An Extended Macro-Finance Model with Financial Factors: Technical Appendix

An Extended Macro-Finance Model with Financial Factors: Technical Appendix June 1, 010 Abstract This technical appendix provides some more technical comments concerning the EMF model, used in the paper " An Extended Macro-Finance Model with Financial Factors". We discuss consecutively the following features: (i) Model speci cation and more speci cally transition and measurement equation; (ii) the econometric methodology including likelihood and priors and (ii) additional results including the tables containing moments of the prior and posterior distributions and impulse response functions. 1

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 1 1 The Model In this section we brie y discuss the state space system estimated in the paper. This state space representation introduces the restrictions that allow us to incorporate stochastic trends into the analysis. This appendix focusses on the speci cation of the transition equation. The identi cation restrictions for money market and return generating factors and the extension of the transition equation (under no-arbitrage) to the yield curve are extensively discussed in the paper. 1.1 The transition equation The state space incorporates a state vector combining observable macroeconomic variables - in ation ( t ); output gap (y t ) and the policy interest rate (i cb t ), collected in the vector X M t = [ t ; y t ; i cb t ] 0 - with a set of latent variables. Depending on their dynamics, we distinguish two types of latent variables: three stationary latent variables l t = [l 1;t ; l ;t ; l ;t ] 0 and two stochastic trends, t = [ 1;t ; ;t ] 0 : The latent variables are collected in the vector X L t = [ l 0 t; 0 t ] 0 : We assume a dynamics for the state variables, X t = [ X M0 t ; Xt L0 ] 0 ; which incorporates following characteristics: (i) the set of stochastic trends, t ; are exogenous and independent, (ii) the stochastic trends are the main drivers of the macroeconomic variables and determine long-run expectations, i.e. lim s!1 X M t = T D t and (iii) we allow for interactions between the stationary latent factors, l t ; and the macroeconomic state. The state space dynamics, consistent with these restrictions can be summarized by a standard VAR(I) representation: X M t l t t 5 = 0 C l 0 5 + MM Ml 0 Xt M 1 I MM Ml 0 T D lm ll 0 l t 1 5 5 + lm I ll 0 0 5 5 t 0 0 I t 1 0 0 0 0 D MM 0 0 " M t " M t + D lm D ll 0 " l t 5 5 ; " l t v MV N(0; I) (1) 5 0 0 S " t " t with [" M0 t ; " t l0 ; " t ] 0 = [" ;t ; " y;t ; " i cb ;t; " l1;t; " l;t; " l;t; " ;t; " ;t] 0, D MM diagonal and and D ll lower triangular, S

An Extended Macro-Finance Model with Financial Factors (Technical appendix) T D = 1 0 0 0 1 1 5 : After some elementary computations, the system can be rewritten as: X M t l t t 5 = C M C l 0 5 + MM Ml M lm ll l 0 0 I 5 X M t 1 l t 1 t 1 5 + D MM 0 D M D lm D ll D l 0 0 S 5 " M t " l t " t 5 () By calling ~ the block of the feedback matrix that refers to the stationary variables [Xt M0 ; lt] 0 0, i.e.: ~ = MM Ml lm ll 5 ; we can express the identi cation restrictions making sure that the stochastic endpoints are de ned as the long-run expectations of the observable variables (see section..1 of the paper) as: M l CM C l DM D l 5 = (I ) ~ T D 0 5 = (I ) ~ 0 1 C l 1 5 = (I ) ~ T D 0 5 ; 5 ; 5 S : Finally, we impose restrictions on ~ by restricting all eigenvalues below one such that the system reverts to the long-run equilibrium implied by the stochastic trends: lim s!1 E t X M s = T D t lim s!1 E t [l s ] = C l 1

An Extended Macro-Finance Model with Financial Factors (Technical appendix) Econometric methodology We use standard Bayesian estimation techniques to estimate the model, consisting of equations (1) and (??) (see below). The vector containing the estimated parameters of the model is denoted by. The posterior of the parameters ; p( j Z T ), is identi ed through Bayes rule: p( j Z T ) = L( j ZT )p() p(z T ; () ) with Z T the data set, L() the likelihood function, p() the priors and p(z T ) the marginal density of the data. In this section (i) we outline the procedure used to nd the mode of the posterior distribution, (ii) we discuss in more detail the prior distributions and (iii) present the measurement equation together with the information variables..1 Mode of the posterior We found the mode of the posterior distribution by selecting the optimal posterior density value out of a set of 1000 optimizations. More speci cally, we ran for 1000 times the following algorithm: 1. Select an initial the initial value, 0 as: 0 = x p + " t ; " t v N(0; V ) with p the priors mean of the parameters and V = diag(0:1 max (j m j ; 0:001)). set 0 opt = 0. for j = 1,..., (a) starting from j 1 opt nd j sa = argmax run the enhanced simulating annealing procedure of Siarry et al. (199) to f() (b) starting from j sa run a simplex algorithm in order to nd j simp = argmax (c) starting from j simp run a Newton-Raphson optimization j NR = argmax (d) starting from j NR run the enhanced simulating annealing procedure of Siarry et al. (199) to nd j opt = argmax f() f() f()

An Extended Macro-Finance Model with Financial Factors (Technical appendix) This procedure was repeated 1000 times. We selected the j NR with the highest posterior density out of these 1000 simulations as the posterior mode.. Priors densities Table 1 lists the speci c prior distributions used for the parameters contained in (see section of the paper):we impose normal priors on all the feedback parameters : The priors related to observable macroeconomic variables, MM, are based on preliminary regression analysis. In particular, we introduce signi cant inertia in the macroeconomic variables (mean auto-regressive parameters between 0.5 for in ation and 0.95 for the output gap). Univariate analysis of the Libor, T-bill and TED spreads suggests that the spreads contain signi cant inertia. This inertia is taken into account in the prior for the feedback of nancial factors by assuming mean autoregressive parameters in LL of 0.. Loose, zero-mean priors are used for most of the o -diagonal elements of. This is the case for the feedback of macroeconomic variables on liquidity and return-forecasting factors, lm, and for the interaction across nancial factors. In line with theory, a negative feedback from spread and return-forecasting factors to macroeconomic variables (i.e. ML ) is imposed N ( 0:5; 0:5), modeling an a priori de ationary impact of liquidity and risk-aversion shocks. Finally, we assume the standard negative feedback from the policy rate to in ation and output gap N ( 0:5; 0:5); a positive impact of output gap on in ation N (0:1; 0:5), and, in line with the Taylor rule, a positive impact of in ation N (0:5; 0:5) and output gap N (0:1; 0:5) on the policy interest rate. The priors with respect to the impact matrix D = S in equation (1) are standard. In particular, we assume an inverted gamma distribution for the diagonal components. Depending on the type of variable, we opt for di erent parameterizations. An important modeling assumption in this respect is the relatively tight prior for the standard deviations of the stochastic trends, S ; IG(0:00; 0:). 1 This choice re ects the belief that the stochastic trends, in line with long-run expectations, move smoothly over time. The priors for the o -diagonal elements of D, contained in D lm ; D MM and D ll, are assumed to be N (0; 0:0). This choice leaves substantial freedom in modeling the covariance between the respective shocks. A crucial set of uniform priors is imposed on the measurement errors. This set of priors aims at facilitating the identi cation of the latent factors by imposing small measurement errors for certain variables in the 1 The rst parameter of the Inverse Gamma distribution refers to the mean of the distribution while the second is the standard deviation.

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 5 measurement equation. In particular, we assume zero measurement error for observable macroeconomic variables M and for the convenience yield, implying cy = 0: The latter assumption implies that the T-bill spread factor, l 1;t ; becomes observable. Second, we impose an upper bound of 0 basis points on the standard deviation of the measurement errors of the Eurodollar - T-Bill spread. These distributional assumptions allow for the identi cation of the credit-related spread factor, l ;t ; which otherwise becomes excessively volatile.. Likelihood and Measurement equation As stated in the paper, we obtain the likelihood function by means of the prediction error decomposition. Obviously, given some of the variables are latent, we use standard Kalman lter procedures to extract the latent factors. The prediction error decomposition is conditioned on the measurement equation, containing the information variables, collected in the vector Z t, and in linearly related to the state vector X t. Z t = A m () + B m ()X t + " m t ; " m t N(0; m 0 m) () A m () = A 0 M ; A0 ; A 0 T b ; A0 y; A 0 s; A 0 cc; A 0 cy 0 ; Bm () = B 0 M ; B0 ; B 0 T b ; B0 y; B 0 s; B 0 cc; B 0 cy 0 ; m = diag [ M ; ; T b ; y ; s ; cc ; cy ] ; where the observable vector, Z t = [ t ; y t ; i cb t ; g t ; y t (1=); y t (1=);..., y t (10); s t (1); s t (10); i Libor t y t (1=); i Eurodollar t y t (1=); i GC repo t y t (1=)], is divided in four blocks: 1. Macroeconomic block The macroeconomic contains in ation, t, output gap, y t ; policy interest rate, i cb t, and trend growth rate of potential output, g t as information variables. We assume perfect updating, implying zero measurement errors.. Yield curve block We include government bond yields spanning maturities between one quarter and ten years. The measurement equation loadings for the yield curve, A y and B y ; are obtained by imposing the noarbitrage conditions listed in the paper: The derivation of these loadings is analogous to Ang and Piazzesi (00).

An Extended Macro-Finance Model with Financial Factors (Technical appendix). In ation survey block We use survey data of the expected average one- and ten-year ahead in ation rates in the observation vector, i.e. s t (1) and s t (10): The respective measurement equation loadings, A s and B s ; can be derived given the transition equation. In particular, solving equation (1) for the in ation expectations, we obtain the respective loadings (see section of the paper) as (see Dewachter (008)): A s(m) = 1 m e mx 1 j=0 A (j) ; B s(m) = 1 m e mx 1 j=0 B (j) ; s 0 (5) and A (j) = A (j 1) + C; B (i) = B (i 1) ; () with e = [1; 0 1 ] and initial conditions A (0) = 0 and B (0) = I 8.. Money market information. Next to the federal funds rate and the yield curve, three money market interest rates are included, i.e. the -month Libor, the -month Eurodollar and the -month GC repo rates. The actual loadings are discussed in the paper. The logarithm function that we maximize, f(x); has the following expression: f(x) = NT log 1 TX log jf t j t=1 1 TX t=1 vt 0 log F 1 t vt with: N = number of measured variables T = total number data points for observable variable v t = Z t [A m + B m (C + X t 1jt 1 )] F t = B m ( t 1jt 1 + DD 0 )B 0 m + m 0 m

An Extended Macro-Finance Model with Financial Factors (Technical appendix) with X t 1jt 1 and t 1jt 1 obtained via the standard Kalman lter recursions: X tjt = C + X t 1jt 1 + K t ([Z t [A m + B m (C + X t 1jt 1 )]]) () tjt = ( t 1jt 1 + DD 0 ) K t B t ( t 1jt 1 + DD 0 ) (8) K t = ( t 1jt 1 + DD 0 )B 0 tf 1 t (9) where C, and D refer to equation () of the paper and X 0j0 has been estimated and 0j0 has been obtained by iterating forward the Riccati equations (), (8) and (9) : Estimation results This section collects the tables containing (i) the description of the posterior densities and (ii) the variance decomposition of the macroeconomic, money market and return forecasting factors. Given the reduced form nature of the transition equation, it is not very informative to discuss the posterior densities in detail. For completeness, however, we include Table ; containing the posterior for the parameters in ; Table containing the posterior for the impact matrix D and Table 5 summarizing the posterior densities for the remaining parameters, including prices of risk and measurement errors. Finally, Table and Figures 1 and allows us to identify the impact of the respective shocks. In particular, the IRFs allow us to verify the identi cation (labeling) of the shocks (see section. note 19). We di erentiate between three types of shocks: macroeconomic, money market and risk premium shocks. Within the class of macroeconomic shocks, we distinguish ve types: three transitory shocks - supply (" ), demand (" y ) and monetary policy (" i ) shocks- and two permanent shocks - long run in ation/in ation target (" ) and an equilibrium growth rate (" ) shock. The responses of the three transitory shocks (Figure 1 panels (a) to (c)) are in line with a structural interpretation of supply, demand and policy rate shocks. Two quali cations should be kept in mind, however: the in ation response to the demand shocks is imprecisely estimated and we observe a price puzzle in the response to a policy shock. The two permanent shocks (Figure, panels (c) and (d)) generate the required permanent e ects. An increase in the in ation target triggers a permanent increase in in ation and interest rate and generates substantial transitory expansionary e ects. Increases in the equilibrium growth rate lead to transitory expansionary e ects in in ation and output and a permanent increase in the interest rate (due to the higher natural

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 8 real interest rate). The IRFs identify two types of money market shocks, which we label respectively as ight to quality and credit crunch shocks. The ight to quality shocks primarily impact on the convenience yield (T-bill spread) while credit crunch shocks a ect the money market spread (Libor spread). The ight to quality shocks typically generate a decrease in government bond yields, a decrease in in ation and a monetary policy easing. Typical credit crunch shocks, increasing the money market spread, have stag ationary e ects on the economy. Somewhat controversial is the response of the policy rate, which (countering higher in ation) increases. Finally, by construction, return-generating factor shocks are neutral with respect to the macroeconomy and the money market. References Dewachter, H., "Imperfect information, macroeconomic dynamics and the yield curve: an encompassing macro- nance model", NBB Working Paper, 1 ( 008). Siarry, P., G. Berthiau, F. Durdin, and J. Haussy (199): Enhanced simulated annealing for globally minimizing functions of many-continuous variables, ACM Transactions on Mathematical Software, (), 09 8.

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 9 Table 1: Prior distributions of the parameters Distr. M ean Std. Dev. Distr. M ean Std. Dev. MM (1; 1) N 0.500 0.50 D ll (j; i) j > i N 0.000 0.00 MM (; 1) N 0.000 0.500 D ll (j; j) j = 1; IG 0.010.000 MM (; 1) N 0.50 0.50 D ll (j; j) j = IG 0.00.000 MM (1; ) N 0.100 0.500 S ; (j; j) j = 1; IG 0.00 0.00 MM (; ) N 0.950 0.50 0 (j) j = 1; :::; 8 N 0.000 0.000 MM (; ) N 0.100 0.500 1 (j; ) j = 1; :::; 8 N 0.000 50.000 MM (1; ) N -0.50 0.50 m (j; j) j = ; :::; 1 U 0.000 0.005 MM (; ) N -0.50 0.50 m (j; j) j = 1 U 0.000 0.00 MM (; ) N 0.800 0.50 A(j) j = N 0.000 0.010 lm (j; i) j; i = 1; ; N 0.000 0.500 A(j) j = 1 N 0.000 0.00 Ml (j; i) j = 1; i = 1; N -0.50 0.50 B(j) j = N 1.000 0.500 Ml (j; i) j = ; ; i = 1; N -0.50 0.500 X 0 (j) j = ; 5 U -0.050 0.050 ll (j; j) j = 1; ; N 0.00 0.500 X 0 (j) j = U -0.100 0.00 ll (j; i) j = i N 0.000 0.500 X 0 (j) j = ; 8 U -0.010 0.050 D MM (j; j) j = 1; ; IG 0.010.000 C (j) j = ; 5 U 0.000 0.015 D MM (j; i) j > i N 0.000 0.00 C (j) j = U 0.000 0.00 D lm (j; i) j; i = 1; ; N 0.000 0.00 M ean= Lower bound, Std. D ev.= Upper bound Notes: These two panels report the priors density for the parameters estimated in the extended Macro-Finance model. N stands for Normal, IG for Inverse Gamma and U for Uniform. The paremeters contained in the table refer to the following state space system: Where the observable and state vectors are Z t = A m + B mx t + " m t ; " m t N(0; m 0 m) (Meas. Eq.) X t = C + X t 1 + S" t; " t N(0; I) (Trans. Eq.) Z t = [ t; y t; i cb t ; gt; yt(1=); :::; yt(10); st(1); st(10); ilibor t X t = [ t; y t; i cb t ; l 1;t ; l ;t ; l ;t ; 1;t ; ;t ] and the parameters of the state equation are given by: C M MM Ml M C = C l 5 ; = lm ll l 0 0 0 I 5 ; D = S = i cb t ; ieurodollar t D MM 0 D M D lm D ll D M 0 0 S with C M C l = I MM Ml 01 lm ll C 1 l Finally the parameters 0 and 1 are related to the stochastic discount factor used for pricing the government bonds: with i t = y t(1=) and t = 0 + 1 X t M t+1 = exp( i t 1 tss0 0 t ts" t+1 ): i cb t ]0 5

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 10 Table : Variance Decomposition State Variables In ation Shocks Supply Demand Mon. pol. Flight to qual. Credit cr. Risk premia In. target Eq. gr. rate 1Q 99.90% 0.00% 0.00% 0.00% 0.00% 0.00% 0.10% 0.00% Q 98.% 0.00% 0.00% 0.9% 0.9% 0.00% 0.5% 0.00% Q 95.1% 0.00% 0.05% 1.1%.9% 0.00% 0.% 0.01% 10Q 88.05% 0.0% 1.0%.1% 5.9% 0.00%.% 0.0% 0Q 8.1% 0.0% 1.%.% 5.0% 0.00%.9% 0.0% 100Q.% 0.01% 1.0% 1.5%.% 0.00%.5% 0.0% Output gap Shocks Supply Demand Mon. pol. Flight to qual. Credit cr. Risk premia In. target Eq. gr. rate 1Q 1.5% 98.9% 0.00% 0.00% 0.00% 0.00% 0.0% 0.00% Q.% 9.99%.8% 0.5% 0.0% 0.00% 0.15% 0.0% Q 5.0% 8.0% 11.1% 0.5% 0.% 0.00% 0.5% 0.0% 10Q 1.8% 50.91%.0% 1.1% 5.1% 0.00% 1.9% 0.19% 0Q.05% 8.% 9.05% 5.0% 10.8% 0.00%.95% 0.5% 100Q.10% 8.15% 9.0% 5.% 10.90% 0.00%.95% 0.5% Fed rate Shocks Supply Demand Mon. pol. Flight to qual. Credit cr. Risk premia In. target Eq. gr. rate 1Q.9% 0.% 9.8% 0.00% 0.00% 0.00% 0.0% 0.00% Q.% 1.5% 89.50%.9% 1.8% 0.00% 0.01% 0.00% Q.1%.81% 5.% 9.0%.9% 0.00% 0.0% 0.0% 10Q 10.%.% 0.% 1.8% 10.89% 0.00% 0.% 0.8% 0Q 10.08%.9% 8.95% 11.% 9.% 0.00% 1.%.0% 100Q.%.1%.0% 8.5%.% 0.00% 9.9% 5.9% T-bill spread Shocks Supply Demand Mon. pol. Flight to qual. Credit cr. Risk premia In. target Eq. gr. rate 1Q.10% 0.9% 5.0%.% 0.00% 0.00% 0.0% 0.0% Q 5.% 0.5% 58.%.5% 0.10% 0.00% 0.80% 0.08% Q 8.58% 1.5% 5.9% 9.% 1.55% 0.00% 1.% 0.1% 10Q 1.%.% 51.0%.15%.9% 0.00% 1.8% 0.15% 0Q 1.01%.95% 9.%.80% 5.9% 0.00% 1.8% 0.15% 100Q 1.0%.95% 9.%.80% 5.9% 0.00% 1.8% 0.15% Libor spread Shocks Supply Demand Mon. pol. Flight to qual. Credit cr. Risk premia In. target Eq. gr. rate 1Q 0.0% 0.9%.% 15.91% 1.% 0.00% 0.00% 0.00% Q 0.% 1.%.5% 1.1% 1.9% 0.00% 0.00% 0.00% Q 1.8%.9% 1.8% 1.0% 0.50% 0.00% 0.01% 0.00% 10Q.5%.99% 1.5% 1.9% 5.5% 0.00% 0.1% 0.01% 0Q.18% 5.0%.08% 1.% 5.8% 0.00% 0.% 0.0% 100Q.0% 5.0%.0% 1.% 5.8% 0.00% 0.% 0.0% Return-forecasting factor Shocks Supply Demand Mon. pol. Flight to qual. Credit cr. Risk premia In. target Eq. gr. rate 1Q.5% 0.5% 18.% 8.% 0.0%.9% 0.00% 0.00% Q 5.8% 0.% 1.1% 8.9% 0.0%.% 0.00% 0.00% Q.9% 0.% 15.0% 9.0% 0.05% 9.9% 0.00% 0.00% 10Q.0% 0.% 1.81% 9.% 0.0%.8% 0.00% 0.00% 0Q.8% 0.% 11.5% 8.% 0.0%.9% 0.00% 0.01% 100Q.8% 0.5% 11.% 8.% 0.0%.9% 0.00% 0.01% Notes: This table reports the forecasting error variance decomposition of the state variables, computed at the mode of the posterior distribution of the parameters. Mon. pol. stands for Monetary policy, Flight to qual. for Flight to quality, Credit cr. for Credit crunch, In. target for In ation target and Eq. gr. rate for Equilibrium growth rate.

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 11 Table : Prior and Posterior Distribution of the Phi Matrix Prior Posterior Distr. Mean Std.Dev. 0.5 % 5 % 50 % 95 % 99.5 % Mode Mean (1; 1) N 0.500 0.50 0.11 0. 0.91 0.58 0.558 0.90 0.89 (; 1) N 0.000 0.500-0.085-0.0-0.008 0.05 0.08-0.01-0.00 (; 1) N 0.50 0.50 0.11 0.15 0.1 0.1 0.1 0.150 0.18 (; 1) N 0.000 0.500 0.01 0.0 0.09 0.118 0.1 0.101 0.098 (5; 1) N 0.000 0.500-0.01-0.011 0.00 0.05 0.059 0.015 0.0 (; 1) N 0.000 0.500-0.8-0.1-0.180-0.1-0.18-0.18-0.181 (1; ) N 0.100 0.500 0.0 0.09 0.05 0.09 0.099 0.08 0.0 (; ) N 0.950 0.50 0.81 0.88 0.91 0.95 0.9 0.9 0.91 (; ) N 0.100 0.500 0.11 0.11 0.11 0.1 0.18 0.1 0.11 (; ) N 0.000 0.500 0.01 0.01 0.0 0.05 0.08 0.0 0.0 (5; ) N 0.000 0.500-0.095-0.09-0.0-0.050-0.0-0.05-0.00 (; ) N 0.000 0.500-0. -0. -0. -0.1-0.1-0. -0. (1; ) N -0.50 0.50 0.015 0.0 0.0 0.101 0.110 0.0 0.0 (; ) N -0.50 0.50-0.11-0.101-0.09 0.01 0.0-0.05-0.0 (; ) N 0.800 0.50 1.058 1.01 1.08 1.099 1.101 1.0 1.0 (; ) N 0.000 0.500 0.0 0.09 0.10 0.18 0.1 0.10 0.098 (5; ) N 0.000 0.500-0.00-0.05-0.0 0.00 0.009-0.018-0.05 (; ) N 0.000 0.500-0.18-0.11-0.1-0.09-0.08-0.110-0.18 (1; ) N -0.50 0.50-0.159-0.19 0.0 0.185 0.1 0.00 0.018 (; ) N -0.50 0.500-0.59-0.558-0.8-0.0-0.10-0.58-0.8 (; ) N -0.50 0.500-0.1-0.00-0.5-0.81-0.1-0.5-0.59 (; ) N 0.00 0.500 0.59 0.8 0.8 0.8 0.50 0.9 0.9 (5; ) N 0.000 0.500 0.009 0.01 0.15 0.1 0.8 0.11 0.15 (; ) N 0.000 0.500 0.51 0.58 0.8 0.8 0.811 0. 0.1 (1; 5) N -0.50 0.50 0.8 0.00 0. 0.585 0.10 0.0 0. (; 5) N -0.50 0.500-0.15-0.19 0.00 0.18 0.19-0.00-0.00 (; 5) N -0.50 0.500 0.01 0.0 0. 0. 0.8 0.1 0. (; 5) N 0.000 0.500-0.15-0.15-0.11-0.00-0.05-0.100-0.11 (5; 5) N 0.00 0.500 0.5 0.8 0.5 0. 0.59 0.0 0.55 (; 5) N 0.000 0.500-0.8-0. -0.9-0.5-0.58-0.18-0.0 (; ) N 0.00 0.500 0. 0.85 0.8 0.891 0.90 0.8 0.8 Notes: This table reports the priors and the posterior density for the parameters of matrix in Eq. 1. The rst three columns report the distributions, means and standard deviations of the prior distributions. The fourth to the eight columns report the.5-th, 5-th, the 50-th and the 95-th 99.5-th percentile of the posterior distributions, respectively. The last two columns report the modes and the means of the posterior distributions. All results were obtained using the Metropolis Hastings algorithm.

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 1 Table : Prior and Posterior Distribution of the Impact Matrix Prior Posterior Distr. Mean Std. Dev. 0.5 % 5 % 50 % 95 % 99.5 % Mode Mean D(1; 1) IG 0.010.000 1.05 1.0 1.1 1. 1.9 1.15 1.1 D(; 1) N 0.000 0.00-0.1-0.18-0.09 0.00 0.0-0.09-0.09 D(; 1) N 0.000 0.00 0.05 0.05 0.10 0.1 0.18 0.11 0.09 D(; 1) N 0.000 0.00-0.01-0.01 0.05 0.10 0.1 0.05 0.0 D(5; 1) N 0.000 0.00 0.00 0.01 0.0 0.1 0.1 0.08 0.0 D(; 1) N 0.000 0.00-0.1-0.1-0.05 0.0 0.05-0.0-0.0 D(; ) IG 0.010.000 0. 0.5 0.1 0. 0.9 0.9 0.1 D(; ) N 0.000 0.00 0.01 0.0 0.0 0.10 0.11 0.0 0.0 D(; ) N 0.000 0.00-0.09-0.08-0.0 0.0 0.0-0.0-0.0 D(5; ) N 0.000 0.00-0.1-0.11-0.05 0.01 0.0-0.05-0.05 D(; ) N 0.000 0.00-0.01-0.01 0.0 0.0 0.0 0.0 0.0 D(; ) IG 0.010.000 1.19 1.1 1. 1. 1. 1.0 1. D(; ) N 0.000 0.00 0.8 0.9 0.5 0.5 0.5 0. 0. D(5; ) N 0.000 0.00-0.1-0.0-0.1-0.0-0.0-0.1-0.1 D(; ) N 0.000 0.00-1. -1.8-1.8-1.11-1.05-1.9-1.1 D(; ) IG 0.010.000 0.5 0. 0.1 0. 0. 0.0 0.1 D(5; ) N 0.000 0.00-0.15-0.1-0.0 0.0 0.0-0.0-0.0 D(; ) N 0.000 0.00-0.5-0.55-0. -0. -0. -0. -0. D(5; 5) IG 0.010.000 0. 0.8 0. 0.8 0.9 0.1 0. D(; 5) N 0.000 0.00 0.0 0.0 0.0 0.51 0.55 0. 0.1 D(; ) IG 0.00.000 1.18 1.1 1. 1.1 1. 1.5 1.1 D(; ) IG 0.00 0.00 0.19 0.19 0.1 0. 0.5 0.1 0.1 D(8; 8) IG 0.00 0.00 0.0 0.0 0.05 0.0 0.0 0.05 0.05 Notes: This table reports the priors and the posterior density for the parameters of S matrix in Eq. 1. The rst three columns report the distributions, means and standard deviations for the prior distributions. The fourth to the eight columns report the.5-th, 5-th, the 50-th and the 95-th 99.5-th percentile of the posterior distributions, respectively. The last two columns report the modes and the means of the posterior distributions. All the statistics of all the posterior distribution are multiplied by 100. The results were obtained using the Metropolis Hastings algorithm.

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 1 Table 5: Prior and Posterior Distribution of the Other Parameters Prior Posterior Distr. Mean StdDev 0.5 % 5 % 50 % 95 % 99.5 % Mode Mean 0 (1) N 0.000 0.000-9.900-9.55 -.0-5.199 -.9 -.5 -.59 0 () N 0.000 0.000.99.0 5.95.985 8...00 0 () N 0.000 0.000 0.1 0.1 0.585 0.899 0.9 0.5 0.50 0 () N 0.000 0.000-0.198-0.0 0.5 0.85 0.98 0.5 0.51 0 (5) N 0.000 0.000 0.9 0.80.1.18.8 1.895 1.9 0 () N 0.000 0.000-0.0-0.5-0.8 0.05 0.08-0.5-0. 0 () N 0.000 0.000 0.0 0.8.11.1.9 1.8.1 0 (8) N 0.000 0.000-8.85 -.90 0.1.580.59-0.90-0.5 1 (1; ) N 0.000 50.000 1. 1..10 0. 5.0 8. 5.90 1 (; ) N 0.000 50.000-9.8-5.88-1.0.0 1.59.0 -.0 1 (; ) N 0.000 50.000-1.99-1.5 -.5.18.18 -.00 -.98 1 (; ) N 0.000 50.000 1.89 15.591.1.11 8.1 5.00. 1 (5; ) N 0.000 50.000-5.91-50.9-5.59-0. -18. -. -5.19 1 (; ) N 0.000 50.000 1.05 1. 19.15..1 18.9 19.1 1 (; ) N 0.000 50.000-9.19-8. -.99 -.5 -. -.885 -.1 1 (8; ) N 0.000 50.000-108.1-95.5-9.5 0.01 55. -1.90-9.9 m (; ) U 0.000 0.005 0.001 0.001 0.00 0.010 0.01 0.001 0.005 m (5; 5) U 0.000 0.005 0.1 0. 0. 0.05 0.1 0.8 0.8 m (; ) U 0.000 0.005 0.19 0.01 0. 0.5 0.5 0.1 0.5 m (; ) U 0.000 0.005 0. 0.1 0.5 0.9 0.00 0. 0. m (8; 8) U 0.000 0.005 0.101 0.10 0.115 0.19 0.11 0.115 0.115 m (9; 9) U 0.000 0.005 0.00 0.00 0.010 0.0 0.01 0.001 0.01 m (10; 10) U 0.000 0.005 0.0 0.11 0. 0.59 0. 0. 0. m (11; 11) U 0.000 0.005 0.0 0.9 0.0 0. 0.8 0.0 0. m (1; 1) U 0.000 0.00 0.00 0.00 0.0 0.055 0.01 0.001 0.05 m (1; 1) U 0.000 0.00 0.189 0.191 0.198 0.00 0.00 0.00 0.19 X 0 () U -0.015 0.015 0.00 0.009 0.0 0.08 0.01 0.05 0.0 X 0 (5) U -0.015 0.015 0.00 0.010 0.08 0.0 0.0 0.09 0.08 X 0 () U -0.100 0.00 0.01 0.09 0.08 0.1 0.1 0.088 0.08 X 0 () U -0.010 0.050-0.00-0.005 0.00 0.01 0.01 0.005 0.00 X 0 (8) U -0.010 0.050 0.01 0.018 0.05 0.0 0.0 0.0 0.05 C () U 0.000 0.015 0.00 0.00 0.005 0.00 0.00 0.005 0.005 C (5) U 0.000 0.015 0.005 0.005 0.00 0.00 0.00 0.00 0.00 C () U 0.000 0.10 0.0 0.0 0.0 0.08 0.090 0.0 0.0 A() N 0.000 0.010-0.0-0.00-0.00 0.005 0.00-0.010-0.00 A(1) N 0.000 0.00-0.001-0.001-0.001-0.001 0.000-0.001-0.001 B() N 0.000 0.00 1.5 1. 1.80..1 1.8 1.818 : for the uniform distribution we report lower and upper bound of the support. Notes: This table reports the priors and the posterior density for the parameters of m in eq.??, 0 and 1 in eq.??, C in eq. 1, and the initial values of the latent variables, X 0. The rst three columns report the distributions, means and standard deviations of the prior distributions. The fourth to the eight columns report the.5-th, 5-th, the 50-th and the 95-th 99.5-th percentile of the posterior distributions, respectively. The last two columns report the modes and the means of the posterior distributions. The statistics of all the posterior distribution of m and A are multiplied by 100. For the uniform distribution the lower and upper bounds are reported instead of mean and standard deviation, respectively. The results were obtained using the Metropolis Hastings algorithm.

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 1 Figure 1: Implulse responses: response to demand, supply, monetary policy and flight to quality shocks. (a) Responses to supply shocks. (b) Responses to demand shocks. (c) Responses to policy shocks. (d) Responses to ight to quality shocks. Notes: This gure depicts the impulse response functions (IRFs) of the state variables to supply shocks (top-left panel), to demand shocks (top-right panel), to policy rate shocks (bottom-left panel) and to ight to quality shocks (bottom-right panel). The IRFs are standardized by the standard deviation the respective shocks. For example, the responses to demand shocks are divided by the standard deviation of the demand shock. The dashed (dark blue) lines are the IRFs evaluated the mode of the posterior distribution of the parameters. The shadowed area refers to the 90 % (darker shadow) and 99 % error bands.

An Extended Macro-Finance Model with Financial Factors (Technical appendix) 15 Figure : Implulse responses: response to credit crunch, risk premia, target inflation and equilibrium growth rate shocks. (a) Responses to credit crunch shocks. (b) Responses to risk premia shocks. (c) Responses to in ation target shocks. (d) Response to equilibrium growth rate shocks. Notes: This gure depicts the impulse response functions (IRFs) of the state variables to credit crunch shocks (top-left panel), to risk premia shocks (top-right panel), to in ation target shocks (bottom-left panel) and to equilibrium growth shocks (bottom-right panel). The IRFs are standardized by the standard deviation the respective shocks. For example, the responses to equilibrium growth rate shocks are divided by the standard deviation of the equilibrium growth rate shocks. The dashed (dark blue) lines are the IRFs evaluated the mode of the posterior distribution of the parameters. The shadowed area refers to the 90 % (darker shadow) and 99 % error bands.