Essays on bank intermediation in developing countries

(1)

Essays on Bank Intermediation

in Developing Countries

Thèse

Pierre Valère Nketcha Nana

Doctorat en économique Philosophiæ doctor (Ph.D.)

Québec, Canada

(2)

(3)

Résumé

Cette thèse comprend trois essais sur des problèmes d’intermédiation bancaire dans les pays en développement. Le premier essai aborde la question des facteurs de l’accumu-lation des réserves bancaires en Afrique. La plupart des travaux antérieurs soulignent le rôle du risque de crédit. Alors que la théorie moderne de l’intermédiation financière montre que l’exposition au risque de liquidité peut inciter les banques à accumuler des actifs liquides, cette hypothèse n’a pas reçu beaucoup d’attention. Nous soutenons que les banques en Afrique sont exposées à un risque de liquidité relativement important ; ensuite, à l’ aide des données couvrant des pays africains sur la période 1994-2008, nous montrons que ce risque est en fait un facteur important de l’accumulation des réserves bancaires. Nos résultats suggèrent que le risque de liquidité réduit significativement la part des dépôts que les banques peuvent distribuer sous formes de crédits ; et par consé-quent, pourrait être un facteur de la disponibilité (ou de la rareté) des financements bancaires en Afrique.

Les deux autres essais de cette thèse abordent la question des déterminants de la dis-ponibilité (ou de la rareté) des prêts bancaires dans les pays en développement. Dans le deuxième essai, nous reconsidérons le rôle du risque de crédit ; plus généralement, des institutions du marché du crédit. Spécifiquement, nous utilisons des données et des mesures plus récentes, publiées par Doing Business, pour réexaminer la question des effets sur le volume des prêts bancaires, de la protection juridique des prêteurs et des emprunteurs, ainsi que du partage des informations sur le crédit. Nos données couvrent un large échantillon de 143 pays sur la période 2006-2010. En accord avec les travaux antérieurs, les résultats de nos analyses indiquent qu’une meilleure protection juridique des prêteurs et des emprunteurs, et un plus grand partage des informations sur le cré-dit sont en général associés à des volumes de prêts bancaires plus importants. Nous trouvons que ces effets demeurent significatifs lorsque l’échantillon inclut uniquement les pays en développement, ou uniquement les pays pauvres.

(4)

Dans le troisième essai, nous considérons le rôle du risque de liquidité et de la poli-tique monétaire. Ces deux facteurs n’ont que très peu été considérés dans les travaux antérieurs sur les déterminants du volume des prêts bancaires dans les pays en dévelop-pement. Nos analyses empiriques, à l’aide des données de panel sur 97 pays pauvres et à revenus intermédiaires, sur la période 2004-2010, montrent que le risque de liquidité et la politique monétaire affectent significativement le volume des prêts bancaires dans les pays en développement. Nous trouvons que ces effets sont très hétérogènes : les effets du risque de liquidité et de la politique monétaire, sur le volume des prêts bancaires, sont plus forts lorsque les conditions du marché du crédit sont favorables ; ces effets sont plus faibles, voire non significatifs lorsque les conditions du marché du crédit sont mauvaises. Ces résultats sont importants pour la politique économique. Car il suggèrent que, dans certains pays en développement tout au moins, notamment ceux où le risque de crédit est relativement faible, des mesures visant à réduire l’exposition des banques au risque de liquidité, et/ou à rendre la politique monétaire moins restrictive, permet-traient d’accroître le volume des prêts bancaires. De telles mesures ne fonctionneraient pas dans les pays où le risque de crédit est relativement plus élevé ; dans ces pays, en effet, il est primordial de réduire le risque de crédit.

(5)

Abstract

This dissertation consists of three empirical essays on issues of bank intermediation in developing countries. The first essays seeks to improve our understanding of why banks in Africa are hoarding large volume of liquid assets. Prevailing explanations of this phenomenon have focused mostly on the role of credit risk. Yet, modern models of financial intermediation show that a high exposure to liquidity risk may also prompt banks to hoard large amounts of (precautionary) liquid reserves. We argue that this risk is important in Africa; and using data over the 1994-2008 period, we provide evidence indicating that it contributes significantly to the hoarding of bank liquid assets. This evidence suggests that liquidity risk reduces the share of deposits that African banks can channel into credits, which therefore, can adversely affect the availability of bank credit.

The second and the third essays focus on the issue of the determinants of the availability of bank credit, or the lack thereof, in developing countries. In the second essay, we (re-)consider the role of credit risk, or more generally, of credit market institutions. Specifically, we use new data and improved measures from Doing Business, to reexamine the issue of the relationships between creditor rights protection and credit information sharing on one hand, and bank credit on the other hand. The data covers a large sample of 143 countries and are taken in averages over the period 2006-2010. Our results indicate the robustness of earlier evidence that both stronger creditor rights protection and better credit information sharing are associated with greater availability of bank credit. We find that these effects are significant even when the sample is restricted to include either developing countries only or poor countries only.

In the third essay we consider the role of liquidity risk and monetary policy. These two factors have not received much attention in previous empirical studies on the determi-nants of bank credit in developing countries. Using a panel dataset which covers 97 low-and middle-income countries over the 2004-2010 period, we show that liquidity risk low-and

(6)

monetary policy are actually important determinants of the availability of bank credit in developing countries. We find important heterogeneity in the results: both liquidity risk and monetary policy have greater effects on bank credit in economies with better credit market conditions, but much smaller and even not statistically significant effects in economies with poor credit market conditions. This result is important because it suggests that, at least in some developing countries, those with a relatively low level of credit risk, reducing the exposure of banks to liquidity risk, and/or implementing a less restrictive monetary policy, are effective channels through which the availability of bank credit could be enhanced. For countries with a relatively high level of credit risk, such channels would be ineffective; in these countries, reducing credit risk is of first order importance to stimulate bank lending.

(7)

List of Tables

1.1 Summary statistics . . . 14

1.2 Econometric results . . . 18

2.1 Descriptions and sources of variables . . . 28

2.3 Correlation matrix. . . 31

2.4 OLS regressions for private credit . . . 33

2.5 OLS regressions for private credit, with alternative measure of information sharing . . . 39

2.6 Checking for the effects of outliers (in the results reported in Table 2.5) . . . 40

2.7 Checking for the effects of outliers and reverse causality (in the results reported in Table 2.4) . . . 42

3.2 Correlation matrix. . . 55

3.3 Main effects of liquidity risk and monetary policy on bank credit . . . 59

3.4 Main effects of liquidity risk and monetary policy: Sensitivity to additional controls . . . 60

3.5 Main effects of liquidity risk and monetary policy: Sensitivity to outliers . . . 63

3.6 Effects of liquidity risk and monetary policy on bank credit: Interaction effects . . . 66

3.7 Effects of liquidity risk and monetary policy on bank credit by levels of credit market conditions . . . 70

(10)

(11)

List of Figures

1.1 Median bank liquid reserves-to-bank assets ratio. . . 6 1.2 Reserve ratio and deposit volatility . . . 15 3.1 Liquid assets to total assets ratio (in %) for the average bank . 47

(12)

(13)

Acknowledgments

This dissertation benefited greatly from the mentoring and support I have received from my advisor, Professor Lucie Samson.

I have also benefited from helpful comments from and conversation with many mem-bers of the Department of Economics at Laval, including, in no particular order, Patrick Gonzalez, Kevin Moran, Stephen Gordon, Sylvain Dessy, Guy Lacroix, Bruce Shearer, and Charles Bellemare. I acknowledge financial support from the Department of Eco-nomics at Laval, and from the African Economic and Research Consortium, in Naïrobi, Kenya.

I thank Patrick Gonzalez, Kevin Moran, and Yazid Dissou for their willingness to spent time and effort to examine this dissertation. A special thank to Patrick Gonzalez who has given me the opportunity to pursue my doctoral studies at Laval. This dissertation is dedicated to him.

Finally, I thank all the members of the staff of the Department of Economics at Laval, the staff of the Library at the Jean-Charles-Bonenfant building, all the members of my family, and all of my friends for their valuable support throughout the completion of this dissertation.

(14)

(15)

General Introduction

As institutions that stand between agents with surplus funds and agents with funding needs, banks traditionally perform two complementaries activities: they accept de-posits, which are typically payable on demand, and they allocate credits to consumers and entrepreneurs.1 _{These activities can be important to economic activity (}_{King and}

Levine,1993;Levine,2005), especially in developing countries where organized financial markets and other forms of intermediaries, such as mutual funds and pension funds, are less developed. Thus, an important issue for development economics is to study the functioning of bank intermediation in developing countries.

What factors explain the current features and levels of bank intermediation in develop-ing countries, and how are these related to economic activity? How best could policy makers actually intervene with regard to enhancing the positive contribution of bank intermediation to economic development? These, among others, are examples of ques-tions that, indeed, may help understand and address some of the challenges faced by the developing world. Part of these questions is examined in this thesis, which consists of three related empirical essays.

The first essay, entitled “Why are banks in Africa hoarding reserves? An empirical investigation of the precautionary motive", examines the importance of liquidity risk in explaining the accumulation of liquid assets in banks in sub-Saharan Africa over the 1994-2008 period. The existing empirical literature on bank intermediation in developing countries has largely neglected the potential detrimental effects that might arise because of a high exposure of banks to liquidity risk. We argue, based on the characteristics of banking systems, that this risk is clearly important in Africa; and we provide suggestive evidence that it reduces the share of deposits that African banks can channel into credits.

1_{In this thesis, banks refer to deposit money banks, i.e. to resident depository corporations and}

quasi-corporations other than the Central Bank which have any liabilities in the form of deposits payable on demand, transferable by cheque or otherwise usable for making payments.

(16)

Despite its focus on Africa, our first essay underscores an hypothesis that might be relevant to help understand the behavior of banks in other developing countries as well. In fact, for more than two decades now, the accumulation of large volumes of liquid assets has been a salient feature of banking systems throughout the developing world (see, for e.g.:Caprio and Honohan, 1993; Deléchat and al., 2012). The precautionary motive seems important to understand this feature and so, it points out to the potential adverse effect that liquidity risk may have on the supply of bank credit in developing countries.

The two remaining essays of our thesis focus on the important issue of the determinants of the availability of bank credit, or the lack thereof, in developing countries. Our second essay, which is entitled “Legal rights, information sharing and bank credit: New cross-country evidence", (re-)considers the role of credit market institutions. More precisely, we use new data and improved measures from Doing Business, to reexamine the issue of the relationships between creditor rights protection and credit information sharing on one hand, and bank credit on the other hand. The data covers a large sample of 143 countries and are taken in averages over the period 2006-2010. This is to facilitate comparisons with the results of previous studies which, in most of the cases, are based on a cross-sectional empirical framework. Our results indicate the robustness of earlier evidence that both stronger creditor rights protection and better credit information sharing are associated with deeper credit markets.

Building on the first two essays, our last essay, entitled “Liquidity risk, monetary policy and bank lending: Evidence from developing countries", questions the empirical impor-tance of liquidity risk as a determinant of bank credit in developing countries. It also focuses on understanding to what extent monetary policy can help in stimulating the supply of bank credit. The potential effects from liquidity risk and monetary policy to bank credit have not been the subject of much attention in previous empirical studies on developing countries. To fill the gap, we use a panel dataset which covers 97 low- and middle-income countries over the 2004-2010 period. Our results show that liquidity risk and monetary policy are actually important to understand the supply of bank credit in developing countries. We find important heterogeneity in the results: both liquidity risk and monetary policy have greater effects on bank credit in economies with better credit market conditions, but much smaller and even not statistically significant effects in economies with poor credit market conditions. This result is important because it suggests that, at least in some developing countries, those with a relatively low level of credit risk, reducing the exposure of banks to liquidity risk, and/or implementing a

(17)

less restrictive monetary policy, are effective channels through which the availability of bank credit could be enhanced. For countries with a relatively high level of credit risk, such channels would be ineffective; in these countries, reducing credit risk is of first order importance to stimulate bank lending.

On the whole, this thesis sheds light on the relative importance of liquidity risk versus credit risk, as well as on the role of monetary policy as determinants of the behavior of banks in developing countries. We view this research as a step towards a more complete understanding of the relationship between financial intermediation and growth and/or poverty reduction in the developing world.

(18)

(19)

Chapter 1 Why are Banks in Africa Hoarding

Reserves? An Empirical Investigation

of the Precautionary Motive

Abstract

For two decades now, many banks in Africa have been holding large amounts of liquid assets. Prevailing explanations of this phenomenon rely on credit rationing models. While modern models of financial intermediation show that a high expo-sure to liquidity risk may prompt banks to hoard large amounts of (precautionary) liquid reserves, this hypothesis has often been overlooked in previous literature. We try to fill the gap in this essay. More specifically, we hypothesize and confirm that bank liquidity hoarding in Africa reflects, at least partially, a precautionary strategy to guard against the risks associated with liquidity services to depositors.

JEL classifications: G11, G21, O16.

Keywords: Bank Liquidity hoarding; liquidity risk; deposit volatility; African banking systems.

(20)

1.1 Introduction

For two decades now, many banks in Africa have been holding large amounts of liquid assets (see, Caprio and Honohan, 1993, Freedman and Click, 2006, and Saxegaard, 2006). For instance, over the period 1990 and 2009, the ratio of liquid reserves to total assets for the median bank in Africa has varied between 11% and 19%. In comparison, over the same period, the liquid reserves to total assets ratio for the median bank in OECD has not exceeded 5% (see Figure 1.1).1

Figure 1.1: Median bank liquid reserves-to-bank assets ratio

0 2 4 6 8 10 12 14 16 18 20 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 SSA OECD

Source: World Development Indicators Database (WDI), The World Bank.

The issue of persistent large bank reserve holdings is critical, especially in Africa where lack of finance if often cited as one the most important constraints on the growth of firms (World Bank, 2007). In fact, the accumulation of large bank reserves displaces funding which could be used to increase the supply of credits to the private sector. It is thus important to find ways of getting a greater share of bank resources flowing to support private sector development.

1_{The statistics in this paragraph are from the World Development Indicators Database (WDI),}

The World Bank. The ratio of bank liquid reserves to bank assets is defined as the ratio of domestic currency holdings and deposits with the monetary authorities to claims on other governments, non-financial public enterprises, the private sector, and other banking institutions.

(21)

Academics and policymakers are confronting this issue (see,Freedman and Click,2006; Saxegaard,2006; Honohan and Beck, 2007;Owoundi,2009). Yet, the build-up of bank reserves in many African countries has mostly been viewed as a consequence of low levels of perceived expected returns on credit.2 In particular, most authors base their analysis on credit rationing models (Stiglitz and Weiss,1981; Jaffee and Stiglitz,1990) and suggest that efforts to address the credit market deficiencies would help increase the extent of bank intermediation.

The main contribution of this essay is to show that another channel may also be ef-fective. More specifically, we hypothesize and confirm that bank liquidity hoarding in Africa reflects, at least partially, a precautionary strategy to guard against liquidity risk. This result is consistent with the liquidity insurance role of banks as put forth by the modern theory of financial intermediation (see, Bryant, 1980; Diamond and Dyb-vig, 1983; Diamond and Rajan, 2001; Kashap and al., 2002; Tirole, 2011). It suggests that the share of deposits banks can channel into credits is constrained by the risks associated with liquidity services to depositors.3 Our empirical analysis shows that a reduction of deposit volatility, which is our primary measure of liquidity risk, will lead banks to significantly reduce their holdings of reserves for precautionary motive. Consequently, it will help expand the availability of loanable funds and eventually the supply of credit to the domestic private sector.

Besides this new result, this essay also contributes to the literature with a new mea-sure that capture bank’s expomea-sure to liquidity risk. In fact, our primary meamea-sure of liquidity risk is constructed based on the standard deviation of bank deposit inflows. An important feature of this indicator is that annual observations are computed from monthly data, which allows us to exploit substantial variation in the time series of deposit inflows and, also, to have a time-varying measure for our analysis. On other hand, an indicator of bank’s exposure to liquidity risk based on the distribution of deposit inflows is especially relevant in the African context, because, as documented in European Investment Bank (2013), deposits are the main source of banks’ liabilities in most African countries.

This essay is related to a recent strand of empirical literature that investigates the importance of the precautionary motive in explaining the holding of bank reserves, by

2_{This is related to credit market deficiencies such as the poor quality and scarcity of information}

about individual borrower risks, and the weak legal and judicial and contract enforcement infrastruc-tures (see,Honohan and Beck,2007).

3_{In the next section, we present a collection of facts that suggest banks’ exposure to liquidity risk}

(22)

examining the build-up of bank reserves in developed countries since the recent financial crisis (e.g.: Ashcraft et al.,2009;Acharya and Merrouche,2012;Cornett and al.,2011). Like most of this literature, we find that there is a powerful relationship between large bank reserves and high exposure to liquidity risk.

We are only aware of two papers, namely Agénor et al. (2004) and Saxegaard (2006), which carried out an empirical analysis closely related to ours, most notably by their fo-cus on developing countries: both papers estimate a demand function for excess reserves (or liquid assets) by commercial banks that captures, in particular, the precautionary motive for holding liquid assets. Nevertheless, our contribution is different from theirs in the approach we use to capture liquidity risk. Moreover, both papers relied on time series data and as such, their findings are likely to suffer from the “individual heterogeneity bias" (see Baltagi, 2008). By using panel data, we are able to control for country specific unobserved time-invariant variables; this enables us to obtain more reliable estimates.

The rest of this essay is organized as follows. Section 1.2 motivates the hypothesis of precautionary hoarding of bank liquidity in Africa. Section 1.3 describes the data. Section 1.4 presents the econometric analysis and results. Section 1.5 concludes.

1.2 Hypothesis motivation

The hypothesis of precautionary hoarding of bank liquidity flows from the modern theory of financial intermediation (e.g.,Bryant, 1980; Diamond and Dybvig, 1983; Di-amond and Rajan, 2001; Kashap and al., 2002; Tirole, 2011). In this literature, the fundamental role of banks is to make illiquid loans to borrowers while providing liq-uidity on demand to depositors. This liqliq-uidity insurance role, however, exposes banks to liquidity risk: demand for cash withdrawals may arrive before the loans mature and force banks to liquidate early and to fail. Hence, to carry out their job effectively, the models of financial intermediation show that banks must invest in a certain costly vol-ume of liquid assets as a hedge against a state of the world where there are unexpected demand for cash withdrawals.

A glance at banking systems in Africa shows that the precautionary motive for banks to hoard liquidity may be especially important. For example, underdeveloped and unreliable payment systems in many countries are such that cash is largely used as medium of exchanges. This implies that banks are likely to face frequent demand for

(23)

cash withdrawals. In addition, the lack of deposit insurance in several countries implies that there is a significant risk that banks may unexpectedly face large outflows of deposits (bank runs). Finally, the fact that capital markets are less developed suggests that banks cannot cannot accommodate liquidity shocks simply by raising new external finance on a moment’s notice.4 _{In such a context, hoarding liquid assets is critical as a}

strategy to mitigate the risk of liquidity shortage.

It is of course true that banks may resort to borrowing from the central bank as a more or less permanent source of funds to cope with liquidity shocks: banks can make use of the central bank’s standing facilities to meet extraordinary liquidity needs at a particular point in time. With this possibility, the precautionary motive to hoard liquid assets may no longer be plausible. However, refinancing conditions are in principle determined by the Central Banks. For example, virtually all liquidity provisions by Central Banks are necessarily based on adequate collateral, where the criteria of adequate collateral are defined by the Central Banks. This implies that if a Central Bank is more conservative in selecting the underlying collateral, banks would be restricted in their access to the Central Bank refinancing facilities.

More formally, Nautz (1998) shows that bank’s demand for liquid assets increases if the access to the Central Bank credits is expected to become more restricted, and if the refinancing is expected to become more expensive. Accordingly, banks reduce their demand for liquid assets if expectations of future refinancing conditions become more optimistic. These results extended earlier models in which bank’s demand for liquid assets depends on the current refinancing conditions set out by the Central Bank (Poole, 1968; Baltensperger,1980) as follows: If the benefits for a bank from resorting to Central Bank funds fail to cover the costs, banks reduce their demand for Central Bank funds and hold liquid assets large enough to make probability of liquidity shortage negligible.

Hence, the view that the build-up of bank reserves in Africa may reflect high exposure to liquidity risk should not be discarded without empirical evidence. While some authors (e.g., Caprio and Honohan, 1993; Nissanke and Aryeetey, 1998; Honohan and Beck, 2007) do acknowledge this point, supporting empirical evidence is surprisingly lacking. This essay will try to fill the gap.

(24)

1.3 Data

We use aggregate data at country-level. The following subsection describes our liq-uidity risk variables, and the one after that describes the other variables included in the empirical analysis. The last subsection presents the sample and some descriptive statistics.

1.3.1 Liquidity risk factors

To capture withdrawal or liquidity risk in connection with banks’ liabilities, our primary variable is a measure of the volatility of deposits. The rationale is simple: the greater banks are uncertain on the volume of deposits they can raise in the near future, the more likely they will keep precautionary reserves high enough so as to make the probability of liquidity shortage very low.5 One measure of such uncertainty for each period (year) can be computed from monthly data as follows:

V olt=

q

1

N −2P(dtn− dtn)2

µt

where: t is for years and n for months; dtn = Log(Dtn) − Log(Dt,n−1); dtn is the mean

of dtn over the N -months period t; µt = _N1 P Log(Dtn); Dtn is total bank deposits of

the nth-month of the period t.

For each year t, the numerator of V oltis the standard deviation of monthly (logarithm

of) deposits.6 It is normalized by the mean of monthly (logarithm of) deposits for the same year, to ensure comparison across countries (because deposits are measured in different units and currencies) and over time (because of widely different means). But the volatility of deposits (V ol) may not be sufficient to capture liquidity risk exposure. For one thing, it does not distinguish between positive and negative changes in deposit inflows, whereas the direction of change may be of critical importance. In fact, positive variability is acceptable while variability on the downside may represent the worst scenario. Moreover, V ol does not account for the possibility of highly asymmetric changes in deposit inflows, whereas such asymmetry may impact liquidity management

5_{Note that deposits are the main source of banks’ liabilities in most African countries (}_European

Investment Bank,2013).

(25)

decisions: positively (negatively) skewed7 deposits implies a probability of very large (low) amounts of deposit inflows. Thus, with other things equal, the risk of liquidity shortage may be lower with greater positive skewness but higher with greater negative skewness. This implies that precautionary reserves decreases as the distribution of deposit inflows becomes more positively skewed, but increases as the distribution of deposit inflows becomes more negatively skewed, ceteris paribus.

On other hand, the structure of bank deposits may also be relevant. In particular, demand deposits, unlike time and saving deposits, can be withdrawn at any time with-out any notice. This implies that, at a given time, banks with large shares of demand deposits are more likely to face large demand for cash withdrawals and may therefore need to hold large amounts of precautionary reserves.

To take these few remarks into account, we consider three additional variables, namely the year average of monthly deposit growth, the skewness of monthly deposits (Skew), and the ratio of demand to time and saving deposits. Regarding deposit skewness (Skew), it worth noting that while we would expect a negative association with re-serves,8 _{we allow this relationship to be nonlinear over the whole range of possible}

values of Skew. In fact, since banks may have a preference for positive skewness - as it would mean lower risk of liquidity shortage-, the association may be stronger for positive skewness (Skew > 0) than for negative skewness (Skew < 0), on the average. This leads us to consider a dummy _{I(Skew > 0), which is equals to 1 if the skewness} is positive, and include this in the regression along with Skew and the multiplicative interaction term Skew ∗_{I(Skew > 0).}

To construct all the variables described in this sub-section, we employ data on monthly deposits from the international Financial Statistics (IFS) database.

1.3.2 Other variables

Besides liquidity risk, cash holdings in the banking system depends on other factors as well. Models of bank reserves (seeBaltensperger,1980;Freixas and Rochet,1997) point

7_{Skewness is a measure of the asymmetry of a probability distribution. It is computed as follows:}

Skew = 1 N P(dti− dti) 3 1 N P(dti− dti) 23/2

8_{Note that when skewness is negative, an increasing Skew implies that the distribution becomes}

(26)

out two other key determinants of precautionary cash holdings. First is the opportunity cost of holding cash, that is the interest which could be earned by investing the cash in alternative assets such as loans; and second is the adjustment or transaction costs which banks may incur in case of emergency borrowing or emergency selling of assets. We use the lending rate -i.e., the bank rate that usually meets the short- and medium-term financing needs of the private sector- to proxy for the opportunity cost;9 _{and discount}

rate -i.e., the cost of borrowing at the central bank discount window- to proxy for the transaction cost.

Another key determinant of the volume of bank’s liquidity is the ease with which banks can access funding liquidity to meet liquidity shortfalls. In large financial systems, banks can access large volume of funding at relatively low transaction costs. One measure of this size effect is the ratio of liquid liabilities to GDP, which is equals to the sum of demand and interest-bearing liabilities of banks and non-banks financial intermediaries, divided by the GDP.

Finally, we should recognize that depending on frictions and constraints on credit mar-kets, such as lack of adequate demand for credits, banks may hold reserves larger than what is required for the precautionary motive. Hence, we extend our list of controls with four additional variables. First is the (logarithm of) gross domestic product (GDP), because larger economies might have larger credit demand. Second is the GDP growth, because, as suggested inDjankov et al.(2007), rapid economic expansion could require more credit. Third is the annual growth rate of the GDP implicit deflator (a measure of inflation), since it is very likely that banks will lend less in economies with high inflation (see, Boyd and al., 2001). The fourth variable is the International Country Risk Guide (ICRG) composite risk rating, to control for the overall quality of each country’s political and legal and economic conditions.10 _{This last variable is highly}

correlated with direct measures of credit risk, such as bad loans ratio (correlation is −0.57, p.value = 0.00), and depth of credit information index11_{(the correlation is 0.76,} 9_{We accommodate for the possibility of nonlinearities between the lending rate and the reserve}

ratio, by including a quadratic term for the lending rate variable (Lending rate2_{). The motivation is}

simple: As shown inStiglitz and Weiss (1981), lending rate affect the riskiness of loans, and there is an “optimal" rate at which the expected return to the bank is maximized; above the “optimal" rate, the best strategy for banks is to ration credit, which may imply that they increase their holdings of liquid assets.

10_{The ICRG composite risk rating covers a broad set of 22 variables in three subcategories of risk:}

political, economic, and financial. The scores on each of the 22 risk components allow to create a risk rating for each of the three risk categories, which are then combined to produce the composite risk rating. The composite rating ranges from zero to 100. A rating of 100 points equates to very low risk and a score of 0 points to very high risk. Source: www.prsgroup.com

(27)

in-p.value = 0.00). We use the ICRG rating because the availability of data is much better in our sample.

All the variables described in this sub-section are collected from the WDI of the World Bank, with the exception of the ICRG composite index which is from the PRS group (www.prsgroup.com).

1.3.3 The sample and some descriptive statistics

We focus our study on the 1994-2008 period.12 _{The dataset includes 18 SSA countries}

for which we have been able to get sufficient data to carry out the econometric analysis. A quick glance at the raw data shows that there are a few number of extreme observa-tions that make the range of some variables very large. For example, the lending rate varies from 10% to 103%, the discount rate from 5% to 150%, and the inflation rate from −17% to 556%. We double-checked the original data sources for any reporting error on our part, but we confirmed the entries.

Table 1.1 presents the summary statistics of our panel data as well as the list of countries included in the sample. It gives the mean, the median, the standard deviation and the range of values (minimum, maximum) of the variables described above.

Note that the panel is unbalanced, with the number of time-series observations varying substantially across countries. Also note that we have log-transformed some of the variables, namely lending rate, GDP and inflation, to curtail the effects of extreme values. Finally, note that we have considered the ratio of discount rate to lending rate (rather than the absolute level of the discount rate), to mitigate multicollinearity -since the correlation between the discount rate and the lending rate is 0.91 (p.value=0.00)-. This approach reduces the range of values and, therefore, it also helps in adjusting for extreme points.

In the first row of the table, we see that the mean (median) of bank reserve ratio in our sample is 15.898% (13.926%), with a standard deviation of 11.039. This shows that there is substantial variation in the level of bank reserve ratios. There is also

formation, from either a public registry or a private bureau, to facilitate lending decisions. Source: WDI.

12_{The period starting in 1994 coincides with major reversal of the financial and monetary policy}

frameworks of post-independence era, when African governments, through financial repression, de-termined directly the behaviour of banks (Allen and al., 2011). The new policy and institutional environment that emerged allows for autonomous management of banking firms, which is a crucial feature that underlies our empirical analysis.

(28)

Table 1.1: Summary statistics

Variables Mean Median St. dev. Min. Max.

Bank reserve ratio 15.898 13.926 11.039 2.172 76.682

Demand to time and savings 0.760 0.609 0.824 0.141 6.952

deposits ratio

Deposit growth 2.138 1.718 2.289 -5.117 21.646

Deposit volatility (V ol) 0.581 0.414 0.623 0.080 4.265

Deposit skewness (Skew) 0.091 0.097 0.924 -3.007 3.283

ICRG composite risk score 62.894 62.347 8.634 36.458 84.479

Lending rate (logged) 3.153 3.030 0.453 2.363 4.636

Discount rate/Lending rate 0.690 0.663 0.314 0.116 1.731

GDP (logged) 22.380 22.527 1.341 19.278 26.056

GDP Growth 4.794 4.629 4.163 -10.240 22.7

Liquid Liabilities /GDP 23.473 18.958 12.711 5.143 79.615

Inflation (logged) 2.394 2.399 1.157 -3.269 6.322

Notes: The sample covers 18 SSA countries over the period 1994-2008. The panel in unbalanced and the total number of observations is 194. The countries included in the sample are: Angola, Botswana, Cameroon, Congo(Dem), Congo(Rep), Ethiopia, Gabon, Gambia, Guinea-Bissau, Kenya, Madagascar, Malawi, Mozambique, Nigeria, South Africa, Tanzania, Uganda, Zambia.

substantial variation in our liquidity risk variables. Our data can thus be used to identify the relationship between liquidity risk factors and bank reserve ratio.

1.4 Econometric analysis and results

This section explores the quantitative importance of the precautionary motive in ex-plaining the patterns of bank reserves in 18 SSA countries over the 1994-2008 period. As a first piece of evidence, Figure 1.2 plots bank reserve ratio against our primary measure of deposit volatility (V ol). Countries with the highest levels of deposit volatil-ity are Congo(Rep), Guinea-Bissau and Malawi, and these countries are also those with the highest levels of bank reserve ratios. South-Africa and Botswana are countries with the lowest levels of deposits volatility, and they are also the countries with the lowest levels of bank reserve ratios. In overall, Figure 1.2 shows that, on average, higher levels of deposit volatility are associated with larger bank reserve ratios.13

13_{In graphs unreported here, we find a very similar pattern when we plot average the bank reserve}

ratio against the average ratio of demand to time and saving deposits. In contrast, the linear fits are much less steep with the average deposit growth and the average deposit skewness variables, even

(29)

Figure 1.2: Reserve ratio and deposit volatility Angola Angola Angola Angola Angola Angola Angola Angola Angola Angola Angola Angola Botswana Botswana Botswana Botswana Botswana Botswana Botswana Botswana Botswana Botswana Botswana Botswana Botswana Botswana Botswana Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon Cameroon CongoDem CongoDem CongoDem CongoRep CongoRep CongoRep CongoRep CongoRep CongoRep CongoRep CongoRep CongoRep CongoRep CongoRep CongoRep Ethiopia Ethiopia Gabon Gabon Gabon Gabon Gabon Gabon Gabon Gabon Gabon Gabon Gabon Gabon Gabon Gabon Gambia Gambia Gambia Gambia Gambia Gambia Gambia Gambia Gambia Gambia Gambia Gambia Gambia GuineaBis GuineaBis GuineaBis Kenya Kenya Kenya Kenya Kenya Kenya Madagascar Malawi Malawi Malawi Malawi Malawi Malawi Malawi Malawi Malawi Malawi Malawi Malawi Malawi Malawi Malawi Mozambique Mozambique Mozambique Mozambique Mozambique Mozambique Mozambique Mozambique Mozambique Mozambique

Mozambique NigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeriaNigeria

SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica SouthAfrica Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania Tanzania TanzaniaUgandaUgandaUgandaUgandaUgandaUgandaUgandaUgandaUgandaUgandaUgandaUgandaUgandaUganda

Zambia Zambia Zambia Zambia Zambia Zambia Zambia Zambia Zambia Zambia Zambia Zambia Zambia Zambia 0 10 20 30 40 A ve ra ge re se rve ra tio 0 .5 1 1.5 2 2.5

Average deposit volatility

We now turn to investigate whether this association is robust. We first present the estimation method and, after that, the results.

1.4.1 Estimation method

Our estimation approach involves using the more recent dynamic system-GMM estima-tor due toArellano and Bover(1995) andBlundell and Bond(1998). In fact, to account for the fact that “equilibrium" may not be achieved each time period, the lagged depen-dent variable is included as a regressor. Hence, strict exogeneity no longer holds; and the straightforward ordinary least squares (OLS) and the Fixed-effects OLS estimators would be biased and inconsistent (see Baltagi, 2008). In contrast, the system-GMM estimator allows obtaining efficient estimates while controlling for time-invariant un-observed heterogeneity. Another key advantage of this estimator is that it is useful to mitigate the potential endogeneity of (some of) the independent variables.

In implementing the system-GMM method, we take a number of steps to improve identification as much as possible. First, we alleviate concerns with reverse causality by lagging our explanatory variables one-period, except the liquidity risk variables as they are computed using lagged data.14 _{Second, we choose forward orthogonal deviations} when we discarded the extreme points.

(30)

(FOD) rather than the first-difference transformation as the former has the virtue of preserving sample size in panels with gaps as ours (see Arellano and Bover, 1995). Third, we restrict the set of internal instruments by using only certain lags instead of all available lags for valid instruments, and by invoking the “collapse" option in “xtabond2". Suggested byRoodman (2009) and Wintoki et al. (2012), these techniques allow us to address the issue of “instrument proliferation" that may over-fit endogenous variables in system-GMM estimations; also, they make the tests of over-identifying restrictions more powerful. Hence, we use lagged variables in levels dated t-3 to t-4, and lagged variables in first-differences dated t-2 to t-3 as “internal" instruments.

Finally, we supplement the system GMM “internal" instruments with one “external" instrument: the political subcomponent of the ICRG composite risk index. The validity of this instrument relies on two facts: i) political risk represents a substantial component of the overall ICRG risk rating, and ii) political risk is largely determined “out of the system". The first fact is obvious, and it means that fluctuations in political risk are a significant source of fluctuations in the ICRG composite index. The second fact relies on the presumption, laid out inAcemoglu et al.(2005a), that political institutions may be think as the state variable which determine economic institutions and outcomes, that political institutions are themselves endogenous, determined by economic institutions and outcomes. This means that fluctuations in political institutions in a given year are largely determined by fluctuations in economic outcomes in previous years, and so are unlikely to be correlated with shocks to economic outcomes, such as bank reserves, in the current year.

To assess the validity of instruments, we use and report the Sargan-Hansen test of over-identifying restrictions and the Arellano and Bond (1991) serial auto-correlation tests.

1.4.2 Results

Our system-GMM regression results are reported in Table 1.2. In parentheses are stan-dard errors, which are robust, clustered at country level, to correct for the presence of any pattern of heteroskedasticity and autocorrelation within the panel. Year dum-mies are included in all the regressions and in each case, they are jointly significant statistically, at the 1% level.

monthly data of the same year. As such, they are less prone to reverse causality, since reserve ratios are recorded at the end of the year, and there is no obvious reason - except the strong assumption of rational expectations- why current levels of bank reserves may influence past deposit inflows.

(31)

In column (1) we present our benchmark system-GMM estimates. These results show that both the estimated coefficients for deposit volatility (V ol) and for demand to time and saving deposits ratio are highly significant statistically, at the 1% level. The estimated coefficients for deposit growth trend and for discount rate are also statistically significant, but only marginally, at the 10% level.

In column (2) we repeat the same regression as in column (1), but adding our “external" instrument as described above. This has little effects on the results. In fact, the statistically significant coefficients and their significance levels are the same as in column (1). Only the magnitude of the coefficients are affected. For example, the coefficient of deposit volatility (V ol) increases from 7.709 to 8.176; and the coefficient of the demand to time and saving deposit ratio increases from 5.643 to 6.210. We may interpret these results as consistent with the argument that our initial system-GMM estimates are relatively robust.

In columns (3) to (5), we assess to what extent our results are sensitive to outliers. We should note that for this exercise, we rerun the same specification as in column (2), dropping one country at a time.15 _{For brevity, we report the results for three}

cases which appeared to be the most sensitive; namely South-Africa, Congo(Rep), and Angola, respectively in columns (3), (4) and (5). Worth noting, these three countries are also the most extreme outliers in our sample.16 It may also be worth noting that the three results, from columns (3) to (5), are illustrative enough of the changes we encountered in the sensitivity analysis.

In column (3), when South-Africa is excluded from the sample, the most significant changes in the results are in the coefficient for discount rate and in that for liquid liabilities to GDP ratio (LL/GDP), which become stronger statistically: The statistical significance of the coefficient for discount rate increases from the 10% level to the 5% level; and the coefficient on LL/GDP become marginally significant, at the 10% level.

15_{Alternatively, we check the sensitivity to outliers by winsorizing the liquidity risk variables at the}

1st and the 99th percentiles. The results we find are qualitatively similar to those reported here, even when the liquidity risk variables are winsorized at the 5th and the 95th percentiles, or when all the variables are winsorized.

16_{The cases of South-Africa and Congo (Rep) are obvious from Figure 1.2: South-Africa is the}

country with the lowest levels of both deposit volatility and reserve ratios; Congo (Rep) is the country with the highest levels of both deposit volatility and reserve ratio. The same is true with the ratio of demand to time and saving deposits. With regard to Angola, we have noticed that it is the country with the highest levels of both deposit growth and inflation. In fact, the maximum value of the deposit growth variable in the sample is 21.646, and the 99th percentile is 12.752; these values are from Angola. Moreover, the maximum value of inflation (log of inflation) in the sample is 556.939% (6.322), and the 99th percentile is 418.232% (6.036); these values are also from Angola.

(32)

Table 1.2: Econometric results

Sensitivity to outliers

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

The dependent variable is bank reserve ratio

Lag1 dep. var. 0.610∗∗∗ 0.594∗∗∗ 0.612∗∗∗ 0.648∗∗∗ 0.645∗∗∗ (0.117) (0.108) (0.119) (0.126) (0.086)

Deposit volatility (V ol) 7.709∗∗∗ 8.176∗∗∗ 8.109∗∗∗ 7.711∗∗∗ 8.053∗∗∗ (1.859) (2.030) (1.682) (2.871) (2.143)

Demand to time and savings 5.643∗∗∗ 6.210∗∗∗ 5.319∗∗∗ 9.034∗ 5.253∗∗∗ deposits ratio (lagged) (1.840) (1.429) (1.331) (5.119) (1.761)

Deposit growth -0.871∗ -0.845∗ -0.869∗ -0.476 -0.397 (0.516) (0.486) (0.457) (0.470) (0.754)

Deposit skewness (Skew) 0.907 0.203 0.131 -2.115 0.114 (2.540) (2.012) (2.467) (2.769) (2.187)

I(Skew > 0) 5.119 5.157 4.688 1.370 5.213

(4.255) (4.259) (4.205) (2.816) (4.273)

Skew *I(Skew > 0) -5.198 -3.785 -2.621 0.118 -2.252 (5.520) (4.199) (4.010) (4.880) (4.143)

ICRG composite risk (lagged) -0.180 -0.054 -0.016 0.094 -0.159 (0.285) (0.184) (0.224) (0.144) (0.203)

Lending rate (lagged) -14.015 -27.909 -36.914 -34.041 -80.321 (56.361) (53.876) (55.476) (24.795) (55.689)

Lending rate2 (lagged) 2.341 4.327 5.572 5.186 12.191 (7.893) (7.573) (7.660) (3.478) (8.294)

Discount rate (lagged) 16.955∗ 17.524∗ 16.974∗∗ 6.607 13.088∗ (9.098) (9.272) (8.474) (4.658) (6.845)

GDP (lagged) -1.828 -2.261 -2.870 -1.512 -1.266 (1.540) (1.436) (2.156) (1.380) (1.283)

GDP Growth (lagged) -0.939 -0.838 -0.864 0.130 -1.372 (0.850) (0.736) (0.694) (0.171) (0.892)

Liquid Liabilities/GDP (lagged) -0.253 -0.251 -0.328∗ -0.138 -0.093 (0.165) (0.159) (0.182) (0.137) (0.082) Inflation (lagged) -3.968 -3.724 -3.781 -0.060 -2.697 (2.659) (2.505) (2.339) (1.151) (2.263) Constant 89.620 111.793 142.227 81.303 177.724 (108.415) (108.771) (141.737) (67.613) (114.032) (to be continued)

(33)

(continued)

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

Number of observations 172 172 158 163 161

Number of time period 15 15 15 15 15

Number of countries 18 17 17 17 17

Sargan test, p-value 0.248 0.241 0.301 0.879 0.176

AR1, p-value 0.046 0.041 0.042 0.032 0.093

AR2, p-value 0.118 0.110 0.117 0.250 0.444

External IV NO YES YES YES YES

Implied cumulative effects

Demand to time and savings 14.469∗∗ 15.295∗∗∗ 13.708∗∗ 25.665 14.797∗∗∗ deposits ratio (lagged)

Deposit volatility (V ol) 17.960∗∗∗ 20.138∗∗∗ 20.899∗∗∗ 21.906∗∗ 22.684∗∗∗

Notes: The estimates are from system-GMM regressions. In parentheses are robust standard errors clustered at country-level. Year dummies are included in all regressions. The sample includes 18 SSA countries over the period 1994 - 2008. The regression in column 2 is the same as in column 1 but with our “external" instruments included. The regressions in columns 3, 4, and 5 are the same as in column 2, but with a sample that excludes South Africa, Congo (Rep), and Angola, respectively. The implied cumulative effects in the bottom lines are computed as follows: marginal effect/(1- coefficient estimate on the lagged dependent variable), and the corresponding p-value are obtained from the non-linear test of significance. ***= Significant at the 1% level; **= Significant at the 5% level; and *= Significant at the 10% level.

In column (4), when Congo (Rep) is excluded, only the demand to time and saving deposit ratio and deposit volatility (V ol) variables display statistical significance at conventional levels, respectively at the 10% level and the 1% level. The coefficient on discount rate has decreased substantially, from 17.724 to 6.607, and has lost its statistical significance. The coefficient on V ol has also decreased but in much smaller proportion, from 8.176 (see column 2) to 7.711. In contrast, the coefficient on the demand to time and saving deposit ratio has become weaker statistically although its magnitude increased by almost 50% (from 6.210 to 9.034).

Finally, in column (5), when Angola is excluded, the only significant change is in the coefficient on the deposit growth variable, which decreases substantially in magnitude, by almost 50%, and looses its (marginal) statistical significance. Both the coefficients on the demand to time and saving deposit ratio and the discount rate variables also decrease in magnitude, but they maintain the same statistical significance as in column (2). With regard to deposit volatility (V ol), the estimated coefficient remains

(34)

statis-tically significant at the 1% level, and its magnitude is virtually unchanged: 8.176 in column (2), and 8.053 in column (5).

We should note in passing that throughout the five columns of Table 1.2, all the sta-tistically significant coefficients are of the expected signs. In addition, the coefficient associated with the lagged dependent variable is always positive and highly statisti-cally significant. Its magnitude, which varies between 0.594 and 0.648, can be seen as evidence of persistence. Finally, all the regressions in Table 1.2 are supported by the diagnostic tests of instrument validity suggested by Arellano and Bond (1991). More specifically, the Sargan and the AR1/AR2 tests, for which the p-values are reported in the table, do not point to significant problem with the instrument sets.17

In overall, the results reported in Table 1.2 show that the estimated coefficient for deposit volatility (V ol) is strongly robust: both its statistical significance and economic magnitude do not change substantially across all regressions, from column (1) to column (5). The estimates suggest that deposit volatility has a positive and highly statistically significant effect on bank reserve ratios. This effect is also economically significant. For example, a one standard deviation reduction in V ol (0.623, see Table 1.1) will reduce the average reserve ratio by approximately 5 percentage points in the same year. This impact persists through the dynamics of reserve ratios, since the estimated coefficient on lagged dependent variable is positive and highly significant statistically. The estimated cumulative effect of a one standard deviation reduction in deposit volatility amounts to a reduction in the average reserve ratio by 11 to 13 percentage points; and this effect passes the statistical significance test at conventional levels (see the bottom line of Table 1.2).

We may also gauge the economic significance of the estimated impact of deposit volatil-ity by comparing two cases taken from our sample, namely Kenya and Malawi. Over the study period, the average value of V ol is much lower in Kenya (0.152) than in Malawi (0.707). Given this difference, and other things being equal, our estimates of the implied cumulative effects predict that the average reserve ratio in Malawi would be higher than that in Kenya by 10 to 12 percentage points. For comparison, the aver-age reserve ratio in Malawi (25.263%) is actually higher than that in Kenya by 14.068 percentage points.

With regard to the other regressors included as controls, particularly the demand to

17_{The Sargan test cannot reject the null of instrument validity; and the serial correlation tests reject}

the null of the absence of first order serial correlation (AR1 test) and not reject the absence of second order serial correlation (AR2 test).

(35)

time and saving deposit ratio, the deposit growth, the discount rate and the LL/GDP variables, empirical evidence of their effects on bank reserve ratios is somewhat fragile. Indeed, across the regressions reported in Table 1.2, we notice some variations in the estimated coefficients of those regressors, either in the statistical significance or in the economic magnitude or in both.

In Table A1 in the appendix, we check the robustness of the above results against alternative econometric techniques. In particular, we repeat the regressions in Table 1.2 using the straightforward OLS and the FE-OLS methods. In spite of the shortcomings of these methods, we find that the overall results are qualitatively similar to those reported here.

1.5 Conclusion

The goal of this essay was to assess empirically to what extent the holding of large amounts of liquid assets by banks in Africa can be attributable to liquidity risk. We construct a time-varying measure of deposit volatility, which we use as our primary approach to capture the exposure of banks to liquidity risk. The empirical analysis is based on a sample of 18 SSA countries over the 1994-2008 period. The estimation strategy involves using dynamic panel data techniques and controlling for potential outliers and endogeneity biases.

We find a strong positive association between higher levels of deposit volatility and large bank reserve ratios. This result implies that deposit volatility in Africa significantly reduces the fraction of deposits that banks can channel to borrowers, since banks in-vest in unproductive liquid assets to self-insure against unpredictable liquidity demand from depositors. Future research may need to assess to what extent this liquidity risk management strategy affects the availability of credit.

(36)

(37)

Chapter 2 Legal Rights, Information Sharing,

and Private Credit: New

Cross-Country Evidence

Abstract

Using improved measures and recent data from Doing Business, this essay reex-amines the effects of legal systems and information-sharing on private credit. The results indicate that stronger legal rights, better contract enforceability and bet-ter information sharing are associated with higher private credit to GDP ratios across countries. These effects are significant even when the sample is restricted to include either developing countries only or poor countries only, but the ef-fects of both legal rights and enforcement are stronger the richer the countries. In overall, this essay may be viewed as enhancing the robustness as well as the generalizability of earlier evidence aimed at establishing a link between legal and information-sharing institutions on one hand, and the size of credit market on other hand.

JEL classifications: G21, G28, K22.

Keywords: Legal rights, contract enforceability, credit information-sharing, private credit.

(38)

2.1 Introduction

Over the past two decades, a growing literature has investigated the role of legal and information-sharing institutions in fostering credit market expansion (e.g.,Pagano and Jappelli,1993;La Porta et al.,1997;Levine,1998;Jappelli and Pagano,2002;Jappelli et al.,2005;Djankov et al.,2007;Qian and Strahan,2007;Bae and Goyal,2009;Acharya et al., 2011a). While a consensus seems to have emerged that better information-sharing enhances financial intermediation, evidence is mixed with regard to the benefits of stronger legal systems.

Indeed, in a pioneering study of 49 countries,La Porta et al.(1997) find that countries with poorer investor protections, measured by both the character of legal rules and the quality of law enforcement, have smaller capital markets. Levine (1998) complements this study by establishing that stronger legal systems are associated with higher share of private credit to GDP in cross-country regressions. In the same spirit, Djankov et al.(2007) examine a larger sample of 129 countries over the period 1978-2003, and find that the effect of strong legal systems is relatively more important in richer countries. At the same time, Acemoglu and Johnson (2005b) contend that “contracting institu-tions", that is, rules and regulations governing transactions between private parties, such as a debtor and a creditor, have no impact on the private credit to GDP ratio. Other cross-country studies, which use micro-level data, suggest that legal rules and enforcement are not equally important in determining the amounts of bank loans. For example, Qian and Strahan (2007) suggest that it is creditor rights, not the enforce-ability of contracts, that matter. In contrast, Bae and Goyal (2009) find that it is enforcement, not the merely existence of laws, that matters. More recently, Acharya et al. (2011a) find that corporate leverage declines when creditor rights are stronger, thereby contradicting the argument that stronger creditor rights brings about a greater level of lending.1

Given these seemingly conflicting evidences, one may be concerned whether building strong legal systems is likely to be an effective avenue to foster credit market develop-ment, especially in less developed countries. With regard to this concern, some cross-country analyses have suggested that information sharing may substitute for stronger legal systems (Jappelli and Pagano, 2002; Djankov et al., 2007).

This essay adds to the literature by using a more recent dataset to revisit the

(39)

country association between the legal rights, information-sharing and private credit. The data are from Doing Business which, starting in 2003, produces quantitative and cross-country comparable indicators of business regulations, including measures of legal rights, contract enforceability, and credit information sharing.2 Not only are these measures available for a more recent period, they also differ in several respects from the measures used in previous literature.

First, the Doing Business “strength of legal rights index" measures the degree to which collateral and bankruptcy laws protect the rights of borrowers and lenders; previous measures of legal rights (see, La Porta et al., 1997; Levine, 1998; Djankov et al.,2007) focus on protecting the rights of creditors only. The difference is critical, at least as far as lending is concerned, since a weak protection of the rights of borrowers may lead them to self-select out of the credit markets.3

Second, Doing Business has three enforcing contract indicators: the time, cost and number of procedures involved in resolving commercial dispute through courts. These indicators have first been developed in Djankov et al. (2003) and the original data, i.e., from 2003, have been used in the literature (e.g., Acemoglu and Johnson, 2005b; Djankov et al.,2007). Doing Business made small changes to the methodology but also, and more importantly, it corrects measurement errors and explicitly supersedes the data of earlier reports over the years. To show the aggregate impact of the improvements since the original 2004 Doing Business report, Spamann (2010a) notes the low rank correlations between the original Doing Business data and data for the same year (2003) from the 2009 edition: 0.33 for procedures, 0.53 for time, and 0.54 for cost.4

Finally, the Doing Business measures of credit information sharing take into account the coverage, scope and accessibility of credit information available through either a public credit registry or a private credit bureau; in general, previous cross-country studies only considered a dummy equals one if either a public or private bureau operates in the country, and zero otherwise.

With these new data, our empirical results strengthen in some ways the main

conclu-2_{For more details, see:} _{http://www.doingbusiness.org/methodology}_.

3_Indeed, _Vig ₍₂₀₁₃_{) documents how a recent regulatory change that strengthened the rights of}

secured creditors in India impose an extra-cost on firms - “the threat of being prematurely liquidated"-, which led to a reduction in the use of secured debt as well as in total leverage. Similar findings are presented in Acharya et al.(2011a).

4_{It may be interesting to note a recent result obtained by}_Spamann₍_2010b_{). He corrected a famous}

index in the law and finance literature, namely the “anti-director right index" developed byLa Porta et al.(1997), and found that the correlation between the corrected and original index is 0.53. He also found that the corrected index fails to support many widely influential claims.

(40)

sions reported in Djankov et al. (2007). More specifically, we find that stronger legal rights, better enforcement systems and better information-sharing are associated with higher private credit to GDP ratios across countries. These effects are statistically sig-nificant even when the sample is restricted to include either developing countries only or poor countries only, but the effects of both legal rights and enforcement are stronger the richer the countries.

It is however important to note that, in contrast to Djankov et al. (2007), we did not find a statistically significant association between the number of days to enforce contracts and private credit to GDP.5 _{In fact, of the three alternative measures of}

contract enforceability we have considered (number of procedures, number of days, and cost to enforce contracts), only the number of procedures to enforce contracts shows a strong and consistent association with private credit to GDP.

Overall, this essay may be viewed as enhancing the robustness as well as the general-izability of earlier evidence aimed at establishing a link between legal and information-sharing institutions on one hand, and the size of the credit markets on other hand. Hence, regarding the recent studies which, by using micro-level data, arrived at the conclusion that loan size is not systematically related to differences in creditor rights (e.g.,Bae and Goyal,2009) or in contract enforceability (e.g.,Qian and Strahan,2007), we can argue that strengthening the legal system affects aggregate lending in large measure through the extensive margin, i.e., by lending to borrowers that were previ-ously out of the credit market. Further, it may be important to stress that efforts to strengthen legal rights, and their enforcement, should consider both creditors’ and borrowers’ rights.

The remainder of this essay is organized as follows: the next section describes the data; section 3 presents and discusses the empirical results, and section 4 concludes.

2.2 Data

We start by gathering raw data on legal rights and credit information sharing for every economy covered by Doing Business since 2004. Data on private credit are from the latest revised version (2012) of the “Financial Development and Structure Dataset" (see Čihák et al., 2012). The private credit variable corresponds to the value of credit provided by deposit money banks to the private sector (in % of GDP). Data on the

5_{However, as explained later, this is probably due to the fact that, in Doing Business data, the}

(41)

control variables are from the World Development Indicators. All these variables and their sources are presented in Table 2.1.

By merging our three data sources and by excluding countries with population less than 1.5 million,6 _{we are left with a sample of 143 countries. To ensure that results are}

not driven by single cycles or exceptional trends, we average data over five years, from 2006 to 2010. We replicate the regressions over longer horizons (that is, over six- and seven-years) as well as over different three-years (e.g., 2004-2006, and 2008-2010) and four-years periods; the broad conclusions we report do not change.

Of greater policy interest are developing countries and poorer countries, because they have both less bank lending and weaker legal and institutional environment.7 Hence, in addition to the full sample of 134 countries, we consider the subsample of developing countries (109 countries), as well as the subsample of poor countries (74 countries).8 Over the five-year period we have considered, 8 countries do not have data on the private credit variable, and one country does not have data on the inflation variable. This leaves us with a maximum of 134 countries for the analysis of private credit, of which 100 developing countries, and 68 poor countries. The full sample is restricted to 106 countries in the analysis with the cost of enforcing contracts variable, since data on this later variable are missing for 28 high-income countries. Table 2.2 reports some descriptive statistics of the variables used in the analysis.

In panel A, Table 2.2, it can be seen that the private credit measure varies substantially across countries, from 3% to 207%. Much of this variation reflects differences in the level of economic development, as can be seen by comparing the mean and median of private credit to GDP across the three panels of the table. The mean (median) is 49% (32%) in the full sample, 29% (23%) in the subsample of developing countries, and 24% (19%) in the subsample of poor countries.

Likewise, the patterns of cross-country variations in legal rights, enforcement, and infor-mation sharing reflect differences in the level of economic development. In fact, moving from the top to the bottom panel of Table 2.2, we see that both the mean and the

6_{Credit markets in small states are often more volatile and more sensitive to international}

fluctu-ations.

7_{For instance, the average private credit is 106 percent in the subsample of developed countries.}

In the subsample of developing countries it is only 29 percent. Also, the average strength of legal right index in the subsample of developed countries is 7. In the subsample of developing countries it is 4.8.

8_{As in}_{Djankov et al.}₍₂₀₀₇_{), poor countries are those having per capita income below the median}

value of the full sample. Developing countries are low and middle income economies according to the World Bank classification.