aa r X i v : . [ ec on . E M ] J un I NFLATION D YNAMICS OF F INANC IAL S HOC KS W ORKING P APER
Olli Palmén
Faculty of Social SciencesUniversity of HelsinkiP.O. Box 17 (Arkadiankatu 7),00014 University of Helsinki, Finland [email protected]
June 8, 2020 A BSTRACT
We study the effects of financial shocks on the United States economy by using a Bayesian struc-tural vector autoregressive (SVAR) model that exploits the non-normalities in the data. We usethis method to uniquely identify the model and employ inequality constraints to single out financialshocks. The results point to the existence of two distinct financial shocks that have opposing ef-fects on inflation, which supports the idea that financial shocks are transmitted to the real economythrough both demand and supply side channels.
Financial shocks have garnered considerable attention in the macroeconomic literature following the Global FinancialCrisis. Drawing from the recent experience, a number of studies find that disturbances in the financial sector play asignificant role in explaining business cycle fluctuations. However, notwithstanding the considerable attention to therole of the financial sector, the inflationary effects of financial shocks are not obvious.According to macroeconomic theory financial shocks may have both inflationary and disinflationary effects. Expan-sionary financial shocks are generally found to lead both to an increase in output and inflation through an increase inhouseholds’ and firms’ demand (Cúrdia and Woodford, 2010; Gertler and Karadi, 2011). However, Gerali et al. (2010)observe that the inflationary response of financial shocks may depend on the specific type of the shock. That is, thesign of the impact effect on inflation is different for shocks that affect banks’ capital and those that have an effect onthe supply of credit, such as household borrowing conditions. More recently, Gilchrist et al. (2017) argue that finan-cial market frictions attenuate the effect of output on prices, which explains the relatively mild deflationary pressuresexperienced during the Great Recession of 2008.Empirical estimates on the inflation dynamics of financial shocks are also not clear. Recent studies that use sign-identified structural vector autoregressive (SVAR) models to estimate the effects of financial shocks generally assumea positive co-movement between inflation and output as predicted by theory (Gambetti and Musso, 2017). How-ever, to account for the fact that theoretical macroeconomic models are consistent with both negative and positiveco-movement between output and inflation, recent empirical studies choose to remain agnostic about the effect of fi-nancial shocks on inflation (see for example, Abbate et al. (2016) and Meinen and Roehe (2018)). Nevertheless, thesestudies provide evidence in favor of both negative and positive co-movement between output and inflation.In this paper, we re-examine the inflation effects of financial shocks using a Bayesian SVAR model by Lanne and Luoto(2020) that exploits the non-normalities in the data. This method allows us to statistically identify the model only basedon the statistical properties of the time series. Given that statistically identified shocks do not have an economic inter-pretation, we use theoretical predictions about the dynamics of the impulse responses and other extraneous informationto label the shocks. Our method stands in contrast to the traditional method of using sign restrictions to identify SVARmodels in two important ways: First, our approach achieves point-identification, whereas the sign identification proce-dure provides a set of admissible models, which are consistent with the restriction and data (Baumeister and Hamilton,nflation Dynamics of Financial Shocks W ORKING P APER
Financial shocks in theoretical models are compatible with both inflationary and disinflationary effects, depending onthe transmission channel and the central banks’ policy rule. In general, the macroeconomic literature finds that finan-cial shocks affect prices through the demand channel. Gertler and Karadi (2011) show that an unexpected decreasein the bank capital leads to a fall in inflation by reducing firms’ and households’ demand for both investment andconsumption. However, the central banks’ response to tensions in the credit markets may offset the disinflationaryeffect of financial shocks. In Cúrdia and Woodford’s (2010) model, a standard Taylor rule that weights output andinflation leads to a positive co-movement between prices and output, whereas a policy rule that takes into account thetightening of private sector borrowing conditions has the opposite effect.More recently, several studies emphasize the role of credit markets in explaining the unexpectedly modest fall in infla-tion during the Great Recession using DSGE models. In Gilchrist et al.’s (2017) model, financially constrained firmsset prices above their marginal cost during financial crises in order to avoid accessing costly external finance, whichleads to opposite movements in inflation and output. Meh and Moran (2010) find that banks’ financial position plays animportant role in explaining macroeconomic fluctuations, given that bank capital mitigates the moral hazard problembetween bankers that finance projects and investors who supply the funding. In their model, an unexpected negativeshock to bank capital leads to a protracted decrease in lending and investment. However, aggregate consumption andinflation slightly increase following the shock, because the subsequent increase in the price of capital goods causehouseholds to increase the consumption of final goods. Financial shocks also behave in the same way as cost-pushshocks in the model of De Fiore and Tristani (2013), where an increase in borrowing costs drive up firms’ marginalcosts, which are subsequently passed to consumers.Theoretical models are also compatible with financial shocks whose inflationary effects depend on the origin of theshock. In Gerali et al.’s (2010) model, a shock that decreases bank capital induces banks to raise the price of lending,which, in turn, reduces firms’ investment demand. However, an increase in firms’ demand for labor drives up wages,and therefore induces a negative co-movement between output and prices. On the other hand, another type of shockoriginating in the financial sector, characterized by an unexpected fall in banks’ loan-to-value ratios, has a negativeeffect on output and inflation .A number of recent empirical studies use SVAR models to identify and estimate the macroeconomic effects of financialshocks and assess their contribution to business cycle fluctuations during the recent economic and financial crises.Sign restrictions have been commonly used to identify financial shocks in these studies. The restrictions are generallybased on the predicted effect of shocks in New Keynesian dynamic stochastic general equilibrium (DSGE) modelsthat include a financial sector. However, inconsistencies in the responses of key variables to financial shocks in DSGEmodels pose an inherent difficulty in assigning sign restrictions. To overcome this problem, sign restrictions havebeen commonly based on features, with which a set of benchmark DSGE models are generally in agreement (see, forexample, (Gambetti and Musso, 2017)).Empirical studies on financial shocks commonly impose a positive co-movement between output and inflation. Thisrestriction is largely motivated mainly by theoretical considerations (Busch et al., 2010; Darracq-Paries and De Santis,2015) but also provides a method to distinguish financial shocks from aggregate supply shocks, which are characterizedas having an opposite effect on output and prices (Furlanetto et al., 2017; Gambetti and Musso, 2017). The restrictionsare commonly imposed on the period during which the shock occurs, but sometimes only on further lags to accountfor the delayed response of real macroeconomic variables. For instance, Busch et al. (2010) impose no restrictions The response of inflation to loan-to-value shocks is based on additional information provided in Gambetti and Musso (2017) W ORKING P APER on the impact response of output and prices but set a positive co-movement for these variables for two periods afterthe initial response. Although the sign of the response of is not restricted before two periods, Busch et al. find thatfinancial shocks also exhibit positive co-movement between output and prices upon impact.In other empirical studies, assumption about the timing of the impact of shocks are used to distinguish real economicshocks, such as shocks affecting aggregate demand or aggregate supply, from financial shocks. (Barnett and Thomas,2014) imposes restrictions on the financial variables on impact following a credit supply shock, but sets the responsesof macroeconomic variables such as prices to be zero on impact and unrestricted for further lags to account for thetime that it takes for financial shocks to be transmitted to the real economy. Moreover, real economic shocks areassumed to have an immediate impact on the output and prices, whereas the impact on financial variables is unrestricted.Peersman (2011) also identifies bank lending shocks by imposing a zero restriction on the real variables on impactand a lagged effect on bank lending, whereas restriction on the financial variables are imposed on impact. Peersman(2011) reports a positive co-movement between output and inflation at longer horizons in the Euro area, whereasBarnett and Thomas (2014) find that expansionary financial shocks have a negative impact on inflation in the UnitedKingdom. Gilchrist and Zakrajšek (2012) impose a recursive structure on the shocks such that unexpected shocks inthe financial markets have no immediate impact on real economic variables. For the United States, they find a non-negligible impact on inflation three quarters after the shock and a positive co-movement between output and inflationat longer horizons.Several recent studies attempt to reconcile the opposing predictions in the theoretical models by remaining agnosticabout the sign of inflation following financial shocks. Abbate et al. (2016) disentangle financial shocks from otherstructural shocks by sign and exclusion restrictions, while imposing no restriction on the sign of the inflation. Theyfind the contractionary financial shock to have inflationary effects, which supports the hypothesis that financial shockspartly explain the missing disinflation following the financial crisis. These findings are somewhat at odds with thoseof Hristov et al. (2012) who observe a negative, albeit non-significant, effect on inflation after expansionary financialshocks in a panel of euro area countries. Moreover, Meinen and Roehe (2018) report an ambiguous effect on inflationfollowing financial shocks in both the euro area and the United States.This paper is related to Brunnermeier et al. (2017) who study the macroeconomic effects of financial stress in a struc-tural VAR model. They similarly use statistical properties of the model to identify the structural shocks, allowing forheteroskedasticity and non-normality of the residuals. Their findings complement our results, in that they succeed inidentifying two distinct shocks that capture tensions in the financial markets that in their effects on prices and creditaggregates. In contrast to Brunnermeier et al., we document negative co-movement between prices and output for onetype of credit-side shock.
The objective of this paper is to study the inflationary impact of financial shocks in the US. To that end, we estimatea Bayesian VAR model with non-normal errors by Lanne and Luoto (2020), who apply the method to the crude oilmarket and monetary policy shocks. The method has also been previously used to study the effects of unconventionalmonetary policy (Puonti, 2019). The model exploits non-normalities in the data to uniquely identify the model, asproposed by Lanne et al. (2017). The advantage of this approach is that restrictions on the model parameters arenot needed for the identification or model inference. However, extraneous information, for example from theoreticalmodels, are needed to give an economic interpretation of the shocks. The labelling of the shocks is discussed in detailin Section 3.4.
The structural VAR model is of the form y t = a + p X l =1 A l =1 y t − l + Bǫ t . (1) y t is an N × vector of endogenous variables. a is an N × vector of constants and A l ( l = 1 , . . . , p ) are the N × N coefficient matrices for the l th lag of the endogenous variables. The N × N matrix B captures the contemporaneousrelations between the structural shocks. The error process ǫ t = [ ǫ ,t , . . . , ǫ n,t ] is a N × vector of independentlydistributed random vectors. Moreover, in contrast to the common assumption of Gaussian errors, we assume that the i th component ǫ i,t , i = 1 , . . . , N follows a t -distribution with degrees of freedom λ i > .3nflation Dynamics of Financial Shocks W ORKING P APER
We assume that the SVAR process is stable such thatdet ( I n − A z − · · · − A p z p ) = 0 , | z | ≤ . (2)and therefore the process has a moving average representation y t = µ + ∞ X j =0 Ψ j Bǫ t − j = µ + ∞ X j =0 Θ j ǫ t − j , (3)where µ = A (1) − a and A ( L ) = I K − A L − · · · − A p L p and the N × N coefficient matrices Θ j can be obtainedrecursively Θ = B Θ j = j X l =1 Θ j − l A l , j = 1 , , . . . , where A l = 0 for l > p . The i th column of Θ j , contains the impulse responses of the endogenous variables to the i thstructural shock ǫ i,t , i = 1 , . . . , N .As shown by Lanne et al. (2017), given the assumptions about the non-normality and independence of error term ǫ t hold matrix and that at most one of the structural shocks is normally distributed, the SVAR model (1) is point-identifiedapart from permutation and scaling of the columns of matrix B . Unique identification is achieved by choosing a singlemodel from n ! observationally equivalent set of models using the algorithm described in Lanne et al. (2017). The model estimation and the choice of priors closely follow Lanne and Luoto (2020), who provide a detailed descrip-tion of the model and the estimation method. Next, we provide an overview of the model priors and the specificationof the model.We operate on the inverse of matrix B , and define a prior for B − such that b ∼ N ( b, V b ) , (4)where b ≡ vec ( B − ) .Moreover, we the prior distribution for each λ i is assumed to be exponential such that λ i ∼ exp ( λ i ) . (5)The prior for a ≡ vec ( A ′ ) is assumed to be multivariate normal: a ∼ N ( a, V a ) . (6)We assume a Minnesota-type prior for the model parameters such that the prior variance for the pq th element of A l ( l = 1 , . . . , p ) is v pq,l = ( κ /l κ ) if p = q and v pq,l = ( κ κ σ p /l κ σ q ) if p = q . The prior variance for the constantterm a is assumed to be ( σ p κ ) .The Minnesota prior for the lagged coefficients of the VAR in Eq. (6) is set up such that κ = 3 , κ = 0 . , κ = 1 . Theprior variance for the constant is assumed to be non-informative such that κ = 100 . Given that the macroeconomicdata enter the model in first differences, we set the first the coefficient for the first lag of each endogenous variableequal to zero. We set the prior mean λ i = 10 for each degree of freedom parameter λ i , such that the prior is close toa normal distribution. The prior for each element in b has mean zero and standard deviation , which results in anon-informative prior for B − . For a detailed description of the identification see Proposition 1 in Lanne et al. (2017) W ORKING P APER
The lag length of the VAR is specified to be two, based on the Akaike information criterion with maximum lag lengthof 10. The results are based on a sample of 1000000 draws after a burn-in of 100000 iterations using the Gibbs samplerdescribed in the appendix of Lanne and Luoto (2020).
The dataset includes quarterly data for the United States, covering the period from 1980Q1 to 2011Q4 . The endoge-nous variables included in the model cover the real and monetary sectors of the economy to capture the role of financialshocks on business cycles: real GDP, consumer price index, lending to households on non-financial corporations, thecomposite lending rate, the three-month Treasury Bill rate. In this section we discuss the economic labelling of the financial shocks. Because the model identification is purelybased on the statistical properties of the data, extraneous information is needed to give the shocks an economic in-terpretation. As discussed above, in theoretical literature, a financial shock is both compatible with a positive (e.g.,Gertler and Karadi, 2011) and a negative (e.g., Gilchrist et al., 2017) co-movement with output and inflation or morethan one financial shock that vary in their inflationary response (Gerali et al., 2010). Based on this observation, incontrast to previous empirical research, we consider the possibility that there exist two financial shocks with oppositeinflationary responses.We proceed to label to shocks by assessing the posterior probability that the structural shocks satisfy a set of inequalityconstraints based on theoretical predictions . To this end, we adopt a set of inequality constraints that have been used inprevious studies that use sign restrictions to identify financial shocks (e.g., Gambetti and Musso, 2017; Hristov et al.,2012). The inequality constraints are summarized in Table 1. The shocks are normalized to have a positive impact onthe lending rate .We are interested in two distinct financial shocks with opposing inflationary responses. Because the inflationaryresponse generally depends on the transmission channel of the shock, we label these shocks as supply side and demandside financial shocks, supply side shocks being those that exhibit negative co-movement between output and inflation,whereas the correlation is positive for demand side shocks.Table 1: Inequality restrictions for identification Variable
Real GDPgrowth Inflation Loan volumegrowth Lendingrate Short-terminterest rateDemand side financial shock ( R ) − − − + − Supply side financial shock ( R ) − + − + − Notes: + / − denote the sign of impact of impulse responses for each variable. The restrictions are imposed on the impulse responseson impact for all variables. As is common in the literature, we assume that recessionary financial shocks increase the composite lending rate anddecrease both the real GDP growth and the growth of bank lending within the same quarter. However, these propertiesare not sufficient to disentangle financial shocks from aggregate supply shocks, which are generally characterized bynegative co-movement between output and prices. Gambetti and Musso (2017) separate financial and aggregate sup-ply shocks by imposing a positive co-movement between inflation and real GDP growth following financial shocks,while only imposing restrictions on the response of inflation and GDP growth to aggregate supply shocks. By con-trast, Hristov et al. (2012) remain agnostic about the sign of consumer price inflation in response to a financial shock.However, in order to distinguish aggregate supply shock from financial shocks, Hristov et al. assign a positive sign for The dataset is obtained from Gambetti and Musso (2017), who also provides a detailed description of the data. The samedataset is used for the comparability of the results. The dataset was downloaded from the Journal of Applied Econometrics DataArchive ( http://qed.econ.queensu.ca/jae/2017-v32.4/ ) Although we assign probabilities to shocks based on the sign of the impulse responses, our leads to point identification, whereasthe traditional sign-restriction approach provides an admissible set of models. The identification is based on the sign pattern of the shocks, and therefore the normalization of the sign of the shock has noreal implication on the identification procedure. W ORKING P APER the short-term interest rate in response to a contractionary aggregate supply shock. This implies that in the absence offinancial tensions the monetary authority increases interest rates in response to higher inflation, despite a fall in output.We assume that the response of the short-term interest rate to financial shocks is negative, which is underpinned by theassumption that the monetary policy authority responds immediately to negative financial shocks independent of theinflationary response. This is consistent with the optimal monetary policy response to financial shocks as described inDe Fiore and Tristani (2013). This also implies that, in addition to its price stability objective, the monetary authorityresponds to contractionary financial shocks by lowering its policy rate despite inflationary pressures.The inference on the structural shocks is based on the posterior distribution of the parameters of interest. The shocksare given an economic label by assessing whether the impulse response functions satisfy a set of inequality constraintsbased on theoretical predictions. In practice, this is done and calculating the posterior probabilities of each pair ofshocks satisfying these constraints .The inequality constraints with respect to two structural shocks are collected in the N × M matrix R m ( m = 1 , ),where N = 5 is the number of variables and M = 5 is the number of constraints imposed on each shock m . Each rowof R m represents one inequality constraint and contains only zeros, and a single entry with or − .Given the constraints described in Table 1, we collect the inequality constraints in the matrices R m , such that R = − − − − and R = − − − . (7)Let B i be the i th column of the matrix B , that is the column of the structural impact matrix corresponding to structuralshock ǫ i,t ( i = 1 , . . . , N ) .The columns of the structural impact multiplier matrix B that satisfy the inequality constraints are then included in theset Q m , such that Q m = { B i : R m B i ≥ M × } . (8)Then, for each pair of structural shocks ǫ i,t and ǫ k,t , for i, k = 1 , . . . , N, and i = k , the posterior probability of theconstrained model conditional on the data isPr (cid:0) B i ∈ Q , B k ∈ Q , B m = i,k ∈ Q C | y (cid:1) , (9)where Q C denotes the complement of the union Q ∪ Q .In practice, we first normalize one row of B in all posterior draws such that the impact effects on the endogenousvariables are compatible with one of the constraints collected in matrix R j . In this case, we normalize the response ofreal GDP growth to be negative on impact for each structural shock. Then the posterior probabilities are calculated bycounting the share of posterior draws for which each of the N ( N − pairs of shocks satisfy the constraints and noother shock satisfies either of the constraints.Based on the posterior probabilities, we then calculate the Bayes factors for each pair of shocks against the uncon-strained model to assess the plausibility of the data supporting the inequality constraints. The Bayes factor is given asthe ratio of the marginal likelihoods of the constrained model ( M r ) against the unconstrained model ( M u ) is given by p ( y | M r ) p ( y | M u ) = R S p ( y | φ ) p ( φ ) dφ R Φ p ( y | φ ) p ( φ ) dφ R S p ( φ ) dφ , (10)where φ is a vector of model parameters p ( φ ) is the prior distribution, and p ( y | φ ) is the marginal likelihood. S denotesthe set of values of φ for which the inequality constraints are satisfied. Although we identify two shocks, the posterior probabilities can be calculated for any number of shocks g , for ≤ g ≤ N , asdescribed in Lanne and Luoto (2020) The inequality constraints can be extended to a longer horizon of the impulse responses by applying constraints on the elementsof Θ j , j > . See Lanne and Luoto (2020). W ORKING P APER
Given that the identification is based on the statistical properties of the time series, it is possible that none, only oneor more than one pair of shocks is consistent with these constraints. If there is no support for any of the shock pairsto satisfy the inequality constraints, we can conclude that there are does not exist two financial shocks with distinctiveinflationary responses. In this case, we assess if a single financial shock with either positive or negative correlationwith output and inflation exists . If the data supports only one pair of shocks over the others, we may infer that thereare two distinct financial shocks with opposite effects on inflation. If more than one pair of shocks are supported bythe data, additional information is needed to distinguish between the shocks of interest.Following Kass and Raftery (1995), we consider a Bayes factor exceeding 3.2 supportive of the constrained model infavor of the unconstrained model. If only one constrained model exceeds the threshold, the pair of structural shocksare labelled accordingly. If the Bayes factor for none of the constrained models exceeds 3.2, the data is not consideredsupportive of the inequality constraints. However, it is also possible that more than one constrained model has Bayesfactor exceeding the threshold. In this case we calculate the Bayes factor comparing the two constrained models. If, inthis case, the Bayes factors exceeds 3.2, we consider that model the one most likely to satisfy the inequality constraints.If the Bayes factors do not exceed the threshold, other criteria are necessary to label the structural shocks. In this section, we discuss the identification of the shocks and their macroeconomic implications on the US economy.
Given that the identification of the shocks depends on the assumption that the structural errors follow a t -distributionand at most one of them is Gaussian, we first examine the posterior densities of the degree of freedom parameters λ i . As can be seen in Figure 1, the posterior distribution of λ i is centered around small values, with the mean rangingbetween 2.22 and 11.7. This implies that the distribution of the structural shocks is indeed heavy-tailed, thus satisfyingthe distributional assumptions for identification.Figure 1: Posterior densities of λ i With the objective of labelling the shocks, we calculate posterior probabilities and corresponding Bayes factors toassess the plausibility that the structural shocks satisfy both of the inequality constraints, as detailed in Table 1. TheBayes factors for the N ( N − pairs of the shocks against the unconstrained SVAR model are reported in Table 2. Model inference for a single shock is discussed in detail in Lanne and Luoto (2020). W ORKING P APER
Table 2: Bayes factors for the combination of shocks satisfying the inequality restrictions R ǫ ǫ ǫ ǫ ǫ R ǫ - 0.01 0 0 0 ǫ ǫ ǫ ǫ Notes:
The values denote Bayes factors for the N ( N − combinations of the columns of B , giventhe inequality constraints R and R . The data provides evidence for two of the shock pairs ((2,3) and (5,3)) satisfying the restrictions with Bayes factorsof 13.3 and 3.6, respectively. Moreover, there is weak or no support for other shock pairs since the Bayes factors forall other combinations are below the threshold of 3.2 . For both of these models, ǫ satisfies the inequality constraintsfor the demand side financial shock and both ǫ and ǫ satisfy the constraints for the demand side financial shock.However, comparison of the two constrained models lends strong support to the shock pair (2,3) over shock pair (5,3)on the basis that the Bayes factor comparing the two constrained model is 3.6. Therefore, we label shocks ǫ and ǫ as the supply side and demand side financial shock, respectively.The data therefore lends support to the existence of two financial shocks that vary in their inflationary response. Thissuggest that financial shocks are transmitted to the real economy via both aggregate demand and aggregate supply chan-nels, which is consistent with predictions in the theoretical literature (see e.g., Gerali et al., 2010). Brunnermeier et al.(2017) also find two distinct shocks relating to financial stress, one relating to unexpected increases in bond spreadsand the other relating to inter-bank lending. However the bond spread shock and the inter-bank shock both lead topositive co-movement between output and prices, although bond spread shock has a more persistent effect on outputand business credit.The existence of these shocks also has implications for the use of sign restrictions to identify financial shocks in struc-tural VAR models. The conventional method of restricting the sign of the response of inflation to financial shocks, asemployed by, for example, Gambetti and Musso (2017) and Barnett and Thomas (2014) will only capture a particulartransmission channel of financial shocks, and might therefore, downplay their macroeconomic importance. Moreover,the decision to remain agnostic about the sign of inflation needs to supplemented with extraneous information that maybe used to disentangle supply and demand side channels of financial shocks. Finally, given that a supply side financialshock and an aggregate supply shock both have opposing effects on prices and output, imposing restrictions only onthese two variables is not sufficient to distinguish between these two shocks. Therefore, for example assumptionsabout the timing of the effects of these two shocks might be considered to disentangle the shocks. Next, we discuss the effects the macroeconomic effects of financial shocks by studying the impulse response functionsto each shock. The impulse responses to demand and supply side financial shocks are presented in Figure 2 andFigure 3, respectively. The solid line is the median and the shaded areas around the median denote the 68% crediblesets. Figure 2 shows that a contractionary demand side financial shock has a negative effect on real GDP and inflation.Real GDP growth falls on impact but the effect of the shock largely dissipates after three quarters. The effect oninflation is more pronounced. Moreover, inflation also decreases immediately, after which the effect gradually beginsto dissipate. Loan volume growth tends to decrease less than growth of real GDP. Loan growth decreases immediatelyfollowing the shock, but recovers within eight quarters following the shock. The short-term interest rate and thelending rate falls for five quarters following the shock, after which it begins to normalize slowly. The results are robust to applying the restrictions to impulse responses four quarters following the shock. W ORKING P APER
Figure 2: Impulse response functions to a demand side financial shock
RealGDP
Consumer priceinflation
Loanvolumes
Lendingrate
Short-terminterest rate
Notes : The solid line is the posterior median and dashed lines are the 16th and 84th percentile of the posteriorprobability. Model is a VAR(2) including a constant. The estimation period is 1980:Q1-2011:Q4As shown in Figure 3, a supply side financial shock also leads to an immediate decline in the real GDP growth andinflation, after which the effects begins to dissipate. Real GDP growth turns positive 8 quarters following the shock,whereas the effect on inflation disappears after a year. Moreover, the decrease in loan volume growth is greater thanthe immediate effect on real GDP growth. Loan volumes decrease on impact and the effect gradually dissipates after20 quarters. The lending rate and the short-term interest rate do not immediately respond to the shock but begin togradually fall, reaching their lowest point after two years, and then slowly recovering from then onwards.Figure 3: Impulse response functions to a supply side financial shock
RealGDP
Consumer priceinflation
Loanvolumes
Lendingrate
Short-terminterest rate
Notes : See notes in Figure 2.Despite the opposing sign of the effect on inflation, the response dynamics to supply side financial shocks are similarto those obtained in related studies, for example Gambetti and Musso (2017) and Mumtaz et al. (2018). However, in9nflation Dynamics of Financial Shocks W ORKING P APER contrast to these studies, we observe both a larger and more persistent effect on loan volumes. On the other hand,demand side financial shocks have a larger effect on inflation in relation to changes in output and the effect on loanvolume growth tends to be smaller compared to what Gambetti and Musso (2017) finds.Figure 4: Forecast error variance decompositions for a supply side financial shock
RealGDP
Consumer priceinflation
Loanvolumes
Lendingrate
Short-terminterest rate
Notes : The solid line is the posterior median and the shaded area is the 68th percentile posterior probability. Model isa VAR(2) including a constant. The estimation period is 1980:Q1-2011:Q4The forecast error variance decomposition of the supply side financial shock is presented in Figure 4. It shows thatsupply side financial shocks initially account for roughly 90 percent of variance in loan volume growth and more than50 percent at the four-year horizon. The contribution to forecast error variance of loan growth at four years’ horizon islarger than Hristov et al.’s (2012) and Abbate et al.’s (2016) estimate of 10 and 44 percent, respectively. Moreover, wefind that the contribution to the forecast error variance of GDP growth is roughly 10 percent. The finding is consistentwith the estimates in other empirical studies that report a contribution to the forecast error variance of output between5 and 20 percent (Abbate et al., 2016; Gilchrist and Zakrajšek, 2012; Hristov et al., 2012; Meeks, 2012; Mumtaz et al.,2018). The effect on inflation is also consistent with earlier studies. The contribution of supply side financial shocks toconsumer price inflation is between 5 and 10 percent at the four-year horizon, which is in the range of 5 to 37 percentcommonly reported in the literature (see e.g., Abbate et al., 2016; Hristov et al., 2012; Gilchrist and Zakrajšek, 2012;Mumtaz et al., 2018).As discussed above, the inequality constraints for supply side financial shocks are also consistent with aggregatesupply shocks as defined in Gambetti and Musso (2017), who only restrict aggregate supply shocks to have opposingeffects on prices and output. Therefore, there is a legitimate concern that financial shocks may be misinterpreted to beaggregate supply shocks. To provide more evidence in support of the identification of the shocks, we turn to forecasterror variance decompositions for additional information.The supply side financial shock accounts for a large share of the variation in loan volume growth at all horizons(see Figure 4). This observation may be interpreted to support of the notion that innovations in loan volume growthoriginate in the financial sector and may therefore also ease worries that the identified supply side shocks are mistakenfor aggregate supply shocks . In a comparison of identification schemes using a Monte Carlo experiment, Mumtaz et al. (2018) find that sign restrictionscombined with a condition that the identified shock maximizes forecast error variance of the quantity of loan supply up to 40quarters, along with a proxy SVAR identification scheme, best match the impulse responses obtained from a DSGE model withfinancial frictions W ORKING P APER
Figure 5: Forecast error variance decompositions for a demand side financial shock
RealGDP
Consumer priceinflation
Loanvolumes
Lendingrate
Short-terminterest rate
Notes : See notes in Figure 4.Figure 5 presents the forecast error variance decomposition for the demand side financial shock. The demand sidefinancial shock accounts for a large share of the variation in inflation but only a small share of the variance of othervariables. The contribution of demand side financial shocks to the forecast error variance of consumer prices is roughly50 percent at the four-year horizon. The effect on inflation is considerably larger than reported in the previous empiricalliterature on financial shocks. However, the findings are consistent with the empirical literature on the macroeconomiceffect of financial shocks. The relatively small contribution to real GDP is consistent with Blanchard and Quah’s(1988) interpretation that demand side shocks have only temporary effect on output. Moreover, Forni and Gambetti(2010) find that a demand side shock explains approximately 50 percent of the forecast error variance of the GDPdeflator at the six-year horizon and under 8 percent of the forecast error variance of the GDP.Next, we discuss the contribution of the shocks to the evolution of consumer price inflation and loan volume growthusing historical decompositions. The historical decompositions for loan volume growth and consumer price inflationare shown in Figure 6 and Figure 8, respectively.Figure 6: Historical decomposition of loan volume growth W ORKING P APER
Figure 7: Historical decomposition of GDP growth
As can be seen in Figure 6, the supply side financial shock ( ǫ ) explains the largest share of the fall in loan volumegrowth at the onset of the financial crisis and contributes most to the increase in loan growth during the recovery.Furthermore, the demand side financial shock seems to have contributed only marginally to the increase in loan volumegrowth during the beginning of the recovery period.Figure 6 shows that both financial shocks have contributed positively to inflation during the recovery period, althoughthe cumulative contribution of supply side financial shocks to inflation is small in relation to demand side financialshocks. Nevertheless, the positive contribution to inflation supports the argument that financial shocks have attenuatedthe disinflationary pressures in the aftermath of the financial crisis (Gilchrist et al., 2017).Figure 8: Historical decomposition of consumer price inflation In this paper, we study the effects of financial shocks on inflation in the United States. We exploit the non-Gaussianfeatures in the time series to uniquely identify the SVAR model, which allows us to assess the inflation effects offinancial shocks. 12nflation Dynamics of Financial Shocks W ORKING P APER
Our results show that the data is supportive of two distinct financial shocks that have opposite inflationary responses.This suggest that a financial shock is transmitted to the real economy via both aggregate demand and aggregate supplychannels. This finding is important from a policy perspective, given that a cost-push shock, which moves output andinflation in opposite directions, poses a trade-off with the central banks’ objective of price and output stability. Thisfinding also has implications for the theoretical work on the transmission of financial shocks. Theoretical modelsthat include the financial sector would benefit from the inclusion of both demand and supply side effects of financialshocks to better capture the effects on the real economy and the impact of monetary policy. Nevertheless, historicaldecompositions suggest that the effect of both financial shocks on consumer prices is small. Therefore, negativefinancial shocks alone do not account for the lack of disinflation in the US during the Great Financial Crisis. Moreover,given that financial shocks seem to exhibit both inflationary and disinflationary effects, as also argued by Gerali et al.(2010), empirically tracing the origins of these different types of financial shocks poses an interesting avenue forfurther research. 13nflation Dynamics of Financial Shocks W ORKING P APER
Acknowledgements
I gratefully acknowledge financial support from Academy of Finland (grant 308628), the Yrjö Jahnsson Foundation(grant no. 6609, 7016), the Foundation for the Advancement of Finnish Securities Markets, and the Nordea BankFoundation. I would also like to thank Markku Lanne and Jani Luoto for their advice in the preparation of themanuscript.
References
Abbate, A., Eickmeier, S., and Prieto, E. (2016). Financial shocks and inflation dynamics. Working paper, DeutscheBundesbank.Barnett, A. and Thomas, R. (2014). Has Weak Lending and Activity in the UK been Driven by Credit Supply Shocks?
The Manchester School , 82(S1):60–89.Baumeister, C. and Hamilton, J. D. (2015). Sign Restrictions, Structural Vector Autoregressions, and Useful PriorInformation.
Econometrica , 83(5):1963–1999.Blanchard, O. J. and Quah, D. (1988). The Dynamic Effects of Aggregate Demand and Supply Disturbances.
NationalBureau of Economic Research Working Paper Series , No. 2737.Brunnermeier, M., Palia, D., Sastry, K. A., Sims, C. A., et al. (2017). Feedbacks: financial markets and economicactivity.
Manuscript, Princeton University .Busch, U., Scharnagl, M., and Scheithauer, J. (2010). Loan supply in Germany during the financial crisis. Technicalreport.Cúrdia, V. and Woodford, M. (2010). Credit Spreads and Monetary Policy.
Journal of Money, Credit and Banking ,42(s1):3–35.Darracq-Paries, M. and De Santis, R. A. (2015). A non-standard monetary policy shock: The ECB’s 3-year LTROsand the shift in credit supply.
Journal of International Money and Finance , 54:1–34.De Fiore, F. and Tristani, O. (2013). Optimal Monetary Policy in a Model of the Credit Channel*.
The EconomicJournal , 123(571):906–931.Forni, M. and Gambetti, L. (2010). Macroeconomic Shocks and the Business Cycle: Evidence from a StructuralFactor Model. Technical Report 040.Furlanetto, F., Ravazzolo, F., and Sarferaz, S. (2017). Identification of financial factors in economic fluctuations.
TheEconomic Journal .Gambetti, L. and Musso, A. (2017). Loan supply shocks and the business cycle.
Journal of Applied Econometrics ,32(4):764–782.Gerali, A., Neri, S., Sessa, L., and Signoretti, F. M. (2010). Credit and Banking in a DSGE Model of the Euro Area.
Journal of Money, Credit and Banking , 42(s1):107–141.Gertler, M. and Karadi, P. (2011). A model of unconventional monetary policy.
Journal of Monetary Economics ,58(1):17–34.Gilchrist, S., Schoenle, R., Sim, J., and Zakrajšek, E. (2017). Inflation Dynamics during the Financial Crisis.
AmericanEconomic Review , 107(3):785–823.Gilchrist, S. and Zakrajšek, E. (2012). Credit Spreads and Business Cycle Fluctuations.
American Economic Review ,102(4):1692–1720.Hristov, N., Hülsewig, O., and Wollmershäuser, T. (2012). Loan supply shocks during the financial crisis: Evidencefor the euro area.
Journal of International Money and Finance , 31(3):569 – 592. Financial Stress in the Eurozone.Kass, R. E. and Raftery, A. E. (1995). Bayes Factors.
Journal of the American Statistical Association , 90(430):773–795.Lanne, M. and Luoto, J. (2020). Identification of Economic Shocks by Inequality Constraints in Bayesian StructuralVector Autoregression.
Oxford Bulletin of Economics and Statistics , 82(2):425–452.Lanne, M., Meitz, M., and Saikkonen, P. (2017). Identification and estimation of non-gaussian structural vectorautoregressions.
Journal of Econometrics , 196(2):288 – 304.Meeks, R. (2012). Do credit market shocks drive output fluctuations? Evidence from corporate spreads and defaults.
Journal of Economic Dynamics and Control , 36(4):568–584.14nflation Dynamics of Financial Shocks W ORKING P APER
Meh, C. A. and Moran, K. (2010). The role of bank capital in the propagation of shocks.
Journal of EconomicDynamics and Control , 34(3):555–576.Meinen, P. and Roehe, O. (2018). To sign or not to sign? on the response of prices to financial and uncertainty shocks.
Economics Letters , 171:189 – 192.Mumtaz, H., Pinter, G., and Theodoridis, K. (2018). What do vars tell us about the impact of a credit supply shock?
International Economic Review , 59(2):625–646.Peersman, G. (2011). Bank Lending Shocks and the Euro Area Business Cycle. Working paper, Ghent University,Faculty of Economics and Business Administration.Puonti, P. (2019). Data-driven structural BVAR analysis of unconventional monetary policy.