Abstract
At Amundi, we have previously explored NLP approaches but were disappointed by their instability - small changes in input words or word order often led to significant shifts in interpretation.
In our recent research, we revisited modern Transformer-based NLP models. Much like their revolutionary impact on language translation, we found that Transformer-based approaches brought an unexpected level of analytical precision to Natural Language Processing tasks.
Our findings show that AI, combined with Big Data, has the potential to deepen our understanding of financial markets. For example, we were able to quantify economic narratives using news data. This allows us to test the intuitions of renowned academic researchers whose work has shaped financial and economic reasoning in our industry.
As finance becomes increasingly intertwined with AI, new questions will emerge for practitioners, especially as knowledge is now processed and accessed in entirely new forms. Concepts have become vectors. It is our responsibility to leverage AI tools to enhance the robustness of portfolio management.
Authors
