|
PureBytes Links
Trading Reference Links
|
On Jun 5, 3:05pm, Jim Johnson wrote:
> Larry Williams (and I believe Tom DeMark) subscribes to notion that as
> prices rise closes tend to be nearer the top of the bar range. But he
> looks for this pattern near the end of a leg--context seems to matter.
> I am about to explore this so I'm starting with following indicator. I
> call it Position In Range Indicator (PIRI). I will be looking for
> ocsillations that track swing hights and lows. I will also look to
> couple it with classical reversal formations. I see it as just another
> hint at a top rather than a stand alone indicator.
The following analysis may offer some ideas on how the position in
range on one day relates to the magnitude and direction of close the
next day. I looked at SPY data (from Yahoo!/CSI) for the period
01/29/93 through 3/12/99 (1544 total samples). I created a dataset
with (x,y) pairs where x was the position in the current bar's range
expressed as a percentage (0..100) and y is the percent change on
a close-to-close basis between the next bar's close and today's
close. I ran these samples through a "crosstab" which separates
the (x,y) pairs into "bins", where each row holds 20% of the x sample,
and each column holds 20% of the y sample. Other statistics are
gathered:
(all samples)
The entries in each cell in the table are:
average
#items in cell
%items of row, falling into this cell
Perent Change of Next Bar (Close-to-Close) Ave. %change
std. dev.
-7.25 -0.48 -0.07 0.26 0.69
-0.48 -0.07 0.26 0.68 5.77
Position in
Range
0.00 -1.31 -0.28 0.11 0.45 1.36 0.13
21.00 59 59 54 63 74 1.09
19.1% 19.1% 17.5% 20.4% 23.9%
21.00 -1.17 -0.26 0.07 0.46 1.15 0.07
45.00 65 49 62 67 66 0.93
21.0% 15.9% 20.1% 21.7% 21.4%
45.00 -1.29 -0.25 0.08 0.46 1.48 0.13
68.00 62 60 54 64 69 1.09
20.1% 19.4% 17.5% 20.7% 22.3%
68.00 -1.08 -0.24 0.05 0.43 1.29 0.08
89.00 60 69 60 60 60 0.85
19.4% 22.3% 19.4% 19.4% 19.4%
89.00 -1.02 -0.25 0.07 0.46 1.27 -0.01
100.00 63 72 79 55 39 0.76
20.5% 23.4% 25.6% 17.9% 12.7%
>From the above:
- Statistically, there are many more samples than expected where
the close was in the top 10% of the range.
- There was a slight tendency for the next day to reverse, and
to be down very slightly, or to be roughly unchanged following
a day where the close was in the high end of the range.
- There was a similar tendency for the market to reverse when
the close was in the low end of the range. There was a noticeable
average positive average gain of 0.13%
These conclusions change (or are reenforced), however, if we look at
only at samples where the day's range was "wide". The following
summary is for the same dataset, limited to days with a range of at
least 1% were studied. This cut the data sample down to about 1/3 of
the total sample:
-7.25 -0.61 -0.10 0.32 0.83
-0.61 -0.10 0.32 0.83 5.77
0.00 -1.61 -0.30 0.13 0.55 1.75 0.25
17.00 24 21 34 33 34 1.31
16.4% 14.4% 23.3% 22.6% 23.3%
17.00 -1.61 -0.33 0.11 0.57 1.38 0.00
43.00 31 28 25 36 26 1.17
21.2% 19.2% 17.1% 24.7% 17.8%
44.00 -1.69 -0.33 0.10 0.52 1.79 0.15
68.00 30 26 27 27 36 1.39
20.5% 17.8% 18.5% 18.5% 24.7%
69.00 -1.18 -0.29 0.05 0.58 1.52 0.06
90.00 34 32 27 25 28 0.99
23.3% 21.9% 18.5% 17.1% 19.2%
90.00 -1.25 -0.34 0.09 0.52 1.51 0.00
100.00 27 39 33 25 21 0.90
18.6% 26.9% 22.8% 17.2% 14.5%
>From the data above:
- there is still a high number of data points where the close was
in the high end of the range.
- there were slightly more samples where the close was in the low
end of the range. This makes sense, because big down days often
have a wider than average range.
- the tendency for the next day to be up, when the close is in the
low end of the range is more noticeable.
- the tendency for the next day to be flat (unchanged) on average,
when the close is in the high end of the range, is still there.
The sample below is restricted to days where the range was at least 2%
of the close. This occurred in roughly 10% of the total sample:
-7.25 -0.58 -0.07 0.47 1.36
-0.59 -0.07 0.46 1.34 5.77
0.00 -1.37 -0.32 0.19 0.80 2.41 0.69
7.00 4 5 6 8 9 1.46
12.5% 15.6% 18.8% 25.0% 28.1%
8.00 -2.15 -0.21 0.10 0.68 2.56 -0.10
30.00 10 5 6 5 6 2.00
31.2% 15.6% 18.8% 15.6% 18.8%
32.00 -2.97 -0.38 0.24 0.87 4.06 0.24
59.00 4 8 7 10 3 2.04
12.5% 25.0% 21.9% 31.2% 9.4%
59.00 -1.34 -0.30 0.26 0.84 1.89 0.30
89.00 7 6 6 5 8 1.27
21.9% 18.8% 18.8% 15.6% 25.0%
90.00 -1.48 -0.26 0.15 0.71 1.65 -0.07
100.00 7 8 7 4 4 1.08
23.3% 26.7% 23.3% 13.3% 13.3%
>From the above data:
- for very wide range days (where the range was >= 2%), there was
a much more noticeable "bar bell" effect, where the close tended
to be either in the very high end of the range, or the low-end
of the range. In fact, the distribution was skewed towards days
where the close was in the low end of the range. This is not
surprising, because big down days often have wide ranges.
- the tendency for the next day to be down-to-flat when the
close is in the high end of the range, is still there, with
an average return of -0.07%.
- the tendency for the next day to be a large up day, when the
close is in the low end of the range is more noticeable. The
average return of 0.69% is probably tradeable.
- however, if the position in close was in the 8 to 30% range, on
average, the next day will be down.
Conclusions:
- during the period studied, a higher than expected number of samples
showed a close in the high end of the range (90 to 100). This may
be due to the overall uptrend during the period studied.
- a close in the high end of the range does not generally predict
that the next day will be an up-day. In fact, on average, the
next day is flat to down.
- the 'barbell' effect increased with the magnitude of the day's range.
That is, the wider the range, the more likely that the position in
the close was either at the high end of the range or the low end
of the range.
- the mean reversion effect increases with the magnitude of the day's
range.
|