forecasting of stock returns with non linear models and the role of trading volume in improving the performance of these models

non-linear time series models have become fashionable tools to describe and forecast stock market returns in recent years. a significant amount of evidence supports a negative relationship between volume and future returns. this suggests that volume could act as a suitable threshold variable in lstar and tar models. in this research, we compared the forecasting ability of lsatr and tar models with arma. moreover, we used lagged volume as the threshold in lstar and tar. daily stock returns and volume of 26 companies were used over the sample period 21/03/2001 to 20/03/2010 years. in order to conduct a forecasting exercise we used the 7 years data as the in sample estimation period and the reminder of the sample as the out of sample period. we used diebold- mariano test to compare forecasting power of the models. results show that non-linear models have higher forecasting power than arma. also using volume did not improve the forecasting performance of lstar and tar.