Skip to content
All library documents

Predicting Bitcoin Volatility from Order-Flow Images

Article arXiv papers · Author: Artem Lensky et al.
Read the original

Summary

This paper predicts short-term realized Bitcoin volatility by encoding snapshots of order-flow data as images. Trade size, direction, and limit order book information are mapped to color channels, then used to train a three-layer convolutional neural network, ResNet-18, and ConvMixer. The study also tests adding handcrafted features and compares the image-based models with GARCH, a multilayer perceptron trained on raw data, and a naive forecast based on current volatility.

Experiments use price data from January 2021 and assess forecast error with RMSPE. The image-based CNN performs best among the described approaches, with lower reported error when supplemented by aggregated features; ConvMixer with features is close behind. The naive current-volatility forecast has higher reported error. These results suggest that image representations can make order-flow patterns useful for short-horizon volatility prediction, but the evidence covers a single month of Bitcoin data. The description does not provide details on validation design or performance across other periods and markets.

Key ideas

  • Order-flow snapshots combine trade sizes, trade directions, and order-book information as image channels.
  • CNN, ResNet-18, and ConvMixer models are evaluated, with and without handcrafted features.
  • The reported CNN forecasts outperform the listed GARCH, raw-data MLP, and naive comparison methods.
  • Adding aggregated features improves the reported CNN result.
  • The evidence is limited to Bitcoin data from January 2021.

Tags

CryptoVolatilityMachine learningMarket microstructureBacktesting
Full text
# Learning to Predict Short-Term Volatility with Order Flow Image Representation


# Learning to Predict Short-Term Volatility with Order Flow Image Representation









Introduction: The paper addresses the challenging problem of predicting the short-term realized volatility of the Bitcoin price using order flow information. The inherent stochastic nature and anti-persistence of price pose difficulties in accurate prediction. Methods: To address this, we propose a method that transforms order flow data over a fixed time interval (snapshots) into images. The order flow includes trade sizes, trade directions, and limit order book, and is mapped into image colour channels. These images are then used to train both a simple 3-layer Convolutional Neural Network (CNN) and more advanced ResNet-18 and ConvMixer, with additionally supplementing them with hand-crafted features. The models are evaluated against classical GARCH, Multilayer Perceptron trained on raw data, and a naive guess method that considers current volatility as a prediction. Results: The experiments are conducted using price data from January 2021 and evaluate model performance in terms of root mean square error (RMSPE). The results show that our order flow representation with a CNN as a predictive model achieves the best performance, with an RMSPE of 0.85+/-1.1 for the model with aggregated features and 1.0+/-1.4 for the model without feature supplementation. ConvMixer with feature supplementation follows closely. In comparison, the RMSPE for the naive guess method was 1.4+/-3.0.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.