Designing a Small Applied Financial Time-Series Project
Summary
The document offers guidance for choosing a short applied mathematics project involving financial data and time-series analysis. It recommends aligning the topic with the supervisor’s research interests and adapting an existing empirical paper into a smaller replication, since a limited project scope makes a broad original study difficult.
Its example examines whether intraday returns improve portfolio allocation compared with daily data. The proposed workflow forecasts covariance from high-frequency realized measures and uses those estimates in a global minimum variance portfolio; the cited study reportedly found lower portfolio volatility from realized covariance estimates. Suggested simplifications include using fewer assets or models, replacing a realized kernel with realized variance, shifting to daily volatility models or cryptocurrencies, or evaluating VaR forecasts. These are topic suggestions, not evidence independently assessed in the document; data access, implementation effort, and the short project length constrain what can be demonstrated.
Key ideas
- Choose a project topic that overlaps with the supervisor’s expertise.
- A small replication of an empirical paper can provide a manageable applied research design.
- The example compares portfolio allocations based on intraday covariance estimates with daily data approaches.
- Realized variance can simplify a high-frequency covariance modeling project.
- Related alternatives include daily volatility forecasting and parametric VaR evaluation.
Tags
Full text
# Topics for a 5 ECTS point project in a Msc Applied Math prorgram # Topics for a 5 ECTS point project in a Msc Applied Math prorgram I need 5 ECTS to complete my education (Msc in applied math, finance track). A statistics professor have agreed to do a project with me but I need to come up with the topic myself. Can you reccomend any topics for this short project? I want the focus to be more on application and not theory. I want the project to include some kind of financial data and ideally involving some kind of time series analysis/modelling, but I do not have many restrictions/requirements. Please suggest and elaborate topics :) ## Answer by Pleb (score 1) https://quant.stackexchange.com/a/63689 Before finding a topic, there are some things that you should consider: - While you are allowed to find your own topic, I would suggest you keep it within the research area of your supervisor. In that way, both you and your professor will be engaged in the project and the professor will likely help you by leveraging his own knowledge from his research interests. - It might be better to try and replicate the empirical study of an interesting paper, albeit at a much smaller scale (eg. smaller asset-dimension, models etc). In the end, you only have 5 ECTS to work with (which amounts to approximately 10 - 15 pages, depending on the university), and thus it can be hard to construct an in-depth study of anything on so few pages. #### An example of a topic: I've provided you with an example of a topic that uses financial time-series. Moreover, the topic can be modified (to some degree) in order to fit your personal interests. - [High frequency volatility modeling & Portfolio management] With high frequency stock data becoming more readily available, a question arises whether we can leverage the intraday stock returns to acquire better portfolio allocations as opposed to daily returns. In Hautsch et al. (2015) they try to answer this question by obtaining high frequency covariance matrix predictions and insert them in the Global Minimum Variance Portfolio (GMVP). Using different smoothing models on realized measures (eg. realized variance) they conclude that realized covariance estimates substantially reduces the portfolios overall volatility. - [Comments] This is indeed a study that is replicable, if you include less assets/models and replace the realized kernel model with a simpler counterpart, eg. realized variance. However, you should skim the paper, in order to get a feel for the topic. You can get high frequency stock data from AlphaVantage. - [General modifications] The study could also be changed from high frequency models to daily volatility models (GARCH etc.) or to completely different assets, such as cryptocurrencies. Furthermore, the end result could be changed to forecasting Value at Risk (VaR) using a parametric VaR framework and an evaluation scheme. The book by McNeil et al. (2015) provides some good information on quantitative risk management techniques. Other than that, the book of Ruey Tsay (2010) might contain further relevant topics of financial time-series than the example topic provided above.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.