The model produced several observable patterns in each market behavior and language structure. These results illustrate how text-based signals correspond to subsequent yield curve movements.
Market structure and curve dynamics
First, short-term volatility within the Brazilian bond market is higher than long-term volatility. This goes against traditional theory and suggests that investors in emerging markets are more sensitive to short-term news and political signals. Long-term instruments appear to trade with comparatively lower volatility, reflecting the dominance of institutional investors at longer maturities.
Furthermore, 84% of day by day yield curve movements fall in 4 of the eleven standard configurations identified within the literature, with parallel upward and parallel downward shifts amongst probably the most common (also confirming this short-term volatility character). This concentration highlights the importance of appropriately classifying a small set of dominant curve dynamics.
Extract signal from speech
To prepare the text data, common words similar to “committee”, “scenario”, “billions” and “prices” were removed as stop words as they don’t contribute to the classification. Word frequencies were then mapped for every category of yield curve movement, allowing comparison of language patterns across different curve configurations.
Seasonality in curve movements
When examining the language related to specific movements, a seasonal pattern emerged. For example, flattening moves in bears were often related to references to August, September, and October, while flattening moves in bulls were more often related to January, February, and March. A chi-square test provided statistical evidence of seasonality across multiple yield curve movements.
