News, Volatility and Jumps: The Case of Natural Gas Futures

Abstract

We investigate the impact of Thompson Reuters News Analytics (TRNA) news sentiment on the price dynamics of natural gas futures traded on the New York Mercantile Exchange (NYMEX). We propose a Local News Sentiment Level (LNSL) model, based on the Local Level model of Durbin and Koopman (2001), to construct a running series of news sentiment on the basis of the 5-minute time grid. Additionally, we construct several return and variation measures to proxy for the fine dynamics of the front month natural gas futures prices. We employ event studies and Granger causality tests to assess the effect of news on the returns, price jumps and the volatility. We find significant relationships between news sentiment and the dynamic characteristics of natural gas futures returns. For example, we find that the arrival of news in non-trading periods causes overnight returns, that news sentiment is Granger caused by volatility and that strength of news sentiment is more sensitive to negative than to positive jumps. In addition to that, we find strong evidence that news sentiment severely Granger causes jumps and conclude that market participants trade as some function of aggregated news. We apply several state-of-the-art volatility models augumented with news sentiment and conduct an out-of-sample volatility forecasting study. The first class of models is the generalized autoregressive conditional heteroskedasticity models (GARCH) of Engle (1982) and Bollerslev (1986) and the second class is the high-frequency-based volatility (HEAVY) models of Shephard and Sheppard (2010) and Noureldin et al. (2011). We adapt both models to account for asymmetric volatility, leverage and time to maturity effects. By augmenting all models with a news sentiment variable, we test the hypothesis whether including news sentiment in volatility models results in superior volatility forecasts. We find significant evidence that this hypothesis holds.