This paper quantifies how variation in real economic activity and inflation in the U.S. influenced the market prices of level, slope, and curvature risks in U.S. Treasury markets. To accomplish this we develop a novel arbitrage-free DTSM in which macroeconomic risks – in particular, real output and inflation risks – impact bond investment decisions separately from information about the shape of the yield curve. Estimates of our preferred macro-DTSM over the twenty-three year period from 1985 through 2007 reveal that unspanned macro risks explained a substantial proportion of the variation in forward terms premiums. Unspanned macro risks accounted for nearly 90% of the conditional variation in short-dated forward term premiums, with unspanned real economic growth being the key driving factor. Over horizons beyond three years, these effects were entirely attributable to unspanned inflation. Using our model, we also reassess some of Chairman Bernanke's remarks on the interplay between term premiums, the shape of the yield curve, and macroeconomic activity.
This paper analyzes how the introduction of macro information into a dynamic term structure model affects model-implied investor demand for long-term bonds and quantifies the risk-adjusted return increment from a strategy conditioned on macro information (economic activity and inflation). The main findings are: (1) When the investor is allowed to condition on macro information, her investment strategy becomes strongly counter-cyclical with respect to economic activity, and depends positively on the real level of interest rates; (2) A strategy that conditions on macro information outperforms a strategy that conditions only on yield information (9.6% vs. 6.7% risk-adjusted annualized return for a typical short-term investor over the sample period considered); (3) For long-term investors, macro information becomes relatively more valuable (over and above yield information) out to an investment horizon of approx. 15 months, and relatively less valuable thereafter. Using a split-sample analysis, I demonstrate that the superior investment performance of the macro strategy is robust and not the result of look-ahead bias. I present complementary theoretical and simulation evidence that the investment performance criterion is not subject to the same in-sample overfitting problem as likelihood-based criteria.
Gaussian dynamic term structure models can produce misleading results when current short-term rates are close to zero, as model-implied rates are not restricted to be non-negative. I estimate models of the U.S. treasury curve with and without an explicit zero bound, and find that mispricing can also occur when current rates are at more moderate levels. Furthermore, ignoring the zero bound can produce biased estimates of the model's pricing parameters as a secondary source of misfit.
This study explores the potential difficulty in identifying risk-neutral parameters when estimating arbitrage-free term structure models. I argue that such identification is accomplished primarily through higher-order principal components of yields. These higher-order PCs are sensitive to the splining procedure used in deriving zero-coupon yields from observed yields on coupon-bearing bonds. I explore the sensitivity of parameter estimates to the splining procedure, and compare them to a model estimated based on coupon yields. I find that when splining noise in the higher-order variation of yields is too high, maximum likelihood estimation of risk-neutral parameters may fail altogether.
This paper develops a two-period moral hazard model in which output is privately observed by the principal. I study whether and when the principal should optimally reveal the history of past output realizations to the agent. The existing literature on subjective performance evaluation finds that, if output is distributed independently over time when conditioned on effort, there should be no interim information revelation. In particular, in a finite period model, incentives to exert effort can be provided most efficiently when all information revelation occurs at the end of the game, and the agent's compensation is made contingent on the entire time path of output realizations. In contrast, I find that an interim performance review may raise ex ante surplus and profits if there is time dependence in output through an unobserved state variable. If the principal reveals early output realizations, this serves as a signal of the state and thus allows the agent to choose a more efficient effort level in later periods. When designing an incentive scheme, the principal thus faces a tradeoff between efficient information revelation and dynamically efficient resolution of moral hazard.
This paper develops a model in which one ruler interacts with two administrators in an uncertain environment with private signals. The ruler must decide whether to call an assembly at which the administrators come together to pool their information. The administrators, in turn, must decide whether or not they will give support to the ruler. Their optimal decision depends on the state of the world. I show that there are some environments in which the ruler optimally calls the assembly, while there are others in which he optimally does not do so. These environments can be characterized in terms of the exogenous parameters of the model. In particular, the ruler should call the assembly if this will help convince some administrators to give their support when they otherwise might not do so. Conversely, the assembly should not be called if there is a risk of discouraging some administrators from supporting the ruler when they otherwise might. I interpret two historical episodes in terms of these findings, and consider various comparative static results and extensions to the basic model.