Anmerkungen:
Nach Informationen von SSRN wurde die ursprüngliche Fassung des Dokuments December 19, 2022 erstellt
Beschreibung:
Quantile forecasting has become an important research topic in econometrics as policy makers and investors are increasingly interested to focus more on downside (upside) risks rather than learning about the most likely outcome. Simultaneously, practitioners have largely used textual data to con- struct new predictors of financial and economic variables. This paper develops a novel methodology for out-of-sample quantile forecasting with textual data. It relies on elastic net quantile regression to incorporate "attention" into the quantile forecasting model, allowing for the dictionary to vary over time and across the quantiles of the conditional distribution. The proposed methodology predicts probabilities of depreciation that are strongly related to the downside entropy of the US-Canada ex- change rate, meaning that such forecasts carry information about downside risks. We show quantile forecasts from the proposed approach outperform the ones from other quantile models by statistically and economically meaningful margins