Abstract
Modern economic literature features quite a number of various indices of economic activity. Some of them are based on public opinion polls (‘manual’ indices), while others are based on unstructured data from the Internet (‘automatic’ indices). However, the question as to which of these approaches is the most effective remains open. In this paper, we compare several different indices of economic activity in terms of their explanatory and predictive power. We build ‘automatic’ indices using machine learning methods. Search queries, news articles and user comments under news posts from social media are used as source data. The analysis of the resulting indices of economic activity shows that the search and news indices Granger-cause ‘manual’ indices and also better explain and predict the set of macroeconomic variables selected for research. The good explanatory power of the current values of macroeconomic indicators by means of current indices of economic activity with a lag in the output of macroeconomic statistics makes them suitable for nowcasting.
Highlights
For a timely assessment of the state of the economy and early detection of economic instability, leading indicator systems are used
Starting with Choi and Varian (2009) and Choi and Varian (2012), many authors suggest that the building of such indicators can be automated by processing large arrays of data generated as a result of human-internet interaction using modern statistical methods
The index value is calculated as the difference between the share of respondents who noted an improvement in the economic environment and the share of those who noted its deterioration. Such indices are constructed in a similar way in many countries; in particular, in Russia the Michigan Index methodology is used to build the consumer sentiment index3 (CSI), which has been calculated by the Levada Centre since the 1990s
Summary
For a timely assessment of the state of the economy and early detection of economic instability, leading indicator systems are used. They help monitor and forecast business activity and reduce the time intervals required for making proactive decisions that are key for the macroeconomic policy. Many such indicators are constructed manually through of consumer and business surveys. Starting with Choi and Varian (2009) and Choi and Varian (2012), many authors suggest that the building of such indicators can be automated by processing large arrays of data generated as a result of human-internet interaction using modern statistical methods. We construct several low-frequency (monthly) indicators reflecting the current state of the economy and compare them with each other and with the indicators constructed manually
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have