The study of chaotic dynamics in financial time series suffers from the nature of the collected data, which is both finite and noisy. Moreover, researchers have become less enthusiastic since a large body of the literature found no evidence of chaotic dynamics in financial returns. In this paper, we present a robust method for the detection of chaos based on the Lyapunov exponent, which is consistent even for noisy and finite scalar time series. To revitalize the debate on nonlinear dynamics in financial markets, we show that the volatility is chaotic. Applications carried out on eight major daily volatility indexes support the presence of low-level chaos. Further, our out-of-sample analysis demonstrates the superiority of neural networks, compared with other chaotic maps, in the forecasting of market volatility.