Research Data for Consumer Sentiment

This data set contains the data used in the research project "Cognitive Biases in Consumer Sentiment: the Peak-End Rule and Herding". The following files and items are includedICSdata.xlsx: Index of Consumer Sentiment and its constituents (sheet 1), and PAGO per region (sheet 2); original source University of Michigan, Survey of Consumers, https://data.sca.isr.umich.edu/ALFRED_data: macro economic series related to economic growth, inflation, (un)employment and consumption, including publication date; original source ArchivaL Federal Reserve Economic Data (ALFRED), https://alfred.stlouisfed.org/; for each series a README sheet is included with metadataFREDdata: financial and economic series related to stock, bond, housing markets, interest rates,gasoline prices and regional unemployment rates; each sheet contains the mnemonic of the donwloaded series.MicroData_20220113: demographic information of each respondent in the Survey of Consumers conducted by the University of Michigan; downloaded from University of Michigan, Survey of Consumers, https://data.sca.isr.umich.edu/Prelim_PA.xlsx: the Index of Consumer Sentiment and its constituent series, as reported in the preliminary annoucement by the University of Michigan (prelim), and the series constructed based on the surveys after the preliminary announcements. The prelim series are publicly available via https://data.sca.isr.umich.edu/ . The pa series have been constructed based on interview datas obtains from the University of Michigan. These data are proprietory and cannot be shared freely.DemographicDifferences.xlsx: average differences between the prelim and pa monthly subsample in the demographic statistics available in MicroData_20220113.xlsx. The difference have been constructed based on interview datas obtains from the University of Michigan. These data are proprietory and cannot be shared freely.<b>Methodology: </b>Linear regressions and time-series methods.<b>Findings</b>: We show that two heuristics, the peak-end rule and herding, generate biases in indexes of consumer sentiment. Both affect respondents' assessment of changes in their financial position over the past year. Conform the peak-end rule, their answers relate more to extreme detrimental monthly than to yearly changes in key financial and macro variables. These effects are stronger for more salient variables. As for herding, we document that respondents interviewed in the second round about past financial changes rely too strongly on future expectations from first-round respondents. These effects persist when we account for structural differences in sample composition or for the effect of other predictive variables. Our research shows the presence of both biases outside controlled environments and sheds new light on the relevance of sentiment indexes.

本数据集收录了研究项目《消费者情绪中的认知偏差：峰终定律与从众效应》的全部研究数据，具体文件及内容如下：
ICSdata.xlsx：包含消费者情绪指数及其构成分项（工作表1），以及各地区PAGO指标（工作表2）；原始数据来源于密歇根大学消费者调查（Survey of Consumers），获取地址：https://data.sca.isr.umich.edu/
ALFRED_data：包含与经济增长、通胀、（非）就业及消费相关的宏观经济序列，含发布日期；原始数据来源于联邦储备经济数据档案库（ArchivaL Federal Reserve Economic Data, ALFRED），获取地址：https://alfred.stlouisfed.org/；每个序列均附带包含元数据的README工作表。
FREDdata：包含与股票、债券、住房市场、利率、汽油价格及地区失业率相关的金融与经济序列；每个工作表均标注了所下载序列的助记符（mnemonic）。
MicroData_20220113：包含密歇根大学消费者调查中每位受访者的人口统计信息；数据来源于密歇根大学消费者调查，获取地址：https://data.sca.isr.umich.edu/
Prelim_PA.xlsx：包含密歇根大学发布的初步公告中披露的消费者情绪指数及其构成序列（初步版，prelim），以及基于初步公告后开展的调查构建的相关序列。其中初步版序列可通过https://data.sca.isr.umich.edu/ 公开获取；PA序列基于密歇根大学提供的访谈数据构建，此类数据为专有数据，不得随意共享。
DemographicDifferences.xlsx：包含基于MicroData_20220113.xlsx中的人口统计数据，计算得到的月度子样本中初步版与PA版序列的平均差异。该差异基于密歇根大学提供的访谈数据构建，此类数据为专有数据，不得随意共享。
<b>研究方法：</b> 线性回归与时间序列分析方法。
<b>研究发现：</b> 我们证实，峰终定律与从众效应这两种认知启发式会引发消费者情绪指数的偏差。二者均会影响受访者对其过去一年财务状况变化的评估。符合峰终定律的特征体现为：受访者的回答更多关联关键金融与宏观变量的极端负面月度变化，而非年度变化；且对于显著性更强的变量，该效应更为显著。关于从众效应，我们发现，在第二轮访谈中被问及过往财务变化的受访者，会过度依赖第一轮受访者的未来预期。在控制样本构成的结构性差异或其他预测变量的影响后，上述效应依然显著。本研究证实了两种偏差在非受控环境下均真实存在，并为消费者情绪指数的研究价值提供了新的视角。