Summary of Paper - Statistical Arbitrage in Cryptocurrency Markets, by Thomas Günter Fischer, Christopher Krauss and Alexander Deinert
Link to paper: https://www.mdpi.com/1911-8074/12/1/31
Link to source: None available
Further steps: This paper has enough to interest to warrant coding up some of the ideas contained within
The authors of this paper use a basket of 40 cryptocurrencies, sampled at 1 minute intervals from Jan 5, 2018 to Sept 7, 2018. They obtain their data from cryptocompare.com. Importantly, they obtain data in USD terms, and many traders are interested in increasing their portfolio in terms of BTC rather than in USD. Obviously, the sample time they worked on was a bear market for crypto in USD terms.
The authors attempt to find cryptocurrencies that will outperform the other cryptocurencies in the sample set While many paper attempt to use returns as the target variable, this method uses whether the instrument outperforms the cross-sectional median of the other coins in the sample set.
As features, the authors use a method that gives returns over certain past periods, with a higher focus on more recent returns. The forumula is in the paper. No other variables are used.
The paper uses logistic regression as a baseline, and Random Forests as the preferred forecasting tool. The method attempts to buy the top 3 cryptos, and short the bottom 3. The authors actually appear to make reasonable estimations for cost and slippage. As would be expected, the short trades do better in the bear market used in the sample window. It’s amazing that the longs are not a complete disaster in USD terms, really.
The authors find that volatility is decreased by their method. The return is bette rthan BTC, but the window is too short and BTC returns are all over. The equity curve is much better with this method that with raw BTC buy and hold. It would be interesting to see how this method performs in terms of BTC denominated pairs in a so-called “alt season”
The authors also note that this strategy would have capacity problems, as it deals with some very low cap coins. However, it could still be an interesting exercise to attempt to replicate some of the method contained within.