The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of Chicago The Term Structure and Inflation Uncertainty Tomas Breach, Stefania D’Amico, and Athanasios Orphanides December 2016 WP 2016-22 The Term Structure and In‡ation Uncertainty Tomas Breach, Stefania D’Amico, and Athanasios Orphanides December 12, 2016 Abstract This paper develops and estimates a Quadratic-Gaussian model of the U.S. term structure that can accommodate the rich dynamics of in‡ation risk premia over the 1983-2013 period by allowing for time-varying market prices of in‡ation risk and incorporating survey information on in‡ation uncertainty in the estimation. The model captures changes in premia over very diverse periods, from the in‡ation scare episodes of the 1980s, when perceived in‡ation uncertainty was high, to the more recent episodes of negative premia, when perceived in‡ation uncertainty has been considerably smaller. A decomposition of the nominal ten-year yield suggests a decline in the estimated in‡ation risk premium of 1.7 percentage points from the early 1980s to mid-1990s. Subsequently, its predicted value has ‡uctuated around zero and turned negative at times, reaching its lowest values (about -0.6 percentage points) before the latest …nancial crisis, in 2005-2007, and during the subsequent weak recovery, in 2010-2012. The model’s ability to generate sensible estimates of the in‡ation risk premium has important implications for the other components of the nominal yield: expected real rates, expected in‡ation, and real risk premia. Keywords: Quadratic-Gaussian Term Structure Models, In‡ation Risk Premium, Survey Forecasts, Hidden Factors. JEL Classi…cation: G12, E43, E44, C58 Breach: Federal Reserve Bank of Chicago. E-mail: tbreach@frbchi.org. D’Amico: Federal Reserve Bank of Chicago. E-mail: sdamico@frbchi.org. Orphanides: MIT Sloan School of Management. E-mail: athanasios.orphanides@mit.edu. For helpful comments, discussions, and suggestions we thank Bob Barsky, Alejandro Justiniano, Don Kim, Thomas King, Anna Paulson, Hiro Tanaka, Min Wei, and seminar participants at the Federal Reserve Board and the Federal Reserve Bank of Chicago. The views expressed here do not re‡ect o¢ cial positions of the Federal Reserve. 1 1 Introduction Longer-term nominal yields contain rich information about real interest rates and in‡ation rates that market participants expect to prevail in the future. Extracting this valuable information, however, is complicated by the presence of unobservable in‡ation risk premia (IRP) and real risk premia (RRP) that are widely acknowledged to vary over time. Monetary policymakers are keenly interested in understanding these premia for multiple reasons. The IRP embedded in nominal yields may re‡ect factors such as uncertainty about in‡ation and the credibility of the monetary authority (e.g., Argov et al. 2007; Palomino, 2012; Du et al., 2016), which may evolve over time with the ability of the central bank to successfully communicate its policy strategy and deliver on its in‡ation objective. Changes in perceptions of in‡ation risks, such as the in‡ation scare episodes of the 1980s (Goodfriend, 1993) or the risk of de‡ation following the last …nancial crisis (Kitsul and Wright, 2013), may cause abrupt changes in long-term yields not necessarily associated with shifts in expectations of future interest rates or in‡ation. At times, signi…cant changes in risk premia may complicate the transmission of monetary policy to longer-term rates, particularly when premia move in the direction opposite to expectations of short-term rates, as has apparently been the case in the "conundrum" period (2004-06) and the "taper tantrum" episode (May-June 2013). Understanding in‡ation uncertainty and associated risk premia is important for the appropriate risk management of monetary policy. (Evans et al, 2015; and Feldman et al, 2016). In recent years, a new generation of dynamic term structure models has been developed to estimate the various components of the term structure by allowing ‡exible speci…cations of risk premia while maintaining analytical tractability (e.g., Dai and Singleton, 2000; Du¤ee, 2002). Despite considerable progress in modelling the term structure, estimation of the IRP has proven challenging. Alternative speci…cations estimated over di¤erent periods have resulted in a broad range of results (surveyed in Bekaert and Wong, 2010). Term structure models estimated using data prior to the last …nancial crisis (e.g., Ang, Bekaert, and Wei, 2008; Buraschi and Jiltsov, 2005; and Chernov and Mueller, 2012) report estimates of the IRP that are larger in magnitude and mainly positive. In contrast, models focusing on more recent data (e.g., Abrahams et al, 2013; Grishchenko and Huang, 2013; and Fleckenstein, Longsta¤, and Lustig, 2014), deliver values of the IRP that are smaller in magnitude and often negative, especially at shorter maturities. One factor that could explain ‡uctuations in the IRP over time is the variation in the level of actual and perceived in‡ation uncertainty. In‡ation uncertainty makes nominal bonds risky, as their real value is eroded by surprise in‡ation, and thus is expected to a¤ect the associated risk premium. Although the speci…c channels may di¤er, a relation between in‡ation uncertainty and the IRP emerges in numerous models, as highlighted for example in the survey by Gurkaynak and Wright 2 (2012). In the data, both actual in‡ation volatility and survey-based in‡ation uncertainty have declined notably since the 1980s (D’Amico and Orphanides, 2008), as the Federal Reserve adopted policies that gradually reestablished its credibility to keep in‡ation low and stable, following a period of monetary neglect. And as shown by D’Amico and Orphanides (2014), real-time measures of perceived in‡ation uncertainty contain meaningful information about future nominal bond excess returns that is not contained in current yields or forward spreads. Another potentially critical factor for the evolution of the IRP is the changing covariance between bond and stock returns, which a¤ects the hedging characteristics of Treasury nominal bonds. As indicated in Campbell, Sunderam, and Viceira (2016), for example, the stockbond covariance was high and positive in the early 1980s but became negative in the 2000s, which in their model mainly re‡ects time-variation in in‡ation volatility and in the covariance between in‡ation and the real economy, with both accounting for signi…cant changes in the sign and size of nominal bond risk premia. We develop a Quadratic-Gaussian term structure model that is ‡exible enough to encompass very diverse dynamic behaviors of the IRP over extreme episodes like the early 1980s, characterized by high actual and expected in‡ation as well as high in‡ation uncertainty, and the post-2008 period, characterized by low in‡ation (and mild de‡ation) as well as very low expected in‡ation and in‡ation uncertainty. The richer dynamic of the IRP is achieved by having time-varying market prices of in‡ation risk in the model, which translates into time-varying in‡ation volatility and time-varying covariances between yield curve factors and in‡ation. To obtain reliable estimates of the parameters governing these sources of in‡ation risk and the IRP, we incorporate information from survey-based in‡ation uncertainty in the estimation. Using this new data input, which captures real-time perceptions of in‡ation risk, proves quite valuable for pinning down the dynamics of the IRP which, in turn, has important implications for the other components of nominal yields: expected real short-term rates, expected in‡ation, and the RRP. The key novelty of this approach is to tackle the di¢ culties in the estimation of the IRP by allowing for very ‡exible market prices of in‡ation risk and by using a real-time measure of in‡ation uncertainty. Introducing survey-based information about second moments may also be seen as an extension of the approach developed in Kim and Orphanides (2012) and Chernov and Mueller (2012) who augment term structure models with information from survey …rst moments. Similarly to the …rst of these studies, we include survey forecasts of short-term interest rates to guard against estimation imprecision and bias due to the highly persistent nature of interest rates; and, similarly to the second study, we include survey forecasts of in‡ation that help pin down expected in‡ation and real rates over a long sample period for which TIPS yields are not available. To account for the possibility that survey forecasts provide noisy information and to let the data determine the extent of this noise, following Kim and Orphanides (2012) we allow for unconstrained variances of the measurement errors around all 3 forecasts in the estimation. Acknowledging the presence of measurement errors is particularly important for survey-based in‡ation uncertainty, which is an imputed variable derived from the subjective probability distributions in the Survey of Professional Forecasters (SPF) using the methodology in D’Amico and Orphanides (2008). In addition, following the intuition in Du¤ee (2011) and motivated by preliminary regressions similar to those conducted by Joslin, Priebsch, and Singleton (2014), in our model in‡ation-related variables are not fully spanned by the current nominal yield curve. This is achieved through two modelling choices. First, we introduce a shock capturing short-run variations in CPI in‡ation that do not require a monetarypolicy response, such as, short-lived changes in energy and food prices. Second, one of the factors is hidden in the nominal yield curve and can only in‡uence the …rst and second moments of in‡ation. In principle, both of these features can be important because, while the conditional volatility of the Brownian shock speci…c to CPI a¤ects in‡ation uncertainty and risk premia but not the forecast of in‡ation (i.e., its expected value is zero), the hidden factor can in‡uence expected in‡ation. Moreover, the conditional volatility of the innovations to the hidden factor can also contribute to ‡uctuations in in‡ation uncertainty and risk premia. The resulting model produces IRP estimates at the 10-year horizon that are larger and positive in the 1980s, then decline by about 1.7 percentage points by the mid 1990s, and subsequently become negative at times, for example during the conundrum period (2005-07) and during the de‡ation scare of 2010-2012. The estimates also capture episodes of sharp increases in the IRP, as for instance during the taper tantrum of May-June 2013. Further, despite being estimated without the use of TIPS yields, the model generates real yields that closely resemble those on TIPS, except for a residual very similar to the liquidity premium estimated in D’Amico, Kim, and Wei (2016). The model also does a good job at …tting the survey-based one-year expected in‡ation and in‡ation uncertainty, and it produces expected short-term rates that closely match those from survey forecasts. With regard to longer-term trends, the model captures the decline in long-term in‡ation expectations from the 1980s to the 2000s that is associated with the Federal Reserve’s overall disin‡ationary policies over this period as well as the decline in long-term expectations of the short-term real interest rate re‡ecting the decline in the equilibrium real interest rate. The rest of the paper is organized as follows. Section 2 sets up the QuadraticGaussian model. Section 3 compares our model to other key studies in the literature. Section 4 presents the state-space form of the model and the data. Section 5 provides details about the identi…cation and estimation methodology. Section 6 presents the main model’s empirical results and comparisons with alternative speci…cations, which help assess the contribution of key elements to its improved performance. Section 7 o¤ers concluding remarks. 4 2 A Quadratic Gaussian Model In this section we develop a Quadratic Gaussian model of the term structure of interest rates that accommodates nominal yields, CPI in‡ation, survey-based expected in‡ation, expected interest rates, and in‡ation volatility. 2.1 The basic building blocks We start with specifying the state factor dynamics under the physical measure, P: dxt = dzt x 2 2 01 2 02 1 z 1 1 ( ( xt )2 zt )1 x z 1 dt + x 2 2 x;z 1 02 1 z dBtx dBtz where zt is a factor hidden in the nominal yield curve in the sense of Du¤ee (2011) but its shocks can be correlated with shocks relevant for nominal interest rates as indicated by the unconstrained x;z , Bt denotes a 3-dimensional standard Brownian motion, and thus all the factors are Gaussian. The nominal pricing kernel under P is given by: dMtN = MtN N0 x t dBt rtN dt where the nominal short rate is an a¢ ne function of only two state variables: rtN = N 0 + N0 1 xt ; and the 2-dimensional vector of the market price of nominal risk is given by: N t = N 0 + N xt ; with N being a 2 2 constant matrix that will be left unrestricted to allow for a ‡exible speci…cation of the market price of nominal risk. However, by preventing N t and rtN from loading on zt , we make sure that this factor is unspanned by nominal yields.1 The log price level follows the process d log Qt = t dt + q0 x t dBt + ! t dBt? and is governed by xt = [xt ; zt ] rather than xt , as the factors underlying the nominal yields are not su¢ cient to span in‡ation and expected in‡ation, which is a¢ ne in xt : 0 t = 0 + 1 xt ; 1 This can be also achieved by imposing restrictions under the risk-neutral measure as in Du¤ee (2011), and we veri…ed that results are not very sensitive to the way we impose the unspanning restrictions. 5 and the two conditional volatility processes are given by q t = q0 + q xt ; ! t = ! 0 + ! 01 xt ; where 0 and ! 0 are scalars, 1 and q0 are 3 1 vectors, ! 01 is a 1 3 vector, q is a 3 3 matrix, and dBtx dBt? = 0. The orthogonal shock speci…c to the in‡ation process is supposed to capture, for instance, short-run variations in in‡ation that do not require a monetary-policy response and thus do not a¤ect the nominal short rate (Kim, 2008). In particular, since we use total CPI in our estimation, not only it is important to have a separate shock for CPI innovations driven by changes in food and energy prices, but since these components are usually more volatile it is key to allow for time variation in the conditional volatility of this shock. Overall, this implies that we treat much of high-frequency variation in in‡ation as unspanned by interest rates. The real pricing kernel is given by MtR = MtN Qt , which by Ito’s Lemma follows the dynamics: dMtR dMtN dQt dMtN dQt = + + = Qt MtR MtN MtN Qt R0 x t dBt rtR dt ! t dBt? where the real short rate becomes a quadratic function of the state variables because dM N t , as each of these elements contains a state-dependent of the interaction term M Nt dQ Qt t N market price of risk, that is, q (xt ) and rtR = R 0 + (xt ), respectively: R0 1 xt + xt 0 R xt ; all parameters are linked by the no-arbitrage conditions: 2.2 R 0 = N 0 0 R 1 R t = = N 1 N t 1 q t R = N0 q + N0 q 0 0 + q0 N 0 1 2 q0 1 ( 2 q0 q 0 0 N0 q 0 + + ! 20 ) q0 q 0 !0!1 1 ! 1 ! 01 : 2 q Bond Pricing Under the risk-neutral measure, Q, xt follows the dynamics: dxt = ( = = e (e dBtx + xt ) dt + xt i 0 i xt ) dt + dBt 6 i t dt xt dt + i t dt dBtx + i t dt where e= i + i 0 i t dt ee = dBt = dBtx + and i = N; R indicating either the nominal or real risk neutral measure. The price of a nominal and real zero-coupon bond with maturity is: Pt;i i Et (Mt+ ) = = EtQ exp i Mt i i0 = exp A + B xt + t+ R rsi ds t x0t C i xt ; i = N; R with the solution satisfying the following di¤erential equations: dAi = d dB i = d dC i = d i 0 i 1 i 1 + B i0 ee + B i0 2 0 B i + tr e0 B i + 2C i ee + 2C i Ci e e0 C i + 2C i 0 0 0 Ci Bi C i0 : In the case of nominal bonds (i.e., i = N ), C N = 0, as in this model we start with specifying an a¢ ne nominal short rate and the real short rate inherits the quadratic component through the no arbitrage condition MtR = MtN Qt , and therefore nominal bonds’prices preserve the same functional form usually obtained in a¢ ne Gaussian models. It follows that since yt;i = 1 log(Pt;i ); nominal and real yields are equal to: yt;N = aN + bN 0 xt yt;R = aR + bR0 xt + xt 0 cR xt ; where ai = 2.3 1 A i , bi = 1 1 B i ; and ci = Ci . In‡ation: Expected and Unexpected In‡ation between t and t + it; , 1 Qt+ 1 log = Qt is de…ned as: Z (xt+s )ds + 0 Z q (xt+s ) dBsx 0 0 and annual average expected in‡ation over horizon 7 + Z !(xt+s )0 dBs? 0 is given by: Et [it+ ] = 1 Z Et [ t+s ] ds 0 therefore unexpected in‡ation can be expressed as follows: Z 1 it; Et [it+ ] = ( t+s Et [ t+s ]) ds+ 0 Z 1 + q 0 + q xt+s 0 dBsx + = 1 0 s ds ) + (xt 0 ( + q 0 ) Z Z s0 e q0 dBsx + dBsx + (! 0 + 0 ! 01 ) 0 Z e s0 ! 1 dBs? + 0 0 0 q 0 0 ! 0 + ! 1 xt+s dBs? = 0 0 Z Z Z dBs? 0 + Z 0 s q0 dBsx + 0 Z 0 ? s ! 1 dBs : 0 It is easy to note that for the unexpected in‡ation to be time varying, that is, to be function of the factors xt , it is su¢ cient that either q or ! 1 are di¤erent from zero, meaning that the time-varying market prices of in‡ation risk are the key features of the model permitting time variation in in‡ation volatility, which we derive below. We can re-write the unexpected in‡ation in matrix form: it; 2 1 0 6 q0 + q 6 6 6 1 C=6 6 6! 0 + ! 01 6 4 1 1 Et [it+ ] = 2 03 603 7 6 7 6I3 7 6 7 7 ; D = 601 6 7 601 7 6 7 4I3 5 01 3 3 (C + Dxt )0 3 7 7 37 7 37 ; 7 37 5 3 37 3 2 R R0 s ds dB x 0 0 s s q0 3 7 6 7 6R x 7 6 e dB s 6 0R 0 q0 x 7 7 =6 s 7 6 0 R s dB ? 7 6 dB 0 0 s 7 6R 4 e s ! 1 dBs? 5 0R 0 ! dBs? 0 s 1 R R 0 where 0 s ds = 1 0 I3 3 e s dBsx : By observing the elements in , it is easy to note that the unexpected in‡ation is driven by all four shocks in the model: the innovations to the yield factors, the innovations to the hidden factor, and the shock speci…c to CPI, as well as the conditional volatility of these shocks. The in‡ation variance is a quadratic function of the state variables: var(it; ) = 1 2 Et (C + Dxt )0 8 0 (C + Dxt ) : 0 , the In the Appendix A, we provide a detailed derivation of all the elements in block matrix whose expected value delivers the variances of and covariances between the shocks that drive unexpected in‡ation (and thus uncertainty). As it will become clear later, having a survey-based measure of in‡ation uncertainty allows us to better pin down some of the parameters in C and : More importantly, the vector of parameters ! 1 can be identi…ed only if we incorporate survey data on this second moment. 2.4 In‡ation Risk Premium We now turn to the main object of interest in this study, that is, the IRP, which is de…ned as follows: IRPt = rtN rtR t q0 N 0 xt 0 N0 q 0 + N0 N0 q 0 0 = q 1 2 + q0 q 0 q0 q 1 2 q0 q 0 0 1 2 ! + 2 0 0 ! 0 ! 1 xt + 1 ! 1 ! 01 xt : 2 Our IRP has a richer dynamic behavior than permitted by previous studies in the literature, for example, Chernov and Mueller (2012) and D’Amico, Kim and Wei (2016), who already allowed for quite ‡exible dynamics. Particularly, in D’Amico, Kim and Wei (2016), the IRP is linear in the state variables and is time varying because of the state-dependent market price of nominal risk–i.e., the time variation is obtained by having just the term N 0 q0 di¤erent from zero in the expression above. In this model, the resulting speci…cation of the IRP has two additional sources of ‡exibility. First, as shown in the last term of the above equation, it is a quadratic function of the state variables because of q and ! 1 . Second, the linear portion can vary because either the market price of nominal risk or the market price of in‡ation risk changes over time, as N , q , and ! 1 multiply xt . This extremely adaptable functional form should allow our model to accommodate very di¤erent dynamic behaviors of the IRP over a long and diverse sample period including the in‡ation scare episodes of the 1980s when, in principle, perceptions of heightened in‡ation risk would have commanded large and positive values of the IRP, and the de‡ation scare episode of 2009-2012, when disin‡ation and low growth made nominal bonds a very good hedge against adverse outcomes possibly pushing the IRP into negative territory. To provide a simple intuition for why the data on real-time in‡ation variance can improve the estimation of the IRP, we rewrite the IRP in the following way: 9 IP Rt; = 1 1 = 1 2 log 41 + Cov Et R Mt+ MtR R Mt+ MtR ; QQt+t Qt Qt+ Et log 1 + Cov rt;R ; it; 3 5 + Jt; =Et rt;R Et (it; ) log 1 + Cov rt;N ; it; var(it; ) =Et rt;R Et (it; ) : where for simplicity we are assuming that the real pricing kernel is mainly driven by the real yield and we are ignoring the Jensen’s inequality term, which in practice is fairly small.2 Based on this simpli…cation, it is easy to see that to the extent that variations in the covariance between the real economy and in‡ation arises from ‡uctuations in the variance of in‡ation, accurate measurement of these speci…c ‡uctuations would be important. Survey data on real-time in‡ation uncertainty serve this purpose, that is, they help identifying ‡uctuations in the variance of in‡ation and thus in the time-varying IRP. Further, as we will explain shortly in Section 4.1 where we describe the covariances of the state variables, having a time-varying market price on in‡ation risk qt also allows time variation in the covariance between nominal interest rates and in‡ation, thus having more data to pin down qt also helps in the estimation of that covariance. 3 Comparison to previous studies This paper draws on contributions from several streams of the term-structural literature. First of all, to achieve time-varying second moments, we favor the use of Quadratic Gaussian (QG) models because a¢ ne term-structure models with stochastic volatility typically fail to produce reasonable risk premia (Dai and Singleton, 2002 and Du¤ee, 2002) and …tted yield volatilities that resemble the time-varying volatilities estimated from semi-parametric time-series models (Ahn, Dittmar, and Gallant, 2002; Collin-Dufresne, Goldstein, and Jones, 2009). For example, Haubrich, Pennacchi, and Ritchken (2012) develop a completely a¢ ne model that has four stochastic drivers and seven factors, but it still generates IRP that do not seem very sensible up to the two-year horizon, as it is mostly negative even in the early 1980s, when most other studies …nd that IRP estimates reach their highest peak. In contrast, as shown in Kim (2004), QG models do not seem to exhibit a tradeo¤ between …tting yield volatility and risk premia, therefore, we build on these type of models (e.g., Kim, 2004; and Kim and Singleton, 2012) and expand on them 2 Jt; ( 1 )[log(Et (Qt =Qt; )) Et (log(Qt =Qt; ))]: 10 by adding ‡exibility to market prices of in‡ation risk and allowing for unspanned in‡ation risk. Particularly, we decided to expand in this direction because Le and Singleton (2013) show that substantial variation in risk premia is unspanned by nominal bond yields and seems to arise from a time-varying market price of in‡ation risk; and, D’Amico and Orphanides (2014) show that perceived in‡ation risk is an important driver of excess bond returns beyond and above the information contained in nominal yields. To allow for unspanned in‡ation risk, our model includes some of the unspanning restrictions emphasized in Du¤ee (2011) and Joslin, Priebsch, and Singleton (2014), and similarly to the latter, we also run preliminary regressions to motivate our hidden factor. Table 1 reports the percentage of variation (R2 ) in in‡ation related variables explained by the 3 latent factors of an a¢ ne term-structure model estimated using only nominal yields and short-term rate forecasts. We …nd that although more than 80% of variation in expected in‡ation is explained by these factors, only half of the variation in in‡ation uncertainty is explained by those same factors. In line with this observation, our unspanning restrictions permit our third factor to drive both expected in‡ation and in‡ation uncertainty while remaining hidden from the nominal yield curve. Our paper is also closely related to studies emphasizing the size and nature of the IRP. For example, similarly to Chernov and Mueller (2012), we use surveybased in‡ation expectations at various horizons, but while their preferred model uses TIPS yields in the estimation, we use short- and long-horizon survey forecasts of nominal interest rates that together with surveys forecasts of in‡ation help to pin down the term-structure of expected real rates over a longer sample. A more important di¤erence is that in this paper, we focus on modeling time variation in the market price of in‡ation risk and incorporate information from survey-based in‡ation variance, which in turn permits us to identify a more ‡exible dynamic of the IRP. Another relevant study that, however, uses a quite di¤erent approach is Buraschi and Jiltsov (2005). Speci…cally, these authors develop a structural model that can identify the underlying nominal and real factors driving the IRP, but also su¤ers from the shortcoming that the market price of risk, even if state dependent, is not as ‡exible as ours, which is based on a more reduced-form approach. Further, their dataset consists only of interest rates, CPI, and money supply, and thus does not include any information from survey forecasts. In addition, di¤erently from our work, in both of these studies, the sample period stops before 2008. Finally, our study is also related to equilibrium term-structure models implying that time-variation in expected excess returns of nominal risk-free bonds is driven by changes in variances of real and in‡ation risks (e.g., Bansal and Shaliastovich, 2012); however, in most of these models, the market price of risk is assumed to be constant and macro risk is fully spanned by nominal yields. This is also true for Campbell, Sunderam, and Viceira (2016), who assume that all time variation in bond risk premia is driven by variation in bond risk and not by variation in the 11 aggregate price of risk. Importantly, their estimates of the variables governing bond risk are informed by realized second moments of high-frequency returns, while our estimates are informed by a real-time measure of perceived in‡ation risk. Moreover, we are more focused on modeling and estimating the IRP, while they emphasize the importance of the time-varying stock-bond covariance for the term structure of interest rates. 4 State-space form and data In this section, we …rst present the state equation and emphasize the role of timevarying market prices of in‡ation risk in generating time variation in the covariances of the state variables, then we turn to the observation equations and highlight how they link the data to our state variables. 4.1 State variables and their covariances We rewrite the model in a state-space form and estimate it by quasi maximum likelihood (QML) using the Augmented State Space Extended Kalman Filter method developed in Kim (2004). The basic idea of his approach is to augment the state vector st with the quadratic term vech(xt xt 0 ), st = [xt ; vech(xt xt 0 ); qt ]0 ; such that the state equations can be written in the usual linear matrix form: st = Gh + h st h + s t; where st = [ t ; vt ; q (xt h )0 t + !(xt h )0 ? t ] is the vector of innovations to xt , vech(xt xt 0 ), and qt , respectively, and vt , Gh and h are de…ned in the Appendix B. The conditional variance of the state variables, st h = V ar(st jIt h ) = Et h ( st st ); is given by: 3 2 0 0 q 0 0 0 t t t vt t t (xt h ) 5= vt 0t 0 vt vt0 vt 0t q (xt h )0 E4 q q q 0 0 0 0 q 0 2 ?2 (xt h ) t t (xt h ) t vt (xt h ) t t (xt h ) + !(xt h ) t 2 3 V art h (xt ) Covt h (xt ; vech(xt xt 0 )) Covt h (xt ; qt ) 4Covt h (xt ; vech(xt xt 0 )) V art h (vech(xt xt 0 )) Covt h (vech(xt xt 0 ); qt )5 0 Covt h (xt ; qt ) Covt h (vech(xt xt ); qt ) V art h (qt ) It is worth noting the di¤erent roles played by q (xt ) and !(xt ): q (xt ) allows covariances between all latent factors and the log price level qt to be time-varying and also contributes to the time variation in the variance of qt ; in contrast, !(xt ) governs only the variance of qt . This suggests that, in principle, the estimated values of !(xt ) should be strongly in‡uenced by data on in‡ation uncertainty, which will also help identifying ‡uctuations in the variable q (xt ). 12 4.2 Observation equations and data From January 1983 to December 2013, we observe seven nominal yields YtN = fyt;N i g7i=1 , the 6-month, 12-month, and 6-to-11 years ahead forecasts of the nominal short rate ft6m ; ft12m ; and ftlong respectively, the survey in‡ation expectations at one- and 11-year horizons EIt1y and EIt11y ; as well as the one-year real-time in‡ation uncertainty IUt1y : We collect all the observable variables in the vector ot = [YtN ; ft6m ; ft12m ; ftlong ; EIt1y ; EIt11y ; IUt1y ]0 and write also the observation equations in a matrix form: ot = a + F st + "t where "t denotes the vector of measurement errors, assumed to be i.i.d., with freely N N (0; 2EI; i ); and N (0; 2N; i ); "ft; i N (0; 2f; i ); "EI estimated variances: "Yt; i t; i "IU N (0; 2IU ): t More details about the functional form of the observation equations and thus of a and F are provided in Appendix C. However, we stress here how each observation equation links speci…c data to all or some of the state variables. Further, it should be noted that in‡ation and survey-based variables are not available for all dates, which introduces missing data in the observation equation and are handled in the standard way by allowing the dimensions of a and F to be time-dependent (see, for example, Harvey 1989). The …rst seven measurement equations relate observable Treasury nominal yields only to the two state variables xt ; due to the unspanning restrictions. Speci…cally, we use the 3- and 6-month Treasury bill rates from the Federal Reserve Board’s H.15 release and converted them to continuously compounded basis. The 1-, 2-, 4-, 7-, and 10-year nominal yields are based on zero-coupon yield curves …tted at the Federal Reserve Board (see Gurkaynak, Sack, and Wright, 2007; Gurkaynak, Sack, and Wright, 2010 for details). We sample yields at the weekly frequency and assume that the monthly CPI-U data is observed on the last week of the current month.34 Similarly, our eighth and ninth measurement equations also link the 6- and 12month-ahead forecasts of the 3-month Treasury bill rate from Blue Chip Financial Forecasts (BCFF), which are available monthly, only to xt : We complement these measurement equations with another one that uses the long-range forecast (6-to-11 years ahead) of the same rate. In BCFF, this forecast is provided only semiannually, but we follow the procedure in D’Amico and King (2015) to convert them to a consistent quarterly frequency, as we think that information from longer-term survey forecasts is very important to correctly estimate the persistency of the yield factors under the physical measure. The basic idea consists of combining the long-range 3 Here we abstract from the real-time data issue by assuming that investors correctly infer the current in‡ation rate in a timely fashion. 4 The data source for the nominal yields and CPI-U is Haver. 13 forecasts from BCFF with those from Blue Chip Economic Indicators (BCEI). This is because BCFF provide these long-range projections in June and December, while the BCEI report them in March and October, these values can then be interpolated to obtain the September value and have a regularly-spaced quarterly time series.5 The eleventh and twelfth equations relate the observed measures of expected in‡ation at the 1- and 11-year horizon to all state variables xt , as in‡ation-related variable are allowed to load on the hidden factor. Speci…cally, we use the median forecast of average in‡ation over the following year from the Survey of Professional Forecasters (SPF) because it is reported at a consistent quarterly frequency and therefore does not require interpolation. However, since the longest available forecasting horizon in these data is one-year ahead, to measure longer-term in‡ation expectations we turn again to the BCS, which has been providing semiannual longrange (2-to-6 and 7-to-11 years ahead) consensus forecasts of CPI since 1983. Once we have converted them to a consistent quarterly frequency using the same methodology described for interest rate forecasts, we can compute the expected average value over the next 11 years— by taking the weighted average of the one-year, 2-6year, and 7-11-year expectations, respectively. Finally, the last observation equation relates the real-time measure of in‡ation variance at one-year horizon to all state variables xt as well as to vech(xt xt 0 ). The real-time measure of in‡ation variance is derived from the subjective probability distributions in the SPF using the methodology of D’Amico and Orphanides (2008), therefore it should capture ex-ante in‡ation risk perceived by investors rather than ex-post realized volatility. 5 Identi…cation and estimation methodology Except for the unspanning restrictions already described in Section 2.1, for all other parameters in the model, we only impose restrictions that are necessary for achieving identi…cation to allow a maximally ‡exible correlation structure between the factors, which has shown to be critical in …tting the rich behavior of risk premia observed in the data. In particular: 2 3 2 3 0 0 1 0 0 11 0 5 ; = 4 21 1 05 = 03 1 ; = 4 0 22 0 0 33 31 32 1 and N is unrestricted. Regarding the set of parameters that allow for time variation in the variance of in‡ation and covariances of in‡ation with the other state variables, we have that q is lower triangular and ! 1 is left unrestricted: 5 For more details see the Appendix in D’Amico and King (2015). 14 q 2 =4 q 11 q 21 q 31 0 q 22 q 32 3 0 0 5 and ! 1= ! 11 ! 12 ! 13 q 33 This implies that the market price of in‡ation risk can be a¤ected by all three factors xt and their interactions, and that the conditional volatility of the shock speci…c to CPI is also a¤ected by the same three factors xt . To facilitate the estimation by starting with reasonable initial values of the parameters and to make the results easily replicable, we break the estimation in a few easier steps: We …rst perform a “pre”-estimation where a set of preliminary parameter estimates governing the nominal term structure is obtained using YtN and survey forecasts of 3-month TBill rate alone;6 second, based on these estimates and data on YtN , we can obtain a preliminary estimate of the state variables, xt and dBt ; third, a regression of monthly in‡ation onto estimates of xt and dBt gives preliminary estimates of 0 ; 1 , q0 ; q , ! 0 ; fourth, a regression of quarterly in‡ation uncertainty on xt and x2t gives preliminary estimates of ! 1 ; and …nally, these preliminary estimates are used as starting values in the full, one-step estimation of all model parameters by QML. 6 Empirical Findings In this section, we …rst provide a summary description of the results based on our "full" model speci…cation, which includes all the features described above and incorporates in the estimation all the information from surveys. Then, we dissect the results to highlight the contribution of key elements of our approach separately, by presenting comparisons with simpler speci…cations and with the estimation that does not make use of survey information on the second moment of in‡ation. 6.1 Full model speci…cation A visual description of our main …ndings is presented in Figures 1, 2, and 3. Specifically, Figure 1 shows the decomposition of the 10-year nominal yield into three components: The real yield (including the RRP), the expected in‡ation at the pertinent horizon, and the corresponding IRP. Figure 2 focuses on the four components of the 10-year nominal yield, as in addition to the expected in‡ation rate and IRP (also shown in Figure 1), it shows the expected future short real rate and the RRP separately. Finally, Figure 3 summarizes the overall …t of the full model, as it compares the model-implied one-year in‡ation variance, 5-year real yield, one-year 6 It is important to keep in mind that in this preliminary estimation we do not impose unspanning and therefore derive 3 latent factors from the nominal term structure. This implies that especially the third factor will have a dynamic quite di¤erent from that one of the hidden factor obtained in the …nal step of the estimation. 15 expected in‡ation, and one-year expectation of the nominal short-term rate to their counterparts in the data (shown in orange). As it can be seen in Figure 1, the model estimation over the 1983 to 2013 period captures the main characteristics of the time variations in longer-term nominal yields that have been discussed in the earlier literature. Overall, in‡ation expectations, real interest rate expectations, the IRP as well as the RRP all trended down during the 1980s and 1990s. Real yields dominate the other components in accounting for the ‡uctuations in nominal yields. However, the major sources of variation di¤er at low and high frequencies. While the expectation component of the yield— the expected real interest rate and expected in‡ation— dominate at business cycle frequencies, the risk premia largely drive higher-frequency ‡uctuations. Focusing on the estimates of the 10-year IRP in Figure 2, our …ndings suggest that it was consistently positive in the …rst part of the sample, reaching its highest peak (about 1:7 percentage points) in the spring of 1984, and then spiked again in May-October 1987. Since the mid 1990s, it has ‡uctuated around zero, reaching its most negative values (about 0:6 percentage points) in 2005-2007, just before the most recent …nancial crisis, and during the subsequent weak recovery, in 2011-2012. The largest ‡uctuations in the estimated IRP capture notable episodes documented over this period that re‡ected changes in perceptions of in‡ation risks. The spikes in 1984 and 1987, for example, coincide with the narrative of the in‡ation scares of the 1980s documented by Goodfriend (1993). Similarly, the substantial decline over the 2010-2012 period largely coincides with the de‡ation scare episode described in Kitsul and Wright (2013). Our estimates of the IRP also capture episodes that have occupied discussions relating to monetary policy. One notable example is the "conundrum" period in the mid-2000s when, as shown in Figure 2, risk premia started declining sharply in 2004 while the Federal Reserve was raising short-term nominal interest rates. Another example is the "taper-tantrum" in the summer of 2013, when longer-term Treasury yields rose dramatically following Fed Chairman Ben Bernanke’s remarks about the possibility of moderating the pace of asset purchases later that year, implying a lower degree of expected monetary policy accommodation. Interestingly, our …ndings also illustrate the time-varying nature of the covariance of yield components. While in much of the 1980s, all four components broadly move in the same direction, after 1987 expectations and risk premia start moving in opposite directions. This pattern is particularly evident in 1987-1992, 2001-02, 2004-08, and 2011-2013. These are periods highlighting the presence of a hidden factor: Changes in the hidden factor would move the IRP and RRP in the same amount of but opposite to the expected future short real rates and the expected in‡ation. This could explain the conundrum period and also indicates that the entire increase in the nominal yields observed during the taper tantrum was indeed due to increases in risk premia. Turning attention to the expected in‡ation and expected short-term real interest 16 rates, Figure 2 also shows that the model captures their secular decline since the 1980s. With respect to in‡ation expectations, this decline is consistent with the Federal Reserve’s successful disin‡ation e¤orts over the 1980s and 1990s and its strategy of maintaining mostly stable in‡ation since then. With respect to the decline in long-term expectations of the short-term real interest rate, the model’s …ndings are consistent with studies suggesting a notable decline in the equilibrium real interest rate over this period (Holtson, Laubach and Williams, 2016). Moving to the overall …t of the full model, Figure 3 suggests that the modelimplied variables match their data counterparts quite well. Starting from the top left panel, it can be noted that the ‡uctuations in the model-implied one-year in‡ation variance track quite closely those in the survey-based in‡ation variance. Further, despite being estimated without the use of TIPS yields, as shown in the top right panel, the full model generates a 5-year real yield that closely resembles that one on TIPS (when available), except for a residual very similar to the liquidity premium estimated in D’Amico, Kim, and Wei (2016). In their study, the estimated TIPS liquidity premium is fairly high and positive in the early years of TIPS, then declines steadily and stays close to zero from 2004 until the height of the 2007-08 …nancial crisis, when it surges to its highest level, to then turn negative around 2011. The two bottom panels indicate that the model can match pretty well one-year survey forecasts of in‡ation and of the short-term rate. This also illustrates that the survey information about …rst moments of key variables like in‡ation and the short rate help the model capturing the slow moving trend in those expectations as well as the ZLB period. 6.2 Dissecting the model’s key features The main empirical contributions of our study can be more easily illustrated and understood by comparing the empirical performance of the full model to the results derived from di¤erent model speci…cations, with each speci…cation obtained by removing from the full model one of its key ingredients. We consider three simpli…cations: 1) the model without time-varying in‡ation volatility, called Model No_TVV (No time-varying volatility, i.e., ! 1 = 03 1 and q = 03 3 ); 2) the model estimated without data on in‡ation uncertainty (called No_IU) and thus without ! 1 ; which cannot be correctly identi…ed without those data; and 3) the model estimated letting also zt to be spanned by nominal yields, called No_Unsp (No unspanning, i.e., N N a 3 3 unrestricted matrix). Table 2 summarizes those 1 (3) unrestricted and model speci…cations and associated parameters restrictions. The …rst exercise quanti…es the contribution of time-varying market prices of in‡ation risk to the overall model performance. The second exercise aims at understanding the value added by survey information about perceived in‡ation uncertainty, and as a consequence the role played by the time-varying conditional volatility of the orthogonal shock speci…c to CPI, which should capture high-frequency 17 variations in in‡ation. Finally, the third experiment is meant to shed light on the importance of the hidden factor for capturing variations in the …rst and second moments of in‡ation. Figure 4 summarizes the comparison between the model with homoskedastic in‡ation shocks (Model No_TVV), whose results are plotted in the left panels, and the full model, whose results are plotted in the right panels. For each model, we show in blue the estimated values of the one-year in‡ation variance, the two-year IRP, the 5-year real yield, and the one-year expected in‡ation, and in orange their data counterparts. For brevity, we do not report the estimates for longer-term variables as they provide the same message and, for the full model, have been highlighted in the Figure 1 and 2. The panel’s …rst row shows the implications of restricting the model to homoskedastic in‡ation shocks. While the full model is able to match quite closely the ‡uctuations in the survey-based in‡ation variance, the Model No_TVV estimates the in‡ation variance to be constant at 0:8 percent which is too low to capture the 1980s and too high to capture the more recent period of relative stability. As shown in the second and third rows, this has important implications for the estimated IRP and real yields. The homoskedastic model generates IRP estimates that are implausible: They are extremely large (in absolute value), as they vary between 10 and +11 percent, and are trending upward over the sample period, with the lowest values in the early 1980s and the highest peak in 2013. In contrast, the full model estimates the 2-year IRP to reach its highest value of about 50 basis points in the early 1980s, then to decline quite consistently through the mid 1990s when it turns negative, particularly in 2001-02, 2004-06, and 2010-12, but also to increase sharply at the height of the recent …nancial crisis in 2008-09 and in the summer of 2013 during the so-called taper tantrum. As shown in the third row, the 5-year real yield implied by Model No_TVV also ‡uctuates within an unreasonable range, as it reaches almost 20 percent in the early 1980s and about 15 percent in 2012; while, on the other hand, the full model generates a 5-year real yield that reaches at most about 7 percent in the early 1980s and closely resembles that one on TIPS (when available), as already noted in the discussion of Figure 3. However, the homoskedastic model …ts the one-year survey expected in‡ation slightly better, indicating that, if survey forecasts of in‡ation are used in the estimation, having time-varying in‡ation volatility does not add much along this dimension. This may be due to the hidden factor, which is responsible solely for variations in in‡ation-related variables. Since the hidden factor has to capture only ‡uctuations in in‡ation expectations but not in the in‡ation variance, it is possible that it is doing a much better job in …tting the survey-based …rst moments. The next four …gures compare results from the full model to the other two simpli…cations we consider, that is, Model No_IU and Model No_Unsp. Figure 5 shows the estimates of the IRP at 2- and 10-year maturity and of the one-year in‡ation variance together with the SPF counterpart across the three model speci…cations. 18 Looking at the third column, it is evident that only the full model is successful in capturing the ‡uctuations in the survey-based in‡ation variance. Of course, this is not that surprising relative to the model estimated without data on in‡ation uncertainty, but is interesting to note the deterioration in the …t when the data on in‡ation uncertainty is used in the estimation of the model without unspanning. Indeed, as illustrated by the contrast between the top and bottom right panels, it seems that allowing for a factor that does not in‡uence nominal interest rates but does in‡uence in‡ation-related variables is important to capture ‡uctuations in perceived in‡ation risk. The regression analysis reported in Table 3 con…rms this observation. The table shows the R2 from regressions of the in‡ation-related concepts onto the three factors implied by the full model. As shown in the last column of the table, the hidden factor explains a large portion of variations in the surveybased in‡ation variance that is not explained by the other two factors: Including the hidden factor in the regression raises the R2 from 47% to 83%. In turn, since Model No_IU and Model No_Unsp do not …t in‡ation variance well, they do not generate very sensible IRP especially in the 1980s. For the Model No_IU, the estimated 2-year IRP is implausibly small and even negative in the early 1980s. This is not consistent with most estimates available in the literature, which tend to be sizable and positive across maturities during those years. In contrast, Model No_Unsp estimates values of the IRP that are as high as 7:3 percent in the early 1980s and are always positive, which is implausibly high. Indeed, most studies obtain estimates of the IRP that hardly reach 2 percent, even at longer maturities, and often turn negative starting in the 2000s (e.g., Buraschi and Jiltsov, 2005; Chernov and Mueller, 2012; Haubrich et al. 2012, Ajello et al., 2012). Based on those previous …ndings, it seems that the estimated IRP from the full model, reported in the left and middle top panels, is much more sensible. In addition to the dynamic behavior of the IRP already described in Figure 2, it is worth noting that the average term structure of the IRP is upward sloping, as it is usually more di¢ cult to predict in‡ation at longer horizons and thus uncertainty about in‡ation is larger. Further, the greater duration of longer-term bonds ampli…es the impact of a given amount of in‡ation uncertainty. Figure 6 illustrates the implications of the IRP estimates for the model-implied real yields and RRP. The bottom row shows quite starkly that, in the case of the Model No_Unsp, the ‡ip side of extremely large and positive IRP is extremely low and ‡at real yields and RRP, which at the 10-year horizon reaches 1 percent in 1984. To a much lesser extent there is a similar trade-o¤ also in the case of the Model No_IU, but only in the early 1980s, which at the 10-year maturity is less evident than at the 2-year maturity (not shown for brevity). In particular, since in the absence of survey data on in‡ation uncertainty, this model produces IRP that are too low or even negative in the early 1980s, it generates real yields and RRP that seem a bit too high in the same period, with the 2-year real yield as high as the 10-year real yield, and the 2-year RRP reaching a peak of about 2 percent in 1984 to 19 counterbalance the negative values of the IRP in the same period. Finally, the full model, similarly to the results for the 5-year real yield already described in Figure 3, generates a 10-year real yield that closely resembles that one on TIPS, again except for a residual very similar to the 10-year TIPS liquidity premium estimated in D’Amico, Kim, and Wei (2016). This model also delivers a 10-year RRP that is mostly positive over the sample period, displaying a marked downward trend as it declines from a level of about 3:5 percent in the early 1980s to almost 0:5 percent at the end of 2013. Figure 7 makes a very simple point: the …t of survey in‡ation expectations across the three models is very similar. This suggests that the data on in‡ation variance and the hidden factor have almost no e¤ect on the model-implied estimates of in‡ation expectations at short and long horizons, when their survey counterparts are included in the estimation. It also implies that these estimates are mainly governed by the two latent factors that extract information mostly from nominal yields and the survey forecasts of the short-term rate. Table 3 con…rms this observation: R2 from regressions of the in‡ation-related concepts onto the …rst two factors (the yield factors) are as high as 84 percent, and the R2 does not increase much once we include the hidden factor in the regression speci…cation. Using long-range survey forecasts of the short-term interest rate in the estimation produces a level yield factor that is quite persistent and is therefore able to capture the gradual downward trend in in‡ation expectations. Finally, …gure 8 clearly illustrates that also the …t of survey forecasts of the nominal short-term rate, at short and long horizons, is very similar across the three models. This, together with the evidence presented in Figure 7, in turn, suggests that expected real rates are well pinned down simply by the di¤erence between survey forecasts of nominal interest rates and in‡ation. Thus, if a term-structure model allows for a ‡exible speci…cation of the IRP, whose richer dynamics are better identi…ed using survey information on in‡ation variance, as it is the case in the full model, then the di¤erence between the observed nominal yields and the sum of expected real rates, expected in‡ation and IRP (all of which are extracting information from survey data), will be su¢ cient to inform the estimates of the RRP. This is the basic intuition to understand the ability of the full model to generate more sensible IRP and RRP over this long sample period. Finally, it is also worth observing that, since all the models …t survey forecasts of the short-term rate very closely, even during the ZLB period, and since these forecasts do not violate the ZLB, then also the model-implied estimates of nominal short rates obey the ZLB at these maturities. In other words, information from surveys is extremely helpful for the estimation of our model also at the ZLB. 20 7 Concluding remarks We show that a Quadratic-Gaussian model of the term structure resulting from a ‡exible speci…cation of the market prices of in‡ation risk and estimated using surveybased in‡ation uncertainty can capture the rich dynamics of in‡ation and real risk premia over the 1983-2013 period. It can also provide guidance on expected real interest rates and expected in‡ation embedded in longer-term yields. In addition to a very ‡exible market price of in‡ation risk, two other features of the model appear particularly useful to capture correctly the dynamics of the in‡ation risk premia over time in our long sample. First, the introduction of timevarying volatility of the shock speci…c to CPI, which mainly captures short-run in‡ation ‡uctuations. Second, the presence of a hidden factor, which is supposed to govern the component of the in‡ation-related variables not spanned by nominal yields. Both of these elements improve the reliability of the estimated in‡ation risk and associated premium, and thus of the decomposition of nominal yields. Interestingly, our results suggest that the hidden factor is important mainly for the in‡ation variance. In contrast, in‡ation expectations load mostly on the level-yield factor, the most persistent state variable implied by our model. With regard to the key novelty in the estimation, the use of real-time data on in‡ation uncertainty proves crucial for pinning down the dynamics of the in‡ation risk premium over our sample that includes both the 1980s, when perceived in‡ation uncertainty was high, and the 2000s and 2010s, when perceived in‡ation uncertainty was low. Use of this information would be much less important if attention were restricted to the more recent period of greater in‡ation stability. The estimated model captures both the decline in in‡ation expectations from the 1980s to the 2000s that is associated with the Federal Reserve’s disin‡ationary e¤orts and the notable decline in the equilibrium real interest rate, as measured by the long-term expectations of the short-term real interest rate. A decomposition of the 10-year nominal yield suggests that the expectation components— the expected real interest rate and expected in‡ation— dominate at business cycle frequencies, while the risk premia largely drive higher-frequency ‡uctuations. Focusing on the IRP, the results con…rm that it was considerably higher in the 1980s than over later periods. The estimates also identify episodes of notably negative IRP, such as the 20052007 period, just before the most recent …nancial crisis, and during the subsequent weak recovery, in 2011-2012. Overall, incorporating available survey information regarding …rst and second moments, allows for the estimation of a ‡exible term structure model that can capture the rich dynamics of risk premia and expectations. 21 8 Appendix A: Components of the in‡ation variance For the interested reader we provide details on the in‡ation variance computation: V art (it; ) = Et [ 1 2 where 0 (C + Dxt )0 2 Vaa 6 Vab0 6 0 6 Vac 6 V =6 6 01 n 6 01 n 6 40n n 01 n Vab Vbb Vbc0 01 n 01 n 0n n 01 n (C + Dxt )] = Vac Vbc Vcc 01 n 01 n 0n n 01 n 1 2 (C + Dxt )0 V (C + Dxt ) 0n 1 0n 1 0n 1 0n 1 0n 1 0n 1 Vdd 0 0 Vee 0n n Vef0 0 0 3 0n 1 0n 1 7 7 0n 1 7 7 0 7 7 0 7 7 0n 1 5 Vgg 0n n 0n n 0n n 0n n Vef Vf f 01 n This can be written as: V art (it; ) = aiu + biu xt + xt 0 Ciu xt 1 2 [ 1 0 q 0 (Vaa 1 + Vab ( q q 0 + q 0 + Vbb ( + ) Vbc ) 0 0 + (! 0 + ! 1 ) (Vee (! 0 + ! 01 ) 2 biu = 2 [ 1 0 Vac + ( q0 + q )0 Vbc ciu = 1 2 ) Vac ) + ( (Vac0 1 Vef 0 + ) Vbc0 ( q0 q 0 + q + q (Vef0 (! 0 )0 (Vab0 22 1 ) Vcc ) + Vdd + ! 01 ) Vf f ) + Vgg ] Vcc + (! 0 + ! 01 )0 Vcf [Vcc + Vf f ] (2) (3) where: aiu = (1) 0 Vf f ] with elements given by: Z 1 (Inxn e ( Vaa = Z0 0 0 1 e = s) ( 0 ) (Inxn s)0 e e ( 10 +e ( ) s) ( s) 0 0 (4) ds s) 0 ( e s)0 ds 10 0 = 1 Vab = 1 0 [ Z 0 (Inxn (Inxn ( e e s) 0 10 ) 1 (Inxn e 0 ) + F0; ( ; 0 ; 0 )] ) ds 0 1 1 [ Inxn (Inxn e )] Z 1 (Inxn e ( s) ) q e s ds Z0 1 q e s e ( s) q e s ds = Vac = = (5) 0 = Vbb = 1 Z q [ (Inxn e ) 1 e F0; ( ; ; q )] (6) Inxn ds = Inxn 0 Vbc = Vcc = Z Z0 0 q e Z s e s0 ds = q q0 s q e 1 (Inxn e ds = F0; ( 0 ; ; (7) ) q0 ; q (8) ) 0q0 q q0 q Vdd = Et [ ) s ds] = G0; ( s 0 Z Vee = ds = 0 Z ! 01 s ds = ! 01 1 (In n e ) Vef = Z0 e s0 ! 1 ! 01 s ds = F0; ( 0 ; ; ! 1 ! 01 ) Vf f = 0 Z 0 0 0 Vgg = Et [ s ! 1 ! 1 s ds] = G0; (! 1 ! 1 ) (9) (10) (11) (12) (13) 0 9 Appendix B: The discrete state equation To estimate the model, we need to discretize it and derive its state-space form. Let h be a very small time interval, it follows that: xt = h + (Inxn h)xt h + 23 t = K + Hxt h + t (14) 10 where is: t N (0; hInxn ). This means the discretized expression for the log price level qt = q t h + 0h + 1 0 xt hh q + (xt 0 h) t + !(xt ? h) t (15) In order to capture the quadratic dynamics of real bond prices, we will have to augment our state vector with the term vech(xt x0t ). We introduce the operator Dn such that vec() = Dn vech(). We also introduce Dn+ = (Dn0 Dn ) 1 Dn0 so that Dn+ vec(xt xt 0 ) = vech(xt xt 0 ). Thus, applying properties of vec(), vech(xt xt 0 ) = Dn+ vec(KK 0 + h 0 ) + Dn+ (H K + K H)xt + Dn+ (H H)vec(xt h xt 0 h ) + Dn+ vec( t 0t 0 h 0 + (K + Hxt h ) 0t 0 + t (K 0 + xt 0 h H 0 )) = vech(KK 0 + h 0 ) + Dn+ (H K + K H)xt h + Dn+ (H H)Dn vech(xt h xt 0 h ) + t h where the errors terms are collected in t = Dn+ ( )vec( 0 t t) 0 vech(h ) + M (xt t (16) Inxn ) (17) h) and M (xt h) = Dn+ (Inxn (K + Hxt h) + (K + Hxt h) This permits us to de…ne a linear state space equation: st = Gh + h st h in which: 2 + 3 3 2 K xt st = 4vech(xt xt 0 )5 ; Gh = 4vech(KK 0 + h 0 )5 ; qt 0h 2 3 H 0 0 4 + K + K H) Dn+ (H H)Dn 05 h = Dn (H 0 0 1 1 h s t s t (18) 2 =4 t t q (xt h) 0 t + !(xt h) ? t 3 5 To perform Kalman …ltering, we will need the conditional moments of the state variables. We can easily see that E[st jFt The conditional variance s t h h] = Gh + = V ar(st jFt 24 h) (19) h st h = V ar( st ) = E[ s s0 t t ]. 10 Appendix C: Observation equations Observed variables ot are linear in the underlying state vector: (20) ot = a + F st + "t Using the expression for the bond prices derived earlier (and solving the di¤erential equations), the continuously compounded yields can be expressed as: YtN = aN + bN 0 xt + "N t ; N 0 where aN = [aN 3m :::a10y ] is the stacked vector of coe¢ cients for each maturity, and aN = 1 AN , where AN are the pricing parameters. The short rate forecasts are ft = Et [rt+ ;3m ] for = 6m; 12m. The 6-to-11 R 11y years ahead forecast is ftlong = 15 6y Et [rt+ ;3m ]d . We can solve for these using our expressions for the yields to …nd: 0 ft = an3m + bN 3m (Inxn N0 ftlong = aN 3m + b3m (Inxn xt + "ft 0 ) + bN 3m e e 1 5 1 (e 6 e 11 (21) 0 )) + bN 3m 1 5 1 (e 6 e 11 )xt + "ft l (22) for = 6m; 12m. The in‡ation expectations can be expressed as: EIt = for 0 + 1 0 1 1 0 1 (Inxn e ) + 1 1 0 1 (Inxn e )xt + "EI t (23) = 1; 11. Lastly, the one-year in‡ation uncertainty can be expressed: IUt1y = C 0 V C + (2C 0 V D)xt + vec(D0 V D)0 Dn vech(xt xt 0 )) + "IU t Thus, collecting all the coe¢ cients of the constant terms in a and the coe¢ cients multiplying the states in F , we have: 2 3 2 3 0 0 0 1 6 aN 7 6 bN 0 0 07 6 f;6m 7 6 f 6m 7 6a 7 6b 0 07 6 f;12m 7 6 f 12m 7 6a 7 6b 7 0 0 7 6 7 a=6 (24) 6 af;l 7 F = 6 bf l 0 07 6 i;1yr 7 6 i;1yr 7 6a 7 6b 0 07 6 i;10yr 7 6 i;10yr 7 4a 5 4b 0 05 au;1yr bu;1yr vec(D0 V D)0 Dn 0 The error vector "t will have a diagonal covariance matrix. 25 References Abrahams, M., T. Adrian, R. K. Crump, and E. Moench, 2013, “Decomposing real and nominal yield curves.” Federal Reserve Bank of New York Sta¤ Reports No. 570, October. Ahn, D., Dittmar, R., and Gallant, A. (2002), "Quadratic term structure models: theory and evidence." Review of Financial Studies Vol. 15, No. 1, 243-288. Ahn, D., Dittmar, R., Gallant, A., Gao, B. (2003), "Purebred or hybrid?: Reproducing the volatility in term structure dynamics." Journal of Econometrics 116, 147–180 Ajello, Andrea, Luca Benzoni, and Olena Chyruk (2012), "Core and ’Crust’: Consumer Prices and the Term Structure of Interest Rates." Working Paper Series WP-2014-11. Federal Reserve Bank of Chicago. Ang, A., G. Bekaert and M. Wei (2007), “Do macro variables, asset markets or surveys forecast in‡ation better?”Journal of Monetary Economics, 54, 1163-212. — — (2008), “The term structure of real rates and in‡ation expectations.”Journal of Finance, 63(2), 797–849. Anh, Le and Kenneth J. Singleton (2013), “The Structure of Risks in Equilibrium A¢ ne Models of Bond Yields.”Working paper, UNC. Argov, Eyal, David Rose, Philippe Karam, Natan Epstein, and Douglas Laxton (2007). "Endogenous Monetary Policy Credibility in a Small Macro Model of Israel." IMF working paper 207. Bansal, R. and Ivan Shaliastovich (2012), “A Long-Run Risks Explanation of Predictability Puzzles in Bond and Currency Markets.”Review of Financial Studies. Bekaert, Geert and Xiaozheng Wang (2010), “In‡ation risk and the in‡ation risk premium.”Economic Policy October 2010 pp. 755-806. Buraschi, Andrea and Alexei Jiltsov (2005), “In‡ation Risk Premia and the Expectations Hypothesis." Journal of Financial Economics 75, 429-490. Campbell, J. Y. and R. Shiller (1996), “A scorecard for indexed government debt,” in B. S. Bernanke and J. Rotemberg (eds.), National Bureau of Economic Research Macroeconomics Annual 1996, MIT Press, Cambridge, MA, pp. 155–97. Campbell J. Y., Sunderam A. and Viceira L., “In‡ation bets or de‡ation hedges? The changing risks of nominal bonds.” Harvard Business School Working Paper, 09-088, 2016. Chernov, Mikhail, and Philippe Mueller (2012), "The term structure of in‡ation expectations." Journal of Financial Economics vol. 106, 367-394. 26 Christensen, Jens H.E. , Jose A. Lopez, and Glenn D. Rudebusch (2012), “Extracting De‡ation Probability Forecasts from Treasury Yields." International Journal of Central Banking, December, 21-60. Cochrane, J. H. and Monika Piazzesi (2005), “Bond Risk Premia." American Economic Review, Vol. 94, No. 1, 138-160. Collin-Dufresne, Pierre , Robert S. Goldstein, Christopher S. Jones, "Can interest rate volatility be extracted from the cross section of bond yields?" Journal of Financial Economics 94 (2009) 47-66. Croushore, Dean (1993), “Introducing: The Survey of Professional Forecasters.” Federal Reserve Bank of Philadelphia Business Review, November/December, 3–13. Dai, Q., Singleton, K. (2002), "Expectations puzzle, time-varying risk premia, and a¢ ne models of the term structure." Journal of Financial Economics 63, 415–441 D’Amico, S., D. Kim and M. Wei (2016). "Tips from TIPS: the information content of Treasury in‡ation-protected security prices", Journal of Financial and Quantitative Analysis, forthcoming. D’Amico S., and Thomas King, (2015), "What does anticipated monetary policy do?" Federal Reserve Bank of Chicago Working Paper 2015-10. D’Amico S., and Athanasios Orphanides (2008), “Uncertainty and Disagreement in Economic Forecasting." Finance and Economics Discussion Series 2008-56. Federal Reserve Board. D’Amico S., and Athanasios Orphanides (2014), "In‡ation Uncertainty and Disagreement in Bond Risk Premia." Federal Reserve Bank of Chicago Working Paper No. 2014-24. David, Alexander, and Pietro Veronesi (2013), “What Ties Return Volatilities to Price Valuations and Fundamentals?" Journal of Political Economy, vol. 21, No. 4 (August), 682-746. Du, Wenxin, Carolin E. P‡ueger, and Jesse Schreger (2016). "Sovereign Debt Portfolios, Bond Risks, and the Credibility of Monetary Policy." NBER Working Paper Series, No. 22592, September. Du¤ee, G. (2002), "Term premia and interest rate forecasts in a¢ ne models." Journal of Finance 57, 405–443. Du¤ee, G. (2011), “Information in (and not in) the term structure.” Review of Financial Studies 24, 2895-2934. Evans, Charles, Jonas Fisher, Francois Gourio and Spencer Krane (2015), "Risk management for monetary policy near the zero lower bound." Brookings Papers on Economic Activity, March 19, 2015 27 Feldman, Ron, Kenneth Heinecke, Narayana Kocherlakota, Samuel Schulhofer-Wohl, and Thomas Tallarini (2016), "Market-Based Expectations as a Tool for Policymakers". Working Paper. Goodfriend, Marvin (1993), "Interest Rate Policy and the In‡ation Scare Problem: 1979–1992." Reserve Bank of Richmond Economic Quarterly Volume 79/1 Winter 1993. Grishchenko, O. and J. Huang (2013), "The in‡ation risk premium: evidence from the TIPS market." The Journal of Fixed Income, Spring. Gurkaynak, Refet S., Brian Sack, and Jonathan H. Wright (2006), “The U.S. Treasury Yield Curve: 1961 to the Present." Journal of Monetary Economics, vol. 54(8), 2291-2304. Gurkaynak, Refet S., and Jonathan H. Wright (2012), “Macroeconomics and the Term Structure." Journal of Economic Literature, 50:2, 331-367. Holston, Kathryn, Thomas Laubach and John C. Williams (2016), “Measuring the Natural Rate of Interest: International Trends and Determinants." Finance and Economics Discussion Series, 2016-073. Washington: Board of Governors of the Federal Reserve System. Haubrich Joseph, George Pennacchi, Peter Ritchken (2012). "In‡ation Expectations, Real Rates, and Risk Premia: Evidence from In‡ation Swaps." Review of Finacial Studies v.25, n 5, 1588-1629. Joslin, Scott, Marcel Priebsch, and Kenneth J. Singleton (2014), "Risk Premiums in Dynamic Term Structure Models with Unspanned Macro Risks." The Journal of Finance, vol. LXIX(3), 1197-1233, June. Kim, Don H. (2004), "Time-varying risk and return in the quadratic-gaussian model of the term-structure." This paper is part of the author’s Stanford dissertation. Kim, Don H. (2008), "Challenges in macro-…nance modeling." Federal Reserve Board Working Paper in the Finance and Economics Discussion Series 2008-06. Kim, Don H. and Athanasios Orphanides (2012), “Term Structure Estimation with Survey Data on Interest Rate Forecasts.” Journal of Financial and Quantitative Analysis, Volume 47, Issue 01, February 2012, 241-272. Kim, Don H., and Kenneth J. Singleton (2012), "Term Structure Models and the Zero Bound: An Empirical Investigation of Japanese Yields." Journal of Econometrics, vol. 170, pp. 32-49. Kim, H. D., Wright, J. H. (2005). “An Arbitrage-Free Three-Factor Term Structure Model and the Recent Behavior of Long-Term Yields and Distant-Horizon Forward Rates." Finance and Economics Discussion Series 2005-33. Federal Reserve Board. 28 Kitsul, Y. and Jonathan H. Wright, “The Economics of Options-Implied Infation Probability Density Functions." Journal of Financial Economics, forthcoming. Longsta¤, Francis A., Matthias Fleckenstein, and Hanno Lustig. (2014). "De‡ation Risk". UCLA Working paper. Palomino, Francisco (2012). "Bond Risk Premiums and Optimal Monetary Policy," Review of Economic Dynamics, vol. 15, no. 1, pp. 19-40. Piazzesi, Monika, and Martin Schneider (2006), “Equilibrium Yield Curves.”NBER Working Paper 12609. Piazzesi, Monika, Juliana Salomao, and Martin Schneider (2013), “Trend and Cycle in Bond Premia.”Working Paper, December. Stark, Tom (2010), "Realistic Evaluation of Real-Time Forecasts in the Survey of Professional Forecasters." Research Special Report, Federal Reserve Bank of Philadelphia, May. Wachter, J. A. (2006), “A consumption-based model of the term structure of interest rates.”Journal of Financial Economics 79 (2), 365-399. Wright, Jonathan H. (2011), “Term Premia and In‡ation Uncertainty: Empirical Evidence from an International Panel Dataset." American Economic Review, 101(4), 1514–34. 29 Table 1: R2 from regressions with nominal yield factors Dependent Variable 1st factor 1st and 2nd factor 1st , 2nd , .05 .06 E[ 1y ] .83 .84 E[ 11y ] .86 .86 V ar( 1y ) .49 .49 and 3rd factor .06 .84 .87 .51 Note: Entries show the R2 of regressions of each of the in‡ation variables (in the …rst column) on the estimated factors from an a¢ ne term structure model. Table 2: Summary of four alternative model speci…cations Model Model Model Model Model Restrictions and Identi…cations N Full ! 1 unrestricted, q lower-triangular, N 1 (3) = 0; 2 2 N (3) = 0; ! 1 = 03 1 ; q = 03 3 ; N No_TVV 2 2 1 N No_IU 1, ! 1 = 03 1 ; q lower-triangular, N iu 2 2, 1 (3) = 0; q N No_Unsp ! 1 unrestricted, lower-triangular, 1 unrestricted, N 3 3. Table 3: R2 from regressing with full model’s factors Dependent Variable 1st factor 1st and 2nd factor 1st , 2nd , and hidden factor .06 .06 .07 E[ 1y ] .80 .83 .90 E[ 11y ] .70 .84 .89 V ar( 1y ) .44 .47 .83 Note: Entries show the R2 of regressions of each of the in‡ation variables (in the …rst column) on the estimated factors xt from the full model. 30 14 Nominal Yield Real Yield ExPi IRP 12 10 8 6 4 2 0 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Figure 1: Decomposition of the 10-year zero-coupon nominal yield. Chart decomposes the 10-year nominal yield into the 10-year real yield (including the RRP), the expected in‡ation over the next 10 years (ExPi) and the corresponding IRP implied by the full model. 31 Expected Real Rate ExPi IRP RRP 4 3 2 1 0 -1 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Figure 2: Expectation components and risk premia in the 10-year zero-coupon nominal yield. Chart shows the expected average short-term real interest rate over the next 10 years and the corresponding RRP together with the expected in‡ation over the next 10 years (ExPi) and the corresponding IRP, implied by the full model. 32 2 1 Yr Inflation Uncertainty 8 5 Yr Real Yield 6 1.5 4 1 2 0 0.5 -2 1990 2000 2010 1990 1 Yr Expected Inflation 2000 2010 1 Yr TBill Fcast 6 10 4 5 2 0 0 1990 2000 2010 1990 2000 2010 Figure 3: Overall Fit of the Full Model. Model estimates (in blue) compared with actual data and surveys (in orange). The top left panel plots the one-year surveybased in‡ation variance versus the model-implied in‡ation variance; the top right panel plots the 5-year actual TIPS yield versus the 5-year model-implied real yield; the bottom left panel plots the survey-based one-year expected in‡ation versus the one-year model-implied expected in‡ation; and the bottom right panel plots the one-year ahead survey forecast of the 3-month T-Bill rate versus the model-implied one-year ahead expectation of the 3-month rate. 33 Infl Uncer (Pct) 1.5 1.5 1 0.5 1 0.5 2 Yr IRP (Pct) 1990 2000 2010 10 0.5 0 0 2000 2010 2000 2010 1990 2000 2010 1990 2000 2010 1990 2000 2010 6 4 2 0 10 0 -10 1990 1 Yr Ex Inf (Pct) 1990 -0.5 -10 1990 5 Yr R Yld (Pct) Full Model Homoskedastic Model 2000 2010 4 4 2 2 1990 2000 2010 Figure 4: Comparison of homoskedastic and full models. The left panels show results from the model with homoskedastic in‡ation shocks and the right panels show the corresponding results from the full model. The …rst row plots the estimated value of the one-year in‡ation variance and the one-year survey-based in‡ation uncertainty (orange). The second row plots the estimated two-year IRP. The third row plots the model-implied 5-year real yield and the actual 5-year TIPS yield (orange). The last row plots the estimated one-year expected in‡ation versus the one-year ahead survey forecast of in‡ation (orange). 34 Full Model 1 2 Yr IRP (Pct) 2 10 Yr IRP (Pct) 1.5 1 0 1 0 0.5 -1 1990 2000 2010 No IU Model 1 1990 2000 2010 1990 2000 2010 2 1.5 1 0 1 0 0.5 -1 1990 2000 2010 No UnSp Model Infl Uncer (Pct) 1990 2000 2010 6 4 1990 2000 2010 1.5 4 1 2 2 0 0.5 0 1990 2000 2010 1990 2000 2010 1990 2000 2010 Figure 5: In‡ation Risk Premium Estimates and Model Fit of In‡ation Uncertainty. Each row plot results from the Full Model, No IU Model, and No Unsp Model, respectively. The left and middle panels plot the estimated values of the 2-year and 10-year IRP. The right panels plot the one-year model-implied in‡ation variance together its survey counterpart (orange). 35 Full Model 8 10 Yr R Yield (Pct) 4 4 2 0 0 No IU Model 1990 2000 2010 8 4 4 2 0 1990 2000 2010 1990 2000 2010 1990 2000 2010 0 1990 No UnSp Model 10 Yr RRP (Pct) 2000 2010 8 2 4 0 0 -2 1990 2000 2010 Figure 6: Model-Implied Real Yields versus TIPS yields and RRP Estimates. Each row plots results from the Full Model, No IU Model, and No Unsp Model, respectively. The left panels plot the model-implied 10-year real yield together with the actual 10-year TIPS yield (orange). The right panels plot the estimated values of the 10-year RRP. 36 Full Model Exp Inflation 1 Yr (Pct) 6 6 4 4 2 2 No IU Model 1990 2000 2010 6 6 4 4 2 2 1990 No UnSp Model Exp Inflation Over Next 11 Yr (Pct) 2000 2010 6 6 4 4 2 2 1990 2000 2010 1990 2000 2010 1990 2000 2010 1990 2000 2010 Figure 7: Model Fit of Survey In‡ation Expectations at one-year and 11-year horizons. Each row plots results from the Full Model, No IU Model, and No Unsp Model, respectively. The left panels plot one-year model-implied in‡ation expectation together with the corresponding survey forecast (orange). The right panels plot average model-implied in‡ation expectation over the next 11 years together with the corresponding survey forecast (orange). 37 Full Model TBill Fcast 6 Mnth (Pct) Fwd Rate Fcast 6 to 11 Yr (Pct) 8 10 6 5 4 0 2 No IU Model 1990 2000 2010 2000 2010 1990 2000 2010 1990 2000 2010 8 10 6 5 4 0 2 1990 No UnSp Model 1990 2000 2010 8 10 6 5 4 0 2 1990 2000 2010 Figure 8: Model Fit of Survey Forecasts of the Short Rate at short and long horizons. Each row plots results from the Full Model, No IU Model, and No Unsp Model, respectively. The left panels plot the 6-month-ahead model-implied expectation of the 3-month rate together with the 6-month-ahead survey forecast of the 3-month T-Bill rate (orange). The right panels plot the 6-to-11 years ahead model-implied expectation of the 3-month rate together with the 6-to-11 years ahead survey forecast of the 3-month T-Bill rate (orange). 38 Working Paper Series A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics. The Urban Density Premium across Establishments R. Jason Faberman and Matthew Freedman WP-13-01 Why Do Borrowers Make Mortgage Refinancing Mistakes? Sumit Agarwal, Richard J. Rosen, and Vincent Yao WP-13-02 Bank Panics, Government Guarantees, and the Long-Run Size of the Financial Sector: Evidence from Free-Banking America Benjamin Chabot and Charles C. Moul WP-13-03 Fiscal Consequences of Paying Interest on Reserves Marco Bassetto and Todd Messer WP-13-04 Properties of the Vacancy Statistic in the Discrete Circle Covering Problem Gadi Barlevy and H. N. Nagaraja WP-13-05 Credit Crunches and Credit Allocation in a Model of Entrepreneurship Marco Bassetto, Marco Cagetti, and Mariacristina De Nardi WP-13-06 Financial Incentives and Educational Investment: The Impact of Performance-Based Scholarships on Student Time Use Lisa Barrow and Cecilia Elena Rouse WP-13-07 The Global Welfare Impact of China: Trade Integration and Technological Change Julian di Giovanni, Andrei A. Levchenko, and Jing Zhang WP-13-08 Structural Change in an Open Economy Timothy Uy, Kei-Mu Yi, and Jing Zhang WP-13-09 The Global Labor Market Impact of Emerging Giants: a Quantitative Assessment Andrei A. Levchenko and Jing Zhang WP-13-10 Size-Dependent Regulations, Firm Size Distribution, and Reallocation François Gourio and Nicolas Roys WP-13-11 Modeling the Evolution of Expectations and Uncertainty in General Equilibrium Francesco Bianchi and Leonardo Melosi WP-13-12 Rushing into the American Dream? House Prices, the Timing of Homeownership, and the Adjustment of Consumer Credit Sumit Agarwal, Luojia Hu, and Xing Huang WP-13-13 1 Working Paper Series (continued) The Earned Income Tax Credit and Food Consumption Patterns Leslie McGranahan and Diane W. Schanzenbach WP-13-14 Agglomeration in the European automobile supplier industry Thomas Klier and Dan McMillen WP-13-15 Human Capital and Long-Run Labor Income Risk Luca Benzoni and Olena Chyruk WP-13-16 The Effects of the Saving and Banking Glut on the U.S. Economy Alejandro Justiniano, Giorgio E. Primiceri, and Andrea Tambalotti WP-13-17 A Portfolio-Balance Approach to the Nominal Term Structure Thomas B. King WP-13-18 Gross Migration, Housing and Urban Population Dynamics Morris A. Davis, Jonas D.M. Fisher, and Marcelo Veracierto WP-13-19 Very Simple Markov-Perfect Industry Dynamics Jaap H. Abbring, Jeffrey R. Campbell, Jan Tilly, and Nan Yang WP-13-20 Bubbles and Leverage: A Simple and Unified Approach Robert Barsky and Theodore Bogusz WP-13-21 The scarcity value of Treasury collateral: Repo market effects of security-specific supply and demand factors Stefania D'Amico, Roger Fan, and Yuriy Kitsul Gambling for Dollars: Strategic Hedge Fund Manager Investment Dan Bernhardt and Ed Nosal Cash-in-the-Market Pricing in a Model with Money and Over-the-Counter Financial Markets Fabrizio Mattesini and Ed Nosal WP-13-22 WP-13-23 WP-13-24 An Interview with Neil Wallace David Altig and Ed Nosal WP-13-25 Firm Dynamics and the Minimum Wage: A Putty-Clay Approach Daniel Aaronson, Eric French, and Isaac Sorkin WP-13-26 Policy Intervention in Debt Renegotiation: Evidence from the Home Affordable Modification Program Sumit Agarwal, Gene Amromin, Itzhak Ben-David, Souphala Chomsisengphet, Tomasz Piskorski, and Amit Seru WP-13-27 2 Working Paper Series (continued) The Effects of the Massachusetts Health Reform on Financial Distress Bhashkar Mazumder and Sarah Miller WP-14-01 Can Intangible Capital Explain Cyclical Movements in the Labor Wedge? François Gourio and Leena Rudanko WP-14-02 Early Public Banks William Roberds and François R. Velde WP-14-03 Mandatory Disclosure and Financial Contagion Fernando Alvarez and Gadi Barlevy WP-14-04 The Stock of External Sovereign Debt: Can We Take the Data at ‘Face Value’? Daniel A. Dias, Christine Richmond, and Mark L. J. Wright WP-14-05 Interpreting the Pari Passu Clause in Sovereign Bond Contracts: It’s All Hebrew (and Aramaic) to Me Mark L. J. Wright WP-14-06 AIG in Hindsight Robert McDonald and Anna Paulson WP-14-07 On the Structural Interpretation of the Smets-Wouters “Risk Premium” Shock Jonas D.M. Fisher WP-14-08 Human Capital Risk, Contract Enforcement, and the Macroeconomy Tom Krebs, Moritz Kuhn, and Mark L. J. Wright WP-14-09 Adverse Selection, Risk Sharing and Business Cycles Marcelo Veracierto WP-14-10 Core and ‘Crust’: Consumer Prices and the Term Structure of Interest Rates Andrea Ajello, Luca Benzoni, and Olena Chyruk WP-14-11 The Evolution of Comparative Advantage: Measurement and Implications Andrei A. Levchenko and Jing Zhang WP-14-12 Saving Europe?: The Unpleasant Arithmetic of Fiscal Austerity in Integrated Economies Enrique G. Mendoza, Linda L. Tesar, and Jing Zhang WP-14-13 Liquidity Traps and Monetary Policy: Managing a Credit Crunch Francisco Buera and Juan Pablo Nicolini WP-14-14 Quantitative Easing in Joseph’s Egypt with Keynesian Producers Jeffrey R. Campbell WP-14-15 3 Working Paper Series (continued) Constrained Discretion and Central Bank Transparency Francesco Bianchi and Leonardo Melosi WP-14-16 Escaping the Great Recession Francesco Bianchi and Leonardo Melosi WP-14-17 More on Middlemen: Equilibrium Entry and Efficiency in Intermediated Markets Ed Nosal, Yuet-Yee Wong, and Randall Wright WP-14-18 Preventing Bank Runs David Andolfatto, Ed Nosal, and Bruno Sultanum WP-14-19 The Impact of Chicago’s Small High School Initiative Lisa Barrow, Diane Whitmore Schanzenbach, and Amy Claessens WP-14-20 Credit Supply and the Housing Boom Alejandro Justiniano, Giorgio E. Primiceri, and Andrea Tambalotti WP-14-21 The Effect of Vehicle Fuel Economy Standards on Technology Adoption Thomas Klier and Joshua Linn WP-14-22 What Drives Bank Funding Spreads? Thomas B. King and Kurt F. Lewis WP-14-23 Inflation Uncertainty and Disagreement in Bond Risk Premia Stefania D’Amico and Athanasios Orphanides WP-14-24 Access to Refinancing and Mortgage Interest Rates: HARPing on the Importance of Competition Gene Amromin and Caitlin Kearns WP-14-25 Private Takings Alessandro Marchesiani and Ed Nosal WP-14-26 Momentum Trading, Return Chasing, and Predictable Crashes Benjamin Chabot, Eric Ghysels, and Ravi Jagannathan WP-14-27 Early Life Environment and Racial Inequality in Education and Earnings in the United States Kenneth Y. Chay, Jonathan Guryan, and Bhashkar Mazumder WP-14-28 Poor (Wo)man’s Bootstrap Bo E. Honoré and Luojia Hu WP-15-01 Revisiting the Role of Home Production in Life-Cycle Labor Supply R. Jason Faberman WP-15-02 4 Working Paper Series (continued) Risk Management for Monetary Policy Near the Zero Lower Bound Charles Evans, Jonas Fisher, François Gourio, and Spencer Krane Estimating the Intergenerational Elasticity and Rank Association in the US: Overcoming the Current Limitations of Tax Data Bhashkar Mazumder WP-15-03 WP-15-04 External and Public Debt Crises Cristina Arellano, Andrew Atkeson, and Mark Wright WP-15-05 The Value and Risk of Human Capital Luca Benzoni and Olena Chyruk WP-15-06 Simpler Bootstrap Estimation of the Asymptotic Variance of U-statistic Based Estimators Bo E. Honoré and Luojia Hu WP-15-07 Bad Investments and Missed Opportunities? Postwar Capital Flows to Asia and Latin America Lee E. Ohanian, Paulina Restrepo-Echavarria, and Mark L. J. Wright WP-15-08 Backtesting Systemic Risk Measures During Historical Bank Runs Christian Brownlees, Ben Chabot, Eric Ghysels, and Christopher Kurz WP-15-09 What Does Anticipated Monetary Policy Do? Stefania D’Amico and Thomas B. King WP-15-10 Firm Entry and Macroeconomic Dynamics: A State-level Analysis François Gourio, Todd Messer, and Michael Siemer WP-16-01 Measuring Interest Rate Risk in the Life Insurance Sector: the U.S. and the U.K. Daniel Hartley, Anna Paulson, and Richard J. Rosen WP-16-02 Allocating Effort and Talent in Professional Labor Markets Gadi Barlevy and Derek Neal WP-16-03 The Life Insurance Industry and Systemic Risk: A Bond Market Perspective Anna Paulson and Richard Rosen WP-16-04 Forecasting Economic Activity with Mixed Frequency Bayesian VARs Scott A. Brave, R. Andrew Butters, and Alejandro Justiniano WP-16-05 Optimal Monetary Policy in an Open Emerging Market Economy Tara Iyer WP-16-06 Forward Guidance and Macroeconomic Outcomes Since the Financial Crisis Jeffrey R. Campbell, Jonas D. M. Fisher, Alejandro Justiniano, and Leonardo Melosi WP-16-07 5 Working Paper Series (continued) Insurance in Human Capital Models with Limited Enforcement Tom Krebs, Moritz Kuhn, and Mark Wright WP-16-08 Accounting for Central Neighborhood Change, 1980-2010 Nathaniel Baum-Snow and Daniel Hartley WP-16-09 The Effect of the Patient Protection and Affordable Care Act Medicaid Expansions on Financial Wellbeing Luojia Hu, Robert Kaestner, Bhashkar Mazumder, Sarah Miller, and Ashley Wong WP-16-10 The Interplay Between Financial Conditions and Monetary Policy Shock Marco Bassetto, Luca Benzoni, and Trevor Serrao WP-16-11 Tax Credits and the Debt Position of US Households Leslie McGranahan WP-16-12 The Global Diffusion of Ideas Francisco J. Buera and Ezra Oberfield WP-16-13 Signaling Effects of Monetary Policy Leonardo Melosi WP-16-14 Constrained Discretion and Central Bank Transparency Francesco Bianchi and Leonardo Melosi WP-16-15 Escaping the Great Recession Francesco Bianchi and Leonardo Melosi WP-16-16 The Role of Selective High Schools in Equalizing Educational Outcomes: Heterogeneous Effects by Neighborhood Socioeconomic Status Lisa Barrow, Lauren Sartain, and Marisa de la Torre Monetary Policy and Durable Goods Robert B. Barsky, Christoph E. Boehm, Christopher L. House, and Miles S. Kimball WP-16-17 WP-16-18 Interest Rates or Haircuts? Prices Versus Quantities in the Market for Collateralized Risky Loans Robert Barsky, Theodore Bogusz, and Matthew Easton WP-16-19 Evidence on the within-industry agglomeration of R&D, production, and administrative occupations Benjamin Goldman, Thomas Klier, and Thomas Walstrum WP-16-20 Expectation and Duration at the Effective Lower Bound Thomas B. King WP-16-21 6 Working Paper Series (continued) The Term Structure and Inflation Uncertainty Tomas Breach, Stefania D’Amico, and Athanasios Orphanides WP-16-22 7