Understanding Reform Narratives: Research Proposal

经济改革是增长和发展的重要组成部分。改革解决危机，纠正低效，提高生产力。近代历史的特点是几次大规模的改革，如后共产主义向市场经济的过渡，拉丁美洲的经济自由化，以及全球金融危机后的结构性改革。然而，改革少之又少。此外，它们是长期公开讨论和政治谈判的结果。本建议概述了一项研究议程，以了解这些改革讨论的性质、它们随时间和各国的演变，以及它们如何有助于最终政策决定的成功。主要数据来源将是各国从各种来源发表的新闻文章。这种方法有三个优点：首先，新闻每天发布，可以提供一个非常高频的改革讨论措施。第二，由于新闻文章是在每个国家发布的，所以这个项目可以使用整个世界作为完整的样本。最后，自20世纪80年代以来，已经存在大型新闻聚合器，提供对新闻数据的访问。这些来源可以在各种维度上进行搜索，以确定改革文章。本项目寻求解决的主要研究问题是：1。针对经济周期的改革讨论是否有系统的模式？这些在不同国家的制度特征和时间上有什么不同？更高强度的改革讨论是否会导致更快地实施实际改革？3. 随着时间和国家的不同，人们对改革的看法有何不同？负面情绪是否预示着拟议改革的失败？4. 重大经济危机如何改变了与经济改革相关的言论？例如，全球金融危机如何影响人们对自由化政策的态度？该项目的方法论将涉及应用著名的机器学习和自然语言处理算法来分析数百万篇与经济改革有关的报纸文章。动态主题建模是一种很有前途的降维方法：这些算法允许研究人员发现潜在的主题，并跟踪特定术语和主题随时间的演变。尽管经济学和政治学文献中已经应用了这种方法，但还没有论文使用这种方法来分析经济改革叙事。主题建模将提供关于叙事如何随时间演变的答案，特别是在大衰退之后。第二种应用的方法是情感分析，使用计算机算法来理解文本数据的情感倾向。文献中提出了许多不同的情感分析方法。对于这个项目，研究对经济改革的态度是至关重要的，因为公众和媒体的情绪往往是改革最终成败的决定性因素。主要的方法挑战将是设计一种结构化和一致的方法来进行多语言的主题建模和情感分析。只关注英语文本是有局限性的，因为在世界上许多地方，当地语言占主导地位。构建算法来利用来自多种语言的文本数据将是这个项目的主要贡献，之后可以应用于许多不同的领域。总之，改革对经济发展至关重要。然而，文献对理解推动改革努力的叙述和讨论给予的关注不够。本提案概述了一个研究议程，以分析这些叙事在不同国家和不同时期的动态。它将提供新的证据，说明人们对经济改革的态度是如何演变的，以及改革成功的关键因素。

Economic reforms are an essential ingredient in growth & development. Reforms resolve crises, correct inefficiencies, and improve productivity. Recent history has been characterized by several large reform episodes, such as the post-Communist transition to market economy, the economic liberalization of Latin America, and the structural reforms following the Global Financial Crisis. Nevertheless, reforms are few & far between. In addition, they are the outcome of long periods of public discussions and political negotiations. This proposal outlines a research agenda to understand the nature of these reform discussions, their evolution over time and across countries, and how they contribute to the success of eventual policy decisions. The main data source will be news articles published in each country from a variety of sources. There are three advantages to this approach: first, news is published daily, and can provide a very high-frequency measure of reform discussions. Second, since news articles are published in every country, this project can use the entire world as the full sample. Finally, large news aggregators already exist which provide access to news data since the 1980s. These sources can be searched across a variety of dimensions in order to identify reform articles.The main research questions this project seeks to address are:1. Are there systematic patterns in reform discussions in response to the business cycle? How do these differ along different institutional characteristics of countries & over time?2. Does higher intensity of reform discussions lead to faster enactment of actual reforms? 3. How does sentiment toward reforms vary over time & across countries? Does negative sentiment predict the failure of proposed reforms? 4. How have major economic crises shifted the rhetoric related to economic reforms? For example, how has the Global Financial Crisis affected attitudes toward liberalization policies? The methodology of this project will involve applying well-known machine learning & natural language processing algorithms to analyze a corpus of millions of newspaper articles relating to economic reforms. One promising approach of dimensionality reduction is dynamic topic modelling: these algorithms allow the researcher to uncover underlying topics and track the evolution of specific terms and topics over time. Although this has been applied in the economics and political science literature, no papers have used this methodology to analyze economic reform narratives. Topic modeling will provide answers to how narratives have evolved over time, particularly in the aftermath of the Great Recession.A second approach to be applied is sentiment analysis, the use of computer algorithms to understand the emotional slant of text data. Many different approaches to sentiment analysis have been proposed in the literature. For this project, studying the attitudes toward economic reforms is crucial, because the sentiments of the public and the media are often decisive for the eventual success or failure of reforms. The main methodological challenge will be designing a structured and consistent approach to conducting topic modeling and sentiment analysis for multiple languages. Focusing exclusively on text in English is a limitation, since local languages dominate in many parts of the world. Constructing algorithms to harness text data from multiple languages will be a major contribution of this project and could thereafter be applied in many different areas. In conclusion, reforms are essential for economic development. However, the literature has given insufficient attention to understanding the narratives and discussions which drive reform efforts. This proposal outlines a research agenda to analyze the dynamics of these narratives, both across countries & over time. It will provide novel evidence on how attitudes toward economic reforms have evolved, and the key ingredients to successful reforms.