What I found to be really valuable and potentially useful were the examples/case histories of how the various machine learning techniques to portfolio management. For me, the most valuable learnings were, regularized regression to compute factor loadings, application of PCA/Clustering and Graphical Approaches to maximize portfolio diversity, and scenario/regime based portfolio models. I fully intend to do some follow-up work in applying those techniques to my personal investment management. So while perhaps not as learner-friendly as the previous two courses, I think the subject matter will prove to be far more valuable if one invests the time after the course.

I think if you want a better understanding of the many machine learning techniques, you might be better served to take a course specifically focused on that. I found the treatment of these techniques, insufficient to gain a solid conceptual understanding of the techniques. With that in mind, the course might be improved be spending even less time on introducing some of the basic machine learning methods / and traditional models, that are well covered elsewhere, and more time on the case histories, and application of the methods to portfolio management and investing.

The subjects addressed in the course, such as models to identify crash regimes, are interesting and important. It points out important implementation issues in Machine Learning like regularization, k-fold validation to choose hyperparameters, and introduces multiple ML algorithms and methods (OLS regression, Logistic regression, Decision trees, Boosting, Graphical analysis functions).

Unhappily the explanations are convoluted and the Python Notebooks only cursorily explained.

Gave the course 3 stars because the Notebooks are 5-star.

I have been more and more frustrated with the course that became less and less explanatory, but more and more descriptive. I still find the topics very interesting, and the first two MOOCs were really amazing, but I find this one much less clear and giving us much less understanding of the coding part. What would be really great would be to get a full description of what the code does, at least much more detailed than at present. As an example, no code was even provided for PCA and graphical networks, that's quite disappointing.

I think the ideas related to this course are interesting, and in concept it's a great follow-on to the previous two. Unfortunately, I don't believe anyone who doesn't already know concepts and techniques of machine learning will come away from this course with any understanding whatsoever of what they are. I am a mentor for a Coursera specialization in Deep Learning, and I found the description of supervised and unsupervised learning here to be unintelligible. I'll be working through the lab code on my own to learn how to use it for portfolio construction, since I didn't bring that away from completing the course. Really a disappointment after the first two courses in the specialization, which I now question whether I will bother to complete.

Lectures were very confusing, poorly explained, poorly structured. Quizzes and tests were extremely bad: questions with often no link to the explained theory, many questions where you could seriously debate about the right answer.

The content was mostly limited to a few cases: a lot about factor shrinking, or better estimations for factor loadings, and a second case of recession prediction. Except for these two, most of the content was vague theoretical examples.

Labs were mixed quality: code was sometimes very clean, sometimes less. The tutors of the labs were sometimes giving a really clear rehearsal or overview of the topics (far better than the gentlemen who give the lectures), but sometimes they just read the text without any contribution.

Overall this MOOC was very disappointing. I have not learned a lot, and most of what I learned I had to google and research myself because the MOOC was too confusing.

This course was horrible, with poor performance on econometric concepts, as well as poor interpretation of ML and inconsistent use of Python.

the coding material shared is very under explained. the lectures are subpar. the audio issues are very bad.

Super interesting, very well explained, with lots of useful resources (links to various papers and textbooks), and, best of all, with very practical, well-annotated notebooks applying the theory covered in the video lessons.

The topics covered in this course are really interesting. I learned a great deal by studying various papers covered in this course - Thank you to both instructors!

Very nice course sharing many types of knowledges around data / cleaning / type of data / several algorithms / organised Python coding

A great course with a Ph Doctoral taste, including amazing and advanced Jupyter Notebooks !!!!

Excellent course¡

Thank you, Princeton crew, for this course. I learned a lot, so Thank you. However, learning was not organized. This course is third in the series but has the lowest rating for a reason. I will summarize the pros and cons and provide a roadmap to improve it.

Pros:

a) The academic referenced material is rigorous and requires familiarity with both investments and machine learning topics. This course is certainly not for a beginner. However, having gone through two courses before, one should be reasonably prepared.

b) This course surely provided ML code that can be expanded to conduct further research. As others have said, this course offers building blocks for ML in the asset management area. However, it does not deliver a finished (or semi-finished product).

c) It provides ML code, which most learners cannot develop on their own.

Cons:

a) Too many cooks in the kitchen. Two instructors and five PhD students is a lot to create confusion among labs, videos, and quizzes. There is inconsistency in each week and across weeks.

b) After going through the first two courses (with just two instructors), the bar is set high. Unfortunately, the bar could not be met.

c) Quizzes are horrible. They are vague and unrelated to labs. In the first two courses, lab content was tested heavily. Here concepts are tested. Quizzes need to be rebuilt.

Improvements:

1) Have only one person present (possibly develop) all the labs (like Vijay did in the first two). Then consistency will be maintained.

2) Have quizzes based on labs (not theory). Or make it 80% lab, 20% theory. Currently, quizzes are 80% theory.

3) Rebuild the weekly quizzes from scratch.

A good course overall but there are significant drawbacks: test questions are sometimes intimidating and overly on theory while Python code is barely covered in the Lab sessions. The papers and materials provided can be of great use for people ready to dive a bit further. Still I think this course lacks a pair of short videos that will cover Python code in detail for learners without strong background in ML and coding. Nethertheless, I don't want to give a poor mark to the course.

The course has seen some improvements since its inception. The subject is still very, very interesting and there is enough materials, code especially for one to explore further on his own. Having a prior knowledge of data science is also probably necessary. Definitely, the tests should be reworked, as they are not adding much value to the understanding of the course.

Overall, it is still a valuable course.

The overall topic of this course is great and very current.

I think the lab sessions can be improved. The Python supporting material is not always available and/or topics are covered at a very high level in the lab sessions.

Given the complexity of some of the sections, it may be worth considering extending this course (from 5 weeks to 7-8 weeks?) so that topics can be covered more in depth.

This is not a watered-down course, and surely is not for beginners as it quickly covers insights instead of details. A very typical course style from top universities. Students need proactively spend time reading and learning materials.

Very good theoretical discussion and practice The practice part is not given as much importance as it possibly deserves and some of the graded questions are a little ambiguous and not very conductive to learning.

would be good to focus more on the jupyter notebooks and less on multiple choice. Really interesting notebooks and quite advanced / technical material which deserves more time and coverage.

Good concepts to touch but lack on coding in granulality example. But overall, I'm get a good example how to implement machine learning technique to finance perspective.

Good overview on Machine Learning techniques, need for some basic knowledge in statistics and Python for an optimized experience.

I would suggest to add the link to the references like pdf docs.

I was thinking very carefully to rate this course and for that I like to refer it first to the article of Claude Shannon (the father of information theory and who worked with Edward Thorp, the first modern mathematician to use quantitative strategies for investments)"A Mathematical Theory of Communication" published in Bell System Technical Journal in 1948 where we are going to take basic elements of communication as we can see in the diagram https://en.wikipedia.org/wiki/File:Shannon_communication_system.svg.

An information source that produces a message

A transmitter that operates on the message to create a signal which can be sent through a channel

A channel, which is the medium over which the signal, carrying the information that composes the message, is sent.

A receiver, which transforms the signal back into the message intended for delivery

A destination, which can be a person or a machine, for whom or which the message is intended

A noise source that can perturbate and corrupt the message.

And second, to the DIKW hierarchy, wisdom hierarchy, knowledge hierarchy, information hierarchy, and the data pyramid as https://en.wikipedia.org/wiki/File:DIKW_Pyramid.svg.

As a conclusion although the information source was a high level, interesting and important, because of the fact that the communication channel was not efficient enough to transmit it, the destination did not receive that information correctly. Related to the flow diagram https://en.wikipedia.org/wiki/File:DIKW_(1).png we can barely knew "how" but we did not went deeply into"why" and further.

On the other hand , the other 3 courses of this specialization arrived to answer "why" and "what is best" questions.

I hope a full revision of this course be performed, in particular: its methodology, the way lessons were taught, the replacement of the non professional LAB's lecturers, the duration of the lessons, etc.

The main 2 problems with the course are: * The lectures by Prof. Mulvey lack the clarity that we got accustomed to from the first 2 courses from Prof. Martellini. For someone without much background in ML they won't make much sense, for someone with - won't provide added value. * The quizzes are poorly designed, mostly testing concepts mentioned during the lectures and very little related to running the labs. Some correct answers are just wrong (e.g. last question of last lab doesn't match the actual results of the lab about best performing method). The labs themselves are quite good and well explained (with the exception of lab1 and lab5). If you get stuck, listen to the labs first to make more sense of the lectures. Wanted to give the course 4 stars but the quality of week 5 (including the lab) made me reconsider the grade.

I must say its been a long journey since first MOOC in this specialization. I had great learning and someone having no past programming background has acquired a lot in this specialization. Fortunately, the first two MOOCs were really well connected since Dr. Vijay Vaidyanathan has explained things so well that at least I could understand the concept as well as the implementation in the real data.. I was really excited for this MOOC but instead of focusing more on the practical part things were taken fast and solely in theory. I wouldn't say it was bad but the lab session could have been more engaging and explanatory like the first two MOOCs since it would have been helpful for non-programming background finance professionals.