Financial Liberalization and Financial Instability
Dirk Bezemer*, Silke Bumann,** and Robert Lensink***
Abstract
The impact of financial liberalization on financial instability is theoretically
ambiguous. By including financial liberalization policies in a standard model of the
loan market with asymmetric information, this article identifies the entry of risky
entrepreneurs and decreased borrowing costs as two channels by which financial
instability exerts an impact. A new data set, covering 85 countries during 2000–2009,
provides proxies for the financial instability of impaired bank loans. Using an existing
measure of financial liberalization, this investigation finds that more liberalized
countries experienced a steeper increase in impaired loans after the 2008 global
financial crisis. These findings have several research and policy implications.
keywords: financial liberalization, financial instability, model, new data
JEL codes: F62, F65, G01, G15, G21
*Dirk Bezemer is a member of the Faculty of Economics and Business, University of
Groningen.
**Silke Bumann is a postdoctoral researcher at the Max Planck Institute for Evolutionary
Biology, August-Thienemann-Straße 2, 24306 Plön, Germany. She can be contacted at
[email protected].
*** Corresponding author: Robert Lensink. E-mail: [email protected]; phone:
+31503633712; address: Faculty of Economics and Business, University of Groningen;
PO box 800, 9700 AV, Groningen, The Netherlands. Robert Lensink is also affiliated
with the Development Economics Group, Wageningen University.
Funding: This project is part of the DFID/ESRC Growth Programme ES/J009067/1
“Politics, Finance and Growth.” Financial support is gratefully acknowledged.
1
Introduction
Financial instability refers to firms’ inability to repay bank loans, and empirical
literature reports both positive and negative impacts of financial liberalization on
financial instability. With this article, we propose a model to identify the channels by
which financial liberalization affects financial instability and understand why the overall
impact is ambiguous. Building on work by De Meza and Webb (1987), we consider two
channels: Financial liberalization decreases the cost of borrowing, making it easier to
repay loans, which reduces financial instability, but it also encourages the entry of risky
borrowers, which increases financial instability.
By integrating these two channels, we introduce a model that also includes an
exogenous shock that decreases firms’ returns and makes over-borrowing more
problematic. That is, the shock unambiguously increases the probability of financial
instability, especially in countries that have liberalized their financial sector. With our
empirical analysis, we accordingly test for the impact of financial liberalization on
financial instability both with and without an exogenous shock, namely, the 2008–2009
global financial and economic crisis for economies other than the United States. The
collapse of Lehman Brothers in September 2008 caused a crisis in the global financial
system that was transmitted, through trade and financial channels, to other economies
with a lag. For example, a 2009 global trade collapse plausibly decreased firms’ returns.
Using a new, cross-country data set related to financial instability, we determine the
share of impaired loans among all bank loans, as a proxy of financial instability.
The estimation results for a sample of 85 economies show no significant relationship
between financial liberalization and financial instability, in line with mixed empirical
findings in prior literature and the theoretical ambiguity of our model. After the 2009
shock though, we find a sharper rise in impaired loans in countries with more financial
liberalization, consistent with our model.
The notion that financial liberalization leads to more financial instability also is
prevalent in financial restraint literature. Financial liberalization intensifies competition,
leading to a decline in the franchise value of banks. Banks may respond by accepting
more risk exposure to maintain their profits. For example, reduced margins might
encourage banks to economize on their screening and monitoring efforts. According to
Dell’Arricia and Marquez (2006), financial liberalization reduces screening efforts,
2
raising project risk and vulnerability to shocks. Banks also may be willing to opt for a
gambling strategy in their loan allocations, by downplaying risk relative to profit.
Although financial liberalization alleviates credit constraints that are due to capital
market imperfections, it may do so at the cost of increased financial instability.
Lorenzoni (2008) notes that competitive financial contracts result in excessive
borrowing ex ante and excessive volatility ex post. His theoretical model thus indicates
that if firms borrow from banks, “[a]ggregate volatility and some degree of financial
fragility are unavoidable … and over-borrowing is a possibility. Therefore, the presence
and the severity of over-borrowing in specific episodes become an empirical issue”
(Lorenzoni, 2008: 826). We also find evidence of fragility (the possibility of financial
instability); we exploit a specific episode (the 2009 economic shock) to identify the
financial instability effect.
Over-borrowing after liberalization often occurs through capital accounts, such that
financial liberalization leads to increased capital inflows, followed by a domestic credit
boom. The likelihood of an over-borrowing scenario differs across economies though.
The threat of crashes seems greatest in emerging markets, especially if the capital
account opens first (Kaminsky and Schmukler, 2003). In Mendoza and Quadrini’s
(2009) open-economy model, financial liberalization leads to a sharp rise in net credit
(see also Mendoza and Terrones, 2008). According to the 2011 IMF World Economic
Outlook survey of 19 advanced and 28 emerging economies over 1960–2011, financial
inflows systematically precede credit booms. Magud et al. (2012) examine 25 emerging
economies and determine that large capital inflows and less flexible exchange rate
regimes significantly increase domestic credit. Calderón and Kubota (2012) instead
examine gross flows and find that gross private capital inflows significantly affect the
probability of domestic credit booms. By estimating impulse response functions of
domestic credit to net capital inflow surges, Furceri et al. (2012) report that shocks in net
debt inflows have the greatest positive effect on credit creation. Moreover, Lane and
McQuade (2014) establish that domestic credit growth relates strongly to net debt
inflows, in both a sample of European countries and an extended sample of 54 advanced
and emerging economies during 1993–2008. Finally, though financially liberalized
economies grow faster than non-financially liberalized ones, they also experience more
crises and are exposed to more severe output contractions during financial crises
3
(Rancière et al., 2006), perhaps because the higher debt levels relative to gross domestic
product (GDP), as caused by financial liberalization and capital inflows, increase the
probability of defaults (Checchetti and Kharroubi, 2012).
Taken together, this prior literature suggests that financial liberalization increases
financial instability. But other studies argue for the opposite effect and suggest that
financial liberalization may promote financial stability through several channels. In
particular, market competition puts pressure on banks to reduce their lending rates.
Therefore, financing costs and constraints decrease for firms, which should reduce their
bankruptcy risk and increase their risk-taking incentives (Boyd and De Nicoló, 2005).
Financing costs also may fall in the wake of capital account liberalization, due to banks’
enhanced opportunities for risk diversification (Bekaert and Harvey, 2000; Henry,
2000; Stulz, 1999). Empirical research offers mixed results about the impact of bank
competition on financial stability though (e.g., Beck et al., 2013; Berger et al., 2009;
Tabak et al., 2012).
Financial liberalization is beneficial if it creates financial development. In the wake of
banking sector liberalization, banks seek to become more efficient by reducing their
overhead costs, improving overall bank and risk management, and offering new financial
instruments and services. Similarly, if domestic markets open to foreign competition,
they can import better bank and risk management techniques and new financial
instruments and services. These developments promote better credit access and credit
allocation (Abiad et al., 2008). More efficient credit allocation contributes to financial
stability, if firms with higher marginal returns borrow more. As Galindo et al. (2007)
show, the share of investment in firms with higher marginal returns to capital has
increased in the wake of financial liberalization. Abiad et al. (2008) also provide
evidence that financial liberalization equalizes the marginal returns of capital across
firms. In a similar vein, Cho (1988) finds that financial liberalization in Korea was
associated with decreased variation in borrowing costs. These channels ultimately
should help improve the efficiency of the financial sector, contributing to higher returns
on investment and thus higher rates of economic growth. Ultimately, higher growth
promotes financial stability, because greater profit opportunities render firms more
robust and stable, lowering the risk of loan defaults.
4
We present our proposed theoretical model of financial liberalization next, which
leads in to our empirical analysis. Finally, we conclude with a summary of our findings
and discussions of limitations, avenues for further research, and policy implications.
Illustrative Model
Using De Meza and Webb (1987) as a foundation, we provide an illustrative, simple
framework for assessing the link between financial liberalization and financial
instability. In this model, entrepreneurs borrow money from banks to launch projects.
We assume the project returns are the same for all entrepreneurs, but they differ in their
success probability. Financial liberalization attracts riskier entrepreneurs with low
project success probabilities into the pool of borrowing entrepreneurs, which lowers the
average project success probability. However, a decline in average project success
probability is not necessarily problematic, because financial liberalization also leads to a
decline in borrowing cost, so more borrowing by risky entrepreneurs might be socially
efficient. Overall then, financial liberalization may increase or decrease the degree of
over-borrowing, and therefore, it has an ambiguous effect on financial stability.
Entrepreneurs
We consider a continuum of risk-neutral entrepreneurs, each of whom undertakes a
risky project. A project requires one unit of labor and one unit of investment.
Entrepreneurs do not possess wealth, so they need to borrow from banks to undertake
their projects. Each loan earns bank interest at the lending rate 𝑟. Projects yield returns
of 𝑅 if successful and 0 if not. Entrepreneurs differ in their probability of success 𝑝𝑖 , and
we assume that 𝑝𝑖 is uniformly distributed between 0 and 1: Higher 𝑝𝑖 implies a safer
project.
An entrepreneur launches a project if the return on investment exceeds a reservation
payoff 𝜇, equal to the opportunity cost of labor. Thus an entrepreneur will invest if
𝑝𝑖 (𝑅 − 𝑟) ≥ 𝜇.
(1)
For a marginal entrepreneur for whom 𝑝𝑖 ≡ 𝑝𝑡 , this expression holds with the equality:
𝑝𝑡 (𝑅 − 𝑟) = 𝜇.
(2)
5
Given Equation 2, the success probability of the marginal entrepreneur’s project equals
𝜇
𝑝𝑡 = 𝑅−𝑟.
In line with De Meza and Webb (1987), the marginal entrepreneur has the lowest 𝑝𝑖
in the group of investing entrepreneurs. That is, for a group of investing entrepreneurs,
(3)
𝑝𝑡 ≤ 𝑝𝑖 ≤ 1.
Asymmetric information also exists, because banks cannot observe the success
probabilities of projects. However, banks are assumed to know the probability
distribution and average project success probability 𝜋 of entrepreneurs launching
𝜇
investment projects. Given Equation 2, all entrepreneurs with 𝑝 ≥ 𝑅−𝑟 launch projects.
Thus, we can write 𝜋 as a function of the lending rate 𝑟:
(4)
𝜇
𝜋(𝑟) = 𝐸 (𝑝|𝑝 ≥ 𝑅−𝑟).
Because 𝑝 is uniformly distributed, 𝑝~𝑈[0,1], the conditional expectation in Equation 4
can be written as:
1
1
2
2 𝑅−𝑟
𝜋(𝑟) = + (
𝜇
(5)
) if 𝑅 > 𝑟.
Accordingly, Equation 5 determines how a change in the lending rate 𝑟 affects the
average project success probability 𝜋 of entrepreneurs launching projects.1 A rise in 𝑟
increases 𝜋.
Banking sector
We assume that banks are risk neutral and operate in a competitive market. They lend to
entrepreneurs at the market rate 𝑟 and take deposits, paying a risk-free rate of interest 𝜌.
In an economy with capital controls, 𝜌 should be greater than it would be in an open
economy with free capital flows (in which case 𝜌 equals the foreign cost of borrowing).
Furthermore, we assume that banks do not observe the success probability of
entrepreneurs’ projects 𝑝 but only know the underlying distribution. As a consequence,
banks set a pooling lending rate of 𝑟.
1
𝜇
If 𝑅 < 𝑟, the average project success probability equals 𝜋(𝑟) = ( ). We do not consider this case here,
2 𝑅−𝑟
because it implies that projects yield negative returns, which means that nobody is willing to invest.
1
6
Banks must hold a fixed percentage 𝑎 of deposits in the form of required reserves at
the central bank. The central bank does not pay any interest on required reserves, which
accordingly operate as a tax on banks. Due to these required reserves, banks only grant a
fraction (1 − 𝑎) of loans for each unit of deposits. For a one-unit increase in loans,
deposits must rise by
1
(1−𝑎)
≡ 𝛽, where 1 ≤ 𝛽 < ∞. Note that an increase in 𝛽 is
associated with an increase in required reserves, whereas a decrease in 𝛽 corresponds to
lowered reserve requirements—one of the financial liberalization policies we consider.
Due to the required reserves, the effective opportunity cost of capital becomes 𝛽𝜌 (where
𝜌 denotes the risk-free rate of interest). In a country without reserve requirements, 𝛽 =
1, and the opportunity cost of capital is 𝜌.
Because banks operate in a competitive environment, they set the lending rate such
that profits are 0. That is, banks grant loans if their expected payment on a project loan
𝜋𝑟 equals the risk-free payment 𝛽𝜌:
𝜋𝑟 = 𝛽𝜌.
(6)
Equation 6 thus specifies the locus of project success probabilities and lending rates that
allow the bank to earn the required rate of return.
Market equilibrium
Equations 5 and 6 determine the equilibrium lending rate 𝑟 ∗ . Given the lending rate,
𝜇
investors decide whether to borrow from banks (i.e., all individuals with 𝑝𝑖 ≥ 𝑅−𝑟
borrow). On the basis of investors’ choices, it is possible to determine an equilibrium
value for average project success probability 𝜋 ∗ . The calculations for 𝑟 ∗ and 𝜋 ∗ are in
Appendix A; here, we provide the graphical analysis in Figure 1.
According to Equation 5, represented by the 𝐵𝐵 curve in Figure 1, an increase in 𝑟 is
associated with a rise in 𝜋. As 𝑟 approaches R, 𝜋 becomes very large. We study situations
for which 𝜋 < 1.2 A lower bound on the average project success probability indicates that
𝜋 can never be below 1⁄2, which would imply that everybody borrows and launches
projects (see Equation 5).
Note that if π>1, no entrepreneur borrows or launches an investment project. The loan market breaks
down. Cases with 𝑟 > 𝑅 are not of interest either, because nobody undertakes projects with negative
returns. In the feasible area, the BB curve can only take positive values.
2
7
Equation 6 describes the 𝐿𝐿 curve in Figure 1, which decreases monotonically in the
lending rate 𝑟, and the risk-free payment 𝛽𝜌 determines the lower bound on 𝑟. Below
this threshold, banks do not lend. If 𝑟 were lower than 𝛽𝜌, banks’ expected payment
would be smaller than 𝜋𝑟, so banks would be earning losses. Moreover, 𝜋 ≥ 1⁄2 implies
that 𝑟 ≤ 2𝛽𝜌, which is an upper bound on 𝑟: Any equilibrium should satisfy 𝛽𝜌 ≤ 𝑟 ≤
2𝛽𝜌. The intersection of the two curves in Equations 5 and 6 determines the equilibrium
values of 𝑟 ∗ and 𝜋 ∗ .
<Figure 1 about here>
Over-borrowing
To determine whether market equilibrium is characterized by over- or under-borrowing,
we compare the market equilibrium with the socially efficient equilibrium in which all
projects with a positive net present value are undertaken. Socially efficient projects have
an expected return above the opportunity cost of labor 𝜇 plus the effective opportunity
cost of capital 𝛽𝜌. Thus, in a socially efficient equilibrium, only projects with 𝑝𝑖 𝑅 ≥ 𝛽𝜌 +
𝜇 are undertaken. The condition for social efficiency is 𝑝𝑖 > 𝑝𝑒 =
𝛽𝜌+𝜇
𝑅
. If over-borrowing
occurs, the marginal borrower has a success probability of 𝑝𝑡 < 𝑝𝑒 . In this case, some
entrepreneurs who borrow and invest should not in the social optimum.
To prove that the market equilibrium is characterized by over-borrowing, we start
with the assumption that there is no over-borrowing, so 𝑝𝑡 ≥ 𝑝𝑒 . The marginal investor
then needs to satisfy the social optimality condition:
𝑝𝑡 𝑅 ≥ 𝛽𝜌 + 𝜇.
(7)
We substitute the participation constraint 𝑝𝑡 (𝑅 − 𝑟) = 𝜇 (Equation 2) to obtain:
𝑝𝑡 𝑅 ≥ 𝛽𝜌 + 𝑝𝑡 (𝑅 − 𝑟),
(8)
which implies that 𝑝𝑡 𝑟 ≥ 𝛽𝜌. In other words, banks earn profits on loans to the marginal
entrepreneur. The marginal entrepreneur is the most risky one in the pool of
entrepreneurs who launch investment projects, so this outcome is incompatible with the
zero-profit condition of banks (Equation 6). As in De Meza and Webb (1987), the market
equilibrium is characterized by over-borrowing: Too many risky entrepreneurs borrow
and launch investment projects. Over-borrowing also implies that the average project
8
1
1
success probability 𝜋 (𝜋(𝑟) = 2 + 2 𝑝𝑡 ) is too low, compared with the average project
1
1
success probability in the socially optimal equilibrium 𝜋 𝑒 (= 2 + 2 𝑝𝑒 ), because 𝑝𝑡 < 𝑝𝑒 .
Over-borrowing means that 𝜋 𝑒 – 𝜋 > 0, as we show in Figure 1.
Impact of financial liberalization on the equilibrium and over-borrowing
We study the impacts of two financial liberalization measures: a decrease in reserve
requirements and a reduction in capital controls. Liberalizing reserve requirements
means that banks hold a smaller fraction of deposits, in the form of required reserves at
the central bank. In the model, we capture this aspect by the decrease in 𝛽. Capital
account liberalization enables entrepreneurs to obtain cheaper foreign funds, leading to
a decrease in 𝜌. The lower 𝛽 and/or 𝜌 implies lower required returns for banks. In this
analysis, we focus on the effect of 𝜌, but the analysis of a change in 𝛽 would be similar.
As we show in Figure 2, financial liberalization entails a downward shift of the 𝐿𝐿
curve, which prompts a decrease in the equilibrium cost of borrowing, from 𝑟 ∗ to 𝑟 ∗′ .
Therefore, riskier entrepreneurs borrow and launch projects, and the equilibrium project
success probability declines from 𝜋 ∗ to 𝜋 ∗′ .
<Figure 2 about here>
Next, we investigate whether financial liberalization increases over-borrowing,
which is characterized by the difference between the socially optimal average project
success probability 𝜋 𝑒 and the average project success probability of the pool of
borrowing entrepreneurs 𝜋. In examining how the difference 𝜋 𝑒 – 𝜋 is affected by
financial liberalization, we denote project success probability after financial
liberalization by 𝜋 ′ . In Appendix B, we derive
𝑑(𝜋 𝑒 −𝜋)
𝑑𝜌
, but here, we concentrate on a
graphical analysis. As Figure 2 shows, financial liberalization, in addition to the shift in
the 𝐿𝐿 curve, entails a downward shift in the horizontal line 𝜋 𝑒 , because the socially
optimum level of the probability of success decreases for each lending rate as costs
decrease. Our main interest lies in assessing whether over-borrowing increases; that is,
whether 𝜋 𝑒′ − 𝜋 ′ > 𝜋 𝑒 − 𝜋. It turns out that the outcome depends on the shape of the
9
curves and thus is unclear ex ante. In Appendix B, we show that the sign of the derivative
𝑑(𝜋 𝑒 −𝜋)
𝑑𝜌
is ambiguous.
Impact of a shock on over-borrowing
We assume that an exogenous shock decreases project returns R. For example, if trade
collapses due to some exogenous event, the average return of domestic firms will
decrease. An exogenous shock has two effects. First, for smaller R, 𝜋(r) is larger
(Equation 5). The BB curve becomes steeper and crosses the LL curve at a greater value
of 𝜋 (Figure 3). Second, it increases the socially optimal probability of success
(horizontal curve increases), because a higher probability of success is needed to make
projects socially efficient when the return decreases. In equilibrium, the difference
𝜋 𝑒′ − 𝜋 ∗′ will be greater, due to the shock. It follows that in a country with greater
financial liberalization, the probability of over-borrowing increases in the event of an
exogenous shock that reduces R.
<Figures 3 and 4 about here>
From modeling to empirical testing
The preceding analysis predicts the impact of financial liberalization on over-borrowing
and therefore on financial instability. However, it also suggests that the extent to which
financial instability increases is a matter of empirical investigation. Borrowing by
entrepreneurs with lower success probabilities can be socially desirable, so overborrowing and financial instability do not necessarily rise with financial liberalization.
We therefore expect an insignificant coefficient for financial liberalization in regressions
of financial instability. However, a negative shock to firm returns also tightens the link
between financial liberalization and financial instability, such that we expect a positive
coefficient on financial liberalization in regressions of financial instability, conditioned
on a shock. In the estimations, we assume that the financial crisis in 2009 induced an
exogenous shock that decreased firm returns R. We rely on the share of defaults in all
bank loans as a proxy for financial instability.
Data and Analysis
Data
10
We use bank-level data indicating the financial instability for 82 countries (see Appendix
E) in 2000–09. We measure financial instability as the share of gross loans3 that are
past due by 90 days or more (i.e., impaired loans), which we refer to as the impaired
loans ratio (ILR). The data cover 23,287 banks in 2000–2009, spanning a wide variety
of types (e.g., commercial banks, investment banks, savings banks, real estate and
mortgage banks, cooperative and Islamic banks). They exclude banks that are rare in
most countries, such as securities firms and microfinance institutions, so the sample
composition is generally comparable across countries (see Appendix D). A country-level
weighted average ILR uses the banks’ total assets as weights. Adrianova et al. (2015)
provide a detailed description of these data.
To capture financial liberalization, we use Abiad et al.’s (2010) measure of
financial reform. This index takes values from 0 (no reform) to 21 (fully reformed),
depending on the policy changes with respect to credit controls, reserve requirements,
interest rate controls, entry barriers, state ownership, policies on securities markets,
banking regulations, and restrictions on capital accounts. The data cover 91 economies
during 1973–2005 (for details, see Abiad et al., 2010). Furthermore, we control for
inflation, the size of the financial sector (credit to the private sector as a percentage of
GDP), constant dollar GDP per capita levels, and size of government expenditures as a
percentage of GDP. Appendix C contains the variable definitions and sources; Table 1
presents the descriptive statistics.
<Table 1 about here>
Specifications, Estimator, and Results
We hypothesized that financial liberalization causes financial instability after a shock to
returns occurs. In this section, we treat the global financial crisis as an external shock to
returns and test whether the effect of financial liberalization depends on this shock. We
take the year 2009 as the time of the shock, because it was the moment of the global
trade collapse—plausibly the most potent channel through which the 2008 Lehman
Brothers collapse and financial crisis reached other developed and emerging economies.
Trade collapsed in 2009 (though capital flows already had dipped in 2008). We include
the treating year 2008 as a shock, as a robustness check. We also include lagged
The value of gross loan equals the loan portfolio plus loss reserves. The total loan portfolio is the sum of
all loan types including residential and other mortgage loans, retail loans, corporate loans and commercial
loans.
3
11
finreform, a variable to reflect the level of financial liberalization, to allow for lagged
responses of impaired loans to financial liberalization. For a shock in 2009, we must lag
financial liberalization by four or more years, because the financial liberalization data
end in 2005. To maximize observations, we thus take a four-year lag, but we also
estimate longer lags as robustness checks.
Our panel estimation thus includes 82 countries. The first model we estimate is:
𝐼𝐿𝑅𝑖,𝑡 = 𝛼0 + 𝛾1 𝑓𝑖𝑛𝑟𝑒𝑓𝑜𝑟𝑚𝑖,𝑡−4 + 𝛾2 (𝑑𝑇2009 × 𝑓𝑖𝑛𝑟𝑒𝑓𝑜𝑟𝑚𝑖,𝑡−4 ) + 𝒙′𝑖,𝑡 𝜷
+ 𝜂𝑖 +𝛿2 𝑑2𝑡 + ⋯ + 𝛿𝑇 𝑑𝑇𝑡 + 𝑢𝑖,𝑡 , with 𝑡 = 1, … ,9,
(9)
where t and i index time and country, respectively. The dependent variable is the
logarithm of the ILR. The variable 𝑓𝑖𝑛𝑟𝑒𝑓𝑜𝑟𝑚𝑖,𝑡−4 denotes the level of financial
liberalization lagged by four periods, to allow for a lagged response of impaired loans to
financial liberalization.4 The variables d2t,…,dTt represent time dummy variables, such
that 𝑑𝑠𝑡 = 1 if 𝑠 = 𝑡, and 0 otherwise. We exclude the first year (2001) from the set of
time dummy variables to avoid multicollinearity. The regressor vector 𝒙′𝑖𝑡 includes the
set of control variables; 𝜂𝑖 are time-invariant country dummy variables; and 𝑢𝑖𝑡 are
identically and independently distributed disturbances. We use the GDP per capita,
government consumption, private credit, and inflation as control variables. The
parameter 𝛾1 captures the direct effect, if any, of financial liberalization on financial
instability. The main parameter of interest is 𝛾2, which measures the extent to which the
relationship between financial liberalization and financial instability changed in 2009
due to the shock.
The next specification includes a lag of the dependent variable. Plausibly, financial
liberalization causes gradual adjustments of bank balance sheets and financial
instability. We also want to account for serial correlation in the disturbances. Using the
same notation, our second empirical model is:
𝐼𝐿𝑅𝑖,𝑡 = 𝛼0 + 𝛼1 𝐼𝐿𝑅𝑖,𝑡−1 + 𝛾1 𝑓𝑖𝑛𝑟𝑒𝑓𝑜𝑟𝑚𝑖,𝑡−4 + 𝛾2 (𝑑𝑇2009 × 𝑓𝑖𝑛𝑟𝑒𝑓𝑜𝑟𝑚𝑖,𝑡−4 )
+ 𝒙′𝑖,𝑡 𝜷 + 𝜂𝑖 +𝛿2 𝑑2𝑡 + ⋯ + 𝛿𝑇 𝑑𝑇𝑡 + 𝑢𝑖,𝑡 , with 𝑡 = 1, … ,9.
(10)
The interpretation of the main parameters of the model, 𝛾1 and 𝛾2, is the same as in
Equation 9, but this dynamic specification produces unbiased standard errors.
4
We lag financial liberalization by four years because we have observations until 2005.
12
Assuming that the decision to liberalize the financial sector is exogenous to the
impaired loans ratio, Equation 9 can be estimated by using the within estimator. In the
dynamic specification in Equation 10, this within estimator no longer yields unbiased
coefficient estimates, because the means of 𝐼𝐿𝑅𝑖,𝑡 and 𝑢𝑖𝑡 contain past, present, and
future values. The average values of the impaired loans ratio ̅̅̅̅̅
𝐼𝐿𝑅 and the error term 𝑢̅
thus are correlated with 𝑢𝑖𝑡 and 𝐼𝐿𝑅𝑖,𝑡−1 , respectively (Nickell, 1981). We take the first
differences of the original model to remove the individual effect 𝜂𝑖 , and we employ
Anderson and Hsiao’s (1982) estimator. However, the differenced lagged dependent
variable and the disturbance term are still correlated: The latter contains 𝑢𝑖,𝑡 , but the
former contains 𝐼𝐿𝑅𝑖,𝑡−1 . Anderson and Hsiao (1982) propose using either the lagged
difference or the lagged level of the variable as an instrument for the differenced lagged
endogenous regressor, because both types of instruments would be uncorrelated with
the differenced disturbance term. As Arellano (1989) points out though, the estimator
based on the level instrument is superior and offers an advantage because fewer
observations are lost. Thus, we employ level instruments in our analysis.
Table 2 contains the estimation results for Equation 9. We do not report coefficients
for year dummies (2001 as the base year), which tend to be significantly negative, to
capture the downward trend in the ILR observed in Figure 5. The coefficient for the
financial reform index is statistically insignificant in all specifications. The coefficient for
the interaction term between the financial reform index and the year 2009 is
significantly positive in columns 2–4. Therefore, banks in more financially liberalized
countries appear less financially stable during the crisis.
<Table 2 about here>
<Figure 5 about here>
Table 3 includes the estimation results for Equation 10. The lagged coefficient on the
ILR is significantly positive throughout, suggesting the need to control for persistence in
this variable. The estimates in Table 3 are subject to Nickell bias, so in Table 4, we
present the results using the instrumental variable (IV) estimator from Anderson and
Hsiao (1981).
<Table 3 about here>
The coefficient for the lagged ILR increases, suggesting a downward bias in the
coefficients for this variable in Table 3. The coefficient for the interaction term between
13
the financial reform index and the year 2009 is again significantly positive in all
specifications. The estimate in column 4 implies that a 10% increase in the financial
liberalization index (in 2005) is associated with an 8% rise in the ILR in 2009.5 For the
average country in the data set, an increase in the financial reform index from 15.5 to 17
leads to an increase in the impaired loans ratio from 0.07 to 0.0756.6 Comparing the
average country in the data set that scores 15.5 on the financial reform index with a
country that is fully liberalized, such that the financial reform index would equal 21, we
determine that the latter has a predicted impaired loans ratio that is 24% greater in the
event of the 2009 shock than the former.7
<Table 4 about here>
To test the robustness of these findings, we first excluded very large observations;
the findings remain similar. In Table 5, we include lags of 4, 5, and 6 years interacted
with the year 2009. The coefficient is always significantly negative. The result for
interacting with the crisis year 2008 is not significant though. Plausibly, the full impact
of the U.S.-originated crisis on other countries was visible only in 2009.
<Table 5 about here>
In Table 6, we also include various bank-level control variables (as country
averages) that might influence the ILF: return on equity, return on assets, the cost-toincome ratio, a measure of bank sector concentration, the interest margin, and a bank’s
international debt. Neither the size nor the significance of the coefficient for the ILR was
substantially affected.
<Table 6 about here>
In summary, financial liberalization leads to a rise in impaired loans following the
crisis shock. This outcome is robust to the inclusion of private credit and other controls
(see columns 3 and 4). In line with prior literature, we find that a larger credit-to-GDP
ratio correlates with a higher ILR, independent of the financial liberalization effect (for
similar findings, see Checchetti and Kharroubi, 2012; Gourinchas and Obstfeld, 2012;
We expect an 8% rise in the impaired loans ratio for a 10% increase in the financial reform index, because
1.1^𝛾2 =1.1^0.84=1.08.
6 For the average country in the data set, the ILR equals 7%, and the financial reform index is 15.5. Thus, a
one-unit increase in the index means that the ILR increases by 0.48 percentage points: 0.06  8 = 0.48.
7 We calculate 1.3^γ =1.3^0.84, because the financial reform index of a fully liberalized countries is 30%
2
greater than that of an average country.
5
14
Mendoza and Terrones, 2008; Schularick and Taylor, 2012). But this effect vanishes
when we include the lagged dependent variable.
Conclusion
We have investigated if and how financial liberalization drives financial instability. One
research stream suggests that financial liberalization, especially capital account
liberalization that increases capital inflows, causes increased leverage. Another stream
cites how greater leverage raises the risk of financial crises and thus increases financial
instability. Thus, financial liberalization appears to drive financial instability. Yet studies
of the lower costs of capital with financial liberalization suggest a greater distance from
default and increased stability. This theoretical ambiguity motivates our research.
With a simple model, we identify two channels through which financial liberalization
can affect the risk of financial instability: entry of risky entrepreneurs and decreased
borrowing costs. In our model, financial instability means over-borrowing relative to the
social optimum. Depending on which channel dominates, more risky borrowing may be
socially optimal or not. The theoretical model helps us identify relevant channels but
offers no conclusive indication of the effect of financial liberalization on financial
instability.
We therefore complemented our theoretical analysis with an empirical assessment of
the impact of financial liberalization on financial instability, conditional on an external
shock. To this end, we used new data on impaired loans as a measure of financial
instability in 85 countries during 2000–2009. Compared with Abiad et al.’s (2010)
financial reform index, which captures financial liberalization, these data suggest that
more financially liberalized countries experience a greater rise in impaired loans during
the financial and economic crisis.
We also tested the liberalization–instability nexus in panel estimations by regressing
the impaired loans ratio on its own lag, four-year lagged values of Abiad et al.’s (2010)
financial reform index, and control variables. The coefficient for the level of financial
reform is statistically insignificant in all specifications. However, the four-year lagged
financial reform index, interacted with a year 2009 dummy variable, yields a
consistently and significantly positive coefficient. That is, banks in more financially
liberalized countries are less stable during a crisis (see also Rancière et al., 2006).
15
Finally, this analysis suggests several avenues for further research. The financial
liberalization transmission channels that we have identified theoretically (i.e., decreasing
cost of borrowing and entry of riskier borrowers) should be assessed separately. Such an
assessment probably requires complementing our country-level analysis with bank-level
data. The theoretical model could be extended by modeling loan defaults. This finding
reinforces some recent concerns about the possible stability implications of capital
account openness (Rey, 2013). It also should prompt researchers and policy makers to
seek a better understanding of these channels of influence.
16
References
Abiad, A., Detragiache, E., Tressel, T., 2010, “A new database of financial reforms”, IMF
Staff Papers, vol. 57, no. 2, pp. 281-302.
Abiad, A., Oomes, N., Ueda, K., 2008. "The quality effect: Does financial liberalization
improve the allocation of capital?" Journal of Development Economics, vol. 87, no.
2, pp. 270-282.
Anderson, T.W., Hsiao, C., 1982, “Formulation and estimation of dynamic models using
panel Data”, Journal of Econometrics, vol. 18, no. 1, pp. 47-82.
Andrianova, S., Baltagi, B., Beck, T., Demetriades, P., Fielding, D., Hall, S.G., Koch, S.,
Lensink, R., Rewilak, J., Rousseau, P., 2015. “A new international database on
financial
fragility”
Working
Paper,
university
of
Leicester.
URL:
http://www.le.ac.uk/economics/research/RePEc/lec/l...
Arellano, M., 1989, "A note on the Anderson-Hsiao estimator for panel data," Economics
Letters, Elsevier, vol. 31, no.4, pp. 337-341.
Bekaert, G. & Harvey, C.R., 2000, "Foreign Speculators and Emerging Equity Markets", Journal
of Finance, vol. 55, no. 2, pp. 565-613.
Beck, T.; De Jonghe, O., Schepens, G., 2013, “Bank competition and stability: Crosscountry heterogeneity”, Journal of Financial Intermediation, vol. 22, no. 2, pp.
218–244.
Berger, A.N.; Klapper, L.F., Turk-Ariss, R., 2009, “Bank competition and financial
stability”, Journal of Financial Services Research, vol. 35, no. 2, pp. 99–118.
Boyd, J.H., De Nicoló, G., 2005, “The theory of bank risk taking and competition
revisited”, Journal of Finance, vol. 60, no. 3, pp. 1329–1343.
Calderón, C., Kubota, M., 2012, “Gross inflows gone wild: gross capital inflows, credit
booms and crises”, Policy Research Working Paper no. 6270, The World Bank.
Checchetti, S., Kharroubi, E., 2012. “Reassessing the impact of finance on growth”, BIS
Working Papers no. 381, Bank for International Settlements.
Cho, Y.J., 1988, “The effect of financial liberalization on the efficiency of credit
allocation”, Journal of Development Economics, vol. 29, no. 1, pp.101–10.
Dell’Ariccia, G., Marquez, R., 2006, “Lending booms and lending standards”, Journal of
Finance, vol. 61, no. 5, pp. 2511–2546.
17
De Meza, D., Webb, D.C., 1987, “Too much investment: A problem of asymmetric
information”, Quarterly Journal of Economics, vol. 102, no. 2, pp. 281-292.
Furceri, D., Guichard, S., Rusticelli E., 2012, “The effect of episodes of large capital
inflows on domestic credit”, North American Journal of Economics and Finance,
vol. 23, no. 3, pp. 325–344.
Galindo, A.; Schiantarelli, F., Weiss, A., 2007. "Does financial liberalization improve the
allocation of investment?: Micro-evidence from developing countries," Journal of
Development Economics, Elsevier, vol. 83, no. 2, pp. 562-587.
Gourinchas, P., Obstfeld, M., 2012, “Stories of the twentieth century for the twenty-first”,
American Economic Journal: Macroeconomics, vol. 4, no. 1, pp. 226–65.
Henry, P. B., 2000, “Stock Market Liberalization, Economic Reform, and Emerging Market
Equity Prices”, Journal of Finance, vol. 55, no. 2, pp. 529-564.
Kaminsky, G.L., Schmukler, S.L., 2003, “Short-run pain, long-run gain: The effects of
financial liberalization”, IMF Working Paper no. 03/34, International Monetary
Fund.
Lane, P.R., McQuade, P., 2014., “Domestic credit growth and international capital
flows”, Scandinavian Journal of Economics, vol. 116, no. 1, pp. 218–252.
Lorenzoni, G., 2008, “Inefficient credit booms”, Review of Economic Studies, vol. 75,
no. 3, pp. 809-833.
Mendoza, E., Quadrini, V., 2010, “Financial globalization, financial crises and
contagion”, Journal of Monetary Economics, vol. 57, no. 1, pp. 24–39.
Mendoza, E.G., Terrones, M.E., 2008, “An anatomy of credit booms: Evidence from
macro aggregates and micro data”, NBER Working Paper no. 14049, National
Bureau of Economic Research.
Magud, N. E., Reinhart, C. M., Vesperoni, E. R., 2012, “Capital inflows, exchange rate
flexibility, and credit booms”, IMF Working Paper no. 41, International Monetary
Fund.
Nickell, S. J., 1981, “Biases in dynamic models with fixed effects”, Econometrica, vol. 49,
no. 6, pp. 1417-26.
Rancière, R., Tornell, A., Westermann, F., 2006, “Decomposing the effects of financial
liberalization: Crises vs. growth”, Journal of Banking & Finance, vol. 30, no. 12, pp.
3331-3348.
18
Rey, Hélène, 2013, “Dilemma not trilemma: The global financial cycle and monetary
policy independence”, Proceedings - Economic Policy Symposium - Jackson Hole,
Federal Reserve Bank of Kansas City, pp. 1-2.
Schularick, M., Taylor, A., 2012, “Credit booms gone bust: Monetary policy, leverage
cycles, and financial crises, 1870-2008”, American Economic Review, vol. 102,
no. 2, pp. 1029-61.
Stulz, R. M., 1999, “Globalization of Equity Markets and the Cost of Capital”, NBER Working
Paper no. 7021, National Bureau of Economic Research.
Tabak, B.M.; Fazio, D.M. and Cajueiro, D.O., 2012, “The relationship between banking
market competition and risk-taking: Do size and capitalization matter?”, Journal of
Banking and Finance, vol. 36, no. 12, pp. 3366–3381.
19
Figure 1: Market equilibrium and over-borrowing
Notes: The intersection of the BB and LL curves determines the
equilibrium values of 𝑟 ∗ and 𝛑∗ . The solid horizontal is the lower
bound on π∗ , and the solid vertical line is the upper bound on 𝑟 ∗ . 𝜋 𝑒
is the socially optimal average project success probability. The
difference between 𝜋 ∗ and 𝜋 𝑒 indicates over-borrowing.
20
Figure 2: Effect of financial liberalization on the
market equilibrium and over-borrowing
Notes: 𝜋 ∗ and 𝜋 ∗′ refer to the equilibrium average project success
probabilities before and after financial liberalization, respectively. 𝑟 ∗
and 𝑟 ∗′ denote the equilibrium interest rates before and after financial
liberalization, respectively. 𝜋 𝑒 is the socially optimal average project
success probability before financial liberalization, whereas 𝜋 𝑒 ′ refers
to the socially optimal average project success probability after
financial liberalization. Over-borrowing equals the difference 𝜋 𝑒 − 𝜋 ∗ .
Financial liberalization can lead to more or less over-borrowing, such
that 𝜋 𝑒 − 𝜋 ∗ can increase or decrease in the wake of financial
liberalization.
21
Figure 3: Effect of a negative shock to R on the
equilibrium
Notes: A shock leads to an anti-clockwise rotation of the BB
curve and an upward shift of the socially optimal average project
success probability 𝜋 𝑒 .
22
Figure 4: Impact of financial liberalization when R
is low
Notes: This figure shows the equilibrium interest rate and average
project success probability before and after financial liberalization,
(𝜋 ∗ , 𝑟 ∗ ) and (𝜋 ∗′ , 𝑟 ∗ ′), respectively. The socially optimal project
success probabilities before and after financial liberalization are
denoted by 𝜋 𝑒 and 𝜋 𝑒′ . It is assumed that project return R is low.
Thus, when R is low, it is less likely that financial liberalization
lowers over-borrowing (i.e., the difference 𝜋 𝑒′ − 𝜋 ∗′ is more
likely to be greater than 𝜋 𝑒 − 𝜋 ∗ ).
23
Figure 5: Impaired loans by level of financial
liberalization
Notes: This figure displays the aggregate development of the
impaired loans ratio (ILR) over time for a sample of 85 economies,
distinguishing between more (solid line) and less (dotted line)
financially liberalized countries. Aggregate refers to the computation
of the arithmetic average over all countries. Financial liberalization
is measured by the financial reform index (Abiad et al., 2010) that
takes values between 0 and 21 (only observed until 2005). A country
with a fully liberalized financial system has an index value of 21.
After studying the distribution of index values, we chose a cut-off
value of 16 (sample mean) to distinguish more versus less financially
reformed countries.
24
Table 1: Descriptive statistics
ILR (%)
Finreform (0–21)
Private credit (%)
Per capita GDP (in 2005 $)
Inflation (%)
Government consumption (%)
Obs.
Mean
SD
815
815
791
791
797
793
6.5
15.6
62
13552
6.8
15.2
6.7
3.8
52.6
13038
8.4
5.1
Min.
value
0.13
3.8
3.8
525
-18.8
4.5
Max.
value
44.1
21
235.5
49942
80.8
29.8
Notes: The definitions and data sources are in Appendix A.
Table 2: Impact of financial liberalization on ILR
Fin. reform(t-4)
(1)
FE
0.12
(0.57)
-0.21
(0.18)
(2)
FE
0.28
(0.56)
1.53***
(0.42)
-4.48***
(1.17)
(3)
FE
-0.20
(0.54)
1.46***
(0.41)
-4.34***
(1.15)
0.35**
(0.17)
1.36
(1.50)
Yes
815
0.15
91
0.95
(1.48)
Yes
815
0.18
91
0.96
(1.38)
Yes
791
0.19
89
Fin. reform(t-4)  09
2009
Private credit
Per capita GDP
Inflation
Gov. consumption
Constant
Year effects
Observations
R-square
Groups
(4)
FE
-0.01
(0.55)
1.03***
(0.39)
-2.93**
(1.14)
0.51***
(0.16)
-1.05**
(0.48)
-0.01
(0.04)
-0.28
(0.40)
9.88**
(4.31)
Yes
712
0.20
85
Notes: The dependent variable is the logarithm of impaired loans ratio (share of loans
past due by 90 days or more in gross loans). A higher impaired loans ratio means more
financial instability. All variables have been log-transformed. Cluster-robust standard
errors are reported in brackets. ***,**,* indicate significance at the 1%, 5%, and 10%
levels, respectively. FE means fixed effect estimator.
25
Table 3: Regressions with lagged dependent variable
L.ILR
Fin. reform(t-4)
(1)
FE
0.53***
(0.05)
0.20
(0.40)
-0.05
(0.13)
(2)
FE
0.52***
(0.04)
0.36
(0.38)
1.25***
(0.31)
-3.56***
(0.89)
(3)
FE
0.49***
(0.05)
0.11
(0.36)
1.15***
(0.32)
-3.39***
(0.92)
0.36**
(0.17)
0.37
(1.03)
Yes
712
0.39
89
-0.04
(0.98)
Yes
712
0.41
89
-0.62
(0.84)
Yes
691
0.41
87
Fin. reform(t-4)  09
2009
Private credit
Per capita GDP
Inflation
Gov. consumption
Constant
Time effects
Observations
R-square
Groups
(4)
FE
0.47***
(0.04)
0.46
(0.36)
0.95***
(0.35)
-2.74***
(1.01)
0.60***
(0.09)
-0.82**
(0.35)
-0.05
(0.04)
-0.42
(0.25)
6.05**
(2.74)
Yes
626
0.44
83
Notes: The dependent variable is the logarithm of the impaired loans ratio (share of
loans past due by 90 days or more in gross loans). A higher impaired loans ratio means
more financial instability. All variables have been log-transformed. Cluster-robust
standard errors are reported in brackets. FE = fixed effects estiamtor. ***,**,* indicate
significance at the 1%, 5%, and 10% levels, respectively.
Table 4: Regressions with lagged dependent variable using the
instrumental variable (IV) estimator
L.ILR
Fin. reform(t-4)
Fin. reform(t-4)  09
2009
(1)
(2)
(3)
IV
IV
IV
0.87*** 0.83*** 0.85***
(0.09)
(0.09)
(0.12)
0.65*
0.60
0.73
(0.39)
(0.39)
(0.48)
0.68** 0.65**
(0.28)
(0.32)
0.23** -1.66**
-1.53*
(4)
IV
0.72***
(0.12)
0.46
(0.45)
0.84**
(0.36)
-2.12**
26
(0.11)
(0.79)
(0.91)
-0.29
(0.51)
-0.10
(0.09)
Yes
619
88
44.52
25.36
-0.09
(0.09)
Yes
619
88
44.05
26.32
-0.09
(0.09)
Yes
600
86
41.05
25.91
Private credit
Per capita GDP
Inflation
Gov. consumption
Constant
Year effects
Observations
Groups
Under-identification test statistic
F-statistic
(1.01)
0.37
(0.26)
-1.36**
(0.67)
0.03
(0.06)
-0.33
(0.55)
-0.04
(0.09)
Yes
536
82
36.55
17.63
Notes: The estimates are based on Anderson and Hsiao’s (1982) first-difference
estimator, where the second lag of the dependent variable is the instrument for the
lagged dependent variable. The dependent variable is the logarithm of impaired loans
ratio (share of loans past due by 90 days or more in gross loans). A higher impaired
loans ratio means more financial instability. All variables have been log-transformed.
Cluster-robust standard errors are reported in brackets. ***,**,* indicate significance
at the 1%, 5%, and 10% levels, respectively.
Table 5: Robustness to different lags of financial reform
(1)
(2)
(3)
(4)
IV
IV
IV
IV
L.ILR
Fin. reform(t-4)
0.72***
(0.12)
0.46
(0.45)
Fin. reform(t-5)
Fin. reform(t-5)  2009
Fin. reform(t-6)  2009
Fin. reform(t-4)  2008
0.70***
(0.11)
0.74***
(0.11)
0.51
(0.45)
0.19
(0.53)
Fin. reform(t-6)
Fin. reform(t-4)  2009
0.70***
(0.11)
-0.73*
(0.39)
0.84**
(0.36)
0.82**
(0.34)
0.80**
(0.33)
-0.30
27
(0.31)
2009
-2.12**
(1.01)
-2.05**
(0.97)
-2.03**
(0.92)
2008
Private credit
Per capita GDP
Inflation
Gov. consumption
Constant
Year effects
Observations
Groups
Under-identification test statistic
F-statistic
0.37
(0.26)
-1.36**
(0.67)
0.03
(0.06)
-0.33
(0.55)
-0.04
(0.09)
Yes
536
82
36.55
17.63
0.40
(0.25)
-1.29*
(0.68)
0.03
(0.06)
-0.33
(0.54)
-0.03
(0.09)
Yes
536
82
36.90
17.37
0.44*
(0.24)
-1.23*
(0.68)
0.04
(0.06)
-0.41
(0.54)
0.04
(0.09)
Yes
536
82
36.58
16.75
1.18
(0.90)
0.38
(0.27)
-1.68**
(0.69)
0.02
(0.06)
-0.31
(0.56)
-0.04
(0.09)
Yes
536
82
36.06
15.01
Notes: This table shows the results for lags 4, 5 and 6 of the financial reform index. Lag 4 of
the financial reform index interacted with 2008 to indicates if financial liberalization had a
significant effect in 2008. The dependent variable is the logarithm of impaired loans ratio
(share of loans past due by 90 days or more in gross loans). A higher impaired loans ratio
means more financial instability. All variables have been log-transformed. Cluster-robust
standard errors are reported in brackets. IV = instrumental variable. ***,**,* indicate
significance at the 1%, 5%, and 10% levels, respectively.
Table 6: Robustness to different sets of controls
L.ILR
Fin. reform(t-4)
Fin. reform(t-4)  2009
(1)
IV
0.72***
(0.12)
0.46
(0.45)
0.84**
(0.36)
Fin. reform(t-4)  2008
2009
Private credit
Per capita GDP
-2.12**
(1.01)
0.37
(0.26)
-1.36**
(2)
IV
0.71***
(0.12)
0.46
(0.45)
0.93**
(0.45)
0.08
(0.37)
-2.12**
(1.01)
0.37
(0.26)
-1.33**
(3)
IV
0.83***
(0.13)
0.77
(0.48)
0.96**
(0.45)
(4)
IV
0.74***
(0.15)
0.15
(0.59)
0.82*
(0.49)
-2.24*
(1.25)
0.10
(0.25)
-0.92
-2.06
(1.37)
0.04
(0.34)
-2.44**
28
Inflation
Gov. consumption
(0.67)
0.03
(0.06)
-0.33
(0.55)
(0.66)
0.03
(0.06)
-0.33
(0.55)
(0.79)
0.06
(0.05)
-0.40
(0.61)
-0.14**
(0.06)
-0.13
(0.14)
-0.39
(0.28)
-0.00
(0.14)
-0.04
(0.09)
Yes
536
82
-0.04
(0.09)
Yes
536
82
-0.12
(0.11)
Yes
465
80
Bank return on equity
Bank cost to income
Bank concentration
Net interest margin
Intl. debt
Return on assets
Constant
Year effects
Observations
Groups
Under-identification test
statistic
F-statistic
36.55
17.63
36.70
18.89
32.45
11.64
(1.10)
-0.01
(0.06)
-0.28
(1.16)
0.14
(0.25)
0.04
(0.18)
-0.38
(0.29)
0.02
(0.20)
-0.25*
(0.14)
-0.23
(0.25)
0.06
(0.14)
Yes
327
56
26.55
10.05
Notes: This table shows the results for alternative sets of control variables. The dependent
variable is the logarithm of impaired loans ratio (share of loans past due by 90 days or more in
gross loans). A higher impaired loans ratio means more financial instability. All variables have
been log-transformed. Bank return on equity is the average return on assets (net income/total
equity); intl. debt is international debt securities (net issues) as a share of GDP; bank cost to
income refers to the total costs as a share of total income of all commercial banks; bank
concentration indicates the assets of the three largest banks as a share of assets of all
commercial banks; net interest margin is the accounting value of a bank's net interest revenue
as a share of its interest-bearing (total earning) assets; and Bank return on assets is an average
(net income/total assets). The data for the additional controls came from the World Bank’s
Financial Development and Structure Dataset. The definitions and sources of the variables are
in Appendix A. Cluster-robust standard errors are reported in brackets. ***,**,* indicate
significance at the 1%, 5%, and 10% levels, respectively.
29
Appendixes
A Calculation of Equilibrium 𝝅 and 𝒓
We derive formal expressions of the equilibrium lending rate 𝑟 ∗ and average project
success probability 𝜋 ∗ . We start with Equations 5 and 6 from the main text, repeated
here for convenience:
𝐵𝐵 curve:
𝜋(𝑟) = 2 + 2 (𝑅−𝑟) if 𝑅 > 𝑟,
𝐿𝐿 curve:
𝜋=
1
𝛽𝜌
𝑟
1
(B.1)
𝜇
(B.2)
.
Setting the two equations to be equal and solving for 𝑟 yields two solutions:
𝑟1∗ =
1
𝑟2∗ =
1
2
2
1
1
2
2
1
1
2
2
𝑅 + 𝜇 + 𝛽𝜌 −
𝑅 + 𝜇 + 𝛽𝜌 +
√2𝑅𝜇 + 𝜇2 + 4𝛽2 𝜌2 + 𝑅2 − 4𝑅𝛽𝜌 + 4𝛽𝜇𝜌, and
(B.3)
√2𝑅𝜇 + 𝜇2 + 4𝛽2 𝜌2 + 𝑅2 − 4𝑅𝛽𝜌 + 4𝛽𝜇𝜌.
(B.4)
The corresponding equilibrium values for the success probability are:
𝜋1∗ =
𝜋2∗ =
𝜇
−
+ 2, and
(B.5)
1
(B.6)
1
𝑅−𝜇−𝛽𝜌+0.5√2𝑅𝜇+𝜇 2 +4𝛽 2 𝜌2 +𝑅 2 −4𝑅𝛽𝜌+4𝛽𝜇𝜌
𝜇
𝜇−𝑅+𝛽𝜌+0.5√2𝑅𝜇+𝜇 2 +4𝛽 2 𝜌2 +𝑅2 −4𝑅𝛽𝜌+4𝛽𝜇𝜌
+ 2.
Whereas the first equilibrium (𝑟1∗ , 𝜋1∗ ) suggests a low lending rate and high success
probability, the second equilibrium (𝑟2∗ , 𝜋2∗ ) has a high lending rate combined with a low
success probability. The second equilibrium is not feasible though, because 𝑅 < 𝑟2∗ , as
we now show. An equilibrium with 𝑅 < 𝑟 is not feasible because no entrepreneur is
willing to invest. We thus focus on the first equilibrium.
Proposition: The second equilibrium is not feasible, because 𝑹 < 𝒓∗𝟐 .
Proof:
Rewrite
𝑟2∗
in
Equation
A.4
as
2𝑟2∗ − 𝑅 = 𝜇 + 2𝛽𝜌 + √Θ
with
Θ=
√2𝑅𝜇 + 𝜇 2 + 4𝛽 2 𝜌2 + 𝑅 2 − 4𝑅𝛽𝜌 + 4𝛽𝜇𝜌. The right-hand side must be positive; the term
under the square root cannot be negative. It follows that 𝑅 < 2𝑟. In this case, 𝑅 can be
smaller than 𝑟, and the right-hand side is still positive. But in the model, it generally
30
holds that 𝜇 + 2𝛽𝜌 > 𝑟. Now we write 2𝑟2∗ − 𝑅 = 𝜇 + 2𝛽𝜌 + √Θ as 𝑟2∗ − 𝑅 + 𝑟2∗ − (𝜇 +
2𝛽𝜌) = √Θ. On the left-hand side, 𝑟2∗ − (𝜇 + 2𝛽𝜌) is negative, and 𝑅 has to be less than 𝑟.
Otherwise, the right-hand side can never be positive.
B Impact of Financial Liberalization on the Equilibrium
We first show how financial liberalization affects the equilibrium cost of borrowing, then
assess the impact of financial liberalization on over-borrowing. Recall that financial
liberalization is reflected by a decrease in 𝛽, capturing lower reserve requirements, or a
decrease in 𝜌, capturing capital account liberalization. In the following analysis, we focus
on the effect of 𝜌. An analysis of a change in 𝛽 is similar.
For the sake of convenience, we employ implicit differentiation methods and write the
functions that describe borrowing (𝐵𝐵 curve) and lending (𝐿𝐿 curve) behavior as 𝜋 𝐵 =
𝐵(𝑟) and 𝜋 𝐿 = 𝐿(𝑟, 𝜌), respectively. Derivatives are denoted by subscripts. For simplicity,
we write the derivatives 𝐵𝑟 (𝑟) and 𝐿𝑟 (𝑟, 𝜌) as 𝐵𝑟 and 𝐿𝑟 , respectively. The equilibrium
requires that 𝐵(𝑟) = 𝜋 = 𝐿(𝑟, 𝜌). We are interested in solving this system for the
equilibrium interest rate and success probability in terms of the parameter 𝜌 with 𝜋 =
𝜋 ∗ (𝜌) and 𝑟 = 𝑟 ∗ (𝜌), or:
(C.1)
𝐵(𝑟 ∗ (𝜌)) = 𝐿(𝑟 ∗ (𝜌), 𝜌).
Then the comparative statics derivative is obtained by totally differentiating Equation
A.7 to 𝜌:
𝑑𝐵 𝑑𝑟 ∗
𝑑𝑟 ∗
𝑑𝜌
𝑑𝐿 𝑑𝑟 ∗
= 𝑑𝑟 ∗
𝑑𝜌
which can be rearranged for
𝑑𝑟 ∗
𝑑𝜌
=𝐵
𝐿𝜌
𝑟 −𝐿𝑟
(C.2)
𝑑𝐿
+ 𝑑𝜌,
𝑑𝑟 ∗
𝑑𝜌
to obtain:
> 0 since 𝐿𝜌 > 0, 𝐵𝑟 > 0 and 𝐿𝑟 < 0.
(C.3)
This result implies that an increase in financial repression will cause the equilibrium
interest rate to rise.
To evaluate the impact of financial liberalization on over-borrowing, we need to
calculate the derivative:
31
𝑑(𝜋𝑒 −𝜋)
=
𝑑𝜌
𝑑𝜋 𝑒
𝑑𝜌
𝑑𝜋
− 𝑑𝜌,
(C.4)
where 𝜋 𝑒 and 𝜋 denote the socially optimal and average project success probability,
respectively. The derivative
𝑑𝜋𝑒
𝑑𝜌
𝑑𝜋 𝑒
𝑑𝜌
equals:
(C.5)
𝛽
= 2𝑅 > 0.
Because 𝐵(𝑟 ∗ (𝜌)) = 𝜋 ∗ , the change in the equilibrium average project success probability
𝜋 ∗ can be found by using the chain rule, or:
𝑑𝜋∗
𝑑𝜌
𝑑𝐵 𝑑𝑟 ∗
= 𝑑𝑟 ∗
𝑑𝜌
(C.6)
.
Combining Equations C.3 and C.6 yields:
𝑑𝜋∗
𝑑𝜌
𝐵𝑟 ∙𝐿𝜌
=𝐵
𝑟 −𝐿𝑟
> 0, 𝑏𝑒𝑐𝑎𝑢𝑠𝑒 𝐿𝜌 > 0, 𝐵𝑟 > 0 and 𝐿𝑟 < 0.
(C.7)
Next, using Equations C.5 and C.7, we evaluate the expression in Equation C.4:
𝑑(𝜋𝑒 −𝜋)
𝑑𝜌
=
𝑑𝜋 𝑒
𝑑𝜌
⏟
+
𝑑𝜋
− 𝑑𝜌 ∓ 0,
⏟
(C.8)
+
which suggests that the overall impact of financial liberalization on over-borrowing is
ambiguous.
32
C Variable Sources and Definitions
Variable
ILR
Finreform
Private credit
Per capita
GDP
Definition
Share of loans past due by 90 days or
more (impaired loans) relative to gross
loans; gross loans equal the loan portfolio
plus loss reserves
Index based on seven dimensions of
financial liberalization, ranging between 0
and 21; a country with a fully liberalized
financial system has an index value of 21
Private credit to the domestic sector as a
percentage of gross domestic product
(GDP).
Real GDP per capita
Inflation
Annual growth rate of the GDP deflator
Government
consumption
Government current expenditures for
purchases of goods and services
(including employee compensation)
Source
DFID project
Abiad et al. (2010)
World Development
Indicators
World Development
Indicators
World Development
Indicators
World Development
Indicators
D Description of the Financial Instability Data Set
A complete description of the data set is available in Adrianova et al. (2015). Here, we
focus on some key features. The data set covers 23,287 banks for 2000–2009 and
includes retail, investment, Islamic, cooperative, real estate and mortgage, and savings
banks. It excludes rare banks, such as securities firms and microfinance institutions.
This exclusion makes the sample composition more comparable across countries, but it
also reduces the number of banks in a few countries where the excluded bank types
dominate the banking system. As Figure D.1 shows, the total number of banks reporting
their impaired loans ratio varies substantially across the 84 countries: 40 countries base
the information on the impaired loan ratio on 41 or more banks, whereas 28 countries
use information from fewer than 30 banks. In a few countries, the total number of banks
ranges between 31 and 49. As Figure D.2 illustrates, the number of banks in the sample
also declines year-on-year over the sample period (cf. 2003), presumably due to
insolvencies and non-reporting.
33
The bank-level data were aggregated to country-level data by computing the
average impaired loans ratio, weighted by total assets over banks in a country. In
particular, we focus on a sample with no restrictions on bank entry into or exit from the
database, or on the number of reporting periods. This approach maximizes the number
of observations, but we acknowledge that the sample composition thus changes over
time.8
number of countries
20
15
10
5
0
1-10
11-20
21-30
31-40
41-50 51-100 101-200 >200
total number of banks
Figure D.1: Distribution of bank-year observations
total number of reporting
banks
Notes: This figure shows the distribution of the number of observations on the
impaired loans ratio. For example, the first bar indicates that in 19 countries,
observations of the impaired loans ratio come from 1–10 banks. Due to
restricted availability, the data in this graph refer to the sample in which banks
report 66% of the time on all five measures of financial instability.
1800
1600
1400
1200
1000
800
600
400
200
0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Alternatively, we could construct a sample that includes only banks that report in all 10 years, such that
Yearobservations, and banks with more impaired
the sample would be balanced over time, but with fewer
loans would be more likely to drop out. This sample also could suffer from bias, because banks with more
bad loans are more likely to stop reporting or go insolvent and thus leave the sample. Excluding those
banks in all years yields a balanced sample, but for our study of financial instability, it would
problematically exclude banks that are likely to be more instable.
8
34
Figure D.2: Number of observations per year
Notes: This figure shows the total number of reporting banks in each year of the
sample period, 2000–2009. For example, the first bar indicates that the total
number of observations in 2000 was 1733. Due to restricted availability of the banklevel data, these data refer to sample in which banks report 66% of the time on all
five measures of financial instability.
E Countries Included in the Data Set
Albania
Algeria
Argentina
Australia
Austria
Azerbaijan
Bangladesh
Belarus
Belgium
Bolivia
Brazil
Bulgaria
Cameroon
Canada
Chile
China
Colombia
Costa Rica
Cote d'Ivoire
Czech Republic
Denmark
Dominican
Republic
Ecuador
Egypt
El Salvador
Estonia
Ethiopia
Finland
France
Georgia
Germany
Ghana
Greece
Guatemala
Hong Kong
Hungary
India
Indonesia
Ireland
Israel
Italy
Jordan
Kazakhstan
Kenya
Korea
Kyrgyzstan
Lithuania
Madagascar
Malaysia
Norway
Pakistan
Paraguay
Peru
Philippines
Poland
Romania
Russia
Senegal
Singapore
South Africa
Spain
Sri Lanka
Sweden
Switzerland
Tanzania
Thailand
Tunisia
Turkey
Uganda
Ukraine
Mexico
Morocco
Mozambique
Nepal
Netherlands
New Zealand
Nicaragua
United Kingdom
United States
Uruguay
Venezuela
Vietnam
35
36