable the country specific effect is uncorrelated to the covariates included. To overcome this, the country specific means of all covariates shall also be included in the estimator, following the methodology proposed by Mundlak (1978). Through the inclusion of the country-specific means for the explanatory covariates suspected to be endogenous to the country intercept, the random effects estimator is now able to control for time-invariant country intercepts (Mundlak 1978) .

The random-effect estimator allows for the inclusion of all countries with available data. This enables those countries which faced a sustained period of high gross capital inflows but not a crisis36 to be included and therefore will find if there is in fact a meaningful association between systemic banking crises and sustained high gross capital inflows within the data. As it is not just those that experienced a banking crisis during the period included, the instances of banking crises are comparatively rarer as the number of country-year observations increases but the number of banking crises does not37_{. The assumption therefore for the random-effect estimators} used in this study is that the Gumbel distribution is followed. The result of this is that complementary logarithmic regressions are utilized for all random-effect estimators carried out in this study.

To ensure that the impact sustained high gross capital inflows has upon the likelihood of a systemic banking crisis the following year is independent from that of a bonanza in net capital inflows during the previous year38, the model is adapted to control for the presence of a net capital inflow surge. This is achieved through the introduction of a dummy variable which takes the value 1 if there is a net capital inflow bonanza present in the previous year and 0 otherwise.

36_{Unlike in the fixed effect estimator which would drop these countries.}

(20)

20

The fixed effect estimator of this model is displayed by equation 2, whereby Ni,t represents the aforementioned dummy variable for the existence of a net capital inflow bonanza in the previous year.

𝑦

_𝑖,𝑡∗

= 𝛼 + 𝛿𝐾𝑠

_𝑖,𝑡∗𝑦

+ 𝜆𝑁

_𝑖,𝑡

+ 𝛽𝑋

_𝑖,𝑡

+ 𝜐

_𝑖

+ 𝜀

_𝑖,𝑡 (2)

𝐾𝑠_𝑖,𝑡∗𝑦 takes the same definitions as previously found for equation 1. 𝑁_𝑖,𝑡 takes the following definitions:

For a Baseline Bonanza:

𝑁

_𝑖,𝑡 =

{1 𝑖𝑓

𝐷𝑖,𝑡−1 > 𝜎

(

𝐷𝑖

)

For an Intense Bonanza:

𝑁

_𝑖,𝑡 =

{1 𝑖𝑓

𝐷𝑖,𝑡−1 > 2𝜎

(

𝐷𝑖

)

The effect of sustained high gross capital inflows upon the likelihood of a banking crisis the following year remains significant controlling for the presence of a net capital inflow bonanza if the estimated coefficient 𝛿̂ remains significant and positive when this net capital inflow bonanza covariate is included.

4.2.3 Model 2

In order to answer the first sub-question of this paper: whether the presence of high leverage within the domestic banking system alters the impact high gross capital inflows with a corresponding asset boom has upon the likelihood of a systemic banking crisis. The first model, represented by equation (1) is altered to include a dummy variable indicating the presence of (sustained) high leverage in the domestic banking system (𝐿∗𝑦_𝑖,𝑡), along with the interaction of sustained high gross capital inflows and high leverage in the domestic banking system (𝐿 × 𝐾𝑠)_𝑖,𝑡∗𝑦. The fixed effect estimator of the model is displayed in equation (3):

𝑦

_𝑖,𝑡∗

= 𝛼 + 𝛿𝐾𝑠

_𝑖,𝑡∗𝑦

+ 𝜌𝐿

∗𝑦_𝑖,𝑡

+ 𝜏(𝐿 × 𝐾𝑠)

_𝑖,𝑡∗𝑦

+ 𝛽𝑋

_𝑖,𝑡

+ 𝜐

_𝑖

+ 𝜀

_𝑖,𝑡 (3)

Where 𝐾𝑠_𝑖,𝑡∗𝑦 takes the same definitions as previously and 𝐿∗𝑦_𝑖,𝑡 takes the following definitions:

𝐿

1𝑦_𝑖,𝑡 =

{1 𝑖𝑓 𝜑

𝑖,𝑡−1

> 0

(21)

21

𝐿

3𝑦_𝑖,𝑡 =

{1 𝑖𝑓 𝜑

𝑖,𝑡−1

> 0, 𝜑

𝑖,𝑡−2

> 0, 𝜑

𝑖,𝑡−3

> 0

= 1

(22)

22

4.2.2 Model 3

Model (1) is then be adapted in order to answer the second sub question of this paper: whether the presence of an asset boom coinciding with sustained high gross capital inflows alters the impact sustained high gross capital inflows has upon the likelihood of a systemic banking crisis. This requires the introduction of a dummy variable which takes the value of 1 in the presence of a boom in domestic asset values, here represented as 𝐴_𝑖,𝑡∗𝑦. This dummy variable is interacted with the sustained high gross capital inflow dummy variable. Equation (4) displays the fixed effect estimator for this second model:

𝑦

_𝑖,𝑡∗

= 𝛼 + 𝛿𝐾𝑠

_𝑖,𝑡∗𝑦

+ 𝛾(𝐴 × 𝐾𝑠)

∗𝑦_𝑖,𝑡

+ 𝛽𝑋

_𝑖,𝑡

+ 𝜐

_𝑖

+ 𝜀

_𝑖,𝑡 (4)

Where

(𝐴 × 𝐾𝑠)

_𝑖,𝑡2𝑦 =

{1 𝑖𝑓𝐾𝑠

𝑖,𝑡2𝑦

= 1,

banking crisis?

5.1.1 Non-parametric Analysis

The unconditional probability of a crisis is the percentage of the total country-year observations whereby a banking crisis has started:𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠_𝑡). The unconditional probability of facing a banking crisis in the next year is 2.79%. The conditional probability represents the percentage of country-year pairings that resulted in a banking crisis given that there was an episode of high gross capital inflows in the previous year 𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠𝑡|𝐾𝑠𝑖,𝑡1𝑦= 1). The conditional probability is 3.58%, with 74 out of a possible 115 banking crises in the sample following on from an episode of high gross capital inflows. All of the independence tests are able to be rejected at the 1% level; thus there is a statistical association between episodes of high gross capital inflows and the occurrence of banking crises the following year39.

If sustained high gross capital inflows increase the likelihood of facing a banking crisis, conditioning by increased lengths of time the gross capital inflows remain high must result in a greater conditional probability. The length of time in which the gross capital inflows remain high is extended to two consecutive years. The unconditional probability of facing a banking crisis the following year in this sample is 2.74%40_{. The conditional probability of a country} facing the start of a systemic banking crisis in the year following a sustained two year period of high gross capital inflows is 4.92%41. All the independence tests able to be rejected at the 1% level; the statistical association between the onset of systemic banking crises and high gross capital inflows remains when this high gross capital inflow episode is sustained for two years.

The length of time in which the high gross capital inflows are sustained is increased further to three years. The number of country-year pairings falls to 3,823 observations, with a total of 104 crises included. The unconditional probability of a banking crisis in the next year is 2.72%. The

39_{In line with the results found in Caballero 2012, when these gross capital inflows fulfill the criteria of a}

baseline bonanza, the conditional probability rises to 7.93%.

40_{This reflects that there are now only a total of 109 crises included}_{and a reduced number of country-year}

observations

41_{i.e 𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠}

(24)

24

conditional probability for the onset of a systemic banking crisis in the year following a three year sustained episode of high gross capital inflows is 6.05%42_{. All independence tests are again} rejected at the 1% level, thus a statistical association between the two.

The number of years in which the high gross capital inflow episode is sustained increases to four. The unconditional probability of experiencing the onset of a systemic banking crisis is 2.62% for this sample of 3,670 country-year pairings. The conditional probability for the start of a systemic banking crisis in the year following a four year sustained episode of high gross capital inflows is 6.33%. All independence tests are rejected; the statistical association between banking crises and episodes of high gross capital inflows remains when the episode is maintained for four consecutive years.

Finally the impact of a five year sustained episode of high gross capital inflows is analyzed. The unconditional probability is 2.62% for 3,517 country year pairings, containing 92 systemic banking crises. The conditional probability for the start of a systemic banking crisis in the year following a five year sustained period of high gross capital inflows is 5.84%. The Pearson chi-squared test of independence is able to be rejected at the 1% level, with the other two tests able to be rejected at the 5% level, implying the statistical association remains.

The non-parametric results identify the potential danger of sustained high gross capital inflows upon a countries banking system. Experiencing gross capital inflows greater than trend leads to an increased probability of facing a systemic banking crisis in the following year, this conditional probability rises with the length of time the high gross capital inflows are sustained as risk builds up in the system. The results suggest a confirmation of the first hypothesis in this paper as when the period of high gross capital inflows is sustained the likelihood of facing the onset of a systemic banking crisis the following year increases. However this non-parametric analysis may over or under estimate the true impact these differing lengths of sustained high gross capital inflows has upon the probability; with the results of the multivariate regression indicating sustained high gross capital inflows have a larger impact upon the likelihood of a banking crisis starting the following year than suggested by the non-parametric analysis.

42_{i.e 𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠}

(25)

25

5.1.2 Multivariate analysis

Table 1 reports the results from the estimated fixed effect logistic model (FE-Clogit) with multiple specifications to control for high gross capital inflows sustained for differing lengths of time. All the results are presented as exponentiated coefficients43, in order to aid in their interpretation and determining the economic impact44. The corresponding z statistic for each coefficient is presented in parentheses.

The first specification estimates the correlation between a one year episode of high gross capital inflows and a banking crisis in the following period, with all control variables included. The coefficient result is significant at the 5% level with a positive odds ratio greater than unity. This indicates that a country which experiences high gross capital inflows, has odds 1.597 times greater of experiencing a systemic banking crisis in the following period than that of a country which did not experience high gross capital inflows45.

Specification 2 then introduces the episode of high gross capital inflows being sustained for two consecutive years. The coefficient for this sustained episode remains statistically significant, yet now at the 1% level. The odds ratio increases to 2.505, thus if a country faces high gross capital inflows sustained for two years its odds of experiencing the onset of a systemic banking crisis the following year are 2.505 times greater than that of a country which did not experience this sustained two year episode of high gross capital inflows over the same time period.

Specifications 3, 4 and 5 increase the length of time that the high gross capital inflows are sustained to three, four and five years respectively. For all three specifications the odds ratios estimated for the episodes of sustained high gross capital inflows remain statistically significant. An episode of high gross capital inflows sustained for three years increases the odds of a country facing the onset of a systemic banking crisis starting the following year by 3.326 when compared to a country which did not experience the high gross capital inflows sustained for three years. If this episode of high gross capital inflows is sustained for four and five years then the odds are 2.224 and 2.484 times greater, respectively.

Table 2 reports the coefficients from the random effects estimator (RE-Mundlak) which includes the country-mean values for all independent variables. The exponentiated coefficients

43_{Also referred to as Odds ratios} 44_{The odds ratio is expressed as 𝑜𝑟 =} 𝑝

1−𝑝, with 𝑝 = Pr(𝑦 = 1| 𝑋) being the probability of the onset of a

banking crisis

(26)

26

provided along with corresponding z statistics in parentheses. The RE-Mundlak estimator provides estimates in line with those found in the fixed effect estimator (Table 1). The advantage being that unlike the fixed effect estimator, the results are no longer based upon those countries that experienced a crisis during the sample period. All the coefficients of interest are positive and greater than unity, indicating an increase in the odds of experiencing the onset of a banking crisis in the following year. Furthermore most of these coefficients are significant at the 1% level.

Overall the results of the multivariate analysis support the finding that a country experiencing high gross capital inflows experiences an increase in the likelihood of facing a systemic banking crisis in the following year46. Crucially it is found that if this episode of high gross capital inflows is maintained for more than one year, the increase in the likelihood is of a greater magnitude. The effect of an episode of three year sustained high gross capital inflows raises the probability of a banking crisis in the following year from an unconditional probability of 4.84%47 to 16%48.

46_{The estimated coefficients for all covariates not of interest are consistent with what has been found}

previously in the literature. The existence of a contemporaneous Currency crisis increases odds by a magnitude of at least 3 in all specifications. An increase in reserves reduces the odds with an estimated odds ratio

consistently below 1. Higher levels of GDP growth, inflation and domestic credit increase the odds of a systemic banking crisis starting the following year. The coefficient on the existence of an explicit deposit insurance scheme is consistently positive yet statistically insignificant.

47_{This is the unconditional probability for the baseline sample used in the regression composed of 1,654}

country-year observations. There are 80 crises present, thus the unconditional probability is equal to 4.84% (80/1,654).

48_{This is using the coefficient from the fixed effect estimator. For the RE-Mundlak estimator the unconditional}

(27)

27

Table 1

Differing Lengths Of Maintained High Gross Capital Inflows and Banking Crises, FE-Clogit

*** p<0.01, ** p<0.05, * p<0.1

Table 2

Differing Lengths Of Maintained High Gross Capital Inflows and Banking Crises, RE-Mundlak

*** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

High Gross Capital Inflows Maintained 1 Year

1.597* (1.91) High Gross Capital Inflows

Maintained 2 Years

2.505*** (3.80) High Gross Capital Inflows

Maintained 3 Years

Maintained 4 Years

2.224** (2.51) High Gross Capital Inflows

Maintained 5 Years 2.484** (1.96) Observations 1,654 1,654 1,654 1,654 1,654 Number of Countries 67 67 67 67 67 Crises 80 80 80 80 80

Controls Yes Yes Yes Yes Yes

(1) (2) (3) (4) (5)

High Gross Capital Inflows Maintained 1 Year

Maintained 2 Years

Maintained 3 Years

Maintained 4 Years

Maintained 5 Years 2.540** (2.19) Observations 2,871 1,654 1,654 1,654 1,654 Number of Countries 141 141 141 141 141 Crises 80 80 80 80 80

Controls Yes Yes Yes Yes Yes

(28)

28

Does any increase in the likelihood occur due to the presence of net capital inflow bonanzas?

Net capital inflow bonanzas have commonly been identified in the literature as increasing the likelihood of a crisis (Caballero, 2012)49, furthermore over half of large net capital inflow episodes are characterized by large gross capital inflows (Benigno et al., 2015). To investigate whether the impact of sustained gross capital inflows is driven by the presence of a net capital inflow bonanza, the model now controls for a net capital inflow baseline bonanza in the previous year50. Table 3 presents the results of the fixed effect estimator including the covariate for a net capital inflow surge in the previous year. All variables of interest remain significant when controlling for net capital inflow bonanzas. This indicates that the impact of sustained high gross capital inflows upon the likelihood of experiencing a banking crisis in the following year is independent to the impact of a net capital inflow bonanza. Even when controlling for the presence of a net capital inflow baseline bonanza in the previous year, if gross capital inflows are sustained at a high level for the previous three years, the odds of facing a bank crisis in the following year are 3.371 times greater than for a country which did not experience these high gross capital inflows.

Table 4 presents the estimated coefficients for the random effect estimator including the control for a net capital inflow bonanza in the previous year. The estimated coefficients from this estimator are in line with those from the fixed effect estimator (table 3). All the variables of interest remain significant, with the majority being at the 1% level.

49_{Non-parametric testing revealed the presence of a net capital inflow bonanza in the previous year raises the}

conditional probability of a banking crisis to 5.03%, which is higher than that of the unconditional probability of 2.79%. This is in line with the results found in Caballero (2012).

(29)

29 Table 3

Differing Lengths Of Sustained High Gross Capital Inflows and Banking Crises, Controlling For The Presence Of A Net Capital Inflow Baseline Bonanza, FE-Clogit

*** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

High Gross Capital Inflows maintained 1 Year

Maintained 2 Years

Maintained 3 Years

Maintained 4 Years

Maintained 5 Years

2.415* (1.88) Net Capital Inflow Baseline

Bonanza Previous Year

1.059 (0.18) 0.927 (-0.24) 0.910 (-0.30) 1.099 (0.31) 1.148 (0.45) Observations 1,654 1,654 1,654 1,654 1,654 Number of Countries 67 67 67 67 67 Crises 80 80 80 80 80

(30)

30

Table 4

Differing Lengths Of Sustained High Gross Capital Inflows and Banking Crises, Controlling For The Presence Of A Net Capital Inflow Baseline Bonanza, RE-Mundlak

*** p<0.01, ** p<0.05, * p<0.1

5.2 Does the increase in likelihood run through the domestic leverage

mechanism?

5.2.1 Non-Parametric Analysis

The conditional probability 𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠𝑡|(𝐿 × 𝐾𝑠)𝑖,𝑡2𝑦= 1) is equal to 6.84% with all three independence tests rejected at the 1% level. This conditional probability is greater than the unconditional probability of 2.58% as well as the conditional probability

𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠_𝑡|𝐾𝑠_𝑖,𝑡2𝑦= 1) which equals 4.92%. The conditional probability

𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠_𝑡|(𝐿 × 𝐾𝑠)_𝑖,𝑡3𝑦= 1) equals 6.96%51, this is greater than both the unconditional probability (2.57%) and the conditional probability from 𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠_𝑡|𝐾𝑠_𝑖,𝑡3𝑦= 1) which is equal to 6.05%. 𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠𝑡|(𝐿 × 𝐾𝑠)𝑖,𝑡4𝑦 = 1) is equal to 8.33%52, which is greater than

51_{Again all independence tests are rejected}

52_{The Pearson test of independence is rejected at the 1% level, while the likelihood-ratio and fishers exact}

independence tests are rejected at the 5% level.

(1) (2) (3) (4) (5)

High Gross Capital Inflows maintained 1 Year

Maintained 2 Years

Maintained 3 Years

Maintained 4 Years

Maintained 5 Years

2.413** (2.03) Net Capital Inflow Baseline

Bonanza Previous Year

1.130 (0.42) 0.940 (-0.21) 0.945 (-0.19) 1.168 (0.54) 1.207 (0.65) Observations 2,841 2,841 2,841 2,841 2,841 Number of Countries 141 141 141 141 141 Crises 80 80 80 80 80

(31)

31

the unconditional probability of 2.51%. In addition to this it is greater than the conditional probability 𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠𝑡|𝐾𝑠𝑖,𝑡4𝑦 = 1) of 6.33%53. These results imply the joint occurrence of sustained high gross capital inflows with a highly leveraged domestic banking system increases the likelihood of a banking crisis the following year.

5.2.2 Multivariate Analysis

Table 5 presents the results of the fixed effect estimator, again the results are shown as odds ratios. Comparing the coefficients present in specifications (1), (2) and (3) we still find a robust and significant54 association between three year sustained high gross capital inflows and banking crises. The magnitude of this coefficient remains similar when discounting for the presence of high leverage (specification 2) and when the high leverage is not present in the domestic banking sector (specification 3). The same pattern is seen when the period of sustained high gross capital inflows is increased to four years; comparing specifications (4), (5) and (6); whereby even in the absence of high leverage the coefficient upon sustained high gross capital inflows remains positive and significant. However when comparing the coefficient upon four year sustained high gross capital inflows with the introduction of the high leverage variable (i.e specification 5 vs specification 4) the coefficient upon the sustained high gross capital inflows reduces slightly, suggesting . As all three interaction variables (specification 3, 6 and 9) are insignificant, this indicates that sustained high gross capital inflows have roughly the same impact upon the likelihood of banking crises occurring in the next year either with or without high leverage in the domestic banking system once all controls are included. These findings remain when undertaking the RE-Mundlak estimator (displayed in Table A1)

53_{The conditional probability of 𝑝(𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠}

(32)

32

Table 5

Interaction between Sustained High Gross Capital Inflows And High Leverage Upon Banking Crises, FE- Clogit

*** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5) (6) (7) (8) (9)

High Gross Capital Inflows Sustained 3 Years 3.013*** (3.95)

3.029*** (3.94)

3.551*** (3.62)

High Gross Capital Inflows Sustained 4 Years 2.616***

(2.87)

2.521*** (2.74)

2.671** (2.49)

High Gross Capital Inflows Sustained 5 Years 3.231**

(2.50)

3.193** (2.46)

3.755*** (2.66)

High Leverage Sustained 3 Years 1.917**

(2.33)

2.227** (2.35)

High Leverage Sustained 4 Years 1.821*

(1.94)

1.906* (1.85)

High Leverage Sustained 5 Years 2.253**

(2.18)

2.488** (2.38) High Gross Capital Inflows X High Leverage

Sustained 3 Years

0.628 (-0.75) High Gross Capital Inflows X High Leverage

Sustained 4 Years

0.806 (-0.27) High Gross Capital Inflows X High Leverage

Sustained 5 Years 0.293 (-0.82) Observations 1,439 1,439 1,439 1,439 1,439 1,439 1,439 1,439 1,439 Number of Countries 57 57 57 57 57 57 57 57 57 Crises 69 69 69 69 69 69 69 69 69

(33)

33

5.3 Does the presence of an asset boom coinciding with high gross capital

inflows impact upon the likelihood of a crisis?

5.3.1 Non-Parametric Analysis

This subsection investigates the association between banking crises and sustained high gross capital inflows with an asset boom, based on unconditional and conditional probabilities, along with independence tests. If the interaction of high gross capital inflows with a domestic asset boom increases the likelihood of a banking crisis then the conditional probability of experiencing a banking crisis in the following year must be higher conditional upon sustained high gross capital inflows with an asset boom present than that from sustained high gross capital inflows without controlling for the presence of an asset boom for the same sample.

The sample of countries with the available data upon sustained high gross capital inflows and asset booms is restricted to the sample with data upon having an interaction between high gross capital inflows sustained for three years with an asset boom sustained for the latter two years of this period. The result of this is 1,709 country-year pairings. The unconditional probability of facing a banking crisis is 2.75% for the sample. The conditional probability of experiencing a banking crisis, given the presence of high gross capital inflows sustained for two years with an asset boom in the last of these two years is 12.50%. This is higher than the conditional probability given the presence of high gross capital inflows sustained for the two previous years, which equals 7.43%55_{. All three independence tests are able to be rejected at the 1% level,} therefore there is a statistical association between bank crisis and high gross capital inflows combined with an asset boom.

When the high gross capital inflows is sustained for three years with an asset boom present for the last two of these, the conditional probability of a banking crisis the following year is 14.04%. All independence tests are rejected at the 1% level, thus the statistical association remains. The conditional probability of a bank crisis given high gross capital inflows sustained for the previous three years is 8.90%56. Therefore for both set time periods sustained high gross capital inflows occurring with an asset boom the conditional probability is greater than that of high gross capital inflows over the period not controlling for the existence of this asset boom.

55_{Therefore 𝑃𝑟(𝐵𝑎𝑛𝑘 𝐶𝑟𝑖𝑠𝑖𝑠}

𝑡|(𝐴 × 𝐾𝑠)𝑖,𝑡2𝑦= 1) > 𝑃𝑟(𝐵𝑎𝑛𝑘 𝐶𝑟𝑖𝑠𝑖𝑠𝑡|𝐾𝑠𝑖,𝑡2𝑦= 1) 56_{𝑃𝑟(𝐵𝑎𝑛𝑘 𝐶𝑟𝑖𝑠𝑖𝑠}

Do Sustained International Gross Capital Inflows Influence The Probability Of Banking Crises?