Factor Variability Test in Stock Return Forecasting Using Dynamic Model Averaging (DMA)

In this study, using dynamic averaging models and monthly data in the period 2001:4 until 2018:3, Tehran Stock Exchange returns be investigated. In this regard, macroeconomics variables and parallel markets indices have been used to forecast the stock returns. Initially, estimating various models such as Recursive models, time-varying parameter models (TVP), dynamic model selection (DMS) and dynamic model averaging (DMA) in Matlab software, It was observed that DMS model with α = β = 0.95 had higher forecast accuracy (based on MAFE, MSFE and Log (PL) metrics). Gold price (48-period), exchange rate (36-period) and inflation rate (30-period) had the highest effect on stock returns, respectively, and global oil prices and GDP had the lowest effect by 28 and 2, respectively. Finally, the results indicate that utilizing dynamic models by considering time variations in parameters and the variation of the model increases the efficiency of forecasting stock returns. Keywords: Forecasting, Stock Returns, time-varying Parameter (TVP), Dynamic Model Averaging (DMA).