The 20DMF is the market direction model of this web site. It is a sector based model that measures
- the money flow in various sectors to issue short or cash signals (Upper panel of the indicator)
- a combination of sector money flow and price trends to catch oversold bounces (Lower panel of the indicator.)
There are two versions of this indicator running in parallel:
1. The EOD 20DMF
We can see below the signals generated by the standard 20DMF:
On the Upper panel:
- Short signals are issued by a break below the 0 line on the upper panel
- Cash signals are issued by a reversal above the 0 line if the signal was in short mode
Note: A Short signal issued when the 20DMF is in a buy mode must be confirmed by the inversed ETFs TEV extension crossing above 4% (This means that traders must have taken significant positions in these ETFs to confirm a short signal.)
On the Lower Panel:
- Buy Oversold signals are issued when we have a 3 points bounce from the oversold level
- Buy 30 signal is when we cross above the +30 level after having hit the -30 level while in short mode
- Short 30 signal is when we cross below the -30 level after having hit the +30 level while in buy mode
- We have fail safe cash signals to avoid market crashed when we are long in oversold and the market decides to become even more oversold.
The Buy 30 and Short 30 signals have been introduced about two years ago to allow to generate signals even with smaller oscillations - QE has indeed generated lower volatility.
2. The RT 20DMF
There are two major improvements in the RT 20DMF.
1. The inversed ETFs are not used anymore to confirm a short signal. The RT 20DMF now uses the Cumulative Tick.
This change was introduced more than one year ago and has produced better results.
2. There is a "Forced Buy Mode" to force the 20DMF into a buy mode from short or neutral when
- The price trend is strong (defined as the 9MA>20MA>50MA)
AND
- The Cumulative Tick is above its average
This change has been introduced to deal with QE manufactured long stretches of bullish activity.
3. Pending issues regarding the 20DMF
1. Use Start of the day prices for The Track Record
I have received requests to use prices at the start of the next day after a signal was generated instead of prices available at the close. The reasoning is that this is what users can only achieve.
I tested the difference two years ago between opening and closing prices and there was none. Sometime the open price was higher than the close and at other time, it was the reverse. I however studied that issue again and I found that opening price outperformed closing prices.
This was not true before December 2012 (except at the far left of the Figure, pointed by the red arrow.) As a reminder, December 2012 is when open QE started. This probably means that funds have been toying the market, pushing it down before the close to get cheap stocks and then gap the market up on the next day knowing that QE would be in their backs.
Since I have been asked, I will switch the track record Figures to reflect the next day open, but will keep the automatic data with an EOD data because the automatic system executes overnight calculations with the latest available price, which is the close of the previous day. The track record Figures are plotted by hand, which means that it is easy to change to the next day open price.
In reality, both prices - open or close - are illusions, because even when you enter an order at the open, you can seldom get the same price as the official opening price. I therefore consider that this is a non-issue.
2. Use The Cumulative Tick as an EOD confirmation tool
In a QE environment, the set of trading rules used by the RT system are better performing than those used by the EOD 20DMF. I believe that after a one year test of these rules, we can roll them into the EOD 20DMF.
The second reason for this decision is that the inversed ETFs are not used as much as they were one or two years ago. I will post in a separate thread a study regarding inversed ETFs, but the short version is that since their use is decreasing, they are less representative of underlying activity or they have become slow to represent the underlying activity.
Below is a comparison between the two algos and the S&P500 since March 2009.
The CTick rules were introduced in early 2012, but back-tested down to 2011.
Before 2011 only one Algo was available.
We can see below that both algos outperformed the S&P500 until about end of 2012 (QE Eternity.)
From that point, the pull-backs were very quickly reversed, which resulted in most short signals failing and most buy signals being of the "forced buy" type.
I have received on a regular basis e-mails asking me to adapt the 20DMF to the new environment.
This can very easily be done by using a longer periodicity on the oscillations: if we use a 35DMF or a 50DMF we will not detect the shallow pull-backs and will stay in a buy mode. That would be fine for 2013, but then how would this indicator have handled the deeper pull-backs of 2009 to 2012?
As a matter of fact, in 2012, I tested a 50DMF and a 200DMF both did fine from 2009 (but not as well as the 20DMF,) but performed very poorly in 2008.
3. Turn a Cash signal into a Buy signal
That is probably the most tricky potential rule change. Before 2009, buying on a neutral signal would have produced a loss, while since March 2009, buying on a cash signal would have been profitable, adding about 10% as we can see in the Table below.
My position has been to say that when a Cover your short signal is issued, traders are free to act as they want: either buy or short.
4. The Forced Buy rule
This rule looks nice and is almost actionable only in a strong QE bull market.
However, if we see below the occurrences when this rule was applied, we can see that between a Forced Buy and a Short signal, the price first increases and then pulls-back. It is not obvious that this rule adds value. I believe that I should test the possibility to sell in strength when such a Forced Buy rule has been activated, so that we do not have to wait for a deep pull-back before getting out.
Conclusions:
In the next days, we will change to the Cumulative Tick rules for the EOD Algo.
More tests will be conducted related to other issues.
One key advantage of the 20DMF is that it is entirely automatic: no human interpretation.
Adaptations of course can be operated so that the indicator is more responsive to some environments without jeopardizing the indicator's reaction in less favorable environments.