Using high-frequency foreign exchange data, we estimate short-lived signals to forecast one-minute-ahead rolling currency prices using other currency pairs, commodities and stock market indices as predictors. With such a large set of covariates, we impose sparsity using Elastic-Net estimator to find unexpected signals in the intraday currency market, what allow us to deal dynamically with multicollinearity, while constraining the size of the estimates of uninformative variables. We investigate the existence of signal patterns for the five most liquid currencies (British Pound, Canadian Dollar, Euro, Japanese Yen, and Swiss-Franc) for the year of 2018. The paper also shed lights on the time-of-day effects reported in the literature.