[Colloquium] [statseminars] Statistics Master's Students Presentations for Friday May 16, 2008: JIN XU, TAKINTAYO AKINBIYI, SANG JIN PARK, YOU CHAN KEVIN CHUNG

Fri May 9 17:12:01 CDT 2008

The University of Chicago
Department of Statistics
Master's Seminar

JIN XU
Departments of Sociology and Statistics
The University of Chicago

GARCH Local Approximation for Non-stationary Financial Data
Through 1958-2007 and Its Influence in Value at Risk

FRIDAY, May 16, 2008 at 8:30 AM
110 Eckhart Hall, 5734 S. University Avenue

ABSTRACT

Local approximation is a useful way to deal with globally non-stationary 
financial data in practice. In this paper, I show that for certain time 
periods when market is unstable, GARCH model with local approximation 
established both adequacy and accurate prediction for the mean and 
volatility structure of the data. Exploring the DJIA index from 1958 to 
2007, I compare the goodness-of fit and root mean square error between 
locally approximated short-term (ST) GARCH models and long-term (LT) 
GARCH models. The findings support my argument that short-term GARCH 
models (ST) excel in both model adequacy and prediction accuracy. 
Besides, the Value at Risk (VaR) for January 2008 is constructed from 
both ST and LT model statistics. It turns out that the VaR from ST 
models reflects the recent market instability better than the VaR from 
the LT models.

--------------------------

The University of Chicago
Department of Statistics
Master's Seminar

TAKINTAYO AKINBIYI
Department of Statistics
The University of Chicago

An Intensity Based Patchwork of Parts Model For 2D Objects

FRIDAY, May 16, 2008 at 10:00 AM
110 Eckhart Hall, 5734 S. University Avenue

ABSTRACT

We develop a deformable model for 2D images and a likelihood based 
estimation process for training the model using the EM algorithm. An 
image is modeled as a patchwork of parts which can overlap and move 
within a small area and can vary in contrast and intensity. A
significant benefit is that whereas most deformable models only model a 
few non-overlapping landmark areas of an image, our model is dense, so 
close to the whole image is covered by parts which are allowed to 
overlap. The ability of the parts to vary in intensity and contrast
allows them to match various lighting sources. Some strategies are 
proposed for how to determine the value of unknown variables in a 
trained model given a new image. These methods prove robust to changes 
in lighting from the original training set. Examples are presented using 
face images.

-----------------------------

The University of Chicago
Department of Statistics
Master's Seminar

SANG JIN PARK
Departments of Sociology and Statistics
The University of Chicago

Hazard of Recurrence in Lymph Node-Negative Breast Cancer:
Treatment Effects over Time by ER Status

FRIDAY, May 16, 2008 at 11:00 AM
110 Eckhart Hall, 5734 S. University Avenue

ABSTRACT

Previous studies of disease recurrence after surgery in breast cancer 
patients have identified time-dependent patterns in the instantaneous 
risk (i.e. hazard) of recurrence, and that these patterns differ by 
disease characteristics such as tumor estrogen-receptor (ER) content.
Benefits from post-surgical chemotherapy and hormonal therapy treatments 
have also been established. Using a cohort of 9,279 participants from 
five randomized trials, we examine and verify these observations using 
standard survival analysis techniques, such as Kaplan-Meier curves, 
nonparametric hazard estimates, and the Cox proportional hazards models.
We further explore time-varying hazards according to ER and estimate 
time-varying treatment effects, using kernel-based smoothing estimation 
of the log hazard ratio with point-wise confidence bounds and 
simultaneous confidence bands over the time span.

Results validated the crossing of hazard functions for patients with 
ER-negative tumors relative to those with ER-positive tumor, with the 
former initially having higher failure hazard that in later follow-up is 
smaller than that for ER-positive tumors. Within both ER groups, the Cox 
regression model suggests significant reduction in risks for all 
treatment types, but the proportionality test fails to confirm the 
constant hazard ratio over time. Nonparametric estimates of log hazard 
ratio over time and their simultaneous confidence intervals suggest that 
all treatments reduce the risk significantly early in the follow up 
time, but that these effects diminish over time.

----------------------------

The University of Chicago
Department of Statistics
Master's Seminar

YOU CHAN KEVIN CHUNG
Department of Statistics
The University of Chicago

Three-Step Disaggregate Forecasts of Personal Health Care Cost

FRIDAY, May 16, 2008 at 1:30 PM
110 Eckhart Hall, 5734 S. University Avenue

ABSTRACT

This paper assesses three methods of forecasting model for the Personal 
Health Care (PHC) costs in the United States. First forecast, the 
simplest approach of the three, is done by taking the aggregate PHC 
datasets and forecasting for 2006–2015 by the ARIMA model. Second 
approach takes into consideration of the differences in PHC expenditure 
by age group and forecasts the future PHC by disaggregating the total 
PHC by age group and extrapolating by the ARIMA model before aggregating 
to find the total PHC estimate. The third approach takes each age group 
and further disaggregates into expected patient number and expected 
average medical cost. The patient numbers were estimated by using the 
ratio between the historic discharge number by age group and the total 
population of that age group for that given year. For the expected 
average medical cost, extrapolation took place by the ARIMA model. The 
result show that the total PHC forecast and the disaggregated PHC 
forecast by age group differ by $64 billion dollars in 2006 and 
increases to $621 billion dollars by 2015. Similarly, for the first and 
the third approach, the forecast differs by $100 billion in 2006 and 
increases to $609 billion dollars by 2015. With three different 
forecasts, the result has a significant implication for future policy 
and business plans.

Information about building access for persons with disabilities may be 
obtained in advance by calling Jewanna Carver
at 773.834-5169 or by email (carver at galton.uchicago.edu).
_______________________________________________
statseminars mailing list  -  statseminars at listhost.uchicago.edu
https://listhost.uchicago.edu/mailman/listinfo/statseminars