We present three essays on the econometric modeling of time-varying densities. In all three studies, we treat density functions themselves as random elements taking values in the Hilbert space of square integrable functions. The first essay introduces functional autoregression of one sequence of time-varying density functions. The second essay develops functional regression of one sequence of densities, the regressand, on another sequence of densities, the regressor. The third essay is concerned with the regression of a sequence of scalar random variables on a sequence of densities.
For all three models, we present methods of estimation, which all involve the solution of an ill-posed inverse problem, and show asymptotic consistency of our estimators. We also outline hypotheses testing strategies based on each model and show asymptotic distribution of test statistics that we have developed. Possible applications of each model in economics and finance are indicated in each essay. In particular, we apply the functional autoregression model to the analysis of intraday return distributions of S&P 500 index and US/UK exchange rate and find that the functional method offers an outstanding performance in out-of-sample forecasting and a unique way to test on the moment dependence structure of time-varying distributions.
| Identifer | oai:union.ndltd.org:RICE/oai:scholarship.rice.edu:1911/20635 |
| Date | January 2007 |
| Creators | Qian, Junhui |
| Contributors | Park, Joon Y. |
| Source Sets | Rice University |
| Language | English |
| Detected Language | English |
| Type | Thesis, Text |
| Format | 183 p., application/pdf |
Page generated in 0.0016 seconds