An agnostic SVAR approach to financial shocks

Transcription

1 An agnostic SVAR approach to financial shocks Jelena Zivanovic Abstract Since the onset of Great Recession financial factors gained a more prominent role in the business cycle research. In particular, economists started to identify financial shocks as important sources of economic fluctuations. While being agnostic about financial shocks, this paper seeks to evaluate what is the role of these shocks for the US business cycle fluctuations. The effects of financial shocks on the real activity are significant and short-lived. Some conditional evidence for the behaviour of corporate debt composition is provided. The shocks originating in financial markets explain a modest part of the output variability, whereas they are the main drivers of financial indicators. Finally, I discuss the implications of my findings for theoretical frameworks. JEL Classification: E32, E44. Keywords: financial shocks, business cycles, structural vector autoregression, sign restrictions. Humboldt-Universität zu Berlin. jelena.zivanovic@wiwi.hu-berlin.de

2 Introduction The Great Recession of has opened many questions related to the relevance of standard drivers of U.S. business cycles. In the aftermath of the latest financial crisis, theoretical frameworks are developed to quantify effects of financial disturbances on the real economy (see Gertler and Karadi (2), Cúrdia and Woodford (2), Christiano et al. (24), among others). Simultaneously, empirical research is used to assess the role of financial shocks as drivers of fluctuations in macroeconomic variables and financial indicators (Gilchrist and Zakrajšek (22), Bekaert et al. (23), Bassett et al. (24) and Caldara et al. (24) to name a few). However, due to the lack of theoretical foundations and ad-hoc assumptions, identifications of financial disturbances in structural vector autoregressions (SVARs) are often problematic. The aim of this paper is to assess the importance of financial shocks in shaping the US business cycles within an SVAR model with sign restrictions by using economic theory as a lens to look at the data. The identification of shocks is based on a common denominator from various DSGE frameworks modelling the financial sector and financial frictions. In particular, an adverse financial shock is specified as a shock to asset prices, which results in a stock market bust combined with a decline in credit volume and an increase in credit spreads. The central feature of the identification is agnosticism about the impact of financial shocks on macroeconomic variables. Recent work using SVAR models has assumed that output, inflation and nominal interest rate move in the same direction upon a financial shock (e.g., Furlanetto et al. (24), Gambetti and Musso (24), Pinter et al. (23)). By imposing these restrictions, they explicitly assume financial shocks act like demand shocks. The predictions of financial DSGE models differ to which extent a financial shock is a demand or supply disturbance. Accordingly, the agnosticism about the nature of financial disturbance, which refers to the lack of restrictions on interest rate and inflation in my identification, is justified by representative theoretical models. The evidence on the importance of financial shocks for business cycle fluctuations has been mixed both in the theoretical and empirical analysis. In the financial accelerator DSGE model, Christiano et al. (24) find that risk shocks in the form of shocks to the variance of idiosyncratic technology explain up to 6% of output fluctuations. On the other hand, Carlstrom et al. (24) find that net worth shocks are responsible for % of output variability, when accounting for the return indexation of loan contract in the financial DSGE model. Similarly, the diversity is present in the empirical studies. For example, Furlanetto et al. (24) assign a dominant role to financial shocks arising in the housing market (accounting for 5% of output variation). On the other hand, other studies ascribe modest relevance to financial shocks (see Gilchrist and Zakrajšek (22), Bassett et al. (24), Meeks (22) among others), by explaining %-2% of output variability. By using the theory more stringently in the identification, my analysis provides evidence on the role of financial shocks The most common identification used in SVARs is the Choleski identification (c.f., Gilchrist and Zakrajšek (22), Bloom (29), Bekaert et al. (23)), whereby fast-moving financial variables are positioned after slow-moving macroeconomic variables. The recursive ordering of financial variables cannot be supported by the economic theory (c.f., Meeks (22)). Other specifications such as SVARs with sign restrictions suppose a priori the behaviour of macroeconomic variables.

3 for shaping the US business cycles. The SVAR framework will be used to assess the dynamics of corporate debt structure. Some recent evidence documents that a surge in bond financing and a decline in bank lending at the aggregate level arise during the US economic downturns (see Adrian et al. (22), Becker and Ivashina (24)), implying that the corporate debt composition is potentially a useful procyclical indicator. Furthermore, DeFiore and Uhlig (24) emphasise that the substitution from loan to bond financing in face of adverse financial shocks can reduce negative effects on the real economy in their DSGE model setup. Against this background, the empirical evidence regarding the dynamics of the debt composition is necessary to understand better the role of debt financing over the business cycle. My results show that financial shocks are associated with a decline in real economic activity and a substantial easing of monetary policy. The effect on output is statistically significant and short-lived. Monetary policy is very responsive to financial shock, since the interest rate is reduced considerably and for a prolonged period of time. The ambiguous response of inflation to financial shocks is in line with the lack of consensus in theory about the nature of financial shocks. I provide some evidence on the substitution of loan to bond financing in the presence of financial shocks. My results assign modest but relevant role to financial shocks since these shocks explain almost 2% of output variability. The remainder of the paper is organised along the following lines. Section 2 presents an overview of financial shocks used in business cycle analysis. Section 3 discusses the identification strategy. Section 4 presents the main results. Section 5 concludes. 2 Business Cycle and Financial Markets The phenomenon of business cycles lies at the core of macroeconomic research. The researchers have identified aggregate demand and supply shocks as key drivers of business cycle fluctuations. 2 The Great Recession of has challenged this view on sources of economic fluctuations. As a result, recent empirical research has emphasised nonstandard shocks originating in financial markets as important forces behind economic fluctuations (see Bassett et al. (24), Gilchrist and Zakrajšek (22), Furlanetto et al. (24), Meeks (22), among others). If the shock is relevant for business cycle fluctuations, it should describe the dynamics observed both in macroeconomic and financial data. My contribution among this line of research is to provide further evidence on the relevance of standard and nonstandard shocks in the ongoing debate about the causes of business cycle fluctuations. 2. Related literature Empirical research has tried to identify financial shocks and quantify their macroeconomic effects. The literature on financial shocks can be largely divided into three broad categories related to credit shocks, credit spread shocks and uncertainty shocks. 2 An excellent review of the business cycle research has been provided in the special issue of Review of Economic Dynamics, see Schmitt-Grohé and Uribe (2) among others. 2

4 The first line of research to assess credit shocks employs the measures of the total credit volume, lending standards and loan rates. In the SVAR with sign restrictions, an adverse loan supply shock is, for example, identified as a decline in loan rate, a contraction in the loan volume, a decline in real GDP, nominal interest rate and inflation (c.f., Gambetti and Musso (24)). This example illustrates that the reactions of macroeconomic variables to the financial shock are specified by assumption. The identified financial shock acts a priori as a demand shock, which is not supported by some theoretical models (e.g., Cúrdia and Woodford (2), Gerali et al. (2)). Alternatively, the recursive VAR identification implies that credit shocks affect macroeconomic variables with a lag, as specified by Bassett et al. (24). Both identification strategies are not fully supported by the theory. To address this gap in the identification of financial shocks, my approach aims to be consistent with the theoretical implications and let the data speak. The second line of research specifies the shocks to corporate bond spreads as the source of financial disturbances. These shocks result in declining output, slow recoveries, disinflation and a decline in interest rate. The shocks to credit spreads explain between and 2% of output fluctuations. Meeks (22) identifies credit spread shocks as exogenous movements that impact the non-default component of high yield spread in a VAR with sign restrictions. The author is rather agnostic about the impact reactions of macroeconomic variables to the shock, but the explanation of his restrictions is not based on the representative financial DSGE models (but rather the mechanism present in Bernanke et al. (999)). Using financial literature on the credit spreads, he argues that macroeconomic shock drive the default part of the spread, whereas the non-default part is explained by financial shocks. My analysis will solely impose the sign restrictions that are line with representative DSGE models, and therefore differ from the identification by Meeks (22). Gilchrist and Zakrajšek (22) construct a corporate credit spread index, consisting of the default part and the excess bond premium, which is interpreted as the changes in the pricing of default risk. The recursive identification of shock to excess bond premium does not have explicit theoretical foundations. Though DSGE models emphasise the role of external finance premium in amplifying the shocks via the mechanism of financial accelerator, it is not clear that the premium should be ordered as a fast-moving variable after output and inflation. Therefore, my agnostic approach to SVAR circumvents this issue by relying on the consensus from the dynamic consequences of DSGE models following financial and macroeconomic shocks. The third line of research refers to uncertainty shocks. After Bloom (29) initiated a discussion about uncertainty shocks, numerous studies have shown that an unexpected increase in uncertainty generates negative effects on real economy (see Bekaert et al. (23), Leduc and Liu (24), Caldara et al. (24), Furlanetto et al. (24), Gilchrist et al. (24) among others). A few recent studies attempt to differentiate between a financial shock (e.g. a credit spread shock) and an uncertainty shock (see Furlanetto et al. (24), Caldara et al. (24), Gilchrist et al. (24)). 3 These studies show that uncertainty shocks are 3 Furlanetto et al. (24) use the magnitude of the response of a proxy for excess bond premium relative to the uncertainty measure (VIX) to differentiate between uncertainty and financial (credit) shocks. On the other hand, Caldara et al. (24) identify two shocks by focusing on the maximal increase in the bond spread and an uncertainty indicator, respectively. 3

5 less important than financial shocks. In particular, Furlanetto et al. (24) find that housing shocks are dominant driving forces behind business cycle fluctuations. Pinter et al. (23) identify risk news shocks by imposing sign restrictions in line with a model by Smets and Wouters (27) including the financial accelerator and a model by Christiano et al. (24). The authors report a limited importance of risk shocks which stands in a stark contrast to an important role of risk shocks for output fluctuations documented in the theoretical framework by Christiano et al. (24). Given these findings, I do not explicitly consider the uncertainty-related financial shocks. The implications of the results on financial disturbances are threefold. First, some identification strategies of financial shocks are problematic. Second, the source of financial shocks remains to be an open question. Third, the evidence on the importance of financial shocks for the business cycle fluctuations is divided. 3 Identification Strategy The identification of financial shocks in the SVAR framework suffers often from the lack of more stringent theoretical basis. In particular, the most common identification, the Cholesky identification, uses the recursive ordering of variables. However, in the context of financial literature, it seems problematic to sensibly order financial variables and find support for a particular recursive order from the theory (see Bjørnland and Leitemo (29), Furlanetto et al. (24)). To avoid making these ad hoc assumptions, SVAR with sign restrictions is used to identify financial shocks. A set of robust sign restrictions 4 is used to identify structural shocks, whereas the dynamics of unrestricted variables should indicate the preferred theoretical models. The justification for the following sign restrictions is grounded in the macroeconomic theory modelling financial markets and financial frictions. A sign is not assigned to a specific variable, when the theory does not provide consensus on the implications of shocks for the variable at question. Table summarises the restrictions used in the SVAR specification. The imposition of restrictions only on impact represents a robust identification scheme and conforms with a larger set of theoretical models (see Canova and Paustian (2) and Gambetti and Musso (24)). Despite different modelling assumptions, nuisance features and transmission mechanisms, dynamic consequences from a set of DSGE models with financial frictions imply that the same sign restrictions hold across most of these model specifications as indicated by Table 2. An adverse financial shock is associated with a decline in asset prices, a decline in credit and an increase in the EFP. By taking a common denominator across the different financial theoretical frameworks, I identify a shock originating in financial markets with a minimum set of sign restrictions, shared by a majority of theoretical models. 5 My approach of finding a cross-section among financial frictions models differs from 4 See appendix A for the summary of SVAR methodology with sign restrictions. 5 Table 2 gives an overview of representative and prominent modelling choices of the financial sector. The work by Brzoza-Brzezina et al. (23) presents two main mechanisms used to model financial frictions, i.e., financial accelerator and collateral constrains. 4

6 Table : Sign restrictions Supply Demand Monetary Financial Y NA π NA R NA - + NA q - NA - - EFP NA NA NA + Credit NA NA NA - Debt composition NA NA NA NA Note: Y stands for real output, π inflation rate, R nominal interest rate, q the price of capital, EFP external finance premium. A + indicates that the impact response is positive; a - indicates that the impact response is negative; NA a impulse response can be on both signs of zero line and therefore no sign is assigned. All the shocks represent adverse disturbances. the method employed by Canova and Paustian (2) and Peersman and Straub (29), who derive restrictions common to different models (e.g., flexible price and sticky price models) and robust to a range of structural parameter values. Additionally, the source of financial shock varies across the models (e.g., a risk shock, a destruction of the productive capital, a credit spread shock, an origination of bad loans to name a few specified in Table 2). The identified shock to asset prices is compatible with the different theoretical counterparts. In the sensitivity analysis, I explore the dynamic consequences of additional financial shocks originating in financial markets. 6 Unfortunately, the theory does not provide a conclusive guidance on how to disentangle different sources of financial shocks and therefore I consider each financial shock separately. It is worth noting that Furlanetto et al. (24) disentangle credit from uncertainty shocks based on the distinction between net worth and risk shocks in the theoretical framework by Christiano et al. (24). 7 As this is not shared by other financial DSGE models (c.f., Carlstrom et al. (24), Gertler and Karadi (2)), I do not find sufficient theoretical support to differentiate across financial disturbances. In DSGE models with the financial accelerator framework, an adverse financial shock affects directly the financial position of firm, and therefore reduces the firm s borrowing capacity specified via a financial contract. The credit channel amplifies shocks, as worsening firm s financial position (lower net worth and price of capital) leads to higher external costs of financing (higher external finance premium), which reduces resources for investment and production. In the model with collateral constrains, a shock resulting in higher spreads tightens the collateral constraint, affecting negatively loans and investment. As the demand for capital decreases, its price goes down. The absence of restrictions on the macroeconomic variables upon a financial shock comes about for two reasons. First, the agnosticism about the reactions of macroeconomic variables allows the data to speak for itself as argued by Uhlig (25). The purpose of the SVAR is to recover the signs of inflation and 6 The two additional shocks will be characterised by the same set of sign restrictions, but correspond to an unexpected decline in credit and credit spread, respectively. 7 Christiano et al. (24) find that conditional on the positive net worth shock, the total credit falls as a result of an increase in the expected return to capital. The decrease in risk, on the other hand, directly increases the amount of borrowing stipulated by the debt contract and brings about a rise in the total credit. 5

7 Table 2: Sign restrictions related to financial shocks Shock Mechanism EFP Credit Q Carlstrom et al. (24) risk debt contract Carlstrom et al. (24) net worth debt contract Christiano et al. (24) risk debt contract Christiano et al. (2b) bank funding debt contract Christiano et al. (2b) liq. buffer debt contract - - Cúrdia and Woodford (2) bank technology heterogeneity + - NA Cúrdia and Woodford (2) bad loans heterogeneity + - NA Gerali et al. (2) bank mon.competition Gertler and Karadi (2) net worth moral hazard Gertler and Karadi (2) capital quality moral hazard Meh and Moran (2) bank funding moral hazard Brzoza-Brzezina et al. (23) net worth debt contract Brzoza-Brzezina et al. (23) riskiness debt contract Brzoza-Brzezina et al. (23) LTV collateral constraint - - Brzoza-Brzezina et al. (23) spread collateral constraint A + indicates that the impact response is positive; a - indicates that the impact response is negative; indicates that the variable is not considered in the model; a indicates a zero-response of the variable on impact; NA indicates that the model does not include a specific variable. LTV stands for loan-to-value ratio. nominal interest rate upon financial disturbances. Recent studies have assumed that output, inflation and nominal interest rate move in the same direction upon a financial shock (e.g., Furlanetto et al. (24), Gambetti and Musso (24), Pinter et al. (23)). To avoid a priori viewpoint on the financial disturbance as a demand-like disturbance, my analysis remains agnostic about the impact responses of inflation and nominal interest rate. For the similar reason, Hristov et al. (22) and Fornari and Stracca (23) take no stand on inflation, however, they specify the reaction of nominal interest rate to financial shocks, which is not fully supported by the theory. Second, the lack of consensus on the impact responses of inflation and nominal interest rate is obvious in Table 5. For example, financial DSGE models with the financial accelerator generate a procyclical interest rate and inflation rate following risk shocks, however, the variables behave differently in the presence of banking shocks (c.f., Meh and Moran (2), Christiano et al. (2b)). A similar argument applies to spread shocks in models containing collateral constraints (c.f., Brzoza-Brzezina et al. (23)). Therefore, the agnosticism is justified by the representative theoretical models. Since theoretical models uniformly imply negative effect of financial shock on output, my sign restriction regarding output complies with theory, which is different from Meeks (22). Whereas most of sign restrictions used to identify aggregate demand, aggregate supply and monetary policy shocks follow the common practice (c.f., Canova and Nicoló (22), Peersman and Straub (26)), signs are additionally imposed on the price of capital, as DSGE models have mostly unanimous implications regarding this financial variable (see Table 4). Aggregate supply shocks move the output together with the price of capital in one direction and inflation in the opposite direction. Aggregate demand shocks are associated with the movements of output, inflation and the interest rate in the same direction. An unexpected 6

8 increase in the nominal interest results in an decrease in output, inflation and price of capital on impact. Examples of aggregate supply shocks include technology shocks, price markup and labour supply shocks. The signs regarding demand shocks are compatible with the dynamic consequences of a contractionary fiscal shock or a negative preference shock in financial DSGE models. The monetary shock does not refer to shocks arising from unconventional monetary policy measures as analysed by Baumeister and Benati (23). The financial sector of the economy is left mostly unrestricted in the presence of the standard economic shocks. The reason for this is that there is no consensus in the theory regarding the responses of financial variables to these shocks. Major DSGE models with financial frictions imply opposing answers regarding the dynamics of financial variables. The dynamic consequences of the external finance premium (EFP) will be used to illustrate this point. The dynamic reaction of EFP to shocks depends on how the capital accumulation and financial frictions are specified. The financial accelerator model à la BGG (999) has a countercyclical premium in the presence of technology, monetary and net worth shock. DeGraeve (28) uses investment adjustment costs in the capital accumulation and finds that the EFP is procyclical following demand shocks and countercyclical following supply and monetary shocks. Christensen and Dib (28) incorporate the debt-deflation channel in the financial accelerator model by specifying the loan contract in the nominal terms. This change results in the procyclical EFP in face of aggregate supply shocks. When the loan contract is indexed to the aggregate return as in Carlstrom et al. (24), the EFP is procyclical in response to monetary policy shock. In the model with collateral constraints as presented by Brzoza-Brzezina et al. (23) the financial spread does not react to standard economic shocks. Similarly, the behaviour is credit is influenced by the modelling choices. Demand shocks in form of unexpected government spending lead to crowding out (countercyclical credit), whereas preference (demand) shock bring about the procyclical credit. The appendix provides an extensive overview of estimated and calibrated DSGE models regarding ambiguous responses of EFP and credit, which justify the absence of sign restrictions. The unequivocal evidence regarding the price of capital from the set of financial DSGE models is used for obtaining robust restrictions on impact in my estimation. Though most DSGE setups do not model more than one debt instrument, noteworthy exceptions represent the models by DeFiore and Uhlig (24) and Verona et al. (23). Verona et al. (23) show that the economy features a substitution between bond and loan financing in the presence of monetary shocks. On the other hand, DeFiore and Uhlig (24) show how an endogenous debt composition, i.e. the shift from loan to bond financing, can shield the economy from negative effects of adverse financial shocks. Given very little theoretical support on the corporate debt composition (the use of loan relative to bond financing), I impose no restrictions on the corporate debt composition. Therefore, the following empirical work will provide necessary conditional evidence regarding the corporate debt composition and present a validation exercise for the existing models using two types of debt financing. 7

9 3. Data The complete dataset used in the VAR specification include: i) the real GDP ii) the GDP deflator; iii) the effective federal funds rate iv) the S&P5 stock market index v) the credit measure vi) the credit spread vii) the ratio of loansto-bonds. All the financial variables except for the interest rates are deflated by the GDP deflator and expressed in growth rates. VAR specification includes a constant and one lag of endogenous variables. The set of financial indicators are supposed to be representative for stock and credit markets. S&P5 stock market index is used as a proxy for price of capital (c.f., Bassett et al. (24), Christiano et al. (2b), Christiano et al. (24)). 8 The credit spread is calculated as a difference between the yield on the BAA corporate bond index and the -year-treasury yield. This spread measure is used as a proxy of an external finance premium (see Carlstrom et al. (24), Furlanetto et al. (24)). The empirical counterpart for total lending to firms in theoretical models is represented by the amount of credit to the corporate nonfinancial sector (a sum of corporate bonds, loans, mortgages and commercial papers). The loan-to-bond ratio is expressed as a difference between the growth rates of loans and bonds and measures changes in the corporate debt composition. Table 3 summarises the unconditional correlations between real GDP growth (i.e., quarter-over-quarter log differences in real GDP) and four financial indicators. These correlations are computed for the entire data set (957Q-24Q2), 9 whereas the second row refers to the data in all recessions. As noted elsewhere in the literature, the credit and stock market indicators are procyclical, whereas the credit spread is countercyclical. Additionally, the corporate debt composition is positively correlated with the real GDP. The economic downturns are associated with the decrease in loan-to-bond ratio (change in the corporate debt composition in favour of bonds), however, the positive correlation between the ratio and the real GDP during US recessions is not significant. Table 3: Correlation between output and financial indicators Credit Stock index Spread Loan-to-bond Entire sample (957Q-24Q2) NBER Recessions (957Q-24Q2) Note: indicates a level of significance of %. indicates a level of significance of 5%. indicates a level of significance of %. 8 The price of capital is one of the major determinants of net worth of firms in theoretical models, which is empirically identified with the value of equity, i.e., the stock market index. Following Christiano et al. (24), the price of capital and the value of equity is the identical concept in the data. 9 The data set begins in 957Q because the data availability of stock market prices. 8

10 4 Results This section presents empirical results for the questions outlined in the introduction. First, what is the effect of financial shocks on the real economic activity? Second, what is the relative importance of financial shocks for business cycle fluctuations? Third, what does the data imply for the nature of financial shocks? Fourth, to which extent do traditional economic shocks drive the dynamics of financial data? Fifth, is there evidence for the change in the aggregate corporate debt composition following financial shocks? By addressing these issues, I try to shed light on the role of financial factors in shaping the US business cycles. 4. Financial shocks Figure displays the median impulse responses of the seven endogenous variables to a one standard deviation adverse financial shock. The shaded bands represent the 6% and 84% percentiles of the impulse-response function distribution at each horizon. A disruption in the financial market generates an economic downturn, which is mitigated by an easing of the monetary policy. The negative effect on real GDP is statistically significant and short-lived. The temporary impact of financial shocks on real GDP has been also documented by Gambetti and Musso (24) in the context of loan supply shocks and Pinter et al. (23) in the context of risk shocks. Figure : Financial shock GDP Inflation.5 Interest rate Stock prices Debt composition Credit Spread

11 Furthermore, if the agnosticism about financial shocks is taken seriously, the results confirm the ambiguous impulse responses of inflation implied by financial DSGE models. Because the response of inflation is statistically insignificant, the data cannot decide between the demand and supply-side financial disturbance. To my knowledge, most of the literature did not address the role of inflation dynamics following financial shocks; or if anything, the empirical research emphasised the disinflationary effect of financial shocks (see Gilchrist and Zakrajšek (22) and Furlanetto et al. (24)). Christiano et al. (2a) find that inflation is lower during US stock market booms than non-boom episodes over the last two centuries. This unconditional evidence is in line with the presence of inflation following negative asset prices shocks, as indicated by my conditional evidence in Figure. The impulse responses of the financial variables mimic the situation of the financial crisis. An adverse shock to asset prices, by assumption, causes a plunge in stock prices, a credit crunch and an elevated level of spreads. The responses of respective financial variables are statistically significant for several quarters after the imposed sign restrictions. The point estimates indicate that the relative shift towards bonds is remarkable and statistically significant at the medium horizon. The results represent, therefore, a conditional empirical evidence for the dynamics of the corporate debt structure, which is in line with unconditional evidence by Adrian et al. (22) and Becker and Ivashina (24). A positive co-movement between real GDP and the loan-to-bond ratio conditional on financial shocks is consistent with the respective correlation presented in Table 3. The results provide an empirical support for the behaviour of corporate debt composition following financial shocks as featured in the DSGE model of De- Fiore and Uhlig (24). Figure 3 shows the historical decomposition in which the contribution of financial shock to the total forecast error at each point in time. Financial shocks play a modest role in affecting the output over the last 6 years, however they are relatively more important in bringing down output during the Great Recession as well as in the late 99s and the beginning of 2s. The empirical assessment of the relevance of financial shocks is conducted on the basis of forecast error variance decomposition, as presented in Figure 3. Whereas asset price shocks explain approximately 2 percent of the output variation and % of inflation variation, the contributions of these shocks to the volatility of credit, stock price index and credit spread is substantial (approximately 5 percent, see Figure 4). My findings join a group of studies ascribing a modest contribution of financial shocks to output variability (see Gilchrist and Zakrajšek (22), Meeks (22), Pinter et al. (23), Peersman and Wagner (24)) and a very limited role of shocks for the inflation variability (similar to Furlanetto et al. (24)). Given this, theoretical frameworks modelling financial sector and respectve frictions associated with this sector should not overestimate the role of financial shocks. 4.2 Dynamics of financial factors Figure 2 plots the median impulse responses of the endogenous variables to adverse aggregate supply, aggregate demand and monetary policy shocks. A countercyclical credit spread arises in response to the aggregate demand and monetary shock, whereas the spread reacts procyclically to the aggregate sup-

12 ply shock. The credit falls over the short term in the presence of negative supply shocks, whereas it rises on impact in response to monetary and demand shocks. The supply-driven economic downturns are associated with an increase in the loan-to-bond ratio (change in corporate debt composition in favour of loans), whereas the ratio decreases in the presence of negative demand shocks. The finding of a positive comovement between output and the measure of corporate debt composition following demand shocks corroborates the unconditional evidence presented in Table 3. Figure 2: Standard economic shocks 4.3 Sensitivity analysis The alternative specification of financial shocks refers to shocks originating in the debt market. Since the same sign restrictions are employed, there are no qualitative differences in the results. The variance decompositions of real GDP, inflation and interest rate following each financial shock (asset price, credit and credit spread shock) are reported in Figure 5. The differences across the shocks in their explanatory ability are negligible. A further exercise refers to assessing the financial shocks by excluding the Great Recession. The results confirm the main results on the economic downturn following adverse financial shocks, however, the credible sets are larger (see Figure 6). Results are available upon request.

13 4.4 Implications What are the lessons from the SVAR estimation? First, the data cannot decide upon the nature of financial shock. The ambiguity about the response of inflation corroborates that both demand-side and supply-side financial disturbance have been present in the data. Second, the theoretical frameworks by Carlstrom et al. (24), Gertler and Karadi (2) and Christiano et al. (24) describe mechanisms, which implications are consistent with my empirical evidence. The shocks related to bank capital and bank funding are not supported by the empirical evidence. One possible explanation is that my SVAR specification focuses on the dynamics of debt and stock markets and does not explicitly include the relevant variables related to the banking sector. Third, the conditional empirical evidence for the corporate debt composition is presented. The substitution between loans and bonds arises in the presence of demand and financial shocks. Thus, this finding provides empirical support for the theoretical model by De- Fiore and Uhlig (24) and is inconsistent with the framework by Verona et al. (23). Fourth, the contributions of financial shocks in explaining output fluctuations are relevant forces behind business cycle fluctuations, however, they are not the dominant ones. In the light of these results, theoretical frameworks should not overemphasise the role of financial factors (c.f., Christiano et al. (24)). 5 Conclusion The aim of the paper was to evaluate the performance of financial and macroeconomic shocks when using the dynamic consequences implied by representative financial DSGE models to identify these shocks. In the SVAR specification I take an agnostic view on the impact responses of macroeconomic variables following financial disturbances and therefore the nature of financial shock. Adverse financial shocks lead to significant negative effects on the economy. A shock to asset prices generates a short-lasting contraction in real GDP, a substantial easing of monetary policy and a relative dominance of bond over loan financing. The data cannot decide upon the nature of financial disturbance. The presence of debt contract (as stipulated by Carlstrom et al. (24), Christiano et al. (24), Brzoza-Brzezina et al. (23)) or moral hazard (as presented by Gertler and Karadi (2)) represent the modelling frameworks which are supported by my empirical analysis. The lack of persistence in the reaction of real output to financial shock has been discussed in the theoretical model with rich financial sector by Canova et al. (25). Understanding the mechanisms under which financial factors play only limited role for the business cycles represents the potential avenue for the further research. References Adrian, T., Colla, P., and Shin, H. S. (22). Which financial frictions? parsing the evidence from the financial crisis of Working Paper 8335, National Bureau of Economic Research. Bassett, W. F., Chosak, M. B., Driscoll, J. C., and Zakrajšek, E. (24). 2

14 Changes in bank lending standards and the macroeconomy. Journal of Monetary Economics, 62:23 4. Baumeister, C. and Benati, L. (23). Unconventional monetary policy and the great recession: Estimating the macroeconomic effects of a spread compression at the zero lower bound. International Journal of Central Banking, 9: Becker, B. and Ivashina, V. (24). Cyclicality of credit supply: Firm level evidence. Journal of Monetary Economics, 62: Bekaert, G., Hoerova, M., and Duca, M. L. (23). Risk, uncertainty and monetarypolicy. Journal of Monetary Economics, 6: Bernanke, B. S., Gertler, M., and Gilchrist, S. (999). The financial accelerator in a quantitative business cycle framework. Handbook of Monetary Economics, (C): Edited by John B. Taylor and Michael Woodford. Bjørnland, H. C. and Leitemo, K. (29). Identifying the interdependence between US monetary policy and the stock market. Journal of Monetary Economics, 56(2): Bloom, N. (29). The impact of uncerteinty shocks. Econometrica, 77(3): Brzoza-Brzezina, M., Kolasa, M., and Makarski, K. (23). The anatomy of standard dsge models with financial frictions. Journal of Economic Dynamics and Control, 37():32 5. Caldara, D., Fuentes-Albero, C., Gilchrist, S., and Zakrajšek, E. (24). The macroeconomic impact of financial and uncertainty shocks. Manuscript. Canova, F., Coutinho, L., Mendicino, C., Pappa, E., Punzi, M. T., and Supera, D. (25). The domestic and the international effects of fnancial disturbances. Manuscript. Canova, F. and Nicoló, G. D. (22). Monetary disturbances matter for business fluctuations in the G-7. Journal of Monetary Economics, 49(6):3 59. Canova, F. and Paustian, M. (2). Business cycle measurement with some theory. Journal of Monetary Economics, 58(4): Carlstrom, C. T., Fuerst, T. S., Ortiz, A., and Paustian, M. (24). Estimating contract indexation in a financial accelerator model. Journal of Economic Dynamics and Control, 46:3 49. Christensen, I. and Dib, A. (28). The financial accelerator in an estimated New Keynesian model. Review of Economic Dynamics, pages Christiano, L., Ilut, C. L., Motto, R., and Rostagno, M. (2a). Monetary policy and stock market booms. Working Paper 642, National Bureau of Economic Research. Christiano, L. J., Motto, R., and Rostagno, M. (2b). Financial factors in economic fluctuations. Working Paper Series 92, European Central Bank. 3

15 Christiano, L. J., Motto, R., and Rostagno, M. (24). Risk shocks. American Economic Review, 4(): Cúrdia, V. and Woodford, M. (2). Credit spreads and monetary policy. Journal of Money, Credit and Banking, 42:3 35. Cúrdia, V. and Woodford, M. (2). The central-bank balance sheet as an instrument of monetary policy. Journal of Monetary Economics, 58(): DeFiore, F. and Uhlig, H. (24). Corporate debt structure and the financial crisis. Technical report, National Bureau of Economic Research. Working Paper 273. DeGraeve, F. (28). The external finance premium and the macroeconomy: US post-wwii evidence. Journal of Economic Dynamics and Control, 32(): Fornari, F. and Stracca, L. (23). What does a financial shock do? First international evidence. Technical report, European Central Bank. Working paper 522. Furlanetto, F., Ravazzolo, F., and Sarferaz, S. (24). Identification of financial factors in economic fluctuations. Working Paper 9/24, Norges Bank. Gambetti, L. and Musso, A. (24). Loan supply shocks and the business cycle. Manuscript. Gerali, A., Neri, S., Sessa, L., and Signoretti, F. M. (2). Credit and banking in a dsge model of the euro area. Journal of Money, Credit and Banking, 42:7 4. Gertler, M. and Karadi, P. (2). A model of unconventional monetary policy. Journal of Monetary Economics, 58():7 34. Gilchrist, S., Sim, J. W., and Zakrajšek, E. (24). Uncertainty, financial frictions, and investment dynamics. Technical Report 238, National Bureau of Economic Research. Gilchrist, S. and Zakrajšek, E. (22). Credit spreads and business cycle fluctuations. American Economic Review, 2(4): Hristov, N., Hülsewig, O., and Wollmershäuser, T. (22). Loan supply shocks during the financial crisis: Evidence for the Euro area. Journal of International Money and Finance, 3: Leduc, S. and Liu, Z. (24). Uncertainty shocks are aggregate demand shocks. Technical Report 22-, Federal Reserve Bank of San Francisco. Meeks, R. (22). Do credit market shocks drive output fluctuations? Evidence from corporate spreads and defaults. Journal of Economic Dynamics and Control, 36: Meh, C. A. and Moran, K. (2). The role of bank capital in the propagation of shocks. Journal of Economic Dynamics and Control, 34(3):

16 Peersman, G. and Straub, R. (26). Putting the new keynesian model to a test. IMF Working Paper 6/35, International Monetary Fund. Peersman, G. and Straub, R. (29). Technology shocks and robust sign restriction in a euro area svar. International Economic Review, 5: Peersman, G. and Wagner, W. (24). Shocks to bank lending, risk-taking, securitization, and their role for US business cycle fluctuations. Manuscript. Pinter, G., Theodoridis, K., and Yates, T. (23). Risk news shocks and the business cycle. Working Paper 483, Bank of England. Schmitt-Grohé, S. and Uribe, M. (2). Introduction to the special issue on the sources of business cycles. Review of Economic Dynamics, 4: 2. Smets, F. and Wouters, R. (27). Shocks and frictions in us business cycles: A bayesian dsge approach. American Economic Review, 67(3): Uhlig, H. (25). What are the effects of monetary policy on output? results from an agnostic identification procedure. Journal of Monetary Economics, 52(2): Verona, F., Martins, M. M. F., and Drumond, I. (23). (Un)anticipated monetary policy in a dsge model with a shadow banking system. International Journal of Central Banking, 9:73 8. Appendix A SVAR with sign restrictions: Consider the following VAR model y t = A + B () y t + B (2) y t B (M) y t M + u t, () where y t is a N vector containing N endogenous variables, A is N vector of constants, B (i) for i =,..M represents N N coefficient matrices and u t is N one-step ahead prediction error with a variance-covariance matrix, of size N N. The prediction error is related linearly to the structural shocks: u t = Sε t, (2) whereby S is a non-singular parameter matrix and ε t N(, I N ). A Bayesian estimation is undertaken to obtain the reduced-form VAR. Following Uhlig (25), both prior and posterior for (B (i), Σ) come from the Normal-Wishart distribution. He shows how the posterior can be analytically obtained. The procedure can be described as follows. In the first step, the Cholesky identification is used to retrieve the matrix S. In the next step, I use the candidate identifications yielding the identity variance covariance matrix. There exists a nonsingular matrix Q, such that the new impact matrix S = SQ and corresponding 5

17 structural shocks ε t = Q ε t, whereby reduced-form residuals u t = S ε t. Assuming that Q is an orthogonal matrix, i.e., Q = Q, the newly generated structural shocks have an identity variance covariance matrix: E[ε t ε t ] = E[Q ε t Qε t] = Q QE[ε t ε t] = I N. (3) Therefore, the candidate structural representations related to each S result in different impulse responses: y t = C(L)S ε t. (4) If the matrices C i from the reduced-form moving average in equation (4) are stacked, the response vector up to horizon h is given by R(h) = E[S S C S C 2... S C h] Q. (5) The sign restrictions related to the specific impulse responses are imposed on the column vectors of the above matrix. The algorithm used to set the sign restrictions is described in Rubio-Ramirez, Waggoner and Zha (2). 2 The functioning of the algorithm can be summarised as follows. I draw a Z matrix such that Z N(, I N ). Afterwards I undertake a QR decomposition of Z. This decomposition enables me to get the orthogonal matrix Q. In the next step, candidate impulse responses are obtained from SQ and B (i) for i =,..M. It is checked whether these generated impulse responses satisfy the sign restrictions. If the sign restrictions are not satisfied, a new Z is drawn and an iteration over the same procedure takes place until the sign restrictions are satisfied. The more sign restrictions are imposed, the longer it takes for the estimation to come to an end. The procedure is repeated so many times as necessary as it is to keep draws that satisfy sign restrictions. The obtained impulse responses are used to compute the statistics as well as to generate the standard error bands. Appendix B The following tables represent the justification for no restrictions on financial variables. The reported signs in Table 4, 6 and 7 related to the responses of respective variables are based on my calculations and results reported in the respective studies. All the models except for Gertler and Karadi (2), Brzoza- Brzezina et al. (23), Meh and Moran (2), Cúrdia and Woodford (2) are estimated. All the shocks represent adverse disturbances. 2 I am grateful to Francesco Ravazzolo for providing me with the algorithm. 6

18 Table 4: Sign restrictions on price of capital Supply Demand Monetary Financial Christensen and Dib (28) NA Carlstrom et al. (24) NA NA - - Christiano et al. (2b) DeGraeve (28) NA Gerali et al. (2) - -/+ - - Gertler and Karadi (2) - NA - - Meh and Moran (2) - NA - - Brzoza-Brzezina et al. (23) - NA - - Brzoza-Brzezina et al. (23) - NA - - Note: A + indicates that the impact response is positive; a - indicates that the impact response is negative; a indicates a zero-response of the variable on impact; NA indicates that the model does not consider a specific shock, +/- indicates that the impact response can be either positive or negative depending on the type of demand shock. Table 5: Sign restrictions on output, inflation and nominal inters rate Shock Mechanism Y π R Carlstrom et al. (24) risk debt contract Carlstrom et al. (24) net worth debt contract Christiano et al. (24) risk debt contract Christiano et al. (2b) bank funding debt contract Christiano et al. (2b) liq. buffer debt contract Cúrdia and Woodford (2) bad loans heterogeneity Gerali et al. (2) bank mon.competition Gertler and Karadi (2) net worth moral hazard Gertler and Karadi (2) capital quality moral hazard Meh and Moran (2) bank funding moral hazard Brzoza-Brzezina et al. (23) net worth debt contract + + Brzoza-Brzezina et al. (23) riskiness debt contract Brzoza-Brzezina et al. (23) LTV collateral constraint Brzoza-Brzezina et al. (23) spread collateral constraint Note: A + indicates that the impact response is positive; a - indicates that the impact response is negative; a indicates a zero-response of the variable on impact. 7

19 Table 6: Sign restrictions on EFP Supply Demand Monetary Financial Christensen and Dib (28) NA Carlstrom et al. (24) NA Christiano et al. (2b) Cúrdia and Woodford (2) + +/- - + DeGraeve (28) NA Gerali et al. (2) Gertler and Karadi (2) + NA + + Meh and Moran (2) + NA + + Brzoza-Brzezina et al. (23) + NA - - Brzoza-Brzezina et al. (23) NA - Note: A + indicates that the impact response is positive; a - indicates that the impact response is negative; a indicates a zero-response of the variable on impact; NA indicates that the model does not consider a specific shock. Table 7: Sign restrictions on credit Supply Demand Monetary Financial Christensen and Dib (28) NA Carlstrom et al. (24) NA Christiano et al. (2b) NA + Cúrdia and Woodford (2) DeGraeve (28) NA Gerali et al. (2) Gertler and Karadi (2) - NA - - Meh and Moran (2) - NA - - Brzoza-Brzezina et al. (23) NA - - Brzoza-Brzezina et al. (23) - NA - Note: A + indicates that the impact response is positive; a - indicates that the impact response is negative; a indicates a zero-response of the variable on impact; NA indicates that the model does not consider a specific shock. 8

20 Figure 3: Historical decomposition GDP Financial Shock Note: Contribution of financial shocks to deviations in GDP growth from its mean value Figure 4: Variance decomposition GDP Inflation Interest rate Stock prices Credit Spread Debt composition Financial shock Other shocks 9

21 Figure 5: Variance decomposition: Alternative specifications of financial shock Shock to asset prices GDP Inflation Interest rate Credit shock GDP Inflation Interest rate Financial shock Spread shock GDP Inflation Interest rate Other shocks Figure 6: Robustness check: Exclusion of Great Recession GDP Inflation.5 Interest rate Stock prices Debt composition Credit Spread