Abstract:In searching for an appropriate time series model based on historical data we applied unit root tests and considered 12 different continuous-times stochastic models prices of six saw log and pulp wood products from two important species of Scots pine (Pinus.sylvesteris) and Norway spruce (Picea.abies), as well as, average softwood log prices for annual long run and shorter monthly time series in the Finnish wood market. For each product we conducted a comparative analysis between models on the basis of Akaike's Information criteria (AIC), the mean square error (MSE) of the models after one period of forecasting, and a likelihood ratio test. Parameter estimation was performed by quasi maximum likelihood estimation and local linearization method. The unit root tests results showed that while in the long run the price of softwood is trend stationary, in short run it shows non-stationary behaviour. Our results also showed that the level of effect of state the variable on volatility has a major role in refining a general model in to simpler models. The model with a general form of diffusion and no drift yields the highest AIC for most products, and the diffusion part of the model plays an important role in ranking by AIC, while in ranking by MSE for one period of forecasting, the drift part of models plays important role.