Financial markets are often modelled as regression problems. The most unpredictable movements were those of cryptocurrencies. Can you predict their future values?
This dataset contains data for 4 cryptocurrencies (Bitcoin, Ethereum, Ripple, Bitcoin-Cash) over a 5 year period 2013-2018. Provided data is all in $USD and includes open/close, high/low, volume traded, spread and close ratio. Close ratio is defined as:
Close Ratio = (Close - Low)/(High - Low)
The dataset is provided in a simple .csv file. It is recommended to use the Python packages numpy and pandas to read and manipulate the data. Choice of programming language is not restricted, however Imperial Strategics & Data Science Society projects/workshops are in Python.
This is not an exhaustive list of tasks, the points are provided in order to guide you:
Limited features are provided in the dataset and so it may be useful to derive your own. Explain the thought process behind any decisions you make here.
Select the most predictive features in the model. Which are useful and which aren't?
Select an appropriate regression model and tune it to improve its performance.
Report your results using appropriate metrics and explain key decisions in your process. Suggest possible improvements.