SUN ZU Lab WEBB3 Analytics #2
By Arthur Serge, Mounir Chaabani and Chadi El Adnani
Predicting volatility is an exciting task due to its importance in creating trading signals and calibrating agents’ positions. We explored at SUN ZU Lab different methods proposed in the literature, and we propose new methods tailored for cryptocurrencies. In this article, we focus on BTC volatility forecasting. This task is non-trivial due to factors such as high noise-to-signal ratio, market microstructure, or heteroscedasticity.
Our method is mainly based on using neural networks adapted for sequential data, as is the case for time series. The advantage of neural networks is their ability to extract complex features and non-linear effects, which is not the case for the majority of statistical models applied to quantify volatility.
We use BTC-USD data from Bitstamp, focusing on transaction and order book data to compute volatility and features for our neural network. The time series runs from February 2022 to April 2022. For this task, we used the Garman-Klass proxy for volatility as we work in the context of high-frequency data, using 10-min volatility for each 5 min. The idea is to predict the volatility for the next 5 minutes based on recent volatility levels and other market indicators.
We jump directly to the conclusion and display a chart summarizing our results:
Prediction of the last 500 out of sample points
Questions and comments can be addressed to email@example.com
About SUN ZU Lab
SUN ZU Lab is a leading data solutions provider based in Paris, on a mission to bring transparency to the global crypto ecosystem through independent quantitative analyses. We collect the most granular market data from major liquidity venues, analyze it, and deliver our solutions through real-time dashboard & API stream or customized reporting. SUN ZU Lab provides crypto professionals with actionable data to monitor the market and optimize investment decisions.