Feedback Proposals 2026
Alex
In section 3.1, is the change in the unit cost in equilibrium, or should I interpret it as a shock to the unit cost because I think it’s an equilibrium?
The Cobb‑Douglass section is interesting, but you could shorten it by focusing less on the technical explanation and equations, and more on the conclusion. The conclusion is quite striking: the impact depends on the elasticity of competition in the output market, not on the technology or the firm’s production function. This is probably not something I would expect, nor is it reflected in your introduction.
I had quite some difficulty following the Acemoglu model, or at least the explanation. I think the shift you identify—from a demand‑side condition to a technology‑side condition, as mentioned in the paper—makes it more relevant. However, it is not clear from the explanation and the equations where this shift originates. It would be useful if you could summarise in one paragraph the key difference in the setup and how it leads to the conclusion that we now have a technology‑side condition. I think that makes sense, but I don’t understand where it comes from in the current write‑up. I think a written up version of Table 1 would accomplish a lot of what I am after.
Just before the hypotheses, I think it might be useful to explain the difference between total factor productivity and labour productivity to really let me and the reader understand what the difference is and why TFP is a better measure.
This is not entirely a fair comment, but one thing your hypotheses and theory ignore is that artificial intelligence—especially the current implementation of chatbots—can also decrease productivity. If employees spend too much time on chatbots, if customers use chatbots to submit complaints, or if hoaxes perpetrated with AI spread more widely, productivity suffers. Coding agents make it easier to hack into databases, requiring more cybersecurity staff. That could be a net productivity‑decreasing factor and might outweigh some of the mechanisms you have highlighted. This may also be reflected in the measure of adoption you use, because in 10K filings you are more likely to report risks, which could be potential factors that have a negative impact on productivity.
Fixed effects do indeed absorb common microeconomic shocks; one of those is the introduction of ChatPT. In effect, what you’re testing for is differences in productivity gains between reporters and non‑reporters and not necessarily the effect of AI itself. This is not necessarily a problem with what you’re doing, but it might inform how you interpret your findings.
Travis
As a non‑specialist, I don’t necessarily understand what “convenience yields” means. So if you use that in the abstract, it’s hard for me to figure out what the contribution is. Similarly CPI and PPI are not defined (I do know what these mean).
I noticed in the introduction that you used a lot of words like “could” or “may”, and you should try to avoid those words and be more specific about what you mean. Instead of saying “convenience yields may serve as a unique proxy for unexpected inflationary pressures”, say something like “under these conditions, convenience yields are unique proxies for unexpected inflationary pressures”. So specify when the thing that you think may be true, when it will be true and when it will not be true.
One thing that I’m missing is the exact question you are trying to answer. I don’t fully understand why you are focusing on China besides its importance. There are hints in the proposal that government interventions might limit some impacts from export prices or that there might be attempts to control the effect of import prices on inflation. However, that is never incorporated into specific predictions or the design.
It could be that you separate commodities that are critically important for China on some metric from those that are not, and then examine whether there is a difference in how those prices and convenience yields affect inflation. Are you try to look at differences between PPI and CPI impacts. That might be relevant in China. You are also not comparing China to other countries directly, making it harder to draw conclusions about whether China is different. So I am struggling with the underlying question you are asking, apart from the fact that convenience yields might be more predictive, which has already been shown in other countries. What are you adding by investigating this question for China specifically?
When I look at the formula for the convenience yields, the way I understand it is that you’re basically incorporating future prices to calculate the convenience yields. So, in a way, you are using forward‑looking information in addition to the spot price in your prediction of inflation. It should not be unexpected that futures prices are predictive over and above spot prices. I wonder whether you’ve thought about that, because once you break down convenience yields, most of the information that is not the spot price comes from future prices. Maybe that is okay, but I think it is worthwhile considering whether you really want to explain this based on inventory holding costs, or acknowledge that other factors may also impact futures prices, which will obviously be related to future inflation.
Ivor
I was most confused by your discussion of information asymmetry throughout the introduction and literature review. I think it would be useful to clarify what you mean by information asymmetry, specifically who has more information than whom and who is communicating in these conference calls and who the audience is.
In your setting there are at least three parties: the acquirer, the target, and the market. The market can play a dual role because it may be investing in the acquirer and in the target. I understand that the conference calls are a way for the acquirer to communicate with the market. However, when you discuss the bargaining process, the information asymmetry flows between the acquirer and the target. The means of communication between the acquirer and the target do not necessarily have to be conference calls. In fact, I am not sure why the acquirer would communicate with the target via conference calls unless that is a specific part of your argument. You probably need to clarify why acquirers would communicate directly with targets through conference calls if that is part of the argument.
If I understand the measurement of the premium correctly, the premium is basically calculated at the time of the conference call. Any effects that the information flow might have will happen after the conference call, but that will not be captured in your measure of the premium. To me, that suggests that you need a theory saying that acquirers have certain private information, which is why they make a bid and why the premium is larger or smaller. At the same time, it is important for them to make that information public, which is why they hold a conference call. So the conference call and the premium are related decisions, both driven by the private information the acquirer has about the target and potential synergies.
Luke
My main point of feedback is that I think you should consider the composition of synergies for each acquisition. You distinguish between operational synergies and corporate synergies, which is fine. You also note that acquisitions within a certain geographical boundary are more likely to have operational synergies, and I am fine with that. However, the prediction you are making—that acquisitions with operational synergies will have higher value—assumes that, on average, corporate synergies are the same across acquisitions. If you design your analysis by categorising acquisitions as having operational synergies or not, you are assuming that the two groups have the same amount of corporate synergies, which may not be true. Some acquisitions may be pursued purely for operational synergies, with favourable corporate synergies, while others with no operational synergies might require sufficiently large corporate synergies to make the acquisition worthwhile. This is the main assumption you need to consider and address. If it might not hold, you could improve your model by introducing control variables or making comparisons that better capture the effect you intend to test with your hypothesis.
As a minor aside, I think the Baker and Gelbach approach is a good idea for your setting. I also think there is a more recent approach that might work and is probably more flexible. For instance, you will not have to select peer firms; you could use principal component analysis to reduce the number of factors to consider instead of directly looking for It would require learning a new R package, which is feasible but perhaps not worthwhile. I have put the link to the paper [here](https://arxiv.org/abs/2511.15123). You can always ask me for more information if you want help with it.
Jiayi
You are writing the introduction from a perspective that contrasts rational decision‑making and biased decision‑making. I don’t think that is the most convincing framing of your story. Towards the end, you discuss the slippery‑slope theory, and I think your overall hypotheses and theory are much better supported by that theory. If you explain the theory at the start, then derive your hypotheses and discuss the literature in light of the slippery‑slope theory, you will have a stronger story.
Your story is essentially about some CEOs taking more risk, perhaps because they are overconfident in their reporting, and then having to cover up when their overconfidence is wrong, leading to further earnings‑management‑type reporting decisions. This fits much more into the slippery‑slope argument.
If you focus too much on rational versus non‑rational decision‑making, you will need to prove that the overconfidence is actually wrong—showing that overconfident CEOs have recognised too many reserves, reported reserves larger than they probably were, and then had to correct them. You would have to demonstrate that they were wrong to label it a bias. The slippery‑slope argument allows you to say that CEOs might be overconfident, which is not a problem per se, but it may make them willing to ignore their mistake, leading to more reporting issues. This is a much stronger story, easier to test, and avoids getting bogged down in the weeds about what is rational or a bias.
For the empirical design, it will be really important to have multiple measures of overconfidence that are strongly correlated with each other. You plan on doing that, which is good, but without it, it will be hard to attribute any findings to overconfidence. One thing that would be interesting to look at is how many CEOs you have per firm in your data set. If you have multiple CEOs for a specific firm, that would allow you to control for a range of firm‑specific factors, creating variation within firms. This could be more convincing if you have a sufficient number of such firms in your sample. For instance, one thing that you could then do is calculate an average overconfidence score for each CEO in the firm, not a year‑by‑year measure but an average per CEO.
Meliani
My main feedback is on the empirical design. One thing to keep in mind when you’re writing is that you have a fairly new design. It would be useful to think a little more about how you will write up the design and explain things. For instance, you’re using the Schwab model, but you don’t really explain the motivation behind that model, why it is appropriate for your specific case, and what the assumptions are. How is it explained in the original paper? Even in your discussion, when you introduce it as the measure of ability, you’re basically saying, “I’m going to use this model, but I’ll add an extra step.” At this point, the reader doesn’t even know what the Schwab model actually does and does not understand what that extra step would be.
Then, if I go to this two‑step model, you basically use the data envelope methods to come up with your tax‑effectiveness measure. It’s not exactly clear what that measure is. I know what you’re doing, but not what measure you’ll get. I have some idea because I know roughly what the method does, so you need to make that clear to the reader. In the second step, you’ll residualise the effectiveness measure you have and calculate the residual.
I have two comments on that two‑step approach. First, it means you use some variables as inputs and other variables in the regression to residualise the measure from the first step. It is not always clear why some measures are used as inputs in the DEA method and others are used as regression controls. Sometimes those measures would be better as inputs, and other measures would be better as regression controls. I may not fully understand the purpose of each step, but it would be useful to think more about that and provide more detail for the reader to evaluate whether what you’re doing makes sense.
The second point, and this is more an opinion, is that it might actually make sense to do the steps in the opposite order. It would make more sense to me to first residualise all the inputs and the outcome measure you’re using in the DEA process, and then calculate the effectiveness measure from that. However, to be sure, I would need more details on the purpose of this DEA step.