公司盈余信息、适应性学习与通货膨胀预期  

Aggregate Earnings, Adaptive Learning and Inflation Expectations

作　　者：孙坚强[1] 赵允宁 蔡玉梅 SUN Jianqiang;ZHAO Yunning;CAI Yumei(School of Economics and Commerce,South China University of Technology;School of Finance and Investment,Guangdong University of Finance)

机构地区：[1]华南理工大学经济与贸易学院,510006 [2]广东金融学院金融与投资学院,510521

出　　处：《经济研究》2019年第10期136-151,共16页Economic Research Journal

基　　金：国家社会科学基金面上项目(17BJY191)“基于公司业绩预警视角的盈利预测对通胀预期的影响机制研究”的研究成果

摘　　要：本文拓展构建增广信息的随机梯度适应性学习预期模型,从学习内容和学习信念的直接视角探讨居民公众对公司盈余等微观信息和经济产出等宏观信息的认知及学习,以及这些信息对通胀预期形成的作用。以人民银行储户问卷调查的通胀预期为样本,采用卡尔曼滤波估计模型,并以滚动的样本外均方预测误差进行嵌套检验。实证发现公众通胀预期形成过程中:(1)认知、学习和使用了公司盈余信息,但学习内容存在选择差异。重视“实现盈余”性质的加总盈余增长信息,忽视“预测盈余”性质的加总意外盈余信息,信息可靠性和获取成本是可能原因。(2)对货币政策和国际输入性因素信息重视程度和学习信念高于其他宏观信息,对货币供应和外汇变动信息的响应程度更高,且近期呈现扩大趋势。此外,尽管价格型货币政策的利率信息已经开始进入公众预期的学习信息集,但受重视程度仍远低于数量型政策的货币供应信息。Inflation expectations have long been an important concern in the macroeconomics literature. The knowledge and ability of the agent and the forecast model and information used are crucial factors in forming inflation expectations. The role of the information contained in the aggregate earnings news in forming inflation expectations has been a significant topic of interest in the recent accounting economics literature (Kothari et al., 2013;Shivakumar & Urcan, 2017;Sun et al., 2018). In particular, studies have used regressions of the inflation forecast error on earnings news to investigate whether the forecasters take earnings news into account when forming their expectations. However, this approach is based on two strong assumptions that have some limitations. First, the economic agent is assumed to have rational expectations;that is, the agent fully uses all of the available information in forecasting without making systematic mistakes. Second, it is assumed that the agent cognitively processes the information in a very short time. In this paper, a stochastic gradient adaptive learning expectation model with augmented information is developed. The agent is assumed to undergo a learning process whereby he/she acquires, updates, and learns newly arrived information in each period and then evaluates the historical prediction errors, re-estimates the model parameters, and forms an updated expectation using the modified forecast model. Starting with a general mixed Phillips curve for inflation, an information variable is integrated into the perceived law of movement and the parameter iteration process. The information variable represents the aggregate earnings or some macroeconomic news such as the output, money supply, of foreign exchange figures. The agents apply the stochastic gradient learning rule and estimate and update the forecast model parameters following a rule of constant gain (CGL) or decreasing gain (DGL). The out-of-sample mean squared forecast error on the rolling-window (R_MSE) is then used to nest t

关 键 词：通胀预期 适应性学习 公司盈余 意外盈余 

分 类 号：F822.5[经济管理—财政学] F275