It is a good class. The instructions are very detailed and the topics are interesting.
If there are any improvements, it would be the question for projects can be a lot more clearer. Sometimes I had to input a lot more codes simply because what it asks for.
A very bad course
Mismanagement by the teaching assistants (TAs). Lackluster video lectures from the professor, who seemed unenthusiastic and merely read from slides instead of engaging the students. Reading directly from textbooks would have been more effective than these lectures.
The course paled in comparison to offerings on platforms like Coursera, which generally provide higher-quality educational content.
I'll say it - I think the bad press this class gets is unfair. There are certainly a few areas where this class could improve - some multiple choice questions would benefit from additional QA, and the TA response times are generally pretty terrible.
However, the pacing of the class, the content covered, and the assignments all felt very fair. In particular, the four data analysis homeworks were really helpful in assessing understanding of course content, and provided me with scripts that I can use in my job. Lectures were on the dense side, but overall provide a lot of useful context and information on classical TS models. Exams are open book, but closed internet, and longer than average (8-9 total hours over the course of the class).
It's clear that they have spent a lot of time revamping the course, and it shows. If time series analysis is a course of interest to you, don't let the reviews scare you away.
This course is highly effective for anyone looking to dive deep into Time Series analysis and its applications in predicting financial data. However, if you're aiming for an easy A, this course may not be the right fit.
The curriculum demands a thorough understanding of univariate and multivariate time series concepts, and exams are designed to challenge you rigorously on these topics. They are far from easy, pushing you to explore the subject matter in depth and test your mastery comprehensively.
If you're genuinely interested in learning Time Series models and willing to put in the effort, this course is an excellent choice. (Dr. Sokol’s instruction in 6501 covers Time Series models at a high level, but this course takes a much more rigorous and detailed approach.)
For me, this was the most challenging and rewarding course in the OMSA program. It truly embodied the rigor and difficulty expected at the graduate level. I feel proud to have learned so much throughout the experience.
A word of caution: the TAs can be slow to respond, especially near deadlines. However, I found the professor to be accessible and supportive whenever I faced such issues, which made a significant difference.
Overall, a solid course and good use of time for those interested in learning the fundamentals of Time Series Analysis!
You can tell Dr. Serban has taken prior feedback and tried to revamp the course to be more applicable, well-structured, and less stressful for learners.
The course is about a 40/60 split between theory and application (and at the end of the day, grading is weighted around 25/75 theory vs application).
While there are an enormous amount of math and statistics that undergird this course content, I felt like one could engage with the material from more of an "applied" lens and not fully master mathematics and still easily get a B. Getting an A may be more difficult without understanding the math, but is probably still obtainable with well-organized notes and code for the exams.
The group project was a great edition! Our team was able to explore some more advanced modeling techniques outside the traditional Regression, ARIMA-GARCH, and VAR models.
It is not a perfect course. The lectures are quite long and full of minutiae. Fortunately, they provide PDF docs with lecture transcriptions, and some prior students created Jupyter Notebooks that walk through the theory/code in step-by-step detail for each type of modeling application in the course. These resources were beyond helpful, and I plan on referencing them in my work.
I found the hardest part of coding sections on exams and homework had more to do with data manipulation in R than a particular time series method. If your programming skills are solid and you have some background in data manipulation using Python or R, you should be fine.
Don't be swayed away from this course just because of rough feedback in the past! For those interested in learning the fundamentals of time series and forecasting, the new version of ISYE 6402 offers solid, applied content at a reasonable pace.
On weeks with exams or final project submissions, I probably spent 15-20 hours. Most other weeks, I probably spent 5-8 hours.
For context, I took Regression Analysis in Spring 2022.
I'll start by saying that my workload is likely more than most. 50% of my time was dedicated towards the project. That project is in a research area I'm interested in, so this number represents a much higher than average workload.
The final project replacing a final exam is new, so there were some items that needed to be addressed as we went along. This likely frustrated some students, but I consider it part of the process and much prefer having an applied project vs. a final exam. I expect that future semesters will benefit from us testing this process.
I believe that students who create their own project idea (and gather/build datasets) will get much more out of this class than the ones using the proposed project datasets. Certainly that is the case for me. There is a great opportunity to use this course to build (or enhance) a research area of interest to you. This also becomes a way to engage with new modules/methods as they are shared.
I thought the homework and exams were challenging, but fair. There are going to be a few True/False and multiple choice exam questions that look nothing like the homework to test your understanding. Coding is close enough to homework and practice exams that a good understanding of past coding will get you a good result. I'd recommend good documentation of past code, especially with parameters and steps (like values for 90%/95%/99% CIs or why a loop was set to a certain end value).
I don't think I have a great aptitude for this type of analysis, but I was vested in the course because of the final project chosen. For those interested in grades, I expect to get an A in the course. More important to me, I was able to use (and talk about results) in recent internship interviews as a way to demonstrate recent experience in R as well as interest and knowledge in the subject.
There are some areas to improve:
Consistency across TAs
TA sessions occasionally yielded different answers than other sessions or the eventual answer on Piazza. Typically these were questions on process or the final project. I'd recommend that questions beyond HW/Exam/Lectures be gathered and listed on Piazza by the TA. They (and Dr. Serban) can collectively answer these process questions.
Saving TA sessions
Early in the semester a handful of video sessions were saved. These stopped and as someone who could not attend weekly, this was a challenge. I was the only attendee (for one TA) the last couple weeks, so these were happening...even if they were not well attended.
This course had interesting material, but was a mess. There would often be days between TAs responding to Piazza, cancelled office hours, poor or misleading instructions and incorrect dates on the syllabus. At times, the homework solutions were inaccurate and based on different data sets, obviously not updated since the last time the class was offered. The tests were open book and open notes, and there are 4 homework sets and a "project" which is poorly designed, easy to complete and adds little value. if you organize the lecture notes and follow the examples, it is easy to obtain a B.
First off I want to say that I enjoyed the regression course with this professor and was expecting something similar from Time Series Analysis. This course started out OK but has steadily got worse throughout the semester. This is the only course I've taken (this is my 8th) where errors in the answer keys and lecture material has made it hard to complete what's required.
As an example the answer key to the most recent coding homework - which is peer graded - was done to a lower quality than the three peer homework submissions I was grading. Graphs were so small they were unreadable and the answer key did not follow the instructions to the questions. One question asked to limit the model order to a certain size and the answer provided a model order higher than the max requested. When classmates pointed out the errors there was no response or guidance. There's been some errors in previous classes but they've always been quickly addressed and guidance given on what to do by the TA's or professor. Not in this case though...
The practice midterm also had multiple errors and made assumptions that weren't in-line with what the questions were asking. Students were asking if they misunderstood what was being asked or if there were errors but no TA responded to clarify for a couple days. This left students including myself anxious that these weren't mistake and we just misunderstood some material as were were going into writing he midterm.
Piazza is also very quiet. In most other courses there has been a lot of questions and discussion but there's usually only a few posts each week from students in the course. I'm guessing that because of the poor review not many students are taking this course this semester which limits the discussion and maybe results in less TA's assigned explaining the delays in answers to questions?
The professor has offered extra credit for students that help improve the quality which I hope can address some of the issues but based on the current state of the course I can't recommend taking it.
Disliked the peer-graded assignments. On top of the Homework workload, you need to review 3 other peers' homework which can be extensively long (plus analysis, code, and graphs it goes up to 10 pages). Peer grade is very unstable. Some people grade more generously, some don't. The final grade is the median of the three. The TA does not review the fairness of the grading. If you fail to grade other people, you get punished and deduct some points from your own homework. The grading for the midterm is harsh as well. A peer got deducted a total of 5 points at two separate places because the generated graphs did not have a clear legend.
The lecture consisted of a lot of math formulation with minimum explanation. I've taken other time series graduate courses at other universities but this is not the structure I preferred.