## _wp11161

## Source details

**Canonical URL:** [_wp11161](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11161.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11161.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11161.pdf.json)

---

### Introduction
- Objective: uncover macrofinancial vulnerabilities from linkages between nonperforming loans (NPL) and macroeconomic performance in advanced economies.
- Sample: 26 advanced countries spanning 1998 to 2009.
- Approaches used:
  - Single-equation panel regressions to address determinants of NPL.
  - Panel vector autoregressive (PVAR) approach and impulse response functions (IRFs) to examine interactions among variables and amplification via falling asset prices and credit constraints.

### Literature — theoretical background and empirical regularities
- Theoretical frameworks emphasized:
  - Business cycle models with financial intermediation (Williamson, 1987).
  - Financial accelerator theory (Bernanke and Gertler (1989); Bernanke and Gilchrist (1999); Kiyotaki and Moore (1997)).
- Empirical regularities:
  - Procyclicality of bank credit, NPL, and loan loss provisions.
  - Loosening of lending standards in upturns can sow seeds of later distress.
  - Determinants of NPL include both institutional/structural factors and macroeconomic variables (GDP growth, unemployment, asset prices, interest rates).

### Empirical approach and data
- Two-pronged empirical approach:
  - Single-equation panel: NPL_it = Y_it́ β + ν_i + ε_it (lags of NPL included; ν_i country fixed effects).
  - Panel VAR: y_it = B(L) y_it + ε_it with IRFs derived from moving-average representation; generalized IRFs used.
- Sample coverage and key variable definitions:
  - Countries: AUS, AUT, BEL, CAN, CZE, DEN, FIN, FRA, GER, GRE, ICL, ISR, ITA, JPN, KOR, LUX, NET, NZL, NOR, POR, SGP, SPA, SWE, SWI, UK, US.
  - NPL: ratio of overdue loans to gross loans, in percent.
  - Other variables: GRW (GDP growth), UNEMP (unemployment), DLHOUSEPR, DLEQUITYPR, INFL, NEER, IPOL, PSCGDP.
- Unit-root and stationarity assessment (Fisher tests; significance markers preserved):
  - Selected entries from Table 1 (Fisher-ADF; Fisher-PP):
    - NPL Level 80.955***; 49.844
    - NPL 1st difference -2.398***; 72.662**
    - IPOL Level 107.931***; 91.916***
    - IPOL 1st difference -8.194***; 151.568***
    - LNEQUITYPR Level 98.626***; 55.185
    - LNEQUITYPR 1st difference -6.847***; 132.034***
    - LNHOUSEPR Level 89.091***; 57.197
    - LNHOUSEPR 1st difference -2.016**; 49.167
    - NEER Level 28.054; 15.671
    - NEER 1st difference 150.440***; 78.829***
    - PSCGDP Level 25.611; 28.326
    - PSCGDP 1st difference -6.170***; 123.911***
    - GRW Level 94.595***; 88.071***
    - GRW 1st difference 95.045***; 117.606***
    - INFL Level 152.803***; 179.480***
    - INFL 1st difference -11.645***; 251.853***
    - UNEMP Level 39.134; 51.613
    - UNEMP 1st difference 157.570***; 110.466***
  - Note: *, **, and *** denote significance at 10 percent, 5 percent, and 1 percent, respectively.
- Cointegration:
  - Johansen trace and maximum eigenvalue tests indicate cointegrating relationships with number of vectors ranging from 3 to 5.

### Stylized facts and NPL patterns across the sample period
- Sub-period behavior:
  - Pre-2003: relatively high or rising NPL (aftereffects of East-Asia crisis and dot-com bubble).
  - 2003–07 precrisis: strong macro performance and low/declining NPL.
  - 2008–09: slower growth or recessions associated with increases in NPL across all countries.
- Extremes and ranges observed:
  - GDP growth: ranges from negative 8 percent in Finland in 2008 to 10.7 percent in Korea in 1999.
  - Unemployment: ranges from about 1 percent in Iceland in 2007 to 18 percent in Spain (year not specified).
  - Maximum NPL observed in Iceland in 2009; minimum observed in Belgium in 1999.
- Cross-correlations:
  - Indicators of good macro performance and rising asset prices are negatively related with NPL.

### Single-equation panel regression estimation approach
- Models estimated:
  - OLS with lagged dependent variables (model 1).
  - Panel-corrected standard error (PCSE) models with fixed effects (models 2 and 3).
  - Alternatives excluding lagged NPL or fixed effects (models 4 and 5).
  - One-step Arellano-Bond GMM (model 6).
- Estimator and diagnostics notes:
  - Fixed effects appropriate for models (1)–(4) (Fisher and chi-square tests reject redundancy).
  - Arellano-Bond diagnostics: reject null of zero first-order serial correlation; do not reject absence of second-order serial correlation where required.
  - One-step GMM preferred in small samples for lower bias; corrected LSDV infeasible due to short time span and unbalanced sample.

### Key regression results (selected coefficients and diagnostics)
- Lagged NPL:
  - NPL first lag: 0.875*** (model 1); 0.608*** (AB_GMM, model 6).
  - NPL 2nd lag: -0.239* (model 1); 0.355*** (model 4); 0.168*** (model 6).
- GDP growth (GRW):
  - GRW: -0.179*** (model 1); -0.157*** (model 6).
  - GRW t-1: 0.168*** (model 2); -0.087* (model 4); -0.040* and -0.042* in other specs.
- Unemployment (DUNEMP):
  - Coefficients around 0.204*** to 0.245*** where included.
- Asset prices:
  - DLHOUSEPR: negative and significant across specs, e.g., -0.026* (model 1), -0.047*** (model 2), -0.027*** (model 3), -0.021*** (model 5), -0.065** (model 6).
  - DLEQUITYPR: negative and significant across specs, e.g., -0.009*** to -0.014***.
- Policy rate (IPOL):
  - Positive where included, e.g., 0.071 (model 1, not significant) and 0.256** (model 5).
- Model fit and observations:
  - Number of observations: 244, 267, 244, 264, 241, 241 (across reported models).
  - R-squared examples: 0.863 (model 1), 0.493 (model 2), 0.543 (model 3), 0.357 (model 4), 0.403 (model 5).
  - Arellano-Bond AR(1) p-value reported as 0.045** and AR(2) p-value as 0.175 where applicable.

### Panel VAR impulse responses — main quantitative findings
- General statement:
  - Over a four-year forecast period NPL both affects and responds to every single variable in the VAR with statistically significant effects and responses.
- Responses to exogenous shocks (selected):
  - House prices shock:
    - A one standard deviation increase in the change in house prices, amounting to 6.1 percentage points, causes NPL to increase by 0.3 percentage points the first year and by 1.5 percentage points by the fourth year.
  - Inflation shock:
    - An increase in inflation by 1.6 percentage points causes NPL to increase by 0.3 percentage points the first year and a cumulative 1.6 percentage points by the fourth year.
  - GDP growth shock:
    - A 2.7 percentage points adverse shock to GDP growth causes NPL to increase by 0.4 percentage point the first year, reaching 1.7 by the fourth year.
  - Credit-to-GDP shock:
    - A one standard deviation decrease in the credit-to-GDP ratio causes NPL to increase by 0.3 percentage points the first year and by 1 percentage point by the fourth year.
- Responses to an NPL shock (one-standard-deviation NPL shock = 2.4 percentage points):
  - Change in house prices subsides by 1.2 percentage points in the first year and reaches 3.2 percentage points by the fourth year (decline).
  - Private sector credit-to-GDP declines by 4.5 percentage points the first year, and the decline reaches 28 percentage points by the fourth year.
  - GDP growth declines by 0.6 percentage points in the first year, reaches a trough decline of 1.4 percentage points in the third year, and subsides to a 1.1 percentage point decline in the fourth year.
  - NPL itself increases by 1 percentage point the first year and reaches 3.8 percentage points by the fourth year.
- Accumulated impulse response table excerpts (selected accumulated responses to generalized one standard deviation shocks; Periods 1 and 4 shown):
  - Response of GRW to shocks (Period 1; Period 4):
    - Period 1: "0.70" (DLEQUITYPR), "0.71" (DLHOUSEPR), "-0.18" (INFL), "0.72" (IPOL), "0.28" (NEER), "1.81" (GRW), "-0.59" (UNEMP), "-0.60" (NPL), "-0.06" (PSCGDP).
    - Period 4: "2.51" (DLEQUITYPR), "1.80" (DLHOUSEPR), "-0.82" (INFL), "0.89" (IPOL), "-0.32" (NEER), "3.20" (GRW), "-0.62" (UNEMP), "-1.10" (NPL), "0.15" (PSCGDP).
  - Response of NPL (Period 1; Period 4, accumulated):
    - Period 1: "-0.45" (DLEQUITYPR), "-0.29" (DLHOUSEPR), "0.32" (INFL), "0.03" (IPOL), "-0.28" (NEER), "-0.36" (GRW), "0.20" (UNEMP), "1.09" (NPL), "-0.29" (PSCGDP).
    - Period 4: "-2.25" (DLEQUITYPR), "-1.54" (DLHOUSEPR), "1.64" (INFL), "0.62" (IPOL), "-1.10" (NEER), "-1.69" (GRW), "0.61" (UNEMP), "3.84" (NPL), "-0.98" (PSCGDP).
  - Response of PSCGDP (Period 1; Period 4, accumulated):
    - Period 1: "3.97" (DLEQUITYPR), "0.68" (DLHOUSEPR), "-1.71" (INFL), "-0.95" (IPOL), "0.58" (NEER), "-0.54" (GRW), "0.71" (UNEMP), "-4.47" (NPL), "16.59" (PSCGDP).
    - Period 4: "24.86" (DLEQUITYPR), "14.53" (DLHOUSEPR), "-10.03" (INFL), "-9.01" (IPOL), "8.84" (NEER), "6.89" (GRW), "2.43" (UNEMP), "-28.30" (NPL), "67.92" (PSCGDP).
  - Note: Bold figures in source indicate statistical significance. Standard errors computed via Monte Carlo (1000 repetitions).

### Discussion, interpretation, and magnitudes
- Main interpretation:
  - Deterioration in macroeconomic environment—slower growth, higher unemployment, falling asset prices—is associated with rising NPL.
  - Asset quality is procyclical and reinforces the business cycle.
  - VAR evidence shows feedback loops: adverse shocks to asset prices, growth, and credit cause loan quality to worsen; NPL shocks feed back to reduce asset prices, credit, and growth.
- Magnitude relevance:
  - Adverse responses in GDP growth of "0.6 to 1.4 percentage points" are significant given advanced economies' median growth of "2.5 percent."
  - An NPL shock of 2.4 percentage points is comparable to increases observed in the Czech Republic, Italy, the United Kingdom, and the United States in either 2008 or 2009.
- Country examples:
  - U.K.: NPL increased from "1.5 percent in 2006" to "1.7 in 2007" and "2.9 in 2008."
  - U.S.: NPL increased from "1.8 percent in 2006" to "2.6 in 2007" and "4.9 in 2008."

### Policy implications, caveats, and suggestions for future research
- Policy implications:
  - NPL remain a permanent feature of banks’ balance sheets; policies should aim to avoid sharp increases that activate adverse macrofinancial feedback loops.
  - Preventing excessive risk-taking during upturns through adequate macroprudential regulations is the first best.
  - The regressions’ coefficients and IRFs can be used in macro-stress testing to assess likely changes in NPL and potential risks to financial stability.
- Caveats and limitations:
  - Stock measures of NPL are a rough measure of credit quality; declines in NPL may reflect loan removals rather than improvement.
  - Flow measures of newly nonperforming loans would be more informative.
  - Cross-border banking and the role of nonbank financial institutions imply heterogeneity in macrofinancial vulnerabilities across countries; focusing only on banks can miss important differences.
- Suggestions for future research:
  - Disaggregate NPL by loan type (mortgage, business, consumer) for single-country studies with long disaggregated series.
  - Explore nonlinear output effects and threshold effects of credit market frictions.
  - Analyze NPL net of provisioning to assess how provisioning affects banks’ resilience and lending capacity.

### Appendix — selected data, summary statistics, and cross-correlations (selected entries)
- Selected summary statistics for full sample (A):
  - DLEQUITYPR: Mean "1.21", Median "1.49", Standard deviation "9.52", Maximum "67.39", Minimum "-230.01".
  - DLHOUSEPR: Mean "5.15", Median "5.34", Standard deviation "4.77", Maximum "25.08", Minimum "-15.23".
  - INFL: Mean "2.23", Median "2.25", Standard deviation "2.17", Maximum "12.41", Minimum "-1.38".
  - IPOL: Mean "3.36", Median "3.33", Standard deviation "3.00", Maximum "18.00", Minimum "0.00".
  - NEER: Mean "106.85", Median "106.89", Standard deviation "106.42", Maximum "157.09", Minimum "47.95".
  - GRW: Mean "2.34", Median "2.33", Standard deviation "2.46", Maximum "10.73", Minimum "-8.02".
  - UNEMP: Mean "6.25", Median "6.37", Standard deviation "6.13", Maximum "18.01", Minimum "1.01".
  - NPL: Mean "2.54", Median "2.44", Standard deviation "1.73", Maximum "15.56", Minimum "0.11".
  - PSCGDP: Mean "125.73", Median "123.01", Standard deviation "106.63", Maximum "527.25", Minimum "31.69".
- Cross-correlations (selected, significance preserved):
  - DLEQUITYPR with DLHOUSEPR: "0.158***"
  - INFL with IPOL: "0.564***"
  - NPL correlations: with DLEQUITYPR "-0.148**"; with DLHOUSEPR "-0.289***"; with GRW "-0.159***"; with UNEMP "0.171***"; with NEER "-0.246***".
  - Note: "267 observations included. *, **, and *** denote significance at 10 percent, 5 percent, and 1 percent."

*Source: _wp11161 — IMF working paper content provided in the supplied PDF excerpt.*

### 1. Panel Unit Root Tests ...............................................................................................

### 1. Panel Unit Root Tests ...............................................................................................

### Introduction
- The deterioration in the quality of banks’ loan portfolio has been at the center of episodes of costly banking system distress and economic crises in both developing and advanced economies.
- The 2008 global financial crisis originated from a sharp increase in mortgage loan defaults in the United States and underscores linkages between financial and macroeconomic shocks.
- Objective: uncover macrofinancial vulnerabilities from the linkages between nonperforming loans (NPL) and macroeconomic performance in advanced economies.
- Sample: 26 advanced countries that spans the period from 1998 to 2009.
- Approaches used:
  - Single-equation panel regressions to address determinants of NPL.
  - Panel vector autoregressive (PVAR) approach and impulse response functions (IRFs) to examine interactions among variables and the extent to which falling asset prices and credit constraints may induce further financial stress and subdued economic activity.

### Literature context and framing
- The paper contributes to empirical literature on interactions between macroeconomic performance and credit market frictions, particularly rising NPL.
- Prior studies cited include: Keeton and Morris, 1985; Rinaldi and Sanchis-Arellano, 2006; Quagliariello, 2007; Jappelli, Pagano, and Marco, 2008; Espinosa and Prasad, 2010.
- Authors claim novelty: “To our knowledge, this paper is the first to look into the macrofinancial vulnerabilities associated with rising aggregate NPL from a sample covering most advanced economies and data related to the 2008 global financial crisis, which is foremost an advanced economies’ crisis.”
- Quotation included: “Often, the banking problems do not arise from the liability side, but from a protracted deterioration in asset quality, be it from a collapse in real estate prices or increased bankruptcies in the nonfinancial sector”. Kaminsky and Reinhart, 1999.

### Key empirical questions
- Determinants of NPL: whether macroeconomic, financial, or institutional factors drive NPL; association of sharp increases in NPL with asset price declines and disruptions to the supply of credit.
- Interactions between NPL and economic performance: whether rising NPL can backfire through falling asset prices and credit constraints to produce additional rounds of financial stress and subdued activity.

### Main findings (as summarized in the source)
- NPL play a central role in the linkages between credit market frictions and macroeconomic vulnerabilities.
- A sharp increase in NPL weakens macroeconomic performance and activates a vicious spiral that exacerbates macrofinancial vulnerabilities.
- The analysis attaches magnitudes to adverse responses in the feedback loop; these magnitudes are meaningful in the advanced economies’ context and can be used for macro stress testing or assessment of macrofinancial vulnerability more generally.

### Policy implications
- NPL remain a permanent feature of banks’ balance sheets.
- Policies and reforms should be geared to avoiding sharp increases in NPL that set into motion the adverse feedback loop between macroeconomic and financial shocks.

### Organization and supporting material (as described)
- Section II: literature review.
- Section III: presentation of models to be estimated and discussion of econometric results.
- Figures referenced in the paper include:
  - Figure 1. Advanced Economies’ Nonperforming Loans
  - Figure 2. Nonperforming Loans and Key Macroeconomic Variables
  - Figure 3. Impulse Response of NPL
  - Figure 4. Impulse Response of Credit-to-GDP
  - Figure 5. Impulse Response of GDP Growth
- Appendix contents referenced:
  - Appendix Tables A1–A5 (Variables’ Description and Data Sources; Summary Statistics; Cross-correlations Between Variables; Vector Autoregression, Lag Length Selection; Impulse Responses—Accumulated responses to Generalized One Standard Deviation)
  - References and additional appendix material noted.

*Source: _wp11161 - 1. Panel Unit Root Tests ...............................................................................................*

### Section IV concludes and discusses the policy implications of our findings.

### _wp11161 - Section IV concludes and discusses the policy implications of our findings.

### II. Literature — theoretical background and empirical regularities
- Theoretical frameworks:
  - Business cycle models with financial intermediation highlight countercyclicality of credit risk and business failures (Williamson, 1987).
  - Financial accelerator theory (Bernanke and Gertler (1989); Bernanke and Gilchrist (1999); Kiyotaki and Moore (1997)) is the prominent framework for macrofinancial linkages and modeling NPL.
- Empirical regularities supporting the financial accelerator:
  - Cyclicality: bank credit, NPL, and loan loss provisions are cyclical — low contemporaneous NPL and subdued provisioning in upturns; higher NPL and tighter credit in downturns.
  - Loosening of lending standards in upturns (affected by regulatory and supervisory framework) can sow seeds of later financial distress.
  - Determinants of NPL are both institutional/structural (financial regulation and supervision, incentive structures) and macroeconomic (GDP growth, unemployment, etc.).
- Findings from micro- and macro-level studies:
  - Within-country bank studies (first strand): management quality and macro fundamentals explain NPL; mortgages least responsive to macro conditions; positive relationship found between NPL and real lending rates; management inefficiency (higher operating expenses-to-operating income) positively associated with NPL.
  - NPL and crisis probability (second strand): large increases in NPL may mark onset of banking crises; high NPL contributes to macrofinancial vulnerabilities.
  - Aggregate NPL studies (third strand): higher debt-to-income linked to higher subsequent NPL; short-run house prices negatively related to NPL (housing wealth as buffer or collateral).
  - VAR-based studies: feedback loops between NPL and macroeconomy; evidence of adverse (sometimes short-lived) feedback from bank losses to economic activity.

### III. Empirical analysis — approach and models
- Two-pronged empirical approach:
  - Single-equation panel regressions to identify determinants of NPL.
  - Panel vector autoregression (VAR) to assess dynamic interactions and impulse responses among NPL and macrofinancial variables.
- Single-equation NPL model (panel):
  - NPL_it = Y_it́ β + ν_i + ε_it
  - Y_it includes endogenous and predetermined variables, lags of dependent variable, with ν_i capturing time-invariant country effects.
  - Lags of NPL included to capture persistence and omitted variables.
- Explanatory variables and expected signs:
  - Real GDP growth and employment: expected negative relationship with NPL.
  - Unemployment: expected positive relationship with NPL.
  - Equity or house price changes: expected negative relationship with NPL (boost wealth/collateral).
  - Interest rates: expected positive relationship with NPL (weaken debt servicing capacity, especially with variable rates).
  - Private sector credit-to-GDP (PSCGDP): contemporaneously expected negative correlation in upturns, but potentially positive with subsequent NPL if high indebtedness reflects lax lending standards.
  - Inflation: ambiguous (can reduce real debt burden or reduce real income if wages sticky; variable-rate loans may transmit higher rates).
  - Nominal effective exchange rate (NEER): indeterminate sign (can hurt exporters or help foreign-currency borrowers).
- Panel VAR specification:
  - y_it = B(L) y_it + ε_it where y_it is k x 1 vector including NPL and macrofinancial variables.
  - Impulse response functions (IRFs) derived from moving-average representation; orthogonalized shocks obtained via matrix A with A A' = Σε (or equivalently diagonalization) to produce contemporaneously uncorrelated u_t shocks for realistic system reactions.

### III.B Data — sample, definitions, and integration properties
- Sample coverage:
  - Annual data from 1998 to 2009 for a sample of 26 advanced economies.
  - Countries: Australia (AUS), Austria (AUT), Belgium (BEL), Canada (CAN), Czech Republic (CZE), Denmark (DEN), Finland (FIN), France (FRA), Germany (GER), Greece (GRE), Iceland (ICL), Israel (ISR), Italy (ITA), Japan (JPN), Korea (KOR), Luxembourg (LUX), the Netherlands (NET), New Zealand (NZL), Norway (NOR), Portugal (POR), Singapore (SGP), Spain (SPA), Sweden (SWE), Switzerland (SWI), the United Kingdom (UK), the United States (US).
- NPL definition and comparability issues:
  - NPL defined as the ratio of overdue loans to gross loans (used consistently throughout the analysis).
  - Cross-country differences in regulation, supervision, and accounting constrain comparability — e.g., some banks restructure or roll-over bad loans while others write them off quickly.
  - Peak NPL levels are generally higher in developing countries than in advanced economies.
- Other variables used (listed in Appendix): GRW (GDP growth), UNEMP (unemployment), DLHOUSEPR (change in house price index), DLEQUITYPR (change in equity price index), INFL (inflation), NEER (nominal effective exchange rate), IPOL (policy rate of interest), PSCGDP (credit to the private sector).
- Unit-root and stationarity assessment:
  - Fisher-ADF and Fisher-PP tests applied (assume individual unit root processes across countries).
  - Fisher-ADF suggests all variables stationary in level except PSCGDP.
  - Fisher-PP indicates somewhat higher stationarity thresholds; LNHOUSEPR null of unit root could not be rejected even after first differencing on Fisher-PP, but LNHOUSEPR considered first-difference stationary based on Fisher-ADF.
  - Different orders of integration imply need to check for cointegration when mixing variables of different integration orders.

- Table 1. Panel Unit Root Tests (assuming individual unit root processes)
  - Variables and test statistics (Fisher-ADF; Fisher-PP)
    - NPL Level 80.955***; 49.844
    - NPL 1st difference -2.398***; 72.662**
    - IPOL Level 107.931***; 91.916***
    - IPOL 1st difference -8.194***; 151.568***
    - LNEQUITYPR Level 98.626***; 55.185
    - LNEQUITYPR 1st difference -6.847***; 132.034***
    - LNHOUSEPR Level 89.091***; 57.197
    - LNHOUSEPR 1st difference -2.016**; 49.167
    - NEER Level 28.054; 15.671
    - NEER 1st difference 150.440***; 78.829***
    - PSCGDP Level 25.611; 28.326
    - PSCGDP 1st difference -6.170***; 123.911***
    - GRW Level 94.595***; 88.071***
    - GRW 1st difference 95.045***; 117.606***
    - INFL Level 152.803***; 179.480***
    - INFL 1st difference -11.645***; 251.853***
    - UNEMP Level 39.134; 51.613
    - UNEMP 1st difference 157.570***; 110.466***
  - Note: *, **, and *** denote significance at 10 percent, 5 percent, and 1 percent, respectively.

### III.C Stylized facts — NPL dynamics across the sample period
- Sample sub-periods and NPL behavior:
  - Pre-2003: relatively high or rising NPL, reflecting subdued macroeconomic performance (aftereffects of East-Asia crisis and dot-com bubble burst).
  - Precrisis 2003–07: strong macroeconomic performance and low or declining NPL.
  - 2008–09: slower growth or recessions associated with increases in NPL across all countries (magnitude varied).
- Observed macroeconomic ranges in the sample:
  - GDP growth: ranges from negative 8 percent in Finland in 2008 to 10.7 percent in Korea in 1999.
  - Unemployment: ranges from about one percent in Iceland in 2007 to 18 percent in Spain (year not specified in supplied text).

*Source: _wp11161 - Section IV concludes and discusses the policy implications of our findings.*

### 2009. The maximum NPL is observed in Iceland in 2009, while the minimum is observed in

### _wp11161 - 2009. The maximum NPL is observed in Iceland in 2009, while the minimum is observed in

### NPL patterns across advanced economies
- The maximum NPL is observed in Iceland in 2009, while the minimum is observed in Belgium in 1999.
- In most countries, NPL is notably much higher during 2008–09 than in the precrisis years.
- Country-specific NPL averages during 2000–07 exceed 2008 and 2009 levels only in a few countries.
- Cross-correlations confirm expected relationships: indicators of good macroeconomic performance and rising asset prices are negatively related with NPL.

### Estimation approach and model specifications
- Single-panel regressions:
  - Models estimated: OLS with lagged dependent variables (column 1); panel-corrected standard error (PCSE) models with unobserved country-specific effects (models (2) and (3)); alternative specifications excluding lagged dependent variables (model 4) or fixed effects (model 5); one-step Arellano-Bond GMM (model 6).
  - Fixed effects are deemed appropriate in models (1) thru (4); Fisher and chi-square tests reject the redundancy of country-specific fixed effects.
  - Arellano-Bond diagnostics: reject null of zero first-order serial correlation and do not reject absence of second-order serial correlation where required.
  - Note on estimator choice: the one-step GMM estimator is chosen because, in small samples, it generally tends to be less biased than the two-step estimator. Corrected least-square dummy variable methods are infeasible due to short time span and unbalanced sample.
- Panel VAR:
  - Variables: DLEQUITYPR, DLHOUSEPR, INFL, IPOL, NEER, GRW, UNEMP, NPL, PSCGDP (nine variables; four in annual percentage change and five in level).
  - Lag selection: one lag chosen based on Schwartz Bayesian information (lag length tests suggested one to three).
  - Cointegration: Johansen’s trace and maximum eigenvalue tests support cointegrating relationships, with the number of cointegrating vectors ranging from 3 to 5.
  - Generalized IRFs used so ordering of variables does not matter.

### Regression results (high-level)
- Key coefficient signs and statistical significance (as presented in Table 2):
  - NPL first lag: 0.875*** in OLS (model 1); 0.608*** in AB_GMM (model 6).  
  - NPL 2nd lag: -0.239* in model 1; 0.355*** in model 4; 0.168*** in model 6.
  - GRW (GDP growth): -0.179*** in model 1; -0.157*** in model 6.
  - GRW t-1 (lagged GDP growth): 0.168*** (model 2); -0.087* (model 4); -0.040* and -0.042* in other specifications.
  - DUNEMP (unemployment change): coefficients around 0.204*** to 0.245*** where included.
  - DLHOUSEPR (change house prices): negative and significant across specifications, e.g., -0.026* (model 1), -0.047*** (model 2), -0.027*** (model 3), -0.021*** (model 5), -0.065** (model 6).
  - DLEQUITYPR (change equity prices): negative and significant across specifications, e.g., -0.009*** to -0.014***.
  - IPOL (policy rate): positive where included, e.g., 0.071 (model 1, not significant) and 0.256** (model 5).
- Model fit and diagnostics:
  - Number of observations across reported models: 244, 267, 244, 264, 241, 241.
  - R-squared examples: 0.863 (model 1), 0.493 (model 2), 0.543 (model 3), 0.357 (model 4), 0.403 (model 5).
  - AR(1) p-value reported as 0.045** and AR(2) p-value reported as 0.175 in Arellano-Bond diagnostics where applicable.

### Panel VAR impulse responses and macrofinancial linkages
- General finding: NPL plays a central role in macrofinancial interactions; over a four-year forecast period NPL both affects and responds to every single variable in the VAR with statistically significant effects and responses.
- Selected quantitative IRF results (responses to one-standard-deviation shocks and interpretation):
  - House prices shock:
    - A one standard deviation increase in the change in house prices, amounting to 6.1 percentage points, causes NPL to increase by 0.3 percentage points the first year and by 1.5 percentage points by the fourth year.
  - Inflation shock:
    - An increase in inflation by 1.6 percentage points causes NPL to increase by 0.3 percentage points the first year and a cumulative 1.6 percentage points by the fourth year.
  - GDP growth shock:
    - A 2.7 percentage points adverse shock to GDP growth causes NPL to increase by 0.4 percentage point the first year, reaching 1.7 by the fourth year.
  - Credit-to-GDP shock:
    - A one standard deviation decrease in the credit-to-GDP ratio causes NPL to increase by 0.3 percentage points the first year and by 1 percentage point by the fourth year.
- Responses to an NPL shock (one-standard-deviation NPL shock = 2.4 percentage points):
  - Change in house prices subsides by 1.2 percentage points in the first year and continues to decline, reaching 3.2 percentage points by the fourth year.
  - Private sector credit-to-GDP declines by 4.5 percentage points the first year, and the decline reaches 28 percentage points by the fourth year.
  - GDP growth declines by 0.6 percentage points in the first year, reaches a trough decline of 1.4 percentage points in the third year, and subsides to a 1.1 percentage point decline in the fourth year.
  - NPL itself increases by 1 percentage point the first year and reaches 3.8 percentage points by the fourth year.
- Magnitudes: the uncovered NPL and GDP growth IRFs are described as very meaningful for advanced economies; an NPL shock of 2.4 percentage points is comparable to increases observed in the Czech Republic, Italy, the United Kingdom, and the United States in either 2008 or 2009.

### Discussion and interpretation
- Panel regression findings confirm stylized facts: deterioration in macroeconomic environment—slower growth, higher unemployment, falling asset prices—is associated with rising NPL.
- Asset quality is procyclical and reinforces the business cycle.
- The VAR results emphasize feedback loops: adverse shocks to asset prices, macroeconomic performance, and credit to the private sector cause loan quality to worsen; conversely, NPL shocks feed back to reduce asset prices, credit, and growth.
- Relative magnitudes highlight the potential for sizable macroeconomic amplification from financial distress in advanced economies over a multi-year horizon.

*Source: IMF working paper content provided in the supplied PDF excerpt.*

### 1.7 percentage points may not mean much as

### _wp11161 - 1.7 percentage points may not mean much as

### Main findings on NPL and macroeconomic performance
- Analysis covers a sample of 26 advanced countries and focuses on aggregate NPL.
- A sharp increase in NPL:
  - "triggers long-lived tailwinds that cripple macroeconomic performance from several fronts."
  - "feeds on itself leading to an almost linear incremental response that continues into the fourth year after the initial shock."
- Impulse response functions (IRFs) results:
  - NPL is the only variable with both a statistically significant response to — and predictive power on — every single variable over a 4-year forecast period.
  - Adverse responses in GDP growth of "0.6 to 1.4 percentage points" are significant given advanced economies' median growth of "2.5 percent."
- Country examples where moderate NPL increases coincided with systemic crises:
  - U.K.: NPL increased from "1.5 percent in 2006" to "1.7 in 2007" and "2.9 in 2008."
  - U.S.: NPL increased from "1.8 percent in 2006" to "2.6 in 2007" and "4.9 in 2008."
- Confluence of adverse responses in GDP growth and unemployment can produce a downward spiral where banking distress and economic deterioration reinforce each other.

### Policy implications and practical applicability
- Econometric relations uncovered can be used for forecasting and stress testing by supervisory authorities and, to some extent, banks.
  - "The regressions’ coefficients and the impulse responses along with assumptions on the evolution of macroeconomic variables and asset prices can be used in a macro-stress testing exercise to assess the likely change in NPL and whether such a change could pose risk of financial instability based on some indicative threshold."
- Broad policy implication:
  - While NPL remain a permanent feature of banks’ balance sheets, policies and reforms should be geared to avoiding sharp increases that set into motion adverse feedback loops between macroeconomic and financial shocks.
  - "Preventing excessive risk-taking during upturns through adequate macroprudential regulations is the first best."

### Caveats and limitations
- Outstanding level of NPL is a rough measure of credit quality; decreases can reflect removal of unrecoverable loans from banks’ balance sheets rather than improvement.
- Flow of newly nonperforming loans would be more informative than stock measures.
- Cross-border bank expansion implies NPL have cross-border implications: losses in one country can prompt tighter credit in another via consolidated balance sheets.
- Macrofinancial vulnerabilities associated with NPL differ across countries because of differing roles of nonbank financial institutions; focusing only on banks’ balance sheets can miss these differences.

### Suggestions for future research
- Disaggregate NPL by loan type (mortgage, business, consumer) for single-country studies with sufficiently long disaggregated NPL time series.
- Explore potential nonlinear output effects and threshold effects of credit market frictions.
- Analyze linkages between macroeconomic indicators and NPL net of provisioning for bad loans to assess how provisioning affects banks’ ability to withstand shocks and continue lending.

### Appendix — Data, summary statistics, and key empirical outputs
- Variable definitions (examples):
  - LNHOUSEPR: Logarithm of the stock price index (Stock price indices from IMF, International Financial Statistics).
  - LNEQUITYPR: Logarithm of the house price index (House price indices from Haver and Bank for International Settlements).
  - DLEQUITYPR: First-difference of LNEQUITYPR in percent (Calculated as noted in the description).
  - DELHOUSEPR: First-difference of LNHOUSEPR in percent (Calculated as noted in the description).
  - GRW: Annual change in real GDP, in percent (IMF, World Economic Outlook database).
  - INFL: Annual change in the consumer price index, in percent (IMF, World Economic Outlook database).
  - IPOL: Policy rate of interest, in percent (Datastream).
  - NEER: Nominal effective exchange rate index, 2000=100 (Global Financial database (GDS)); "An increase denotes an appreciation."
  - NPL: Ratio of non-performing loans to gross loans, in percent (Bankscope, US Federal Reserve Database (for the United States), IMF, Global Financial Stability report).
  - PSCGDP: Credit to the private sector in percent of GDP (IMF, International Financial Statistics).
  - UNEMP: Rate of unemployment in percent (IMF, World Economic Outlook database).
- Summary statistics (selected means, medians, standard deviations, maxima, minima are reported in the source tables; note: columns A relate to the full sample while columns B exclude Iceland):
  - Example reported statistics for full sample (A):
    - DLEQUITYPR: Mean "1.21", Median "1.49", Standard deviation "9.52", Maximum "67.39", Minimum "-230.01".
    - DLHOUSEPR: Mean "5.15", Median "5.34", Standard deviation "4.77", Maximum "25.08", Minimum "-15.23".
    - INFL: Mean "2.23", Median "2.25", Standard deviation "2.17", Maximum "12.41", Minimum "-1.38".
    - IPOL: Mean "3.36", Median "3.33", Standard deviation "3.00", Maximum "18.00", Minimum "0.00".
    - NEER: Mean "106.85", Median "106.89", Standard deviation "106.42", Maximum "157.09", Minimum "47.95".
    - GRW: Mean "2.34", Median "2.33", Standard deviation "2.46", Maximum "10.73", Minimum "-8.02".
    - UNEMP: Mean "6.25", Median "6.37", Standard deviation "6.13", Maximum "18.01", Minimum "1.01".
    - NPL: Mean "2.54", Median "2.44", Standard deviation "1.73", Maximum "15.56", Minimum "0.11".
    - PSCGDP: Mean "125.73", Median "123.01", Standard deviation "106.63", Maximum "527.25", Minimum "31.69".
- Cross-correlations (selected):
  - DLEQUITYPR with DLHOUSEPR: "0.158***"
  - INFL with IPOL: "0.564***"
  - NPL correlations: with DLEQUITYPR "-0.148**"; with DLHOUSEPR "-0.289***"; with GRW "-0.159***"; with UNEMP "0.171***"; with NEER "-0.246***".
  - Note: "267 observations included. *, **, and *** denote significance at 10 percent, 5 percent, and 1 percent."
- VAR lag selection (reported criteria and selected orders):
  - Log likelihood and information criteria reported for lags 0 to 3; asterisk indicates lag order selected by the criterion.
- Impulse responses — accumulated responses to generalized one standard deviation shocks (selected entries, accumulated over periods 1 to 4):
  - Response of GRW (Period 1 to 4, response to NPL shock column values shown in table):
    - Period 1: "0.70" (DLEQUITYPR), "0.71" (DLHOUSEPR), "-0.18" (INFL), "0.72" (IPOL), "0.28" (NEER), "1.81" (GRW), "-0.59" (UNEMP), "-0.60" (NPL), "-0.06" (PSCGDP).
    - Period 4: "2.51" (DLEQUITYPR), "1.80" (DLHOUSEPR), "-0.82" (INFL), "0.89" (IPOL), "-0.32" (NEER), "3.20" (GRW), "-0.62" (UNEMP), "-1.10" (NPL), "0.15" (PSCGDP).
  - Response of NPL (Period 1 to 4, accumulated):
    - Period 1: "-0.45" (DLEQUITYPR), "-0.29" (DLHOUSEPR), "0.32" (INFL), "0.03" (IPOL), "-0.28" (NEER), "-0.36" (GRW), "0.20" (UNEMP), "1.09" (NPL), "-0.29" (PSCGDP).
    - Period 4: "-2.25" (DLEQUITYPR), "-1.54" (DLHOUSEPR), "1.64" (INFL), "0.62" (IPOL), "-1.10" (NEER), "-1.69" (GRW), "0.61" (UNEMP), "3.84" (NPL), "-0.98" (PSCGDP).
  - Response of PSCGDP (Period 1 to 4, accumulated):
    - Period 1: "3.97" (DLEQUITYPR), "0.68" (DLHOUSEPR), "-1.71" (INFL), "-0.95" (IPOL), "0.58" (NEER), "-0.54" (GRW), "0.71" (UNEMP), "-4.47" (NPL), "16.59" (PSCGDP).
    - Period 4: "24.86" (DLEQUITYPR), "14.53" (DLHOUSEPR), "-10.03" (INFL), "-9.01" (IPOL), "8.84" (NEER), "6.89" (GRW), "2.43" (UNEMP), "-28.30" (NPL), "67.92" (PSCGDP).
- Impulse response tables note: "Bold figures suggest significantly different from zero at the standard significance levels: 1 percent, 5 percent, and 10 percent." Standard errors computed via "Monte Carlo (1000 repetitions)."

*Source: IMF working paper content provided in the supplied PDF excerpt.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11161.pdf_