Systemic Risks and the Macroeconomy
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: gdenicolo@imf.org; lucchett@unive.it;

This paper presents a modeling framework that delivers joint forecasts of indicators of systemic real risk and systemic financial risk, as well as stress-tests of these indicators as impulse responses to structural shocks identified by standard macroeconomic and banking theory. This framework is implemented using large sets of quarterly time series of indicators of financial and real activity for the G-7 economies for the 1980Q1-2009Q3 period. We obtain two main results. First, there is evidence of out-of sample forecasting power for tail risk realizations of real activity for several countries, suggesting the usefulness of the model as a risk monitoring tool. Second, in all countries aggregate demand shocks are the main drivers of the real cycle, and bank credit demand shocks are the main drivers of the bank lending cycle. These results challenge the common wisdom that constraints in the aggregate supply of credit have been a key driver of the sharp downturn in real activity experienced by the G-7 economies in 2008Q4- 2009Q1.

Abstract

This paper presents a modeling framework that delivers joint forecasts of indicators of systemic real risk and systemic financial risk, as well as stress-tests of these indicators as impulse responses to structural shocks identified by standard macroeconomic and banking theory. This framework is implemented using large sets of quarterly time series of indicators of financial and real activity for the G-7 economies for the 1980Q1-2009Q3 period. We obtain two main results. First, there is evidence of out-of sample forecasting power for tail risk realizations of real activity for several countries, suggesting the usefulness of the model as a risk monitoring tool. Second, in all countries aggregate demand shocks are the main drivers of the real cycle, and bank credit demand shocks are the main drivers of the bank lending cycle. These results challenge the common wisdom that constraints in the aggregate supply of credit have been a key driver of the sharp downturn in real activity experienced by the G-7 economies in 2008Q4- 2009Q1.

I. Introduction

The recent financial crisis has underscored the need for a deeper understanding of the key drivers of systemic financial risk and its two-way relationship with real activity. We believe that to accomplish these goals, at least two requirements need to be met. First, measures of systemic risk need to be associated with the potential for undesirable welfare consequences, such as extreme adverse real effects. Second, the interplay between real and financial activity needs to be assessed through the implications of some theoretical model, and correspondingly quantified. Importantly, detecting macro-financial linkages through a consistent and tractable framework may make it feasible to design risk monitoring tools implementable in real time. Contributing to accomplishing these goals is the main objective of this paper.

We design a modeling framework that aims at tracking and quantifying the impact and transmission of structurally identifiable shocks within/between the macroeconomy, financial markets and intermediaries, as well as their “tail” realizations. In terms of Figure A below, the proposed framework aims at identifying which sectors of the economy are most affected by a shock at impact, to gauge size and persistence of shocks’ propagation within and between sectors, and forecast their systemic real and financial outcomes.

Figure A
Figure A

Financial exposures (stocks and flows) between sectors

Citation: IMF Working Papers 2010, 029; 10.5089/9781451962567.001.A001

Ideally, a computable general equilibrium model specified at a suitable level of dis-aggregation would allow us to identify the sources of shocks as well as the linkages through which they are propagated. In practice, formulating and implementing such a model is a formidable theoretical and computational task. At present, an increasing number of research resources are devoted to develop macroeconomic models with meaningful interaction between financial and real sectors. However, work in this direction is still in its infancy, since work-horse Dynamic Stochastic General Equilibrium (DSGE) models do not yet embed essential financial structure or sectors, being their modeling of financial markets and institutions highly stylized.2

As a result, the available modeling technologies are still relatively underdeveloped. Some models analyzing the impact of macroeconomic shocks on segments of the financial sector have been developed recently in some central banks and international organizations. Yet, the feedback effects of financial vulnerabilities on the macroeconomy has been usually left unmodeled, since the output of these models is used mainly for financial supervisory purposes.3

Our modeling framework delivers joint forecasts of indicators of systemic real risk and systemic financial risk, as well as stress-tests of these indicators as impulse responses to structurally identifiable shocks. This framework is novel in two respects. First, it uses a dynamic factor model with structural identification based on theory. This permits to extract information on common sources of shocks contained in a large set of time series, and to characterize their economic content. Second, it integrates the dynamic factor model with quantile regressions techniques, which allow us to estimate and forecast the size of tail realizations of systemic risks.

We make a distinction between systemic real risk and systemic financial risk, based on the notion that real effects is what concerns policymakers most since they are likely to entail welfare consequences. Our systemic real risk indicator is GDP-at-Risk (GDPaR), defined as the worst predicted realization of quarterly growth in real GDP at 5 percent probability over a pre-determined forecasting horizon. Our indicator of systemic financial risk (FSaR) is defined as the worst predicted realization of a system-wide financial risk indicator at 5 percent probability over a pre-determined forecasting horizon.

The underlying joint dynamics of GDP growth and the system-wide financial risk indicator is modeled through a factor-augmented VAR (FAVAR) model, following variants of the methodology detailed in Stock and Watson (2002, 2005). Estimates of GDPaR and FSaR indicators are obtained through quantile regressions.

Forecasts of GDPaR and FSaR indicators are obtained by inputting the predicted values of factors obtained from the companion factor-augmented VAR into the relevant quantile regressions. Identification of structural shocks is accomplished with an expanded version of the sign restriction methodology introduced by Canova and De Nicolò (2002), where shocks are identified based on standard macroeconomic and banking theory. Stress-tests of both systemic risk measures are obtained inputting impulse responses to shocks identified in the FAVAR model into the relevant quantile regressions.

We implement this framework using a large set of quarterly time series of financial and real activity for the G-7 economies during the 1980Q1-2009Q3 period. We obtain two main results. First, we find evidence of out-of sample forecasting power of the model for tail risk realizations of real activity for several countries. This suggests the usefulness of the model as a risk monitoring tool. Second, in all countries we identify aggregate demand shocks as the main drivers of the real cycle, and bank credit demand shocks are the main drivers of the bank lending cycle. This result is consistent with the hypothesis that shocks to the real economy are the main drivers of both real and financial risks. Importantly, this finding challenges the common wisdom that constraints in the aggregate supply of credit have been a key driver of the sharp downturn in real activity experienced by the G-7 economies in 2008Q4-2009Q1.

The remainder of the paper is composed of four sections. Section II defines systemic risks and describes indicators consistent with these definitions. Section III outlines the model setup, estimation and forecasting, and the procedure used to identify structural shocks. Section IV describes the implementation of the modeling framework on data for the G-7 countries and the relevant results. Section V concludes.

II. Systemic risks

A. Definitions

We adopt the following definitions:

Systemic financial risk is the risk that a shock will trigger a loss of economic value or confidence in, and attendant increases in uncertainty about, a substantial portion of the financial system.

Systemic real risk is the risk that a shock will trigger a significant decline in real activity.

As in Group of Ten (2001) and De Nicolò and Kwast (2002), these definitions embed a key necessary condition for a financial shock to induce adverse systemic real risk realizations: financial shocks must be highly likely to induce significant adverse real effects, such as substantial reductions in output and employment. Thus, it explicitly requires that the negative externalities of a financial shock that extend to the financial system also extend to the real economy. Financial markets turbulence, attendant increases in volatility and/or failures of financial intermediaries that are devoid of significant and widespread real effects are not classified as systemic.

We adopt these definitions for two reasons. First, distinguishing systemic financial risk from systemic real risk allows us to better assess the extent to which a realization of a financial shock is just amplifying a shock in the real sector, or originates in the financial system. Second, financial events that carry significant adverse real effects, such as sharp reductions in output and increases in unemployment, are what may affect welfare significantly and, as noted, are the ultimate concern of policy-makers.

B. Measurement

To control risk in financial institutions, risk managers track Value-at-Risk (VaR). VaR measures the worst possible portfolio loss over a given time horizon at a given probability. To control risk in the economy, policy makers may wish to track measures of worst possible real macroeconomic outcomes. One such a measure is GDP-at-Risk (GDPaR), defined here as the worst predicted realization of quarterly growth in real GDP at 5 percent probability.

To control risk in the financial system, policy makers may also wish to track measures of worst possible system-wide financial outcomes. One such a measure is financial system-at-risk (FSaR), defined as the worst predicted realization of the market-adjusted return of a large portfolios of financial firms at 5 percent probability. Following Campbell, Lo and MacKinlay (1997), this market-adjusted return is the return of a portfolio of financial firms less the return on the market. We chose this measure for the easiness with which can be embedded in the model described below. However, other indicators can be adapted to our framework, such as those based on distance-to-default measures as in De Nicolò et al. (2004), or based on CDS spreads, as in Huang, Zhou and Zhu (2009a, 2009b).

III. A Dynamic Factor Model of Systemic Risks

Denote real GDP growth with GDPGt, and the indicator of system-wide financial risk with FSt The joint dynamics of GDPGt and FSt is modeled by a version of the Dynamic Factor Model (DFM) detailed in Stock and Watson (2002, 2005).

The model is described by the following equations:

GDPGt=λR(L)ft+γ11(L)GDPGt1+γ12(L)FSt1+ut1(1)
FSt=λF(L)ft+γ21(L)GDPGt1+γ22(L)FSt1+ut2(2)
Xit=λi(L)ft+δiXit1+vit(3)
ft=Γ(L)ft1+ηt(4)

Equations (1) and (2) describe a VAR in GDPGt and FSt augmented with a factor structure. The dynamics of a (large) vector of series (predictors) Xt indexed by iN is represented by the factor model (3), where ft is a set of dynamic factors.4 Equation (4) describes the dynamics of these factors through a VAR.

As in Stock and Watson (2005), factors and idiosyncratic errors ut1, ut2, and vit are assumed to uncorrelated at all leads and lags. Assuming finite lags up to p, and defining the vector of static factors with Ft[ft,ft1,,ftp1], one obtains the static form representation of the DFM:

GDPGt=ΛRFt+γ11(L)GDPGt1+γ12(L)FSt1+ut1(5)
FSt=ΛFFt+γ21(L)GDPGt1+γ22(L)FSt1+ut2(6)
Xit=ΛiFt+δiXit1+vit(7)
Ft=Φ(L)Ft1+Gηt(8)

Note that Φ(L) includes Γ(L) and 0’s, while G is a matrix of coefficients of dimension rxq, where r is the number of static factors and q that of dynamic factors. If r = q, then Φ(L) = Γ(L) and G = I, that is, (8) is equivalent to (4).

Substituting (8) in (5) and (6), we obtain a Factor-Augmented VAR (FAVAR) representation of the DFM, akin to that adopted by Bernanke, Boivin, and Eliasz (2005):

Ft=Φ(L)Ft1+Gηt(9)
GDPGt=ΛRΦ(L)Ft1+γ11(L)GDPGt1+γ12(L)FSt1+ut1(10)
FSt=ΛFΦ(L)Ft1+γ21(L)GDPGt1+γ22(L)FSt1+ut2(11)

A. Systemic Risk Measures

Using estimates of the static factors Ft, the systemic risk indicators GDPaR and FSaR are obtained by estimating the following quantile regressions:

GDPGt=α1q+ΛqRFt+γ11q(L)GDPGt1+γ12q(L)FSt1+ut1q(12)
FSt=α2q+ΛqFFt+γ12q(L)GDPGt1+γ22q(L)FSt1+ut2q(13)

Denoting the estimated coefficients of (12) and (13) with a “hat”, GDPaRt and FSaRt are the fitted values of the quantile regressions (12) and (13) with q = 0.05 :

GDPaRt=α^1q+Λ^qRFt+γ^11q(L)GDPGt1+γ^12q(L)FSt1(14)
FSaRt=α^2q+Λ^qFFt+γ^12q(L)FSt1+γ^22q(L)GDPGt1(15)

B. Measures of Systemic Risk Spillovers

It can be useful and informative to compute measures of systemic risk spillovers from real activity to the financial sector (and viceversa) that are net of the impact of common factors on GDPaR and FSaR measures. These can be obtained by using the Covar measures introduced by Adrian and Brunnermeier (2008).

Estimates of Co(GDPaRt) and Co(FSaRt) are given by:

Co(GDPaRt)=α^1q+β^1qFt+γ^11q(L)GDPaRt1+γ^12q(L)FSaRt1(16)
Co(FSaRt)=α^2q+β^2qFt+γ^12q(L)GDPaRt1+γ^22q(L)FSaRt1(17)

The existence of systemic risk spillovers can be gauged comparing Co(GDPaR)t with GDPaRt, and Co(FSaR)t with FSaRt. For example, if Co(GDPaR)t < GDPaRt, then negative risk spillovers in the real sector arise from negative risk spillovers either in the real sector, or in the financial sector, or both. However, positive risk spillovers cannot be ruled out, since improvements in real activity, or a reduction in system-wide financial risk, can have positive feedback effects on either sectors. This is apparent noting that the differences between the Covar and the systemic risk measures are given by:

Co(GDPaR)tGDPaRt=γ^11q(L*)(GDPaRtGDPGt)+γ^12q(L*)(FSaRtFSt)(18)
Co(FSaR)tFSaRt=γ^12q(L*)(GDPaRtGDPGt)+γ^22q(L*)(FSaRtFSt)(19)

IV. Estimation and Forecasting

The first estimation step is to compute static factors and choose their number. Since our focus is on forecasts of systemic risk indicators, we adopt the following forecasting criterion to select both number of static factors and lags of the FAVAR (10)-(11).

First, we use principal components to extract all factors with eigenvalues greater than 1, in number R. Second, we order factors according to their explanatory power of the variance of the data, and construct F˜={(Fr=1),(F1,Fr=2),,(F1,F2,,Fr=R)}. Lastly, we choose the number of lags L and the number of static factors rF˜ that maximize FPE(L, r) + AIC(L, r), where FPE is the Final Prediction Error Criterion and AIC is the Akaike Information Criterion. As detailed below, our forecasting criterion turns out to yield an optimal number of static factors close to the number of dynamic factors obtained by applying the statistical criterions based on Bai and Ng (2003).

In the second estimation step, we use the optimal number of lags L* and number of static factors r* obtained in the previous step to estimate quantile regressions (12)-(13) Note that these quantile regressions can be viewed as forecasting equations of systemic risk indicators. Using the VAR of static factors described by equation (9), we compute dynamic forecasts of static factors k quarters ahead. Then, these forecasts are used to obtain recursive forecasts of indicators of systemic risk using estimated coefficients of regressions (12)-(13).

In sum, the foregoing procedure yield forecasts of GDPaR, FSaR, co(GDPaR) and co(FSaR) indicators k quarters ahead. 5

V. Identification and stress tests

We would like to know how systemic risk indicators respond to structural shocks in the economy. To this end, we can use impulse responses to identified structural shocks through the FAVAR. These impulse responses can be viewed as stress tests of systemic risk indicators to these structural shocks. At a given date, the size of these responses provides a gauge of the sensitivity of systemic risk indicators to shocks of a given (standardized) size. Between dates, changes in the size of impulse responses of the systemic risk indicators to a given shock can provide a measure of changes in the resilience of an economy to a given shock.

A. Orthogonalization

We can obtain impulse responses of “factors” to their orthogonalized innovations, and translate them into impulse responses of indicators of systemic risk in(14)-(15) via the estimated coefficients of the quantile regressions. Yet, orthogonal innovations extracted from the FAVAR estimation do not have any “economic” interpretation, although they have the useful property of being contemporaneously and serially uncorrelated. Their economic interpretation can be obtained through identification based on some underlying theoretical model, as detailed next.

Inverting (9) yields the Moving Average (MA) form of the factor VAR (equation(9)):

Ft=A(L)ηt(9a),

where A(L) = (1 − Φ(L)L)−1G.

Substituting (9a) in (10) and (11), we obtain:

GDPGt=ΛRA(L)ηt+γ11(L)GDPGt1+γ12(L)FSt1+ut1(10a)
FSt=ΛFA(L)ηt+γ21(L)GDPGt1+γ22(L)FSt1+ut2(11a)

For the sole purpose of identification, we make the simplifying assumption that the dynamic impact of FS on GDPG, and of GDPG on FS, is entirely captured by the dynamics of factors. This amounts to posit γ12(L) = γ21(L) = 0, and converts our forecasting model into the standard Factor VAR detailed in Stock and Watson (2005). Under this assumption, inverting (10a) and (11a) yields the MA representation of the FAVAR:

GDPGt=BR(L)ηt+wt1(10b),
FSt=BF(L)ηt+wt2(11b),

where BR(L) = (1 − γ11(L)L)−1 ΛR A(L), BF(L) = (1 − γ22(L)L)−1 ΛF A(L), wt1=(1γ11(L)L)1ut1 and wt2=(1γ22(L)L)1ut2.

Likewise, the MA representation of the systemic risk indicators is:

GDPaRt=BqR(L)ηt+v11q(14a),
FSaRt=BqR(L)ηt+vt2q(15a),

where BqR(L)=(1γ11q(L)L)1ΛqRA(L),BqF(L)=(1γ22q(L)L)1ΛqFA(L),vtq1=(1γ11q(L)L)1ut1q and vtq2=(1γ22q(L)L)1ut2q.

B. Theory-based identification

Extending the identification procedure introduced in Canova and De Nicolò (2002), we identify a chosen set of orthogonal innovations as structural shocks if they satisfy certain sign restrictions on key variables derived from aggregate dynamic macroeconomic theory and a simple banking model.

Specifically, the theoretical restrictions on the responses of key aggregates to structural shocks implied by an aggregate macroeconomic model are as follows. If a positive temporary orthogonal innovation represents a positive transitory aggregate supply shock, then it should generate transitory weakly positive output responses and weakly negative transitory responses in inflation, depending on capacity utilization. On the other hand, if it is a real aggregate demand shock, it should generate weakly positive transitory responses in output and inflation. Canova and De Nicolò (2002) show that these sign restrictions can be derived from a wide class of general equilibrium monetary macroeconomic models with different microfoundations.

What are the implications of these theoretical responses for the demand and supply of bank credit? To answer this question, we use the implications of textbook partial equilibrium banking models, as for example described in Chapter 3 of Freixas and Rochet (2008), or the simple model in Boyd, De Nicolò and Loukoianova (2009). In these models, aggregate shocks can have an impact on both the demand for credit and the supply of funding for intermediaries. Specifically, the theoretical restrictions on the responses of bank credit growth and changes in loan rates implied by these banking models are as follows. If there is a positive transitory shock to the demand for bank credit (e.g. because of a positive technology shock to firms generating an increase in demand for investment, or an increase in the quality of investment prospects), then we should observe a transitory increase in bank credit growth and an increase in loan rates.

We call a shock generating these responses a positive credit demand shock. Conversely, if there is a positive transitory shock to the supply of bank credit (e.g. the supply of bank liabilities increases or banks expand by raising capital), then we should observe a transitory increase in bank credit growth but a decline in loan rates. We call a shock generating these responses a positive credit supply shock. Of course, negative shocks have all the signs of these responses reversed.

Note that real aggregate demand or supply shocks can affect the underlying drivers of the supply and demand for bank credit simultaneously. For example, a negative aggregate demand shock can induce firms and household to decrease their demand for bank credit, shifting the demand for bank credit to the left: this would result in a decline in loan rates ceteris paribus. At the same time, the adverse wealth effects of a negative aggregate demand shock may induce investors to reduce their supply of loanable funds to banks, or banks could reduce their supply of credit as they may become increasingly capital constrained or risk averse: this would result in a leftward shift in the supply of credit ceteris paribus. Which effect dominates on net will be reflected in movements in loan rates and bank credit growth. If negative credit demand shocks dominate, then loan rates and bank credit growth should decline, while the converse would be true if negative credit supply shocks dominate.

Table A below summarizes the responses of GDP growth, inflation, bank lending growth, and changes in loan rates in response to positive structural shocks implied by standard aggregate macroeconomic models and partial equilibrium banking models:

Table A.

Theoretical responses of key variables to positive shocks

article image

Identification of structural shocks will be conducted by checking whether a subset of orthogonal innovations of the FAVAR produces responses of the four variables considered that match the signs of the responses implied by theory.

VI. Implementation

Our modeling procedure is implemented using quarterly macroeconomic and financial series for the G-7 economies for the period 1980:Q1-2009:Q3. All series are taken from Datastream.

For each country, the vector of quarterly series Xt in equation (3) includes about 95 series, which are detailed in the Appendix. They can be classified into three main groups. The first group comprises equity markets data, including prices, price/earnings ratios and dividend yields for the entire market and by sector. The inclusion of all sectors spanning from manufacturing to services allows us to gauge the differential impact of shocks on different sectors of the economy, as well as to capture the impact of specific sectors on systemic risks. The second group includes financial, monetary and banking variables related to credit conditions, namely: interest rates for different maturities, monetary policy rates, bank prime rates and interbank rates, bank lending, and monetary aggregates. The third and last group includes price and quantity indicators of real activity. This set of variables includes net exports, capacity utilization, firms’ investment, consumer confidence, unemployment, consumption and saving for firms, government and household, a consumer price index, industrial production, house prices and manufacturing orders.

In the reminder of this section, we first report some descriptive statistics, then we detail the results of the forecasting model of systemic risks, and lastly, we carry out a benchmark identification of structural shocks, examining the responses of the systemic risk indicators to these shocks.

A. Descriptive Statistics

Table 1 reports basic statistics for GDP growth (GDPG) and our system-wide indicator of financial risk (FS). Three facts are worth noticing. First, ranges as well as volatilities of GDPG and FS appear to differ markedly across countries, suggesting differential sensitivities of these indicators to underlying shocks. Second, means of FS are generally small and not different from 0 according to simple t-statistics tests: this is expected, as in the long-run the evolution of bank stock returns tracks that of the market. Third, the contemporaneous correlation between GDPG and FS appears relatively small, with no significant correlation for the U.S., Canada,, Japan and Italy, and a positive and significant—albeit small—correlation for the U.K., France and Germany.

Table 1.

Descriptive Statistics of Real GDP Growth (GDPG) and the System-wide Financial Risk Indicator (FS)

article image
N.B.: Bold values indicate an estimate significantly different from 0 at a 5 percent confidence level

As shown in Figure Set 1, however, the comovement between GDPG and FS appears to be the most pronounced during recessions and the latest “crisis” period in all countries. This suggests either an increase in the sensitivities of both indicators to common shocks, or a significant increase in risk spillovers between real and financial activity, or a combination of both.

Figure Set 1.
Figure Set 1.
Figure Set 1.

GDP Growth and FS Indicators

Citation: IMF Working Papers 2010, 029; 10.5089/9781451962567.001.A001

Assessing to what extent movements in real activity and the financial risk indicator are primarily driven by common shocks or primarily by spillovers is especially important during periods of both real and financial instability. Whether the recent crisis has been one in which the sharp contraction in real activity registered at end-2008 and beginning 2009 has been caused by sharp declines in the aggregate supply of bank credit, or alternatively, sharp declines in real activity are the main drivers of the reduction in the demand for bank credit, is still an open issue. Indeed, the conventional wisdom has been one in which the credit crunch has prompted banking systems to curtail lending, and banks’ increasingly binding capital constraints have forced banks to de-leverage, with the attendant contraction of their asset size and further constraints in their lending capacity. Yet, bank loan growth in the U.S. and the Euro area, for example, has been buoyant since the start of the crisis, although it has decelerated since September 2008. This may suggest that the contraction in bank lending growth reflects primarily the sharp decline in the demand for credit resulting from the severe contraction in consumption growth and investment.6

Identification is essential to address these issues, and this is exactly what we do. Capturing the main drivers of the demand and supply of credit, and assessing whether shifts in the demand or supply of bank credit dominate on net requires identification of structural shocks.

B. Estimation and Forecasting

We estimated static factors and autoregressive coefficients of each variable by principal components according to the iterative procedure described in Stock and Watson (2005), and chose their number and the lags of equations (12) and (13) according to the forecasting criterion described previously. Notably, for all datasets of the seven countries our forecasting criterion selected the same number of static factors and lags: five factors and one lag.

As a cross-check, we also estimated the number of static factors chosen according to the Bai and Ng’s ICp1 and ICp2 criterions, obtaining 11 static factors for the U.S.—consistent with Stock and Watson (2005) results—and between 9 and 12 static factors for the other countries. We also estimated the number of dynamic factors as principal components of the residuals of each variable in equation (10) and (11), obtaining 6 dynamic factors for the U.S., and between 4 and 6 dynamic factors for the other countries.

In light of these results, and because our focus is on forecasting and on identification with restrictions dictated by theory, we acted conservatively by treating the five estimated static factors equal to the number of dynamic factors, essentially assuming Ft = ft, so that in equation (8) G = I.

We used these five estimated factors as independent variables of quantile regressions (14) and (15) specified with one lag. The resulting GDPaR and FSaR estimates were also used to compute Covar measures (16) and (17).

As detailed in the previous section, forecasts of GDPaR and FSaR eight quarters ahead were obtained projecting forward the factors through the VAR of equation (8) and using the estimated quantile coefficients to project forward GDPaR and FSaR values. Forecasts were undertaken with all data available as of September 25, 2009, that is, at end-2009Q3. Note, however, that at that time actual real GDP was available only up to 2009Q2, so that the first effective forecast date for GDPaR is 2009Q3 and the estimated 2009Q3 GDP growth is a “nowcast”.

Figure Set 2 reports estimated GDPaR and FSaR series, together with their forecasts eight quarters ahead of 2009Q3. Table 2 reports basic descriptive statistics of the systemic risk indicators, as well as the difference between Covar and at-risk measures. As noted, the latter measure is useful to gauge risk spillovers in excess of those implied by the dependence of both measures on common factors.

Figure Set 2.
Figure Set 2.
Figure Set 2.

GDPaR and FSaR Estimates and Forecasts (2009q3-2011q2)

Citation: IMF Working Papers 2010, 029; 10.5089/9781451962567.001.A001

Table 2.

Descriptive Statistics of Systemic Risk Indicators

article image
N.B.: GDPaR is GDP at risk; FsaR is the Financial-system at risk indicator; dcoGDPaR = co(GdPaR)-GDPaR, where co(GDPaR) is the Covar version of the systemic real risk indicator; dcoFSaR = co(FSaR)-FSaR, where co(FSaR) is the Covar version of the systemic financial risk indicator.

We point out two main findings. First, means of FSaR estimates are very similar across countries, but their standard deviations vary significantly across countries. The converse is true for GDPaR, whose measures exhibit marked cross-country variations, while their standard deviations do not appear to vary markedly. Second, risk spillovers are present for GDPaR measures, as Table 2 exhibits negative values for all countries, while spillovers for FSaR measures are on average small and not significantly different from 0. Overall, common factors appear to be the dominant drivers of systemic risk indicators, whereas risk spillovers (net of common factors) seem relatively small in all countries.

Turning to GDPaR and FSaR forecasts, Figure Set 2 indicates for all countries a V-shaped pattern of systemic risk indicators, with forecasts pointing at a return of these systemic risk indicators to their historical mean by mid-2010. This means that the model predicts a significant decline in the size of real and financial losses associated with tail risk events.

One intuitive—albeit informal—way of judging the forecasting ability of the model is to assess whether out-of sample forecasts of the systemic risk indicator GDPaR move in the same direction of subsequent actual values of GDP growth. A full formal evaluation of the forecasting performance of the model is outside the scope of this paper. However, here we report perhaps the most demanding assessment of the model’s forecasting ability. Namely, we assess if the model signals a decline in GDPaR prior to one of the largest historical declines in real activity: that experienced in 2008Q4-2009Q1 in all G-7 countries.

Figure Set 3 reports the results of this comparison: the blue line is the out-of-sample GDPaR forecasts made in 2008Q3, while the red line is actual GDP growth. Predicted changes in GDPaR and actual GDP growth go in the same direction for at least 1 quarter ahead within a three quarters’ horizon (up to 2009Q1) in all countries. Although informal, we view this evidence as notable. The out-of sample consistency of GDPaR forecasts with the future evolution of actual GDP growth for the most unpredictable event in decades suggests the potential usefulness of our model as a real-time risk monitoring tool.

Figure Set 3.
Figure Set 3.

GDPaR Out-of-Sample Forecasts and Actual GDP Growth (2008q3-2009q1)

Citation: IMF Working Papers 2010, 029; 10.5089/9781451962567.001.A001

C. Identification of Structural Shocks

We implemented the identification procedure outlined previously following three steps. First, we selected an orthogonal decomposition of the MA representation (9a). Second, we computed impulse responses of FAVARs for GDP Growth, Inflation, Bank Lending Growth and first differences in Loan Rates for each country. Lastly, we checked whether the joint signs of the responses of these variables conformed to the signs predicted for different shocks by the basic macro and banking models summarized in Table A.

As a benchmark orthogonalization, we chose a Choleski decomposition with factors ordered according to their explanatory power of the common variations in the data, with factor 1 ordered first, factor 2 second, and so on, and with GDPG, Inflation, Bank Lending Growth and first differences in loan rates ordered last in each FAVAR equation. The simple assumption underlying this choice is that the casual ordering implied by this decomposition reflects the relative importance of factors in explaining variations in the data, and each idiosyncratic component of the observable variables does not affect any of the factors at impact.

To check robustness, however, we examined alternative decompositions with inverted ordering of the variables, obtaining similar signs of the responses of each of the observable variables to shock to orthogonalized innovations. We also examined the covariance matrix of innovations of the VAR of each country, and such matrices appeared approximately diagonal in all cases, indicating that the ordering of variables in the VAR was not likely to change results under the casual ordering selected. Furthermore, the approximate diagonality of these covariance matrices also suggests that our results may be robust to alternative orthogonal decompositions— not necessarily recursive—that can be extracted applying the systematic statistical search implemented by Canova and De Nicolò (2002).

Figure Set 4 reports impulse responses of GDP growth, Inflation, Bank Lending Growth and changes in Lending Rates for each of the G-7 countries. Strikingly, the response of all variables to all shocks at impact or for at least up to two quarters after impact is either strictly positive (in most cases) or non negative (in few cases).7 Hence, according to Table A, under the assumed benchmark orthogonalization, all structural shocks in these economies can be identified as aggregate demand shocks associated with bank credit demand shocks.

Figure Set 4.
Figure Set 4.
Figure Set 4.
Figure Set 4.
Figure Set 4.
Figure Set 4.
Figure Set 4.
Figure Set 4.

Impulse Responses of GDP Growth, Inflation, Bank Lending Growth and Change in Lending Rate to Shocks to Factors and Own Shock

Citation: IMF Working Papers 2010, 029; 10.5089/9781451962567.001.A001

The finding of aggregate demand shock as the predominant drivers of real cycles in the G-7 economies is matching exactly the findings by Canova and De Nicolò (2003), who used only a small dimension VAR for the G-7 countries, but implemented a full search for shocks interpretable according to aggregate macroeconomic theory in the entire space of non-recursive orthogonalizations of the VAR of each country.

The finding that aggregate bank demand shocks are the predominant drivers of cycles in bank credit growth is consistent with their being prompted by aggregate demand shocks. This result also supports the conjecture that slowdowns in aggregate bank credit growth are primarily the result of downturns in real activity, as they reflect declines in the aggregate demand for bank credit by households and firms, rather than a reduction in the aggregate supply of bank credit.

Notably, the five identified aggregate demand and bank credit demand shocks are not all the same, as they have a differential impact on GDP growth, inflation, bank lending growth and changes in loan rates within as well as between countries. This suggests that the sectors of the economy where they originate are different. As shown in Table 4, the variance decompositions of the four variables VAR in each country show that the variance explained by each shock varies across both variables and countries, with most shocks resulting relevant in each country.8

Table 3.

Variance Decomposition of GDP Growth, Inflation, Bank Lending Growth and Changes in Loan Rates to Identified Aggregate Demand and Bank Credit Demand Shocks

article image
N.B.: Boldfaced values denote estimates significantly different from 0 at 5 percent confidence levels.
Table 4.

Variance Decomposition of GDPaR and FSaR to Identified Aggregate Demand and Bank Credit Demand Shocks

article image
N.B.: Boldfaced values denote estimates significantly different from 0 at 5 percent confidence levels.

Similar results are obtained when we look at the impulse responses and variance decompositions of GDPaR and FSaR measures. As shown in Figure Set 5, the sign of the impact of each shock on GDPaR is essentially the same in each country, although magnitude and persistence of these shocks widely differ. As shown in Table 5, the relevant variance decompositions indicate the importance of each of the identified shocks for the systemic risk indicators in each country.

Figure Set 5.
Figure Set 5.
Figure Set 5.
Figure Set 5.
Figure Set 5.

Impulse Responses of GDPaR and FSaR to Identified Aggregate Demand and Bank Credit Demand Shocks and Own Shock

Citation: IMF Working Papers 2010, 029; 10.5089/9781451962567.001.A001

In sum, all identified structural shocks are aggregate demand shocks associated with bank credit demand shocks, this identification is the same for all countries considered, and appears robust to alternative orthogonalizations of the innovations in the FAVAR.

VII. Conclusion

This paper has developed a modeling framework that delivers forecasts of indicators of systemic real and financial risks that can be updated in real time. In addition, the proposed identification procedure allows gauging the sensitivity of these indicators to structural shocks identified by theory, giving economic content to stress tests. The implementation of such framework appears promising as a risk monitoring tool.

We view this framework as a first building block for an analysis of the determinants of systemic risks. As it can be inferred from our discussion, refinements and extensions of our framework are aplenty, since we have exploited the rich information provided by the factor model only in a limited way.

There remain deeper questions that need yet to be answered: where do these structural shocks originate? And, to which other sectors are they transmitted? In terms of Figure A of the introduction, answering these questions amounts to identifying in which “box” shocks originate, and disentangle the linkages between the originating box and other boxes in the picture, that is, the web of linkages implied by the transmission mechanism of these shocks.

Answering these questions amounts to exploit more fully the rich information structure provided by the factor model. We believe that such an exploration is likely to yield increasing returns. It can guide a more effective integration of financial frictions into current macroeconomic modeling, encourage the development of more disaggregated versions of such macroeconomic modeling by incorporating the insights of models of financial intermediation, and can be a powerful monitoring tool available to policy-makers. Carrying out some of these extensions is already part of our research agenda.

Appendix: List of Variables

All variables below are extracted for each country in the G-7 group during the 1980. Q1-2009.Q3 period. The frequency of all series is quarterly. Data transformations are implemented to make all series stationary. Δln = log level difference; Δlevels = level difference.

article image
article image

References

  • Adrian, Tobias, and Markus K. Brunnermeier, 2008, “Covar,” Federal Reserve Bank of New York Staff Report No. 348, September.

  • Bai, Jushan, and Serena Ng, 2002, “Determining the Number of Factors in Approximate Factor Models,” Econometrica, Vol. 70, No. 1: pp. 191-221.

    • Search Google Scholar
    • Export Citation
  • Bernanke, Ben, Jean Boivin, and Piotr Eliasz, 2005, “Measuring the Effects of Monetary Policy: A Factor-Augmented Vector Autoregressive (FAVAR) Approach,” Quarterly Journal of Economics, Vol. 120, pp. 387-422.

    • Search Google Scholar
    • Export Citation
  • Bernanke, Ben, Mark Gertler, and Simon Gilchrist, 1999, “The Financial Accelerator in a Quantitative Business Cycle Framework,” in Handbook of Macroeconomics, John Taylor and Michael Woodford eds., North Holland, Amsterdam, pp. 1341-393.

    • Search Google Scholar
    • Export Citation
  • Boyd, John, Gianni De Nicolò, and Elena Loukoianova, 2009, “Banking Crises and Crisis Dating: Theory and Evidence,” IMF Working Paper No. 09/141.

    • Search Google Scholar
    • Export Citation
  • Canova, Fabio, 2007, Methods for Applied Macroeconomic Research, Princeton University Press, Princeton, New Jersey.

  • Canova, Fabio, 2002, “Monetary Disturbances Matter for Business Cycle Fluctuations in the G-7,” Journal of Monetary Economics, Vol. 49, pp. 1121-159.

    • Search Google Scholar
    • Export Citation
  • Canova, Fabio, 2003, “On the Sources of Business Cycles in the G-7,” Journal of International Economics, Vol. 59, pp. 77-100.

  • Chari, V.V., Lawrence Christiano, and Patrick J. Kehoe, 2008, “Facts and Myths About the Financial Crisis of 2008,” Federal Reserve Bank of Minneapolis Working Paper 666, October.

    • Search Google Scholar
    • Export Citation
  • Christiano, Lawrence, Roberto Motto, and Massimo Rostagno, 2009, “Financial Factors in Economic Fluctuations. Manuscript,” May.

  • Cohen-Cole, Ethan, Burcu Duygan-Bump, José Fillat, and Judit Montoriol-Garriga, 2008, “Looking Behind the Aggregates: A Reply to Facts and Myths about the Financial Crisis of 2008,” Working Paper No. QAU08-5, Federal Reserve Bank of Boston.

    • Search Google Scholar
    • Export Citation
  • De Nicolò, Gianni, Peter Hayward, and Ashok Bathia, 2004, “U.S. Large Complex Banking Groups: Business Strategies, Risks and Surveillance Issues,” Special Issue Paper, IMF Country Report no. 04/228, July.

    • Search Google Scholar
    • Export Citation
  • De Nicolò, Gianni, and Myron Kwast, 2002, “Systemic Risk and Financial Consolidation: Are They Related?”, Journal of Banking and Finance, Vol. 26, No. 5, pp. 86180.

    • Search Google Scholar
    • Export Citation
  • Foerster, Andrew T., Pierre-Daniel G. Sarte, and Mark W. Watson, 2008, “Sectoral vs. Aggregate Shocks: A Structural Factor Analysis of Industrial Production,” NBER Working Paper No. 14389.

    • Search Google Scholar
    • Export Citation
  • Freixas, Xavier, and Jean-Charles Rochet, 2008, Microeconomics of Banking, 2 nd Edition, MIT Press, Cambridge, Massachusetts.

  • Group of Ten, 2001, “Report on Consolidation in the Financial Sector (Basel: Bank for International Settlements). http://www.bis.org/publ/gten05.pdf.

    • Search Google Scholar
    • Export Citation
  • Huang, Xin, Hao Zhou, and Hai Zhu, 2009a, A Framework for Assessing the Systemic Risk of Financial Institutions, BIS Working Paper No. 281, April.

    • Search Google Scholar
    • Export Citation
  • Huang, Xin, Hao Zhou, and Hai Zhu, 2009b, “Assessing the Systemic Risk of a Heterogeneous Portfolio of Banks During the Recent Financial Crisis,” FEDS Working Paper 2009-44, Board of Governors of the Federal Reserve System, Washington D.C., November.

    • Search Google Scholar
    • Export Citation
  • International Monetary Fund, 2009, Global Financial Stability Report, October, Washington, D.C.

  • Ivashina, Victoria, and Scharfstein, David, 2008, “Bank Lending During the Financial Crisis of 2008,” Working Paper, Harvard Business School.

    • Search Google Scholar
    • Export Citation
  • Stock, James, and Mark Watson, 2002, “Macroeconomic Forecasting Using Diffusion Indexes,” Journal of Business Economics and Statistics, April, pp. 147-62.

    • Search Google Scholar
    • Export Citation
  • Stock, James, and Mark Watson, 2005, “Implications of Dynamic Factor Models for VAR Analysis,” NBER Working Paper No. 11467.

  • Stock, James, and Mark Watson, 2006, “Forecasting with Many Predictors,” Chapter 10 in Handbook of Economic Forecasting, Graham Elliott, Clive W.J. Granger and Allan Timmermann Eds, North Holland, Amsterdam, pp. 516-54.

    • Search Google Scholar
    • Export Citation
  • Sorge, Marco, 2004, “Stress Testing Financial Systems: An Overview of Current Methodologies,” BIS Working Paper No. 165.

  • Walsh, Carl, 2009, “Using Monetary Policy to Stabilize Real Activity,” manuscript, September.

1

De Nicolò, IMF Research Department; Lucchetta, University of Venice. We thank without implications Fabio Canova, David Romer, Ken West, Hao Zhou, Harry Mamaysky and seminar participants at the IMF and at the November 2009 NBER/Fed Cleveland Research Conference on “Quantifying Systemic Risk” for comments and suggestions.

2

However, a rapidly growing literature, briefly reviewed by Walsh (2009), is exploring the implications of specific financial frictions in the context of extensions of the “financial accelerator” model of Bernanke, Gertler and Gilchrist (1999), with work by Christiano, Motto and Rostagno (2009) at the forefront of this effort.

3

See Sorge, 2004 for a review of stress testing, and Huang, Zhou and Zhu (2009a,2009b) for recent contributions.

4

Following Stock and Watson (2006), we do not include GDP growth and the FS indicator in the vector Xt of predictors.

5

Differing from Stock and Watson (2002), we obtain multistep-forecasts using the FAVAR rather than k-step projections. Assessing the relative merit of these procedures in terms of their out-of sample forecasting ability is a worthwhile enterprise in future applications.

6

For the U.S., Chari, Christiano and Kehoe (2008) made assertions at variance with the common wisdom, which were countered by Cohen-Cole et al., 2008 and Ivashina and Sharfstein (2008), to whom the former authors further replied. The issue is still open. For example, the IMF GFSR (2009) states that “This GFSR contends that the credit disruption has been an exogenous and significant factor in the global recession that began in 2008. However, it could be argued that the slowdown in credit is a symptom rather than a cause of the economic slowdown and merely reflects the lower demand of credit – by households and corporates – rather than a supply disruption.”

7

The only exception is the shock associated with the third factor for Canada, whose responses do not satisfy any of the sign restrictions in Table A, and thus it results unidentified.

8

The results echo the findings of an increased impact of sect oral shocks on aggregate industrial production indexes documented recently by Foerster, Sarte and Watson (2008).

Systemic Risks and the Macroeconomy
Author: Marcella Lucchetta and Mr. Gianni De Nicolo