The search algorithm of optimal time series model for forecasting Latvian GDP. In this scientific paper there is developed algorithm for a finding optimal time series model for GDP forecasting. Latvian GDP statistical data with quarterly observation frequency is taken as a time series. ARMA Analysis of Latvian GDP Time series is performed and described. The set of model has been constructed. For check of quality of models Residual tests are performed and models are compared between themselves. Using econometric software EViews 6.0 forecasts for best models are made and results are compared with real data of last three squares of year 2009.