Friday, August 17, 2007

Pullbacks and Up-Down Patterns

[Sunday, 8 October 1995]

PULLBACKS:

In order to help identify when a stock has truly pulled back and is ready to move higher, I used a simple set of criteria and tested it for 4 stocks and a mutual fund. This was a limited data set, spanning only about a year, but it was somewhat helpful.
Rule: if a stock has declined 2 days in a row, then surges up on heavy volume (defined as volume > 50% above the 5 day average of volume), then buy at the next day's open.
PB% = pullback % = % price is above or below this next day's open;
PD = pullback days = # of days since pullback occurred.
RAT = volume ratio (defined earlier).

Stock: Criteria: n: Max % down Weekly Moly Min
next week: % +/-: +/-: Moly:
(avg)

LYTS Avge: 149 -9.6 (-3.1) 3.0 9.8% -7.0%
+PD < 3 12 -8.1 (-2.6) 5.6 9.8
+PD=1 6 -8.1 (-3) 6.1 10.2
+PD=1,
RAT<10 4 -1.8 (-.8) 8.5 19.5
PD<5,RAT<5 10 -1.8 (-.3) 9.8 18.2
PB<10% 55 -8.7 (-3) 3.2 12.6
PB<5% 36 -8.7 (-3.7) 3.7 14.7 -7.0%

A very interesting effect is noted here. As it turns out, the volume ratio is INVERSELY correlated to safety and future performance. In other words, the time to be in a stock is just after heavy volume has been pushing it down, and a sudden reversal on heavy volume occurs (2 days of light volume decline, followed by a heavy volume up day). Nevertheless, insuring we are within 10% of the pullback price (the open on the day after such a reversal), keeps our maximum weekly loss to -8.7% and improves the next mean monthly gain to 12.6% from an average of 9.8%.

Amgen: Amgen weakly illustrates the power of PB and PD. All takers averaged a maximum weekly decline of -9.3% and a weekly/monthly gain of 1.4%/6.1%. However, if you look at PD<= 3 and RAT < 0, the loss drops to -4.7% and the weekly/monthly gains shoot up to 5.3% and 16.5%. There were only four cases of this, however.

Intel: Intel averaged -18.5% as a weekly loss maximum with week/month gains of 1.5/7.4%. When PD <= 3, the weekly maximum loss was - 9.0%, but the week/month gains were unaffected 1.4/-1.1. Interestingly, if you include only RAT > 10, the average weekly gain drops to -10.9%, with a max of -18.5% and the average monthly gain increases to -5.0%.

Gateway 2000: Gateway is marginally improved using PD <=3 and MP B/S = +, which goes from a maximum weekly loss of -14.5% to -11.6%. The average weekly gain is improved to 2.9% from 1.5%.

As a result, to avoid the specific problem of being stopped out of a stock within the first week of purchase, using PB < 10% as a buy screening criteria would probably be a good move. However, this should be refined further.
Up-Down Patterns:

The next very interesting piece of the puzzle came from a simple observation I made that periods of runs in one direction tend to be followed by a counterreaction in the opposite direction. I began with Micron Technology:

MU:
RR, week Avge max loss: RR, mo:
Pattern: (min): (min): (min): Odds (week):
All: 2.9 (-15.7) -5.9 (-19) 14.3 (-3.1) 64
--- 9.8 (1.6) -3.4 (-6) 18.2 (9.0) 100%
--+ 2.8 (-8.3) -7.5 (-15) 12.6 (3.9) 64
-++ 3.3 (-14) -5.8 (-17) 14.2 (-1.6) 62
+++ 0.2 (-16) -7.1 (-19) 12.3 (-2.8) 53
++- 3.3 (-9) -5.8 (-15) 13.0 (2.2) 73
+-- 2.4 (-4) -5.1 (-15) 14.5 (3.3) 71
+-+ 4.9 (-6) -5.0 (-13) 18.7 (-3.1) 80

This was fascinating. As one might predict, periods of short term price declines, such as runs of 3 days in a row down, are followed by weekly returns of 9.8% on average, with a minimum weekly return of 1.6%, and a biggest weekly loss of only -6% (by weekly loss, I mean the lowest point during the next week relative to the buy price; the buy price is defined as the OPEN price the next day following the pattern). Note that the rate of return the next month is high also, with 18.2% being the average and the market rising 100% of the time.

AAMS:
RR, week Avge max loss: RR, mo:
Pattern: (min): (min): (min): Odds (week):
All: 3.5 (-11.8) -3.3 (-15) 14.3 (-9.3) 55 n=148
--- 4.1 (-2) -2.0 (-6) 8.0 (6.5) 67
+-- 3.4 (-7) -2.4 (-7) 15.9 (1.0) 56
+++ 6.7 (-6) -5.4 (-15) 13.7 (-9.3) 69
++- 3.6 (-5) -2.9 (-6) 16.1 (-5.9) 67

Note that this stock was in a defined uptrend during this study period (by the way, this extended from approximately 9/94 to 5/95), so the results might be a bit skewed. Nevertheless, the "---" pattern appears quite safe, with a maximum weekly low point of -6%, and only 2/3ds of the average next weekly low. However, what also must be noted is that it did not tend to be predictive of next month's gain (as one might expect). Note also that buying a powerful stock as it's moving higher ("+++") can lead to powerful results, e.g. a next week gain of 6.7%, double the average.

I also tested informally some of these patterns, using the following set of rules:

Trade with the trend (as defined by MP):
Buy whenever the closing price is lower for three days in a row, setting a buy stop an 1/8th of a point above that day's high, then resetting each day.
Once the buy stop is hit, sell whenever the close is less than the low of the day before. Using this strategy with Intel was quite profitable:

41.01 -> 43.53 +6%
43.32 -> 56.57 +31%
56.63 -> 62.63 +11%
65.76 -> 67.13 +2%
60.00 -> 60.13 +0% => Total: 57% B&H: 46.6%.

The remarkable thing about this series of trades, however, was that the total time of exposure was about 10 trading days! With not a single losing trade! This means that one could scan all markets and look for opportunities to do this, achieving several hundred percent per year...

No comments:

Post a Comment