The Hindenburg
Omen is a technical analysis
signal that attempts to predict a forthcoming stock market crash.
It is named after the Hindenburg disaster,
the crash of the German zeppelin of the same name in
May 1937. The Hindenburg Omen is the alignment of several technical factors
that measure the underlying condition of the stock market - specifically the NYSE
- such that the probability that a stock market crash
occurs is higher than normal, and the probability of a severe decline is quite
high. The rationale behind the indicator is that, under normal conditions,
either a substantial number of stocks establish new annual highs or a large
number set new lows - but not both. When both new highs and new lows are large,
it indicates the stock market is
undergoing a period of extreme divergence. Such divergence is not usually
conductive to future rising prices. A healthy market requires some degree of
internal uniformity, whether the direction of that uniformity is up or down
The
traditional definition of a Hindenburg Omen has five criteria:
These
measures are calculated each evening using Wall Street Journal figures for
consistency. The occurrence of all five criteria on one day is often referred
to as an unconfirmed Hindenburg Omen. A confirmed Hindenburg Omen occurs if a
second (or more) Hindenburg Omen signals occur during a 36-day period from the
first signal.
Looking
back at historical data, the probability of a move greater than 5% to the
downside after a confirmed Hindenburg Omen within the next 41 days after its occurrence
was 77%, the probability of a panic sellout was 41% and the probability of a
major stock market crash was 25%. The occurrence of a confirmed Hindenburg Omen
does not necessarily mean that the stock market will go down, although every NYSE
crash since 1985
has been preceded by a Hindenberg Omen.
Because
of the very specific and seemingly random nature of the Hindenberg
Omen criteria, it is likely that this phenomenon is simply a case of overfitting. That is,
if one backtests through a large data set and tries
enough different variables, eventually correlations are bound to be found that
don't really have any predictive significance.