The election polling error fiasco has brought into sharp focus problems with blind random surveying of the public. The political poll errors immediately prior to this election were apparently as high as 10 to 15% which is astounding.
I have long suspected some problem with the AAII polling data since it has remained mired in the 30's as the stock market rocketed up to record highs early this year and recently. Apparently the political polls under reported conservatives possibly due to people answering political polls being embarrassed to admit supporting conservatives or simply not connecting with a proper sample size of conservatives. Considering that the stock market is so closely connected to conservative philosophy, what if respondents to stock market sentiment polls are similarly hesitant to proclaim bullishness which they might fear associates them with being conservative or are similarly more difficult to connect to. If the stock market sentiment polls are in error by an amount similar to the 10 to 15% in the political polls, the 38% bulls this week in the AAII survey suddenly becomes as high as 53% in reality which makes a lot more sense given the market performance.
Rather than fudge all the bullish values by a fixed amount, what I have done to cope with this potential measurement error is to massage the data by calculating a long slow stochastic value for the sentiment readings which measures their position relative to their range over the last six months since its been more than two and half years since the AAII survey had more than 50% bulls. This results in more obvious sell signals above 80% and buy signals below 20%. This calculation can be done at Stockcharts.com using the !AIIBULL symbol and the Slow Stochastic Indicator.
I'm fully aware of the risk in too much data manipulation to force a result which makes figures lie when liars figure, but it sure seems to improve the AAII sentiment indicator's signals.
Edited by Douglas, 06 November 2020 - 06:14 AM.