Sentiment of the FOMC--论文代写范文精选

这篇essay代写范文挖掘技术应用到公开市场委员会记录，可以帮助量化这些信息来提供丰富的实时分析资源反映经济和财务分析。话语参与者选择特定主题。

Abstract

The Federal Open Market Committee (FOMC) meets eight times each year to set monetary policy. During these meetings, a changing cast of participants engages in presentations and discussions, drawing on the perspectives of research staff and community and business leaders as they formulate their views on economic conditions and determine the stance of monetary policy. Determining what the FOMC finds relevant to policy discussions and how these discussions might have changed over time can be challenging. Although the Committee releases carefully constructed statements and meeting minutes to the public, some marketwatchers have argued these pieces have only rendered proceedings more mysterious or opaque. The full transcripts offer a more complete picture of Committee meetings; however, these transcripts are only released to the public after five years. 

Furthermore, the transcripts can be somewhat difficult to parse: the texts contain a wealth of disparate information ranging from casual anecdotes to research findings to staff economic forecasts. Nevertheless, meeting transcripts offer readers the unique opportunity to examine the original expressions of individual meeting participants prior to being distilled and summarized into the statement and minutes. Applying text-mining techniques to FOMC transcripts can help quantify this information to provide a rich analytical resource reflecting real-time economic and financial analysis. The words participants choose for particular topics allow text analysts to measure the tone of the overall discussion in a way not possible in statements or minutes. In addition, researchers can measure the tone of individual speakers. Unlike the minutes, which attribute general summary discussions to unidentified “Committee members,” the transcripts identify speakers along with their contributions. 

This identification invites comparisons between individual speakers or classes of individuals such as Board members and Bank presidents. In this article, I study the tone and diction, or word choice, of the meeting participants to better understand how the discussions are formed, how they related to the performance of the economy, and how they may have changed with movements toward greater transparency. Using some fairly simple language-processing tools, I measure the tone of FOMC deliberations, explore differences across speakers, and examine how the tone of the discussions relates to a measure of economic activity. I find first that the composition and tone of the discussions have changed over time. 

More specifically, the length of comments, the uniqueness of word choice, and the measure of the tone display distinct patterns from the late 1970s through 2009. Second, I find measurable differences in the diction and tone of different classes of speakers who participate in the discussions. The contributions of Board members, for example, have a different composition and tone than that of Reserve Bank presidents or Federal Reserve System staff. Finally, I find measures of the relationship between the tone of the discussions and economic activity also show differences across time and speaker. Section I provides background information on the transcripts and the text-mining tools used to extract information. Section II calculates the tone measure for each discussion and explores how the role of individual speakers has changed over time. Section III examines the relationship between the tone measure and real economic activity and assesses what effect a move toward greater transparency in the Committee might have had.

Extracting Text from the Transcripts 
Committee discussions generate an extensive amount of text. Although the Federal Reserve Act only mandates four FOMC meetings per year, the Committee met as often as monthly up until the early 1980s and has met eight times each year since. Conference calls may also occur between scheduled meetings. In addition, the number of meeting attendees contributing to the deliberations can add significantly to the text. The Committee comprises all sitting members of the Federal Reserve Board of Governors—usually seven but at times as few as four—as well as five Reserve Bank Presidents who serve on the Committee on a rotating basis. Reserve Bank presidents who are not voting members of the Committee attend and participate in all meetings as do staff members from Reserve Banks and the Board. The meetings are closed to the public, but the Committee releases an official statement at the close of the meeting to convey its monetary policy decision. Minutes from the meeting are available several weeks later, and the Committee releases full transcripts of the discussion with a five-year lag.1 

Not all of these communication pieces may be suitable for text analysis. The official statements, for example, are perhaps too carefully crafted, as the media and market participants vigilantly parse them. Indeed, The Wall Street Journal dedicates a column to outlining changes in the wording of FOMC statements from meeting to meeting. The transcripts, on the other hand, are ideal for text analysis, as they capture each part of the meeting from roll call to parliamentary procedures for policy votes. The transcripts include the entire discussion, indicating who was speaking and what was said with little editing except for the potential removal of “a very small amount of information received on a confidential basis from, or about, foreign officials, businesses, and persons that are identified or identifiable” (Board of Governors 2014). They show how Reserve Bank Presidents provide important regional context and information, how Governors voice opinions or ask questions, and how Board staff present information on economic output and other relevant topics. Such detail makes the text of the transcripts an excellent source of information to be mined.

Text mining 
Text mining creates structured data out of unstructured data, allowing a quantitative analysis of qualitative information. Traditional methods of assessing relationships and patterns in data deal exclusively with structured data—numeric information generally well-formatted in tables or databases. However, much of the data created or captured today is far less structured or in many cases unstructured, such as the text of tweets, blog posts, emails, or documents. Analyzing such inputs first requires transforming them from the raw data format to a format that can effectively use methods identical or analogous to those used to analyze structured numeric data. 

Researchers can mine text using different methods, each suitable for answering a different set of questions. More specifically, new research in this field applies a variety of methods to FOMC documents.2 I focus here on the specific words FOMC participants choose during their discussions. Note that studying the words used is different than examining the topics discussed. The former is more closely aligned with expression, the latter with content. As expression and word usage relate more to how ideas are conveyed than to the ideas themselves, they are a more appropriate way to address sentiment in a document like the FOMC transcripts. To assess how someone feels, examining their actual choice of words may be more instructive than attempting to attach a sentiment to a particular topic. Much of the current work on sentiment analysis focuses on consumer opinions expressed in tweets, online reviews, and other social media outlets. I apply similar techniques here with some changes to acknowledge the important differences between social media posts and monetary policy discussions.

Processing the transcripts 
Some written records of the Committee’s meetings are available from the Federal Reserve Board from as early as 1936. I start the sample with 1977, as this is the first year for which records are identified as transcripts. First, I extract the text from the digital file, parse it into words based on spaces and punctuation, and remove the preliminary Committee procedures (for example, roll call). I then group the text pieces into individual comments by speaker. For each named speaker, I collect the text of that person’s comment until the next speaker is identified. For some entries, this text is as short or simple as “yes” or “thank you”; for others, a speaker giving a presentation or answering a question at length can have a single comment that runs for pages. I apply the extraction method to 362 complete transcripts and five partial transcripts over 33 years, yielding 114,912 individual comments.3 I

While many pre-processing options are available in different textmining applications, I choose to minimally pre-process the text. Removing stop words, for example, is helpful for exercises involving word counts or relative frequencies but may not be helpful for other analyses. In addition, I have chosen not to weight the words when calculating the sentiment measure. None of the commonly used weighting schemes is an obvious choice for this exercise, and though some evidence suggests weights can help decrease the noise in certain measures, it is not clear they would improve this analysis. Simply counting the number of comments in the transcripts highlights changes in the nature of the FOMC’s discussion over time. Chart 2 shows the average number of comments made by each speaker each year during the meetings or conference calls that occurred that year. 

The number of individual contributions per speaker has varied greatly over time with a distinct downward trend through about 2005 and a steady upward trend since then. There does not appear to be any clear cyclical pattern. But fewer comments do not mean less discourse. Indeed, while the number of comments decreased, their average length increased. Chart 3 shows the number of words per comment appears to have increased significantly from 1993 to 2005. While the number of comments has decreased from its peak in 2005, it is still significantly higher than in the early years of the sample.

51Due网站原创范文除特殊说明外一切图文著作权归51Due所有；未经51Due官方授权谢绝任何用途转载或刊发于媒体。如发生侵犯著作权现象，51Due保留一切法律追诉权。
更多essay代写范文欢迎访问我们主页 www.51due.com 当然有essay代写需求可以和我们24小时在线客服 QQ:800020041 联系交流。-X