The Past Performance of the Hindenburg Omen Stock Market Crash Signals 1985 - 2005
by Robert McHugh http://safehaven.com/article-3880.htm
Evolution of the Signal: Peter Eliades (www.stockmarketcycles.com) traces the origins of this potential stock market crash signal to the work of Norman Fosback, author of Stock Market Logic, back in the 1970s. Fosback did a lot of research on Highs and Lows and developed an indicator that differed from the one we have now. Credit for discovery of the Omen is given to Jim Miekka, a friend of Kennedy Gammage who wrote a report called the Sudbury Report. Kennedy, who is probably the foremost expert on the Omen, suggested to Jim that it be dubbed the Hindenburg Omen after that ill-fated dirigible doomed to crash. Perfectly appropriate name based upon our research of its past performance. Kennedy has a missive in our Guest articles section at www.technicalindicatorindex.com on the Hindenburg Omen, for those of you interested.
During the past two weeks, we have had six Hindenburg Omen signals by the traditional definition, and based upon the research I am about to share with you, I would say we have had five signals as I have added two more filters to its qualification. Still quite a significant cluster.
So what is a Hindenburg Omen? It 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. This Omen has appeared before all of the stock market crashes, or panic events, of the past 21 years. All of them. No panic sell-off occurred over the past 21 years without the presence of a Hindenburg Omen. The way Peter Eliades put it in a recent Daily Update, September 21, 2005 (Peter is well worth the read, believe me), "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 market is undergoing a period of extreme divergence — many stocks establishing new highs and many setting new lows as well. Such divergence is not usually conducive to future rising prices. A healthy market requires some semblance of internal uniformity, and it doesn't matter what direction that uniformity takes. Many new highs and very few lows is obviously bullish, but so is a great many new lows accompanied by few or no new highs. This is the condition that leads to important market bottoms."
How has this signal performed over the past 21 years, since 1985? The traditional definition of a Hindenburg Omen is that the daily number of NYSE New 52 Week Highs and the Daily number of New 52 Week Lows must both be so high as to have the lesser of the two be greater than 2.2 percent of total NYSE issues traded that day. However, this is just condition number one. The traditional definition had two more filters: That the NYSE 10 Week Moving Average is also Rising (condition # 2), and that the McClellan Oscillator is negative on that same day (condition # 3). These measures are calculated each evening using Wall Street Journal figures for consistency. Critics have taken this definition and pointed rightly to several failed Omens (although the correlation was still quite good).
But if we add two more filters, the correlation to subsequent severe stock market declines is remarkable. Condition # 4 requires that New 52 Week NYSE Highs cannot be more than twice New 52 Week Lows, however it is okay for New 52 Week Lows to be more than double New 52 Week Highs. Our research found that there were two incidences where the first three conditions existed, but New Highs were more than double New Lows, and no market decline resulted. There were no instances noted where if 52 Week Highs were more than double New Lows, while the first three conditions were met, that a severe decline followed. So condition # 4 becomes a critical defining component. The fifth condition we found important for high correlation is that for a confirmed Hindenburg Omen, in other words for it to be "official," there must be more than one signal within a 36 day period, i.e., there must be a cluster of Hindenburg Omens (defined as two or more) to substantially increase the probability of a coming stock market plunge. Our research noted seven instances over the past 21 years where - using the first four conditions - there was just one isolated Hindenburg Omen signal over a thirty-six day period. In six of the seven instances, no sharp declines followed. In only one instance did a sharp subsequent sell-off occur based upon a non-cluster single Omen, but in that case it was incredibly close to having a cluster of two Omens as the previous day's McClellan Oscillator just missed being negative. We included this instance in our data below.
So to recap, we have an unconfirmed Hindenburg Omen if the first four conditions are met, but the fifth is not - in other words we only have one signal within a 36 day period. Once a second or more Omen occurs, we then have a confirmed Hindenburg Omen signal with substantially higher odds that a subsequent stock market plunge is coming.
Our research noted that plunges can occur as soon as the next day, or as far into the future as four months. In either case, the warning is useful. It just means, if you want to play the short side after a confirmed signal, or move out of harms way, you must be prepared to see it happen as soon as the next day, or four months from now, possibly after you forgot about it. About half occurred within 41 days.
Based upon the five parameters noted above, here's what we found: Confirmed Hindenburg Omens are very rare. Excluding the confirmed Hindenburg Omen we have now, September 2005, there were only 22 confirmed Hindenburg Omen signals over the past 21 years. This is amazing when you consider that during that time span, there were roughly 5,000 trading days. Of those 5,000 trading days where it was possible to generate a Hindenburg Omen, only 160 (3.2 percent) generated one, clustering into 22 confirmed stock market crash signals.
If we define a crash as a 15% decline, of the 22 confirmed Hindenburg Omen signals, six (27.2 percent ) were followed by financial system threatening, life-as-we-know-it threatening stock market crashes. Three (13.6 percent) more were followed by stock market selling panics (10% to 14.9% declines). Three more (13.6 percent) resulted in sharp declines (8% to 9.9% drops). Five (22.7 percent) were followed by meaningful declines (5% to 7.9%), three (13.6 percent) saw mild declines (2.0%to 4.9%), and two were failures, with subsequent declines of 2.0% or less. Put another way, there is a greater than 25 percent probability that a stock market crash - the big one - will occur after we get a confirmed (more than one in a cluster) Hindenburg Omen. There is a 41 percent probability that at least a panic or crash sell-off will occur. There is a 54.5 percent probability that a sharp decline greater than 8.0 % will occur, and there is a 77.2 percent probability that a stock market decline of at least 5 percent will occur. Only one out of roughly 7.5 times will this signal fail.
All the biggies over the past 21 years were identified by this signal (as defined with our five conditions). It was present and accounted for a few weeks before the stock market crash of 1987, was there three trading days before the mini crash panic of October 1989, showed up at the start of the 1990 recession, warned about trouble a few weeks prior to the L.T.C.M and Asian crises of 1998, announced that all was not right with the world after Y2K, telling us early 2000 was going to see a precipitous decline. The Hindenburg Omen gave us a three month heads-up on 9/11, and told us we would see panic selling into an October 2002 low. And now we have another confirmed Hindenburg Omen signal, here in the autumn of 2005.
Here's the data:
Date of first
Omen Signal # of Signals
In Cluster DJIA
% Decline Time Until
9/21/2005 5 ? ?
4/13/2004 (1) 5 5.4% 30 days
6/20/2002 5 15.8% 30 days
23.9% 112 days
6/20/2001 2 25.5% 93 days
3/12/2001 4 11.4% 11 days
9/15/2000 9 12.4% 33 days
7/26/2000 3 9.0% 83 days
1/24/2000 6 34.2% 44 days
6/15/1999 2 6.7% 122 days
12/22/1998 (2) 2 0.2% 1 day
7/21/1998 (3) 1 19.7% 41 days
12/11/1997 11 5.8% 32 days
6/12/1996 3 8.8% 34 days
10/09/1995 6 1.7% 1 day
9/19/1994 7 8.2% 65 days
1/25/1994 14 9.6% 69 days
11/03/1993 3 2.1% 2 days
12/02/1991 9 3.5% 7 days
6/27/1990 17 16.3% 91 days
11/01/1989 36 5.0% 91 days
10/11/1989 2 10.0% 5 days
9/14/1987 5 38.2% 36 days
7/14/1986 9 3.6% 21 days
(1) In April 2004, the Fed pumped $155 billion in liquidity from the last week in April - right after the Hindenburg Omens were generated - to the third week of May, a 22 percent annual rate of growth in M-3, to stave off a crash. Even with the liquidity, the market still fell 5.0 percent.
(2) The 12/23/1998 signal barely qualified, as the McClellan Oscillator was barely negative at -9, and New Highs were nearly double New Lows. Had this weak signal not occurred, condition # 5 would not have been met. This skin-of-the-teeth confirmation may be why it failed. It says something for having multiple, strong confirming signals.
(3) This signal came close to having two confirming signals, which may be why as a non-cluster signal, it produced a strong sell-off.
Another point to make here is that the actual stock market declines are often greater than the measures in the prior data chart. That's because oftentimes the decline from a top has already occurred before the Hindenburg Omens have been generated. These percent declines are only measuring the declines from the first Omen in a cluster. If we measured declines from the tops, it would be worse in many cases.
Another observation is that once you get two solid Hindenburg Omens in a cluster, the probability of a severe decline does not seem to increase as more Omens occur within the cluster. Sometimes a two signal cluster produced a worse decline than a 5, 11, or 17 signal cluster. But what can be said about multiple signal clusters is that the warnings are being given further out in time, keeping us on the alert. More signals also assures us a greater likelihood of better quality signals, which seems to matter. Multiple signals are telling us things are not getting better, that something continues to remain wrong with the market.
As far as September 2005, so far, here are the five signals that meet all five of the conditions required for a potential stock market crash warning:
September 21st, 2005: The figures were 3,463 total issues traded on the NYSE Wednesday, with 136 New 52 Week Highs and 149 New 52 Week Lows. The common number of new highs and lows is 136, which is 3.93 percent of total issues traded, well above the minimum threshold of 2.2 percent. The McClellan Oscillator came in at negative 188, and the 10 week NYSE was rising.
September 22nd, 2005: We saw 102 New Highs and 165 New Lows on the NYSE, with total issues traded of 3,447. Taking the lowest common figure for Highs and Lows, 102, and dividing it by 3,447, we get 2.95 percent, well above the 2.2 percent threshold. The McClellan Oscillator was negative (-184), and the 10 week NYSE moving average was rising.
September 27th, 2005: There were 86 New NYSE Highs and 108 New NYSE Lows, on 3,463 issues traded for a 2.48 common percent, and likewise the 10 week NYSE was rising and the McClellan Oscillator was negative at -124.
September 28th, 2005: According to the Wall Street Journal online, the NYSE had 153 New Highs and 102 New Lows, from 3,433 issues traded. The common percentage is 102/3,433 = 2.97 percent. Since this ratio is above 2.2 percent, it satisfies one of five conditions we are looking for. The 10 week moving average for the NYSE was rising (it was 7,513 last week and was 7,526 Wednesday), thus condition number two was met. The McClellan Oscillator is negative, satisfying condition number three, and New Highs were not more than New Lows. This is the fourth solid Omen within the cluster.
September 29th, 2005: There were 178 NYSE New 52 Week Highs and 89 New Lows, the lowest of the two coming in at 2.60 percent of 3,417 total issues traded, above the 2.2 percent threshold. The McClellan Oscillator came in at negative -24, and the NYSE 10 Week Moving Average was rising. This is the least solid of the five signals as we almost had New highs more than twice new lows, and the McClellan Oscillator is heading toward positive territory. But it still qualifies.