Keynote Lectures

We propose a new class of autoregressive conditional interval (ACI) models for interval-valued time series data. As an appealing measure for variability and uncertainty, intervals are often of direct interest for forecast. An interval-valued observation in a time period contains more information than a point-valued observation in the same time period. Examples of interval data include the maximum and minimum temperatures in a day, the maximum and minimum GDP growth rates in a year, the maximum and minimum stock prices in a trading day, the ask and bid prices in a trading period, and the 90 and 10 incomes of a cohort in a year, etc. Estimation methods, particularly a minimum distance estimation method, are proposed to estimate the parameters of an ACI model. We establish the consistency and asymptotic normality of the proposed estimators and construct some hypothesis tests for model parameters. Simulation studies show that the use of interval time series data can provide more accurate estimation for model parameters than point-valued observations in terms of mean squared error criterion, even when partial information (e.g., the difference or range between the right and left bounds) of an interval is of interest.

Is it possible to develop simplified models to gain insights for large and complex biologic networks? This talk will discuss our attempts to develop statistical methods for this purpose that include network reconstruction by Boolean networks, studies of yeast transcription factors and evolution of the yeast protein interaction network. Future developments regarding this direction will be discussed as well.

Based on detailed trade records of individual investors who participated in both China’s A- and B- share markets, we find investors are more likely to buy A (B) shares when the A-share premium is lower (higher), when they have already held the same firm’s A (B) shares, when they have previously traded the same firm’s A (B) shares and when their A-share portfolio outperforms B-share portfolio. Given that the correlation between the same firm’s A and B shares is below 70% and that A shares are more expensive, it is sensible for investors to invest more into the B shares. Our study suggests that investors accept a less than optimal portfolio due to lack of investment experience.

The curve time series framework provides a convenient vehicle to accommodate some nonstationary features into a stationary setup. We propose a new method to identify the dimensionality of curve time series based on the dynamical dependence across different curves. The practical implementation of our method boils down to an eigenanalysis of a finite-dimensional matrix. Furthermore, the determination of the dimensionality is equivalent to the identification of the non-zero eigenvalues of the matrix, which we carry out in terms of some bootstrap tests. Asymptotic properties of the proposed method are investigated. In particular, our estimators for zero-eigenvalues enjoy the fast convergence rate n while the estimators for non-zero eigenvalues converge at the standard root-n rate. The proposed methodology is illustrated with both simulated and real data sets.