The CBOE VIX measures market volatility implied in the options pricing of S&P stocks. Although designed as input for options trading, VIX is also known for its market direction implications. Very low VIX readings occur near market tops, and very high readings around market bottoms.
Here is an actual DejaVu output for VIX condition (level and change) as of March 31, 2003, with 1-month ensuing S&P outcomes. Of the 1,168 market days since 06/30/98*, only 23 days were "like" the condition prevailing at Q1/03. ("Like" is defined here by "F5" thresholds; ask us for details).
A mere 23 precedent days seems thin at first. But those 23 days led to far higher return (+3.07% vs -0.52%), with notably smaller variance (3.25% StD, vs 5.80% StD) compared with "all other" days. Altogether, the difference (qualifying vs non-qualifying days) is statistically significant.
Significance testing is critical. Statistical significance is the touchstone for determining whether an apparently attractive outcome is sufficiently large, consistent and numerous to warrant reliance. The role of the sig tests is to mediate the trade-offs among magnitude, variance and number.
DejaVu analysis verifies the general VIX effects, identifies useful thresholds, quantifies probable outcomes and estimates expected returns. When market relationships evolve (as VIX seems to be doing in 2003-04), the model readily folds new data into new inference.
The DejaVu user selects data, defines thresholds, and sets outlook horizon. The model compiles histories, profiles the outcomes, tests for significance, and computes expected returns.
* Prior to the summer of 1998, VIX and other volatility measures were notably lower than after. When indicators undergo a secular change, eras must be analyzed separately. (Extreme low VIX of early 2004 must soon be analyzed for possible new new era.) VIX levels here are based on the "old" VIX of the S&P 100.