Intelligence Authentic + Cryptocurrencies

Deep Learning To Predict Ethereum Prices - Beating the Benchmark Block by Block

 

Challenge

Ethereum prices have fluctuated wildly in the past year, with daily double digit percent gains and losses the norm in the current state of affairs. Intelligence Authentic LLC endeavors to understand the dynamics of Ether’s price movements at the narrowest levels, starting with prices at each batch of trade executions on the Ether blockchain, with an analysis provided for the June 15 2018 to July 15 2018 timeframe.

Solution

In effect, our problem is to determine, block by block, how Ethereum’s price movements iterate. Data was collected from etherchain.org, utilizing block information to build features sets for prediction. Models were built via Keras and Tensorflow, data preparation in Pandas and Scikit-Learn, utilizing variable transformations of momentum, gas prices, transaction counters, and prior log returns.

Results

After correcting for transaction costs and changing our execution model to be more conservative, our algorithm beat the benchmark (buy at the beginning of the period, sell at the end) by over 60%, and achieved out of sample F1-scores of 60-65% depending on the iteration of the model used.

Works conclude that there is predictability in the markets to be discovered by machine learning, especially deep learning. Feel free to contact us at info@intelauthentic.com with any comments, thoughts, or suggestions.

After correcting for transaction costs and changing our execution model to be more conservative, our algorithm beat the benchmark (buy at the beginning of the period, sell at the end) by over 60%, and achieved out of sample F1-scores of 60-65% depending on the iteration of the model used.

Works conclude that there is predictability in the markets to be discovered by machine learning, especially deep learning. Feel free to contact us at info@intelauthentic.com with any comments, thoughts, or suggestions.

Correlations.png