Untitled

On the other hand, the size of the projected weekly countertrend or
new trend move was substantial and very tradable. There still remained
the problem of being able to predict when the divergence would actually
create the resulting change in trend. Nonetheless, a weekly divergence is
far more tradable than a daily divergence.

It is a common belief that divergences do predict changes in trend,
even if only a short countertrend move. But it is important to realize that
you can’t simply buy or sell when the divergence occurs. In virtually all
cases, the market continues its trend for several more bars before suc-
cumbing to the power of the divergence. Many traders point out how great
divergences called highs and lows but few, if any, argue that the divergence
actually lasted for some period of time before the price turned direction.
Let’s look at a couple of examples.

In Figure 5.3, we see the price make a signiﬁcant low about nine bars
from the left (mid-December). The stochastics also make a low. However,
six bars later, the market makes a new low close while the stochastics
have not.

Note that I said a new low close, not a new low. That is because the
relationship of the closes to the closes is the important factor, not the rela-
tionship of the lows to the lows. This is because stochastics are based on
the closing price.

In this case, there was stochastic divergence but the price did notmake
a new signiﬁcant low for three more bars. In our study, the divergence
came three days in advance of the actual bottom. We then measure the
distance from the actual bottom to the subsequent high to get an idea of
the magnitude of the countertrend move. We measure this as a percent of
the price of the underlying instrument to normalize the study over many
futures contracts. For example, a move from 100-00 to 101-00 on the bonds
would equal 1 percent. Obviously, this is not perfect, but it is still indicative
of the magnitude when tested over 30 futures contracts.
Look at Figure 5.3 again and you will see several other divergences.
The most signiﬁcant one is the one about three-quarters through the chart.
The market leaps to a new high with a very wide range bar but the stochas-
tics do not make new highs. Eventually, a high is made and the market sells
off but not before mounting a very strong rally.
In our study, we went back over one year of data on about 30 different
commodities. These commodities were in various types of markets: bull,
bear, and chop city. As a result, we get a good indication of the validity of
using stochastic divergences.

First, we tested divergences on daily charts. We found 58 different
divergences. They led a turn in the market by an average of 4.5 days with
a standard deviation of 5.7. However, the mode was only 1 and the median


was 2. The longest lead time was 22 days. As you can see, the lead time
of daily divergences was random. This tends to reduce the value of daily
divergences.
There were obviously 58 retracements from these 58 divergences. The
average percentage retracement was 5.9 percent in a range of essentially
0 to a 21 percent retracement. The mode was 4 percent, and the median
was 5 percent; the reliability of the retracement ﬁgures seems reasonable
even though the standard deviation was 4.4 percent.
There were a lot more weekly divergences than daily divergences. We
counted 124 during our sample period. They led turning points by an av-
erage of 2.8 weeks with a standard deviation of 3.4 weeks. The range was
from calling the turning point exactly to 16 weeks in advance. Actually, the
modewas calling the exact turning point. In otherwords, themost common
occurrence of a weekly divergence was that week was the actual turning
point, which is very interesting to me.
The average retracement of the 124 retracementswas 15.3 percentwith
a standard deviation of 13.6 percent. Obviously, this is a far more powerful
indicator of subsequent price movements than the daily divergences. The
mode was, however, only 4 percent with a median of 11 percent though the
maximum was 80 percent. Talk about a big move!
I’m about to say something that is astounding and perhaps heretical:
It could be that divergences are simply illusions. Perhaps we are creating
a pattern where none exists. For example, I could create a system that
predicted a countertrend move soon after every day named Tuesday. Well,
sure enough, this system will call 98 percent of every countertrend move
within two weeks. In other words, only rarely will a market go two full
weeks without retracing at least one day. Be aware of this potential prob-
lem. Having said that, I continue to use stochastic divergences on the daily
and particularly on the weekly charts because I am making money using
them.

A failure occurs when the %K changes direction, doesn’t cross the %D,
and reverses back to the original direction. Figure 5.4 shows two different
failures. Arrows in the stochastic part of the chart show the failures. The
ﬁrst one occurs in early March when the %K actually drops but does not
actually cross the %D. The second one occurs in late April. This is not a
classic failure because it doesn’t actually drop but it doesn’t sharply slow
and almost cross the %D. They often occur when there is a sharp retrace-
ment against the main trend but then an equally sharp resumption of the
main trend.
The classic interpretation is that failure shows that the original trend
will continue and that you should trade in that direction. In this case, that
means jumping in the direction of the trend as soon as the %K resumes its
original trend.