Essays on Non-Gaussian Time Series Analysis

Abstract

This thesis is a compilation of essays on the extension of financial econometric techniques to various fields of financial and non-financial risk management-- namely, longevity risk, disaster risk and food security risk.

First, longevity risk is quantified by proposing a mortality forecasting methodology based on a modified survival function and nonparametric residual-based bootstrapping. The parameters of the survival function are estimated through time and are modelled with a time series model structure. The estimated model is used to generate forecasts of parameter values and life expectancy. Confidence intervals are generated by residual-based bootstrapping through an autoregressive sieve based on the estimated model. The methodology is applied to life tables of male and female subjects from the United States, Australia and Japan, and compared with the Lee-Carter model in terms of forecasting life expectancy. From the results for the three countries, the proposed survival function has better long-term forecasting performance than does the Lee-Carter model.

Second, a proposed methodology for estimating disaster risk is devised using bootstrapped multivariate extreme value theory methods. A disaster risk measure called storm-at-risk is created. The risk measure can be estimated through semiparametric and nonparametric approaches and is applied to weather extremes data generated by typhoons that enter the western North Pacific basin. Robustness checks on the performance of the approaches are conducted. The semiparametric approach performs better than the nonparametric approach in longer periods, but not in smaller periods.

Third, food security risk is quantified by proposing risk measures for hierarchical agricultural time series data, which are generated for national and sub-national levels. The risk measures are created by a combination of forecast reconciliation methods for hierarchical time series data and residual-based bootstrapping methods. The methodology is applied to Philippine rice production time series data that are collected from the regions and are aggregated to the macro-regional and national levels. Show more