Purifying Stock Market Sentiment Indicators

It is arguable that sentiment indicators derive substantially from what just happened in the stock market and that they therefore add little or no value to price action itself in predicting future returns. In their May 2009 paper entitled “Purified Sentiment Indicators for the Stock Market”, David Aronson and John Wolberg investigate this thesis by removing the influence of recent stock market price dynamics (defined by 18 variations of price velocity, acceleration and volatility) to produce multiple “purified” versions of each of five sentiment indicators: (1) the CBOE Implied Volatility Index (VIX); (2), the CBOE Equity Put-to-Call Ratio (PCR); (3) the American Association of Individual Investors Bulls minus Bears (AAII); (4) the Investors Intelligence Bulls minus and Bears (INV); and, (5) Hulbert’s Stock Newsletter Sentiment Index (HUL). They then measure the power of the purified sentiment indicators to generate profitable trading signals by testing 100 signaling rules for each indicator. Using data for the five sentiment indicators from initial availability (ranging from January 1963 to July 1987) through October 2008, along with contemporaneous daily closes of the S&P 500 index, they conclude that:

• Since January 1990, specific past price dynamics (the most predictive) predict all five raw sentiment indicators, with R-squared statistics ranging from from 0.27 to 0.70. The most predictive past price dynamic varies from indicator to indicator and somewhat overstates dependence of the sentiment indicators on past price action.
• Poll-based sentiment indicators (AAII, INV and HUL) are especially sensitive to the recent simple stock market trend (velocity).
• Purification generally reduces the drift and stabilizes the volatility of the sentiment indicators, suggesting greater usefulness of purified indicators for fixed threshold trade signaling.
• However, purification reduces the profit factor (ratio of gains from profitable signals to losses from unprofitable signals) of trade signals for four of five sentiment indicators (all but VIX), bringing into question whether these indicators augment the inherent predictive power of price dynamics. Profit factor analysis on past price dynamics alone suggests that the poll-based AAII, INV and HUL are proxies for simple price trend (velocity) and that AAII, INV, HUL and PCR add minimal value to price-based indicators.
• In contrast, purification significantly enhances profit factors for VIX signals, implying that VIX contains predictive information above and beyond price dynamics but that past price dynamics mask this information. Purified VIX is superior to all other indicators tested. (See the charts below.)

The following charts, taken from the paper, compare the profit factors for 50 long and 50 short trading signals based on raw VIX and purified VIX (and benchmark random signals). Profit factor values appear above the bars, with significance at the 0.05 level relative to the random signal benchmark indicated by asterisks and boxes. Results show that: (1) most profit factors for raw VIX signals are not significantly different from those for random signals; and, (2) profit factors for purified VIX signals are significantly stronger than those for random signals.

In summary, evidence indicates that many sentiment indicators add little or no value to simple price action indicators, but VIX purified of price action contains significant predictive power for future stock market returns. However, price action masks the predictive information contained in raw VIX.

A reader asked: ” How do you ‘purify’ the VIX of price action?”

Within the source paper, subsection “B. Method Used to Derive Purified Sentiment Indicators” (pages 6-9) provides a conceptual description of the regression-based method use to purify various measures of investor sentiment. Essentially, the method uses regressions of each investor sentiment measure (such as VIX) versus 18 “price dynamics indicators” to isolate variations in the sentiment measure that past price action does not explain. This unexplained residual is the purified sentiment measure. See the paper for additional description of the method and identification of the 18 indicators.

You May Also Enjoy...

• De-Snooping Market Timing Rules Based on Fundamental and Sentiment Indicators
• Comparing German and American Investor Sentiment Indicators
• Monthly News Sentiment Predicts Stock Market Returns?
• When Market Sentiment Works
• An International Test of Common Stock Return Indicators