| The dissertation reviews some popular nonparametric and semiparametric regression methods and applies them in Finance and Economics.;In Finance, we re-examine the linear relationship between the anomalous stock returns and the firm characteristic (anomaly) variables from a nonparametric perspective. The dataset is divided into five groups -- Micro, Small, Big, All but Micro (Small plus Big) and All (Micro plus Small plus Big). The breakpoints separating Micro from Small and Small from Big are the 20th and 50th percentiles of market capitalization for NYSE stocks. We find that the linear relationship does not exist across all groups. But the nonlinearity problem is minor except in the All group. The presence of extreme values of the anomaly variables and the group parameter heterogeneity appear to be the cause of nonlinearity in the All group.;In Economics, we argue that the insignificance of public capital on state level production is caused by the linear assumption of the time and state effects and not by the Cobb-Douglas specification. To avoid the linear assumption on the state and time effects, we use the semi-parametric smooth coefficient model to get the estimates of input elasticities. In contrast to Henderson and Kumbhakar (2006), we find that the return to public capital is positive and significantly different from zero assuming a Cobb-Douglas production function. Further, we argue that the relief funds from the economic stimulus package to state infrastructure may be necessary and productive. |
