WPS8465 Policy Research Working Paper 8465 Anatomy of Credit-Less Recoveries Luisa Corrado Isolina Rossi Macroeconomics, Trade and Investment Global Practice June 2018 Policy Research Working Paper 8465 Abstract The recovery from the global crisis that erupted in 2007 on a sample of advanced and emerging countries between shows that the decoupling between real and financial vari- 1980 and 2014. Using a simple endowment economy ables during the business cycle can lead to negative and model, the paper shows that credit-less recoveries are cor- long-lasting consequences for the economy. A key feature of related with liquidity shocks in real and financial markets the past global crisis in many countries is that the recovery and with the pace of private sector deleveraging. The empir- in aggregate output has not been accompanied by a con- ical analysis shows that during these episodes demand-side temporary pick-up in lending flows to the private sector, frictions played a relatively larger role in predicting the rendering the recovery credit-less. This paper uses data on occurrence of the episodes, reflecting weak demand for output and credit to study the relative roles of demand and liquidity by the private sector in the aftermath of the crisis. supply drivers of credit growth during economic recoveries This paper is a product of the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at irossi1@worldbank.org and luisa.corrado@uniroma2.it. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Anatomy of Credit-less Recoveries Luisa Corradoy Isolina Rossi z Key words: Credit-less Recoveries; Liquidity; Probit Models JEL Codes: C23, E32, E51, E41. Authors are reported in alphabetical order of last names; the …ndings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its a¢ liated organizations, or those of the Executive Directors of the World Bank or the governments they represent. y University of Rome Tor Vergata, Department of Economics and Finance and University of Cambridge, Faculty of Economics. E-mail: luisa.corrado@uniroma2.it, lc242@cam.ac.uk. z Corresponding author: University of Rome Tor Vergata, Department of Economics and Finance. E-mail: isolina.rossi@uniroma2.it 1 Introduction There is a unanimous consensus among economists on the central role played by bank lending in supporting economic activity, especially in the aftermath of severe economic downturns. Recent economic history shows that credit ‡ows are characterized by boom-bust cycles which can lead to high market volatility and damaging consequences for the economy (Claessens et al. 2011). Financial frictions play a crucial role in determining the shape of these cycles and during recessionary episodes they can magnify the e¤ects of macroeconomic ‡uctuations. The recent global crisis that erupted in 2007 sparked renewed interest on the drivers of such frictions and di¤erent methodologies have been developed to model new narratives of business and …nancial cycles. It remains unclear, however, whether these frictions translate into shocks to the demand or to the supply of credit (Adrian et al. 2011). A key feature of the recent global crisis is that in many countries worldwide the recovery in aggre- gate output has not been accompanied by a contemporary pick-up in bank lending ‡ows, rendering the recovery credit-less. During these episodes GDP growth is found to be on average between 2 and 3 percent lower than during normal recoveries, and investment remains weak and below pre-crisis levels with potentially negative long-term e¤ects for the economy. A recent growing body of liter- ature focusing on the liquidity-output nexus during recoveries shows that the decoupling between credit and output is not uncommon. On average, between 19 and 26 percent of all recoveries occur in the absence of a pick-up in lending ‡ows (Abiad et al. 2011; Bijsterbosh and Dahlhaus, 2011 and Sugawara and Zalduendo, 2013). Deepening the understanding of credit-less recoveries is important to analyze the underlying frictions and demand-side dynamics at play during these episodes. In addi- tion to exhibiting systematic lower growth and investment performance, recent literature shows that credit-less recoveries may also re‡ect ine¢ cient reallocation of credit among sectors, with resources shifting away from sectors more dependent on external …nancing (Coricelli and Frigerio, 2015). Dur- ing these episodes GDP growth is found to be on average between 2 and 3 percent lower than during normal recoveries, and investment remains weak and below pre-crisis levels with potentially negative long-term e¤ects for the economy. Credit-less episodes provide an ideal experiment to test whether …nancial frictions are predom- inantly supply or demand-driven, and to study their impact on the liquidity-output nexus during recoveries. At a theoretical level, credit-less recoveries could be the outcome of demand or supply- side constraints. Demand-side constraints to credit growth can arise from the reluctance of …rms and households to resume borrowing in the aftermath of a recession due to weak growth and employment prospects. At the same time, weak credit demand could be driven by a deterioration of borrowers’ credit-worthiness position or excessive debt levels. On the contrary, supply-side constraints manifest through higher lending costs and could be the outcome of a temporary shortage of liquidity following a crisis accompanied by …nancial sector stress and excessive pressure on banks’balance sheets. In this context, failure to restore adequate liquidity levels may result in a tightening of credit supply to the economy. The paper uses a novel panel dataset on output and domestic bank credit for a sample of advanced and emerging countries between 1980 and 2014 to analyze the role of demand and supply-side drivers of bank lending growth during credit-less recoveries. We argue that during these episodes the evidence points to the prevalence of demand-side frictions to credit growth, re‡ected in lower demand for liquidity by the private sector in the aftermath of recessions. During these episodes output recovery takes place while the economy is deleveraging, i.e. the stock of credit is decreasing. However, as in Biggs et al. (2009) we show that as long as the pace of deleveraging is decreasing a rebound in output can occur in the presence of negative credit growth. Our work is carried out in two steps. In the …rst one, we model credit-less recoveries in a simple endowment economy framework featuring representative in…nitely lived households. In this model, credit dynamics are a function of the policy rate and of liquidity shocks. These shocks re‡ect frictions in …nancial intermediation and a¤ect the value of liquidity needed to …nance consumption and investment. In the second step, we test empirically the hypotheses. We …rst exploit simple information on the correlation between prices and output during the recession phase of the cycle to identify demand and supply driven recessions originating in the real and …nancial markets. Subsequently, we examine the role of demand and supply-driven recessions in in‡uencing the likelihood of a credit-less recovery in our sample of advanced and emerging countries. Related literature: Our work builds on the theoretical literature on credit constraints and is closely related to empirical studies on …nancial crises and credit-less recoveries. From an economic theory perspective, the relationship between output and credit has been analyzed in the context of macroeconomic equilibrium models featuring real and …nancial markets. A large body of literature concerned with the notion of …nancial accelerator (Bernanke and Gertler, 1989; Bernanke, Gertler and Gilchrist, 1999 and Kiyotaki and Moore, 1997) has documented the channels through which …nancial markets a¤ect the real economy. In these models, the interaction between real and …nancial variables stems from the need of …rms to access funds in credit markets. Due to …nancial market imperfections, …rms’ ability to access liquidity is dependent on the value of their …nancial wealth, which is assumed to be pro-cyclical. Shocks a¤ecting the economy impact on borrowers’ credit- worthiness and on their ability to access funds to …nance investment, contributing to amplify the magnitude of business cycle ‡uctuations. 3 The …nancial ampli…cation mechanism can work through distinct channels, which can poten- tially a¤ect both the demand and supply of liquidity in the economy. When credit constraints arise from high collateral requirements (Kiyotaki and Moore, 1997), a drop in the value of collateral dur- ing recessions limits borrowers’ ability to access funds resulting in a contraction in the amount of credit supplied. At the same time, a tightening of agents’borrowing capacity can lead consumers to increase precautionary savings, while reducing consumption and ultimately credit demanded (Guer- rieri and Lorenzoni, 2011). Recent empirical literature has also investigated the link between lending standards, informational asymmetries and access to credit during the business cycle. For example, Ariccia and Marquez (2006) present a framework showing that when informational asymmetries Dell’ in loan markets become less severe, as may occur during the expansionary phase of the cycle, lend- ing standards become less stringent. This leads to higher credit growth and to increased …nancial systemic vulnerability. Credit demand and supply can also be a¤ected by excessive leverage in the economy. A high debt overhang can make …rms and households reluctant to resume borrowing in the aftermath of reces- sions, keeping credit demand low. Similarly, excessive indebtedness among …nancial intermediaries, including banks can prevent their ability to secure liquidity in the inter-bank market, thus a¤ecting the liquidity supplied to the economy (IMF, 2013). Chada et al. (2010) study how liquidity e¤ects a¤ect monetary conditions in a simple endowment economy model, providing a comprehensive strategy to analyze the demand and supply drivers of liquidity levels in the economy. Our work also contributes to the recent empirical literature on credit-less recoveries, examining the lending-growth nexus during the business cycle. The …rst authors to document these episodes were Calvo et al. (2006). In their seminal work on post-collapse crisis recoveries in emerging countries they name these episodes Phoenix Miracles, as output is found to “rise from its ashes” without any recovery in the stock of credit. The recovery in output is explained by a process of …nancial engineering whereby …rms discontinue long-term investment projects in order to restore liquidity to …nance activity amidst …nancial constraints. More recent studies document signi…cant cross-country heterogeneity in the distribution of post- crisis performance. Using the same sample of crisis episodes of Calvo et al. (2006), Huntley (2008) re-examines the evidence on post-crisis performance using GDP per-capita data. The path of post- crisis recoveries is characterized by a bi-modal distribution leading to two di¤erent types of recoveries: slow and fast recoveries. Sixty percent of all recovery episodes are found to take place in …ve years, accompanied by strong investment levels, credit and consumption; in the remaining cases, output 4 recovery takes place in 15 years or more without any rebound in domestic credit.1 Credit-less recoveries are also a central feature of business cycles in advanced countries (Abiad et al. 2011; Claessens et al. 2009) and tend to follow recessions accompanied by …nancial sector stress, including boom-busts in credit markets. Analyzing the …nancial cycles of a sample of advanced economies, Claessens et al. (2008) show that recessions associated with strong credit contractions and house prices busts tend to be more severe and last longer. More often than not, following these episodes a recovery of the real economy occurs ahead of improvements in …nancial conditions. Similarly, Abiad et al. (2011) show that when credit-less recoveries are preceded by systemic banking crises, their frequency is more than three times higher. If a credit boom and a banking crisis precede the recovery, the relative frequency of such episodes in the sample of countries under analysis increases to 77 percent. Other macroeconomic indicators positively correlated with the probability of credit-less recovery are the size of output contractions and the extent of external adjustment during the recession phase of the cycle (Sugawara and Zalduendo, 2013 and Bijsterbosch and Dahlhaus, 2011). During credit- less episodes sectors more dependent on external …nance tend to grow disproportionately less then during normal recoveries (Abiad et al. 2011) and resources tend to depart from sectors which are more (bank) credit dependent (Coricelli and Frigerio, 2015). Our work di¤ers signi…cantly from current studies. While existing literature predominantly fo- cuses on the macroeconomic conditions preceding credit-less recoveries, our analysis focuses on the study of demand and supply dynamics underlying credit patterns during crisis episodes, using price- output correlations to disentangle the relative contributions of distinct (demand and supply) frictions during the recession phase. In particular, the work contributes to the above mentioned literature in several ways. First, it is among the …rst works to provide a comprehensive theoretical framework to analyze credit-less recoveries, which places major emphasis on the role of demand and supply side frictions in credit markets to explain liquidity dynamics during the recovery phase of the cycle. Second, it employs a novel panel database of crisis episodes between 1980 and 2014 - including data covering the global …nancial crisis - to examine the relative contributions of demand and supply drivers of credit growth. By analyzing distinct drivers of output recovery, our work shows that sub- dued lending activity in the aftermath of recessionary episodes, re‡ects (on average) weak demand for liquidity by the private sector. These results suggest that when constraints to credit growth are mainly demand-driven, policy interventions aimed at stimulating aggregate demand making full usage of the …scal and monetary levers should be prioritized. The remainder of the paper is organized as follows. Section 2 presents the model and the testable 1 The role of cross-country hetereogenity during credit-less recoveries has also been documented by Ayyagari, Demirgüç-Kunt and Maksimovic (2011). 5 assumptions. Section 3 describes the data and the empirical strategy and then discusses the results and presents the robustness analysis. Finally, section 4 concludes. 2 Model To analyze the liquidity-output nexus during economic recoveries we extend the theoretical framework of Chadha et al. (2010). We analyze the demand and supply drivers of liquidity levels and growth in the economy. At the beginning of the period households, who are both consumers and entrepreneurs, receive an exogenous endowment, yt 1 . Following receipt of the endowment, any deposit transfer obtained from the previous period, returns from maturing bonds, return on capital, the representative dt household decides how to allocate its wealth over real (cash) deposits pt , investments, It = kt kt 1 , and a one-period nominal bond, bt . The household then receives its nominal endowment income, pt yt , which cannot be spent until the following period. 2.1 Households At the beginning of each period, households maximize their utility: P 1 i t max U = Et ln ci ; (1) i= t where is the intertemporal discount rate, Et are expectations formed at time t and ci is the consumption level. The representative household is also subject to the cash-in-advance constraint: dt 1 ct t 1: (2) pt implying that the current consumption of liquidity-constrained households is …nanced by cash de- posits accumulated in the previous period (dt 1 ). A stochastic shock to liquidity, t 1, re‡ecting frictions in …nancial intermediation alters the value of liquidity, t 1 dt 1 , which in turn impacts on consumption in the following period. The individual budget constraint is: st+1 dt pt 1 y t 1 bt dt 1 ct + (kt kt 1 ) + bt+1 + = + + ik;t kt 1 + t 1 (3) pt pt pt pt pt where the nominal bond in time t + 1 yields a unit of currency, a nominal return equal to it+1 and 1 a price of st+1 = 1+it+1 . The Lagrange multiplier on the …rst constraint is denoted 1;t and to the second one as 2;t . Di¤erentiating with respect to ct , bt+1 , dt and kt , brings to the following …rst 6 order conditions: 1 = 2;t + 1;t ; (4) ct st+1 1;t+1 1;t = Et ; (5) pt pt+1 1;t pt pt = Et 2;t+1 + Et 1;t+1 : (6) t pt+1 pt+1 1;t = (1 + ik;t ) Et 1;t+1 (7) Using (4) and rearranging expression (6) yields: 1 pt 1;t = t Et (8) ct+1 pt+1 Given (8) we can express the inter-temporal equilibrium condition (5) as: ct+2 t+1 pt+1 Et = Et (1 + it+1 ) (9) ct+1 t pt+2 This shows that under the optimal equilibrium path, households’consumption will depend on devi- ations in the nominal interest rate, in‡ation and liquidity shocks.2 2.2 Consumption, Credit and Spreads The log-linearized form of the optimality condition in (9) can be expressed as: 4ct+2 = it+1 4pt+2 + 4 t+1; (11) implying that consumption growth is tilted by liquidity shocks in …nancial markets and by the current stance of monetary policy.3 Similarly, the cash in advance constraint (2) implies that deposit growth in t + 1 yields the following: 2 The set of optimality conditions (5) and (7) also imply the following asset equation: (1 + it+1 ) (1 + ik;t ) Et 1;t+1 = Et 1;t+1 (10) pt+1 pt according to which at the optimum, a household is indi¤erent between the two assets (capital and bonds) since the expected bene…t in terms of utility is the same. 3 We omit the expectation operator for notational convenience. 7 4ct+2 = 4dt+1 4pt+2 + 4 t+1 (12) i.e. in the short-term, deviations of consumption from its long-run path will be determined by the policy rate and the liquidity premium. By equating (11) and (12): 4dt+1 = it+1 (13) which shows that in the long run higher deposit growth increases the nominal rate. In this model the liquidity shock creates a gap between the policy rate, it , and the cost of liquidity provision by commercial banks, im t . Speci…cally, the external …nance premium, ef pt , depends on the …nancial market shocks: it im t = ef pt = d t + s t = t: (14) where t is a composite supply/demand …nancial shock and t = t 1 +" ;t with < 1 and " ;t is a normally distributed error term. Under the simpli…ed assumption that banks transform deposits into loans, we introduce the growth of credit as a function of the growth of deposits. Under this framework, any change in credit growth, 4ds m t , will depend on the market interest rate, it : 4d s m t = it (15) Therefore, any deviation between the actual growth of credit from the one expected by the monetary authority when setting the policy rate, it , will depend on the ef pt : 4d s t 4 dt = i m t it = ef pt = d t + s t (16) which implies: 4ds t 1 4 dt 1 = d t 1 + s t 1 (17) i.e. the historical disequilibrium in the credit market will depend on the occurrence of liquidity demand or supply shocks. 2.3 Firms The consumer good is produced by …rms owned by households in a competitive market via the production function: 8 ct = kt (18) We assume that kt = f (ds t 1 ), i.e. capital is a¤ected by the credit level provided by commercial banks to …rms which becomes e¤ective one period ahead due to the sluggish response of investments in long-term capital (Smets and Wouters, 2007; Faccini and Yashiv, 2015). Investments are therefore given by: ds t 1 ds t 2 It = (kt kt 1 ) = (19) pt and consumption is: ds t 1 ct = pt We now analyze the e¤ects of the growth of credit in the aggregate economy. 2.4 Credit Growth and the Aggregate Economy during Recessions In our model, output is the sum of consumption and investments: yt = ct + It (20) By replacing (18) and (19) in (20) we obtain an expression for nominal output, yt pt : yt pt = ds t 1 + ds t 1 ds t 2 (21) Hence, as stressed by Biggs et al. (2009), output is correlated with both the ‡ow and the change in the ‡ow of credit in the economy. The latter is named credit impulse. Expressing the above expression in terms of changes, nominal output growth is: 0 1 + B C 4y t + 4p t = 4dst 1 + @ 4ds 4ds A (22) | {z } | t 1 {z t 2 } Deleveraging Credit Impulse Equation (22) implies that when the economy is deleveraging, i.e. 4ds t 1 < 0; output will decrease. However, as long as deleveraging occurs at a decreasing rate (or if its pace slows down), 4ds t 1 < 4ds t 2 < 0, a recovery in output can occur in the presence of negative credit growth. In fact, while the reduction in the stock of credit impacts directly on investment in long-term capital and therefore on potential output, any development in the ‡ow of credit may explain the business cycle rebounding e¤ects during recoveries. 9 Next, we model in‡ation dynamics assuming that current in‡ation depends on the output gap in previous periods and on the occurrence of demand and supply shocks. Negative values of the output F gap in t 1, yt 1 yt 1 , signal that the economy is operating below its potential, and accumulating unused capacity which …rms can absorb during the recovery phase: 0 1 B F d;s C pt = pt 1 + @ yt 1 yt 1 + zt 1A (23) | {z } U nused Capacity d;s zt 1 are a real demand/supply shock during the recessionary phase that a¤ects the accumulation of unused capacity in t 1 and, therefore, in‡ation in period t . We assume the following autoregressive d d s s processes zt = d zt 1 + "d;t , zt = s zt 1 + "s;t with d, s < 1 where "d;t and "s;t are normally distributed error terms. Finally, we allow for hysteresis in potential output: F d;s yt 1 = yt 2 + zt 2 (24) d;s i.e. potential output is not only a function of past demand/productivity shocks, zt 2 , but also a function of past actual output, yt 2 . Following the recent crisis episodes worldwide a number of economic institutions and research hubs, such as for example the IMF (2009) and the European Commission (2009) and researchers, see for example Pisani-Ferry and van Pottelsberghe (2009) and Furceri and Mourougane (2009), have emphasised the negative e¤ects of deep economic recessions on potential output. To this extent, we extend the basic model by allowing the path of potential output to be in‡uenced by lagged actual output. Therefore, by replacing (24) in (23): d;s 4p t = 4y t 1 + 4zt 1 (25) in an economic downturn the term in round brackets will re‡ect change in unused capacity. In the empirical section we de…ne output growth, 4yt 1 , as (yt 1 yt 2 ) where yt is real GDP. 2.5 Testable Implications By replacing 4pt from (25) in (22) we derive the following evolution of aggregate demand: d;s 4yt = 4ds t 1 + 4d s t 1 4 ds t 2 4 yt 1 zt 1 (26) The supply of credit 4ds t 1 implied by (17) is: 4d s t 1 = 4dt 1 + d t 1 + s t 1 (27) 10 Credit ‡ow between t 1 and t 2 is obtained by adding and subtracting 4ds t 2 = 4d t 2 + d s t 2 + t 2 on the LHS and RHS of (27): 4ds t 1 4 ds t 2 = (4dt 1 4dt 2 ) + 4 d t 1 +4 s t 1 (28) d s where 4 t 1 and 4 t 1 denote supply/demand …nancial shocks during the recessionary phase. We de…ne the probability of credit-less recoveries as the joint probability that yt > 0 and 4dt 6 0. The determinants of this probability can be identi…ed from the reduced form model implied by the three equations (26), (27) and (28): 0 + 1 B (4dt 1 ) ; (4dt 1 4dt 2 ); 4yt 1 ; C B | {z } | {z } | {z } C B Deleveraging Credit Impulse U nused Capacity C Pr ( yt > 0; 4 dt 6 0) = f B B + + + + C C (29) B (4 d s t 1; 4 t 1) ; d zt s 1 ; zt 1 C @ | {z } A F inancial Demand=Supply F rictions | {z } Real Demand=Supply F rictions Hence our testable implications on the drivers of credit-less recoveries imply deleveraging, credit impulse, underutilization capacity or dominance of frictions in real/…nancial demand and supply during recessions. As we document in Figure 1 during credit-less episodes, real GDP growth was on average lower than during "credit-with" recoveries. Indeed, investment and consumption are the two components of demand mostly a¤ected by the bounce-back e¤ect channelled by changes in liquidity, as illustrated in the model above. 3 Liquidity Shocks and the Shape of The Recovery: Empir- ical Analysis 3.1 Data, Methodology and Stylized Facts Our sample includes 42 advanced and emerging economies, covering the period between 1980 and 2014.4 Crises episodes and credit-less recoveries are identi…ed following a methodology similar to Braun and Larrain (2005) and Abiad et al. (2011). A contraction in output is considered as a 4 The sample of advanced economies includes: Australia, Austria, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, New-Zealand, Norway, Spain, Sweden, Switzerland, United Kingdom and United States. The sample of emerging economies includes: Algeria, Argentina, Brazil, Chile, China, Colombia, Hun- gary, India, Indonesia, The Republic of Korea, Malaysia, Mexico, Peru, Philippines, Poland, The Russian Federation, South Africa, Thailand, Turkey, Ukraine, Uruguay and Venezuela. 11 crisis episode each time the HP-…ltered cyclical component of real GDP falls one standard deviation below the mean. Letting t be the year of the trough of the crisis, a recovery is de…ned as the …rst four years following the crisis trough t. The business cycle was then described around the peak, the trough and the recovery of the cyclical component of GDP. To de…ne credit-less recoveries we use a methodology similar to Bijsterbosch and Dahlhaus (2011), employing …ve di¤erent de…nitions of credit-less recoveries. These are summarized in Table 1. According to de…nition 1 a recovery is de…ned as credit-less if the annual growth of domestic (real) bank credit is negative during the …rst three years of the recovery (i.e in t + 1; t + 2 and t + 3); in the case of de…nition 2, a recovery is de…ned as credit-less if the annual growth rate of domestic real credit is negative during the …rst two years of recovery (i.e in t + 1 and t + 2 ). In the case of de…nition 3, a recovery is de…ned as credit-less if the annual growth rate of domestic real credit is negative during the …rst year of recovery (i.e in t + 1). According to de…nition 4, a recovery is de…ned as credit-less if the average annual growth rate of domestic real bank credit is negative during the …rst four years of the recovery (i.e. between t + 1 and t + 4); In the case of de…nition 5, a recovery is de…ned as credit-less if the average annual growth rate of real bank credit is negative during the …rst four years of the recovery, starting from t+2 (i.e. between t + 2 and t + 5). Using de…nition 2, we identify 113 output collapses of which 21 - or more than 18 percent - showed a credit-less recovery. Tables 3 and 4 present a full list and description of the growth performance of these episodes. Descriptive evidence shows that stressed credit conditions signi…cantly impacted the pace of re- covery in the sample of countries, a result in line with …ndings of previous empirical literature on credit-less recoveries (Abiad et al. 2011; Bijsterbosch and Dahlhaus, 2011). Our analysis con…rms that credit-less recoveries bring signi…cant costs. Figure 1 documents that during these episodes, real GDP growth was on average more than 2 percent lower than during "credit-with" recoveries; investment and consumption were the two most hard hit components of demand, growing on average 4.3 and 1.7 percent less. Figure 3 presents descriptive statistics on the number of demand and supply-driven recessions, and credit-less recoveries over time in the sample of advanced and emerging countries. In the real sector of the economy, we …nd that demand and supply driven recessions are homogeneously distributed over the crisis episodes that occurred between 1980 an 2014, with the exception of the time window of 2007-2014 when demand driven recessions prevailed. In the …nancial sector, supply driven recessions prevailed during the crisis episodes that occurred between 1980 and 2006, while demand driven recessions were dominant in the time window 2007-2014. We then move to assess how credit-less episodes are linked to the nature of the preceding recession, 12 i.e. whether they are driven by changes in supply or demand conditions in the …nancial and real sectors. In this paper, to identify (real and …nancial) demand and supply driven recessions, we follow the literature on the cyclical relationship between output and prices (den Haan, 2000; Cooley and Ohanian, 1991; Pakko, 2000; Smith, 1992 among others). To identify real (demand and supply) driven recessions, we draw upon information on prices and output growth during the recession phase of the crisis episodes identi…ed in our sample. In a model with ‡exible price-adjustment, during recessions in which supply-side drivers dominate, the correlation between in‡ation and output growth is negative; when demand-side drivers prevail, the same correlation is assumed to be positive. If the whole process only takes a short period of time, we will typically observe changes in quantities and relatively stable prices so the correlation will be smaller or close to zero. Since we are focussing on changes in demand and supply during the whole recessionary phases it is reasonable to assume that the correlation will re‡ect long term changes in demand and supply conditions with a resulting non zero correlation between output and prices. To identify …nancial (demand and supply) driven recessions, we exploit data on credit growth and on the external …nance premium (EFP).5 The underlying assumption is that a negative correlation between the EFP and credit growth during the recession phase signals the prevalence of supply- side constraints to lending, whereby increases in borrowing costs are accompanied by a contraction in credit ‡ows. On the contrary, a positive correlation between the two variables would signal a demand-side constraint, as contractions in the demand for credit would lead to lower borrowing costs (provided that credit supply remains unchanged). If the whole process only takes a short period of time, we will typically observe changes in liquidity but smaller changes in the cost of liquidity provision. That implies a stable EFP and smaller (or close to zero) correlation between liquidity and EFP. Table 2 provides descriptive statistics on the correlation between price and quantities in the real and …nancial sectors for our sample of countries. To investigate the channels of the relationship between credit and output during recoveries, we employ a …xed e¤ects Probit model similar to Abiad et al. (2011), where the dependent variable re‡ects the probability that the recovery will be credit-less. Let a credit-less episode Yi;t in country i = 1; 2; :::; N in time t = 1; 2; :::; T be modeled by: Yit = X0it + uit Yi;t = I (Yit > 0) (30) Where Yit indicates the occurrence of a credit-less recovery in country i in time t, and Yit is an 5 The external …nance premium (EFP) signals the tightness of borrowing constraints and is calculated by taking the di¤erence between the bank lending rate and the risk-free rate at which short-term government bonds are issued. The variable is constructed using data on bank lending and risk-free rates sourced from the International Financial Statistics (IFS/IMF) database. 13 indicator function transforming Yit into a dummy variable signaling the occurrence of a credit-less recovery. Xit is a vector of regressors including drivers of credit-less recoveries. We assume that the error term uit in equation (30) has a normal distribution with unit variance. Model (30) makes no assumption on the correlation between country-speci…c time-invariant e¤ects s covariates Xit . In order to take into account the role of panel-level correlation, we and the model’ also estimate the same model with a) a random e¤ects Probit model and b) a Mundlak-type (1978) correction model. While the assumption under the random e¤ects estimator is that the panel-level s correction model introduces an entities are uncorrelated with the independent variables, Mundlak’ alternative assumption on the characteristics of the country-level speci…c e¤ects. In Mundlak (1978), the country-speci…c e¤ects in the error term uit = i + it are related to the time-invariant averages s independent variables: of the model’ i = X0i + vi E( i j Xi ) = X0i (31) Where Xi is the time invariant average of each covariate Xit . The full model can be described by the following speci…cation: Yit = X0it + X0i + vi + it (32) s covariates In equation (32), the term vi is assumed to be random and uncorrelated with the model’ Xit . The coe¢ cient vector estimates the e¤ect of Xit on the probability of a recovery being credit- less, holding the panel level means …xed in time. As the covariates re‡ect the characteristics of the recessions prior to the credit-less episode, this addresses potential endogeneity problems of all regressors in the …xed and random e¤ects models. 3.2 Explanatory Variables Our work analyzes the drivers of credit-less recoveries looking separately at the demand and supply- side contributions to credit growth in real and …nancial markets. It focuses on the role of the following covariates in predicting the occurrence of a credit-less recovery: Demand-side covariates GDP growth : The inclusion of this variable aims at capturing a "bounce-back" e¤ect, as in Bijsterbosch and Dahlhaus (2011). In equation (29), negative values of this variable signal that the economy is operating below its potential, and accumulating unused capacity which …rms can absorb during the recovery phase. This in turn may limit their need for new borrowing keeping demand for credit low. Real demand shock dummy variable : This is a dichotomous variable taking the value of 1 if the 14 correlation between output growth and in‡ation during the recession phase is positive; 0 otherwise. As explained in section 2, in our model real demand shocks impact aggregate demand through their e¤ect on accumulation of unused capacity by …rms. This variable aims at testing whether low aggregate demand following the recession could make the private sector reluctant to resume borrowing to …nance consumption or investment, leading to a contraction in the demand for credit and a higher likelihood of credit-less recovery. Financial demand shock dummy variable : This variable aims at capturing the role played by demand-side constraints to credit growth during the recovery. It consists of a binary variable taking the value of 1 if the correlation between the external …nance premium (EFP) and credit growth is positive during the recession ; 0 otherwise. Contractions in the demand for credit can result, for example, from a deterioration of borrowers’balance-sheets, which limits agents’ability to access new funding. Supply-side covariates Real supply shock dummy variable : This is a dichotomous variable taking the value of 1 if the correlation between output and in‡ation during the recession is negative; 0 otherwise. Inclusion of this variable is motivated by the aim of investigating whether or not the occurrence of a real supply shock in the economy increases the likelihood of a credit-less recovery. Financial supply shock dummy variable : This is a dichotomous variable taking the value of 1 if the correlation between the external …nancial premium (EFP) and the growth of domestic private credit is negative during recessions; 0 otherwise. A crisis episode accompanied by a …nancial supply shock can lead to excessive pressure on …nancial intermediaries’ balance-sheets. In turn, this may a¤ect overall levels of liquidity available for lending and lead to increases in borrowing costs. Credit ‡ow : This variables represents the annual growth rate of domestic bank credit (4dt 1 ). A reduction in the amount of credit in the economy, what we call deleveraging, reduces output. We expect this variable to be negatively correlated with the probability of a credit-less recovery. Credit impulse : As suggested by Biggs et al. (2009), a contraction in credit ‡ows reduces out- put. However, if credit reduces at a decreasing rate (4dt 1 < 4dt 2 ), a credit impulse e¤ect (4dt 1 4dt 2 ), may partially o¤set the negative e¤ect on output resulting from deleveraging, 4dt 1 :6 There are additional factors that can potentially impact economic conditions in the aftermath of a crisis, hence credit dynamics during the recovery. For example, if crisis episodes are preceded 6 Assume that a crisis in t 1 causes a contraction in credit and credit growth turns negative. If in year t, the pace of de-levering slows then 4dt or credit ‡ ow still remains negative but 4dt 4dt 1 or credit impulse turns positive. It is then possible that the growth of cyclical GDP is positive and there is an economic recovery that is accompanied by “de-leveraging” but is supported by a positive credit impulse. 15 by a credit boom, then credit dynamics during the recovery may look di¤erent than the case where the crisis was preceded by average credit growth. Similarly, credit dynamics during recoveries are in‡uenced by policy responses a¤ecting monetary aggregates. An expansionary monetary stance similar to the one that occurred in the Euro area following the global crisis contributes to increase liquidity levels and leads to more favorable funding conditions. To control for these conditions, we include two additional dummy variables, one signaling the occurrence of a credit boom prior to the crisis and another one controlling for the presence of a monetary expansion during the crisis.7 Our empirical strategy allows to test the hypotheses on credit-less recoveries through a compre- hensive framework of analysis featuring demand and supply side covariates, while at the same time controlling for country speci…c factors in‡uencing the drivers of credit growth. 3.3 Main results Our …rst step consists in estimating the …xed e¤ects Probit model (30) on the entire sample of advanced and emerging countries. Table 5 presents probit estimations for demand-side drivers of credit-less recoveries. Countries with negative GDP growth have a higher probability of experiencing a credit-less recovery. During economic downturns …nancially constrained …rms may be able to accu- mulate unused capacity, which can be used during the recovery phase to restore activity8 . Because of this “bounce-back”e¤ect, larger contractions of GDP during the recession increase the likelihood that production can recover without need of further lending, as …rms have access to unused capacity limiting their need for lending during the recovery phase of the cycle. In a similar way, large GDP contractions can make households reluctant to resume borrowing during the recovery, especially in the presence of a weak growth outlook. In our estimates, the occurrence of a real demand-side shock does not signi…cantly increases the probability of credit-less recovery episodes. When looking at credit growth dynamics, estimates indicate that the recovery in output during these episodes has a closer relationship with the change in the ‡ow of credit (i.e. 4dt 1 4 dt 2 or what is named credit impulse in our model) rather than with the change in the stock of credit.9 In other words, our results show that a rebound in output can occur in the presence of negative credit growth (or when the economy is deleveraging). However, in order to have an economic recovery, credit must be 7 Both of these variables are taken fom Laeven and Valencia’ s (2008) database on banking crisis. Credit booms are identi…ed as those during which the deviation of credit-to-GDP ratio relative to its trend is greater than 1.5 times its historical standard deviation and its annual growth rate exceeds 10 percent, or years during which the annual growth rate of the credit-to-GDP ratio exceeds 20 percent. Monetary expansions are computed as the change in the monetary base between its peak during the crisis and its level one year prior to the crisis. 8 This result is also in line with the …ndings of Sugawara and Zalduendo (2013) and Bijsterbosch and Dahlhaus (2011). 9 Biggs et. al. (2009) were the …rst authors to refer to a credit impulse e¤ect deriving from the change in the pace of deleveraging and these results are in line with their …ndings. 16 reducing at a decreasing rate (i.e. the pace of deleveraging is decreasing) so the credit impulse turns positive and this e¤ect may partially o¤set the negative consequences of deleveraging. This happens as investments in long-term capital respond sluggishly to credit, so when deleveraging slows down output can face a rebounding e¤ect during the recovery phase. Our model, in line with Biggs et al. (2009), con…rms the nature of the output-credit link during recoveries, showing that the rebound economic activity is linked to the change in the ‡ow rather than in the stock of credit. In principle, beyond the assumed sluggish e¤ect on capital investments, the positive credit impulse e¤ect can also re‡ect an improvement in general economic activity sup- ported by other factors such as real exchange rate devaluations or enhanced non-bank …nancing. The question of whether the reduction in the pace of deleveraging re‡ects improvements in general economic activity, or if it directly a¤ects aggregate output growth, represents an interesting path for new research. The occurrence of a …nancial demand shock signi…cantly increases the probability of credit-less episodes. This con…rms that demand-side frictions in …nancial markets may have indeed an impor- tant role in contributing to weak credit growth during the recovery phase of the cycle in the sample of countries. These frictions may manifest in diverse ways across countries. For example, in …nancial markets low demand for credit can be the result of a deterioration of borrowers’creditworthiness, or of the reluctance of the private sector (…rms and households) to resume borrowing in the presence of negative growth and employment prospects for the economy. However, these results do not exclude that supply constraints may have also played a role during recoveries. This is likely the case for recoveries preceded by …nancial sector stress (e.g. banking crises) which impact on …nancial interme- diaries’availability of liquidity for lending purposes. Our aim in this analysis is to demonstrate that during these episodes, demand constraints to credit growth weighed more than supply-side frictions. Table 5 shows that the occurrence of a credit boom prior to the crisis increases the likelihood of a credit-less recovery in our sample of countries, a result in line with the …ndings of Abiad et al. (2011). We expect that following a rapid increase in credit growth prior to the crisis, the economy will need to deleverage, leading to low credit demand in the aftermath of the recession. The likelihood of a credit-less recovery is in‡uenced by policy responses during the crisis. Mone- tary policy expansions during the crisis increase the probability of a credit-less recovery, suggesting that more favorable funding conditions resulting from an accomodating monetary stance did not support the pick-up of credit ‡ows to the private sector. This suggests that the failure of credit growth to pick-up during these episodes may have resulted from weak demand for liquidity by the private sector, corroborating our hypothesis of demand-driven credit-less recoveries presented above. To take into account the role of a potential correlation between panel-level entities and indepen- 17 dent variables, we estimate the same model using a random e¤ects panel Probit speci…cation and a Mundlak-type speci…cation. Results are presented in columns 2 and 3 of Table 5. Results obtained with these speci…cations are in line with the …xed e¤ects model estimations, although some of the coe¢ cients vary in signi…cance levels. The outcome of the Hausman test (p-value = 0:92) implies that the null hypothesis that the models yield similar coe¢ cients is not rejected at any reasonable signi…cance level, and that the random e¤ect model yields a more e¢ cient estimator of the probabil- ity of credit-less recovery. We perform an additional test on the panel-level means of the Mundlak error-correction model presented in column 3 of Tables 5. The p-value of the test is 0.60 suggesting that the time-invariant unobservables are not related to the independent covariates, and that the model satis…es the random e¤ects model assumptions. In our sample, seven (or 33 percent) of the overall credit-less recoveries occurred after the global …nancial crisis erupted in 2007. This period has been characterized by a sharp contraction in credit in many countries worldwide, driven by both demand and supply factors. To assess how our results change when these observations are removed, we estimate our model excluding recovery episodes that occurred after 2010. Results are presented in Tables 7 and 8. Although some of our coe¢ cients vary in signi…cance, overall results are in line with our estimates and con…rm that on average demand-side factors have a stronger predictive power on credit-less recoveries. Table 6 presents estimation results obtained analyzing supply-side drivers of credit-less recov- eries. Evidence from the real side of the economy con…rms that negative GDP growth increases the probability of credit-less episodes. Recessionary episodes featuring real supply shocks increase the likelihood of credit-less episodes. When looking at the role played by …nancial sector drivers of credit growth, …ndings show that supply-side shocks in …nancial markets did not play on average a signi…cant role in favoring the occurrence of credit-less recoveries. Although we cannot exclude the possibility that in many crisis episodes, especially those associated with banking crises, disruptions in the supply of credit may have played a role in keeping credit low during the recovery, our results suggest that on average credit-less episodes have been an outcome of subdued demand for liquidity by the private sector in the aftermath of the crisis. Columns 2 and 3 of Table 6 present estimations of the random e¤ects panel Probit speci…cation and a Mundlak-type speci…cation. Similarly to above, we perform a Hausman test for the hypothesis that the di¤erences between the coe¢ cients of the …xed and random e¤ects models are not systemic, …nding a p-value of 0.86. Testing at the 5 and 10 percent signi…cance levels, the null hypothesis that both the …xed and random e¤ects models yield similar coe¢ cients is not rejected. Estimation coe¢ cients presented in Table 5 and Table 6 are not interpretable using standard inference methods. As the next step, we compute the marginal e¤ects for changes in the explanatory 18 variables. Table 11 reports the marginal e¤ects of the probability of credit-less recovery and Figure 4 presents the predicted probabilities of credit-less recovery using our demand and supply-side models. The coe¢ cients of the the marginal e¤ects and patterns of predicted probabilities for di¤erent values s covariates con…rm our results. They show that when a recovery is preceded by a of the model’ demand shock in …nancial markets, the average probability of a credit-less recovery is 6 percent higher. There are no signi…cant e¤ects found for supply-side shocks. 3.4 Robustness This section performs a robustness analysis of the models presented in the previous section. The goal is to assess the overall performance of the models in predicting credit-less episodes. The analysis is carried out following two procedures similar to Bijsterbosch and Dahlhaus (2011). As a …rst step, we assesses the sensitivity of the results presented in the previous section to an alternative de…nition of credit-less recovery. According to this one, a recovery is identi…ed as credit-less when annual real credit growth is negative during the …rst three years of recovery. Model estimations for the demand and supply-side of the economy are presented in Table 9 and Table 10. For each model, most of the signs of the covariates remain unchanged, although some of the regressors vary in signi…cance levels. Overall, results suggest that the models are robust to alternative de…nitions of credit-less recoveries. As a second step, we analyze the baseline models’predictive performance by calculating type I and type II errors (or false positive and false negative respectively).10 Calculating these errors requires s predicted probability for observation setting a positive outcome threshold : Letting pi be the model’ i, and xi a binary variable signaling the occurrence of the actual outcome. The classi…cation signals a false positive (type I error) if pi > and xi = 0. On the contrary, if pi < and xi = 1, it signals a false negative (type II error). Following the literature, we set the thresholds equal to 30, 40 and 50 percent. Table 12 reports type I, type II errors and overall success rates for the models presented in section 3.2 for di¤erent thresholds values of . By setting = 0:5, the models have a success rate between 88 and 87 percent. The probability that the models incorrectly predict a credit-less episode (type I error) is 5.3 and 4.5 percent for the demand and supply-side models respectively. The probability that the models fail to predict a credit-less episode (type II error) is between 27 and 33 percent for the demand and supply-side respectively. Overall, the analysis suggests that both models have a good predictive power. 10 A similar analysis is also done by Bijsterbosch and Dahlhaus (2011) and Sugawara and Zalduendo (2013), to assess the predictive ability of a probabilistic model of credit-less recoveries. 19 4 Conclusions The disruption of bank lending following the global …nancial crisis that erupted in 2007 resulted in sharp output contractions in both emerging and advanced countries. This brought up the question of how the duration of credit tightening and impaired bank intermediation a¤ect output and its recovery patterns. Recent trends in private sector deleveraging and balance-sheet adjustment suggest that subdued credit growth during the recovery represented an obstacle for growth. Against this background, policy responses have been substantial, ranging from increases in …scal spending, sizable monetary expansions, and bank recapitalizations.11 Despite these interventions, in many countries worldwide recoveries have been weak, and often accompanied by a modest pick-up in credit ‡ows. In their pioneering study on sudden stops in capital ‡ows, Calvo et al. (2006) were the …rst authors to document credit-less recoveries in emerging markets, naming them Phoenix Miracles. Subsequent empirical literature shows that these episodes are a common feature of business cycles in both advanced and emerging economies, and that they bring relevant costs for countries. Output growth is between 2 and 3 percent lower than during "normal" recoveries, and it takes longer for it to recover to its pre-crisis levels. An open question remains whether weak credit growth is a result of demand or supply constraints to lending activity. Providing evidence on this point is important not only for economic theory but also for policy design. The lack of comprehensive empirical evidence in this area has limited economic research in providing detailed insights on the role of policy responses following these episodes. The primary goal of this work was to analyze demand and supply drivers of bank credit growth during economic recoveries. Building on Chadha et al. (2010), we analyze how liquidity shocks impact on credit conditions in an endowment economy model. In our framework, credit-less episodes are correlated with demand and supply shocks in real and …nancial markets and with the pace of private sector deleveraging. As in Biggs et al. (2009), we show that a recovery in output can occur in the presence of negative credit growth as long as the pace of deleveraging is decreasing. We use panel data on output, credit and the external …nance premium for a sample of 42 advanced and emerging countries between 1980 and 2014 to identify demand and supply shocks in real and …nancial markets. Through panel Probit model techniques we analyze the relative impact of these shocks in predicting the onset of a credit-less recovery. Findings show that in the sample of advanced and emerging countries, stagnant credit growth during credit-less recoveries has been (on average) primarily demand-constrained. These episodes appear the outcome of weak demand for liquidity by the private sector in the aftermath of recessionary 11 See for example Jiménez et al. (2017) for evidence on the e¤ect of countercyclical bank capital bu¤ers and credit supply. 20 episodes. Our results con…rm that during these episodes output growth is correlated with the pace of private sector deleveraging and that …rms may be able to restore activity by accessing unused capacity accumulated during the recession, while postponing additional credit-intensive investments. These results, however, do not exclude that bottlenecks to the supply of credit have also contributed to sluggish credit growth. This likely occurred during recessionary episodes accompanied by …nancial sector stress (e.g. banking crisis) leading to contractions of liquidity levels and higher lending costs. These results are relevant from a policy perspective and show that demand stimuli policies during credit-less recoveries are likely to lead to higher growth. To the extent that credit-less recoveries are undesirable outcomes from a growth perspective, policy measures should aim at preventing large contractions in aggregate demand and at promoting macroeconomic stability ex-ante. In the case of occurrence of these episodes, …scal and …nancial sector policy can tackle directly demand constraints to credit growth. For example, low demand for credit may be caused by high collateral requirements or by excessive private sector indebtedness. In these cases, policy responses should prioritize measures aimed at loosening collateral rules and favoring debt restructuring keeping demand for credit low. 21 References Ariccia G and Li B. (2011). “Creditless Recoveries” [1] Abiad A, Dell’ . IMF Working Paper 11/58, IMF, Washington, DC. [2] Adrian T, Colla P and Shin H. (2011). “Which Financial Frictions? Parsing the Evidence from the Financial Crisis of 2007-09”. Federal Reserve Bank of New York Sta¤ Report No. 528. [3] Ayyagari M, Demirguc-Kunt A and Maksimovic V. (2011). “Do Phoenix Miracles Exist? Firm- Level Evidence from Financial Crises”. Policy Research Working Paper 5799, World Bank, Wash- ington, DC. [4] Bernanke B and Gertler M. (1989). “Agency Costs, Net Worth, and Business Fluctuations” American Economic Review 79: 14-31. [5] Bernanke B and Gertler M. (1990). “Financial fragility and economic performance”. The Quar- terly Journal of Economics 105(1): 87-114. [6] Bernanke B, Gertler M and Gilchrist S. (1999). “The …nancial accelerator in a quantitative business cycle framework”. In J. Taylor and M. Woodford, editors, The Handbook of Macroeco- nomics, pages 1341–1393. Elsevier Science B.V., Amsterdam, The Netherlands [7] Biggs M, Mayer T and Pick A. (2009). “Credit and Economic Recovery”. DNB Working Paper 218, De Nederlandsche Bank, Amsterdam, Netherlands. [8] Bijsterbosch M and Dahlhaus T. (2011). “Determinants of Credit-Less Recoveries”. ECB Work- ing Paper 1358, European Central Bank, Frankfurt, Germany. [9] Bodnár K, Kovalszky Z and Hudak EK. (2014). “Recovery from Crisis and Lending”. Financial and Economic Review 13 (4): 57-85. [10] Braun M and Larrain B. (2005). “Finance and the Business Cycle: International,Inter-Industry Evidence,”Journal of Finance 15(3): 1097-128. [11] Calvo GA, Izquierdo A and Talvi E. (2006). “Phoenix Miracles in Emerging Markets: Recovering without Credit from Systemic Financial Crisis”. NBER Working Paper No. 12101. [12] Chadha J, Corrado L and Sun Q. (2010). “Money, Prices and Liquidity E¤ects: Separating Demand from Supply”. Journal of Economic Dynamic and Control 34(9): 1732-1747. 22 [13] Claessens S, Kose MA and Terrones M. (2011). “Financial Cycles: What? How? When?". Working Paper 11/76. International Monetary Fund. Washington DC. [14] Claessens S, Kose MA and Terrones M. (2009)."A Recovery Without Credit: Possible, but...". Voxeu.org May 22nd 2009 [15] Claessens S, Kose MA and Terrones M. (2008). " What Happens During Recessions, Crunches and Busts?". IMF Working Paper No. 08/274. [16] Cooley, T. F. and L. H. Ohanian (1991). "The Cyclical Behavior of Prices." Journal of Monetary Economics 28: 25-60. [17] Coricelli F and Frigerio M. (2015). "The Credit-Output Relationship During the Recovery from Recession". Open Economies Review 26(3):551-579. Ariccia G and Marquez R. (2006). "Lending Booms and Lending Standards". The Journal [18] Dell’ of Finance 61(5): 2511-2546. [19] den Haan, Wouter J. (2000). "The Comovement between Output and Prices." Journal of Mon- etary Economics 46: 3-30. [20] European Commission (2009). "Impact of the Current Economic and Financial Crisis on Poten- tial Output". Occasional paper, European Commission. [21] Faccini, R and Yashiv E. (2015). "Frictions in DGSE Models: Revisiting New Kenesian vs New Classical Results. Working paper. [22] Furceri, D. and Mourougane, A. (2009). "The E¤ect of Financial Crises on Potential Output: New Empirical Evidence from OECD Countries." OECD Economics Department Working Pa- pers 699, OECD Publishing. [23] Guerrieri V and Lorenzoni G. (2011). “Credit Crises, Precautionary Savings and the Liquidity Trap”. NBER Working Paper No. 17583. [24] Holmstrom B and Tirole J. (1997). “Financial Intermediation, Loanable Funds and the Real Sector”. The Quarterly Journal of Economics 112(3): 663-691. [25] Huntley J. (2008). “Phoenix Falling: Recovering from Sudden Stops in Emerging Markets”. Unpublished mimeo, Northwestern University, Evanston, IL. 23 s the Damage? Medium-term Output Dy- [26] International Monetary Fund IMF (2009). “What’ namics After Banking Crises”. IMF Working Paper 09/245, Washington DC. [27] International Monetary Fund IMF (2013). “Global Financial Stability Report - Transition Chal- lenges to Stability”. Washington DC. [28] Jiménez G, Ongena S, Peydró JL and Saurina J. (2017). “Macroprudential Policy, Countercycli- cal Bank Capital Bu¤ers and Credit Supply: Evidence from the Spanish Dynamic Provisioning Experiments“. Journal of Political Economy 125 (6): 2126-2177. [29] Kiyotaki N and Moore J. (1997). “Credit cycles”. Journal of Political Economy 105(2): 211-248. [30] Laeven L and Valencia F. (2012). "Systemic Banking Crisis Database: An update. IMF Working Paper No.12/163, IMF, Washington DC [31] Mundlak Y. (1978). “On the pooling of time series and cross section data”. Econometrica 46: 69-85. [32] Pakko, Michael R. (2000). "The Cyclical Relationship between Output and Prices: An Analysis in the Frequency Domain." Journal of Money, Credit and Banking 32(1): 382-399. [33] Pisani-Ferry J. and van Pottelsberghe B. (2009). "Handle with Care! Post-crisis Growth in the EU." Policy Briefs 301, Bruegel. [34] Smetz F and Wouters R. (2007). "Shocks and Frictions in US Business Cycles: A Bayesian DGSE Approach". American Economic Review 97(3): 586-606. [35] Smith, R. T. (1992). "The Cyclical Behavior of Prices." Journal of Money, Credit, and Banking, 24: 413-30. [36] Sugawara N. and Zalduendo J. (2013). "Credit-less Recoveries: Neither a Rare nor an Insur- mountable Challenge". World Bank Policy Research Working Paper No. 6459. The World Bank, Washington DC. 24 Figure 1: The business and the …nancial cycles,1980 - 2014 United kingdom Thailand 25 Figure 2: Anatomy of Credit-less Recoveries Average percentage (%) growth during recoveries, 1980-2014 GDP Investment Consumption 26 Figure 3: Frequency of Aggregate Shocks and Credit-less Recoveries Number of demand and supply shocks and credit-less recoveries, by year and country group Real Sector Advanced countries Emerging countries Financial Sector Advanced countries Emerging countries 27 Figure 4a: Predicted probabilities with 95% CI Demand Supply .8 .5 .6 .4 Pr(Y=1) .4 Pr(Y=1) .3 .2 .2 .1 0 -10 -7 -4 -1 2 5 8 11 14 -10 -7 -4 -1 2 5 8 11 14 GDP growth (%) GDP growth (%) .8 .8 .6 .6 Pr(Y=1) .4 Pr(Y=1) .4 .2 .2 0 0 0 1 0 1 Real Demand Shock Real Supply Shock .8 .6 .6 .4 Pr(Y=1) Pr(Y=1) .4 .2 .2 0 0 -12 -4 4 12 -12 -4 4 12 Credit flow (%) Credit flow (%) 28 Figure 4b: Predicted probabilities with 95% CI Demand Supply .8 1 .8 .6 .6 Pr(Y=1) .4 Pr(Y=1) .4 .2 .2 0 0 -5 5 15 25 35 45 -5 5 15 25 35 45 Credit im pulse Credit impulse .6 1 .8 .4 .6 Pr(Y=1) Pr(Y=1) .4 .2 .2 0 0 0 1 0 1 Fiancial Demand Shock Financial Suppl y Shock 29 Table 1: Frequency of credit-less recoveries Sample No. of recoveries Credit-less Credit-less Def. 2 Credit-less Credit-less Credit-less Def. 1 Def. 2 Def. 3 Def. 4 Def. 5 Advanced countries Number 55 5 9 15 7 10 Percentage (%) 51 9 16 27 12 18 Emerging countries Number 52 7 12 26 15 10 Percentage (%) 48 13 23 50 29 19 Whole sample Number 107 14 21 41 22 20 Percentage (%) 100 13 19 38 20 18 Note: According to De…nition 1 a recovery is de…ned as credit-less if the annual growth rate of domestic real bank credit is negative during the …rst three years of the recovery (i.e t+1,t+2 and t+3);. In the case of De…ntion 2, a recovery is de…ned as credit-less if the annual growth rate of domestic real credit is negative during the …rst two years of recovery (i.e t+1 and t+2 ). Based on De…ntion 3, a recovery is de…ned as credit-less if the annual growth rate of domestic real credit is negative during the …rst year of recovery (i.e t+1). According to De…nition 4, a recovery is de…ned as credit-less if the average annual growth rate of domestic real bank credit is negative during the …rst four years of the recovery (i.e. between t+1 and t+4); In the case of De…nition 5, a recovery is de…ned as credit-less if the average annual growth rate of real bank credit is negative during the …rst four years of the recovery, starting from t+2 (i.e. between t+2 and t+5). Table 2: Price-quantity correlations during recessions Sample Correlation output gap/in‡ation Correlation EFP/Credit growth Advanced countries 1980-1990 0:05 0:29 1991-2000 0:06 0:33 2001-2014 0:33 0:04 Emerging countries 1980-1990 0:06 0:22 1991-2000 0:25 0:09 2001-2014 0:41 0:32 Whole sample 1980-1990 0:00 0:25 1991-2000 0:15 0:12 2001-2014 0:37 0:18 Note: The table reports correlation coe¢ cients between output gap growth and in‡ation and between the external …nance premium (EFP) and credit growth during the recession phase of the cycle. The length (years) of the recession varies by country. Table 3: Credit-less Recoveries and Growth Performance in Advanced Countries Country No. of Output Peak to Recovery Credit-less GDP Growth (%) GDP growth (%) GDP growth (%) GDP Growth (%) Credit Growth (%) Collapses Dates Recovery at peak during recession Trough to Peak During Recovery During Recovery Australia 2 1981 1987 No 3:35 0:54 5:58 4:13 15:41 1990 1996 No 3:52 0:00 3:13 3:98 7:80 Austria 3 1983 1991 No 2:97 1:55 1:61 3:74 7:29 2000 2007 No 3:36 1:25 2:61 2:95 5:02 2008 2013 No 1:54 3:37 5:34 1:45 0:05 Canada 3 1981 1986 No 3:50 3:02 6:52 3:75 2:10 1989 1996 No 2:37 0:37 1:52 2:89 3:58 2007 2013 2:00 0:76 4:71 2:56 Denmark 4 1985 No 3:63 10:13 1986 1997 No 4:94 0:78 5:03 3:68 1:43 2000 2007 No 3:74 0:55 3:35 2:42 9:72 2007 2013 Y es 0:82 2:90 5:91 0:40 2:32 Finland 2 1990 1997 Y es 0:67 3:32 1:41 4:51 5:91 2008 2013 No 0:72 8:26 8:98 0:75 3:34 France 4 1981 1991 No 1:07 1:97 1:49 3:26 6:92 1990 1997 No 2:91 0:67 3:52 2:03 0:78 2000 2007 No 3:87 1:29 3:05 2:28 11:02 2007 2013 No 2:36 1:37 5:30 1:22 1:41 Germany 2 2001 2009 No 1:69 0:29 0:98 2:68 0:08 2008 2013 No 1:08 5:61 6:70 2:11 4:76 Greece 3 1985 1991 No 2:50 0:87 4:76 2:79 0:72 2004 2009 No 5:06 0:59 4:46 1:07 5:39 2008 current Y es 0:33 6:30 6:96 Iceland 3 1981 1987 No 4:26 0:00 6:41 5:55 4:08 1987 1996 No 8:54 0:45 11:91 2:45 3 :5 2000 2007 No 4:72 2:31 1:98 6:96 29:85 Ireland 4 1985 1990 No 3:08 0:42 3:51 6:04 8:82 1990 1998 No 8:46 3:43 2:71 9:74 31:76 2000 2008 No 10:22 5:00 5:82 4:00 23:0 2007 2013 Y es 5:54 3:90 11:18 1:14 7:90 Israel 2 1987 1998 No 7:18 5:07 0:25 9:08 8:97 2000 2007 No 8:94 0:45 7:76 5:36 2:23 Italy 3 1980 1987 No 3:43 0:80 2:26 3:01 3:74 1989 1997 Y es 3:38 0:87 4:24 2:04 0:13 2007 2013 No 1:47 3:26 6:95 0:56 1:63 Japan 3 1985 1991 No 6:33 3:46 2:22 5:35 7:69 1991 1998 No 3:32 0:61 2:46 1:03 3:48 2007 2013 No 2:19 3:28 7:71 1:89 2:50 New Zeland 4 1981 1986 No 4:65 0:93 3:72 3:15 11:55 1986 1996 No 2:70 0:15 1:59 4:93 8:66 1996 2002 No 3:62 1:28 3:01 4:14 3:31 2007 2013 2:95 0:93 3:21 2:07 Norway 2 1980 1986 No 4:56 0:91 4:32 4:90 19:84 1997 2007 No 5:28 2:04 4:36 2:97 Spain 3 1980 1990 No 2:20 1:70 1:04 4:81 9:90 1991 1997 No 2:54 0:05 3:57 2:87 3:55 2008 current 1:11 1:77 2:78 Sweden 2 1990 1997 Y es 0:75 1:45 2:82 3:13 1:03 2007 2013 No 3:40 2:87 8:58 2:40 3:63 Switzerland 3 1981 1987 No 1:60 0:33 0:96 2:53 5:41 2000 2007 No 3:94 0:54 3:89 3:50 5:75 2008 2013 No 2:27 2:12 4:40 1:91 3:45 UK 3 1985 No 3:02 12:55 1988 1996 No 5:93 0:56 5:48 3:56 5:04 2007 2013 Y es 2:58 2:32 6:77 1:71 4:79 USA 4 1986 No 4:91 7:52 1989 1995 Y es 3:68 0:92 3:75 3:26 1:56 2000 2006 No 4:09 1:38 2:30 3:15 6:52 2007 2013 Y es 1:77 1:53 4:55 2:16 Note: Tables 1 and Table 2 report average annual growth rates of real GDP and real domestic bank credit in selected advanced and emerging countriess. Asterisks (*) indicate missing data Table 4: Credit-less Recoveries and Growth Performance in Emerging Countries Country No. of Output Peak to Recovery Credit-less GDP Growth (%) GDP growth (%) GDP growth (%) GDP Growth (%) Credit Growth (%) Collapses Dates Recovery at peak during recession Trough to Peak During Recovery During Recovery Algeria 3 1985 No 5:27 12:89 1985 1992 Y es 3:69 0:43 4:7 0:74 29:04 1992 1998 No 1:80 1:5 2:7 3:52 3:31 Argentina 2 1987 1994 No 2:90 4:15 5:30 9:08 17:43 1998 2006 Y es 3:85 4:86 14:74 8:86 0:46 Brazil 3 1980 1987 No 9:11 2:40 11:52 6:20 1987 1996 No 3:59 0:22 4:06 4:15 4:21 2008 2013 No 5:01 0:23 5:25 3:99 12:32 Chile 2 1981 1987 No 4:73 7:05 8:52 6:82 1:03 2007 2013 No 5:16 1:12 6:19 5:31 7:95 China 2 1986 12:10 1988 1995 No 11:30 5:80 2:03 13:07 12:38 Colombia 4 1981 1098 2:26 2:24 0:82 4:67 1990 1997 No 6:04 3:22 3:67 4:13 5:98 1997 2003 No 3:43 1:91 7:63 3:13 5:32 2007 2014 No 6:90 3:05 2:92 5:18 12:67 Hungary 2 1997 Y es 1:97 1:24 2008 2013 No 0:83 6:55 7:39 0:67 5:45 India 4 1983 1991 No 7:28 4:45 3:32 5:54 4:03 1990 1997 No 5:53 3:76 0:78 6:45 6:19 1999 2006 No 8:84 4:15 5:04 8:58 17:50 2007 2012 No 9:80 3:89 5:91 7:61 9:35 Indonesia 1 1997 2002 Y es 4:69 13:12 17:82 3:46 13:37 Korea 2 1984 No 9:43 11:85 1997 2002 No 5:76 5:71 11:48 7:87 27:81 Malaysia 3 1984 1991 No 7:76 1:80 2:37 9:38 5:56 1997 2002 Y es 7:32 7:35 14:68 5:22 0:70 2008 2013 No 4:83 1:51 6:34 5:72 7:73 Mexico 3 1981 1987 No 8:77 2:41 12:96 1:08 2:09 1994 1999 No 4:72 5:75 10:48 5:05 1:94 2007 2013 No 3:22 1:68 7:96 3:63 9:34 Peru 2 1981 1987 Y es 5:55 5:31 15:96 6:20 8:11 1987 1994 No 9:72 8:91 14:70 4:80 20:59 Philippines 1 1983 1989 No 1:87 7:31 9:18 5:17 2:52 Poland 1 1995 No 4:62 1:73 Russia 3 1991 1999 Y es 2:11 7:12 1997 2002 No 1:4 5:3 6:7 6:55 11:34 2002 2013 No 4:74 4:99 12:56 3:37 7:63 South Africa 3 1981 1987 No 5:36 1:11 7:20 1:50 1:06 1989 1996 No 2:39 1:15 4:53 2:95 4:43 2008 2013 Y es 3:19 1:53 4:72 2:67 0:11 Thailand 3 1984 1990 No 5:75 5:09 0:21 11:54 22:88 1996 2002 Y es 5:65 5:19 13:28 4:65 6:22 2007 2013 No 5:43 0:49 6:17 4:61 10:25 Turkey 3 1993 1998 No 7:65 4:66 12:31 6:28 12:12 2000 2005 No 6:77 5:69 12:47 7:29 18:83 2007 2013 No 4:66 2:08 9:49 6:06 23:19 Ukraine 2 1992 1998 Y es 6:77 2008 2013 Y es 2:3 14:8 17:1 2:4 2:91 Uruguay 2 1981 1987 No 1:55 10:01 11:83 4:28 4:96 1998 2006 Y es 4:51 3:86 12:25 4:34 17:17 Venezuela 3 1988 1993 No 5:82 8:56 14:39 5:63 0:67 2001 2007 No 3:39 8:30 11:14 11:80 42:87 2008 2014 No 5:27 2:34 6:76 3:71 Table 5: Determinants of Credit-less Recoveries: Demand-side Contributions (1) (2) (3) Pooled Probit Panel Probit Mundlak Correction GDP growtht 1 0:064* 0:054** 0:047** (0:038) (0:022) (0:021) Real Demand Shockrecession 0:527 0:537 0:394 (0:392) (0:578) (0:623) Credit ‡owt 1 ( Dt 1 ) 0:072*** 0:059*** 0:062*** (0:026) (0:019) (0:021) Credit impulset 1 ( Dt 2 Dt 1 ) 0:035** 0:028*** 0:029*** (0:013) (0:009) (0:011) Financial Demand Shockrecession 1:472*** 1:252** 1:373*** (0:333) (0:487) (0:517) Developing dummy 0:601 0:168 0:048 (0:801) (0:619) (0:791) Credit boomrecession 1:675*** 1:580 1:688 (0:501) (0:990) (1:120) Monetary expansion 1:687*** 1:753* 1:750* (0:497) (0:919) (0:983) Constant 1:006*** 3:336*** 4:215*** (0:281) (0:814) (1:151) Country E¤ects Yes No Yes Observations 179 450 450 Pseudo-Likelihood -42.0 -73.3 -70.5 Note: Regressions are Probit estimates, values in parenthesis are robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. Real demand shock is a DV variable taking the value of 1 if during the recession, the correlation between output gap and in‡ation is positive; 0 otherwise. Financial demand shock is a DV taking the value of 1 if during the recession, the correlation between the EFP and credit growth is positive; 0 otherwise. Credit boom is a DV taking the value of 1 if during the recession the country experienced a credit boom, 0 otherwise. Monetary expansion is a DV taking the value of 1 if during the recession the country experienced a monetary expansion, 0 otherwise. Table 6: Determinants of Credit-less Recoveries: Supply-Side Contributions (1) (2) (3) Pooled Probit Panel Probit Mundlak Correction GDP growth t 1 0:079** 0:063*** 0:056** (0:035) (0:022) (0:022) Real Supply Shock recession 0:545 0:287 0:528 (0:432) (0:633) (0:632) Credit ‡ow t 1( Dt 1) 0:048*** 0:044*** 0:046*** (0:018) (0:012) (0:014) Credit impulset 1( Dt 2 Dt 1) 0:022** 0:020*** 0:021*** (0:009) (0:006) (0:007) Financial Supply Shock recession 0:272 0:226 0:228 (0:389) (0:541) (0:486) Developing dummy 0:752 0:222 0:157 (0:700) (0:528) (0:577) Credit boom recession 2:116*** 1:772** 1:912** (0:364) (0:718) (0:741) Monetary expansion 1:607*** 1:693** 1:667** (0:427) (0:712) (0:726) Constant 0:525* 2:524*** 1:809* (0:287) (0:745) (0:951) Country E¤ects Yes No Yes Observations 179 450 450 Pseudo-Likelihood -53.6 -85.2 -80.5 Note: Regressions are Probit estimates, values in parenthesis are robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. Real supply shock is a DV variable taking the value of 1 if during the recession, the correlation between output gap and in‡ation is negative positive; 0 otherwise. Financial supply shock is a DV taking the value of 1 if during the recession, the correlation between the EFP and credit growth is negative; 0 otherwise. Credit boom is a DV taking the value of 1 if during the recession the country experienced a credit boom, 0 otherwise. Monetary expansion is a DV taking the value of 1 if during the recession the country experienced a monetary expansion, 0 otherwise. Table 7: Determinants of Credit-less Recoveries: Demand-side Contributions Prior to the global …nancial crisis (<2010) (1) (2) (3) Pooled Probit Panel Probit Mundlak Correction GDP growtht 1 0:005 0:029 0:014 (0:030) (0:026) (0:025) Real Demand Shockrecession 0:656 0:609 0:494 (0:470) (0:620) (0:668) Credit ‡owt 1 ( Dt 1 ) 0:065** 0:048*** 0:055*** (0:029) (0:017) (0:020) Credit impulset 1 ( Dt 2 Dt 1 ) 0:031** 0:023*** 0:026** (0:015) (0:009) (0:011) Financial Demand Shockrecession 1:688*** 1:338** 1:474*** (0:433) (0:597) (0:653) Developing dummy 0:308 0:092 0:020 (0:773) (0:609) (0:808) Credit boomrecession 2:184*** 1:803* 2:113* (0:842) (1:037) (1:195) Monetary expansion 1:473** 1:790** 1:663* (0:658) (0:837) (0:915) Constant 1:209*** 3:425*** 4:185*** (0:294) (0:716) (1:108) Country E¤ects Yes No Yes Observations 144 375 375 Pseudo-Likelihood -28.6 -58.7 -55.6 Note: Regressions are Probit estimates, values in parenthesis are robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. Real demand shock is a DV variable taking the value of 1 if during the recession, the correlation between output gap and in‡ation is positive; 0 otherwise. Financial demand shock is a DV taking the value of 1 if during the recession, the correlation between the EFP and credit growth is positive; 0 otherwise. Credit boom is a DV taking the value of 1 if during the recession the country experienced a credit boom, 0 otherwise. Monetary expansion is a DV taking the value of 1 if during the recession the country experienced a monetary expansion, 0 otherwise. Table 8: Determinants of Credit-less Recoveries: Supply-Side Contributions Prior to the global …nancial crisis (<2010) (1) (2) (3) Pooled Probit Panel Probit Mundlak Correction GDP growth t 1 0:085** 0:068*** 0:055** (0:037) (0:027) (0:026) Real Supply Shock recession 0:543 0:292 0:522 (0:508) (0:688) (0:682) Credit ‡ow t 1( Dt 1) 0:056** 0:405*** 0:048*** (0:024) (0:011) (0:014) Credit impulset 1( Dt 2 Dt 1) 0:027** 0:019*** 0:023*** (0:012) (0:006) (0:007) Financial Supply Shock recession 0:248 0:175 0:151 (0:448) (0:649) (0:566) Developing dummy 0:650 0:139 0:135 (0:830) (0:484) (0:576) Credit boom recession 2:326*** 1:645** 1:988** (0:659) (0:755) (0:821) Monetary expansion 1:250** 1:536** 1:409* (0:576) (0:732) (0:788) Constant 0:055* 2:413*** 1:828** (0:321) (0:681) (0:899) Country E¤ects Yes No Yes Observations 144 375 375 Pseudo-Likelihood -37.9 -67.5 -62.7 Note: Regressions are Probit estimates, values in parenthesis are robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. Real supply shock is a DV variable taking the value of 1 if during the recession, the correlation between output gap and in‡ation is negative positive; 0 otherwise. Financial supply shock is a DV taking the value of 1 if during the recession, the correlation between the EFP and credit growth is negative; 0 otherwise. Credit boom is a DV taking the value of 1 if during the recession the country experienced a credit boom, 0 otherwise. Monetary expansion is a DV taking the value of 1 if during the recession the country experienced a monetary expansion, 0 otherwise. Table 9: Robustness Analysis: Demand-side Drivers of Credit-less Recoveries Alternative de…nition of credit-less recovery (Def. 1) (1) (2) (3) Pooled Probit Panel Probit Mundlak Correction GDP growtht 1 0:008 0:015 0:009 (0:044) (0:030) (0:029) Real Demand Shockrecession 0:068 0:022 0:045 (0:560) (0:732) (0:773) Credit ‡owt 1 ( Dt 1 ) 0:049** 0:053*** 0:051** (0:024) (0:019) (0:020) Credit impulset 1 ( Dt 2 Dt 1 ) 0:014* 0:017*** 0:016*** (0:008) (0:005) (0:006) Financial Demand Shockrecession 1:178*** 1:046* 1:014* (0:434) (0:539) (0:560) Developing dummy 1:656** 0:562 0:481 (0:783) (0:733) (1:079) Credit boomrecession 1:998*** 1:824* 1:791* (0:444) (0:972) (1:017) Monetary expansion 0:700 0:510 0:576 (0:496) (0:921) (0:999) Constant 1:708*** 3:309*** 2:687** (0:325) (0:824) (1:3232) Country E¤ects Yes No Yes Observations 117 450 450 Pseudo-Likelihood -32.6 -57.1 -55.5 Note: Regressions are Probit estimates, values in parenthesis are robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. A recovery is de…ned as credit-less if the annual growth of domestic bank credit is negative during the …rst 3 years of recovery. Real demand shock is a DV variable taking the value of 1 if during the recession, the correlation between output gap and in‡ation is positive; 0 otherwise. Financial demand shock is a DV taking the value of 1 if during the recession, the correlation between the EFP and credit growth is positive; 0 otherwise. Credit boom is a DV taking the value of 1 if during the recession the country experienced a credit boom, 0 otherwise. Monetary expansion is a DV taking the value of 1 if during the recession the country experienced a monetary expansion, 0 otherwise. Table 10: Robustness Analysis: Supply-side Drivers of Credit-less Recoveries Alternative de…nition of credit-less recovery (Def. 1) (1) (2) (3) Pooled Probit Panel Probit Mundlak Correction GDP growth t 1 0:013 0:030 0:025 (0:045) (0:030) (0:030) Real Supply Shock recession 0:939* 0:711 0:657 (0:552) (0:734) (0:761) Credit ‡ow t 1( Dt 1) 0:041* 0:048*** 0:048*** (0:022) (0:013) (0:015) Credit impulset 1( Dt 2 Dt 1) 0:012* 0:015*** 0:015*** (0:007) (0:004) (0:005) Financial Supply Shock recession 0:525 0:379 0:213 (0:514) (0:626) (0:595) Developing dummy 2:112* 0:687 0:372 (0:828) (0:717) (0:943) Credit boom recession 2:400*** 2:068** 2:069** (0:438) (0:971) (1:018) Monetary expansion 0:635* 0:415 0:411 (0:385) (0:749) (0:755) Constant 1:730*** 3:053*** 3:000* (0:332) (1:025) (1:807) Country E¤ects Yes No Yes Observations 117 450 450 Pseudo-Likelihood -35.2 -60.0 -57.9 Note: Regressions are Probit estimates, values in parenthesis are robust standard errors. *** p<0.01, ** p<0.05, * p<0.1. A recovery is de…ned as credit-less if the annual growth of domestic bank credit is negative during the …rst 3 years of recovery. Real supply shock is a DV variable taking the value of 1 if during the recession, the correlation between output gap and in‡ation is negative positive; 0 otherwise. Financial supply shock is a DV taking the value of 1 if during the recession, the correlation between the EFP and credit growth is negative; 0 otherwise. Credit boom is a DV taking the value of 1 if during the recession the country experienced a credit boom, 0 otherwise. Monetary expansion is a DV taking the value of 1 if during the recession the country experienced a monetary expansion, 0 otherwise. Table 11: Marginal E¤ects for Changes in Explanatory Variables Demand-Side Marginal E¤ects GDP growtht 1 0:002* (0:001) Real Demand Shockrecession 0:026 (0:028) Credit ‡owt 1 0:002** (0:001) Credit impulset 1 0:001** (0:000) Financial Demand Shock recession 0:061* (0:032) Developing dummy 0:008 (0:030) Credit boomrecession 0:077 (0:047) Monetary expansion 0:085* (0:041) Supply-Side Marginal E¤ects GDP growtht 1 0:002* (0:001) Real Supply Shockrecession 0:013 (0:029) Credit ‡owt 1 0:002** (0:000) Credit impulset 1 0:000* (0:000) Financial Supply Shock recession 0:226 (0:541) Developing dummy 0:010 (0:024) Credit boomrecession 0:081** (0:036) Monetary expansion 0:078** (0:031) Note: Marginal e¤ects on Random e¤ects probit estimates, values in parenthesis are standard errors. *** p<0.01, ** p<0.05, * p<0.1. Table 12: Type I and II errors for di¤erent values of =0.3 =0.4 =0.5 Demand Type I error (%) 11:4 8 :4 5:3 Type II error (%) 12:5 16:6 27:0 Success rate (%) 88:2 89:3 88:8 Supply Type I error (%) 11:4 7 :6 4:5 Type II error (%) 25 29:1 33:3 Success rate (%) 84:9 86:5 87:1