Essays on Non-linearities in Stock and Bond Returns: A Density-Based Approach.

摘要

The dissertation consists of three essays that revolve around non-linearities embedded in asset returns.;In the first essay "The Role of Slope Heterogeneity in Bond Excess Returns Predictability", I investigates bond excess return forecastability using current forward rates. The dynamics of excess return are modeled non-parametrically. Estimation shows heterogeneous slopes for independent variables, indicating the existence of non-linearity. Empirically, I find this non-linearity plays an important role in excess return prediction both in- and out-of-sample. By including non-linearity in the model, the in-sample R2 jumps to as high as 91%. Meanwhile, lagged forward rates are no longer statistically significant, in contrast to the results documented in previous research. The out-of-sample forecasts also favor the non-parametric model. Findings in this paper suggest a potential important information source embedded in the current forward rates cross-section. Information associated with non-linearity is largely ignored in the existing literature as it is averaged out by linear model settings.;The second essay "Do Non-Linearities Matter in the Yield Curve?" tries to answer the question that do non-yield variables contain information beyond what is contained in the yield curve? Using a non-linear factor extracted from the yield curve, I find nonyield factors, which are constructed from a large panel of macro-finance data, are no longer significant in predicting future bond excess returns both in- and out-of-sample. Moreover, my non-linear factor generates countercyclical and business cycle frequency bond risk premia. The findings underscore the importance of non-linearities embedded in the term structure, suggesting a fully spanned term structure model with non-linear state factors may be capable of matching features observed in the data.;In the third essay "A Test on Asymmetric Dependence" (joint work with Prof. Maasoumi, Lei Jiang and Ke Wu), we provide a model-free test for asymmetric dependence between stock and market returns, based on the Kullback-Leibler mutual information measure. Our test has greater power in small samples than previous tests of asymmetric correlation proposed by Hong, Tu and Zhou (2007). Empirically, we find that asymmetric dependence is a prevailing phenomenon in most commonly used portfolios.