7

I am looking for a thesis topic for my Applied Economics degree where I can use Data Mining/Machine Learning for some kind of prediction.

Do you guys have any good ideas about what I could do?

My first thought was credit scoring, but that seems a little boring considering the number of papers out there and the lack of available data.

I don't want to write about stock market forecasting because that's what I did for my mathematician thesis.

A good topic would be predicting some sort of an economical indicator but I don't really have any good ideas so far.

Thank you!

Gyula
  • 211
  • 1
  • 3
  • 5
  • 1
    Have you tried asking a professor in your department for ideas? –  Sep 23 '13 at 16:47
  • The economics department is not very big on data mining, I asked them but they couldnt give me any good ideas – Gyula Sep 23 '13 at 18:02
  • 2
    This question appears to be off-topic because it is about getting a thesis topic rather than a statistical question itself – Andy Jun 03 '14 at 16:05

3 Answers3

5

Here's a nice survey by Einav and Levin of what's been done and some directions for further research (Section 5). Here is an application to Netflix data and price discrimination where ML techniques do the heavy lifting in feature selection and demand estimation. Here's a paper on Yelp review fraud, where Yelp's fraud detection algorithm provided identification. This is a more gingerly use of these methods.

Here's another nice survey by Hal Varian (chief economist at Google).

dimitriy
  • 31,081
  • 5
  • 63
  • 138
2

There's a recent article that uses Google Trends to predict financial crises or other economic events: http://www.nature.com/srep/2013/130425/srep01684/full/srep01684.html

A similar approach can use Twitter (as in here), to predict other economic events or variables (e.g. marginal utilities, prices, etc).

Diego
  • 156
  • 3
1

Hidden Markov models (HMMs) in the form of "regime switching models" have been used to determine growth/decline periods (I know, they're not technically HMMs, but they basically are). Does that count? Read up on the work of James Hamilton if you're interested: http://weber.ucsd.edu/~jhamilto/

David Marx
  • 6,647
  • 1
  • 25
  • 43