Thursday, August 20, 2009

Distribution Days Quantified

I'm out of action most of this week and unable to post much. Therefore I decided to take an excerpt from this past weekend's Subscriber Letter and post it to the blog:

One sign of a potential top that some traders monitor is distribution days. It was popularized by Investors Business Daily. The essential idea is that when the market falls on increasing volume that suggests institutions are selling stock. When clusters of distribution days occur, it is a topping signal. Below are some quotes from an IBD column published on August 4th that discusses distribution.

“What you're looking for is distribution. If one or more of the major indexes (the NYSE composite, the Nasdaq, the S&P 500 or Dow industrials) falls more than 0.2% in higher volume than in the prior session, that's a distribution day.

Distribution means the big money — mutual funds, investment banks and other institutional investors — is dumping shares. That's bad news for the little guy, because institutions make up roughly three-fourths of the market and chart its direction.

IBD studies show that when you get a series of three to five distribution days over a few weeks during an uptrend, that's a red flag.”

“Identifying distribution days is crucial: If you don't, you might have the wrong take on the market's direction. Then you'll be wrong about every move you make. That's a nice recipe for financial agony…Once distribution days pile up, it's wise to scale back your portfolio. Ease off margin, and get rid of any laggard stocks first. Raise cash and move entirely off stocks if necessary.”

So the bottom line is that if the market rallying, and you see a cluster of distribution days occur within a fairly short time period, you should begin selling stocks. The market is likely heading for a tumble. Let’s take a quantified look at it.

First, before I show test results I will say that clusters of distribution days do often occur near market tops – so they got that part right. But are they predictive of a top and should they be used for purposes of early identification?

I actually devised this test 4 years ago when I wrote an article for TradingMarkets about distribution days. To test the concept I looked for the following criteria:
1) The S&P closes above the 200-day moving average (remember – we’re looking for a top.)
2) Sometime within the last 12 days the S&P closed within 1% of its 200-day high. (Again, confirming we are near a top.)
3) Over the last 12 days there have been at least 4 distribution days.

Looking for 4 distribution days within 12 trading days was the criteria I used in the original test based on information at that time. It still seems like a good number according to the above article that suggests “a series of three to five distribution days over a few weeks during an uptrend, that's a red flag.”

So if you wanted to use this red flag as a short signal, how would you do looking out over the next 1, 2, and 3 month periods?
(click to enlarge)

Needless to say these results are horrible. It appears that following a bout of distribution is NOT a good time to be selling. What if we flip the study on its head and instead BUY after such instances when distribution day counters are unloading positions?
(click to enlarge)

Not the most explosive results I’ve ever posted in term of average trade, but a decent edge nonetheless. Wins are bigger than losses and the winning percentage is pretty good. This makes for decent looking profit factors (gross gains / gross losses = profit factor).

Is this a new phenomenon? Did distribution day counting formerly work and in recent years it has failed? That might explain why IBD has discussed it for so long. Sadly, no. Below is the equity curve for the 1st test above using a 20-day holding period.

While the results have been helped out by some horrible bear markets in the last decade, it’s never been a winning concept.

So why preach it? Well, it’s rare that you’ll get a top without a bout of distribution days. Therefore, when a top actually does occur, the service or person who talks about their importance can point to the top and say “See, the distribution days signaled it. You would’ve been fine if you’d just used this tool.”

This is somewhat similar to the perception that has been created with regards to follow through days for calling market bottoms. They occur there, but they are not predictive and are pretty much a worthless tool. For detail on follow through days, you may refer to the series I wrote last year. The primary difference here is that while follow through days are generally worthless, counting distribution days to try and identify tops is worse than that – it’s hazardous.

The bottom line reality of distribution days is that when the market endures a pullback after an extended uptrend, it’s often a buying opportunity and NOT a time to sell.

Of course there are still reasons to be cautious here, but of the things I am seeing that are of concern, distribution days are at the bottom of the list.

Since I’m out of action most of this week and I’m sure this post will be viewed as controversial in the eyes of some, I’ve decided to give away the code. Rather than debate or re-run the tests different ways (I’ve already done that myself), Tradestation users may feel free to download the code themselves from the free download page on the website. Login is required. Sign-up is free and only requires a name and email address.


Unknown said...

Rob, twelve days is too short. I'm pretty sure that IBD uses a month time frame which would be about 22 trading days and in the past, I think they usually use a count of 5 distribution days.

JimmyC said...

gingersue, then IBD needs to be more clearer than the "series of three to five distribution days over a few weeks during an uptrend." What is "a few weeks?" It's vague and open for different interpretation. yes, 22 days is about a month then IBD should state such a simple time frame.

Unknown said...

Back in June, IBD said this..

"If you see three to five distribution days accumulate on any one index within four to five weeks, chances are good the market is set to take a tumble.

It's also important when distribution days fade from that time period or when the indexes keep rallying past them. These can be positive signs for the market.

You can keep track of distribution days by reading the Market Pulse, published daily alongside IBD's Big Picture. The column analyzes each day's market action.

If you track the Market Pulse, you'll notice that distribution days occasionally disappear.

Readers following the Market Pulse on May 6-8, for example, may have noticed a drop in such days.

In the May 5 edition, the Pulse put the Dow at four distribution days, the other indexes at one or two each.

The next day, the Dow had dropped to three distribution days.

What gives? The March 30 distribution day was older than five weeks, and therefore no longer relevant to the current market action.

Remember, you're looking for distribution days occurring in a four-to-five-week span. Older distribution days shouldn't count.

Also, if an index climbs significantly above a distribution day's close — about 5% — it also loses its significance.

The May 6 Big Picture explained that the Dow's then most-recent distribution day, logged March 30, had dropped away and was no longer relevant.

This is because 25 market sessions had passed — the standard shelf life for distribution days.

That same week, other distribution days became irrelevant because of distance — the market had risen, leaving the distribution days in the dust.

The May 8 Market Pulse showed that the Dow, Nasdaq and S&P 500 had all added a distribution day the previous session.

The NYSE composite, on the other hand, went from two distribution days to one.

Here's the explanation:

All the indexes slumped in higher volume in the May 7 session, logging distribution days all around.

Despite that, the NYSE composite had already risen more than 5% above its distribution days on April 14 and 22.

That wiped those two demerits off its record, leaving it with just the single distribution day chalked up in the March 7 session.""

Unknown said...

I've been observing IBD's market calls for a few years now via the big picture, and they have beaten any other method I am familiar with over the same time frame. (and I am familiar with numerous ones).

I think this backtest really misses the mark, and I think the biggest flaw in the test is the 200 day MA, and 1% from the high within 12 days criteria. IBD called for corrections in Dec 07, end of May 08, end of Aug 08, Jan 09. They called the rally beginning March 09, and short of a couple whipsaws since, have been bullish the whole way up. Show me another system with this track record, because I would like to know what it is.

I like this blog, but this analysis has really missed by a mile.