Open iangow opened 3 months ago
My forward-looking sentences are intended to implement exactly Muslu et al. [2015] with the help from the appendix to Muslu et al. [2015].
The time frame keyword list in Brochet, Loumioti and Serafeim [2015] are adapted in my paper.
I checked your uploaded fls_lists.ipynb file in https://github.com/yiyangw2/time_frame_gold_corpus/blob/main/fls_lists.ipynb and will correct the word list for an exact matching of the forward-looking words.
I checked your uploaded fls_lists.ipynb file in https://github.com/yiyangw2/time_frame_gold_corpus/blob/main/fls_lists.ipynb and will correct the word list for an exact matching of the forward-looking words.
You could use just the code in fls_lists.ipynb
(e.g., include it in time_frames_py
).
Professor Gow,
Thank you so much for your advice!
Yours sincerely, Yiyang
Professor Gow,
I've updated with the code you suggested for forward-looking keywords. I was wondering if you could kindly give me some advice on next step forward to improve the paper? Thanks again for your help and guidance!
Yours sincerely, Yiyang
It seems that your code does not address the issues made evident in fls_lists.ipynb
(e.g., "planed" instead of "planned").
Hi Professor Gow,
Sorry for the confusion! I updated and ran the code on Rstudio before but not on Github. Now the Github is also updated.
Yours sincerely Yiyang
You don't follow Brochet, Loumioti and Serafeim [2015] closely at all. You should document the differences between your approach and theirs. Also it's not clear why their word list is applicable to your purpose. You should explain this.
Also, it seems that you almost follow Muslu et al. [2015] exactly. The one difference I noticed is the inclusion of "'ll"
in the "first search" conducted by Muslu et al. [2015]. I think this makes sense given the different context (conference call versus SEC filings), but you should note this.
Hi Professor Gow,
Thank you for your advice!
Yours sincerely Yiyang
The reason the keyword list in Brochet et al. (2015) is chosen for the aim of this paper is that Brochet et al. (2015) talk about time horizon that senior executives emphasize in their communications and provide evidence on construct validity. However, different from their research, my research does not distinguish between short-term and long-term specifically. Besides, the reference to a future year is obviously a time frame keyword, and other changes to the original wordlist are based on manually identifying time frame phrases to attain satisfactory classification performance.
You need to explain what you're trying to measure and why the Brochet et al. (2015) approach is good for your purposes and why it makes sense to deviate from the Brochet et al. (2015) approach to the extent that you do deviate from that
Thank you so much! Professor Gow.
I will look into the issue you pointed out for me.
Yours sincerely, Yiyang
Hi Professor Gow,
I rewrote the part explaining the measure as follows with your guidance on last Friday:
The reason the keyword list in Brochet et al. (2015) is chosen for the aim of this paper is that Brochet et al. (2015) provides a starting point of a list of time horizon terms. Basically the authors “create a proxy for the time horizon that senior executives emphasize in their communications” and find that their “measure of disclosure time horizon is associated with capital market pressures and executives’ short-term monetary incentives”. However, different from their research, my research does not distinguish between short-term and long-term specifically, and therefore I combine their word lists for short time horizon and long time horizon. Starting from this combined word list, the reference to a future year is obviously a time frame keyword, and therefore added additionally to the list. The other changes to the original word list are based on manually identifying time frame phrases to attain satisfactory classification accuracy with a random sample of conference calls’ transcript (this validation step is detailed in the Appendix A2 to the paper with supporting materials on Github). This step classifies FLS into two categories – those with and without time frames.
Look at the code in
fls_lists.ipynb
. It seems your updated code does not address the differences I highlight there at all. It is still unclear whether your "forward-looking sentences" implements the Muslu et al. [2015] approach exactly or not. Is it your intention to implement this?Also why is the Brochet, Loumioti and Serafeim [2015] word-list the right one for your purposes? You should explain this.
I updated my version of the notebook to be a simpler implementation of the core code.