Autoregressive LLMs generate text by sampling from estimated probability distributions over the next token, conditional on preceding context. We leverage these conditional probabilities to construct an entropy-based measure of prediction uncertainty, which we term inner confidence. Predictions with higher inner confidence are systematically more accurate.
Forecasting the option implied volatility (IV) surface is difficult with standard time-series models because of its time-varying granularity. We propose a new two-step real-time sequential forecasting framework. The first step fits the daily surface and can accommodate any underlying specification for option prices or IVs, including dynamic option-pricing models, nonparametric methods, and machine-learning techniques. In the second step, we sequentially estimate a dynamic IV model using an updating rule.