JMLR: Workshop and Conference Proceedings vol (2013) 1–13 Online Learning for Time Series Prediction

We address the problem of predicting a time series using the ARMA (autoregressive moving average) model, under minimal assumptions on the noise terms. Using regret minimiza-tion techniques, we develop effective online learning algorithms for the prediction problem, without assuming that the noise terms are Gaussian, identically distributed or even indepen-dent. Furthermore, we show that our algorithm’s performances asymptotically approaches the performance of the best ARMA model in hindsight