Subscribe to RSS feeds Bulkowski Blog via RSS

Support this site! Clicking the links (below) takes you to If you buy ANYTHING, they pay for the referral. My books...

Encyclopedia of Chart Patterns 2nd Edition book.
Chart Patterns: After the Buy
Getting Started in Chart Patterns, Second Edition book.
Trading Basics: Evolution of a Trader book.
Fundamental Analysis and Position Trading: Evolution of a Trader book.
Swing and Day Trading: Evolution of a Trader book.
Visual Guide to Chart Patterns book.
Picture of Bumper.
Picture of the head's law.

Bulkowski's Distribution Days

Class Elliott Wave Fundamentals Psychology Quiz Studies Setups Software Tutorials More...
Candles Chart
Small Patterns
As of 02/20/2019
  Industrials: 25,954 +63.12 +0.2%
  Transports: 10,627 +9.81 +0.1%
  Utilities: 746 +2.71 +0.4%
  Nasdaq: 7,489 +2.30 +0.0%
  S&P 500: 2,785 +4.94 +0.2%
Tom's Targets    Overview: 02/14/2019
26,000 or 24,600 by 03/01/2019
10,900 or 9,900 by 03/01/2019
755 or 725 by 03/01/2019
7,700 or 7,050 by 03/01/2019
2,825 or 2,650 by 03/01/2019

Written by and copyright © 2005-2019 by Thomas N. Bulkowski. All rights reserved. Disclaimer: You alone are responsible for your investment decisions. See Privacy/Disclaimer for more information. Some pattern names are the registered trademarks of their respective owners.

This article discusses distribution days, what they are and how they perform.


Distribution Day Summary

A cluster of distribution days within 21 calendar days during a rising price trend is supposed to predict a price drop. It doesn't. Clusters of distribution days only work in a falling price trend, regardless of whether volume is heavy or light. This finding applies to individual stocks as well as the S&P 500 index.

Rob Hanna found the same thing. In his blog, he writes,

Needless to say these [Hanna's] results are horrible. It appears that following a bout of distribution is NOT a good time to be selling. ...Is this a new phenomenon? Did distribution day counting formerly work and in recent years it has failed? That might explain why IBD [Investor's Business Daily] has discussed it for so long. Sadly, no. While the results have been helped out by some horrible bear markets in the last decade, it's never been a winning concept.

Distribution Day Background

A SFO magazine article in the May 2007 issue by Kate Stalter titled, Stock picking: the formula for success, prompted this research. She wrote,

Look for a day when an index sells off in heavier volume than in the previous session. That's known as a distribution day...Investor's Business Daily's research into past corrections and downtrends shows that five or six distribution days over a period of about four weeks can be a signal that the market's uptrend is weakening.

Another source said that a distribution day means a major index closes lower more than 0.2% on higher volume 4 or 5 times within 2 or 3 weeks.


Distribution Day Methodology

I used 568 stocks from January 1, 2005 to July 1, 2010 in the test. I chose those dates because they were recent and because they nearly filled my spreadsheet with data (usually over 50,000 samples). I pulled the stocks from the list of those I follow on a daily basis. Stocks with closing prices below $5 were excluded from the test. I also threw out distribution days if the stock did not drop by more than 0.2%, so that it matched the Investor's Business Daily implementation.

I ran 16 different tests to uncover the goods on distribution days, mostly on individual stocks, but on the S&P 500 index as well. I started with two time periods, 14 days and 21 days (both calendar days, not trading days) for the window size where a cluster of distribution days should occur. For "heavy" volume, I tried a 21-trading day (about a month) moving average of volume and compared the distribution day to see if it was above one, two, and four times the average volume. Since this did not match the intent of the Investor's Business Daily implementation, I discarded the results.

I focused on a 21-calendar day window for a cluster of distribution days to appear, with volume either above or below the prior day's volume, and price trending higher or lower before the clusters began. Those match the parameters that others use.

Since Stalter's magazine article references an uptrend, I used a 10-trading day exponential moving average to determine the trend (up or down) and also conducted tests where I ignored the price trend.

For the distribution day count, which ranges from four to six times in the references cited above, I just let the median cluster count determine the number. Why? Because a number too high means the sample count drops. Few samples can be unreliable. If the technique really works, then it should appear in counts above versus below the median cluster count.

For each test, I measured how far price moved over time (from 1 week to 3 months) after the cluster of distribution days ended. Then I compared the moves for stocks having above or below the median number of clusters and the resulting performance over time.

For the S&P 500, I started from January 1950 and ran it to July 1, 2010 using the same methods.


Distribution Day: Methodology Example

Picture of my dog.

To make this methodology clear, let me give you an example.

On January 21, 2005, 3M (MMM) had a distribution day occur but it was the only one within a 3-week window, giving a cluster count of 1. As another example, on April 5, 2005, 3M had 4 distribution days that occurred within a 3-week window ending on that date.

I let the program continue finding distribution days and counting how often they appeared within a 3-week window. The program found over 58,800 samples. The median number of distribution days was 3. That means half the samples had 3 or fewer distribution days within a 3-week window and the other half had more than 3 clusters within 21 days.

When I compared the performance, I found that a week after the cluster of distribution days ended, price was higher by 1% for those with 1-3 distribution days compared to a drop of 1% for those with more than 3 distribution days. I repeated the comparison for weeks 2, 3, and so on up to 3 months later.

Once I had the performance numbers for those above and below the median cluster count, I counted the number of times a distribution day above the median showed a larger drop over time than did those below the median, in a series of competitions. The results of those competitions appears below.


Distribution Day Results

The following table shows the results for stocks in a series of competitions as explained in the prior section.

VolumePrice Trend1 wk2 wks3 wks1 mo2 mos3 mos

For example, if volume was heavier than the prior day and price was trending higher in the prior 10 trading days, none of the contests with above median distribution days had price drop further during the test periods from 1 week to 3 months later.

The next row down shows that if volume on the distribution day is higher than the prior day and price trends lower leading to the distribution day, price drops more if there is a cluster of distribution days than if there is not a cluster.

Comparing the volume column and the results, we see that volume is irrelevant to performance. If price trends down on heavy or light volume, price drops further after more distribution days.

The results show that the idea behind a distribution day is wrong on two counts. 1) Price only drops after a cluster of distribution days if price is trending lower, and 2) price drops further even after a light volume day.

You can download the summary spreadsheet of the tests by clicking on the link. It loads a 34k Excel spreadsheet.

Distribution Day: S&P Results

The results for the S&P 500 index is the same as that shown in the table above except for one cell: heavy volume, downward price trend for 1 week shows 0% instead of 100%. In other words, distribution days for the S&P 500 index don't work as advertised either.

-- Thomas Bulkowski


See Also

Written by and copyright © 2005-2019 by Thomas N. Bulkowski. All rights reserved. Disclaimer: You alone are responsible for your investment decisions. See Privacy/Disclaimer for more information. Some pattern names are the registered trademarks of their respective owners. Cows are machines which make grass fit for people to eat.