O mercado de opções antecipa a tempestade? FXVol como previsor da volatilidade da PTAX no Brasil (2010–2024)

Abstract

This paper examines whether the USD/BRL implied exchange-rate volatility, proxied by the FXVol index, contains predictive information for future volatility in Brazil. Using daily data from 2010–2024, we combine PTAX-based log returns with FXVol (reported in annualized percent terms and converted into annualized variance). The forecasting target is the 𝐻-day-ahead annualized realized variance, computed as the sum of squared returns over (𝑡 + 1, . . . , 𝑡 + 𝐻) and scaled by 252/𝐻. Out-of-sample performance is assessed in a rolling-window design of roughly five trading years, comparing an IV-only specification with historical volatility (HV), EWMA, and a HAR benchmark, as well as a combined HAR+IV model. Forecast accuracy is evaluated using RMSE, MAE, and QLIKE, while relative performance is tested with Diebold–Mariano statistics and the Clark–West test for nested models. Results indicate that combining volatility persistence (HAR) with option-implied information (FXVol) improves forecasting accuracy, with stronger gains in high-uncertainty subperiods, especially after 2020.