Assessing the sustainability of credit growth: the case of central and Eastern European countries.

• Author: Coudert, Virginie
• Position: Report

1. Introduction

   Credit booms are generally identified as a key factor behind financial crises, in particular in the emerging countries, as they tend to fuel excessive demand, inflationary pressures and speculative asset price bubbles. In this view, the severe financial crisis that hit some of the central and eastern European countries (CEECs) in 2009 could be attributed to previous excesses. Although the crisis was clearly triggered from abroad by the global financial turmoil, its severity is likely to have overwhelmed the mere contagion effects, especially in the Baltic States. In those latter countries, credit was soaring by 40% to 70% a year in 2006-2007, and has subsequently dried up in 2009. Most other CEECs have followed the same pattern, although with less extreme variations.

   An important question is therefore whether the credit growth had been in excess in the CEECs in the years preceding the 2008-2009 financial crisis. This question is justified since credit growth has been shown to often precede credit crunches and financial crises. (Kaminsky and Reinhart, 1999). The theoretical literature on bubbles gives rationales for that, as leverage amplifies speculative behaviour as shown for example by Allen and Gale (2000). However, assessing the excessiveness of credit is tricky, especially in the case of the CEECS, because of their particular economic situation. As they are meant to catch up rapidly with the previous EU members, their levels of capital, productivity and income are converging towards those of advanced countries. Against this backdrop, it is not surprising that credit growth had been particularly strong, exacerbating external deficits and debt (Duenwald et al., 2005, Coricelli et al., 2006, Diev and Pouvelle, 2008).

   Hence, the strong credit growth that was observed in the CEECs can be interpreted in two ways. First, it may have been part of a normal catching-up process. At the start of transition, between 1991 and 1993, the existing credit stock was eliminated by hyperinflation in some countries (in particular Poland and the Baltic States). Then, during the stabilisation phase, the pace of financial liberalisation and financial deepening steadily picked up. For instance, in 1997, the level of credit stock of these economies was still very low in percentage of GDP: less than 20% in the Baltic States, Poland and Romania (compared with, for example, 82% in France and 106% in Germany in the same period). Second, credit growth may also have been excessive, resulting in an overheating of the economy and inflationary pressures. This could be a concern for some of these countries that are expected to adopt the euro in the future and must therefore comply with the Maastricht convergence criteria, in particular the price stability criterion.

   Two types of approach are used in economic literature to identify credit booms. The first is a purely statistical approach, based on deviations of credit series from their long-term trend, such as in Gourinchas et al. (2001), Tornell and Westermann (2002), IMF (2004) and Sa (2006). The second is econometric and seeks to explain the level of credit or credit growth as a function of economic fundamentals (Cotarelli et al. (2005), Boissay et al. (2005), Egert et al. (2006), Kiss et al. (2006)).

   This article applies both types of approach using a large sample of emerging and developed countries, with a view to identifying a behaviour pattern that may be specific to countries of eastern and central Europe. The goal we pursue by using alternative calculations is to determine which one seems to be the better indicator of excessive credit growth. In the statistical approach, we test possible thresholds and indicators to define credit boom periods. In the econometric approach, we use an error-correction model. We first determine the equilibrium level of the credit/GDP ratio corresponding to the fundamentals in the sample as a whole. If the credit/GDP ratio has not yet reached its estimated equilibrium level in CEECs, the rapid credit growth may stem from the catching-up process. Credit growth is then explained as a function of deviations of the credit/GDP ratio from its equilibrium level, estimated in the previous stage, and other macroeconomic variables such as the growth of GDP per capita. An error-correction model had already been used by Boissay et al. (2005). Here, we use a large reference sample including both developed, emerging and transition countries in order to take account of the interactions between the initial level of credit and the speed of convergence towards the new long-term equilibrium.

   The rest of the paper is organized as follows. Section 2 compares the credit/GDP ratio and real credit growth with their long-term trend; beyond a certain threshold, positive deviations are classified as credit booms. In Section 3, we provide econometric estimates of the credit/GDP ratio relative to macroeconomic variables and estimate the credit growth rates; we then compare the estimated values with the observed figures in the CEECs in 2007 and 2008.

2. Deviations from the long-term trend

2. 1 The principle

   Comparing time series with their long-term trend is a straight way to identify outstanding observations. The time series is decomposed into its long-run and short-run components by a filtering method, the most popular being the two-sided linear Hodrick-Prescott (1980) filter. In the case of credit, if a credit indicator significantly exceeds its long-term trend at a certain date, this can be considered as signalling a credit boom. In the following, we present this methodology by generalising the method adopted by Gourinchas, Valdes and Landerretche (2001), IMF (2004) and Sa (2007). These different studies vary according to the credit indicators used, the way in which deviations from trend are calculated and thresholds defined.

   2.1.1 The credit indicator used

   Credit boom periods are generally estimated by using panel data, since too few of these events occur in a single country. The sample covers a set of countries, denoted i = 1,.., n, over a period t = 1,.., T.

   The first relevant indicator is the credit/GDP ratio, as a percentage, denoted [c.sub.1,i,t]:

   [c.sub.1,i,t] = 100*[C.sub.i,t] /[Y.sub.i,t] (1)

   where [C.sub.it] denotes the outstanding stock of loans of country i at date t, and [Y.sub.it] its GDP. This is the indicator used by Gourinchas, Valdes and Landerretche (2001).

   The second possible indicator is the real credit growth rate:

   [c.sub.2,i,t] = 100 [[C.sub.i,t]/[C.sub.i,t-1]/1 + [[pi].sub.i,t - 1] (2)

   where [[pi].sub.i,t] denotes the inflation rate of country i in time t. Tornell and Westermann (2002), IMF (2004) and Sa (2007) use this indicator.

   2.1.2 Calculating the deviation from the trend

   The long-term trend, denoted [[??].sub.k,i,t] for, k = 1,2 is generally estimated by a Hodrick-Prescott (HP) filter. In this paper, we also use a fixed-length symmetric band-pass Baxter-King (1999) filter (BK). The deviation from the long-run trend is equal to the difference between the indicator and its trend. (3)

   [e.sub.k,i,t] = [c.sub.k,i,t] [[??].sub.k,i,t] (3)

   It is therefore expressed as a percentage, corresponding to GDP percentage points, for [c.sub.1,i,t] and real growth points for [c.sub.2,i,t].

   By definition, we consider that a credit boom is identified at period t in country i if and only if the deviation [e.sub.k,i,t] exceeds a certain threshold [S.sub.k,i]

   [e.sub.k,i,t] > [S.sub.k,i] (4)

   The thresholds [S.sub.k,i] are set either separately for each of the countries or are the same across the sample, depending on the method used. We construct a dummy variable, denoted [I.sub.k,i,t], that indicates the credit boom being equal to 1 when the country experiences a credit boom, and to 0 otherwise.

   [I.sub.k,i,t] = 1, if [e.sub.k,i,t] > [S.sub.k,i]

   [I.sub.k,i,t] = 0, otherwise. (5)

   2.1.3 The two methods for defining the thresholds

   By varying the threshold [S.sub.k,i], the definition of the credit boom is more or less restrictive: the higher the threshold, the rarer the cases of credit booms. The thresholds can be defined in two ways.

   The first method defines them for each country individually as a multiple of the standard deviation of credit fluctuation around the trend:

   [S.sub.k,i] = a[[phi].sub.k,i]

   where [[phi].sub.ki] denotes the standard deviation of the credit fluctuation around the trend for country i, a is an arbitrarily chosen coefficient. The IMF (2004) uses this approach by setting the coefficient a at 1.75. A credit boom is thus defined as credit growth that exceeds its long-term trend by 1.75 times the standard deviation of the fluctuation around the trend. With this figure, assuming a normal distribution there would be a 5% probability that the gaps would lie above the threshold, which yields 5% of credit booms in the sample.

   The second method consists in calibrating thresholds to obtain a given proportion p(0

   [S.sub.k] such that 1/NT [N.summation over (i=1)] [T.summation over (t=1)] [I.sub.kit] = P (7)

   In this case, the threshold is set as a single value for all countries. Note that for each given proportion of crises p, Equations (5) and (7) implicitly defines a unique value of the threshold [S.sub.k]. This is the method used by Gourinchas, Valdes and Landerretche (2001), who take a sample of 91 countries over the 1960-1996 period and set different thresholds in order to obtain a given number of booms.

   Another more expeditious technique consists in choosing arbitrary thresholds for the credit growth in all countries (Tornell and Westerman, 2002). These authors take three different definitions of a boom: period of cumulative real credit growth over the two previous years of more than 20%, 30% and 40%, based on a sample of 39 countries, over the 1980-1999 period.

2. 2 Estimates

   For each country, we consider two indicators: the credit/GDP ratio and real credit growth.4 We estimate their long-run trend using a Hodrick-Prescott filter and...