[Paper Review] Can social microblogging be used to forecast intraday exchange rates?

Panagiotis Papaioannnou, Lucia Russo+2 more|RePEc: Research Papers in Economics|Oct 20, 2013

Complex Systems and Time Series Analysis4 citations

TL;DR

This study investigates whether Twitter microblogging data can enhance intraday forecasting of the EUR/USD exchange rate beyond traditional models. Using time series analysis and neural networks on sentiment-rich Twitter data, the authors find that social media information significantly improves directional change predictions in the very short term, outperforming random walk and AR models.

ABSTRACT

The Efficient Market Hypothesis (EMH) is widely accepted to hold true under certain assumptions. One of its implications is that the prediction of stock prices at least in the short run cannot outperform the random walk model. Yet, recently many studies stressing the psychological and social dimension of financial behavior have challenged the validity of the EMH. Towards this aim, over the last few years, internet-based communication platforms and search engines have been used to extract early indicators of social and economic trends. Here, we used Twitter's social networking platform to model and forecast the EUR/USD exchange rate in a high-frequency intradaily trading scale. Using time series and trading simulations analysis, we provide some evidence that the information provided in social microblogging platforms such as Twitter can in certain cases enhance the forecasting efficiency regarding the very short (intradaily) forex.

Motivation & Objective

• To test whether social microblogging data can improve short-term exchange rate forecasting beyond the random walk model.
• To evaluate the predictive power of publicly available Twitter sentiment and information flow in the context of the Efficient Market Hypothesis (EMH).
• To investigate whether trader beliefs, as reflected in microblogging content, carry information useful for forecasting intraday exchange rate movements.
• To compare the forecasting performance of models incorporating Twitter data against baseline models using only historical exchange rates.
• To explore the potential of social media as a source of early indicators for market dynamics in high-frequency foreign exchange markets.

Proposed method

• Collected and processed real-time Twitter data related to the EUR/USD exchange rate using keyword-based filtering and sentiment analysis.
• Constructed time series models using ARX (AutoRegressive with eXogenous inputs) and Artificial Neural Networks (ANN) to incorporate Twitter data as exogenous variables.
• Applied bootstrap resampling (5000 iterations) to validate model robustness and compare forecasting performance against baseline models.
• Used log-differenced returns to address non-stationarity and trend effects in exchange rate data.
• Conducted moving average trading simulations to evaluate economic performance of the models in a practical trading context.
• Evaluated forecasting accuracy using the proportion of correctly predicted directional changes (Sgns) at one-step-ahead horizon.

Experimental results

Research questions

• RQ1 Can social microblogging data such as Twitter content improve the forecasting accuracy of intraday exchange rates beyond traditional time series models?
• RQ2 To what extent does incorporating public sentiment and trader beliefs from Twitter enhance the prediction of directional changes in the EUR/USD exchange rate?
• RQ3 Does the inclusion of social media data lead to statistically significant improvements in forecasting efficiency compared to random walk and AR models?
• RQ4 How does the forecasting performance vary across different forecasting horizons, particularly in the very short term?
• RQ5 Can model performance be validated through simulated trading strategies that reflect real market behavior?

Key findings

• The ARX and ANN models incorporating Twitter data significantly outperformed baseline models without social media inputs, particularly at the one-step-ahead forecasting horizon.
• The proportion of correctly predicted directional changes (Sgns) for the best ARX and ANN models exceeded the maximum values of the bootstrap distributions, indicating statistical significance.
• For short-term forecasts (nₖ = 1), the models using Twitter data achieved Sgns values well above 0.5, suggesting meaningful predictive power beyond random chance.
• At larger forecasting horizons (nₖ > 1), no significant improvement was observed, as Sgns values converged to approximately 0.5 with small deviations.
• Trading simulations based on the Twitter-informed models demonstrated economic performance gains, indicating practical utility in real-time forecasting.
• The results were robust to non-stationarity after log-differencing the exchange rate data, confirming the stability of the findings under trend-adjusted conditions.