Time series forecasting by finite-state automata

We have developed automata to address the problem of time series forecasting. After turning the time series into sequences of letters, Mohri's algorithm constructs an automaton indexing this text that, once a given word is read, can be used to obtain the set of its positions. By using the automaton to determine what letter usually follows the last sequence of length L, we have developed a one step predictor. This predictor will be used to analyze sunspot activity data.