The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Monetary Policy and the Term Structure of Nominal Interest Rates: Evidence and Theory Charles L. Evans and David A. Marshall Working Papers Series Macroeconomic Issues Research Department Federal Reserve Bank of Chicago December 1997 (WP-97-10) ■ 1 FEDERAL RESERVE B A N K OF CHICAGO Monetary Policy and the Term Structure of Nominal Interest Rates: Evidence and Theory* Charles L. Evans* David A. Marshall* November 3, 1997 Abstract This paper explores how exogenous impulses to monetary policy affect the yield curve for nominally risk-free bonds. We identify monetary policy shocks using three distinct variants of the identified V A R methodology. All three approaches imply simi lar patterns for the effect of monetary policy shocks on the term structure: A contrac tionary policy shock induces a pronounced positive but short-lived response in short term interest rates, with a smaller effect on medium-term rates and almost no effect on long term rates. Because of their transitory impact, monetary policy shocks account for a relatively small fraction of the long-run variance of interest rates. The response of the yield curve to a monetary policy shock is unambiguously a liquidity effect rather than an expected inflation effect. We then ask whether a dynamic stochastic equi librium model that incorporates nominal rigidities can replicate these patterns. We find that the limited participation model of Lucas (1990), Fuerst (1992), and Christiano and Eichenbaum (1995), is broadly consistent with the data, provided modest adjustment costs are imposed on monetary balances available to satisfy households’ cash-in-advance constraint. ’T his paper represents the views of the authors and should not be interpreted as reflecting the views of the Federal Reserve Bank of Chicago. We thank Marty Eichenbaum and Larry Christiano for many helpful discussions. We thank Wendy Edelberg for superlative research assistance at an early stage of this research. Comments are welcome. ^Federal Reserve Bank of Chicago; Research Department, RO. Box 834, Chicago, IL 60690; (312) 3225812; cevans@frbchi.org ^Federal Reserve Bank of Chicago; Research Department, RO. Box 834, Chicago, IL 60690; (312) 3225102; dmarshall@frbchi.org 1. Introduction Monetary policy is the natural starting point for an inquiry into the economic determinants of the nominal term structure. Bond traders and other non-academic observers often cite monetary policy as a major factor in term structure movements.1 Academic observers have also argued that the term structure is intimately linked to monetary policy and its goals. For example, Bernanke and Blinder (1992), Estrella and Hardouvelis (1991), and Mishkin (1990) explore using the spread between long-term and short-term yields as an indicator ofmonetary policy, future economic activity and future inflation. However, empirical models of the term structure typically used in the finance literature do not explicitly incorporate monetary policy. Rather, they characterize the nominal term structure as driven by unobserved latent factors. For example, Litterman and Scheinkman (1991), and Dai and Singleton (1997) estimate three-factor models, with the factors associated with the level, slope, and curvature of the yield curve. A n open question is whether one or more of these factors corresponds, in part, to monetary policy shocks. In this paper, we ask how exogenous impulses to monetary policy affect yields on zerocoupon bonds of various maturities. W e investigate the impact of these shocks on the shape of the yield curve, as well as on term premiums, and ex ante real rates. Having documented these empirical patterns, we ask if a dynamic stochastic equilibrium model of aggregate economic activity can replicate the empirical patterns we find in the data. Standard equi librium macroeconomic models have had little success at modelling the term structure.2 W e ask whether the performance of this class of models can be improved by incorporating ex plicit nominal rigidities. Such an inquiry is a critical step in matching financial factors with economic determinants. The fundamental empirical problem in assessing the effects of monetary policy shocks is the identification problem: how to distinguish exogenous monetary policy shocks from the endogenous response of the monetary policy instrument to other, nonmonetary, exogenous *For example, the Wall Street Journal of December 13,1995 describes February 1994 as the month “when the Fed began raising short-term interest rates and set off the year’s bond-market slaughter.” 2 See, for example, den Haan (1995), Backus, Gregory, and Zin (1989), and Bekaert, Hodrick, and Marshall (1997b). 1 impulses. There isdisagreement in the profession on the best way to resolve this identification problem. Rather than taking a stand on this controversy, we use three different identification strategies that have been proposed in the literature. All are variants of the identified vector autoregression (VAR) approach proposed by Sims (1986), Bemanke (1986), and Blanchard and Watson (1986). In particular, we use the recursive identification strategy of Christiano, Eichenbaum and Evans (1996a,b), the non-recursive identification strategy advocated by Sims and Zha (1995a), and the approach of Gali (1992) that utilizes long-run restrictions as part of the identification strategy. While these three identification strategies have differing implications for the effect of monetary policy shocks on real economic variables, it is interesting that their implications for the effect of monetary policy shocks on the term structure are broadly similar. All three strategies imply that a contractionary policy shock induces a pronounced positive, but transitory response in short term interest rates, with a smaller effect on medium-term rates and almost no effect on long term rates. This finding stands in contrast to the popular opinion, often expressed in the financial press, that changes in monetary policy systematically affect long-term bond prices.3 Our empirical results imply that the main effect of monetary policy shocks is to shift the slope of the yield curve. Because of their transitory impact, monetary policy shocks account for a relatively small fraction (less than 15%) of the long-run variance of interest rates. This shock roughly corresponds to the slope factor in the models of Litterman and Scheinkman (1991), Knez, et. al (1994), and Dai and Singleton (1997). The response of the yield curve to a monetary policy shock is unambiguously a liquidity effect rather than an expected inflation effect, since the response of expected future inflation to the policy shock is opposite to the interest rate response. W e find some evidence that a contractionary policy shock increases term premiums, at least for the shorter maturities. Having documented these empirical patterns, we ask whether they are consistent with a dynamic stochastic equilibrium model that incorporates nominal rigidities. W e focus on the limited participation model suggested by Lucas (1990), Fuerst (1992), and Christiano and Eichenbaum (1995). W e calibrate the money-growth process to the results from our 3 Of course, our empirical experiment has a precise definition. Statements about monetary policy in the financial press undoubtedly confound monetary policy shocks and normal responses of policy to nonmonetary shocks. 2 estimated VARs. The theoretical model captures the broad features found in the data. In particular, a contractionary monetary shock causes a short-lived rise in the short-term yields, with the response decreasing in the maturity ofthe bond. These responses are unambiguously liquidity effects, with the real yields rising substantially more than the nominal yields. In addition, the monetary contraction induces a rise in term premiums, which also decreases with maturity. Finally, a number of recent studies are related to our empirical analysis. Each of our identification strategies have the property that the monetary authority does not respond to developments in the bond market contemporaneously. The studies by Leeper, Sims and Zha (1996) and Bernanke, Gertler, and Watson (1997) also maintain this assumption in their analysis with both short and long-term interest rates. While the focus of the latter article is on the way the monetary authority’s response function amplifies non-monetary impulses, they report a number results that are qualitatively similar to our findings. Gordon and Leeper (1994) and McCallum (1994), however, take the view that the monetary authority responds contemporaneously to information conveyed in long-term interest rates. The plan of the remainder of the paper is as follows: In section 2 we describe the three strategies we use to identify monetary policy shocks. In section 3, we present the implica tions of each of these strategies for the effect of monetary policy shocks on the yield curve. Section 4 sets out the equilibrium model with limited participation constraints, describes our calibration of the model, and compares the implications of the theoretical model to our empirical results. Section 5 concludes. 2. Identifying monetary policy shocks Since Sims (1980), numerous proposals have been made for identifying fundamental economic impulses using V A R methods.4 In an attempt to characterize the facts about monetary policy and the term structure robustly, we use three alternative strategies for identifying monetary policy shocks. Each of the three strategies requires estimation of an identified VAR. To 4 The articles mentioned in the introduction are a small subset of the empirical literature that uses VARs to understand economic fluctuations. Surveys by Watson (1995), Christiano, Eichenbaum, and Evans (1997b), and Leeper, Sims, and Zha (1996) provide a fuller description of this literature. 3 conserve space, the discussion in this section focuses on identification issues. For a discus sion of econometric issues in estimating these models, the reader is referred to Christiano, Eichenbaum and Evans (1997b) and Sims and Zha (1995b). The empirical approaches we use are: (1) a recursive strategy studied by Christiano, Eichenbaum, and Evans (1996a, b); (2) a nonrecursive strategy studied by Sims and Zha (1995a); and (3) a strategy which employs a combination of long-run and contemporaneous restrictions studied by Gali (1992). Much of the literature focuses on quarterly time series analysis, while the frequency of our data analysis is monthly. Consequently, the robust macroaggregate responses to the three measures of monthly monetary policy shocks is of some independent interest. 2.1. M onetary policy rules In allof the identification strategies we use, itisassumed that the monetary policy instrument is the federal funds rate, denoted F F t. W e assume that F F t is determined by a relationship of the form F F t = f ( n t ) + cret (2.1) In (2.1), fit is the information set available to the monetary authority at date t, / is linear function that describes the monetary authority’s reaction to the state of the economy, and et isan exogenous shock to the monetary policy reaction function with unit variance. The policy reaction function / incorporates the authority’s preferences regarding counter-stabilization actions, inflation-fighting activity, and so on. The residual £ reflects random, nonsystematic factors that affect policy decisions, such as political factors and the personalities, views and composition of the Federal Open Market Committee. W e will be considering in detail the responses to monetary policy shocks of bond yields of various maturities. To insure that the shocks do not change as we move across maturities, we exclude the bond yields from Clt . This requires excluding the bond yield data from the V A R equations that describe the evolution of the macrodata. Consequently, in all of our V A R systems, the non-yield equations include no lagged values of the yields. Of course, the yield equation in each V A R includes lagged values of all of the V A R system’s variables. 4 Finally, we do not consider specifically a T a y l o r r u le for monetary policy in any of our identification strategies (Taylor 1993). However, it is not possible to observe that the mone tary policy rule is (2 .1) and be certain that it is inconsistent with a Taylor rule. To see this, consider a simple-variant of a Taylor rule: FFt= c*0 + yt+ a i where y is Taylor’s output gap and p a 2E [p{+s - pt \0.t]+ <?£t (2.2) is the log price level, so the expectation in brackets is the s-period-ahead inflation rate. Clearly, in evaluating the conditional expectation of the s-period-ahead inflation rate, any variable which Granger-causes the inflation rate will appear in the conditional expectation. Without substantially more structure placed on the analysis, equations (2.1) and the reduced form implied by (2.2) are indistinguishable. 2.2. Christiano-Eichenbaum -Evans recursive identification strategy In the Christiano, Eichenbaum, and Evans (1996b) (denoted flfEE) variables. Since (C E E ) strategy, the information set includes all lagged variables in the system, plus certain contemporaneous C E E (1996b) employs monthly data, we have modified the data vector only slightly in order to facilitate comparisons with the theoretical model in Section 4. Specifically, the data vector is given by where: E M Z CEE = (E M , P, P C O M , F F , N B R / T R , denotes the logarithm of nonagricultural payroll employment; P A M 2), denotes the logarithm of the personal consumption expenditures deflator in chain-weighted 1992 dollars; P C O M denotes the smoothed change in an index of sensitive materials prices; F the federal funds rate; N B R / T R credit to total reserves; and A M 2 F denotes denotes the ratio of nonborrowed reserves plus extended denotes the log growth rate ofthe monetary aggregate M 2 .5 As Christiano, Eichenbaum, and Evans (1996a) discuss, the inclusion of commodity prices in a recursively-identified V A R mitigates anomalous responses of the price level from monetary policy shocks (the “price puzzle” described by Sims (1992) and Eichenbaum (1992)). 5 For all our VARs, the logged data is also multiplied by 100 so that the impulse responses can be interpreted as percent deviations in all of our figures. 5 The P ,and C E E monetary policy reaction function includes the contemporaneous values of E P C O M in Q $ E E in equation (2.1). Specifically, F F t and e fEE M , = A i( L ) Z fJ { E + 01 E M t + a 2P t + a 3P C O M t + c re fE E is assumed to be orthogonal to all other right hand-side variables. This identi fication strategy has two important properties: (1) E M , P , and P C O M do not respond contemporaneously to the monetary policy shock, and (2) all of the other variables in the Z f EE reSp0n(l contemporaneously to the monetary policy shock. In this sense, the iden tification of the monetary policy shock e E E B is recursive. The VARs were estimated over the sample period 1965:1 to 1995:12. Twelve lagged values were estimated in each equation, with the initial lags beginning in 1964:1. Column one in Figure 1 displays the impulse response functions for the C E E R e c u r s iv e monetary policy shocks. Monte Carlo bootstrap methods were used to compute 95% confidence bands. The confidence bands are displayed around the point estimates of the impulse response functions, leading to generally asymmetric error bands (as suggested by Sims and Zha 1995). The monetary policy shock leads to a 46 basis point increase in the federal funds rate on impact. The funds rate rises to its maximum of 58 basis points in the second month before falling thereafter. The funds rate response, therefore, is persistent but transitory: this response pattern holds for each of the identifications we consider. The other variable’s responses seem consistent with most economists’ prior expectations for a monetary policy shock. The increase in the federal funds rate occurs simultaneously with reductions in nonborrowed reserves relative to total reserves as well as M2. Employment and the P C E deflator are unchanged for several periods before falling persistently. Employment begins to fall before the P C E deflator, while commodity prices fall almost from the outset. Finally, the P C E deflator’s response is n e g lig ib ly positive for about six months. Of the three monetary policy identifications we consider, this is the largest price puzzle in Figure 1. 6 2.3. Sims-Zha nonrecursive identification strategy Sims and Zha (1995a) propose an alternative strategy for identifying monetary policy shocks which is a nonrecursive scheme. In their analysis, the monetary authority’s information set (denoted fif2) includes all lagged variables in the system, plus certain contemporane ous variables. Sims and Zha’s empirical analysis used quarterly data. W e use monthly analogues to their quarterly data series. Specifically, our data vector is given by (P c m , A T R , F F , P im , P , W , Y ) , A T R where: P e rn denotes the log of intermediate goods prices; Y = denotes the logarithm of crude materials prices; denotes the log growth rate of total reserves; F denotes the log of the real wage; and Z sz P F denotes the federal funds rate; P i m denotes the log of the P C E deflator; W denotes the log of real G D P .6 When Sims and Zha analyze the federal funds rate as the monetary policy instrument, they select total reserves to be the monetary aggregate in the analysis. Our analysis follows their variable selection.7 The and SZ A T R monetary policy reaction function includes the contemporaneous values of P c m in fl f z in equation (2.1). Specifically, FFf = A\(L)Zf3i + OL\Pcmt + GL2 A T Ri + (2.3) (j e ^z . The Sims-Zha strategy isnonrecursive, because e f z isallowed to be correlated with P c m t and A T R t. (That is, P c m t and A T R t are allowed to respond contemporaneously to a monetary policy shock.) This correlation implies that e fz cannot be recovered as the residual from an OLS regression. Furthermore, Sims and Zha’s system of equations does not possess any predetermined variables which can be used as instruments for P c m t and A T R t in equation (2.3). This leads Sims and Zha to full-information estimation methods. For the remainder of this subsection, let e refer to the vector of structural (SZ) shocks, 6Our real GDP data is interpolated (from Leeper, Sims and Zha (1996)). Sims and Zha used the quarterly GDP implicit deflator; our use of the PCE deflator is consistent with their choice of a price index with timevarying commodity bundle weights. In all nonrecursive cases we considered, using the PCE deflator resulted in fewer price puzzles than LSZ’s preferred use of the CPI. 7 We also considered VARs with A M 2 in place of A T R , and the results were similar to the responses reported in figure 1 . 7 and u refers to the vector of VAR innovations. That is, A (L )Z fz = ut where A ( - ) denotes the matrix polynomial of V A R coefficients. Sims and Zha assume that e and u are related via a linear transformation £ t = B 0U t, where Bo is a square nonsingular matrix. To achieve identification of the S Z monetary policy shock £ m p i we follow Sims and Zha relatively closely in specifying the B £Pcm £m d £m p £Pi m £p £w/p Ey £yield B n B\2 0 B22 Bz\ B32 Ba 0 B51 0 B qi 0 B71 0 B&i B s 2 B\z B u B 1 5 Bie Bzz 0 B 2 5 0 Bzz 0 0 0 0 B u B45 B 4 & 0 0 B 5 5 B 5q 0 0 0 Bee Bn B27 0 0 0 0 B47 0 B57 0 Be7 0 0 0 0 B77 0 0 Bzz B&4 B z z B&e B z 7 Bzz q matrix: UPcm Utr Upp Upim Up V-w/p Uy Uyield W e depart from Sims and Zha by excluding personal bankruptcies from the VAR. The first row indicates that P e rn is an i n f o r m a t i o n v a r ia b le in the economy (other than the yield shock £yieid)- and responds to all structural shocks £ The second row is a money demand relationship. Our estimation constrains the coefficients on sign, while uy and u tr and u r f to have the same u p have opposite signs from u TR.8 The third row is the monetary policy 8These sign restrictions insure that the interest elasticity of money demand is negative, the output 8 reaction function. Rows four through seven indicate that the monetary policy shock on impact on P cm . P im , P , it/, and Y respond to o n ly in d ir e c t ly th ro u g h th e e ffe c t o f m o n e t a r y p o l i c y Finally, in order to maintain the same identification of monetary policy in each of our VARs, the yields are influenced by all the other variables but influence no variables themselves. The VARs were estimated over the sample period 1964:7 to 1995:12. Six lagged values were estimated in each equation, with the initial lags beginning in 1964:1. Column two in Figure 1 displays the impulse response functions for the SZ monetary policy shocks. Following Sims and Zha (1995a), Bayesian Monte Carlo methods were used to compute 95% confidence bands. The confidence bands are displayed around the point estimates of the impulse response functions, leading to generally asymmetric error bands. The SZ monetary policy shock leads to a 50 basis point increase in the funds rate on impact, rising to 64 basis points in the second period. As with the C E E case, the funds rate falls thereafter. Total reserves fall over this period, although the initial response is close to zero. Prices fall on impact, with the response larger for crude materials prices and smallest for the P C E deflator. Real G D P and wages rise negligibly and insignificantly for the first five months before falling. Broadly speaking, these responses are qualitatively similar to the recursive results, although there is greater uncertainty as measured by the confidence bounds. 2.4. G ali id en tificatio n strategy using long-run restrictions Gali (1992) uses an alternative identification strategy which imposes a mixture of long-run restrictions and contemporaneous impact restrictions to identify four economic shocks, one of which is a monetary policy shock. Gali’s empirical analysis used quarterly data, while our analysis uses monthly data. The monetary policy reaction function can be represented by equation (2 .1). Specifically, we follow Gali in considering a four-variable autoregression. The data vector is given by log difference of GDP; F F Z G = (A Y , F F , F F — A P, A M denotes the federal funds rate; F — A P ), F — A P where: A Y denotes the denotes the real interest elasticity of money demand is positive, and that the price elasticity of demand for nominal balances is positive. 9 rate where A P is the log difference of the CPI; and A M — A P denotes real M l balances. To maintain comparability with the other procedures, we use the level of the federal funds rate. (Gali used the first difference of the interest rate.) Since Gali’s data is quarterly, we use the Leeper-Sims-Zha monthly data set for monthly GDP, fed funds rate, CPI and M l .9 Gali’s monetary policy reaction function can be represented as F F t and e f = A { L ) Z f _ l + a xA Y t + a 2{ F F t is potentially correlated with time t - A P t) + a 3( A M t - A P t) + a e f (2.4) variables (like the Sims-Zha system). Identifi cation is achieved with six restrictions on the covariance structure of the innovations. To understand Gali’s identifying restrictions, we must provide the other three structural shocks with economic labels: supply shock, money demand shock, and IS shock. Gali’s restric tions can be described as follows. First, the monetary policy, money demand and IS shocks have no long-run effect on output; these restrictions identify the supply shock. Second, the monetary policy and money demand shocks have no contemporaneous effect on output; knowledge of the supply shock and these two restrictions identifies the IS shock. Third, one additional identifying restriction is necessary to identify the remaining two shocks. One of the restrictions that Gali considers deletes the price data from the monetary authority’s contemporaneous information set. In equation (2.4), this imposes the coefficient restriction that <*2 = — 03,leaving only two contemporaneous coefficients to estimate with two available instruments (the supply and IS shocks). This identifies the monetary policy shock.10 The VARs were estimated over the sample period 1964:8 to 1995:12. Six lagged values were estimated in each equation, with the initial lags beginning in 1964:2.11 Column three in 9We follow Gali in using the CPI rather than the PCE deflator for two reasons. First, Gali conducted several sets of unit root and cointegration tests in order to justify his data transformations. Consequently, we used his P and M l to maintain comparability. Second, the CPI data system delivered more plausible impulse response functions for most of the shocks than the PCE deflator data system. Since the CPI does not get revised, its stochastic trend properties may be more consistent with Gali’s unit root assumptions. 10 Gali (1992) alternatively considers: (1 ) deleting only output from the contemporaneous information set of the monetary authority, and (2 ) explicitly imposing a homogeneity restriction on the money demand equation in his structural VAR. Gali reported that his results were largely robust across these alternative identification restrictions, and our implementation of these restrictions also produced qualitatively similar results. 11 “Double-differencing” to impose the long-run restrictions in identifying the supply shock uses up the 10 Figure 1 displays the impulse response functions for the G a l i monetary policy shocks. Monte Carlo bootstrap methods were used to compute 95% confidence bands. The confidence bands are displayed around the point estimates of the impulse response functions. The G a li monetary policy shock increases the federal funds rate on impact by 41 basis points. In the second period it increases by 53 basis points, and falls thereafter. M l growth falls during this period, indicating a liquidity effect. As with the SZ policy shock, the price level falls on impact and declines further after about six months. As with the C E E policy shock, real activity (as measured by monthly real GDP) is about flat for four months and then falls. 3. The response of bond yields to exogenous monetary policy shocks 3.1. Im pulse responses Figure 2 plots the estimated responses of bond yields to a one-standard-deviation contrac tionary monetary policy shock. Bond yields are measured as continuously-compounded an nualized returns on zero-coupon bonds. The yields from 1959:01 -1991:02 are monthly data taken from McCulloch and Kwon (1993). For the period 1991:03 - 1995:12, we use yields computed by Robert Bliss using the McCulloch/Kwon procedure. (See Bliss (1994).)* 12 The solid lines give the point estimates of the impulse responses; the upper and lower dashed lines give the boundaries of the 95% confidence region. The plots trace the responses over 24 months. Each of these responses is measured in percent deviation from the non-stochastic steady state. W e display the responses for bond maturities of one month, six months, one year, 3 years, and 10 years. According to all three of the identification strategies, the policy shock increases the one-month rate by approximately 20 basis points in the period when the shock occurs. This response is statistically significant in each case. The one-month additional lag beginning in 1964:1. 12 McCulloch and Kwon’s (1993) data on zero-coupon bond yields are derived from a tax-adjusted cubic spline discount function, as described in McCulloch (1975). A more detailed explanation can be found in McCulloch and Kwon (1993). Unlike McCulloch and Kwon (1993), Bliss (1994) does not tax-adjust the bond yields. However, under the current tax code, the requisite tax-adjustment in the McCulloch-Kwon procedure is negligible. From 1987:01 through 1991:02 (the last date where we have an overlap between the two data sets), the McCulloch/Kwon data and the Bliss data are virtually identical. 11 rate continues to climb in the following month, and then falls rapidly, with the effect of the shock insignificant after 6 months. The six-month and the twelve-month rates display qualitatively similar response patterns, although the magnitude of the response decreases for the longer-term bonds. When we move to even longer-term bonds, the initial effect dimin ishes substantially as maturity increases: The initial response of the three-year bond is only around 9 basis points, falling to less than 5 basis points for the 10 year bonds. The initial response for this last bond is significant (barely) for only the C E E identification. (Under the C E E identification, we found that the response to a federal-funds shock of all bonds longer than 13 years is insignificant at the 5% marginal significance level.) The main qualitative discrepancy among the three identification strategies is that the bond yield responses die off somewhat more slowly in the SZ identification than either in the C E E or the Gali iden tifications. Interestingly, these results are roughly comparable to Cook and Hahn’s (1989) estimates of the effects on interest rates after a publicly announced change in the federal funds rate. They find that in response to a 100 basis point increase, short rates rise about 50 basis points, while long rates rise about 10 basis points. The results are straightforward: There is a significant, relatively large, but relatively short-lived effect on the short rates, with a decreasing and less significant effect at longer maturities. In other words, there is not a parallel upwards shift of the term structure in response to these monetary policy shocks; rather, the shock causes the yield curve to flatten. A n alternative way to portray these patterns is to look at the effect of a monetary shock on the shape of the yield curve. One way to summarize this shape is to take a quadratic approximation of the yield curve at each date. W e do so by regressing all interest rates at a given date on a constant, maturity, and squared maturity, and treating the parameter estimates (denoted i n t e r c e p t , s lo p e , and cu rv a tu re respectively) as the coefficients of this quadratic approximation. (Note that these coefficients are time-varying, since the regression only involves interest rates at a given date, and is re-estimated anew each month.) To portray the way the shape of the yield curve responds to a monetary shock, we estimate VARs, analogous to those described above, in which the interest rate is replaced by one of these three coefficients. The resulting impulse responses are displayed in Figure 3. A monetary shock raises 12 the level of the yield curve, decreases the slope, and reduces the curvature. (The positive response of c u r v a t u r e denotes a reduction in curvature because the average yield curve is concave, so the average value of c u r v a t u r e is negative.) The positive response of le v e l looks very much like the response of the one-month interest rate. In the C E E identification, this response becomes insignificant in about 6 months. The effects on s lo p e and cu rv a tu re are significant only for the first four months. 3.2. Variance decompositions The impulse responses suggest that monetary policy is an important determinant of short run interest-rate variability, at least for the shorter-term rates. To study this question directly, consider the variance decompositions displayed in Table 1. The table gives the point estimates of the fraction of the one-month ahead, six-month ahead, and 24-month ahead conditional variance of five bond yields attributable to the monetary policy shock, as identified by each of the three identification strategies. According to this table, monetary policy shocks account for 17% - 18% of the six-month ahead conditional variance of the one-month interest rate. The fraction of the six-month ahead variance accounted for by the monetary policy shock decreases sharply with maturity. It isstill non-trivial for the one-year interest rate (9% - 11%), but rapidly becomes negligible as maturity lengthens. The 24-month ahead conditional variance can be interpreted as a proxy for the uncon ditional interest rate variance. According to the C E E and Gali identifications, monetary policy shocks account for a relatively smaller fraction of long-run variance of interest rates (Around 7% for the one-month rate, less for the longer-term rates). This reflects the rapid decay in the impulse responses implied by these identification strategies. The SZ identifica tion attributes somewhat more of the long-run variance to monetary policy shocks, due to the greater persistence of the impulse responses implied by that identification strategy. The impulse responses and variance decompositions suggest that the monetary policy shock resembles the “slope” factor identified in the finance literature. In particular, Litterman and Scheinkman (1991) and Dai and Singleton (1997) estimate factor models of the term structure in which the three factors shift the level, slope, and curvature of the yield curve, respectively. Litterman and Scheinkman (1991) find that the level factor accounts 13 for about 90% of the unconditional variability of yields across the maturity spectrum, with the slope factor accounting for most of the rest. However, Dai and Singleton (1997) note that the slope factor accounts for a good deal of the short-run variability of the short-term interest rate. The slope factor is less important for the unconditional variability because it has only a transitory impact, with a half lifeof about 4 months. (In contrast, the level factor induces more persistent responses in the yield curve. The half-life of a level-factor impulse is estimated at approximately 4 1/2 years.) All of these characteristics correspond closely our result for the monetary policy shock. One might conjecture that the slope factor identified by Litterman and Scheinkman (1991) and Dai and Singleton (1997) is, in part, driven by monetary policy. 3.3. Term prem ium s Monetary policy shocks could affect longer rates either through their effect on expected future short rates or by affecting term premiums. To distinguish between these two alternatives, let B ? denote the T-period continuously-compounded bond yield, and let us define the T-period term premium, T P ?, by T P J = fl? - i £ 1 t=0 E tH u , (3.1) That is, the term premium is the difference between the T-period interest rate H ? and the average of expected future 1-period interest rates over the next T periods. The expecta tions theory of the term structure is the hypothesis that term premiums are time-invariant. It can be shown (see Bekaert, Hodrick, and Marshall (1997a)) that the expectations hy pothesis is equivalent to the hypothesis that the intertemporal marginal rate of substitution in nominal wealth is conditionally homoskedastic, in the strong sense that a ll conditional higher moments are time-invariant. Of course, the expectations theory has been rejected decisively in US data,13 so, empirically, T P ? varies through time. It is of interest, therefore, to see whether monetary policy shocks affect longer yields primarily through their effect 13 However, the expectations theory fares far better in other countries. For example, the expectations theory cannot be rejected using data from the UK, and the rejections are far less decisive with German data. See Bekaert, Hodrick, and Marshall (1997a), Hardouvelis (1994), Jorion and Mishkin (1991). 14 on expected future short yields, or whether they directly affect term premiums. This is an important issue in its own right, and it may serve to indirectly inform us about the way monetary policy affects the elusive intertemporal marginal rate of substitution in wealth. To help understand the sources of time-variation in term premiums, we compute the response of T P j to the monetary policy shock in our model as the difference between the contemporaneous response of R j and the average of the first T -step responses of R}. In Figure 3, we display these responses, along with the 95% confidence intervals, for T = 6 months through 3 years.14 For all three identification strategies, the point estimates imply that the six-month term premium responds positively to a contractionary monetary policy shock, with the maximal response of approximately 10 basis points occurring between two and five months after the initial impulse. However, the significance of these estimates is questionable: The maximal response barely creeps into significance for the C E E and Gali identifications. For the longer maturities, the C E E identification implies a significant and rather long-lived term premium response. For example, the 12-month term premium rises to 12 basis points after one year. However, these responses are not found in the other identification strategies. W e conclude that the evidence for longer term premium responses is decidedly mixed. 3.4. Real yields, and expected in fla tio n premiums According to Figures 1 and 2, nominal bond yields and inflation move in opposite directions in response to a monetary policy shock, so the nominal yield response clearly represents a liquidity effect, rather than an expected inflation effect. To quantify the magnitude of this liquidity effect, we compute the response of the real T-month yield to the monetary policy shock. W e do so by subtracting from the i th response of the T-month nominal yield the average (annualized) inflation response from step i + 1 to step i + T. These computations are displayed in Figure 4 for T = 6 month through 3 years. According to all three identification 14To estimate the response of term premia, we must estimate a 7-variable VAR, including both the 1-mont interest rate and the T-period rate. It is problematical to perform this exercise for the longer interest rates, since it involves summing the first T responses of the long interest rate. In the case of the 10-year rate, for example, we would have to sum the first 120 responses. We have little confidence in the point estimates over this long a horizon. 15 strategies, real yields respond significantly to monetary policy shocks for all maturities dis played. Notably, both the C E E and Gali identifications imply that the positive response of the 12-month and 3-year real yields is economically meaningful (about 20 basis points) and highly persistent, remaining statistically significant for a year or more. (The point estimates from the SZ identification tell a similar story, although the confidence bands are much wider for this identification strategy.) 4. Can a dynamic equilibrium model replicate these patterns? The empirical results for each of the three identification strategies are remarkably similar. A contractionary monetary shock causes a substantial rise in short-term nominal yields, with a progressively smaller response as the bond maturity is lengthened. This in turn flattens the slope and curvature of the yield curve. These responses are rather transient, fully dissipating between six months and one year. They represent pure liquidity effects: The responses of real yields are significant, and generally exceed the response of the nominal yields. There is some evidence that term premiums also respond positively, at least for the shorter yields. W e now ask whether these patterns are consistent with a dynamic equilibrium of rational agents. To answer this question, we study the implications for the term structure of nom inal interest rates of a class of dynamic models that incorporate the limited participation assumption introduced in Lucas (1990) and studied in Fuerst (1992) and Christiano and Eichenbaum (1995). The benchmark model we use is a variant of the model analyzed in Christiano, Eichenbaum, and Evans (1997c). The nominal rigidity in this class of models is that households must decide how much cash (denoted b e fo re “ Q t ') to use in the goods market the monetary policy shock is revealed. Furthermore, it is assumed that Q t cannot be adjusted without cost. Rather, there is a cost in leisure time that must be paid whenever consumption money is changed from period to period. This adjustment cost allows real effects of a monetary policy shock to be propagate dynamically through time. The following is a brief description of the model. (See Christiano, Eichenbaum, and Evans (1997c) for a more detailed description.) 16 4 .1 . B a s i c S e t - u p There are three types of entities: households, firms, and a financial intermediary, plus a government whose sole function is to create money. The households own the firms and the financial intermediary, so all profits from these entities are paid to the households. Households’ purchases of consumption and new capital are subject to a cash-in-advance constraint, to be described below. Since the focus of this paper is on the effects of monetary policy, we assume that, unlike the empirical models of section 2, monetary policy is the only source of randomness in this economy. Each period, the government injects a quantity of money X t as a transfer to the financial intermediary. The total quantity of money in the economy evolves according to M t+i = M t + X t. For convenience in calibrating the model to the impulse responses described above in section 3, we assume that net money growth xt = — 1 evolves as a stationary moving average process: X t = X + 90e t + 6 \ £ t —l + @ 2 £ t—2 + @ z £ t—3 + 9 ± £ t —4 (4.1) where x denotes the nonstochastic steady state rate of money growth, and {e(} is a sequence of i.i.d. standard normal shocks. Equation (4.1) is an exogenous monetary policy rule, while equation (2 .1) is an endogenous monetary policy rule. Notice that the monetary authority could be using (2.1) to set the federal funds rate and yet the Wold representation for money growth would be a moving average of current and lagged exogenous shocks, as approximated by the finite-order M A in (4.1). The observational equivalence of these rules allows us to estimate an endogenous monetary policy rule from the data, and then use the equivalent exogenous policy rule as the driving process in the model. The {0i}*=o coefficients in (4.1) are computed directly as the impulse response coefficients from row four in Figure l.15 In all cases we truncate the M A process at an MA(4). As the confidence bands in Figure 1 show, this truncation seems reasonable. 15 Recall that the log money growth data has been scaled up by a factor of 100. 17 4 .2 . H o u s e h o ld s The representative household’s choice variables in period cash set aside for purchases Q t, money t M t+ 1, capital K t+i , bonds with maximum maturity of n periods, denoted ^B3 tj are: consumption C t, labor L t, and a portfolio of zero-coupon ^ In this notation B { denotes a bond purchased at date t paying one dollar at the end of date t + j — 1. The household takes as given the nominal rental rate on capital, rf,the dollar price bj. of a bond maturing at the end of period t + W t and j — 1,as well as the dollar prices of labor and consumption goods, P t. The timing is as follows: At the beginning of period t, the household carries over from the end of the previous period M t, K t, and bonds of maturities 2 through n (One-period bonds purchased in period — 1 pay off at the end of period t t — 1.) Before the monetary policy shock in period t is revealed, the household must set aside Q t dollars to finance purchases subject to the cash-in-advance constraint. The household takes its remaining financial assets (money holdings M t — Qt and holdings of zero-coupon bonds) to the financial intermediary. The monetary policy in period t is then revealed. Having seen the shock, the household rebalances its portfolio by purchasing from the intermediary bonds of maturities 1 through n. The portfolio constraint facing the household in these transactions is:.16 (4.2) j =i i=2 The household then rents its capital K a firm for nominal wage W t- t at nominal rental rate r t and sells its labor L t to It is assumed that wages are paid in money that can be used immediately for purchases of consumption and new capital. The cash-in-advance constraint can therefore be written: Pt {Ct + ( K t+1 - (1 - 6 ) K t)) < Q t + W tL t . (4.3) where 6 denotes the capital depreciation rate. Finally, at the end of the period, the firm pays6 1 16As long as b\ < 1 (equivalently, the one-period net nominal interest rate is positive), the household sells its entire money holdings (net of Qt) to the intermediary. 18 out all profits to the household as a dividend D t, and the financial intermediary redeems all maturing bonds B } and pays out all of its profits to the household as a dividend F t . The flow budget constraint for nominal household wealth can therefore be written: M t+ i < F t + D t + B \ + T tK t + Q t + W t L t — P t ( C t + (-Kt+i —(1 —$ ) K t ) ) ■ (4-4) Let H t denote the time cost of adjusting Q . This adjustment cost is assumed to have the following form: Ht = H + exp c( ^ " i_x '_ Q t_ (4.5) — l— X ,Q t-1 - 2} ’ where x (the steady state growth rate of money in equation (4.1)) is the net growth rate in Qt in a nonstochastic steady state. In period t, the household chooses C t , Q t, L t, Mi+1, K t+ i, and jl?* | _i to maximize OO E - i ^ 2 f t U (Ct , L t , H t ) t =o subject to (4.2), (4.3), and (4.4), where U (C ,L ,H ) (.L + H f 1+ +v,) (1-7) /(I ~7)- (4-6) This utility function has the property that the income effect on leisure is zero. Everything else equal, this tends to magnify the output response from a monetary shock. Intuitively, the household’s labor supply does not decrease when money expands. Parameter ijj is the inverse of the elasticity of labor supply. Parameter 7 is a curvature parameter that affects the household’s degree of risk aversion. Parameter ipo is purely a scaling parameter. Households make all date t choices except one as functions of information known at date t and earlier. The exception, Q t , is restricted to be a function of date t — 1 and earlier information only. 19 This informational constraint on Q t reflects the limited-participation feature of the model. Both the informational restriction on the choice of Q t and the cost of adjusting Q t can be interpreted as ways of capturing, in a representative-agent model, more fundamental microeconomic frictions affecting household portfolio adjustment. For example, Caballero (1993), Marshall and Parekh (1994), and Schroder (1995) show that extremely small fixed costs of adjusting an economic choice variable at the individual level can imply extremely sluggish behavior of the corresponding macroeconomic aggregate. Even small costs of portfolio ad justment can imply a sluggish response of the aggregate household portfolio to monetary policy shocks. Unfortunately, it is extremely difficult to formulate dynamic equilibrium models with cross-sectional heterogeneity that explicitly incorporate fixed adjustment costs. (The models of Caballero (1993), Marshall and Parekh (1994), and Schroder (1995) are all partial equilibrium models.) Our formulation is an attempt to incorporate these effects into an equilibrium model in a tractable fashion. 4.3. Firms We adopt a Blanchard-Kiyotaki (1987) type of monopolist competition framework. At time t, a final consumption good, Yt , is produced by a perfectly competitive firm. It does so by combining a continuum of intermediate goods, indexed by * € (0 , 1), using the technology: (4.7) Yt = where 1 < fi < oo and Yu denotes the time t input of intermediate good i . 17 Let P t and P it denote the time t price of the consumption good and intermediate good i, respectively. Profit maximization implies the Euler equation: = \ P j Yu Yt ' (4.8) 17The model with competitive identical firms is a special case of this model in which \i is set to unity. This monopolistic competition paradigm is typically adopted in limited participation models because it implies a larger output response and a more realistic investment response to monetary policy shocks. However, the responses of bond yields change very little when fi is set to unity. 20 Integrating (4.8) and imposing (4.7), we obtain the following relationship between the price of the final good and the price of the intermediate goods: Pt = (4.9) r rr ' p ; r d i [Jo Intermediate good i is produced by a monopolist who uses the following technology: K % i\ r - if $ K ° i L ) r > </, (4.10) Y it = < 0 where 0 < a < 1. Here, i th Lu and o th e r w is e Ku denote time t labor and capital used to produce the intermediate good. The parameter (j> denotes a fixed cost of production. We rule out entry and exit into the production of intermediate good i. Intermediate firms rent capital and labor in perfectly competitive factor markets. Economic profits are distributed to the firms’ owner, the representative household. The firm’s choices are affected by monetary policy through a cash-in-advance constraint. Firms retain no cash from period to period, but workers must be paid in advance of produc tion. As a result, firms need to borrow their wage bill, W tL it, from the financial intermediary at the beginning of the period. Repayment occurs at the end of time period t, at the gross interest rate, R t . Given that the firm’s only source of finance is through the financial inter mediary, this feature of the model is one possible articulation of the credit channel in the monetary transmission mechanism. Profit maximization leads the intermediate good firm to set its price equal to a constant markup over marginal cost: P it = implying W tR t Pt where f^ t = (1 —a) (K it/ L i t ) a _ fL,t _ fn ,t V Pt (4.11) /JL is the marginal product of labor and fx ,t = a ( L it/ K i t )^ is the marginal product of capital; and we have imposed the equilibrium condition, Pit = Pt 21 for all z.18 Note that, in equation (4.11), the nominal interest rate R t is determined in part by the marginal product of la b o r. This reflects the cash-in-advance constraint on labor inputs, described above. In equilibrium, all intermediate goods firms choose the same labor and capital combinations, so we henceforth drop the i subscript. 4.4. Financial Intermediary At time t, a perfectly competitive financial intermediary buys and sells bonds with the household at prices l?t , j — 1,...,n. The net amount of funds transferred from households to the financial intermediary is M injection, X t, t — Q t- The intermediary also receives a lump sum cash from the monetary authority. These funds are supplied to the login market at the gross interest rate R t. Demand in the loan market comes from the intermediary good producers, who seek to finance their wage bill, W tL t. Clearing in the loan market requires: (4.12) W tL t = M t - Q t + X t . At the end of the period the intermediary pays off all maturing bonds B} to households, and distributes its profits (revenue from loan repayments minus the cost of paying off maturing bonds) to households as a dividend F t : F t = R tW tL t - B\. The assumption of perfect competition insures that the nominal return earned by the house hold on one-period bonds equals R t ,the nominal return earned by the intermediary from its 18In deriving equation (4.11), we also use the following characterization of the marginal cost of the inter mediate good firm: 1 / T \ a 1 / PC \ a) u ) 22 r' one-period loans to the firms. That is, (4.13) * 4 4.5. Equilibrium Let At, ut , and denote the Lagrange multipliers associated with constraints (4.2), (4.3), and (4.4), respectively. The first-order conditions of the household are: For Q t: —“ —P U H it+i H ft+1 E t-i ■+ v t + £,t + (4.14) = o. Qt For L t: U L ,t + ( v t + Z t) W i = (4.15) o. For C t : U c ,t For K = iy t + (4.16) it ) P t - t + i: + Z t) P t ( K t+1 - (1 - S ) K t) = 0 E k {^t+1rt+i + (ut+ l + em)^+i(l ~ S )} (4.17) For M t+i: Zt = /3 E t { At+i } . (4.18) At6j = 6 (4-19) X M = 0 E t { X t+ 1b i ~ l } . (4.20) For B \ : For B \ ,j = 2 ,...,n: Notice that the conditional expectation in equation (4.14) is with respect to period 23 t — 1 information. This reflects the limited participation feature of the model. Using equations (4.16) and (4.18) we can eliminate multipliers v t and £t. W e can then use equations (4.11) and (4.13) to obtain the following equilibrium conditions: E t -1 {At} — (3U h ,,£-hlH 't+ 1 E t~ i - U c,t Q t+ i U c ,t Qt Pt (4.21) = P E k |/5A(+2^ i i P t +i + Uc,t+ i(l - 6 ) © II £1* J" + (4.22) W tR t f L<t rt f K ,t Pt A* Pt t* X tH = 1,... , n j = 0 j E t { \ t+ j}, (4.23) (4.24) Equation (4.24) implies that the marginal utility of nominal wealth is At, and the in tertemporal marginal rate of substitution (IMRS) in nominal wealth is . This IMUS determines the bond yield of maturity j , denoted K [ , according to , r^At+J- (4.25) p i The behavior of the term structure is therefore determined by the stochastic process for At. Using (4.13), (4.21), (4.22), (4.23), and (4.24) evaluated at j = 1,we obtain the following expression for At as a function of the processes for quantity variables: At — P (1 - a) ^ E t ju^t+i^+i^- - p u H)t+2m . lQt In the absence of adjustment costs on Q t (that is, ifH 24 Q t+ 2 Qhi ' = U c ,t+ 1 (4.26) 1 0), equation (4.21) implies that the conditional expectation of Atissimply the expected marginal utility of nominal consumption, as in the standard model. The marginal product of labor enters equation (4.26) due to the cash-in-advance constraint faced by the firms on labor inputs. Note that the model can be solved recursively. According to (4.26), At is a function only of quantity variables, whose determination does not depend on \ t o r R%. Therefore, bond yields can be computed by first solving for the laws of motion of aggregate quantities, using techniques standard in the equilibrium business cycle literature, and then using (4.25) and (4.26) to compute R f . Details of the solution procedure we use can be found in the technical appendix. 4.6. Calibration In choosing parameters for the model, we adhere closely to the equilibrium business cycle literature. First, we choose a , a = 0.36, (3, 7 ,ip, /3 = 1.03-(1/12), 7 = 1, p , 6, and ip = 2 / 3 , x as follows: p = 1.40, 5-0.00667, x = 0.00667. The values of a (capital’s share in the Cobb-Douglas production technology), (3 (the monthly subjective discount factor) and 8 (the monthly capital depreciation rate) are standard choices. The value of 7 implies a logarithmic specification in (4.6). The value of x implies a yearly monetary growth rate of 8%. The wage elasticity of labor supply in this model is 1/ ip , so our choice of ip implies a labor supply elasticity of 1.5. This value is somewhat higher than most microeconomic estimates (for example, Card (1991), Killingsworth (1983) and Pencavel (1986), estimate elasticities near zero for males; whereas it is in the range of 0.5 to 1.5 for females, Killingsworth and Heckman (1986)). However, the implied labor supply elasticity in most real business cycle models substantially exceeds the value we use (for example, Christiano and Eichenbaum’s (1992) model parameter estimates imply a Frisch labor supply elasticity in excess of 5.0). The markup parameter p is at the high end of the range used in the literature. Rotemberg and Woodford (1995) survey the evidence on markups and select a markup of this size. 25 W e set ipQ to imply that in nonstochastic steady state, employment is unity. (This is a normalization that is without loss of generality.) In particular, we set a (3 (1 - a ) //(1 + x ) /i(l + 1—a 1 a(3 x) + 6 - 1 Our calibration implies that ipo = 2.95 . W e set the fixed cost (j> to imply that pure profits are zero in nonstochastic steady state, as follows: * = \ r ) Ka’ where K (427) denotes the nonstochastic steady state stock of capital. Our calibration implies that K = 181.45, so <(>= 1.86.19 There is no literature to draw on in choosing values for the adjustment cost parameters c and and d. H ' Furthermore, the adjustment cost function H (•) is constructed so that both H are zero in the nonstochastic steady state, so steady state properties cannot be used to calibrate c and d. In our baseline calibration we choose c and d to imply a reasonable response of the one-month interest rate to a monetary policy shock. In particular, we set c = 2 and d = 1. Finally, we choose the coefficients {#i}^=0 in the money growth rule (4.1) to match the response of the monetary aggregate to a policy shock (as described in section 4.1). The three sets of values are given in Table 2. 4.7. Implications of the model 4.7.1. Macroeconomic variables Figure 5 displays responses of various macroeconomic aggregates other than the bond yields to a one-standard-deviation monetary contraction. These responses are qualitatively similar to the empirical responses displayed in Figure 1, and often correspond quantitatively as 19Equation (4.27) implies that the ratio of the fixed cost <p to the steady state output level is \x —1. Our parameterization therefore implies that, on average, 40% of output goes to pay the fixed cost. 26 well.20 The money growth process is displayed in row four of Figure 5 and by construction is identical to the first five coefficients of the impulse response functions in row four of Figure 1. Depending on the calibration of the M A process, the price level declines to a level between 20% and 30% below the steady state. By comparison, the price responses in Figure 1 range between 15% (in the C E E identification to 50% (in the Gali identification). However, these empirical responses do not appear to have attained the new steady state after 24 months, while the price response in the model is close to the new steady state after 8 months. As in the empirical results, the output level in the model declines in response to a contractionary monetary policy shock. However, the response in the model is both smaller in magnitude and less persistent than in the data. In particular, the maximal output responses in the model variants is a decline of about 0.045%, as compared to a decline of around 0.2% in the empirical exercises. Furthermore, the maximal response in the model is after three or four months, with the effect of the shock largely dissipated after 10 months. In contrast, the point estimates in allthree empirical exercises show a sustained response even after 24 months (although this long-lived response is significant only for the C E E identification). As noted by Christiano, Eichenbaum, and Evans (1997a), the magnitude of the output response in a limited participation model of this type is determined largely by the elasticity oflabor supply. (A monetary injection increases output by relaxing the firm’s cash-in-advance constraint on labor inputs and clearly the response of labor input to this increase in wages is critical.) For example, when we increase this elasticity to 2.5 (from our baseline calibration of 1.5), the maximal output response increases to 0.085%. Finally, to see whether our specification of the adjustment cost function is reasonable, the last row of Figure 5 plots the response of the time-cost H t of adjusting Q state labor supply is normalized to unity, the units for H t t. (Since the steady are the fraction of steady-state labor used in adjusting Q t - ) As can be seen, the maximal adjustment cost engendered by the monetary policy shock is in the period following the shock, and the magnitude is between 0.0002 and 0.0004 (depending on the model variant). W e conclude that the costs implied by our specification are trivially small. 20 Output, employment, price and money growth have been scaled up by a factor of 100 so the units are percent deviations from steady state. This makes the responses directly comparable to Figure 1. 27 4.7.2. Bond Yields W e find that the model replicates qualitatively the responses of bond yields to monetary policy shocks. To preview our results, we find that a contractionary monetary policy shock causes a positive but transient response in nominal bond yields, and an even bigger positive response in real bond yields. These responses are largest for the short-term yields, with the magnitude of the response declining as maturity increases. There is a positive response in the term premiums for the shorter-maturity bonds. In the rest of this section we describe these findings in greater detail. In Figure 6 we display the responses of yields of different maturities to the contractionary monetary policy shock. In presenting our model’s results, the bond yield responses are annualized percentage point deviations from steady state. These responses are qualitatively similar to those found in the data, but the magnitudes are somewhat too large, at least for the shorter maturities. Consider the C E E calibration of the money growth process: The contemporaneous response of the one-month yield in the model is 38 basis points, rising to 62 basis points three months after the shock. In contrast, our empirical exercises give a contemporaneous response of approximately 20 basis points, rising to a maximal response of 28 or 30 basis points one month later. The contemporaneous response in the model of the six month yield is 59 basis points, which actually exceeds the contemporaneous response of the one-month yield. This is largely due to the persistence of the short-rate response in the model in the model. (Recall that the expectations hypothesis would imply that the contemporaneous response of the six-month rate should equal the average of the first six months’responses of the one-month rate.) However, the high contemporaneous response of the six month rate is also due, in part, to the response of the six-month term premium, as discussed below. The remaining yields’responses decline in magnitude as maturity increases, much as in the empirical exercises. The initial responses of the three- and ten-year yields are 11 basis points and 3 basis points, respectively. These numbers are rather close to the responses found in the data. The responses implied by the Gali calibration are similar to those of the C E E calibration. The responses of the one-month yield implied by the Sims-Zha calibration differs somewhat from the other two, in that the contemporaneous response to the policy shock is rather small (about 15 basis points). Still, the maximal response in this 28 calibration (which occurs 4 months after the shock) is over 60 basis points, substantially exceeding any of the point estimates in the empirical exercises. 4.7.3. Shape of the term structure Figure 7 displays the model’s implications for the i n t e r c e p t , s lo p e and c u r v a t u r e term struc ture descriptors. These responses can be computed directly from the individual bond yield responses as follows. Consider a bond yield of maturity m months, and let R ™ denote its response j months following a monetary policy shock . The j t h response of the three descrip tors — m, in t e r c e p t , s lo p e and c u rv a tu re — can be computed by projecting i?"1 on a constant, and m 2. For each of the model parameterizations reported, the i n t e r c e p t increases for about four months and falls rapidly. The maximal in t e r c e p t responses are about 50 basis points, whereas in the data (Figure 3), the maximal responses are around 18 basis points. This result highlights again that the model’s implications for short-maturity bond yields ex ceed the empirical responses. The s lo p e response falls, consistent with the declining influence of the monetary policy contraction on longer-maturity yields. The quantitative responses of the model and data are quite similar; for example, the C E E impact response in Figure 3 is -0.015 and the model’s impact response is -0.013. Finally, the model’s implication for cu rv a tu re is qualitatively consistent with the empirical responses in Figure 3. However, the magnitude of this effect in the model is severely diminished relative to the data. 4.7.4. T erm premiums The empirical exercises provide some evidence that term premiums increase following a contractionary monetary policy shock. Our model does imply a positive term-premium response. As shown in Figure 7, the six-month term premiums in all three variants of the model display a positive response of 16-18 basis points. This is somewhat larger than the point estimates in our empirical exercises (in which the maximal response was 8 - 1 0 basis points). Furthermore, the term-premium response in the model dissipates rapidly, becoming essentially zero after three or four months. In contrast, the response of the six-month term premium in the empirical exercises appears to increase over the first four or five months. The 29 response of term premiums declines monotonically with maturity, being three basis points for the three-year premium and less than one basis point for the 10-year premium. 4.7.5. Real yields As in the data, we find that the real yields actually display a more pronounced response than nominal yields. (This, of course, is a simple implication of the negative response of the inflation rate.) Figure 8 displays the real yield responses. The responses of shorter-term real yields in the model are larger and less persistent than in the data. For example the maximal response of the six- and 12-month real yields in the C E E calibration of the model are 97 basis points and 53 basis points, respectively. In the data, the maximal responses from the C E E and SZ monetary policy shocks are under 40 basis points. The Gali identification produces a maximal response of 45 points (recall that the Gali monetary policy shocks generate the largest fall in the price level). The maximal response in the model is the contemporaneous response, while the maximal response in the data occurs 4 - 5 months after the date of the shock. The response of the three-year yield in the model (18 basis points) is close to that in the empirical exercises, so the excessive responses in the model appear to be confined to the shorter-term yields. 4.8. Sources of the Liquidity Effect in the Model It is of interest to see which elements of the model account for its ability to replicate the qualitative features found in our empirical exercises. To this end, we experimented with simpler versions of the model. These results are displayed in Figures 9. First, suppose money growth were i.i.d., rather than MA(4). That is, we replace equation (4.1) with xt = (4-28) x + OoEt where, to insure that x t has the same variance as before, we set 0 q = \jY^i=Q O b The results of this exercise are in the first column of Figure 9. Not surprisingly, the general response patterns are quite similar, although the dynamics are less complex. (Basically, all responses 30 appear to die off exponentially.) The price level response is smaller in the i.i.d. case, but this is purely an artifact of our normalizing the variances to be equal.21 The most notable effect of removing serial correlation in money growth is to completely eliminate the term-premium responses. This in turn reduces the responses of the longer nominal yields, and (along with the reduced inflation response) reduces the responses of the real yields. The second column of Figure 9 keeps the MA(4) structure in equation (4.1) (calibrated to the C E E identification), but sets the Q t adjustment cost (in equation (4.5)) to zero. By construction, the impact responses of all variables to the contractionary monetary policy shock are identical to those in the full model. However, the response patterns following the date of the shock differ substantially. In particular, the smooth responses in our baseline calibration are replaced by a rapid reversal of the contemporaneous response. Consider the output response. The initial liquidity effect reduces output, but thereafter the expected inflation effect dominates. That is, the contractionary shock signals that future inflation will be lower. (Indeed, our estimated M A process implies a greater reduction in money growth the period after the shock is observed.) Inflation acts like a distortionary tax, so reduced expected inflation tends to increase output. W e next explore the role of the limited participation feature in the model. Consider first a model x v ith o u t the limited participation features — that is, the household now observes the monetary policy shock before choosing Q t. Furthermore, we fix the adjustment cost at zero, and the money growth rate is i.i.d., as in equation (4.28). This model now behaves as a conventional cash-in-advance model with i.i.d. shocks. The model’s responses are completely neutral. No endogenous variables (other than the price level) respond to the policy shock. That is, there is neither a liquidity effect nor an expected inflation effect. The third column in Figure 9 suppresses both the limited participation constraint and the adjustment costs, but assumes that the money growth rate follows the MA(4) process estimated from the C E E identification. Now, the only effect of a monetary policy shock is the expected inflation effect: As a result, the contractionary monetary shock causes nominal 21That is, y / z l o < X^i=o^»> so while the unconditional variance of xt is the same for both processes, the cumulative effect of a monetary shock on the money stock (and therefore the price level) is greater in the MA(4) specification. 31 interest rates to decline. That is, the impact effect of a contractionary monetary policy shock on bond yields has a counterfactua! response relative to Figure 1. As described above, output responds positively to this reduced expected inflation. Interestingly, the term premium responses are virtually identical to those found in the baseline calibration. Evidently, the term premium responses are driven not by the limited participation feature of the model, but purely by the serial correlation in the money growth process. As a result of this positive term premium response, the response of the intermediate-term nominal yields turns positive after the initial decline. In the fourth column of Figure 9 we add a Q t-adjustment cost (as in equation (4.5)) to the model in the third column. The adjustment cost not only smooths the impulse responses, but also attenuates the contemporaneous response: Households do not reduce their demand for Q t-money as much as they would in the absence of adjustment costs, so less funds flow from the household to the financial intermediary. On impact the variables other than the long yields respond like the cash-in-advance model of column three. After the initial impact, the model’s responses are like the baseline calibration, since unanticipated money movements axe absent. The impact response of the rc-period long yield incorporates the average of the first n responses of the one-month yield, which is positive for moderately large n. A s a result, the impact response of the longer yields is positive. Of course, the positive term premium response magnifies the impact on the long yields. Perhaps the most curious result from these exercises is that the term premium response arises neither from the adjustment costs nor from the limited participation constraint, but purely from the serial correlation in the money growth process. More generally, it is the information content in the monetary policy shock that causes term premiums to respond. Term premiums are a reflection of time-varying conditional higher moments in the logarithm of the marginal utility of wealth22 (our variable At). While the exogenous shocks in this model are conditionally homoskedastic, log(At) responds nonlinearly to these shocks. These nonlinearities in turn can induce conditionally heteroskedastic behavior in log(A£). Evidently, a monetary policy shock conveys not only information about the conditional first moment of future money growth, but also information about conditional higher moments of log(A£). 22For a formal derivation of this result, see Bekaert, Hodrick, and Marshall (1997a). 32 5. C o n c lu sio n s The results of this paper are straightforward and quite intuitive. W e find clear evidence that the short-term effect of monetary policy takes the form of a liquidity effect: A monetary contraction raises interest rates while reducing expected inflation, inducing a pronounced rise in real interest rates. This effect is rather transitory, dissipating between 6 to 12 months after the initial impulse. Monetary policy shocks primarily affect short-term interest rates, with a diminishing effect on longer-term rates. Much of the response of longer-term rates can be explained by the expectations hypothesis. There is some weak evidence that the response of the shorter-maturity bonds is enhanced by a positive response of term premiums. Finally, most of the variance of interest rates is due to sources other than monetary policy. In particular, monetary policy is a non-trivial source of the short-run variability of short term rates, but it represents a negligible source of variability for long-term rates. W e are encouraged by the apparent robustness of these results: These basic patterns emerge under three rather different identification approaches W e also find that a simple dynamic equilibrium model, in which nominal rigidities take the form of a limited participation constraint, is consistent with the broad patterns we have detected in the data. This suggests that our empirical evidence is in no way anomalous, but has a simple equilibrium explanation. W e believe that models of this sort may help us understand the interaction between monetary policy and asset markets generally. Having said this, we note that the model does require some degree of sluggishness in household portfolio adjustment. W e have modelled this by imposing a simple adjustment cost. It would be more satisfactory, from a theoretical standpoint, to be explicit about the microeconomic frictions that underlie this slow aggregate portfolio adjustment. W e also note that our model only includes monetary policy shocks. In principle, nonlin earities of the type we encountered in pricing the zero-coupon bonds imply that the impulse responses to monetary shocks need not be invariant to the presence of other shocks in the model. It is an open question whether these nonlinearities substantially affect the economic analysis. It would be of interest to explore the role of monetary policy in a model of interest rates that incorporates a full set of exogenous impulses, such as technology shocks, prefer 33 ence shocks, and shocks to the transactions technology. (This last type of shock can induces exogenous movements in money demand.) In addition, a model of this sort can be used to explore the effect of changes in the monetary policy reaction function. Such an analysis would surely encounter a host of issues not discussed here. 6. T ech n ica l a p p e n d ix : S o lv in g th e m o d e l W e solve the model using the partial linearization method described in Christiano, Eichenbaum, and Evans (1997c). Let us define qt model has two endogenous state variables: [xt,e t> = Q t/ M t. K t+ 1 and The variable qt is stationary. The qt ] the exogenous state variables are £4—3]• W e can reduce the equilibrium conditions of the model to two Euler equations involving only the processes for these state variables. A linear approximation to the integrands of these Euler equations is taken, and the resulting system of stochastic difference equations is solved in the usual manner to yield a linear law of motion for the endogenous state vector [ K t+ i,q t ] ' (regarded now as differences from the steady state values): K t K t+1 = A qt where A and B + qt- (6.1) B [ x t , e t , £ t ~ i , e t- 2 i S t s } ' 1 are coefficient matrices of the appropriate dimensions. All other variables of interest are known functions of the processes for [ K t , q t - 1]/ and [xt,s(,et-i,£t-2,et-3]7) so they can be computed exactly once laws-of-motion (4.1) and (6.1) are known. That is, no other linear approximations are used other than equation (6.1). In order to compute n-period bond yields, the conditional expectation E t [At+n] must be evaluated. For n equal to 12 months or less we do so by Gauss-Hermite quadrature, using a two-point discretization23 over {et+i, For n equal to 36 months or 120 months, the quadrature procedure is computationally infeasible, so we use a Monte Carlo method. In particular, we simulate a time series (At, K t , q t - i X t >£t> }t=i°°> and we regress At+n 23 When the order of the discretization is increased to six, the implications for the impulse responses we study are virtually unaffected. 34 on a third-order Chebyshev polynomial function of fitted regression as an approximation for E t using the [At+n]. As a check on the accuracy of this procedure, we compute the impulse responses for the 12 month yield using both the Monte Carlo procedure and the quadrature procedure. The resulting impulse responses are virtually indistinguishable. 7. 35 R e fe r e n c e s [1] Bekaert, G., R. Hodrick, and D.A. Marshall (1997a), “‘Peso Problem’ Explanations for Term Structure Anomalies”, Federal Reserve Bank of Chicago Working Paper No. WP-97-7. [2] Bekaert, G., R. J. Hodrick, and D. Marshall, (1997b), ’’The Implications of First-Order Risk Aversion for Asset Market Risk Premiums,” J o u r n a l o f M o n e ta r y E c o n o m ic s , vol. 40, pages 3-41 . [3] Bernanke, Ben S. (1986), ‘Alternative Explanations of the Money-Income Correlation’, C a r n e g i e - R o c h e s t e r C o n f e r e n c e S e r i e s o n P u b l ic P o l i c y , Vol. 25, pages 49 - 99. [4] Bernanke, Ben S. (1990), “On the Predictive Power of Interest Rates and Interest Rate Spreads,” N e w E n g l a n d E c o n o m ic R e v i e w ,Federal Reserve Bank of Boston, Nov/Dec: 51-68. [5] Bernanke, Ben S. and Alan S. Blinder (1992), ‘The Federal Funds Rate and the Channels of Monetary Transmission’,A m e r i c a n E c o n o m ic R e v ie w , Vol. 82, No. 4, pages 901 - 921. [6] Bernanke, Ben S., Mark Gertler, and Mark Watson (1997), “Systematic Monetary Policy and the Effects of Oil Price Shocks,” B r o o k in g s P a p e r s o n E c o n o m i c A c t i v i t y , Vol. 1, pages 91-142. [7] Blanchard, Olivier and Nobu Kiyotaki (1987), “Monopolistic Competition and the Ef fects of Aggregate Demand.” A m e r ic a n E c o n o m ic R e v ie w 77: 647-666. [8] Blanchard, Olivier and Mark Watson (1986), “Are Business Cycles All Alike?” in A m e r i c a n B u s i n e s s C y c le : C o n t i n u i t y a n d C h a n g e , The Robert J. Gordon, ed. (Chicago, IL: University of Chicago Press). [9] Bliss, Robert, 1994, ‘Testing term structure estimation methods,” Working Paper, Fed eral Reserve Bank of Atlanta. [10] Caballero, R.J, 1993, “Durable Goods: An Explanation for their Slow Adjustment,” J o u r n a l o f P o litic a l E c o n o m y 101, 351-384. 36 [11] Card, David (1991), “ Intertemporal labor supply: An assessment,” National Bureau of Economic Research Working Paper No. 3602. “ [12] Christiano, Lawrence J. and Martin Eichenbaum (1992). Current real-business cycle theories and aggregate labor-market fluctuations. A m e r ic a n E c o n o m ic R e m e w 82 (3): 430-50. [13] Christiano, Lawrence J. and Martin Eichenbaum (1995), ‘Liquidity Effects, Monetary Policy and the Business Cycle’, (1995), Jo u rn a l o f M o n ey, C r e d it a n d B a n k i n g , Vol. XXX, Part 1,pages 1113 - 1136. [14] Christiano, Lawrence J., Martin Eichenbaum and Charles L. Evans (1996a), ‘The Effects of Monetary Policy Shocks: Evidence from the Flow of Funds’, R e v ie w a n d S t a t is t ic s o f E c o n o m ic s Vol. 78, No. 1,pages 16 - 34. [15] Christiano, Lawrence J., Martin Eichenbaum and Charles L. Evans (1996b), ‘Identifi cation and the Effects of Monetary Policy Shocks’, in Mario Blejer, Zvi Eckstein, Zvi Hercowitz and Leonardo Leiderman, G ro w th , F i n a n c i a l F a c t o r s i n E c o n o m i c S t a b ili z a t io n a n d Cambridge: Cambridge University Press, pages 36 - 74. [16] Christiano, Lawrence J., Martin Eichenbaum and Charles L. Evans, (1997a), ‘Sticky Price and Limited Participation Models: A Comparison’, E u r o p e a n E c o n o m i c R e v ie w , Vol. 41, No. 6,pages 1201 - 1249. [17] Christiano, Lawrence J., Martin Eichenbaum and Charles L. Evans, (1997b), ‘Monetary Policy Shocks: What Have W e Learned and to What End?’ prepared for the H a n d b o o k o f M a c r o e c o n o m ic s . [18] Christiano, Lawrence J., Martin Eichenbaum and Charles Evans, (1997c) ‘Modeling Money’,manuscript, Northwestern University. [19] Cook, T. and T. Hahn (1989) The Effect of Changes in the Federal Funds Target on Market Interest Rates in the 1970s’, J o u r n a l 37 o f M o n e t a r y E c o n o m ic s , 24: 331-351. [20] Dai, Qiang and Singleton, Kenneth J. (1997), “Specification Analysis of Affine Term Structure Models,” N BER Working Paper No. 6128. [21] den Haan, W. (1995), The Term Structure of Interest Rates in Real and Monetary Economics, J o u r n a l o f E c o n o m ic D y n a m ic s a n d C o n tr o l 19, 909-40. [22] Eichenbaum, Martin, 1992, “Comment on ‘Interpreting the Macroeconomic Time Series Facts: The Effects of Monetary Policy’,” E u r o p e a n E c o n o m i c R e v ie w ,36:1001-1011. [23] Estrella, Arturo and Gikas Haxdouvelis, “The Term Structure as a Predictor of Real Economic Activity,” J o u r n a l o f F in a n c e 46 (June 1991), 555-576. [24] Fuerst, Timothy (1992), ‘Liquidity, Loanable Funds, and Real Activity’, J o u r n a l M o n e t a r y E c o n o m ic s , of Vol. 29, No. 1,February, 3 - 24. [25] Gali, Jordi (1992), ‘How Well Does the IS-LM Model Fit Post War Data?’, Q u a r t e r l y J o u r n a l o f E c o n o m ic s , Vol. 107, No. 2,pages 709 - 738. [26] Gordon, David B. and Eric M. Leeper (1994), ‘The Dynamic Impacts of Monetary Policy: An Exercise in Tentative Identification’, J o u r n a l o f P o lit ic a l E c o n o m y , Vol. 102, No. 6,pages 1228 - 1247. [27] Hardouvelis, G.A., (1994), The Term Structure Spread and Future Changes in Long and Short Rates in the G7 Countries: Is There a Puzzle? J o u r n a l o f M o n e t a r y E c o n o m ic s 33, 255-284. [28] Jorion, P., and F. Mishkin, (1991), A Multicountry Comparison of Term-Structure Forecasts at Long Horizons, [29] Killingsworth, M. 1983. J o u r n a l o f F in a n c ia l E c o n o m ic s L a b o r S u p p ly . 29, 59-80. New York: Cambridge University Press. [30] Killingsworth, M., and Heckman, J. 1986. Female labor supply: A survey. In o f L a b o r E c o n o m ic s , Handbook ed. O. Ashenfelter and R. Layard, Handbooks in Economics Series, no. 5, Vol. 1,pp. 103-204. North-Holland: Elsevier Science. [31] P.J. Knez, R. Litterman, and J. Scheinkman, (1994), Explorations into Factors Explain ing Money Market Returns, J o u r n a l o f F in a n c e 38 49, 1861-1882. [32] Kwon, H. and J.H. McCulloch (1993) U.S. term structure data, 1947-1991’, O h io S ta te U n i v e r s i t y W o r k in g P a p e r , 93-6. [33] Leeper, Eric M., Christopher A. Sims and Tao Zha (1996), ‘What Does Monetary Policy Do?’, B r o o k in g s P a p e r s o n E c o n o m ic A c tiv ity , Vol. 2, pages 1 - 63. [34] Litterman and Scheinkman (1991), Common Factors Affecting Bond Returns, n a l o f F ix e d I n c o m e The Jou r 1,54-61. [35] Lucas, R.E., Jr. 1990. Liquidity and interest rates. J o u r n a l o f E c o n o m ic T h e o r y 50 (2): 237-64. [36] McCallum, B. T., 1994, “Monetary Policy and the Term Structure of Interest Rates,” NBER Working Paper No. 4938. [37] McCulloch, J. H., 1990, ”U.S. Term Structure Data, 1946-87,”Handbook of Monetary Economics Volume I, pp. 672-715. [38] Marshall, D.A., and Parekh, N.G., 1994, “The Effect of Costly Consumption Adjust ment on Asset Price Volatility,” working paper WP-94-21, Federal Reserve Bank of Chicago. [39] Mishkin, F., 1990, “The Information in the Longer Maturity Term Structure about Future Inflation,” Q u a r te r ly J o u r n a l o f E c o n o m ic s 105: 815-828. [40] Pencavel, J. 1986. Labor supply of men: A survey. In H a n d b o o k o f L a b o r E c o n o m ic s , ed. O. Ashenfelter and R. Layard, Handbooks in Economics Series, no. 5, Vol. 1, pp. 3-102.North-Holland: Elsevier Science. [41] Rotemberg, J. and M. Woodford, 1995, ‘Dynamic General Equilibrium Models with Imperfectly Competitive Product Markets’,in F r o n t i e r s o f B u s i n e s s C y c le , ed. Thomas F. Cooley, Princeton University. [42] Schroder, Mark, (1995), “Optimal Portfolio Selection with Fixed Transaction Costs,” Working Paper, State University of New York at Buffalo. 39 [43] Sims, Christopher A. (1980), ‘Macroeconomics and Reality’,E c o n o m e t r ic a , Vol. 48, No. 1,pages 1 - 48. [44] Sims, Christopher A. (1986), ‘Are Forecasting Models Usable for Policy Analysis?’, Federal Reserve Bank of Minneapolis Q u a r t e r l y R e v ie w , Vol. 10, No. 1, pages 2 - 16. [45] Sims, Christopher A. (1992), “Interpreting the Macroeconomic Time Series Facts: The Effects of Monetary Policy,” E u r o p e a n E c o n o m i c R e v ie w ,36:975-1000. [46] Sims, Christopher A. and Tao Zha (1995a), ‘Does Monetary Policy Generate Reces sions?’,Manuscript. Yale University. [47] Sims, Christopher A. and Tao Zha (1995b), ‘Error Bands for Impulse Responses,’Work ing paper number 95-6, Federal Reserve Bank of Atlanta. [48] Taylor, John (1993), “Discretion versus Policy Rules in Practice,” C o n f e r e n c e S e r ie s o n P u b l i c P o l i c y C a m e g ie - R o c h e s t e r 39, December 1993, 195-214. [49] Watson, Mark (1995), “Vectorautoregressions and Cointegration,” in the E c o n o m e t r ic s . 40 Handbook o f Table 1 Fraction of yield variance explained by monetary policy shock A. C E E Identification Bond Maturity Horizon One month Six months One year Three years Ten years 1-month 16.7% 14.8% 11.0% 6.5% 2.9% 6-months 16.9% 12.9% 10.8% 4.8% 1.5% 24-months 6.9% 5.3% 4.6% 3.0% 1.4% B. SZ Identification Bond Maturity Horizon One month Six months One year Three years Ten years 1-month 13.6% 12.3% 10.2% 6.6% 1.5% 6-months 18.1% 6.9% 8.7% 7.6% 3.6% 24-months 15.4% 3.7% 11.3% 14.5% 10.3% C. Gali Identification Bond Maturity Horizon One month Six-months One year Three years Ten years 1-month 7.2% 4.9% 3.5% 1.9% 0.0% 6-months 17.1% 13.7% 11.2% 5.0% 3.2% 24-months 7.4% 2.9% 2.9% 2.2% 0.2% 41 Notes to Table 1 : The estimated fraction of the one-month ahead, six-month ahead, and 24-month ahead conditional bond-yield variance attributable to monetary policy shocks is displayed. Panel A displays results for the C E E recursive strategy for identifying monetary policy shocks (described in Section 2.2); Panel B displays results for the Sims-Zha nonrecursive identification strategy (described in Section 2.3); and Panel C displays results for the Gali identification strategy that incorporates long-run restrictions (described in Section 2.4). For each identification strategy, results are displayed for bonds maturing in one month, six months, one year, three years, and ten years. 42 Table 2 : Parameter Calibrations for the M o n e y Growth Process Identification Strategy 00 0i 02 03 CEE .00041 .00081 .00067 .00038 -.00008 Gali .00047 .00087 .00082 .00036 .00035 SZ .00016 .00092 .00064 .00099 .00069 04 Notes for Table 2: This table gives the values of the moving average coefficients {0i}*=o in money growth process (4.1) implied by our estimates of the C E E recursive strategy for identifying monetary policy shocks (described in Section 2.2); the Sims-Zha nonrecursive identification strategy (described in Section 2.3); and the Gali identification strategy that incorporates long-run restrictions (described in Section 2.4). 43 C EE Recursive Sims-Zha Gali MP Shock=>EM MP Shock=>GDP MP Shock=>GDP months MP Shock=>PCEPriceDeflator months MP Shock=>PCEPriceDeflator months MP Shock=>CPIPrice § s. & months MP Shock=>FF months MP Shock=>FF months MP Shock=>FF § o & months MPShock=>GrowthinTR I £ months MPShock=>CommodityPrices 0.036 0.018 •0.054 ■0.072 months MP Shock=>NBRX & 0.14•0.14■02*-0.42•0.56■0.70- 0.10- \ /N 0.05-0.00-0.05-0.10------& 0.15-020- / V _. , __ -025months MP Shock=>CrudeMaterialsPrices 0.50 025 •025 •0.50 & -0.75 -1.0 0 -125 -1.50 months MP Shock=>IntermediateMaterialsPrices months MP Shock=>GrowthinM1 percent months MP Shock=>GrowthinM2 0.100.05-0.06-0.10-0.15•020- r'.*Nv ^ \ ^ ^- __ s J ) s. J/ •0.50-0.75 -1.00 months MP Shock=>RealWages months o & 0.15 0.10 0.05 •0.00 -0.05 •0.10 •0.15 •020 Figure 1: Responses of Macroeconomic Variables to Monetary Policy Shocks __ months Notes to Figure 1: For each of the three identification strategies described in section 2, this figure displays the responses of variables in the V A R system (other than bond yields) to a onestandard deviation monetary policy shock. The first column displays impulse responses for the CEE recursive identification, described in section 2.2. The responses reported are: employment (EM), personal consumption expenditure deflator (PCE Price Deflator), federal funds rate (FF), growth in M 2 money, an index of Commodity Prices, and the ratio of non-borrowed reserves to total reserves (NBRX). The second column displays impulse responses for the Sims-Zha nonrecursive identification, described in section 2.3. The responses reported are: GDP, personal consumption expenditure deflator (PCE Price Deflator), federal funds rate (FF), growth in total reserves (TR), an index of Crude Material Prices, an index of Intermediate Material Prices, and Real Wages. The third column displays impulse responses for the Gali identification, described in section 2.4. The responses reported are: GDP, consumer price index (CPI), federal funds rate (FF), and growth in Ml money. For all variables except the federal funds rate, the responses are in percentage deviations from the steady state. For the federal funds rate, the units are percentage points per annum. The solid lines plot the point estimates for the impulse responses; dashed lines give 95% confidence intervals, as described in sections 2.2 - 2.4. Sims-Zha Gali MP Shock=>FF MP Shock=>FF MP Shock=>FF 1 6 11 months 16 21 S» ^_ 6 11 months 16 21 / 1 11 1 °‘ ° -0.2•0.4- (/ 1 1 i | Q. 2 © Q l 0.80.60.4| 0.2Q. 00" -0.2-0.4- / // - // 0.80.80.4- >> CEE Recursive --— _______ months MP Shock=>1-monthyield MP Shock=>1-monthyield MP Shock=>1-monthyield months months months MP Shock=>6-monthyield MP Shock=>6-monthyield MP Shock=>6-monthyield 0o) 0) o. ♦§ 1 c s? 2L \y 6 11 months ^ 16 21 0.50.4°J0.2- ✓' •0.1- \V. ■0.2■........« — -0.3-.. .it.Trv 6 11 16 21 months 0.50.4^ 0.3<5 0.2mP 0.1C ’ -0.1•0.2-0.3- B 11 16 months MP Shock=> 12-monthyield MP Shock=>12-monthyield ______ MP Shock=>12-monthyield c I & 2L a 1 months 6 11 months 16 MP Shock=>3-yearyield MP Shock=>3-yearyield MP Shock=>3-yearyield months months months MP Shock=>10-yearyield MP Shock=>10-yearyield MP Shock=>10-yearyield 0.25 0.20 © O 0.15 0.10 a 0.05 •0.05 •0.10 Figure 2: Responses of Bond Yields to Monetary Policy Shocks 21 months Notes to Figure 2: For each of the three identification strategies described in section 2, this figure displays the response to a one-standard deviation monetary policy shock of the federal funds rate (first row) and of the continuously-compounded yields for zero coupon bonds of maturities one month, six months, one year, three years, and ten years (rows two through six). The first column displays impulse responses implied by CEE recursive identification (described in section 2.2); The second column displays impulse responses implied by the Sims-Zha nonrecursive identification (described in section 2.3); the third column displays impulse responses implied by the Gali identification (described in section 2.4). For all these impulse responses, the units are percentage points per annum. The solid lines plot the point estimates for the impulse responses; dashed lines give 95% confidence intervals, as described in sections 2.2 - 2.4. C E E Recursive Sims-Zha Gali MP Shock=>TS Intercept MP Shock=>TS Intercept MP Shock=>TS Intercept months months months MP Shock=>TS Slope MP Shock=>TSSlope MP Shock=>TS Slope months months 6 11 months MP Shock=>TS Curvature MP Shock=>TSCurvature MP Shock=>TS Curvature months months months MP Shock=>6-monthTermPremium MP Shock=>6-monthTermPremium MP Shock*>6-monthTermPremium months months 6 11 months MP Shock=>12-monthTermPremium MP Shock=>12-monthTerm Premium MP Shock=>12-monthTermPremium months months months MP Shock=>3-yearTermPremium MP Shock=>3-yearTermPremium MP Shock=>3-yearTermPremium o <D C L a 1 16 21 & 1 16 21 0.25 0.20 0.15 0.10 0.05 •0.05 •0.10 -0.15 Figure 3: Responses of Term Structure Descriptors and Term Premiums to Monetary Policy Shocks Notes to Figure 3: The interpretation of Figure 3 is identical to that of Figure 2, except that, instead of displaying responses of bond yields to a monetary policy shock, the first three rows display responses of the three term structure descriptors in te rc e p t, s lo p e , and c u r v a t u r e . These term structure descriptors are computed as described in Section 3.1. The last three rows display the responses of the T-period term premium, T P ] , defined in equation (3.1). C E E R e c u rsiv e S im s-Z h a G a li MP Shock => 6-month real rate MP Shock => 6-month real rate MP Shock => 6-month real rate months months months MP Shock => 12-month real rate MP Shock => 12-month real rate MP Shock => 12-month real rate months months months MP Shock => 3-year real rate MP Shock => 3-year real rate MP Shock => 3-year real rate ____________________ m n n t h c _______________________________________ m n n t h c ________________________________________m n n t h c Figure 4: Responses of Real Interest Rates to Monetary Policy Shocks Notes to Figure 4: The interpretation of Figure 4 is identical to that of Figure 2, except that, instead of displaying responses of nominal bond yields to a monetary policy shock, the responses of real bond yields of maturities six months, 12 months, and three years are displayed. The first column displays impulse responses for the CEE recursive identification, described in section 2.2. The second column displays impulse responses for the Sims-Zha non-recursive identification, described in section 2.3. The third column displays impulse responses for the Gali identification, described in section 2.4. In computing the real bond yields, the CEE identification (column 1) and the Sims-Zha identification (column 2) compute the inflation rate using the personal consumption expenditure deflator; the Gali identification (column 3) uses CPI inflation. In all cases, units are in percentage points per annum. CEE Recursive Sims-Zha Gali MP Shock => Output, Employment MP Shock => Output, Employment MP Shock => Output, Employment months months months MP Shock => Price MP Shock => Price MP Shock => Price months months months MP Shock => 1-month rate MP Shock => 1-month rate MP Shock => 1-month rate months months months MP Shock => Growth in Money MP Shock => Growth in Money MP Shock => Growth in Money months months months MP Shock => Adjustment Cost MP Shock => Adjustment Cost MP Shock => Adjustment Cost c © o 6 11 Figure 5: Responses of Macroeconomic Variables to Monetary Policy Shocks in the Calibrated Limited Participation Model Notes to Figure 5: This figure displays the responses to a one-standard deviation monetary policy shock, implied by the model described in sections 4.1 - 4.4, of output, the price level, the one-month interest rate, the growth rate of money, and the adjustment cost of Q , (as given in equation (4.5)). For output, the price level, and the growth rate of money, the responses are in units of percentage deviations from the steady state. For the one-month interest rate, the units are percentage points per annum. For the adjustment cost, the units are fractions of time devoted to employment in the steady state. The three columns calibrate the law of motion for money growth (equation (4.1)) to estimates from the CEE identification, the Sims-Zha identification, and the Gali identification, as given in Table 2. The remaining parameter calibrations are: a = 0.36, P = 1.03'(1/12), a = 1, \|/ = 2/3, p = 1.40, 6 = 0.00667, x = 000667, c = 2, d = 1. CEE Recursive Sims-Zha Gali MP Shock => 1-month yield MP Shock => 1-month yield MP Shock => 1-month yield months months months MP Shock => 6-month yield MP Shock => 6-month yield MP Shock => 6-month yield months months months MP Shock => 12-month yield MP Shock => 12-month yield MP Shock => 12-month yield c <D O 1 6 11 16 1 6 11 16 months months MP Shock => 3-year yield MP Shock => 3-year yield § g <D MP Shock => 3-year yield 0.30-0.25- 0.30-0.25- 0.20 - 0.20- 0.150.10- 0.050.00-- -0.05-0.10 -- 1 6 11 16 -0.050.10 - I T . . . . ......................... 1 6 11 !- ■ ■■ 16 months months months MP Shock => 10-year yield MP Shock => 10-year yield MP Shock => 10-yearyield Figure 6: Responses ofBond Yields to Monetary Policy Shocks inthe Calibrated Limited Participation Model Notes to Figure 6: The interpretation of Figure 6 is identical to that of Figure 5, except that the responses, implied by the model of sections 4.1 -4.4, of the continuously-compounded yields for zero coupon bonds of maturities one month, six months, one year, three years, and ten years are displayed. The units are percentage points per annum. The calibration is as in Figure 5. CEE Recursive Sims-Zha Gali MP Shock => TS Intercept MP Shock => TS Intercept MP Shock => TS Intercept months months months MP Shock => TS Slope MP Shock => TS Slope MP Shock => TS Slope months months 6 11 months MP Shock => TS Curvature MP Shock => TS Curvature MP Shock => TS Curvature months months MP Shock => 6-month term premium MP Shock => 6-month term premium & 1 1 6 11 18 MP Shock => 6-month term premium 16 months months months MP Shock => 12-month term premium MP Shock => 12-month term premium MP Shock => 12-month term premium 0.100C <D E 8. 0.0750.0500.025-0.025- months 1 6 11 months MP Shock => 36-month term premium MP Shock => 36-month term premium ^ § S a E)A 0.040.030.020.010.00■0.01• ■0.02•0.03- 10 months MP Shock => 36-month term premium 11 1 ft Figure 7: Responses of Term Structure Descriptors and Term Premiums to Monetary Policy Shocks in the Calibrated Limited Participation Model Notes to Figure 7: The interpretation of Figure 7 is identical to that of Figure 5, except that the responses, implied by the model of sections 4.1 -4.4, of the three term structure descriptors in te r c e p t, s lo p e , and c u r v a tu r e (rows 1 - 3) and of the 6-month, 12-month, and 36-month term premiums are displayed. The three term structure descriptors are computed as described in section 3.1. The T-period term premium, T P ] is defined in equation (3.1). The calibration is as in Figure 5. CEE Recursive Sims-Zha Gali MP Shock => 1-month real rate MP Shock => 1-month real rate MP Shock => 1-month real rate months months months MP Shock => 6-month real rate MP Shock => 6-month real rate MP Shock => 6-month real rate months months months MP Shock => 12-month real rate MP Shock => 12-month real rate MP Shock => 12-month real rate months months months MP Shock => 36-month real rate MP Shock => 36-month real rate MP Shock => 36-month real rate months months months MP Shock => 120-month real rate MP Shock => 120-month real rate MP Shock => 120-month real rate c© O <D Q. Figure 8: Responses ofReal Interest Rates to Monetary Policy Shocks in the Calibrated Limited Participation Model Notes to Figure 8: The interpretation of Figure 8 is identical to that of Figure 5, except that the responses, implied by the model of sections 4.1 -4.4, of the real bond yields of maturities one month, six months, 12 months, three years, and ten years are displayed. The units are percentage points per annum. The calibration is as in Figure 5. Lim. Part., Adjust. Costs,iid Lim. Part., No Adjust Costs, MA(4) No Lim. Part., No Adjust. Costs, MA(4) No Lim. Part., Adjust. Costs, MA(4) MP Shock => Output MP Shock => Output MP Shock => Output MP Shock => Output © o G> Q. 1 4 7 months months months months MP Shock => Price MP Shock => Price MP Shock => Price MP Shock => Price 0.16• — --------- 1 .... 0.10- 0.16- 8 V © Q. •4X32• °*° 41.16• 1 C ^ © Q. -0.16■ 8 V -0.32- 4 7 4X3S- 10 months months months months MP Shock => 1-month yield MP Shock => 1-month yield MP Shock => 1-month yield MP Shock => 1-month yield c 0o © © \ _ V -0.7- 0.7- 4X7- 4X7- 7 10 months months months MP Shock => 6-month yield MP Shock => 6-month yield MP Shock => 6-month yield MP Shock => 6-month yield ? ? £ percent ? _"v. ------------------------ £ 4)26- 10 ? “ • 00 7 (XiO- 0*0percent | V 7 10 months months months months MP Shock => 12-month yield MP Shock => 12-month yield MP Shock => 12-month yield MP Shock => 12-month yield 032\ | ojoo S. 032- 032• | 000 y_ Q- 0.00 _ ------------ 1 8. \ 4X16- 4X16months months months months MP Shock => 3-year yield MP Shock => 3-year yield MP Shock => 3-year yield MP Shock => 3-year yield 0.10006■ c “ 8 & 8 & 7 c 4 months 0.50■ Q. jercent 1.4* 10 4X10- 4 7 C 006 8i -006- 4X06■ -aio- 10 aio- Oj0 &- 1 4 7 7 10 10 months months months months MP Shock => 10-year yield MP Shock «> 10-year yield MP Shock => 10-year yield MP Shock => 10-year yield 0-025 c 8 & 8 & 7 10 Figure 9: Im pulse Responses Im plied by V ariants o f the M odel o f Section 4 C 8 8. 4X005 -0.060 Lim. Part., Adjust. Costs,iid Lim. Part., No Adjust. Costs, MA(4) No Lim. Part., No Adjust. Costs, MA(4) No Lim. Part., Adjust. Costs, MA(4) MP Shock => 6-month term premium MP Shock => 6-month term premium MP Shock => 6-month term premium MP Shock => 6-month term premium months months months months MP Shock => 12-month term premium MP Shock => 12-month term premium MP Shock => 12-month term premium MP Shock => 12-month term premium months months months months MP Shock => 36-month term premium MP Shock => 36-month term premium MP Shock => 36-month term premium MP Shock => 36-month term premium months months months months MP Shock => 1-month real rate MP Shock => 1-month real rate MP Shock => 1-month real rate MP Shock => 1-month real rate 0.2 --------------------------------- c 0 a,‘ o 0 -------------CL -O.tJ-i-. -. ---. -.-■ -r—«-. -.---- 0.100 -I--------------------------------- C g Q. oooo---------------------------------------------------------------- •0.040 .-■* ■ . . ---- - -----, -. -1 4 7 10 © O 0 CL 002 . ooo - ■ota - -- '■ ' — .— .— ,— .— 1 4 ■ — 7 — .— ---- 10 2.4 C 0 2 0 o. 00 1 . - — ........ — •a* - —— ,— ,— .— _ I 4 ............— — ,— ,— ,— ,— ,—-------- 7 to 1 4 7 M> months months months months MP Shock => 6-month real rate MP Shock => 6-month real rate MP Shock => 6-month real rate MP Shock => 6-month real rate 0.75 -0.25 1 4 7 10 months months months months MP Shock => 12-month real rate MP Shock => 12-month real rate MP Shock => 12-month real rate MP Shock => 12-month real rate aso •0.25^ --1 ---.-. -. -1 -, ---1 -1 -1 -1 4 7 10 1 4 7 10 months months months months MP Shock => 3-year real rate MP Shock => 3-year real rate MP Shock => 3-year real rate MP Shock => 3-year real rate 02 02 c © ------------------ o M8. o,.-r=^_________ 0.1 - 'figure 9 (Continued) Notes to Figure 9: The responses implied by four simple variants of the model of sections 4.14.4 are displayed for the following variables: output (row one); the price level (row two); bond yields of maturities 1 month, 6 months, 12 months, 3 years, and 10 years (rows three through seven); term premiums of maturities 6 months, 12 months, and 3 years, as defined in equation (3.1) (rows eight through ten); and real bond yields of maturities one month, 6 months, 12 months, and three years (rows 11 through 14). The simple model variants are as follows. Column 1: baseline model of Figure 5, but with money growth i.i.d.; Column 2: baseline model of Figure 5, but with adjustment cost set equal to zero; Column 3: baseline model of Figure 5, but without the limited participation constraint, and with adjustment cost set equal to zero; Column 4: baseline model of Figure 5 without the limited participation constraint, but with adjustment costs as in Figure 5.