اندازه‌گیری شاخص نا اطمینانی اقتصادی رسانه‌بنیان با الگوریتم‌های یادگیری ماشین در ایران و تاثیر آن بر نرخ ارز

هدف اصلی مقاله اندازه‌گیری شاخص نا اطمینانی اقتصادی با استفاده از اخبار منتشرشده در شبکه‌های اجتماعی است. این روش اندازه‌گیری با فراگیری استفاده از شبکه‌های اجتماعی اهمیت بالایی پیدا کرده است. در این مقاله، با پایش و تحلیل 3,117,960 خبر از 28 کانال تلگرامی پرمخاطب و اثرگذار ایران، شاخص نا اطمینانی اقتصادی در ایران را از ژانویه 2017 تا دسامبر 2020 اندازه‌گیری شد. برای تحلیل این اخبار از روش‌های «یادگیری ماشین با ناظر» بهره ‌گرفته ‌شد. در مرحله اول 13404 خبر توسط ارزیابان انسانی برحسب اثرگذاری بر نا اطمینانی برچسب‌گذاری شد. سپس با استفاده از چهار الگوریتم («c4. 5» از روش‌های درخت تصمیم، «پرسپترون چندلایه» از روش‌های شبکه عصبی مصنوعی، «لجستیک» از روش‌های تابع‌محور و «بیز ساده» از روش‎‌های بیزی) برچسب‌گذاری کل اخبار انجام شد. شاخص نا اطمینانی اقتصادی به‌صورت شمارشی و بر اساس تعداد اخباری که اثرگذار بر نا اطمینانی اقتصادی هستند، اندازه‌گیری و مقدار این شاخص استاندارد شده و سپس کیفیت شاخص با شواهد تاریخی، برچسب‌گذاری مجدد و مقایسه با شاخص مبتنی بر داده‌های گوگل ارزیابی شد. شاخص محاسبه شده با وقایع مهم دوره مطالعه مانند خروج آمریکا از برجام، تحریم نفتی و بالا گرفتن تقابل آمریکا با ایران در ترور سردار سلیمانی همخوانی دارد. برآورد تاثیر نا اطمینانی اقتصادی رسانه‎‌بنیان بر نرخ ارز با مدل گارچ، اثر مثبت و معنی‌دار این نا اطمینانی را نشان می‌دهد.

measuring the media-based economic uncertainty index by machine learning algorithms in iran and its effect on the exchange rate

1- introduction economic uncertainty is one of the important and influential factors on economic policies and their results, and in such a situation, rational decisions are replaced by other methods. various studies has shown the effect of economic uncertainty on inflation, investment, economic growth, consumption and demand for money. uncertainty is difficult to measure due to its invisibility, and as the uncertainty measurement methods improve, the measurement of its effect on various economic variables and markets and the prediction of their behavior in response to the actions of economic agents will be more accurate. the main aim of this article is to measure the economic uncertainty index by using news published in social networks. this method of measurement has become very important with the widespread use of social networks. 2- theoretical framework uncertainty is one of the most controversial concepts in the philosophy and methodology of economics. the history of the concept of economic uncertainty goes back to david hume .there are three categories of theories about economic uncertainty . the first group believes that the future reality is unchangeable and predetermined and economic decision makers have perfect information. in this view, there is no such thing as uncertainty and the world is in complete certainty. 18th century the economists of were the first group to present this theory . the second group believes that the reality of the future is unchangeable and predetermined and the decision makers are able to know the future. these economists use objective conditional probability functions to solve the future uncertainty problem . the third class considers the future reality to be changeable and unknown. the starting point of these theories started from the study of the chicago school economist frank knight titled risk, uncertainty and profit. he clearly distinguished between the two concepts of risk and uncertainty. keynes also reached the same results as knight. in general, in a situation where the economy has a high level of uncertainty, the theories of the first and second category have a good explanation. but in confronting with exogenous shocks such as the corona virus epidemic, war and financial crisis, the concept of uncertainty will be more appropriate in the theories of the third category. this study will measure this index based on fundamental uncertainty (the third category). 3- methodology in this article, the economic uncertainty index in iran was measured from january 2017 to december 2020 by monitoring and analyzing 3,117,960 news from 28 popular and influential iranian telegram channels. to analyze these news, we used supervised machine learning methods. in the first step, 13,404 news items were labeled by human evaluators according to their impact on uncertainty. the labels had two modes affecting uncertainty and neutral. then by using four algorithms (c4.5 from decision tree methods, multilayer perceptron from artificial neural network methods, logistics from function-oriented methods and simple bayes from bayesian methods) labeling of the whole news was done. the economic uncertainty index was calculated numerically and based on the number of news items that affect economic uncertainty, the measurement and value of this index was standardized, and then the quality of the index was evaluated with historical evidence, relabeling and comparison with the index based on google data. 4- results & discussion among the 4 media-based uncertainty indicators, 3 indicators can better explain the historical events of this period. among them, the best performance is determined by c4.5 algorithm from the decision tree methods. after this algorithm, multilayer perceptron, logistic has the best performance and the weakest performance belongs to the simple bayes method. media-based economic uncertainty index trend with c4.5 method is consistent with the important events of