The Riskiness of Credit Allocation and Financial Stability

Luis Brandao Marques; Qianying Chen; Claudio Raddatz; Jerome Vandenbussche; Peichu Xie

doi:10.5089/9781513513775.001.A001

Jump to Content Jump to Main Navigation

The Riskiness of Credit Allocation and Financial Stability

Author: Mr. Luis Brandao Marques¹, Qianying Chen², Claudio Raddatz³, Mr. Jerome Vandenbussche, and Peichu Xie⁴

¹ 404811396https://isni.org/isni/0000000404811396International Monetary Fund
| ² 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund
| ³ 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund
| ⁴ 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: LMarques@imf.org; QChen@imf.org; CRaddatz@imf.org; Jvandenbussche@imf.org; PXie@imf.org

Publication Date:: 27 Sep 2019

eISBN:: 9781513513775

Language:: English

Keywords:: WP; credit allocation; real GDP; credit quantity; credit quality indicator

Download PDF (2 MB)

We explore empirically how the time-varying allocation of credit across firms with heterogeneous credit quality matters for financial stability outcomes. Using firm-level data for 55 countries over 1991-2016, we show that the riskiness of credit allocation, captured by Greenwood and Hanson (2013)’s ISS indicator, helps predict downside risks to GDP growth and systemic banking crises, two to three years ahead. Our analysis indicates that the riskiness of credit allocation is both a measure of corporate vulnerability and of investor sentiment. Economic forecasters wrongly predict a positive association between the riskiness of credit allocation and future growth, suggesting a flawed expectations process.

Abstract

I. Introduction

The recent global financial crisis has renewed interest in understanding the role played by credit to the private sector as a source of financial instability. A sizable literature has focused on changes in the aggregate volume of credit and pointed to the dangers of fast credit expansions because they lead to increased leverage in the banking sector and the private nonfinancial sector, making the economy more vulnerable to negative shocks (Schularick and Taylor, 2012).² A more recent literature has documented the additional role of credit spreads and shown that low spreads precede episodes of financial instability as they set the stage for large spread reversals and large associated credit losses in financial institutions (Lopez-Salido et al. 2017, Krishnamurthy and Muir, forthcoming).

In this paper, we show that a third dimension of credit to the private sector—the extent to which the distribution of credit is tilted towards riskier borrowers—contains relevant information about future financial stability. To capture this new dimension, we build on work by Greenwood and Hanson (2013) –henceforth GH– who propose an indicator, labeled ISS, measuring the dispersion of firm-level credit quality across buckets of firms sorted by net debt issuance. Since this indicator is related both to borrower default risk and to the cross-sectional distribution of credit flows, we refer to it as the riskiness of credit allocation.

It may seem intuitive that a measure that captures the extent to which credit flows correlate with firm credit quality should provide information on future financial stability outcomes. However, this proposition has remained at best a conjecture in the financial stability literature.³ A small number of recent papers has analyzed the relevance of the distribution of credit across heterogeneous borrowers for excess bond returns (the focus of the GH paper), or GDP growth (Lopez-Salido et al., 2017; Gomes et al., 2018; Kirti 2018), but not on downside risks to GDP growth and the probability of financial crises, which is our area of attention in this paper.

Our contribution is twofold. First, we contribute to the literature on the dynamics of the composition of corporate credit flows by showing that the riskiness of credit allocation has followed a procyclical pattern at the global level over our 25-year-long sample period. We further provide cross-country evidence that ISS is positively associated with contemporaneous GDP growth and change in the credit-to-GDP ratio. This latter association is stronger when domestic financial conditions are looser, when bank lending standards are easier, and when credit spreads are lower, pointing to shifts in credit supply as an important driver of the riskiness of credit allocation. ⁴

Our second and main contribution is to the financial stability literature. We show that while the riskiness of credit allocation is not significantly associated with future GDP growth, it helps better predict downside risks to GDP growth and banking crises at horizons of up to three years.⁵ This predictive power—which we document based on a large sample of advanced and emerging economies over 1991–2016—is additional to that of changes in credit-to-GDP and of price of risk proxied by a financial conditions index.

We explore two plausible mechanisms for this key result linking the riskiness of credit allocation and financial stability outcomes. We find that variations in ISS, conditional on size of credit expansion and level of financial conditions, captures variations in the size of the weak tail of firms. Recent research has shown that firms’ ability to access credit and make investments when financial conditions tighten is related to their degree of financial vulnerability (Duval et al., forthcoming). Therefore a higher level of ISS means greater amplification of negative financial shocks. Second, we find that ISS helps predict reversals of financial conditions and corporate spreads. These two results indicate that ISS combines features of a corporate financial vulnerability measure and of a risk sentiment measure.

Several financial-frictions-based theories are relevant in explaining the relationship between ISS and the size of credit expansions. First, with costly state verification, the availability of credit to high-risk firms follows from their net worth (including collateral values) and is procyclical, generating a financial accelerator effect (Bernanke and Gertler 1989). Second, banks’ capacity and incentives to screen borrowers can deteriorate in periods of significant credit expansions, reinforcing the procyclical nature of lending to relatively riskier firms (Berger and Udell 2004; Dell’Ariccia and Marquez 2006). These theories, however, cannot provide a full account of our evidence, as they do not explain why the riskiness of credit allocation helps predict financial stability episodes controlling for the size of the credit expansion, financial conditions, and macroeconomic conditions.

Belief-based theories provide another class of possible explanations. In the narratives of Minsky (1977), Kindleberger (1978) and Bordalo et al. (2018), variations over time in investor beliefs and risk appetite can cause more credit being supplied to riskier firms during periods of optimism and/or neglect of risk. During those times, the composition of the flow of credit does not follow mechanically from aggregate credit volumes and current economic conditions. A further exploration of our data provides additional support for a role played by belief-based explanations. Extending the result obtained by Mian et al. (2017) for aggregate credit volumes, we show that ISS is positively related to the IMF’s GDP growth forecast and GDP growth forecast error, indicating a role for flawed expectations.

We submit our key findings to a comprehensive set of robustness checks. Importantly, we show that our results are robust with respect to the way we specify the ISS variable, because different indicators to measure firm-level credit quality may be suitable to different market and data environments. As GH, the ISS indicator we use in our regressions is based on the expected default frequency (EDF). Because its value as a credit risk indicator may be deemed problematic for countries with illiquid equity markets, we also calculate ISS using three common accounting ratios—debt-to-assets (leverage) ratio, interest coverage ratio (ICR), and debt-to-EBITDA ratio. We provide a full set of empirical results for the leverage ratio in the core of the paper, and use the other two ratios to check the robustness of our benchmark results. Overall, we find that our results are similar regardless of the underlying firm vulnerability indicator used to construct ISS. Our key findings are also robust to perturbing various other elements of the definition of ISS, and to using alternative aggregate credit series as well as a large set of controls following the literature on financial crises.

The remainder of the paper is organized as follows. Section II discusses data sources and describes the construction of the two core (EDF-based and leverage-based) ISS indicators. Section III describes their evolution at the global level and in selected countries. Section IV discusses their cyclical properties, and their relationship to indicators of domestic financial conditions. The paper then turns to the empirical analysis of the relationship between the riskiness of credit allocation and downside risks to GDP growth (Section V), and to the occurrence of banking crises (Section VI). Section VII presents robustness checks. Section VIII discusses possible mechanisms linking variations in the riskiness of credit allocation and future financial instability. Section IX concludes and is followed by a data appendix (Appendix A). A second appendix (Appendix B) is available online. It provides additional results and further robustness checks.

II. Construction of the Riskiness of Credit Allocation and Data Sources

A. Construction of the Riskiness of Credit Allocation indicator

We build on GH to construct the ISS indicator for a set of 55 countries (26 advanced economies and 29 emerging markets) at the annual frequency over the 1991–2016 period using firm-level data. These data are sourced from the Worldscope database, which provides a rich set of financial statement variables for listed firms. Appendix A provides details on the country sample and Online Appendix B provides explanations on the data cleaning process. We use only country-years for which observations for at least 40 firms are available to reduce potential volatility associated with a small number of firms as well as firms’ entry and exit.

In their analysis, GH use the expected default frequency (EDF) as the preferred firm-level measure of credit quality and demonstrate the robustness of some of their key results to the use of leverage instead. We treat EDF and leverage more symmetrically because the low liquidity of some stock markets outside the largest advanced economies makes the EDF – a market-based measure – less obviously superior as a measure of credit risk in a broad crosscountry sample.⁶ Therefore, we construct two main ISS measures: ISS^EDF and ISS^Leverage and two additional measures based on the interest coverage ratio (ICR), and the debt-to-EBITDA ratio for our robustness analysis.⁷

For each firm-level indicator, ISS is built as follows: First, in each year, each firm is assigned the value (from 1 to 10) of its decile in the distribution of the indicator in the country where it is located. A higher decile represents a larger value of the underlying indicator. Second, firms are similarly sorted by the change in net debt to lagged total assets into five equally-sized buckets. Firms in the bucket with the largest increases in debt (relative to their lagged assets) are called “top issuers,” and firms in the bucket with the largest decreases in debt are called the “bottom issuers.” A raw ISS measure is computed as the difference between the average vulnerability decile for the top issuers and the corresponding average for the bottom issuers:

\begin{matrix} I S S_{c, t}^{r a w, X} = \frac{Σ_{i \in T o p_I s s u e r_Q u int i l e} D e c i l e_{i, c, t}^{X}}{N_{c, t}^{T o p_{-} I s s u e r_{-} Q u i n t i l e}} - \frac{Σ_{i \in B o t t o m_{I s s u e r_{Q u i n t i l e}}} D e c i l e_{i, c, t}^{X}}{N_{c, t}^{B o t t o m_{I s s u e r_{Q u i n t i l e}}}}, & (1) \end{matrix}

where X ∈ {EDF,leverage}, Decile^X is the decile in the distribution of the vulnerability indicator X, N is the number of firms, i is the firm, c is the country, and t is the year. The use of deciles abstracts from changes in the mean and shape of the distribution of the credit quality indicator, focusing only on the ranking of a firm in the distribution of that indicator.⁸ Because the focus of the paper is on the dynamics of the riskiness of credit allocation within countries and not on its cross-country variation, we normalize this raw measure by subtracting its country-specific mean.⁹ This removes any influence of the country-specific sectoral composition of firms and ensures greater cross-country and cross-measure comparability. Since both a higher EDF and higher leverage are indicators of lower credit quality, an increase in ISS signals higher vulnerability. Figure 1 shows the distribution of the two indicators, which have the shape of a bell curve and have a standard deviation of about one.

The share of high yield bond issuance (HYS) is an alternative measure of debt issuer quality (Lopez-Salido et al. 2017, Kirti 2018). However, in addition to the reasons provided by GH, we have one important reason to prefer ISS to HYS: ¹¹ bond market development was limited in most advanced economies outside the U.S. until the late-1990’s and remains limited in most emerging markets and small advanced economies today. This, in light of the crosscountry context of our study, makes HYS a very noisy and unduly volatile indicator.¹²

B. Other data

Macroeconomic data series, including credit series, are sourced from the International Monetary Fund (IMF)’s International Financial Statistics (IFS) and World Economic Outlook databases, the Bank for International Settlements (BIS), as well as Haver Analytics. Our baseline credit series is that from IMF’s International Financial Statistics as it provides the greatest coverage. Financial variables are sourced from Bloomberg and Thomson Reuters. Lending standards are obtained from Haver Analytics. Financial conditions indices are constructed for 43 countries over 1990–2016 as described in Online Appendix B. Data on financial crises are obtained from Laeven and Valencia (2018). The change of the credit-to-GDP ratio is winsorized at the 1 percent level to reduce the influence of outliers. Country coverage is summarized in Appendix Table A1. Further details on data sources and definitions are provided in Appendix Table A2.

III. The Riskiness of Credit Allocation and Its Evolution Across Countries

The evolution of the riskiness of credit allocation across countries suggests clear global patterns, as shown on Figure 2 which plots the two-year moving average of the two core ISS indicators for the median country. Its dynamic at the global level is broadly the same for ISS^EDF and ISS^Leverage. Starting from elevated levels in the late 1990s, it fell in 2000–04 in the aftermath of the Asian and Russian crises and of the burst of the dot-com equity bubble, reached its historical low in 2002 for ISS^EDF and in 2004 for ISS^Leverage, rose steeply afterwards and hit a peak at the onset of the global financial crisis. It then declined sharply over the next two years and was slightly below its pre-crisis level at the end of 2016, the latest data point in our analysis.

This global dynamic is reflected at the country level, with some country-specific nuances. Figure 3 shows the evolution of ISS^EDF and ISS^Leverage in six major economies during 1995– 2016. The two measures display similar patterns in the six countries.

The dynamics in the United States (Figure 4, panel 1) and Japan (Figure 4, panel 2) are very similar in both cyclicality and magnitudes.¹³ The most recent period (2014–16), however, suggests a divergence: the riskiness of credit allocation decreased in the United States to a relatively low level while in Japan it remained at a level that is relatively high in historical perspective.¹⁴
Figure 4, panels 3 and 4 show contrasting developments in two of the largest euro area countries. Spain (Figure 4, panel 3) had a credit boom from the late 1990s to the mid-2000s, which was followed by a deep recession during the global financial crisis and the euro area sovereign debt crisis. Measures of the riskiness of credit allocation for this country reflect these developments quite well: a steep rise in riskiness took place in the mid-to-late 1990s, leading to very high levels of riskiness until the crisis of 2008, which triggered a sudden and large fall of the indicator. This pattern is consistent with findings of Banco de España (2017). By contrast, variations in the riskiness of credit allocation in Germany (Figure 4, panel 4), a country that did not have a credit boom during the 20-year period, have remained within the same narrower range as the United States and Japan, and the measure has moved into positive territory in recent years, suggesting a higher level of risk-taking.
The evolution of the riskiness of credit allocation in India (Figure 4, panel 5) has broadly followed global patterns, and the measure was at a relatively low level in 2016. The synchronization of China (Figure 4, panel 6) with global developments is weaker—peaks and troughs appear to occur with a two-to-three-year lag. The finding of a peak in 2009– 10 is consistent with recent evidence that the implementation of a large stimulus plan beginning at the end of 2008 led to a misallocation of credit (Cong and others 2017).

IV. The Cyclicality of the Riskiness of Credit Allocation

These patterns raise several questions regarding the cyclicality of the riskiness of credit allocation. Does it systematically comove with GDP growth and credit growth? If so, does the association with credit growth depend on measures of financial conditions that signal expansions in credit supply, such as credit spreads or a broad price-based financial conditions index? We shed light on these questions using standard cross-country panel regressions.

To analyze the dynamics of the composition of corporate credit flows, we estimate the following equation:

\begin{matrix} I S S_{i, t}^{V} = α_{i}^{V} + γ_{t}^{V} + β_{1}^{V} Δ G D P_{i, t} + β_{2}^{V} Δ C r e d i t_{i, t} + β_{3}^{V} C o n t r o l_{i, t} + ε_{i, t}^{V}, & (2) \end{matrix}

in which V ∈ {EDF’, leverage} represents a firm-level vulnerability indicator, and correspondingly $I S S_{i, t}^{V}$ represents the riskiness of credit allocation based on indicator V for country i at time t. ΔGDP is real GDP growth, and ΔCredit is the change in the ratio of bank credit to the nonfinancial private sector to nominal GDP. Control is the domestic currency appreciation against the U.S. dollar, and helps control for a potential mechanical valuation effect on ISS from debt denominated in foreign currency.¹⁵ Both country $(α_{i}^{V})$ and year $(γ_{t}^{V})$ fixed effects are included. The standard errors are clustered at the country level.

Results are provided in Table 1. Whether EDF-based (column (1)) or leverage-based (column (2)), the riskiness of credit allocation increases when GDP growth or changes in the domestic credit-to-GDP ratio are stronger. These findings are consistent with standard financial accelerator mechanisms, and with mechanisms in which credit supply shocks affect macrofinancial outcomes through a risk-taking channel. The association of credit expansion with greater riskiness of credit allocation is statistically significant for both measures. A one standard deviation increase in the change of the credit-to-GDP ratio (equivalent to an increase of 5.5 percentage points) is associated with an increase in the riskiness of credit allocation of 0.12–0.25 standard deviation, depending on the ISS measure.¹⁶ If the specification is enriched by adding a credit boom dummy (constructed as in Dell’Ariccia et al. 2016), a variable capturing the length of a credit boom, or dummies to capture different phases of a credit boom, none of these variables is significant. This points to the absence of nonlinearities in the relationship between the size of a credit expansion and the riskiness of credit allocation, but also that the relationship is not simply driven by extreme episodes of large credit expansions.

Table 1.

Cyclicality of the Riskiness of Credit Allocation

Source: Authors’ estimatesNotes: All regressions are OLS and include country and time fixed effects. Standard errors are clustered at the country level and shown in parentheses. *** p<0.01; ** p<0.05; *p<0.1.

Table 1.

Cyclicality of the Riskiness of Credit Allocation

	(1)	(2)
	Dependent Variable:
	ISS^EDF	ISS ^Leverage
Real GDP Growth	0.08*** (0.02)	0.08*** (0.02)
Δ(Credit-to-GDP Ratio)	0.02*** (0.01)	0.05*** (0.01)
Appreciation against the US dollar	-0.01 (0.01)	-0.04*** (0.01)
Number of Observations	936	986
Number of Countries	53	55
R²	0.15	0.31

Table 1.

Cyclicality of the Riskiness of Credit Allocation

	(1)	(2)
	Dependent Variable:
	ISS^EDF	ISS ^Leverage
Real GDP Growth	0.08*** (0.02)	0.08*** (0.02)
Δ(Credit-to-GDP Ratio)	0.02*** (0.01)	0.05*** (0.01)
Appreciation against the US dollar	-0.01 (0.01)	-0.04*** (0.01)
Number of Observations	936	986
Number of Countries	53	55
R²	0.15	0.31

While the relationships documented above only establish the cyclical patterns of ISS and do not speak to causality, we can use shed some additional light on the mechanisms behind changes in the riskiness of credit allocation by noting that supply-driven credit expansions are likely to be accompanied by looser financial conditions or looser lending standards. To analyze the relationship between size of credit expansion, financial conditions, and riskiness of credit allocation, and shed further light on the role of credit supply shifts in the dynamics of ISS, we therefore enrich equation (2) as follows:

\begin{matrix} I S S_{i, t}^{V} = α_{i}^{V} + γ_{t}^{V} + β^{V} C o n t r o l s_{i, t} + δ^{V} F_{i, t} + θ^{V} \times F_{i, t} \times Δ C r e d i t_{i, t} + ε_{i, t}^{V}, & (3) \end{matrix}

in which Controls_i,t is a vector of control variables including real GDP growth, change in the credit-to-GDP ratio, and domestic currency appreciation as discussed above. The term F _i,t represents either a financial conditions index (FCI), a survey-based measure of bank lending standards, or a corporate credit spread (capturing credit market conditions). Both the change in the credit-to-GDP ratio and the FCI are demeaned at the country level, while lending standards and corporate spreads are transformed into a z-score to ensure greater cross-country comparability. The estimated coefficient ${\hat{θ}}^{V}$ captures the marginal effect on the credit cyclicality of the riskiness of credit allocation of a change in the financial conditions variable. The standard errors are clustered at the country level, as before.

Table 2 reports the results. The association between larger credit expansions and riskier allocations is stronger when the price of risk is low (columns 1 and 4), when lending standards are loose (columns 2 and 5), or when corporate credit spreads are low (columns 3 and 6), indicating that outward credit supply shifts are associated with riskier allocations. To capture a possible effect of search for yield motives (Rajan, 2005), we also examined a possible role for the long-term rate, either transformed into a z-score or a dummy indicating a value in the lowest quartile of the country-specific distribution. While the sign of the coefficients indicates an association between lower long-term rates and higher riskiness of credit allocation that is consistent with a search for yield motive, their statistical significance is weak.

Table 2.

Credit Expansion, Financial Conditions, and Riskiness of Credit Allocation

Source: Authors’ estimates.Note: All regressions are OLS, include country and time fixed effects, and control for real GDP growth and domestic currency appreciation against the US dollar. An increase in the “bank lending standards” variable means stricter bank lending standards. An increase in the financial conditions index means tighter financial conditions. Standard errors are clustered at the country level and shown in parentheses. *** p<0.01; ** p<0.05; *p<0.1.

Table 2.

Credit Expansion, Financial Conditions, and Riskiness of Credit Allocation

	(1)	(2)	(3)	(4)	(5)	(6)
	Dependent Variable: Riskiness of Credit Allocation Based on EDF			Dependent Variable: Riskiness of Credit Allocation Based on
Δ(Credit-to-GDP Ratio)	0.04*** (0.01)	0.03*** (0.01)	0.03*** (0.01)	0.05*** (0.02)	0.05*** (0.01)	0.05*** (0.01)
Financial Conditions Index (FCI)	-0.11 (0.12)			-0.12 -0.08
A(Credit-to-GDP Ratio) x FCI	-0.01** (0.01)			-0.01** (0.00)
Bank Lending Standards		-0.11 (0.07)			-0.10 (0.07)
Δ(Credit-to-GDP Ratio) x Bank Lending Standards		-0.04*** (0.01)			-0.03* (0.02)
Corporate Credit Spread			-0.05 (0.06)			-0.07 (0.06)
Δ(Credit-to-GDP Ratio) x Corporate Credit Spread			-0.02** (0.01)			-0.02** (0.01)
Number of Observations	812	257	655	849	266	663
Number of Countries	41	21	37	42	21	37
R²	0.16	0.31	0.23	0.34	0.39	0.33

Table 2.

Credit Expansion, Financial Conditions, and Riskiness of Credit Allocation

	(1)	(2)	(3)	(4)	(5)	(6)
	Dependent Variable: Riskiness of Credit Allocation Based on EDF			Dependent Variable: Riskiness of Credit Allocation Based on
Δ(Credit-to-GDP Ratio)	0.04*** (0.01)	0.03*** (0.01)	0.03*** (0.01)	0.05*** (0.02)	0.05*** (0.01)	0.05*** (0.01)
Financial Conditions Index (FCI)	-0.11 (0.12)			-0.12 -0.08
A(Credit-to-GDP Ratio) x FCI	-0.01** (0.01)			-0.01** (0.00)
Bank Lending Standards		-0.11 (0.07)			-0.10 (0.07)
Δ(Credit-to-GDP Ratio) x Bank Lending Standards		-0.04*** (0.01)			-0.03* (0.02)
Corporate Credit Spread			-0.05 (0.06)			-0.07 (0.06)
Δ(Credit-to-GDP Ratio) x Corporate Credit Spread			-0.02** (0.01)			-0.02** (0.01)
Number of Observations	812	257	655	849	266	663
Number of Countries	41	21	37	42	21	37
R²	0.16	0.31	0.23	0.34	0.39	0.33

V. The Riskiness of Credit Allocation and Downside Risks to Growth

We move on to ask whether the riskiness of credit allocation helps predict future GDP growth. One might expect that a relatively larger expansion of credit to riskier firms would have benefits for future economic activity if these firms were previously credit-constrained and were now able to boost their investment to seize growth opportunities. To investigate this possibility, we estimate the following equation:

\begin{matrix} Δ y_{i, t, t + h} = α_{i}^{V} + γ_{t}^{V} + β Δ {(\frac{C r e d i t}{G D P})}_{i, t - 1}^{m v 3} + δ F C I_{i, t - 1}^{m v 3} + θ I S S_{i, t - 1}^{m v 3} + μ C o n t r o l s_{i, t - 1}^{m v 3} + u_{i, t}, & (4) \end{matrix}

in which Δy_i,t,t+h is cumulative real GDP growth rate from t to t+h, where h=1,2,3. The change of the credit to GDP ratio, FCI, and ISS are the same as in the previous section. The choice to include the FCI instead of the corporate spread as a measure of the price of risk is dictated by the smaller availability of the latter. Both country $(α_{i}^{V})$ and year $(γ_{t}^{V})$ fixed effects are included. Controls include real GDP growth. All explanatory variables enter the equation as the first lag of their simple three-year moving average.

Results, provided in Table 3, indicate that although credit expansions and tight financial conditions tend to forecast future GDP declines, as already documented in the literature, the relationship between ISS and future GDP growth is never positive and rarely significant. This suggests that a riskier credit allocation does not provide any extra kick to future GDP growth.¹⁷

Table 3.

Riskiness of Credit Allocation and Cumulative GDP Growth

Source: Authors’ estimates.Note: All regressions are OLS, include country and time fixed effects. Explanatory variables enter the regression as the lag of their simple three-year moving average. The change in the credit-to-GDP ratio is winsorized at 1 percent. Standard errors are clustered at the country level and shown in parentheses. *** p<0.01; ** p<0.05; *p<0.1.

Table 3.

Riskiness of Credit Allocation and Cumulative GDP Growth

	(1)	(2)	(3)	(4)	(5)	(6)
	1 year	2-years	3-years	1 year	2-years	3-years
Real GDP Growth	-0.00413 (0.0635)	-0.0947 (0.101)	-0.201 (0.130)	0.0347 (0.0575)	-0.0657 (0.0912)	-0.339*** (0.121)
Δ(Credit-to-GDP Ratio)	-0.102*** (0.0261)	-0.198*** (0.0416)	-0.292*** (0.0536)	-0.106*** (0.0250)	-0.238*** (0.0396)	-0.411*** (0.0524)
Financial Conditions Index	-0.524** (0.238)	-0.810** (0.379)	-0.889* (0.488)	-0.353 (0.232)	-0.641* (0.368)	-0.978** (0.487)
ISS^EDF	-0.0763 (0.133)	-0.195 (0.213)	-0.250 (0.274)
ISS^Leverage				-0.260* (0.141)	-0.319 (0.224)	-0.231 (0.296)
Number of Observations	586	586	586	658	658	658
Number of Countries	40	40	40	42	42	42

Table 3.

Riskiness of Credit Allocation and Cumulative GDP Growth

	(1)	(2)	(3)	(4)	(5)	(6)
	1 year	2-years	3-years	1 year	2-years	3-years
Real GDP Growth	-0.00413 (0.0635)	-0.0947 (0.101)	-0.201 (0.130)	0.0347 (0.0575)	-0.0657 (0.0912)	-0.339*** (0.121)
Δ(Credit-to-GDP Ratio)	-0.102*** (0.0261)	-0.198*** (0.0416)	-0.292*** (0.0536)	-0.106*** (0.0250)	-0.238*** (0.0396)	-0.411*** (0.0524)
Financial Conditions Index	-0.524** (0.238)	-0.810** (0.379)	-0.889* (0.488)	-0.353 (0.232)	-0.641* (0.368)	-0.978** (0.487)
ISS^EDF	-0.0763 (0.133)	-0.195 (0.213)	-0.250 (0.274)
ISS^Leverage				-0.260* (0.141)	-0.319 (0.224)	-0.231 (0.296)
Number of Observations	586	586	586	658	658	658
Number of Countries	40	40	40	42	42	42

This insignificant relationship, however, may mask heterogeneity across different parts of the future GDP growth distribution. In the spirit of Adrian et al. (2018), who show that the distribution of GDP growth evolves over time as a function of economic and financial conditions, we re-examine these results using quantile regressions. To do so, we replace the left-hand-side of Equation (44) by $Δ y_{i, t, t + h}^{d}$ ,where d represents a decile, and use Powell (2016)’s fixed effect quantile panel estimator for each decile of the 3-year cumulative GDP growth distribution.¹⁸

Columns (1) through (9) of Table 4 show the results for individual deciles. They reveal that a greater riskiness of credit allocation shifts the whole left tail and the median – in other words, the bottom five deciles- of the growth distribution to the left, and that it moves the top deciles to the right, although generally not significantly so. In other words, the riskiness of credit allocation has a significant impact on downside risks to growth.

Table 4.

Riskiness of Credit Allocation and Risks to GDP Growth (all Deciles)

Source: Authors’ estimates.Note: The estimates shown in columns (1)-(9) are obtained through quantile regressions with nonadditive fixed effects (Powell 2016). The dependent variables are all deciles of the 3-year cumulative GDP growth. Standard errors are in parentheses. Explanatory variables enter the regression as the lag of their simple three-year moving average and are demeaned at the country level; the change in the credit-to-GDP ratio is winsorized at 1 percent. Real GDP growth is controlled for. *** p<0.01, ** p<0.05, * p<0.1.

Table 4.

Riskiness of Credit Allocation and Risks to GDP Growth (all Deciles)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Decile 1	Decile 2	Decile 3	Decile 4	Decile 5	Decile 6	Decile 7	Decile 8	Decile 9
Panel A: ISS based on EDF
Δ(Credit-to-GDP Ratio)	-0.137** (0.0626)	-0.257*** (0.0293)	-0.261*** (0.0342)	-0.265*** (0.0224)	-0.237*** (0.0191)	-0.274*** (0.0229)	-0.317*** (0.0348)	-0.321*** (0.0225)	-0.339*** (0.0731)
Financial Conditions Index	1.172 (0.940)	-0.641 (0.587)	0.0771 (0.374)	-0.371 (0.365)	-0.351 (0.365)	-0.390 (0.434)	-1.150** (0.490)	-1.120** (0.496)	-1.661** (0.759)
ISS	-0.993*** (0.254)	-1.321*** (0.164)	-0.853*** (0.145)	-0.639*** (0.109)	-0.565*** (0.134)	-0.146 (0.142)	0.129 (0.267)	0.306 (0.310)	0.519 (0.502)
Number of Observations	586	586	586	586	586	586	586	586	586
Number of Countries	40	40	40	40	40	40	40	40	40
Panel B: ISS based on leverage
Δ(Credit-to-GDP Ratio)	-0.259*** (0.0462)	-0.264*** (0.0459)	-0.270*** (0.0234)	-0.267*** (0.0164)	-0.244*** (0.0309)	-0.307*** (0.0266)	-0.395*** (0.0304)	-0.462*** (0.0442)	-0.600*** (0.0455)
Financial Conditions Index	2.222*** (0.704)	0.209 (0.758)	-0.0118 (0.452)	-0.487** (0.234)	-0.422 (0.299)	-0.828* (0.467)	-1.506*** (0.426)	-1.906*** (0.616)	-2.519*** (0.801)
ISS^Leverage	-0.650** (0.323)	-0.614*** (0.200)	-0.743*** (0.230)	-0.689*** (0.0748)	-0.559*** (0.170)	-0.163 (0.152)	0.169 (0.173)	0.531 (0.356)	0.832*** (0.294)
Number of Observations	658	658	658	658	658	658	658	658	658
Number of Countries	42	42	42	42	42	42	42	42	42

Table 4.

Riskiness of Credit Allocation and Risks to GDP Growth (all Deciles)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	Decile 1	Decile 2	Decile 3	Decile 4	Decile 5	Decile 6	Decile 7	Decile 8	Decile 9
Panel A: ISS based on EDF
Δ(Credit-to-GDP Ratio)	-0.137** (0.0626)	-0.257*** (0.0293)	-0.261*** (0.0342)	-0.265*** (0.0224)	-0.237*** (0.0191)	-0.274*** (0.0229)	-0.317*** (0.0348)	-0.321*** (0.0225)	-0.339*** (0.0731)
Financial Conditions Index	1.172 (0.940)	-0.641 (0.587)	0.0771 (0.374)	-0.371 (0.365)	-0.351 (0.365)	-0.390 (0.434)	-1.150** (0.490)	-1.120** (0.496)	-1.661** (0.759)
ISS	-0.993*** (0.254)	-1.321*** (0.164)	-0.853*** (0.145)	-0.639*** (0.109)	-0.565*** (0.134)	-0.146 (0.142)	0.129 (0.267)	0.306 (0.310)	0.519 (0.502)
Number of Observations	586	586	586	586	586	586	586	586	586
Number of Countries	40	40	40	40	40	40	40	40	40
Panel B: ISS based on leverage
Δ(Credit-to-GDP Ratio)	-0.259*** (0.0462)	-0.264*** (0.0459)	-0.270*** (0.0234)	-0.267*** (0.0164)	-0.244*** (0.0309)	-0.307*** (0.0266)	-0.395*** (0.0304)	-0.462*** (0.0442)	-0.600*** (0.0455)
Financial Conditions Index	2.222*** (0.704)	0.209 (0.758)	-0.0118 (0.452)	-0.487** (0.234)	-0.422 (0.299)	-0.828* (0.467)	-1.506*** (0.426)	-1.906*** (0.616)	-2.519*** (0.801)
ISS^Leverage	-0.650** (0.323)	-0.614*** (0.200)	-0.743*** (0.230)	-0.689*** (0.0748)	-0.559*** (0.170)	-0.163 (0.152)	0.169 (0.173)	0.531 (0.356)	0.832*** (0.294)
Number of Observations	658	658	658	658	658	658	658	658	658
Number of Countries	42	42	42	42	42	42	42	42	42

Our findings indicate that a riskier credit allocation does not result in a trade-off between greater mean growth and greater downside risks to growth. Instead, it increases downside risks without a clear impact on average growth nor in the upper tail of the future GDP growth distribution. This has some similarity with the findings of Baron and Xiong (2017), which document that strong banking sector credit growth is associated with both a greater likelihood of bank equity price crash and lower mean equity returns, indicating a neglect of crash risk.

In Table 5, we zoom in on the bottom two deciles of the 1-year, 2-year, and 3-year ahead GDP growth distributions. Variations of the riskiness of credit allocation appear strongly related to movements of the left tail of the growth distribution over all horizons. Panel A in the top half of the table shows the results obtained with ISS^EDF, while Panel B at the bottom shows the results obtained with ISS^Leverage The change in credit-to-GDP is always has a negative and significant effect. The coefficient for the FCI is positive and is significant mostly in the first year. The EDF-based riskiness of credit allocation is negative and significant for both deciles over two- and three-year horizons, but only for the second decile during the first year, while the leverage-based riskiness of credit allocation is always negative and significant. In quantitative terms, an increase in the riskiness of credit allocation by one standard deviation shifts the left tail of the 3-year cumulative growth distribution to the left by 1–1.3 percentage point for the EDF-based measure and by 0.6–0.7 percentage point for the leverage-based measure.

Table 5.

Riskiness of Credit Allocation and Downside Risks to Growth

(1^st and 2^nd Decile of the Cumulative GDP Growth Distribution)

Source: Authors’ estimates.Note: The estimates are obtained through quantile regressions with nonadditive fixed effects (Powell 2016). Standard errors are in parentheses. Explanatory variables enter the regression as the lag of their simple three-year moving average and are demeaned at the country level; the change in the credit-to-GDP ratio is winsorized at 1 percent. Real GDP growth is controlled for. *** p<0.01, ** p<0.05, * p<0.1.

Table 5.

Riskiness of Credit Allocation and Downside Risks to Growth

(1^st and 2^nd Decile of the Cumulative GDP Growth Distribution)

	(1)	(2)	(3)	(4)	(5)	(6)
	1 year		2 years		3 years
	Decile 1	Decile 2	Decile 1	Decile 2	Decile 1	Decile 2
Panel A: ISS based on EDF
A(Credit-to-GDP Ratio)	-0.0158*** (0.00454)	-0.0718*** (0.0188)	-0.129*** (0.0320)	-0.159*** (0.0205)	-0.137** (0.0626)	-0.257*** (0.0293)
Financial Conditions Index	1.517*** (0.0309)	0.0236 (0.399)	0.472 (0.915)	0.265 (0.485)	1.172 (0.940)	-0.641 (0.587)
ISS	-0.0342 (0.0487)	-0.181** (0.0717)	-0.656*** (0.207)	-0.708*** (0.135)	-0.993*** (0.254)	-1.321*** (0.164)
Number of Observations	549	549	549	549	549	549
Number of Countries	40	40	40	40	40	40
Panel B: ISS based on leverage
A(Credit-to-GDP Ratio)	-0.0441*** (0.0137)	-0.0586*** (0.0156)	-0.188*** (0.0389)	-0.164*** (0.0249)	-0.259*** (0.0462)	-0.264*** (0.0459)
Financial Conditions Index	0.781** (0.349)	0.433*** (0.143)	1.830 (1.187)	0.241 (0.525)	2.222*** (0.704)	0.209 (0.758)
ISSLeverage	-0.239** (0.114)	-0.220*** (0.0537)	-0.625** (0.306)	-0.621*** (0.157)	-0.650** (0.323)	-0.614*** (0.200)
Number of Observations	658	658	658	658	658	658
Number of Countries	42	42	42	42	42	42

Table 5.

Riskiness of Credit Allocation and Downside Risks to Growth

(1^st and 2^nd Decile of the Cumulative GDP Growth Distribution)

	(1)	(2)	(3)	(4)	(5)	(6)
	1 year		2 years		3 years
	Decile 1	Decile 2	Decile 1	Decile 2	Decile 1	Decile 2
Panel A: ISS based on EDF
A(Credit-to-GDP Ratio)	-0.0158*** (0.00454)	-0.0718*** (0.0188)	-0.129*** (0.0320)	-0.159*** (0.0205)	-0.137** (0.0626)	-0.257*** (0.0293)
Financial Conditions Index	1.517*** (0.0309)	0.0236 (0.399)	0.472 (0.915)	0.265 (0.485)	1.172 (0.940)	-0.641 (0.587)
ISS	-0.0342 (0.0487)	-0.181** (0.0717)	-0.656*** (0.207)	-0.708*** (0.135)	-0.993*** (0.254)	-1.321*** (0.164)
Number of Observations	549	549	549	549	549	549
Number of Countries	40	40	40	40	40	40
Panel B: ISS based on leverage
A(Credit-to-GDP Ratio)	-0.0441*** (0.0137)	-0.0586*** (0.0156)	-0.188*** (0.0389)	-0.164*** (0.0249)	-0.259*** (0.0462)	-0.264*** (0.0459)
Financial Conditions Index	0.781** (0.349)	0.433*** (0.143)	1.830 (1.187)	0.241 (0.525)	2.222*** (0.704)	0.209 (0.758)
ISSLeverage	-0.239** (0.114)	-0.220*** (0.0537)	-0.625** (0.306)	-0.621*** (0.157)	-0.650** (0.323)	-0.614*** (0.200)
Number of Observations	658	658	658	658	658	658
Number of Countries	42	42	42	42	42	42

VI. The Riskiness of Credit Allocation and the Occurrence of Banking Crises

Having shown that the riskiness of credit allocation has a strong effect on the left tail of the future growth distribution, we now revisit the more classic literature on the occurrence of banking crises by augmenting the literature’s typical specification with the riskiness of credit allocation as an additional explanatory variable. In other words, using cross-country logit regressions, we analyze whether ISS constitutes an early warning indicator of a systemic financial crisis.

Before turning to the formal econometric analysis, it is worthwhile looking at Panel 1 of Figure 4, which illustrates that the riskiness of credit allocation has a very clear inverted-U shape around systemic financial crisis episodes: it rises gradually during the five years preceding the crisis, reaches a relatively high level, and then falls following the onset of the crisis. In our data, credit expansions are also large before a crisis (Panel 2), which is consistent with findings by Schularick and Taylor (2012), and corporate spreads are low (Panel 3), which is consistent with findings by Krishnamurthy and Muir (forthcoming). By contrast, conventional corporate vulnerability indicators (Panel 4) pick up significantly only when the crisis has already struck.

We analyze the logarithm of the odds ratio of the start of a systemic banking crisis using the following conditional fixed-effects logistic regression model:

\begin{matrix} l o g \frac{P [C r i s i s s t a r t_{t} = 1 | X_{i, t - 1}]}{P [C r i s i s s t a r t_{t} = 0 | X_{i, t - 1}]} = α_{i} + β Δ C r e d i t_{i, t - 1}^{m v 3} + γ F C I_{i, t - 1}^{m v 3} + δ I S S_{i, t - 1}^{V, m v 3} + μ C o n t r o l s_{i, t - 1}^{m v 3} + u_{i, t} & (5) \end{matrix}

in which Crisisstart is a dummy variable equal to 1 at the start of a systemic banking crisis and equal to 0 otherwise, X refers to the vector of explanatory variables, α_i is a country fixed effect, ΔCredit is the change of the ratio of bank credit to the nonfinancial private sector to nominal GDP, FCHs the financial conditions index, ISS is the riskiness of credit allocation, and V ∈ {EDF, leverage}. Controls includes controls for the macroeconomic environment, namely the change in the current-account-balance-to-GDP ratio and real GDP growth, as in Jordà, Schularick, and Taylor (2016a).¹⁹. All explanatory variables enter the equation as the first lag of their simple three-year moving average (hence the mv3 superscript) and are demeaned at the country level. Standard errors are clustered at the country level.²⁰ Because country fixed effects are included, the regression sample shrinks to include only countries that have had at least one crisis.²¹

Regression results are reported in Table 6. Column (1) shows that changes in the credit-to-GDP ratio predict crises, in line with the bulk of the literature, most notably Schularick and Taylor (2012). Column (2) shows that the price of risk also predicts crises. However, when the change in credit-to-GDP and the price of risk enter the regression together in Column (3), the change in credit-to-GDP ratio ceases to be significant.

Table 6.

Crisis Prediction Model

Source: Authors’ estimates.Note: The estimates are obtained through a conditional fixed effects logit regression. Standard errors are shown in parentheses. Explanatory variables enter the regression as the lag of their simple three-year moving average and are demeaned at the country level. The change in credit-to-GDP ratio is winsorized at 1 percent. Controls include the change in current account-to-GDP ratio and the real GDP growth rate. *** p<0.01, ** p<0.05, * p<0.1.

Columns (4) to (6) show our key results when ISS is added to the specification. The riskiness of credit allocation is significant when it enters separately (column (4)), together with the change in credit-to-GDP (column (5)), and together with change in credit-to-GDP and price of risk combined (column (6)). Thus, for a given size of credit expansion and a given level of the price of risk, a greater level of the riskiness of credit allocation implies a higher probability of financial crisis. A one standard deviation rise in the ISS measure increases the odds of a crisis by a factor of about four. Comparing columns (3) and (6), one can observe that the Pseudo-R2 shows a sizable improvement when the riskiness of credit allocation is added to the regression.

Conditional fixed effects logit is known to give consistent estimates (Chamberlain, 1980) but does not provide estimates of the individual fixed effects, which are needed if one wants to compute statistics such as the area under the ROC curve (AUROC) which are often found in the literature to assess the performance of a logistic regression model. The unconditional-with-dummies estimator provides estimates of the individual fixed effects but leads to inconsistent estimates due to the incidental parameter problem, although Coupé (2005) finds that the bias is small when T is large.²² Online Appendix Table B2 shows that coefficients obtained from an unconditional-with-dummies estimator are close to those obtained using the conditional estimator, and that the addition of the riskiness of credit allocation variable to the regression specification boosts the AUROC statistics in this model.

VII. Robustness Analysis

In this section, we present evidence that our main findings are robust: they hold regardless of the choice of the specific credit series and of the choice of firm-level vulnerability indicator, and they are not sensitive to perturbations of the ISS definition.

A. Alternative aggregate credit series

Our baseline credit series is credit to the private nonfinancial sector provided by domestic banks, as in Schularick and Taylor (2012), Gourinchas and Obstfeld (2012), and Baron and Xiong (2017). It is sourced from IFS and has the best country-year coverage. Results for the downside risks to growth model (second decile, 3-yar horizon) using these series are repeated in Column (1) of Table 7a. A broader concept of credit, i.e. total credit provided to the private non-financial sector by all sectors (both domestic and foreign), both bank and nonbank) is used in Column (2). These data series are sourced from the BIS. Since the recent literature (Jordà et al 2016, Mian et al 2017, Aldasoro et al. 2018, Alter et al. 2018) has emphasized the role that credit to household plays in overall financial vulnerability,we also split the total credit series by source: results for credit by domestic banks are shown in Column (5), and results for credit by cross-border sources are shown in Column (6). Aldaroso et al.(2018) also find that cross-border claims on banks and nonbanks is a predictor of banking crises, so in Column (7) and Column (8), we use cross-border claims on banks, and cross-border claims on banks plus non-banks. In Column (9), we focus on the credit gap, constructed with the IFS credit data, as Borio et al. (2011) suggest that this indicator a high signal-to-noise ratio to forecast episodes of financial instability.²³

Table 7a.

Downside Risks to Growth – Alternative Credit Series

Table 7a.

Downside Risks to Growth – Alternative Credit Series

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Panel A: Riskiness of Credit Allocation Based on EDF
Real GDP Growth	-0.0546 (0.144)	0.0236 (0.0773)	-0.0543 (0.127)	-0.168* (0.0998)	-0.0274 (0.140)	0.0300 (0.0883)	0.0868 (0.115)	0.0757 (0.0842)	-0.161 (0.118)
Δ(Domestic Bank Private Nonfinancial Credit-to-GDP) (IFS)	-0.257*** (0.0293)
Δ(Total Private Nonfinancial Credit-to-GDP) (BIS)		-0.260*** (0.0221)
Delta Total Privated Nonfinancial Credit-to-GDP: to Corporate			-0.275*** (0.0472)
Delta Total Privated Nonfinancial Credit-to-GDP: to Household				-0.835*** (0.0768)
Delta Total Privated Nonfinancial Credit-to-GDP: from Domestic Bank					-0.239*** (0.0398)
Δ(Cross-Border Credit to Non-Banks to GDP) (BIS)						-0.197*** (0.0603)
Δ(Cross-Border Credit to Banks to GDP) (BIS)							-0.175*** (0.0380)
Δ(Cross-Border Credit to Banks and Non-Banks to GDP) (BIS)								-0.145*** (0.0343)
Credit-to-GDP Gap (based on IFS)									-0.0778*** (0.00983)
Financial Conditions Index	-0.641 (0.587)	0.0545 (0.479)	1.224* (0.735)	0.618 (0.773)	-0.247 (0.562)	0.713 (0.588)	0.167 (0.709)	0.536 (0.396)	0.120 (0.457)
ISS^EDF	-1.321*** (0.164)	-1.426*** (0.202)	-0.983*** (0.206)	-0.662*** (0.231)	-1.349*** (0.156)	-1.288*** (0.182)	-1.478*** (0.215)	-1.435*** (0.254)	-1.349*** (0.298)
Number of Observations	586	567	490	490	567	592	592	592	557
Number of Countries	40	36	36	36	36	40	40	40	36
Panel B: Riskiness of Credit Allocation Based on Leverage
Real GDP Growth	-0.0316 (0.118)	-0.0898 (0.0858)	-0.388** (0.161)	-0.389*** (0.150)	0.0561 (0.0916)	-0.165 (0.114)	-0.00524 (0.0808)	0.0561 (0.100)	-0.0871* (0.0469)
Δ(Domestic Bank Private Nonfinancial Credit-to-GDP) (IFS)	-0.264*** (0.0459)
Δ(Total Private Nonfinancial Credit-to-GDP) (BIS)		-0.223*** (0.0435)
Delta Total Privated Nonfinancial Credit-to-GDP: to Corporates			-0.170** (0.0691)
Delta Total Privated Nonfinancial Credit-to-GDP: to Households				-1.037*** (0.100)
Delta Total Privated Nonfinancial Credit-to-GDP: from Domestic Banks					-0.279*** (0.0736)
Δ(Cross-Border Credit to Non-Banks to GDP) (BIS)						-0.383*** (0.0605)
Δ(Cross-Border Credit to Banks to GDP) (BIS)							-0.193*** (0.0283)
Δ(Cross-Border Credit to Banks and Non-Banks to GDP) (BIS)								-0.149*** (0.0414)
Credit-to-GDP Gap (based on IFS)									-0.102*** (0.00602)
Financial Conditions Index	0.209 (0.758)	0.725 (0.541)	0.657 (0.499)	-0.438 (0.805)	0.116 (0.678)	0.607 (0.491)	0.0116 (0.430)	1.067* (0.559)	0.558** (0.238)
ISS^Leverage	-0.614*** (0.200)	-0.321 (0.222)	-0.351 (0.305)	0.314 (0.237)	-0.585** (0.295)	-0.912*** (0.206)	-1.349*** (0.0802)	-1.029*** (0.117)	-0.101* (0.0517)
Number of Observations	658	636	537	537	636	668	668	668	620
Number of Countries	42	38	38	38	38	42	42	42	37

Table 7a.

Downside Risks to Growth – Alternative Credit Series

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Panel A: Riskiness of Credit Allocation Based on EDF
Real GDP Growth	-0.0546 (0.144)	0.0236 (0.0773)	-0.0543 (0.127)	-0.168* (0.0998)	-0.0274 (0.140)	0.0300 (0.0883)	0.0868 (0.115)	0.0757 (0.0842)	-0.161 (0.118)
Δ(Domestic Bank Private Nonfinancial Credit-to-GDP) (IFS)	-0.257*** (0.0293)
Δ(Total Private Nonfinancial Credit-to-GDP) (BIS)		-0.260*** (0.0221)
Delta Total Privated Nonfinancial Credit-to-GDP: to Corporate			-0.275*** (0.0472)
Delta Total Privated Nonfinancial Credit-to-GDP: to Household				-0.835*** (0.0768)
Delta Total Privated Nonfinancial Credit-to-GDP: from Domestic Bank					-0.239*** (0.0398)
Δ(Cross-Border Credit to Non-Banks to GDP) (BIS)						-0.197*** (0.0603)
Δ(Cross-Border Credit to Banks to GDP) (BIS)							-0.175*** (0.0380)
Δ(Cross-Border Credit to Banks and Non-Banks to GDP) (BIS)								-0.145*** (0.0343)
Credit-to-GDP Gap (based on IFS)									-0.0778*** (0.00983)
Financial Conditions Index	-0.641 (0.587)	0.0545 (0.479)	1.224* (0.735)	0.618 (0.773)	-0.247 (0.562)	0.713 (0.588)	0.167 (0.709)	0.536 (0.396)	0.120 (0.457)
ISS^EDF	-1.321*** (0.164)	-1.426*** (0.202)	-0.983*** (0.206)	-0.662*** (0.231)	-1.349*** (0.156)	-1.288*** (0.182)	-1.478*** (0.215)	-1.435*** (0.254)	-1.349*** (0.298)
Number of Observations	586	567	490	490	567	592	592	592	557
Number of Countries	40	36	36	36	36	40	40	40	36
Panel B: Riskiness of Credit Allocation Based on Leverage
Real GDP Growth	-0.0316 (0.118)	-0.0898 (0.0858)	-0.388** (0.161)	-0.389*** (0.150)	0.0561 (0.0916)	-0.165 (0.114)	-0.00524 (0.0808)	0.0561 (0.100)	-0.0871* (0.0469)
Δ(Domestic Bank Private Nonfinancial Credit-to-GDP) (IFS)	-0.264*** (0.0459)
Δ(Total Private Nonfinancial Credit-to-GDP) (BIS)		-0.223*** (0.0435)
Delta Total Privated Nonfinancial Credit-to-GDP: to Corporates			-0.170** (0.0691)
Delta Total Privated Nonfinancial Credit-to-GDP: to Households				-1.037*** (0.100)
Delta Total Privated Nonfinancial Credit-to-GDP: from Domestic Banks					-0.279*** (0.0736)
Δ(Cross-Border Credit to Non-Banks to GDP) (BIS)						-0.383*** (0.0605)
Δ(Cross-Border Credit to Banks to GDP) (BIS)							-0.193*** (0.0283)
Δ(Cross-Border Credit to Banks and Non-Banks to GDP) (BIS)								-0.149*** (0.0414)
Credit-to-GDP Gap (based on IFS)									-0.102*** (0.00602)
Financial Conditions Index	0.209 (0.758)	0.725 (0.541)	0.657 (0.499)	-0.438 (0.805)	0.116 (0.678)	0.607 (0.491)	0.0116 (0.430)	1.067* (0.559)	0.558** (0.238)
ISS^Leverage	-0.614*** (0.200)	-0.321 (0.222)	-0.351 (0.305)	0.314 (0.237)	-0.585** (0.295)	-0.912*** (0.206)	-1.349*** (0.0802)	-1.029*** (0.117)	-0.101* (0.0517)
Number of Observations	658	636	537	537	636	668	668	668	620
Number of Countries	42	38	38	38	38	42	42	42	37

Panel A at the top of Table 8a provides results for ISS^Leverage, and Panel B at the bottom provides results for ISS^EDF. Results are very stable across columns and panels. ISS and the credit quantity variable are always significant, while the FCI is generally insignificant.

Table 8a.

Downside Risks to Growth – Alternative Firm Vulnerability Indicators

Table 8a.

Downside Risks to Growth – Alternative Firm Vulnerability Indicators

	(1)	(2)	(3)	(4)	(5)	(6)
	1 year		2 years		3 years
	Decile 1	Decile 2	Decile 1	Decile 2	Decile 1	Decile 2
Panel A: ISS based on ICR
Real GDP Growth	0.0995* (0.0574)	0.102*** (0.0359)	-0.0329 (0.0698)	0.0374 (0.0453)	-0.104 (0.0658)	-0.0950 (0.0848)
Δ(Credit-to-GDP Ratio)	-0.0568*** (0.0186)	-0.0704*** (0.0143)	-0.114*** (0.0315)	-0.168*** (0.0248)	-0.199*** (0.0593)	-0.211*** (0.0204)
Financial Conditions Index	0.526* (0.277)	0.115 (0.168)	1.764*** (0.198)	-0.328 (0.423)	1.846*** (0.360)	0.0361 (0.541)
ISS	-0.160 (0.219)	-0.186** (0.0810)	-1.004*** (0.377)	-0.691*** (0.177)	-1.500*** (0.293)	-1.398*** (0.238)
Number of Observations	651	651	651	651	651	651
Number of Countries	42	42	42	42	42	42
Panel B: ISS based on debt/EBITDA
Real GDP Growth	0.106* (0.0554)	0.124*** (0.0397)	0.0322 (0.0573)	0.0852* (0.0436)	-0.123 (0.133)	-0.0841 (0.0992)
Δ(Credit-to-GDP Ratio)	-0.0601*** (0.0165)	-0.0748*** (0.0159)	-0.140*** (0.0303)	-0.184*** (0.0223)	-0.307*** (0.0417)	-0.222*** (0.0271)
Financial Conditions Index	0.614** (0.241)	0.107 (0.196)	1.325* (0.718)	-0.0525 (0.491)	1.849*** (0.177)	-0.0268 (0.621)
ISS^Debt/EBITDA	-0.0987 (0.166)	-0.0866 (0.0839)	-0.296 (0.199)	-0.172 (0.125)	-0.322 (0.235)	-0.609*** (0.172)
Number of Observations	648	648	648	648	648	648
Number of Countries	42	42	42	42	42	42

Table 8a.

Downside Risks to Growth – Alternative Firm Vulnerability Indicators

	(1)	(2)	(3)	(4)	(5)	(6)
	1 year		2 years		3 years
	Decile 1	Decile 2	Decile 1	Decile 2	Decile 1	Decile 2
Panel A: ISS based on ICR
Real GDP Growth	0.0995* (0.0574)	0.102*** (0.0359)	-0.0329 (0.0698)	0.0374 (0.0453)	-0.104 (0.0658)	-0.0950 (0.0848)
Δ(Credit-to-GDP Ratio)	-0.0568*** (0.0186)	-0.0704*** (0.0143)	-0.114*** (0.0315)	-0.168*** (0.0248)	-0.199*** (0.0593)	-0.211*** (0.0204)
Financial Conditions Index	0.526* (0.277)	0.115 (0.168)	1.764*** (0.198)	-0.328 (0.423)	1.846*** (0.360)	0.0361 (0.541)
ISS	-0.160 (0.219)	-0.186** (0.0810)	-1.004*** (0.377)	-0.691*** (0.177)	-1.500*** (0.293)	-1.398*** (0.238)
Number of Observations	651	651	651	651	651	651
Number of Countries	42	42	42	42	42	42
Panel B: ISS based on debt/EBITDA
Real GDP Growth	0.106* (0.0554)	0.124*** (0.0397)	0.0322 (0.0573)	0.0852* (0.0436)	-0.123 (0.133)	-0.0841 (0.0992)
Δ(Credit-to-GDP Ratio)	-0.0601*** (0.0165)	-0.0748*** (0.0159)	-0.140*** (0.0303)	-0.184*** (0.0223)	-0.307*** (0.0417)	-0.222*** (0.0271)
Financial Conditions Index	0.614** (0.241)	0.107 (0.196)	1.325* (0.718)	-0.0525 (0.491)	1.849*** (0.177)	-0.0268 (0.621)
ISS^Debt/EBITDA	-0.0987 (0.166)	-0.0866 (0.0839)	-0.296 (0.199)	-0.172 (0.125)	-0.322 (0.235)	-0.609*** (0.172)
Number of Observations	648	648	648	648	648	648
Number of Countries	42	42	42	42	42	42

Similar robustness exercises for the crisis model are shown in Panel A and Panel B of Table 7b. Results are very stable across columns and panels. Change in credit-to-GDP is rarely significant, while ISS and FCI are always significant.

Table 7b.

Crisis prediction – Alternative credit series

Source: Authors’ estimates.Note: The estimates are obtained through a conditional fixed effects logit regression. Standard errors are shown in parentheses. Explanatory variables enter the regression as the lag of their simple three-year moving average and are demeaned at the country level; the change in credit-to-GDP ratio is winsorized at 1 percent. Controls include the change in current account-to-GDP ratio and the real GDP growth rate. *** p<0.01, ** p<0.05, * p<0.1.

B. Alternative firm-level vulnerability indicators

Our baseline firm-level indicators are the expected default frequency and the ratio of debt to assets. We examine the results obtained with two alternative vulnerability indicators sometimes used in the literature, the interest coverage ratio (ICR), and the debt-to-EBITDA ratio. Table 8a shows the results obtained in the downside risks to growth model, while Table 8b shows the results obtained in the crisis occurrence model. Results are very similar to those obtained with the two core vulnerability indicators: ISS is always significant in the crisis model and is associated to shifts in the left tail of the future GDP growth distribution, although more significantly so for the ICR indicator.

Table 8b.

Crisis prediction – Alternative Firm Vulnerability Indicators

Source: Authors’ estimates.Note: The estimates are obtained through a conditional fixed effects logit regression. Standard errors are shown in parentheses. Explanatory variables enter the regression as the lag of their simple three-year moving average and are demeaned at the country level; the change in credit-to-GDP ratio is winsorized at 1 percent. Controls include the change in current account-to-GDP ratio and the real GDP growth rate. *** p<0.01, ** p<0.05, * p<0.1.

Table 8b.

Crisis prediction – Alternative Firm Vulnerability Indicators

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Δ(Credit-to-GDP Ratio)	0.202*** (0.0699)		0.120 (0.0750)		0.185** (0.0839)	0.0593 (0.109)		0.168** (0.0815)	0.0584 (0.108)
Financial Conditions Index		-2.468** (0.963)	-2.234** (0.947)			-4.581*** (1.615)			-3.616*** (0.867)
ISS^Debt/EBITDA				1.393*** (0.404)	1.401*** (0.494)	2.749*** (0.672)
ISS^ICR							1.557** (0.624)	1.488* (0.781)	3.082*** (1.132)
Number of Observations	443	443	443	431	431	431	432	432	432
Number of Countries	21	21	21	20	20	20	20	20	20
Number of Crisis Episodes	21	21	21	20	20	20	20	20	20
Pseudo R²	0.243	0.383	0.353	0.298	0.342	0.611	0.262	0.304	0.548

Source: Authors’ estimates.Note: The estimates are obtained through a conditional fixed effects logit regression. Standard errors are shown in parentheses. Explanatory variables enter the regression as the lag of their simple three-year moving average and are demeaned at the country level; the change in credit-to-GDP ratio is winsorized at 1 percent. Controls include the change in current account-to-GDP ratio and the real GDP growth rate. *** p<0.01, ** p<0.05, * p<0.1.

Table 8b.

Crisis prediction – Alternative Firm Vulnerability Indicators

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Δ(Credit-to-GDP Ratio)	0.202*** (0.0699)		0.120 (0.0750)		0.185** (0.0839)	0.0593 (0.109)		0.168** (0.0815)	0.0584 (0.108)
Financial Conditions Index		-2.468** (0.963)	-2.234** (0.947)			-4.581*** (1.615)			-3.616*** (0.867)
ISS^Debt/EBITDA				1.393*** (0.404)	1.401*** (0.494)	2.749*** (0.672)
ISS^ICR							1.557** (0.624)	1.488* (0.781)	3.082*** (1.132)
Number of Observations	443	443	443	431	431	431	432	432	432
Number of Countries	21	21	21	20	20	20	20	20	20
Number of Crisis Episodes	21	21	21	20	20	20	20	20	20
Pseudo R²	0.243	0.383	0.353	0.298	0.342	0.611	0.262	0.304	0.548

Source: Authors’ estimates.Note: The estimates are obtained through a conditional fixed effects logit regression. Standard errors are shown in parentheses. Explanatory variables enter the regression as the lag of their simple three-year moving average and are demeaned at the country level; the change in credit-to-GDP ratio is winsorized at 1 percent. Controls include the change in current account-to-GDP ratio and the real GDP growth rate. *** p<0.01, ** p<0.05, * p<0.1.

C. Alternative constructions of the riskiness of credit allocation

To provide further evidence that the composition of credit matters to financial stability outcomes, we explore seven perturbations of the construction of the riskiness of credit allocation variable:

First, we use the raw vulnerability measure (instead of its decile) in the ISS formula, as Gomes et al. (2018) do for the United States.
Second, we keep the deciles, but weigh them by debt instead of taking their simple average. Firms with relatively larger debt will therefore have a greater impact in this version.
Third, we sort firms by net issuance in US dollars (instead of net issuance normalized by lagged assets). Large firms will have a greater influence on this version of the measure.
Fourth, we sort firms by their level of vulnerability (instead of their net issuance to lagged assets) and use deciles of net issuance to lagged assets (instead of deciles of vulnerability). This is a “reverse” ISS.
Fifth, we combine the first and the third perturbation, i.e. we sort firms by net issuance in US dollars and we use the raw vulnerability measure.

The coefficients of interest for the downside risks to growth model are shown in Table 9a, and in Table 9b for the crisis prediction model. All five alternatives of ISS based on EDF have the right sign and are significant, and four out of five alternatives of the ISS based on leverage have the right sign and are significant (the second alternative is not significant), .

Table 9a.

Downside Risks to Growth – Alternative constructions of the Riskiness of Credit Allocation

Source: Authors’ estimates.Note: The table shows estimated coefficients of the riskiness of credit allocation in the downside-risks-to-growth model, as in Table 5, column (6). Row (0) reproduces the results obtained with the baseline ISS. The alternative indictors are constructed as follows: in row (1), the raw vulnerability measure is used (instead of its decile); in row (2), the vulnerability measure transformed into a decile is weighted by debt; in row (3), firms are sorted by the absolute amount of their debt issuance (instead of their debt issuance normalized by lagged assets); in row (4), firms are sorted by vulnerability level (instead of net debt issuance) and we take the difference in average net debt issuance to assets (transformed into deciles) across top vulnerability and bottom vulnerability firms; in row (5), we combine perturbations (1) and (3). All explanatory variables enter the regression as the lag of their simple three-year moving average. The change in the credit-to-GDP ratio, the FCI, and real GDP growth are included in all regressions. *** p<0.01; ** p<0.05; * p<0.1.

Table 9a.

Downside Risks to Growth – Alternative constructions of the Riskiness of Credit Allocation

Alternative ISS constructions	(1)	(2)	(3)	(4)	(5)	(6)
	Based on EDF			Based on Leverage
	Beta	Std	Obs.	Beta	Std	Obs.
(0) Baseline	-1.321***	(0.164)	586	-0.614***	(0.200)	658
Alternative vulnerability scale
(1) Simple average of raw vulnerability	-0.120***	(0.0229)	586	-7.586***	(2.397)	658
Alternative weighting
(2) Debt-weighted deciles	-0.856***	-0.239	586	0.0246	(0.334)	658
Alternative sorting
(3) By net debt issuance in USD	-0.962***	(0.193)	586	-0.948***	(0.193)	658
(4) By vulnerability	-1.339***	(0.132)	586	-0.953***	(0.253)	658
Alternative vulnerability scale and sorting
(5) = (1) & (3)	-0.0563***	(0.0241)	586	-9.957*	(2.948)	658

Table 9a.

Downside Risks to Growth – Alternative constructions of the Riskiness of Credit Allocation

Alternative ISS constructions	(1)	(2)	(3)	(4)	(5)	(6)
	Based on EDF			Based on Leverage
	Beta	Std	Obs.	Beta	Std	Obs.
(0) Baseline	-1.321***	(0.164)	586	-0.614***	(0.200)	658
Alternative vulnerability scale
(1) Simple average of raw vulnerability	-0.120***	(0.0229)	586	-7.586***	(2.397)	658
Alternative weighting
(2) Debt-weighted deciles	-0.856***	-0.239	586	0.0246	(0.334)	658
Alternative sorting
(3) By net debt issuance in USD	-0.962***	(0.193)	586	-0.948***	(0.193)	658
(4) By vulnerability	-1.339***	(0.132)	586	-0.953***	(0.253)	658
Alternative vulnerability scale and sorting
(5) = (1) & (3)	-0.0563***	(0.0241)	586	-9.957*	(2.948)	658

Table 9b.

Crisis prediction – Alternative constructions of the Riskiness of Credit Allocation

Source: Authors’ estimates.Note: The table shows estimated coefficients of the riskiness of credit allocation in the crisis prediction model, as in columns (7) and (10) of Table 3. Row (0) reproduces the results obtained with the baseline ISS. The alternative indictors are constructed as follows: in row (1), the raw vulnerability measure is used (instead of its decile); in row (2), the vulnerability measure transformed into a decile is weighted by debt; in row (3), firms are sorted by the absolute amount of their debt issuance (instead of their debt issuance normalized by lagged assets); in row (4), firms are sorted by vulnerability level (instead of net debt issuance) and we take the difference in average net debt issuance to assets (transformed into deciles) across top vulnerability and bottom vulnerability firms; in row (5), we combine perturbations (1) and (3). All explanatory variables enter the regression as the lag of their simple three-year moving average. Country fixed effects, the change in the credit-to-GDP ratio, the FCI, the change in current-account-to-GDP ratio, and real GDP growth are included in all regressions. *** p<0.01; ** p<0.05; * p<0.1.

Table 9b.

Crisis prediction – Alternative constructions of the Riskiness of Credit Allocation

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Alternative ISS constructions	Based on Expected Default Frequency				Based on Leverage
	Beta	Std	Pseudo R²	Obs.	Beta	Std	Pseudo R²	Obs.
(0) Baseline	3.594***	(1.042)	0.749	361	2.560***	(0.746)	0.55	443
Alternative vulnerability scale
(1) Simple average of raw vulnerability	0.381***	(0.111)	0.7	361	40.81***	(14.79)	0.543	443
Alternative weighting
(2) Debt-weighted deciles	1.761***	(0.624)	0.701	361	1.913**	(0.865)	0.485	443
Alternative sorting
(3) By net debt issuance in USD	3.159***	(1.075)	0.704	361	1.786***	(0.572)	0.466	443
(4) By vulnerability	3.007**	(1.305)	0.739	361	1.537**	(0.729)	0.498	443
Alternative vulnerability scale and sorting
(5) = (1) & (3)	0.393***	(0.122)	0.685	361	29.63***	(7.599)	0.475	443

Table 9b.

Crisis prediction – Alternative constructions of the Riskiness of Credit Allocation

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Alternative ISS constructions	Based on Expected Default Frequency				Based on Leverage
	Beta	Std	Pseudo R²	Obs.	Beta	Std	Pseudo R²	Obs.
(0) Baseline	3.594***	(1.042)	0.749	361	2.560***	(0.746)	0.55	443
Alternative vulnerability scale
(1) Simple average of raw vulnerability	0.381***	(0.111)	0.7	361	40.81***	(14.79)	0.543	443
Alternative weighting
(2) Debt-weighted deciles	1.761***	(0.624)	0.701	361	1.913**	(0.865)	0.485	443
Alternative sorting
(3) By net debt issuance in USD	3.159***	(1.075)	0.704	361	1.786***	(0.572)	0.466	443
(4) By vulnerability	3.007**	(1.305)	0.739	361	1.537**	(0.729)	0.498	443
Alternative vulnerability scale and sorting
(5) = (1) & (3)	0.393***	(0.122)	0.685	361	29.63***	(7.599)	0.475	443

D. Other robustness checks

Online Appendix B provides additional results of and further robustness checks. We show that the results for the crisis model hold when including including individual lags of ISS (instead of its moving average), and when excluding three years post crisis so as to avoid the post-crisis bias discussed in Bussière and Fratscher (2006). Results in the downside risks to growth model are robust to using GDP per capita. Results of both models are both robust to the inclusion of other early warning indicators identified in the crisis literature are controlled for, including real effective exchange rate and foreign exchange reserves (as Gourinchas and Obstfeld, 2012) or aggregate corporate vulnerability indicators (as Lee et al., 2018). .

VIII. Discussion of Possible Mechanisms

We have provided strong evidence of a positive association between the riskiness of credit allocation and both future downside risks to GDP growth and the probability of financial crisis. In this section we explore two mechanisms that could plausibly explain these associations.

First, the riskiness of credit allocation is likely to at least partially capture variations in lending standards to the corporate sector, and therefore the extent to which the most vulnerable firms accumulate further debt, thereby becoming even more vulnerable. Recent research suggests that higher firm vulnerability leads to lower access to credit and lower investment when financial conditions tighten (Duval et al., forthcoming). If it led to a fatter tail of vulnerable firms, a higher level of the riskiness of credit allocation would then amplify the effect of a negative shock on investment and economic activity. To check whether this meachanism is at play, we run the following regression:

\begin{matrix} Δ S h a r e_{i, t}^{H V} = α_{i} + β Δ {(\frac{C r e d i t}{G D P})}_{i, t} + γ F C I_{i, t} + δ I S S_{i, t}^{V} + u_{i, t} & (6) \end{matrix}

where Share^HV is the share of assets in high vulnerability firms, and V ∈ {EDF, leverage} as above.²⁴

Table 10 presents the results. High vulnerability firms are defined alternatively as those with a leverage ratio above the 75^th percentile of their country-specific distribution (columns (1) and (4)), those with an ICR below the 25^th percentile of their country-specific distribution (columns (2) and (5)), and those with a debt-to-EBITDA ratio above the 75^th percentile of their country-specific distribution (columns (3) and (6)). The regression results confirm that this mechanism plays a role. Regardless of the ISS indicator and the high vulnerability measure used, the coefficient of ISS is very significant, which indicates that changes in ISS capture changes in the distribution of corporate vulnerabilities.

Table 10.

Riskiness of Credit Allocation and Change in the Share of Vulnerable Assets

Source: Authors’ estimatesNote: All regressions are OLS and include country fixed effects. Standard errors are clustered at the country level and shown in parentheses. *** p<0.01; ** p<0.05; *p<0.1.

Table 10.

Riskiness of Credit Allocation and Change in the Share of Vulnerable Assets

	(1)	(2)	(3)	(4)	(5)	(6)
	Dependent Variable: ∆Share of Assets in Firms with High Vulnerability
	Leverage>p75	ICR<p25	Debt-to-EBITDA Ratio P75	Leverage>p75	ICR<p25	Debt-to-EBITDA Ratio>P75
∆Credit-to-GDP	0.18*** (0.05)	0.32*** (0.09)	0.36*** (0.09)	0.10* (0.06)	0.21** (0.09)	0.24*** (0.08)
FCI	0.85*** (0.23)	1.49*** (0.23)	1.46*** (0.26)	0.89*** (0.21)	1.42*** (0.21)	1.45*** (0.26)
ISS^EDF	1.11*** (0.30)	1.08*** (0.31)	0.65** (0.32)
ISS^Leverage				2.50*** (0.31)	2.81*** (0.51)	2.24*** (0.46)
Number of Observations	812	812	805	842	842	833
Number of Countries	41	41	41	42	42	42
R²	0.04	0.07	0.08	0.11	0.14	0.12

Table 10.

Riskiness of Credit Allocation and Change in the Share of Vulnerable Assets

	(1)	(2)	(3)	(4)	(5)	(6)
	Dependent Variable: ∆Share of Assets in Firms with High Vulnerability
	Leverage>p75	ICR<p25	Debt-to-EBITDA Ratio P75	Leverage>p75	ICR<p25	Debt-to-EBITDA Ratio>P75
∆Credit-to-GDP	0.18*** (0.05)	0.32*** (0.09)	0.36*** (0.09)	0.10* (0.06)	0.21** (0.09)	0.24*** (0.08)
FCI	0.85*** (0.23)	1.49*** (0.23)	1.46*** (0.26)	0.89*** (0.21)	1.42*** (0.21)	1.45*** (0.26)
ISS^EDF	1.11*** (0.30)	1.08*** (0.31)	0.65** (0.32)
ISS^Leverage				2.50*** (0.31)	2.81*** (0.51)	2.24*** (0.46)
Number of Observations	812	812	805	842	842	833
Number of Countries	41	41	41	42	42	42
R²	0.04	0.07	0.08	0.11	0.14	0.12

Second, the riskiness of credit allocation could capture a dimension of investor sentiment that is not captured either by the financial conditions index of the change in credit-to-GDP. If so, it should help predict reversals of financial conditions and/or corporate spreads. We thus follow Lopez-Salido et al. (2017) and estimate:

\begin{matrix} Δ F_{i, t} = α_{i} + β Δ {(\frac{C r e d i t}{G D P})}_{i, t - 1}^{m v 3} + γ F_{i, t - 1}^{m ν 3} + δ I S S_{i, t - 1}^{V, m ν 3} + u_{i, t}, & (7) \end{matrix}

where F is either the financial conditions index or the corporate spread, and ΔF is its first difference.

Table 11 presents evidence that the riskiness of credit allocation indeed helps predict reversals in financial conditions (columns (1)-(4)) and the corporate spread (columns (5)-(8)). For the reversal in financial conditions, the effect is more than twice stronger when financial conditions are loose (columns (2) and (4)). The riskiness of credit allocation therefore has features of a risk sentiment indicator.

Table 11.

Riskiness of Credit allocation and Reversal in Financial Conditions

Source: Authors’ estimates.Note: All regressions are OLS and include country fixed effects. The dependent variable is the change in FCI for columns (1)–(4) and the change in the corporate credit spread in columns (5)-(8). Explanatory variables enter as the lag of their simple three-year moving average. In columns (2) and (4), the sample is restricted to observations for which the lag of the FCI’s simple three-year moving average is negative. In columns (6) and (8), the sample is restricted to observations for which the lag of the spread’s simple three-year moving average is negative. Standard errors, clustered at the country level, are shown in parentheses. *** p<0.01; ** p<0.05; *p<0.1.

Table 11.

Riskiness of Credit allocation and Reversal in Financial Conditions

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Dependent Variable:
	∆FCI				∆spread
FCI	-0.68*** (0.03)	-1.45*** (0.08)	-0.67*** (0.03)	-1.41*** (0.08)
Corporate Credit Spread					-0.35*** (0.04)	-0.41*** (0.08)	-0.36*** (0.04)	-0.34*** (0.08)
ISS^EDF	0.11** (0.05)	0.27*** (0.09)			0.12** (0.04)	0.12** (0.06)
ISS^Leverage			0.08** (0.04)	0.25*** (0.08)			0.05* (0.03)	0.13** (0.05)
Sample	Full	I.FCI_mv3<0	Full	I.FCI_mv3<0	Full	I.spread_mv3<0	Full	I.spread_mv3<0
Number of Observations	707	348	780	377	552	328	580	345
Number of Countries	40	40	42	41	37	37	37	37
R²	0.28	0.30	0.29	0.31	0.15	0.06	0.14	0.08

Table 11.

Riskiness of Credit allocation and Reversal in Financial Conditions

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Dependent Variable:
	∆FCI				∆spread
FCI	-0.68*** (0.03)	-1.45*** (0.08)	-0.67*** (0.03)	-1.41*** (0.08)
Corporate Credit Spread					-0.35*** (0.04)	-0.41*** (0.08)	-0.36*** (0.04)	-0.34*** (0.08)
ISS^EDF	0.11** (0.05)	0.27*** (0.09)			0.12** (0.04)	0.12** (0.06)
ISS^Leverage			0.08** (0.04)	0.25*** (0.08)			0.05* (0.03)	0.13** (0.05)
Sample	Full	I.FCI_mv3<0	Full	I.FCI_mv3<0	Full	I.spread_mv3<0	Full	I.spread_mv3<0
Number of Observations	707	348	780	377	552	328	580	345
Number of Countries	40	40	42	41	37	37	37	37
R²	0.28	0.30	0.29	0.31	0.15	0.06	0.14	0.08

We conclude this discussion by asking in the spirit of Mian et al. (2017) whether there could be a role for behavioral biases in explaining the negative relationship between riskiness of credit allocation and downside risks to future growth. While this question is difficult to answer in the absence of data on expectations of downside risks to growth, one can nevertheless approach it by examining data on GDP growth forecasts. We thus ask whether professional economic forecasts –specifically, the IMF’s World Economic Outlook (WEO) forecasts– properly capture the relationship between ISS and future GDP growth documented in Table 3 and discussed in Section V above. To that effect, we estimate versions of Equation (5) where the dependent variable is forecasted GDP growth or the GDP growth forcast error.

Results are presented in Table 12. Columns (1) and (4) reproduce the results obtained for 3-year ahead cumulative growth. Columns (2) and (5) show that when ISS is high, professional forecasters are also mistakenly anticipating higher GDP growth in the future. Column (3) and (6) then show that the forecast error is positive, significant, and quantitatively large, with a one standard deviation increase in ISS being associated with a forecast error greater that 0.5 percentage points. This result seems difficult to reconcile with a rational expectations-based model, and suggests that economic agents fail to understand the some of the negative effects (such as an increase in downside risks to growth) of an increase in the riskiness of credit allocation.

Table 12.

Riskiness of Credit allocation, GDP Growth Forecast, and Forecast Error

Source: Authors’ estimates.Note: The estimates are obtained through OLS with time and country fixed effects. Standard errors are in parentheses. Explanatory variables enter the regression as the lag of their simple three-year moving average and are demeaned at the country level; the change in the credit-to-GDP ratio is winsorized at 1 percent. Real GDP growth is controlled for. *** p<0.01, ** p<0.05, * p<0.1.

Table 12.

Riskiness of Credit allocation, GDP Growth Forecast, and Forecast Error

	(1)	(2)	(3)	(4)	(5)	(6)
	Actual	Forecast	Error	Actual	Forecast	Error
Panel A: ISS based on EDF
Δ(Credit-to-GDP Ratio)	-0.29*** (0.05)	-0.01 (0.02)	0.28*** (0.06)	-0.41*** (0.05)	0.00 (0.02)	0.41*** (0.06)
Financial Conditions Index	-0.89* (0.49)	-0.13 (0.21)	0.76 (0.51)	-0.98** (0.49)	-0.01 (0.21)	0.97* (0.51)
ISS^EDF	-0.25 (0.27)	0.28** (0.12)	0.53* (0.29)
ISS^Leverage				-0.23 (0.3)	0.31** (0.13)	0.55* (0.31)
Number of Observations	586	586	586	658	658	658
Number of Countries	40	40	40	42	42	42

Table 12.

Riskiness of Credit allocation, GDP Growth Forecast, and Forecast Error

	(1)	(2)	(3)	(4)	(5)	(6)
	Actual	Forecast	Error	Actual	Forecast	Error
Panel A: ISS based on EDF
Δ(Credit-to-GDP Ratio)	-0.29*** (0.05)	-0.01 (0.02)	0.28*** (0.06)	-0.41*** (0.05)	0.00 (0.02)	0.41*** (0.06)
Financial Conditions Index	-0.89* (0.49)	-0.13 (0.21)	0.76 (0.51)	-0.98** (0.49)	-0.01 (0.21)	0.97* (0.51)
ISS^EDF	-0.25 (0.27)	0.28** (0.12)	0.53* (0.29)
ISS^Leverage				-0.23 (0.3)	0.31** (0.13)	0.55* (0.31)
Number of Observations	586	586	586	658	658	658
Number of Countries	40	40	40	42	42	42

IX. Conclusion

In the conclusion to their paper, Greenwood and Hanson (2013) extrapolate their results obtained in the context of the U.S. corporate bond market and suggest that to identify the existence of a sentiment-driven credit boom and implement countercyclical credit policy “[…] looking at credit quantities or credit spreads is not enough – policy makers should also consider the credit quality of debt market financing”. In this paper, we show that their ISS measure of debt issuer quality, which we refer to as the riskiness of credit allocation, helps predict shifts in the left tail of the GDP growth distribution as well as systemic banking crises 2 to 3 years ahead in a sample of 55 countries covering the 1991–2016 period. We show that this predictive power of riskiness of credit allocation is additional to that of changes in aggregate credit quantities and that of the price of risk typically emphasized in the financial stability literature. Further, we provide evidence that shifts in credit supply play a role in explaining variations in the riskiness of credit allocation, that these variations are associated with variations in the thickness of the weak tail of the distribution of corporate vulnerability measures, and with future reversals of financial conditions. We also show that economic forecasters wrongly associate increases in ISS with increases in future GDP growth.

Our analysis has implications for macroprudential policy-makers. The calibration of the countercyclical capital buffer currently gives a prominent role to the quantity of aggregate credit and the so-called credit-gap. Our findings suggest that policy-makers also need to take the riskiness of credit allocation into account when seeking to prevent financial instability episodes and to differentiate good credit booms from bad credit booms.

Our findings on the dynamics of the composition of corporate credit flows and on the mistaken perception that riskiness of credit allocation and future GDP growth are positively associated also have implications for theoretical models of credit expansions. They favor models that emphasize credit supply shocks and where behavioral biases play an important role.

We established the predictive performance of the riskiness of credit allocation in sample. Although Appendix Table B8 provides evidence of ISS’s predictive power for downside risks to growth in the pre-2008 sample too, reducing concerns that ISS captures only developments around the Great Recession, it would be interesting to establish ISS’s out-of-sample performance. This would require longer series, or series at a higher frequency, and is left for future research.

X. References

Acharya, Viral V., Tim Eisert, Christian Eufinger, and Christian W. Hirsch. 2019. “Whatever It Takes: The Real Effects of Unconventional Monetary Policy.”, The Review of Financial Studies, in print.
- Search Google Scholar
- Export Citation
Adrian, Tobias, Nina Boyarchenko, and Domenico Giannone. 2019. “Vulnerable Growth.” American Economic Review.
- Search Google Scholar
- Export Citation
Adrian, Tobias, Federico Grinberg, Nellie Liang, and Sheheryar Malik. 2018. “The Term Structure of Growth-at-Risk”, IMF Working Paper 18/180.
- Search Google Scholar
- Export Citation
Aldasoro, Iñaki, Claudio Borio, and Mathias Drehmann. 2018. “Early Warning Indicators of Banking Crises: Expanding the Family.” BIS Quarterly Review, March: 29–45.
- Search Google Scholar
- Export Citation
Alp, Aysun. 2013. “Structural Shifts in Credit Rating Standards”, Journal of Finance, 68(6), December, pp 2435–2470.
- Search Google Scholar
- Export Citation
Alter, Adrian, Alan Xiaochen Feng, and Nico Valckx, “Understanding the Macro-Financial Effects of Household Debt: A Global Perspective”, IMF Working Paper 18/76
- Search Google Scholar
- Export Citation
Baghai, Ramin P., Henri Servaes, and Ane Tamayo. 2014. “Have Rating Agencies Become More Conservative? Implications for Capital Structure and Debt Pricing”, Journal of Finance, 69(5), October, pp 1961–2005.
- Search Google Scholar
- Export Citation
Banco de España. 2017. “Financing and Investment Decisions of Spanish Non-Financial Corporations.” 2016 Annual Report.
- Search Google Scholar
- Export Citation
Baron, Matthew, and Wei Xiong. 2017. “Credit Expansion and Neglected Crash Risk.” Quarterly Journal of Economics 132 (2): 713–64.
- Search Google Scholar
- Export Citation
Berger, Allen N., and Gregory F. Udell. 2004. “The Institutional Memory Hypothesis and the Procyclicality of Bank Lending Behavior.” Journal of Financial Intermediation 13 (4): 458–95.
- Search Google Scholar
- Export Citation
Bernanke, Ben S., and Mark Gertler. 1989. “Agency Costs, Net Worth and Business Fluctuations.” American Economic Review 79 (1): 14–31.
- Search Google Scholar
- Export Citation
Bernanke, Ben S., and Mark Gertler, and Simon Gilchrist. 1996. “The Flight to Quality and the Financial Accelerator.” Review of Economics and Statistics 78 (1): 1–15.
- Search Google Scholar
- Export Citation
Bordalo, Pedro, Nicola Gennaioli, and Andrei Shleifer. 2018. “Diagnostic Expectations and Credit Cycles.” Journal of Finance, published online.
- Search Google Scholar
- Export Citation
Borio, Claudio, and Philip Lowe. 2002. “Assessing the Risk of Banking Crises.” BIS Quarterly Review, December: 43–54.
- Search Google Scholar
- Export Citation
Borio, Claudio, and Mathias Drehmann. 2009. “Assessing the Risk of Banking Crises -Revisited.” BIS Quarterly Review, March: 29–46.
- Search Google Scholar
- Export Citation
Borio, Claudio, Mathias Drehmann, and Kostas Tsatsaronis (2011). “Anchoring countercyclical Capital Buffers: The role of Credit Aggregates,” International Journal of Central Banking 7 (4), 189–240.
- Search Google Scholar
- Export Citation
Bussière, Matthieu, and Marcel Fratzscher. 2006. “Towards a New Early Warning System of Financial Crises”, Journal of International Money and Finance, 25(6), 953–973
- Search Google Scholar
- Export Citation
Büyükkarabacak, Berrak, Valev, Neven T., 2010. “The role of household and business credit in banking crises”. Journal of Banking and Finance, 34 (6), 1247–1256
- Search Google Scholar
- Export Citation
Chamberlain, Gary. 1980. “Analysis of Covariance with Qualitative Data.” Review of Economic Studies 47: 225–238
- Search Google Scholar
- Export Citation
Coupé, Tom. 2005, “Bias in Conditional and Unconditional Fixed Effects Logit Estimation: A Correction”, Political Analysis, 13: 292–295.
- Search Google Scholar
- Export Citation
Dell’Ariccia, Giovanni, Deniz Igan, Luc Laeven, and Hui Tong. 2016. “Credit Booms and Macrofinancial Stability.” Economic Policy 31 (86): 299–355.
- Search Google Scholar
- Export Citation
Dell’Ariccia, Giovanni, Luc Laeven, and Robert Marquez. 2014. “Real Interest Rates, Leverage, and Bank Risk-Taking.” Journal of Economic Theory 149 (1): 65–99.
- Search Google Scholar
- Export Citation
Dell’Ariccia, Giovanni, Luc Laeven, and Gustavo A. Suarez. 2017. “Bank Leverage and Monetary Policy’s Risk-Taking Channel: Evidence from the United States.” Journal of Finance 72 (2): 613–54.
- Search Google Scholar
- Export Citation
Dell’Ariccia, Giovanni, and Robert Marquez. 2006. “Lending Booms and Lending Standards.” Journal of Finance 61 (5): 2511–46.
- Search Google Scholar
- Export Citation

Topics

Countries

Series

About

Resources

Contributor Notes

Abstract

I. Introduction

II. Construction of the Riskiness of Credit Allocation and Data Sources

A. Construction of the Riskiness of Credit Allocation indicator

B. Other data

III. The Riskiness of Credit Allocation and Its Evolution Across Countries

IV. The Cyclicality of the Riskiness of Credit Allocation

V. The Riskiness of Credit Allocation and Downside Risks to Growth

VI. The Riskiness of Credit Allocation and the Occurrence of Banking Crises

VII. Robustness Analysis

A. Alternative aggregate credit series

B. Alternative firm-level vulnerability indicators

C. Alternative constructions of the riskiness of credit allocation

D. Other robustness checks

VIII. Discussion of Possible Mechanisms

IX. Conclusion

X. References