Journal Issue
Share
Article

Portugal: Selected Issues

Author(s):
International Monetary Fund. European Dept.
Published Date:
September 2018
Share
  • ShareShare
Show Summary Details

Financial Conditions and Growth at Risk in Portugal1

A. Introduction

1. The macro-finance literature and recent experience provide compelling evidence that financial imbalances grow in good times, creating downside risks to economic growth. Indeed, when confidence in economic prospects is high and financial conditions benign, households, firms and financial intermediaries tend to take excessive risk, leading to an increase in leverage, maturity mismatches and other balance sheet weaknesses.2 In presence of elevated imbalances, a negative shock can force borrowers into default or fire sales, putting pressure on lenders’ profits and collateral values, and disrupting financial intermediation. This can result in higher odds of severe and prolonged economic contraction.3 Thus, financial conditions that cause financial vulnerabilities to build up in good times could convey valuable information about downside risks to future GDP growth.

2. A recent strand of empirical research has undertaken to investigate the predictive power of financial conditions for downside risks to economic growth. Adrian, Boyarchenko and Giannone (2016) showed that financial conditions improve the accuracy of predicting economic contractions in the case of the United States. The Chapter 3 of the October 2017 IMF Global Financial Stability Report (GFSR) developed a conceptual framework that links financial sector risks to GDP growth risks through macro-financial linkages. Applying this “Growth-at-Risk” approach (GaR) to a set of major advanced and emerging market economies, the authors found that financial conditions are leading indicators of risks to GDP growth in both advanced and emerging market economies. The estimated impact of financial conditions on growth is stronger at the left tail of the distribution of future GDP growth than on the median and upper percentiles growth rate for advanced economies, suggesting the existence of asymmetries. While rising price of risk4 is a powerful indicator of near-term risks of large contractions, growing leverage and higher credit growth signal downside risks to GDP growth over the medium term.

3. This paper applies the GaR methodology developed in the October 2017 GFSR to Portugal. After three years of lackluster economic recovery, GDP growth gained strength in 2017, and staff’s baseline scenario assumes that economic prospects will remain positive over the medium term. On the other hand, this strong recovery is taking place in an environment of low interest rates and volatility, and compressed risk premia, which may incentivize borrowers and lenders to take excessive risks, aggravating the still-elevated leverage and posing risks to financial stability in case of a negative shock. This paper attempts to provide an estimate of the downside risks to future GDP growth, based on the current level of GDP growth and current financial conditions.

4. The GaR methodology involves three key steps. First, financial conditions indicators are partitioned into a predetermined number of subgroups using a data reduction technique. Second, a model of future output growth is estimated as function of current economic conditions and the partitioned financial conditions indicators using quantile regressions. Finally, the conditional quantile function (or inverse cumulative distribution function) is transformed into a probability density function by fitting a skewed t distribution. This probability density function is then exploited to quantify downside tail risks to future GDP growth.

5. The paper is organized as follows. The next section summarizes the prevailing macro-financial environment in Portugal. The subsequent section describes the methodology for building Portugal’s financial conditions measures and discusses the results. The section after provides estimates for downside risk to GDP growth. The last section concludes with policy implications.

B. Macro-Financial Environment in Portugal

6. Portugal’s economic activity picked up in 2017, and the gap between the economic and financial cycle is narrowing. After three years of moderate growth, real GDP accelerated to 2.7 percent in 2017, the highest since 2000, and the output gap is closing. Although the credit cycle continues to trail the cyclical recovery, the gap is narrowing driven by a strong increase in new lending to households for consumption and house purchases, as well as to nonfinancial corporates with good risk profiles. The robust growth in new lending has supported aggregate demand and thus the economic recovery, but may have also slowed the deleveraging process.

7. Despite recent progress, financial imbalances remain elevated. While the nonfinancial private sector’s debt-to-GDP ratio fell 47 percentage points since 2012, it is still among the highest in the euro area, with total household and corporate debts standing at 74 and 138 percent of GDP at end-2017 (unconsolidated basis), respectively. For its part, public debt remains elevated at 126 percent of GDP. In the banking sector, the still high stock of non-performing loans (13.3 percent of total loans at end-2017) and modest profitability prospects remain concerns. In the housing markets, prices continue to increase (about 20 percent in real terms since 2013 compared to seven percent in the euro area), and there are concerns that the current easy financial conditions could boost mortgages and further drive up prices. In this environment, a sudden repricing of risks could affect financial stability and thus economic growth.

Figure 1.Macro-Financial Developments

Sources: Eurostat, Banco de Portugal, NBB, Haver Analytics, and IMF staff calculations.

C. Measuring Financial Conditions

8. The literature offers two approaches to constructing measures of financial conditions from a wide range of financial indicators. The “univariate” approach summarizes into a single index the salient features of a large set of financial indicators. This approach avoids overfitting issues due to the inclusion in a model of a large number of predictors, but comes at the price of suppressing separate signals provided by different types of financial indicators, which could be relevant at different horizons (for instance asset prices and leverage).5 The “partitioning” approach, developed in the October 2017 GFSR, seeks to differentiate information about a variety of horizons conveyed by financial variables, while avoiding overfitting problems. Financial indicators are organized into a few groups based on their economic similarity, and then their information is aggregated at group level, which facilitates the interpretation of the results.

9. Following the October 2017 GFSR, a large set of financial variables is partitioned into three groups (see Table 1 below). We use the Linear Discriminant Analysis (LDA), a data-reduction technique, to determine the weights to aggregate the financial variables in each group. LDA is like principal component analysis (PCA) as it seeks for linear combinations of variables which best explain the data, but it diverges from PCA as it attempts to maximize the separability between the classes of data or categories. Here, the categorical variable (Y) is a dummy variable, taking on the value of 1 when future GDP growth at the one-year horizon is below the 20th percentile of historical outcomes, and 0 otherwise.6 As such, the weights (loadings) to aggregate financial indicators in each group are determined in a way that maximizes their contribution to discriminating between periods of low GDP growth (below the 20th percentile) and the rest of the time. This approach is suitable from this paper’s perspective as it allows to link financial indicators and GDP growth outcomes.

Table 1.Portugal Partition Groups
Price of RiskCredit AggregatesExternal Conditions
Interbank rateCredit to GDPVIX
Real long-term rateCredit growth toVSTOXX
Sovereign spreadHouseholdExchange rate (US
Term spreads(quarterly)dollar to Euro)
P/E of PSE-20Credit growth toOil prices
House price returnsNon-FinancialReal GDP growth
One-year CDSCorporate (quarterly)(euro area, excluding Portugal)
Source: IMF staff.
Source: IMF staff.

10. Overall, the dynamics of the price of risk, credit aggregates and external conditions dynamics capture well key global and Portuguese events, and predict well episodes of severe economic weakness (GDP growth below the 20th percentile, see figure 2).7 More specifically:

  • Price of risk. The dynamics of the price of risk closely tracks the impact of global and euro area financial events, including the 2002 financial turbulence following the discoveries of accounting irregularities at large US corporations (Enron, WorldCom), the 2008 global financial crisis, the 2012 euro area crisis, the 2013 taper tantum episode, and the 2015 global bond sell off. As shown in figure 2, these events triggered spikes in the domestic price of risk and predicted well the subsequent economic downturns. Since mid-2013, the price of risk has been low along with the ECB’s accommodative monetary policy and with improvements in sovereign spreads as fiscal consolidation took hold.
  • Credit aggregates. Its evolution reflects well Portugal’s monetary experience, notably the launch of the euro zone in 1999, the introduction of euro currency in 2002, and the deleveraging since 2012.
  • External conditions. Its evolution captures correctly the episode of stress and relaxation in global financial markets described above, which triggered spikes in external financial conditions.

Figure 2.Partitioned Financial Indicators 1999–2018

Source: IMF Staff calculations.

Note: The grey‐shaded areas correspond to periods where GDP growth was below the 20th percentile of historical outcomes.

D. Estimating Risks to GDP Growth

11. The empirical strategy for estimating risks to GDP growth follows the two-step procedure of Adrian et al (2016). In the first step, quantile regressions of Koenker and Bassett (1978) are used to forecast the conditional quantiles of future GDP growth at several time horizons. In the second step, the quantile regression (inverse cumulative distribution function) is transformed into a probability density function using a parametric fit (a skewed t distribution). Once the curve is fitted, the growth-at-risk is estimated for the selected horizons.

Quantile Regressions

12. The recourse to quantile regressions is justified by the possibility of non-linear output responses to financial conditions documented in the literature.8 In addition, quantile regressions are robust to non-normal errors and outliers, and invariant to monotonic transformation. 9

13. Quantile regressions are estimated on the following specification:

where Yt+h,qrepresents the quantiles (q) of the future distribution of GDP growth (y) h quarters ahead; PR, CA, and EC are the three predictors corresponding to the price of risk, credit aggregates and external conditions derived from the LDA, respectively; Ct,qh and ϵt,qh are the intercept and error term, respectively.

14. Overall, the results from the quantile regressions support the view of a nonlinear relationship between financial conditions and future GDP growth, with the various financial indices being more or less informative depending on the time horizon (see figure 3).

  • Price of risk is a powerful forecaster for downside risks to GDP growth at horizons of one to eight quarters. The estimated impact of tighter price of risk appears to be mostly stronger at the tails of the GDP growth distribution than around the median, supporting the view of asymmetries in the output response. The price of risk becomes uninformative over longer horizons.
  • Credit aggregates (leverage and credit growth) signal downside risks to GDP growth at horizons of four to 12 quarters (one to three years). The significant predictor power of credit aggregates at short-term horizon (one year) is unsurprising given that the Portuguese economy is already-highly leveraged. That said, rising credit aggregates signal higher downside risks over the medium term (two to three years) than in the short term (one year).
  • External conditions provide some valuable information about downside risks to growth both over short and medium term. The strongest effect of tighter external conditions is mainly at the three-year horizon and at the right tail of the GDP growth distribution.

Figure 3.Quantile Regressions Coefficients

Source: IMF staff calculations.

Note: The vertical lines in the blue bars denote confidence intervals at 10 percent and, when they cross the x‐axis, this signals the absence of statistical significance of the predictor.

Estimation of Downside Tail Risk to GDP Growth

15. To estimate the tail risks around the baseline, a skewed t distribution is fitted on the empirical conditional quantile function for each specific time horizon. Further details are discussed in the Adrian et al (2016). The distribution is calibrated so that the mode (which is the most likely outcome taken by the distribution) is equal to the staff’s baseline scenario. Using the skewed t fitted curve, a probability density function can be derived for future GDP growth at each time horizon.

16. Overall, the GaR model suggests moderate risks around the baseline for Portugal’s GDP growth. Based on the financial conditions at 2018:Q1, a severely adverse outcome (given by the 5 percent left tail) is for GDP growth to fall below 1.3 percent one-year ahead, and below 0.9 percent in the two-to-three-year horizons. This is a rather benign risk outlook, given the still-elevated leverage, and reflects the dominating effect of the low price of risk, itself a reflection of supportive monetary policies at the euro area level and tight fiscal policies domestically.

Figure 4.Probability Densities of GDP Growth Four and Eight Quarters Ahead

Source: IMF staff calculations

E. Policy Implications and Conclusion

17. The analysis highlights the importance of the price of risk, leverage and credit growth as leading indicators of risks to GDP growth. The price of risk appears to provide the most powerful signal in the short term, while credit aggregates are the most significant predictor in the medium term. This finding is consistent with the volatility paradox (Brunnermeier and Sannikov, 2014), and is line with other empirical studies (GFSR, 2017).

18. The GaR model suggests contained downside risks to Portugal’s growth projections at the current juncture based on financial conditions data, but credit growth should continue to be monitored given still high leverage. The moderate risk to growth identified by the GaR model reflects the impact of low credit spreads and volatility in the financial markets, in their turn reflecting the prevailing policy mix. Still, a repricing of risks and other shocks could be magnified by the still-high leverage, and lead to less favorable growth outcomes. Such context puts a premium on the timely detection of signs of excessive risk taking and the deployment of macroprudential policies.

Annex I. Data Sources
VariablesDescriptionSource
Interbank rateChange in three-month EuriborBloomberg Finance L.P.; IMF staff
Change in Long-term Real Interest RatePercentage point change in the 10-year government bond yield, adjusted for inflationBloomberg Finance L.P.; IMF staff
Equity ReturnsLog difference of the PSE-20 indexBloomberg Finance L.P.
Sovereign SpreadsYield on 10-year government bonds minus the benchmark country’s yield on 10-year government bondsBloomberg Finance L.P.; IMF staff
House Price ReturnsLog difference of the house price indexBank for International Settlements; IMF staff
CDSChange in expected default probability of sovereign bond at one yearBloomberg Finance L.P.; IMF staff
Household debt to GDPAt market valueBank for International Settlements
Non-financial corporate debt to GDPAt market valueBank for International Settlements
Government debt to GDPGeneral government, consolidated dataEurostat; IMF staff
Credit growthChange in credit provided by domestic banks, all other sectors of the economy, and nonresidents; quarterly growthBank for International Settlements; IMF staff
VIXChicago Board Options Exchange Market Volatility IndexBloomberg Finance L.P.
VSTOXXEURO STOXX 50 Options PricesBloomberg Finance L.P.
Exchange rateChange in US dollar per EuroBloomberg Finance L.P.
Oil pricesChange in Brent pricesBloomberg Finance L.P.
Real GDP growthCompound annual growth rateHaver Analytics; IMF staff
Source: IMF staff.
Source: IMF staff.
References

    Adrian, Tobias,NinaBoyarchenko, andDomenicoGiannone,2016, “Vulnerable Growth,” Federal Reserve Bank of New York Staff Report 794.

    Azzalini, Adelchi, andAntonellaCapitanio,2003, “Distributions Generated by Perturbations of Symmetry with Emphasis on a Multivariate Skew t Distribution,” Journal of the Royal Statistical Society: Series B65: 367–89.

    • Search Google Scholar
    • Export Citation

    Claessens, Stijn,AyhanKose, andMarcoTerrones,2011a. “Financial Cycles: What? When? How?” CEPR Discussion Paper 8379, Centre for Economic Policy Research, London.

    • Search Google Scholar
    • Export Citation

    Claessens, Stijn,AyhanKose, andMarcoTerrones,2011b, “How Do Business and Financial Cycles Interact,” CEPR Discussion Paper 8396, Centre for Economic Policy Research, London.

    • Search Google Scholar
    • Export Citation

    Duda,RichardO.,Peter E.Hart, andDavid G.Stork,2001, Pattern Classification, New York: Wiley-Interscience.

    European Central Bank, 2018, “A New Financial Stability Risk Index (FSRI) to Predict Near Term Risks of Recessions,” Financial Stability Review, May 2018.

    • Search Google Scholar
    • Export Citation

    Giot, Pierre, andSébastienLaurent,2003. “Value-at-Risk for Long and Short Trading Positions.” Journal of Applied Econometrics18 (6): 641–64.

    • Search Google Scholar
    • Export Citation

    International Monetary Fund (IMF), 2017. “Financial Conditions and Growth at Risk.” Global Financial Stability Report, Chapter 3, October 2017, Washington, D.C.

    • Search Google Scholar
    • Export Citation

    Izenman,AlanJ.,2013, Modern Multivariate Statistical Techniques: Regression, Classification, and Manifold Learning, New York: Springer-Verlag.

    • Search Google Scholar
    • Export Citation

    Koenker, Roger,2005, Quantile Regression, New York: Cambridge University Press.

    Koop, Gary, andDimitrisKorobilis,2014, “A New Index of Financial Conditions,” European Economic Review71 (C): 101–16.

    LafarguetteR.,2018, “Using Growth-at-Risk for Macrofinancial Analysis and Forecasting Application on Peru,” Peru, FSAP Technical Note.

    • Search Google Scholar
    • Export Citation

    Lambert, Philippe, andSébastienLaurent,2002, “Modelling Skewness Dynamics in Series of Financial Data using Skewed Location-Scale Distributions,” Discussion Paper, Institut de Statistique, Université Catholique de Louvain, Belgium.

    • Search Google Scholar
    • Export Citation
1

Prepared by Mesmin Koulet-Vickot.

2

See Brunnermeier and Sannikov (2014) for a discussion on the volatility paradox, in which benign financing conditions (low volatility) hides risks that become apparent when financial conditions tighten.

4

The price of risk includes variables such as term spreads, credit spreads, equity, and house prices.

5

See the 2017 October GFSR (Chapter 3).

6

Details on the LDA approach are discussed in the annex of the October 2017 GFSR. See also Izenman 2013 for a thorough description of the LDA technique.

7

The data cover 1999:Q1 to 2018:Q1, on a quarterly frequency. The choice of financial variables has been influenced by data availability (see Annex 1).

8

Standards linear regressions consider the impact of the regressors on the conditional mean of the dependent variable, while quantile regressions investigate the impact of the regressors on various points (quantiles) of the dependent variable’s conditional distribution.

9

For an introduction to quantile regression, see Koenker, (2005).

Other Resources Citing This Publication