ارزيابي قدرت پيش بيني قيمت سهام با استفاده از مدل هاي خاكستري، شبكه هاي عصبي ايستا و پويا (مطالعه موردي: شركتهاي فعال در صنعت بيمه عضو بورس اوراق بهادار تهران)

Evaluating Forecasting ability of Stock Price by Grey Models, Static and Dynamic Neural Networks (Case Study: Insurance Companies of Tehran Stock Exchange)

ﭘﯿﺶﺑﯿﻨﯽ ﻗﯿﻤﺖ ﺳﻬﺎم ﻣﻮﺿﻮﻋﯽ ﻣﻬﻢ در ﻫﺮ دو دﯾﺪﮔﺎه ﻧﻈﺮي و ﮐﺎرﺑﺮدي اﺳﺖ. ﻫﺪف ﻣﺤﻘﻘﺎن، ﺗﻮﺳﻌﻪ روشﻫﺎي ﭘﯿﺶﺑﯿﻨﯽ ﺑﻪ ﻣﻨﻈﻮر ﭘﯿﺶﺑﯿﻨﯽ دﻗﯿﻖﺗﺮ اﺳﺖ. ﺳﺮﻣﺎﯾﻪﮔﺬاران ﺳﻌﯽ در ﯾﺎﻓﺘﻦ ﺑﻬﺘﺮﯾﻦ ﺑﺮﻧﺎﻣﻪ ﺳﺮﻣﺎﯾﻪﮔﺬاري دارﻧﺪ ﮐﻪ اﯾﻦ اﻣﺮ ﻧﯿﺎزﻣﻨﺪ ﭘﯿﺶﺑﯿﻨﯽ آﯾﻨﺪه ﺑﺎزار ﻣﯽﺑﺎﺷﺪ. ﻫﺪف اﯾﻦ ﻣﻘﺎﻟﻪ ﻣﻘﺎﯾﺴﻪ روشﻫﺎي ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ )ANN(، ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﭘﻮﯾﺎ )NARX( و ﻣﺪل ﺧﺎﮐﺴﺘﺮي )GM( ﺑﺮاي ﭘﯿﺶﺑﯿﻨﯽ ﻗﯿﻤﺖ ﺳﻬﺎم ﻣﯽﺑﺎﺷﺪ. دادهﻫﺎي ﺳﺮيزﻣﺎﻧﯽ ﺑﻪ ﺻﻮرت روزاﻧﻪ ﻣﺮﺑﻮط ﺑﻪ ﺷﺮﮐﺖﻫﺎي ﺑﯿﻤﻪاي ﻋﻀﻮ ﺑﺎزار ﺑﻮرس ﺗﻬﺮان ﻣﯽﺑﺎﺷﺪ ﮐﻪ در ﺑﺎزه زﻣﺎﻧﯽ 1388/7/15 ﻟﻐﺎﯾﺖ 17/ 1396/7 ﮐﻪ در ﺑﺎزار ﺑﻮرس ﻓﻌﺎﻟﯿﺖ داﺷﺘﻪاﻧﺪ. ﻣﺘﻐﯿﺮﻫﺎي ﻣﯿﺎﻧﮕﯿﻦ ﻣﺘﺤﺮك ﺳﺎده ﭘﻨﺞ روزه )5-MA(، ﻣﯿﺎﻧﮕﯿﻦ ﻣﺘﺤﺮك ﺳﺎده ﺑﯿﺴﺖ روزه )20-MA(، ﻣﯿﺎﻧﮕﯿﻦ ﻣﺘﺤﺮك ﻫﻤﮕﺮا واﮔﺮا )MACD(، ﻗﯿﻤﺖ ﻃﻼ، ﻗﯿﻤﺖ ﻧﻔﺖ و ﻧﺮخ ارز ﺑﻪ ﻋﻨﻮان ﻣﺘﻐﯿﺮﻫﺎي ﻣﺴﺘﻘﻞ در ﻧﻈﺮ ﮔﺮﻓﺘﻪ ﺷﺪهاﻧﺪ. ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﻣﺘﻐﯿﺮﻫﺎي ﻣﺴﺎﻟﻪ، از ﺳﻪ ﻣﺪل ﺧﺎﮐﺴﺘﺮي )1,1(GM(1,4), GM و 1,7(GM ﺟﻬﺖ ﭘﯿﺶ ﺑﯿﻨﯽ اﺳﺘﻔﺎده ﺷﺪه اﺳﺖ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن ﻣﯽدﻫﺪ روشﻫﺎي ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ و ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﭘﻮﯾﺎ داراي ﮐﺎراﯾﯽ ﯾﮑﺴﺎن ﻣﯽﺑﺎﺷﻨﺪ در ﺣﺎﻟﯿﮑﻪ ﻣﺪلﻫﺎي ﺧﺎﮐﺴﺘﺮي ﮐﺎراﯾﯽ ﭘﺎﯾﯿﻦﺗﺮي دارﻧﺪ. ﺷﺒﯿﻪﺳﺎزيﻫﺎي ﻋﺪدي ﻧﺸﺎن ﻣﯽدﻫﺪ ﮐﻪ روشﻫﺎي ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ و ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﭘﻮﯾﺎ ﺑﺎ ﻣﯿﺎﻧﮕﯿﻦ ﺧﻄﺎ 0.2=RSME ﭘﯿﺶ-ﺑﯿﻨﯽ ﻗﺎﺑﻞ ﻗﺒﻮﻟﯽ اراﯾﻪ ﻣﯽﮐﻨﻨﺪ.

Predicting stock price is an important issue in both theoretical and practical aspects. Researchers develop prediction methods to get more accurate forecasting and investors try to find best investing program which depends on future prediction of their markets. The aim of this paper is comparing artificial neural network (ANN), nonlinear autoregressive exogenous model (NARX) and grey model (GM) for predicting stock price. The stock prices of insurance companies in Tehran Stock Exchange are considered in the period 7-10-2009- 9-10-2017. The variables 5 days simple moving average (MA-5), 20 days simple moving average (MA-20), moving average convergence divergence (MACD), gold price, oil price and exchange rate are considered for the prediction. Based on these variables, the models GM(1,1), GM(1,4) and GM(1,7) are selected for the prediction. The results show that ANN and NARX are in the same performance level while grey models have lower performance. The numerical simulations demonstrate that ANN and NARX provide reasonably good prediction with the average error RSME=2.04.