Advice for seminar papers and theses
In this blog post, I gather general advice on how to write theses and seminar papers. References can be found below. The recommendations here are not necessarily well-structured but are more like a list of things that I observe(d) through time and want to share with future students of mine.
A thesis is not a novel. Hence, state very early what you are doing. People want to know in which direction you are heading. This is also helps to understand your priorities in the literature section. General statements like “Volatility investing is important to investors and market participants.” are kind of useless as your reader already knows this. My colleague Andreas Pick has a nice discussion about writing introductions: https://apick.eu/teaching/2022/05/08/Academic_writing.html.
Don’t structure your results section w.r.t. models but w.r.t. evaluation criteria.
- Hence, the section of your results should not be structured like Subsection 1: Model 1, Subsection 2: Model 2, and so on.
- Instead, structure your results by evaluation criteria; for example, Subsection 1: Forecast performance, Subsection 2: Economic significance, and so on. Discuss all models jointly w.r.t. these different criteria. For this it also helps to make a smaller number of possibly large tables. Jumping from one table to the other is tiring for the reader.
For a thesis, it is best if you have at least one own extension; be it theoretical or an alteration of a model that is out there but needs a twist to be applied in your setting. A horse race of methods that are already out there will not bring you an excellent grade, no matter how well-executed it is.
Econometricians like statistical tests. A lot of my supervision project touch the current forecasting literature and quantitive finance. Hence, a small number of references.
- Diebold-Mariano test: Diebold and Mariano (1995)
- Model confidence set: Hansen et al. (2011). If you use R for you computations, please use the implementation in the rugarch package instead of the implementation in the MCS package. The model confidence set can be though of being an extension of the Diebold-Mariano test for evaluating multiple models at once. In this sense, I would prefer it in most situations because you don’t need to choose a benchmark and you don’t run too deep into the multiple testing problem.
- Sharpe ratio test: Ledoit and Wolf (2008).
Econometricians don’t like inconsistent notation. Triple-check everything in this regard including all formulas etc. This holds especially if you merge notation from different papers.
There are general guidelines about punctuation and equations; see, for example, https://www.chicagomanualofstyle.org/book/ed17/part2/ch12/psec018.htmls/ and the following sections. You can only access this resource from the university network.
Minor details but still important:
- In English, there are hyphens and two types of dashes for different use cases https://www.merriam-webster.com/words-at-play/em-dash-en-dash-how-to-use * Footnotes go after punctuation https://proofed.com/writing-tips/should-footnote-markers-go-after-the-punctuation/.
- Citations are written as Author (Year) or (Author, Year) but never (Author (Year)).
- In general, no abbreviations in title or abstract. If you need them in the abstract, always define them in the abstract.
How to make nice figures and tables
Tables and figures should be self-containing. If you read this blog post, I am probably your main supervisor. Hence, I do a shameless selfplug and recommend Kleen and Tetereva (2023) as the best inspiration for your designs. In general, I think it is better to separate the caption from the more detailed notes below the figures and tables. The notes shouldn’t include any interpretation but enough info to understand what is presented.
It is important to structure your day. One example could be to reserve 2 hours for writing in the morning and start coding after that.
Starting into the day with a plan makes it easier to get going in the morning. Hence, I recommend to use the last 15 minutes of your working day to plan your next working day. This can be done through making appointments with yourself in your calendar app.
- For joint seminar papers, it might be best to write in Overleaf. Some institutions like the EUR have a subscription for their students.
- Writing in Overleaf is nice but I think texifier is by far the best LaTeX editor when writing on your own. Depends on your budget of course as you need to pay for it.
- Think about using ToDo apps like Microsoft’s To Do or Things. The latter is paid but has a nicer user interface but your mileage may vary.
Coding and computational facilities
It is not so easy to write code that runs fast. However, often only a handful of sections in your code cause most of the delay; for example, due to copying large amounts of data back and forth. Hence, it is good to start using profiling early; for R, profvis is the best option in my opinion.
A great reference for nice coding in financial econometrics is tidyfinance.
Students of mine had great experience with Google Colab when making use of computational resources beyond their own computers.
- Especially, when working together for seminar papers, Google Colab makes it way easier to work on code together.
- There are two upsides to using something like Google Colab when writing code together: First, no sync issues and all teammates have the same computational backbone. Once, I had a seminar group with two students having very potent computers but the other two had not. As a consequence, one group was always wondering why two were complaining about performance issues. Sometimes the code wouldn’t even run due to different amounts of RAM.
- There is a reasonably potent free tier but also the paid plans are relatively cheap.
Diebold, F. X., & Mariano, R. S. (1995). Comparing predictive accuracy. Journal of Business and Economic Statistics, 13(3), 253–263. https://doi.org/10.1198
Hansen, P. R., Lunde, A., & Nason, J. M. (2011). The model confidence set. Econometrica, 79(2), 453–497. https://doi.org/10.3982/ECTA5771
Kleen, O., & Tetereva, A. (2023). A Forest Full of Risk Forecasts for Managing Volatility. Available at SSRN: http://doi.org/10.2139/ssrn.4161957
Ledoit, O., & Wolf, M. (2008). Robust performance hypothesis testing with the Sharpe ratio. Journal of Empirical Finance, 15(5), 850–859. https://doi.org/10.1016/j.jempfin.2008.03.002