Since Bansal and Yaron (2004), the …nance literature has stressed the importance of long-run risk in consumption growth. In contrast, there has been little attention devoted to long-run nominal risks in the economy, speci…cally, time-variation in the economy’s long-run in‡ation rate, even though such risk would be very relevant for pricing nominal bonds. We therefore consider the case where the monetary authority’s target rate of in‡ation, t, varies over time. Certainly,
…nancial market perceptions of the long-run in‡ation rate in the United States appear to have varied considerably in recent decades: Kozicki and Tinsley (2001) show that survey data on long-run in‡ation expectations have varied substantially over the past 50 years, Rudebusch and
Wu (2007, 2008) estimate a similar degree of variation in a macro-…nance no-arbitrage model, and Gürkaynak, Sack, and Swanson (2005) …nd that the “excess sensitivity”of long-term bond yields to macroeconomic announcements is consistent with …nancial markets perceiving the long-run in‡ation rate in the economy to be less than perfectly anchored.
From the point of view of modeling the term premium, long-run in‡ation risk has a number of advantages over long-run consumption risk. First, estimates of the low-frequency component of consumption are extremely imprecise, so it is very di¢ cult to test empirically the direct predictions of a Bansal-Yaron long-run consumption risk model with observable macroeconomic variables. In contrast, survey data on long-run in‡ation expectations are readily available and show considerable variation. Second, the idea that long-term nominal bonds are risky because of uncertainty about future monetary policy and long-run in‡ation is intuitively appealing.
Third, estimates of the term premium in the …nance literature are low in the 1960s, high in the late 1970s and early 1980s, and then low again in the 1990s and 2000s (e.g., Kim and Wright, 2005), which suggests that in‡ation and in‡ation variability are highly correlated with the term premium, at least over these longer, decadal samples. Modeling the linkage between long-run in‡ation risk and the term premium thus seems to be a promising avenue for understanding and modeling long-term bond yields.
Following the empirical evidence in Gürkaynak et al. (2005),we assume that t loads to some extent on the recent history of in‡ation:
t = t 1+ # ( t t) + "t : (43)
There are two main advantages to using speci…cation (43) rather than a simple random walk or AR(1) speci…cation with # = 0. First, (43) allows long-term in‡ation expectations to respond to current news about in‡ation and economic activity in a manner that is consistent with the bond market responses documented by Gürkaynak et al. Thus, # > 0 seems to be consistent with the data.33 Second, if # = 0, then even though t varies over time, it does not do so systematically with output or consumption; as a result, long-term bonds are not particularly risky, in the sense that their returns are not very correlated with the household’s stochastic discount factor. In fact, long-term bonds even have some elements of insurance in this case, because a negative shock to "t leads the monetary authority to raise interest rates and depress
33 Gürkaynak et al. …nd that a value of # = :02 is roughly consistent with the bond market data.
output at the same time that it causes long-term bond yields to fall and bond prices to rise, which results in a negative term premium on the bond. By contrast, if # > 0, then a negative technology shock today raises in‡ation and long-term in‡ation expectations and depresses bond prices at exactly the same time that it depresses output, which makes holding long-term bonds quite risky and helps the model to match the positive mean term premium we see in the data.
We add equation (43) to our DSGE model from the preceding section, setting the baseline value of # = :02, which is consistent with the high-frequency bond market evidence in Gürkay-nak et al. (2005). We set the baseline values for and equal to .99 and 5 basis points, respectively, consistent with the Bayesian DSGE model estimates in Levin et al. (2005).
As can be seen in Figure 1, the e¤ects of the long-run in‡ation risk on the term premium are indeed substantial. As the quasi-CRRA is varied along the horizontal axis, holding the other parameters of the model …xed at their baseline values, the term premium is always the highest for the version of the model with long-run in‡ation risk. That is, by making long-term bonds in the model riskier, the model can generate any given level of the term premium with a lower value for the quasi-CRRA than was possible without long-run in‡ation risk.
Table 3 reports the macroeconomic and …nancial moments that result from introducing long-run in‡ation risk into our DSGE model. The …rst column repeats the empirical moments from the U.S. data, and the second column reports results for a version of the model with long-run in‡ation risk and expected utility preferences (that is, with the parameters of the model set to their baseline values, except for = 0). The introduction of time-variation in makes the macroeconomic variables a little more volatile on average, but the …t of the model to the macro data is about as good overall as for the baseline model. The …t of the model to the …nancial moments, however, is also no better— the term premium is still less than one basis point, and its variation is still only about one-tenth of one basis point, far smaller than the data (and this result is extremely robust to varying the parameters of the model over wide ranges). Intuitively, long-run in‡ation risk increases the quantity of nominal bond risk in the model, but households simply aren’t risk-averse enough for that greater quantity of risk to have a noticeable e¤ect on bond prices in the model.
With Epstein-Zin preferences, however, introducing long-run in‡ation risk into the model has substantial e¤ects. The third column of Table 3 reports results for the model with
Epstein-Table 3
Empirical and Model-Based Unconditional Moments with Long-Run Risk Model with Model with Model with Unconditional U.S. Data, EU Preferences EZ Preferences EZ Preferences
Moment 1961-2007 and LR Risk and LR Risk and LR A Risk
sd[C] 1.19 1.70 2.01 2.37
sd[L] 1.71 3.02 1.37 2.13
sd[wr] 0.82 2.40 1.52 1.81
sd[ ] 2.52 3.65 3.25 2.95
sd[i] 2.71 3.32 2.94 2.86
sd[r] 2.30 2.39 1.71 1.55
sd[i(40)] 2.41 1.71 1.89 1.66
mean[ (40)] 1.06 :003 1.05 0.98
sd[ (40)] 0.54 .001 0.51 0.28
mean[i(40) i] 1.43 .10 0.96 0.89
sd[i(40) i] 1.33 1.73 1.10 1.36
mean[x(40)] 1.76 .003 1.04 0.96
sd[x(40)] 23.43 13.07 11.64 12.20
distance to:
macro moments 6.62 2.48 3.88
…nance moments 7.80 2.18 2.27
all moments 14.42 4.67 6.15
memo:
quasi-CRRA 2 90 90
IES 0.5 1.1 0.5
1.5 0.5 1.5
0.75 0.65 0.8
A .9 .95
A .01 .005 .001
.99 .99
# .02 .015
5bp 1bp
A .97
A .005
All variables are quarterly values expressed in percent. In‡ation and interest rates, the term premium ( ), and excess holding period returns (x) are expressed at an annual rate.
Zin preferences and long-run in‡ation risk, where we have searched over values for , # , and as well as the quasi-CRRA, IES, , , A, and A to …nd the best …t to the empirical moments in the …rst column.34 Relative to the the second column, the term premium and other
…nancial moments generated by the model are much larger and much more in line with the data. Relative to the case of no long-run risk— the last column of Table 2— the term premium is far more variable once long-term in‡ation risk is incorporated into the model.35 Intuitively, when # > 0, technology (and other) shocks have an ever more persistent e¤ect on in‡ation because of the pass-through from t to t, which makes long-term nominal bonds in the model even more risky. As a result, the term premium in the model is larger and more volatile.
The estimation achieves this improvement in …t by choosing a high value for the quasi-CRRA, which helps to …t the term premium and other …nancial moments with relatively moderate consumption volatility. (Alternatively, the model with long-run in‡ation risk can …t the macro and …nancial moments just as well as the model without long-run risk, using a lower value for the quasi-CRRA). The time-variation in makes the model as a whole more volatile, so the estimation compensates for this by choosing a lower technology shock variance, A= :005; the greater degree of nominal volatility in the model due to time-varying , together with the smaller degree of real volatility due to technology shocks, improves the overall …t of the model to the macro moments. The estimation also prefers a higher value for the IES, 1.1, which helps shift some of the volatility of consumption over to short-term real interest rates, in line with the data, and a low value for (a high Frisch elasticity of labor supply), which helps shift some of the volatility of real wages over to labor. Finally, a low value for …ts the data the best— as discussed above, exogenous shocks to actually imply a lower term premium, all else equal, because long-term nominal bonds in the model act like insurance against this particular type of shock. It is the loading # of on current in‡ation that makes time-variation in costly in the model, not exogenous shocks to .
34 In addition to the range of parameter values considered in the previous section, we searched over values of 2 f:98; :99; :995; :997; :998g, # 2 f0; :005; :01; :015; :02g, and 2 f1; 2; : : : ; 15g basis points.
35 These results hold for a ten-year zero-coupon bond in the model as well: the term premium has a mean of 76.2 bp and a standard deviation of 39.7 bp. These are a few basis points less than for the consol, but the main points in the text are all unchanged.