نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشجوی دکتری، گروه اقتصاد، دانشکده علوم اداری و اقتصاد، دانشگاه فردوسی، مشهد، ایران
2 استادیار، گروه اقتصاد، دانشکده علوم اداری و اقتصاد، دانشگاه فردوسی، مشهد، ایران
3 استاد، گروه اقتصاد، دانشکده علوم اداری و اقتصاد، دانشگاه فردوسی، مشهد، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
This study examines textual data's ability to nowcast Iran's gross domestic product (GDP). To this end, 301,498 economic news articles from March 2005 to December 2023 were extracted from the Fars news agency website using a web crawling technique. Following initial preprocessing, the news texts were sorted into various categories via the Dirichlet Latent Allocation (LDA) model, wherein each category corresponds to a distinct news topic. Subsequently, to ascertain whether an article conveys a positive or negative sentiment, we executed lexicon-based sentiment analysis utilizing SentiStrength. Ultimately, by aggregating the news sentiment scores seasonally under each topic, we constructed a seasonal sentiment time series. These time series were then assessed for their efficacy in nowcasting Iran's quarterly GDP, employing ridge regression, lasso regression, elastic net, and gradient boosting methods. The findings reveal that incorporating textual data can reduce prediction errors by 12 to 18 percent relative to a univariate time series model. Moreover, our results suggest that sentiment extracted from textual content, particularly news articles, is a viable approach. This strategy could potentially enable the provision of immediate GDP estimates following the end of each reference quarter.
کلیدواژهها [English]