STAT 4870/7870 - Fall 2019
Final Project
Due Thursday, December 5, 2019 (by 5:00 pm by email)
If you have any questions please come to office hours or email me or set up an appointment!
Note: Projects will be worked on individually or in groups of two. Note that projects worked on in groups
of two require prior approval of the group formed and should be obtained as soon as possible (by 10/8/19).
Option 1:
STAT 4870:
Choose an interesting time series that can be modeled using a seasonal ARIMA(p, d, q) × (P, D, Q)s model
(see Section 3.9 tsaEZ). Determine two good models for your data using the methods learned throughout
the semester. Obtain forecast intervals for 2s time periods into the future. Write a report based on your
analysis.
STAT 7870:
A real time series, should be used for the analysis. You must choose a dataset where the analysis involves
a method of analysis that we have not formally discussed in the class. Please see me to discuss this when
you have some ideas.
Guidelines for all students:
1. Briefly describe the process from which your realization came and give the source of your data.
Explain the purpose of your analysis. Include a table of the data.
2. Decide on two final models, both of which fit the data well. Compare forecasts for these best two
models, and briefly comment on any differences.
3. Your reports should be concise but complete. The final written report should be typed and include
STAT 4870/7870作业代写、R程序设计作业调试
a one-page non-technical summary of the findings, followed by the background of the scientific
questions, the body of technical analyses with interpretations, a conclusion and the listing of the
data. Including graphics, the report should not exceed 8 pages in length (unless prior approval is
given). Reports should not include any raw computer output. Use cutting and pasting to present
only relevant computer output and graphs. Output is relevant if you refer to it in your write up.
4. Your project will be graded on accuracy, completeness, relevance of your graphical displays, tabular
comparisons, and your statements, and on the report’s organization, conciseness, and neatness.
Whenever possible use graphs plus a few words to make your points. Number your figures and refer
to them in your write up. Reports must be typed or done using a word processing system.
5. On a separate (but attached) page provide a one paragraph description (for each student) detailing
the contributions made to the final project.
How to choose data for the final project:
Most importantly, the process that generates data should be interesting to YOU. You should understand
the basics of the data you work with (For example, do not choose outcomes of rugby tournaments if you
1
do not know the rules of the game and do not want to learn them as a part of your project.) The data
should be long enough to ensure that you can perform meaningful statistical analysis. I suggest you stick to
time series with at least 180 observations (the more the better). DO NOT choose a dataset just because
you found how other people analyzed it and want to use this as a guide. This will seriously hurt your grade.
A one-page proposal outlining the scientific questions to be addressed and the relevant techniques
to be employed, with a separate listing of the data, has to be turned in by email on
Friday October 25 (by 5 pm).
Option 2 (STAT 7870 Only): Use R to carry out a simulation study
Propose a simulation study to be investigated in R. Note that this option is suggested for graduate students
only, or those with good programming skills and a good background in statistical inference. Feel free to
run your proposed idea by me before beginning. I can also email you with ideas if you are having a hard
time coming up with one. Some example problems include:
1. Consider the ARIMA (p, 0, 0) model, that is, a pure AR model. Design a simulation with studies the
effect of the error distribution on the behavior of the MLE estimator(s) φi
. You should have factors
sample size n, true parameter value(s), and error distribution.
2. Consider the ARIMA (p, 0, 0) model, that is, a pure AR model. The R function for fitting an autoregressive
model an output vector asy.var.coef of asymptotic variance of the MLE parameter
estimates. Design a simulation with compares the Monte Carlo estimate of variance to these asymptotic
variances, as function of n (The two should get closer as n increases.)
3. Design a simulation with studies the performance of the estimated autocorrelation function. (Recall
that the true autocorrelation function for ARMA model can be obtained using the R command
ARMAacf.) There are many situations here, and you will have to choose a particular one. Certainly
sample size will be a factor, but you should also study the effect of other factors.
A one-page proposal outlining the specific simulation study to be addressed and the relevant
techniques to be employed has to be turned in by email on Friday October 25 (by 5 pm).
Some possible DATA Sources: (You can use others that are not listed here.)
Time Series Data Library:
https://datamarket.com/data/list/?q=provider:tsdl
StatLib Datasets Library at Carnegie Mellon:
http://lib.stat.cmu.edu/datasets/
NBER Data - National Bureau of Economic Research
http://www.nber.org/data/
Federal Reserve Economic Data
http://research.stlouisfed.org/fred2/
DATA.GOV
https://www.data.gov/
因为专业，所以值得信赖。如有需要，请加QQ：99515681 或邮箱：[email protected]
微信：codehelp