تعيين طول دوره زماني بهينه در مدل هاي پيش بيني سود

Usefulness of earnings forecasting, and determining a suitable model that can forecasts earnings with less error, have an important role in stakeholders decisionsʹʹ Models. In this regard, random walk

هدف اين پژوهش مقايسه ي مدل هاي گام تصادفي و سري زماني در پيش بيني سود عملياتي شركت‌هاي پذيرفته شده در بورس اوراق بهادار تهران، با آزمودن مقادير قدرمطلق درصد خطاهاي پيش بيني، و نيز محاسبه ي طول دوره ي زماني بهينه براي استفاده در مدل سري زماني سود، با استفاده از آماره ي آكاييك (AIC)، مي باشد. داده هاي مورد نياز شامل 828 سال- شركت، از صورت هاي مالي سال هاي 1372 تا 1389 (46 شركت) استخراج گرديده است. نتايج نشان مي دهد در مدل هاي ساده ، مدل گام تصادفي نسبت به مدل اتورگرسيو به طور عمده و معني دار، و در مدل هاي تركيبي، به طور جزيي، پيش بيني را با خطاي كمتري انجام مي دهد. نتايج همچنين نشان مي دهد در مدل سري زماني، استفاده از وقفه هاي بالاتر باعث بهبود مدل ها نخواهد شد و در بين وقفه هاي 1 تا 6 مورد بررسي، وقفه 1 براي مدل هاي ساده و وقفه 2 براي مدل هاي تركيبي، بهينه مي باشد.

Usefulness of earnings forecasting, and determining a suitable model that can forecasts earnings with less error, have an important role in stakeholders decisionsʹ Models. In this regard, random walk and time series models have been investigated in researches that tried to find the appropriate models for earnings forecasting. The purpose of this study is comparing the random walk and time series (autoregressive) models in forecasting of operating earnings for companies listed in Tehran Stock Exchange, and identifying the optimal length of time period to use in time series models. This research investigates the difference between absolute percentage errors(APEs) of random walk and autoregressive models in earnings forecasting. Furthermore, this study has a special emphasis on identifying the optimal time period for application in the earnings time series, and so uses the Akaike information criterion (AIC). The required data (including 828 years– company), from 1372 to 1389 fiscal year (46 companies listed in Tehran Stock Exchange) has been collected. The results show that, random walk model have less APEs, but the differences are significant only in simple models (explanatory variables are not used). The results suggest that time series models will not be improved by using the greater lags. Also the lags of 1 and 2 (from 1 to 6) are optimum in simple and combined time series models, respectively.