Tom Demark Sell Signal–Market Top, from Steve Ellison

January 24, 2012 |

I had some difficulty finding a concise definition of the DeMark Sequential indicator. From an article written by Mr. Burke in the 1990s and other sources, I constructed a test, but there are variations in calculation and execution, and there are rumors that Mr. DeMark adds proprietary logic.

1) The method A setup for a buy (sell) occurs when there are 9 consecutive bars with closing prices lower (higher) than 4 bars earlier, and the high (low) of either the 8th or 9th bar is higher (lower) than the low (high) price of a bar at least 3 bars earlier in the sequence.

When setup is complete, begin the countdown. Count any bar in which the close is lower (higher) than the close 2 bars earlier. This time, the counted bars do not need to be consecutive.

If the price goes above (drops below) the highest high (lowest low) of the bars in the setup sequence, the setup and countdown are canceled.

If a new setup occurs in the same direction, the countdown resets to zero.

If a new setup occurs in the opposite direction, begin a new countdown and cancel the previous setup and countdown.

If the count of bars that close lower (higher) than 2 bars before reaches 13, and the 13th counted bar closes lower (higher) than the 8th counted bar, it is a buy (sell) signal. It was not clear to me what should happen if the count reached 13, but the 13th bar did not close lower (higher) than the 8th. I decided to cancel the setup and countdown in that case.

2) The test Using the above rules, I tested daily bars of the S&P 500 futures from 1982 to 1989. I checked the net change over various periods following the buy and sell signals. Results of buy signals:

Change next
Date                          12 days 18 days 28 days 42 days 63 days
1/3/1984    2.0%    0.3%   -5.2%   -3.8%   -4.8%
3/22/1984   -0.9%    0.8%    2.1%   -2.8%   -3.1%
7/20/1984    9.5%    9.2%   11.5%   11.1%   12.5%
10/20/1987   18.1%   14.5%    7.3%   14.6%   12.2%
12/4/1987   11.4%    9.1%   11.2%   11.9%   18.6%

Average                          8.0%    6.8%    5.4%    6.2%    7.1%
Median                           9.5%    9.1%    7.3%   11.1%   12.2%
Average of all periods in sample 0.5%    0.8%    1.3%    1.9%    2.9%

Results of sell signals:

Change next
Date                          12 days 18 days 28 days 42 days 63 days
10/12/1982   -0.1%    6.8%    1.6%    3.7%    9.9%
4/20/1983    4.6%    2.5%    1.2%    6.1%    5.9%
6/17/1983   -1.0%   -2.5%   -3.4%   -3.4%   -2.4%
7/11/1985   -2.3%   -3.2%   -2.5%   -4.1%   -6.6%
3/11/1987    1.9%    4.2%    1.1%   -0.2%    1.9%
1/29/1988    1.1%    2.3%    2.3%    0.0%    1.0%
3/14/1989   -0.4%    0.4%    4.1%    5.7%    8.4%
5/12/1989    1.7%    3.8%    2.2%    4.0%    8.1%

Average                          0.7%    1.8%    0.8%    1.5%    3.3%
Median                           0.5%    2.4%    1.4%    1.8%    3.9%
Average of all periods in sample 0.5%    0.8%    1.3%    1.9%    2.9%

There were only 5 buy signals in 8 years, but they worked out very well, including a buy signal on the day after the 1987 crash. The record of the sell signals was decidedly mixed. The best that can be said is that the 28-day net change was lower than average after a sell signal, although still positive. I decided 28 days was the optimal holding period and considered only 28-day net changes after later signals.

Change next
Date               28 days
11/15/1991    6.0%
11/7/1994   -1.2%
3/19/2001    6.6%
7/19/2002    8.6%
5/10/2005    4.2%
7/14/2006    4.5%
3/17/2008    9.2%
1/12/2009  -14.2%
10/4/2011   13.3%

Average             4.1%
Median              6.0%
Average of
all 28-day periods  0.6%

Sell signals since 1990:

Change next
Date              28 days
8/23/1994   -0.9%
6/19/1995    2.7%
11/8/1995    3.0%
9/16/1996    1.9%
11/4/1996    2.6%
2/18/1997   -8.3%
7/31/1997   -4.5%
3/18/1998   -0.6%
1/7/1999   -3.1%
4/23/1999   -4.4%
11/7/2001    2.2%
6/3/2003    3.3%
11/28/2003    6.0%
1/22/2004    0.6%
12/29/2004   -1.2%
9/14/2006    4.2%
6/1/2007    0.1%
4/29/2009    8.1%
9/16/2009    0.3%
12/8/2010    4.5%
2/2/2011   -1.5%

Average             0.7%
Median              0.6%
Average of
all 28-day periods  0.6%

The results of buy signals have continued to be on average very good, although also very rare. The results after sell signals appear consistent with randomness.

Anatoly Veltman writes in:

May I ask, why would 7- or alternatively 8- or alternatively 9- or alternatively 10- or alternatively 11- ….be my guest to go on forever… "work"?

I understand where you are going, but your critique will apply to all of technical analysis, and not just DM. I am not a follower of DM, but I believe that technical analysis is based on psychology. At 80F, humans can only live 9 days without water — so there is some cognitive explanation why we are counting 9 days of frustration to capitulate those who traded counter-trend, before the real counter-trend arrives.

Anatoly Veltman writes:

Now, don't ever talk "all T/A". The reason previous volume areas tend to hold the price is because people tend to transact (again) at their former prices. I've always had a beef with time counters who have no accounting for price. The Chair rightfully refers to many as charlatans; but do I understand his page-sized color-coding scheme correctly: the Bond move +0'02 = Bond move +2'00? SP change of 0.50 gets same color as SP change of 15.00??

I bet you that a system that buys a 38%, a 50%, or a 62% retracement of preceding impulsive up-wave will produce better result than a system that buys exactly the "8th declining" 5-min bar, or 15-min bar, or 30-min bar, or 60-min bar or daily bar or weekly bar. Isn't the 8th 5-min bar getting you to where the 4th 10-min bar would get you? What's the magic of counting those bars?