Source-linked AI summary
Social signals and algorithmic trading of Bitcoin
David Garcia, Frank Schweitzer
TL;DR
The paper examines how large-scale digital traces can be converted into actionable knowledge for algorithmic trading, addressing the gap between social-signal analysis and demonstrated trading profitability. It develops a multidisciplinary framework, applies it to Bitcoin using economic and social signals, and reports profitable strategies supported by statistical evaluation, while cautioning that backtested performance may not persist.
Problem
The paper addresses the gap between statistical analysis or prediction of social and financial signals and evidence that such methods produce profitable trading strategies.
Method
The authors combine multidimensional time-series analysis with social, economic, information-retrieval, and computational-finance methods to design and backtest Bitcoin trading strategies.
Results
Opinion polarization and exchange volume precede rising Bitcoin prices, while emotional valence precedes polarization and rising exchange volumes; strategies based on these findings achieve very high backtested profits.
Takeaways & Limitations
The framework provides tractable principles for designing and evaluating algorithmic trading strategies when social signals are available.
Takeaways & Limitations
Backtested historical profits may not predict future profits because information sources can be adopted by traders and markets may absorb the strategy’s knowledge.
Abstract
from arXiv · showhide
The availability of data on digital traces is growing to unprecedented sizes, but inferring actionable knowledge from large-scale data is far from being trivial. This is especially important for computational finance, where digital traces of human behavior offer a great potential to drive trading strategies. We contribute to this by providing a consistent approach that integrates various datasources in the design of algorithmic traders. This allows us to derive insights into the principles behind the profitability of our trading strategies. We illustrate our approach through the analysis of Bitcoin, a cryptocurrency known for its large price fluctuations. In our analysis, we include economic signals of volume and price of exchange for USD, adoption of the Bitcoin technology, and transaction volume of Bitcoin. We add social signals related to information search, word of mouth volume, emotional valence, and opinion polarization as expressed in tweets related to Bitcoin for more than 3 years. Our analysis reveals that increases in opinion polarization and exchange volume precede rising Bitcoin prices, and that emotional valence precedes opinion polarization and rising exchange volumes. We apply these insights to design algorithmic trading strategies for Bitcoin, reaching very high profits in less than a year. We verify this high profitability with robust statistical methods that take into account risk and trading costs, confirming the long-standing hypothesis that trading based social media sentiment has the potential to yield positive returns on investment.
1 Introduction
The paper addresses the gap between analyzing social signals and demonstrating their profitability in trading. It presents a multidisciplinary framework and applies it to Bitcoin using social and economic digital traces.
- Research motivation: Large-scale observational data create a need for methods that extract actionable knowledge beyond descriptive analysis, especially in computational finance.Technical analysis based mainly on price time series is often insufficient to derive satisfactory returns.
- Contribution: The authors present a framework that combines social psychology and economics with information retrieval, time series analysis, and computational finance.The framework derives stylized facts from multidimensional signals and uses them to design and evaluate algorithmic trading strategies.
- Contribution: The Bitcoin application monitors economic and social digital traces at daily resolution, including market activity, adoption, searches, word of mouth, emotional valence, and polarization.The strategies are evaluated through backtesting and compared with technical-analysis strategies.
- Research gap: Prior studies found temporal relationships between social signals and financial outcomes, but predictive accuracy did not establish profitability in trading scenarios.The paper identifies this missing link across sentiment, media, Twitter, and discussion-pattern studies.
2 Trading strategy framework
The framework converts asset prices and related social or economic signals into returns, identifies temporal signal patterns with a multivariate time-series model, and turns them into evaluated trading strategies.
- Signal preparation: The framework requires an asset price series and can incorporate economic and social signals associated with market agents.Prices are converted into returns to analyze profitability.
- Multidimensional analysis: A lag-1 Vector Auto-Regression models linear relations among all signals while testing and correcting assumptions about stationarity and error correlations.The single multivariate model is used to avoid false positives from pairwise Granger tests.
- Impulse analysis: Impulse Response Functions correct for correlated errors and identify which signals precede changes in price returns.The resulting patterns are treated as stylized facts for strategy design.
- Trading strategy design: A signal predictor uses the direction of its estimated return response and recent signal changes to predict the next period’s price movement.The Combined strategy aggregates predictors through a majority vote.
- Evaluation: Strategies are evaluated through daily data-driven backtesting on a leave-out sample and compared with standard benchmark strategies.The study allocates roughly one year for daily-trading evaluation, depending on expected profitability and variance.
- Bitcoin signals: The Bitcoin implementation combines price, exchange volume, transaction volume, adoption, search, Twitter attention, emotional valence, and polarization signals.The monitoring system combines real-time retrieval with historical time series.
3 Results
The analysis identifies temporal relationships among Bitcoin price returns and social or economic signals, then converts the strongest patterns into trading strategies evaluated against benchmarks and risk measures.
- Signal analysis: Polarization and exchange volume significantly increased price returns one day after shocks, while polarization also increased exchange volume instantaneously.The return response was measured in return percentages and declined rapidly after the first day.
- Strategy design: Valence, polarization, and exchange volume exceeded the 0.1% cumulative-effect threshold, reaching effects up to 0.5% in one day and motivating four trading strategies.The resulting strategies were Valence, Polarization, FXVolume, and Combined.
- Strategy evaluation: Valence, Polarization, and Combined outperformed random traders, while Polarization and Combined clearly outperformed the RSI and Momentum technical strategies.The FXVolume strategy was not far from random-trader performance.
- Strategy evaluation: The Combined strategy produced profits beyond 100% for most of the trading period, with profitability assessed using profit distributions and Wilcoxon tests.The study also evaluated risk-corrected returns with the Sharpe Ratio and examined the distribution and stationarity of Combined-strategy returns.
- Strategy evaluation: The Combined strategy achieved the highest Sharpe Ratio, above 1.75, with mean daily returns above 0.3%.Table 1 compares these measures across the strategies, the DJIA, and 10,000 random traders.
4 Concluding remarks
The framework integrates time-series analysis, strategy design, and backtesting to identify actionable relations between Bitcoin prices and social or economic signals. In simulations, combining valence, polarization, and exchange volume produced very high profits, while the authors caution that historical backtesting may not predict future performance.
- The framework integrates analysis, design, and evaluation of trading strategies with social and economic signals.
- The approach identifies temporal patterns in Bitcoin prices, exchange volume, Twitter valence, and Twitter polarization, using robust methods designed to address noise correlations and finite time series.
- A strategy combining valence, polarization, and exchange volume reached very high profits in less than a year in data-driven simulations.
- The framework offers tractable explanations for strategy mechanisms and can be applied to other trading scenarios with available social signals.
- Historical backtested profits do not necessarily predict future profits, and market adoption, scalability, regulation, and shorting risks may change performance.
- The study presents social signals as convertible into trading profits and as a framework for validating strategy profitability and understanding system dynamics.
5 Materials and methods
The study constructs daily Bitcoin economic and social time series, tests their stationarity and temporal responses with a VAR-based framework, and converts predictions into backtested trading decisions. The simulations incorporate transaction costs, full capital allocation, and limited short selling.
- Time-series preparation: The analysis differences each signal until stationarity, using ADF and KPSS tests before fitting the VAR model.The ADF test tests non-stationarity, while KPSS tests stationarity.
- Temporal analysis: Impulse responses use orthogonalized one-standard-deviation shocks and bootstrap-resampled residuals to compute confidence intervals under VAR error correlations.
- Trading rules: Predictions map to buy, hold, sell, or short positions depending on forecast sign and current ownership, with technical strategies using the price series.
- Trading simulation: The backtest charges proportional buy and sell costs, invests 100% of capital at each timestep, and limits short positions to one iteration.
- Economic signals: Bitcoin prices span daily observations from February 1, 2011, to December 31, 2014, while exchange-rate evaluation uses actual BTC/USD exchange data rather than only the BPI.
- Social signals: The social data include Google search interest, 19,578,671 Bitcoin-related tweets, daily tweet volume, emotional valence, and lexicon-based opinion polarization.