Ascertaining price formation in cryptocurrency markets with machine learning

Abstract

The cryptocurrency market is amongst the fastest-growing of all the financial markets in the world. Unlike traditional markets, such as equities, foreign exchange and commodities, cryptocurrency market is considered to have larger volatility and illiquidity. This paper is inspired by the recent success of using machine learning for stock market prediction. In this work, we analyze and present the characteristics of the cryptocurrency market in a high-frequency setting. In particular, we applied a machine learning approach to predict the direction of the mid-price changes on the upcoming tick. We show that there are universal features amongst cryptocurrencies which lead to models outperforming asset-specific ones. We also show that there is little point in feeding machine learning models with long sequences of data points; predictions do not improve. Furthermore, we solve the technical challenge to design a lean predictor, which performs well on live data downloaded from crypto exchanges. A novel retraining method is defined and adopted towards this end. Finally, the trade-off between model accuracy and frequency of training is analyzed in the context of multi-label prediction. Overall, we demonstrate that promising results are possible for cryptocurrencies on live data, by achieving a consistent 78% accuracy on the prediction of the mid-price movement on live exchange rate of Bitcoins vs. US dollars.

Online risk-based portfolio allocation on subsets of crypto assets applying a prototype-based clustering algorithm

“…Livieris et al (2020) ensemble different learning strategies that exhibit high efficiency and reliability mainly for low-frequency applications. Fang et al (2021) analyze a data-driven approach with a retraining method to predict successful mid-price movements in cryptocurrency markets. The disadvantage of the learning algorithms that take advantage of the market inefficiency is that the methods are data-hungry (Marcus 2018) and the forecasting benefit decay in non-stationary time series.…”

Cryptocurrency market trend and direction prediction using Machine Learning: A Comprehensive Survey

“…[44]Gil Cohen(2019) They also proposed a trading strategy based on their proposed ML model. [46]Fan(2021) compared dynamic and fixed retraining of the ML model, the effects of retraining on prediction accuracy, and the use of auto-encoder for mid-price movement prediction. Researchers discussed the long-term and short-term predictions and stated the ML model specs that performed better for both cases.…”

Feature Selection for Forecasting

“…Machine learning models are more general, make fewer simplifying assumptions, and offer better model-fitting capabilities. Due to these reasons, techniques such as ML and deep learning created a new research direction in the financial literature (Fang et al, 2021). To forecast the future value of financial assets and find the reason for these assets' behavior, numerous machine learning (ML) techniques were employed, such as SVM (Akyildirim et al, 2021), Support Vector Regression (SVR) (Kara et al, 2011), Random Forest (RF) (Patel et al, 2015a), and convolutional neural networks (CNN) (Tsantekidis et al, 2017).…”