search

gtrebilcock / BitcoinEconometrics

0 stars 1 forks source link

Midterm Peer Review #10

Open wenjia-z opened 4 years ago

wenjia-z commented 4 years ago

This project is trying to predict the prices of bitcoin using some theoretically related features such as gold price, inflation rate, GPU, etc. After their initial analysis and regression, they have picked out some useful features, but they still face the under-fitting problem. It would be better to prediction the price movement direction instead of the price itself.

Things I like about this project:

  1. They use various data visualization methods. It gives me a clear and direct intuition about the data they use, like what is the price trend and how the features correlated with each other. In addition, the overall form of the midterm report is professional and elegant.
  2. Their project is based on their review of previous literature, which provides some rationale for their research. Also, they explain their feature meanings in detail, and it can let people without related knowledge to better understand the whole process.
  3. They analysis the data and regression results in depth. Like, they notice the negative values in the prediction and try to figure out the reason. They also try to explain the sudden drops in their data with the missing value.

Things that I think can improve in this project.

  1. In stock price analysis, time series models are widely used. Considering that Bitcoin is a tradable asset and shares some similarities with stock, maybe they could make use of the previous bitcoin prices to make new features.
  2. If the simple models do not work well, then I think they can try other more complex machine learning methods.
  3. Maybe it would be better to transform the existing features into other forms.