Blog - Investing Notes
August 31, 2007 - Does the Bullish
Percent Index Predict Market Direction?
A reader asked whether the Bullish
Percent Index is a useful indicator of overall stock market or sector
direction. Does it reliably identify overbought/oversold conditions
from which stock prices are likely to revert? In a study published in
the 2005
Journal of Technical Analysis, Andrew Hyer relates the simple
average Bullish Percent across 40 stock market sectors (BPAVG) to future
broad stock market returns. Using weekly levels of BPAVG as calculated
by Dorsey, Wright
& Associates and overall stock market returns over the next
100 calendar days based on the Value
Line Geometric Index for a total sample period of 1/6/98-1/24/05
(about 368 weeks or 26 intervals of 100 calendar days), he concludes
that:
- BPAVG provides very useful information about the future performance
of the broad market over the next 100 calendar days.
- When BPAVG is below (above) 40%, average returns for the broad market
over the next 100 calendar days are positive (mostly, but not consistently,
negative). This indicator is more reliable for identifying oversold
than overbought conditions. Investors may want to increase equity
exposure when BPAVG is low.
- Future returns for the broad market are on average stronger when
BPAVG is rising.
- Low values of BPAVG usually correspond to low levels of bullishness
from American Association
of Individual Investors investor surveys.
The following table, reformatted from one in the paper, summarizes
the principal findings of the study. It shows that average returns for
the Value Line Geometric Index are generally strongest for lows levels
of BPAVG.
Is this study convincing?
The study has methodology issues that potentially invalidate its conclusions,
as follows:
- The summary table presents 369 observations, which seems a fairly
large sample. However, the number of independent measurements
of broad market returns equals the number of non-overlapping intervals
of 100 calendar days within the total sample, about 26. In other words,
the study uses the information in each of the independent 100-day
return intervals an average of 14 times. A sample of 26 is too small
to infer a reliable relationship between BPAVG levels and stock returns.
- The study methodology does not account for potential clustering
of BPAVG values over relatively short calendar intervals. For example,
if the four observations in the 10-19% range occurred in closely bunched
weeks, all four would relate to largely overlapping future 100-day
return intervals. From a statistical point of view, such overlap would
mean that information in all four observations is hardly greater than
the information in one of them. From a practical point of view, a
trader committing funds to the first signal for 100 days would not
be able to use the next three signals.
- The study offers no rationale for using a forecast interval of 100
calendar days. If the author chose that interval after testing
many others to find one with convincing results, then the conclusions
probably overstate any actual relationships due to data mining bias
(see the summary of Chapter 6 in our blog
entry of 12/11/06).
If the study were recast such that the sampling interval is comparable
to the forecast interval, its statistical conclusions would likely be
unconvincing and its economic value for systematic trading unattractive.
In summary, weaknesses in the methodology of this study substantially
undercut its conclusion that the average Bullish Percent across sectors
is very useful for predicting stock returns.
It may be that such indicators add no information to that provided
by past returns alone.
See Blog
Synthesis: Some Trading Indicators for analyses of the usefulness
of other technical indicators. See Blog
Synthesis: Sentimental Journey for analyses of various sentiment
indicators.
