This thesis proposes an algorithm, which is an adaptation of evolution strategies, for time series forecasting. The proposed methodology enables the search for a prediction function without the knowledge of the functional form a priori. The original evolution strategies are designed for real value optimization of the coefficients, while the exact functional form is still required as an input. However, the proposed adaptive evolution strategies adjust the functional form as well as the coefficients via the evolution process. This approach is, therefore, suitable for various applications where the functional forms are not known a priori. In this algorithm, the functional form is randomly generated and evolved via mutation and selection in order to minimize an error. We applied the proposed adaptive evolution strategies to forecast the Baht/US-dollar exchange rate, the bank deposit and the Thailand stock exchange index. The results show that the proposed method can successfully formulate a prediction functionfor these applications and yield errors of less than 5% in all cases.