Feedback Proposals 2025
Giancarlo Kain
The introduction is quite focused on the empirical part but it’s missing to an extent what the overall story is. I think the story you are trying to tell is focused on the energy transition and how it will affect different carbon commodities differently. That’s not always clear.
This might not be super important but in terms of the story I was not 100% clear on what the Oil and Gas effect is compared to Coal. Is the idea that Oil and Gas will be less affected compared to Coal by any new legislation/regulation but still affected? Or is the idea that Oil and Gas will actually benefit from the new legislation/regulationb because they are going to eat Coal’s market? Maybe you do not need to hypothesise the difference but I think the take-away message will be different depending on which story has the most support.
I understand that the energy shock/Ukrain in Europe is an important motivator. It might be useful to pay some attention to your sample. Are their Russian companies in the sample (that are now cut out of some markets)? If not, can you distinguish between just a move away from Russian companies to Western companies?
Empirically, there is maybe a trend in the Carbon prices and capex. Is this what you want to pick-up in your analysis? I don’t really know the answer but it is something you will need to think about. You can also just plot the data and see whether there are obvious trends. Similarly, I would make a plot of all the pricing variables you are going to use as controls (Brent, LNG, coal price). If these are strongly correated (which I think they are), and they are also strongly correlated with the carbon price, I am not sure you will be able to distinguish any sepearate effects for the carbon price and the commodity prices. It might at least be useful to look at the models without these controls.
Minor comments
- p1: Is it standard in the literature to see carbon taxes as compliance?
- O+G firms -> Why not just say Oil and Gas?
Kaitlyn Bent
I like the general structure of the introduction and the overall story that you are presenting. The biggest missing piece for me is that you don’t make it explicit that you are interested in the quality of the lit market. I think making the difference between the lit and dark market more explicit can also help your hypothesis development. Is the aggressive/passive HFT behavior mainly in the dark market or in the lit market? If it is the dark market what are the knock-on effects on the lit market? Will non-HFT traders move to the lit/dark market and will this have a further knock-on effect? These are the type of questions that I would try to clarify in the theoretical explanations that you provide.
As specific example, when you discuss the Aquilina et al. (2024) study in section 2.3, I am not really sure whether you interpret the results as bad for market quality or good for market quality.
Like I said for the simulation assignment, I am not convinced that the interaction hypotheses (\(\beta_3\)) in the two regression models for Spread and Depth are capturing your theory. Especially given that you also expect an affect of CP on HFT. I am not sure what to recommend though.
# Lucia Li
I like the introduction and the general story you set up in the introduction. However, the literature review feels a bit too shallow and could benefit from a more indepth, critical review of the different arguments for the relation between accounting variables and disappearances.
My feeling is that you maybe have too many moving parts. One things that could help you is to connect the explations for the first hypotheses more with the potential effects of the peers. If the disappearances are the result of a general regulatory crackdown on an industry (for H1), this will probably have different effects on industry peers (H2) than when the disappearances are the result of CEO corruption (assuming that the peers are not equally corrupt). I would also advice to limit the number of peer types (competitors, suppliers, or customers) that you are looking at to keep the scope of the thesis manageable.
The H1 test should probably have more variables and ideally have some kind of regularisation as I presented in the Machine Learning lecture. It might also be worthwhile to think about not including fixed effects and whether you want to predict the exact year of the disappearance or the firm without worrying about the specific year.
Rory Rutherford
I like the idea. It’s really original. The biggest challenge will be to convince your assessors that the LLM classification is trustworthy and valuable. The paper you have submitted and discussed for the machine learning assigment is good example you can follow for all possible pitfalls of using LLMs for these kind of classification tasks. Also keep in mind that the main task in that paper is to summarise a bigger text while you require the LLM to make a decision.
I would say that is a more complex task for the LLM and you will want to make sure that your assessors are convinced that the classification actually works. You cannot just rely on the outcome tests (e.g. brown firms and green firms have different abnormal returns) because the arguments become quite circular (the classification works because you find an effect and you find an effect because you were successful in categorising different types of firms).
Writing wise, you should be careful how you present existing classification methods because you will want to use them to validate that your classification outcomes. If you present the existing frameworks as too bad, they are not valid as comparisons either.
Meiqi Wang
My general comment is about the framing of the research project. I understand that in practice corporate site visits by analysts are to help analysts and for them to gain more information on the company. However, in a lot of jurisdictions that would still legally be insider trading or some kind of disclosure misconduct by the CEO because they are releasing non-public information to a select group of investors and not too the general public.
In your literature review and hypothesis development, I believe it would be helpful to distinguish between two different effects: (1) the effect of the visits and (2) the effect of disclosure of the visits. It was not always clear to me whether you were discussing the first of the second effect. When you hypothesize the frequency of the visits, I think that refers to (1). The disclosure would as a first approximation affect the general market which might influence the behavior of the insiders. However, it’s not clear to me how you expect the market to react to the news that there are more informed investors in the market (because of the visits).
I know you have changed the empirical approach. I just want to highlight that the design in the proposal uses the log + 1 and the abnormal variable calculation in a two step regression process which both can lead to invalid results. Just be aware of those when you fine tune your research design.