REM1
Coursework Feedback
Professor Thies Lindenthal
htl24@cam.ac.uk
March 2024
## General observations <span class="subtitle"><br>My perspective</span> * It was a (relatively) pleasant experience to read your work! In general... - ... one could see that time, effort and heartblood went into the reports. - ... the quality of the reports was high (OK, there is a distribution) - ... it was evident that you have learnt a thing or two in the first residential. - ... you made it your own, included prior knowledge, techniques, experience, or data. * There was so much variation! Not N times the same set of steps, repeated again and again. And that is excellent! * However, how to mark all of this, in absence of a single master solution?
## Mark distribution <span class="subtitle"><br>This is a typical Cambridge distribution... before grade inflation.</span> * Median: 72.5; range: 65–78 - 1st—4, 2:1—8, 2:2—6, fail—0 <br> <img src="imgs/hist.png" height=450>
## Literature work <span class="subtitle"><br/>Reading is research!</span> * Most were avid readers. Great! - Many comments on "effective/efficient use of literature". This saves time and space! - Is it OK if 90% of "my" findings have been shown by others before? YES! The last few % are the contribution! * Literature everywhere, not just in literature review - Link results, wherever possible, back to the literature ('I find X which is/is not in line with Y's finding...')
## Empirics <span class="subtitle"><br/>Collecting quantitative evidence is hard</span> * Identification is difficult (we try to teach you a bit more) * It's a dilemma: - Reliable identifcation is rare case. Limited to few, often not that interesting minor aspects. - Great, big topics are usually not identifiable. "Effect of work from home" Good luck * Acknowledge challenge, discuss a perfect case. Make concessions - and justify them. * Tackle question from several imperfect angles and aggregate evidence? * Economic story? * Empirically feasible? * Real estate data literacy * <mark>All of this is difficult and needs time, skills and experience! Yes, we were lenient.</mark>
## Writing <span class="subtitle"><br/></span> * C-C-C: Context - Content - Conclusions * Emulate economic papers, not industry reports - Bullet - points - are - ugly. * Flow of argument is all-important. Bad story-telling kills the most beautiful regression. * Focus on tables and charts: 20/80 rule - They receive most of the attention. * Proofreading is for wimps. Not.
## That mix of theory and empirics... <span class="subtitle"><br>We want that balance... and it saved someone here...</span> * Example - "The two reports are like day and night. The first is strong, well-reasoned, good critique of the ML text. Critical reading, good use of literature. <br><br>The second is really confused. The data are not introduced well. Why did you select that market? What is special about it? What data do you use? How are they sourced? Are they representative? Be really critical here. The MPT implementation is broken. No numerical optimisation, no efficient frontiers. Portfolio variance formula in spreadsheet does not make much sense, should include CRE, not based on theory.<br><br> Good bibliography, overall. Some references need to be formatted differently (No Google searches!)"
## Interesting data choices <span class="subtitle"><br>Thanks for something new!</span> * Example - "Discussion of LLM text is structured well. Good use of literature, overall solid. The second report has problems, though. First of all, the choice of market is interesting. US or UK would have been the safe and boring option. The property index needs a critical discussion. Just from looking at the graphs, we can see that the returns must be radically smoothed. FOAC must be close to 1! (Minor point: The figures are copied from the land departments website. That is OK, but give a proper reference.) Those data need some desmoothing, despite being a transaction based index. And even then, they might never be informative enough to be used in a MPT setup. Offer descriptive statistics etc in the report. The spreadsheet is only meant to show the data series, main anaylsis needs to be shown in report too. -97% cannot be the average return for RE... it is -0.97%, check the spreadsheet again. "
## <span class="subtitle"><br></span> * Example - "Exhibit 4, minor comment: The efficient frontier is plotted only paritally. Should end in the blue dot indicating bonds. Exhibit 5 is a prime example of a corner solution. Slight changes in the parameters/subperiod definitions etc change the outcome disproportiopnately. Sensitivity analysis? How are expectations for these short time-periods formed? Only few observations available."