Abstract: In finance,scholars applied no nlinear models to describe and predict securities price movement in recent years,especially using Smooth Transition Auto Regression(STAR)model and Artificial Neural Network(ANN)model.In this paper,we examine the out-of-sample fo recasts performance of several linear and no nlinear models for ShangZheng 180 index in short term,middle term and relatively long term as well.We compare their forecast precision by statistical and investment criteria.The results indicate that securities price has nonlinear characteristics in our country,and is able to be predicted in some extent, which put the Efficient Markets Hypothesis(EMH)into question.Furthermore,ANN models mostly produce better forecasts than the RW,AR model as well as STAR model.Nevertheless,on average,investors can obtain a higher net return according to a market-timing strategy based on the forecasts fro m ANN models than a risk-adjusted buy and hold strategy.

Key words: efficient market, predictable, nonlinear model, investment strategy