2019.01.07

The application of machine learning on Cryptocurrency Market

Pocket


This article is divided into three sections:
At first, let’s see the current situation in cryptocurrency Market.
Then, we will introduce the main content from the paper (reference 1)
Finally, we will also summarize the recent hot topics in Cryptocurrency area.




Top 5 Cryptocurrencies




Top 5 cryptocurrencies in Coin Market Capital by 2019/01/07





The total capital is 137M+ dollar by 2019/01/07. By contrast, the peak value was 800M+ dollar in 2018/01. The whole market shrinks significantly at market value.




Machine learning the Cryptocurrency Market
However, cryptocurrency still attracts much interest from many areas including academic (reference 1). Next, I will briefly introduce one recent paper published by Laura Alessandretti etl in 2018/05, and the objective of this paper is evaluating the performance of machine learning models on predicting daily cryptocurrency price for 1,681 currencies.

-Dataset Description
Data source: Coin Market Capital
Time span from 2015/11/11 to 2018/04/24
Trading volume > 100,000$ at any given day
The dataset is split into train and test according to the time.




この画像には alt 属性が指定されておらず、ファイル名は image-2.png です
Data split





Each coin is represented in the following features: price, market capitalization, market share, rank, trading volume within 24 hours, lifetime of the currency in days.
Two algorithms are evaluated in this paper.Random forest is one commonly utilized algorithm today. While LSTM is one well known deep learning algorithm which is mainly used to deal with temporal problem. Actually, we could find a lot of papers or articles trying to exploit LSTM in stock market (reference 2).




この画像には alt 属性が指定されておらず、ファイル名は image-8.png です




Evaluations
-Baseline: simple moving average (SMA)
-Metrics ( Geometric mean return and Sharp ratio)
We could found that Method 3 (LSTM) is the best option pointed in this paper.




Results





The other hot topics
The application of bitcoin ETF is ongoing. Though such kind of applications have been rejected several times by SEC in USA, people never give it up. In fact, the success of bitcoin ETF in the future will definitely open the bright door for the cryptocurrency. More financial institutes could take part in the development through bitcoin ETF just what happened in gold market.




https://www.theverge.com/2017/12/4/16732952/winklevoss-twins-bitcoin-billionaires-surge





The release of more stable coins like Gemini Dollar(GUSD). Many people in cryptocurrency field believe the appearance of more stable coins could make the whole mark stable and mature.








The development of coin fund. Coin fund is another attractive field, and could push forward the development of crypotocurrency. One representative operation is the quant contest (reference 2).




https://cryptofundresearch.com/





Crypotocurrency market is still in early and active stage. More new trying e.g., Ethereum (ETH) Constantinople is coming. Let’s look forward to it.




Reference:
1) Machine learning the cuypotocurrency market
2) Predicting Stock Prices Using LSTM
3) Quant contest in cryptocurrency




次世代システム研究室では、グループ全体のインテグレーションを支援してくれるアーキテクトを募集しています。インフラ設計、構築経験者の方、次世代システム研究室にご興味を持って頂ける方がいらっしゃいましたら、ぜひ募集職種一覧からご応募をお願いします。
皆さんのご応募をお待ちしています。