In his excellent article on sentiment analysis based on American Association of Individual Investors (AAII) sentiment surveys, Charles Rotbult segmented market’s six and twelve month forward returns based on various levels of bullish and bearish sentiment readings. In the article, Mr. Rotbult defined excessive optimism/pessimism based on the deviation of bullish/bearish readings from their means. Mr. Rotbult’s analysis showed that extraordinarily low levels of bullish readings, defined as below two standard deviations from the mean, to be a good contrarian indicator with average 6-month return of 14.0% and average 12-month return of 20.7%. Further, market returns were positive in each one of the six and twelve month periods. However, predictability wasn’t as good when sentiment was extremely high defined as being three standard deviations above the mean. Also, bearish sentiment readings had a mixed record at predicting subsequent market returns.
In this post, I present a model for AAII sentiment analysis and analyze results following excessive sentiment readings. The model that I use to perform sentiment analysis is based on bullish readings divided by the total of bullish and bearish readings. Let’s call this ratio as the AAII ratio. To identify bouts of excessive optimism and excessive pessimism, I use the deviation from mean with two standard deviation readings on either side considered to be extremes. As against using the full period data for calculating mean and standard deviation bands, I use four-year moving averages and four-year moving standard deviations. This is done to avoid any look-ahead bias from seeping into the model.
In general, when using mean-reversion models like these, I like to see some evidence of the mean-reversion process getting a start, i.e., as against jumping right in front of a speeding freight train, I like to see some evidence that the brakes are coming into action. For the purposes of this analysis, I define a sentiment sell signal as the AAII ratio climbing above +2SD (two standard deviation above the mean) and then dropping back below +2SD level. A buy signal is generated with the AAII ratio dropping below -2SD (two standard deviation below the mean) and then climbing back above the -2SD level. In addition, I apply some continuation/reversal rules such that it is always a switching signal, i.e., the model is always on a buy or sell signal.
Figure 1 shows the AAII ratio historically along with the mean and standard deviation bands around the mean.
Figure 2 plots the S&P 500 index (right axis) along with buy and sell signals based on the AAII ratio model. Clearly, the ability of these signals to indicate good/poor market environments is somewhat mixed.
The table below summarizes the signal’s performance. The numbers below suggest that AAII ratio based sentiment analysis model has some value in identifying constructive market environments with average annualized return of 13.5% during buy signals. Over the last 24 years, this model was in the market 61% of the time. The rest of the time, when sell signals were in force, market’s annualized returns averaged 4.5%.
|Returns||% of Time in Market|
|Average Annualized Return: Buy Signal||13.49%||61%|
|Average Annualized Return: Sell Signal||4.46%||39%|
In summary, the AAII model discussed here has some value as a market timing tool. However, it is not a tool to be used without regard to other factors.
 AAII has been surveying its members since 1987 about the direction of equity markets over the next six months. Member responses are classified in three categories: bullish, neutral, and bearish. Results of the survey are published every Thursday morning by AAII and is made available on this page. AAII’s member survey is a widely followed sentiment measure and is used by a multitude of market participants as a market sentiment analysis tool. The good folks over at AAII offer the historical data for download from the sentiment survey page. You can click here to download historical AAII survey results in an excel file.