How Well Do Economists Forecast Stock Market Prices? A Study of the Livingston Surveys
利用1955年至1985年的61次利文斯顿调查数据，发现股市预测在统计上无偏且误差方差最小，与早期研究结论相反，并指出早期研究因算法问题可能掩盖了真实关系。
The desire to forecast stock prices appears to be eternal. Alfred Cowles (1933) was one of the first researchers to evaluate statistically how well professional forecasters predict stock prices. In this tradition, our paper will explore the statistical properties of the Livingston stock market forecasts.' A number of recent studies, including those by Josef Lakonishok (1980), Bryan Brown and Shlomo Maital (1981), and Douglas Pearce (1984), finds that the Livingston stock market forecasts are statistically biased and/or do not utilize all available information (i.e., are informationally inefficient). In contrast to earlier studies, we find, using 61 semiannual Livingston surveys from June 1955 through June 1985, that stock market forecasts are statistically unbiased and minimum error variance estimators. In Section I, we show that earlier studies are likely to mask the true relationships between stock market forecasts and realized rates of return because of inappropriate computational algorithms. In Section II, we present our statistical results based upon the corrected data algorithm. We provide a brief summary of the paper in the last section. 1. Computing Livingston Stock Market Rate of Return Forecasts