Machine Learning Strategies for Inflation Swaps Using News Sentiment

Highlights:

• While traditional financial data remains the primary input for assessing financial markets, alternative data is becoming an increasingly valuable complement. In particular, sentiment data has demonstrated strong potential for providing market insights that support investment decisions.
• Integrating quantitative techniques into the investment process as a complement to human expertise is essential for enriching decision-making.
• One of the initial applications of alternative data is the generation of proprietary market sentiment indicators. By lever-aging these indicators, we applied machine learning and deep learning techniques to forecast US Inflation Swaps move-ments in a one-month horizon.
• This paper presents our ongoing development of expertise in analysing news using the Global Database of Events, Language, and Tone (GDELT) dataset. We construct indices and generate signals through machine learning (ML) and deep learning (DL) techniques, detailing the methodological framework adopted to process and model this high-dimen-sional data.
• Backtesting results pave the way for discussing prac-tical applications. Further research could focus on dif-ferent instruments and/or geographies. We recom-mend using XGB for its robustness and extremely high hit ratio. Note that 3/5 models generate negative re-turns, which is to be expected given high transaction costs and their lower hit ratio, for the reasons dis-cussed in dedicated sections.

Download the full publication below