Академический Документы
Профессиональный Документы
Культура Документы
MTA JOURNAL
Summer-Fall 2000
Issue 54
A Publication of
MARKET TECHNICIANS ASSOCIATION, INC.
One World Trade Center
Suite 4447
212/912-0995
Fax: 212/912-1064
e-mail: shelley@mta.com
Incorporated 1973
www.mta.org
ISSUE 54
EDITORIAL COMMENTARY
ADDRESS TO MTA 25TH ANNIVERSARY SEMINAR MAY 2000 LIVING LEGENDS PANEL
Robert J. Farrell
1
2
3
4
5
MECHANICAL TRADING SYSTEM VS. THE SP 100 INDEX: CAN A MECHANICAL TRADING SYSTEM
BASED ON THE FOUR W EEK RULE BEAT THE SP 100 INDEX?
13
19
TESTING THE EFFICACY OF THE NEW HIGH/NEW LOW INDEX USING PROPRIETARY DATA
25
31
MTA JOURNAL
Summer-Fall 2000
ISSUE 54
EDITOR
Henry O. Pruden, Ph.D.
Golden Gate University
San Francisco, California
ASSOCIATE EDITORS
David L. Upshaw, CFA, CMT
Lake Quivira, Kansas
MANUSCRIPT REVIEWERS
Connie Brown, CMT
Aerodynamic Investments Inc.
Pawley's Island, South Carolina
Cornelius Luca
Bridge Information Systems
New York, New York
Robert B. Peirce
Cookson, Peirce & Co., Inc.
Pittsburgh, Pennsylvania
PRODUCTION COORDINATOR
Barbara I. Gomperts
Financial & Investment Graphic Design
Marblehead, Massachusetts
PUBLISHER
Market Technicians Association, Inc.
One World Trade Center, Suite 4447
New York, New York 10048
MTA JOURNAL
Summer-Fall 2000
MTA JOURNAL
Summer-Fall 2000
AFFILIATE
Affiliate status is available to individuals who are interested in technical analysis, but who do not fully
meet the requirements for Member, as stated above; or who currently do not know three MTA members for
sponsorship. Privileges are noted below.
DUES
Dues for Members and Affiliates are $200 per year and are payable when joining the MTA and thereafter
upon receipt of annual dues notice mailed on July 1. College students may join at a reduced rate of $50 with
the endorsement of a professor.
APPLICATION FEES
Applicants for Member will be charged a onetime, nonrefundable application fee of $25; no fee for
Affiliates.
Members
Affiliates
MTA JOURNAL
Summer-Fall 2000
Management Committee
Director: President
Accreditation
Journal
Body of Knowledge
Membership
Richard A. Dickson
Scott & Stringfellow Inc.
804/780-3292
Fax: 804/643-9327
E-mail: DDickson@ScottStringfellow.com
Director: Secretary
Bruno DiGiorgi
Lowry's Reports Inc.
561/842-3514
Fax: 561/842-1523
E-mail: bdigiorgi@lowrysreports.com
Director: Treasurer
Andrew Bekoff
Bloomberg Financial Markets
212/495-0558
Fax: 212/809-9143
E-mail: abekoff@bloomberg.net
Director: Past President
TBA
Distance Learning
Richard A. Dickson
Scott & Stringfellow Inc.
804/780-3292, Fax: 804/643-9327
E-mail: DDickson@ScottStringfellow.com
Education
Director
Mike Epstein
NDB Capital Markets Corp.
617/753-9910
Fax: 617/753-9914
E-mail: mepstein@ndbcap.com
IFTA Liaison
Internship Committee
MTA JOURNAL
Summer-Fall 2000
Library
Larry Katz
Market Summary & Forecast
805/370-1919
Fax 805/777-0044
E-mail: lpk1618@aol.com
Newsletter
Michael N. Kahn
Bridge Information Systems
212/372-7541
E-mail: mkahn@bridge.com
Placement
Rick Bensignor
Morgan Stanley Dean Witter
212/761-6148
Fax: 212/761-0471
E-mail: rick.bensignor@msdw.com
Programs (NY)
Bernard Prebor
MCM MoneyWatch
212/908-4323
Fax: 212/908-4331
E-mail: bprebor@mcmwatch.comt
Regions
M. Frederick Meissner
404/875-3733
E-mail: fmeissner@mta.org
Rules
Nina G. Cooper
Pendragon Research, Inc.
815/244-4451
Fax: 815/244-4452
E-mail: ngcooper@grics.net
MTA JOURNAL
Summer-Fall 2000
ROBERT J. FARRELL
MTA JOURNAL
Summer-Fall 2000
INTRODUCTION
For options-based trading, the price action of any freely-traded
asset (e.g., stocks, futures, index futures, etc.) can be grouped into
three generic categories (however defined by the trader): (a) bullish price action; (b) bearish price action; (c) congestion/trading
range price action. Specific options-based strategies can be implemented which result in profits if any two out of the three outcomes
unfold. For example, the purchase of both call and put options on
the same underlying asset for the same strike price and same expiration date is termed a straddle position (e.g., buying XYZ $100
strike March 1999 call and put options = XYZ $100 March 1999
straddle). This straddle position can be profitable if either (a) or
(b) quickly occur with significant magnitude (i.e., price volatility)
prior to option expiration. In this sense, a straddle trade is nondirectional since it can profit in both bull and bear moves.
Price volatility can be described by several common technical
indicators including ADX, average-true-range, standard deviation,
and statistical volatility (also called historical volatility). Volatility
has been observed to be mean-reverting. Periods of abnormally
high or low short-term price volatility are followed by price volatility that is closer to the long-term price volatility of the underlying
asset.(1,3) A short-term drop in price volatility (volatility implosion)
can be reliably expected to be followed by a sudden volatility increase (volatility explosion). Connors, et. al. have shown that multiple days of short-term volatility implosion is a predictor of a strong
price move.(1,2)
The volatility implosion does not predict the direction of the
impending price move, but only that there is a high probability
that the underlying asset is going to move away from its current
price and by a significant amount. In addition, the volatility implosion does not predict when (how quickly) the explosion price move
will develop. We can predict which direction the price of the stock,
commodity, or market is not going to move with a high degree of
probability. It most likely will not move side-ways indefinitely. Knowing this, one can devise a trading strategy that is able to profit, or at
least not lose money, if the stock moves quickly higher or lower
such as the straddle strategy earlier described above.
In the option straddle strategy described above (e.g., XYZ $100
March 1999 straddle), as the price of the underlying asset moves
away from the option's strike price in either direction, the option
that is gaining in value will increase at a greater rate than the opposing option that is losing value. The position is said to be gamma
positive in both directions. The straddle will lose if the price of the
asset stays at or near the strike prices of the options, i.e. the stock
moves side-ways. The straddle position deteriorates because of
continued decrease in the volatility of the underlying asset, plus
the time-decay value of the option as it approaches expiration.
This study was designed to explore the potential investment returns that could be obtained using the basic option straddle strategy. At PRISM Trading Advisors, Inc., this strategy has been successfully implemented to generate superior returns at lower risk
than traditional investment portfolio benchmarks.
MTA JOURNAL
Summer-Fall 2000
RESULTS
DISCUSSION
It has been observed that short-term volatility will have a tendency to revert back to its longer-term mean.(1,3) Connors et.al.(1)
have published the Connors-Hayward Historical Volatility System
and showed that when the ratio of the 10-day versus the 100-day
historical volatilities was 0.5 or less, there was a tendency for strong
stock price moves to follow.
In this study, PRISM Trading Advisors, Inc., have confirmed the
phenomenon of volatility mean reversion by presenting the first
large scale option-based analysis while maintaining a strict marketneutral/delta-neutral (gamma positive) trading program. We have
shown that a significant price move occurs 75% of the time following a short-term volatility implosion (as defined in the Methods
and Materials section).
For this analysis we chose a relatively straightforward strategy:
to purchase a straddle. A straddle is the proper balance of put and
call options that produce a trade with no directional bias. A straddle
is said to be delta neutral and will generate the same profit whether
the underlying assets price moves higher or lower. As the asset
price moves away from its initial price one option will increase in
value while the other opposing option will decrease in value. A
profit is generated because the option that is increasing in value
will increase in value at a faster rate than the opposing option is
decreasing in value. The straddle is said to be gamma positive in
both directions.
This option strategy has a defined maximum risk of the trade
that is known at the initiation of the trade. This maximum risk of
loss is limited to the initial purchase costs of the straddle (premium
costs of both put and call options). There is no margin call with
this straddle strategy. There is an additional way that this strategy
can profit. Since the options are purchased at the time there has
MTA JOURNAL
Summer-Fall 2000
10
MTA JOURNAL
the RSI, ROC or MACD indicator. One could design a more sophisticated exit strategy such as exiting the position if the stock
price exceeds a predetermined price objective as defined by price
channels, parabolic functions, etc. An additional possibility would
be to re-establish a nondirectional options position at a predetermined price objective, thereby locking in all the profits generated up to that point. The myriad of options-based strategies available to adjust back to a delta neutral position based on technical
indicators and predetermined price objectives are beyond the scope
of this paper.
Although both systems had positive expectations based on 280
trades, there are several limitations of the study design. Although
moderate slippage was used in all the calculations, the robustness
of this study might have been improved if access to real-time stock
option bid-ask prices were available for all of the trades investigated. Unfortunately, such a large, detailed database is not readily
available. Given that the real-time bid-ask prices were not available, the use of the Black-Scholes formula with the known historical inputs (stock price, implied volatility, 90-day T-Bill yield) is an
acceptable alternative thereby minimizing any pricing differences
between the actual and theoretical option prices systematically
throughout the time period used in the study.
The current study revealed that a simple straddle options-based
strategy designed to exploit a sudden implosion of a stocks volatility with time as the only existing criteria produced draw-downs that
preclude it as a viable trading strategy in its own right. However,
this simple strategy had a positive expectation of generating superior returns, and therefore can be used as the basis to develop trading strategies capable of producing superior returns without the
need to correctly predict the direction of a given stock, commodity, or market being traded. The addition of some simple money
management rules dramatically improved the overall returns while
simultaneously decreasing the excessive draw-downs that plagued
the original trading strategy, thereby transforming it into a applicable trading system for every day use. This volatility-based, deltaneutral strategy also is independent of market direction. A market-neutral strategy and portfolio may be considered as a separate
asset class by portfolio managers in the efficient allocation of their
clients investment portfolios to boost returns while simultaneously
decreasing their clients risk exposure.
In conclusion, this is the first large-scale trading research study
to be shared with the trading public that clearly demonstrated how
the phenomenon of price volatility mean-reversion can be exploited
by using an options-based delta-neutral approach. Price, time and
volatility factors using options-based strategies to further maximize
positive expectancy represent active areas of real-time trading research at PRISM Trading Advisors, Inc. These results will be the
subject of future articles.
REFERENCES
1. Connors, L. A., and Hayward, B.E., Investment Secrets of a Hedge
Fund Manager, Probus Publishing, 1995.
2. Connors, L. A: Professional Traders Journal, Oceanview Financial
Research, Malibu, CA. March 1996, Volume 1, Issue 1.
3. Natenberg, S., Option Volatility and Pricing. Advanced Trading Strategies and Techniques, McGraw Hill, 1994.
(Over)
Summer-Fall 2000
11
TABLE 1
TABLE 2
2 Week
4 Week
6 Week
-191.9%
+334.7%
-84.3%
Total Return
-0.69%
+1.20%
-0.30%
Maximum Draw-Up
+373.9%
+933.2%
+948.2%
Maximum Draw-Down
-424.3%
-450.8%
-763.3%
91
106
100
185
174
177
Total Return
Average Return per Trade
4 Week
6 Week
+993.4%
+1188.6%
+3.55%
+4.25%
Maximum Draw-up
+641.4%
+704.1%
Maximum Draw-Down
-188.1%
-246.2%
120
117
160
161
14
13
+87.8%
+132.1%
+109.0%
+132.1%
+109.0%
-48.0%
-51.8%
-59.0%
-10.0%
-10.0%
Figure 1
Figure 2
MTA JOURNAL
Summer-Fall 2000
12
PREFACE
Results
System
NS-20BS-EQ
3.
4.
5.
6.
MTA JOURNAL
Equal
7.58%
9.00%
Sharpe Ratio
4.15%
3.85%
0.45
0.34
Maximum Loss
0.36
0.58
It is clear that the above system is performing less than the Buyand-Hold Strategy of S&P 100 Index.
There are several choices to improve performance of the system by modifying system parameters. The most natural change is
to search for better performance by modifying the number days
used for the Buy and Sell rules. The performance of the NS-xBSEQ system where x is the number of days for Buy and Sell rules was
tested for x between 10 days and 90 days. Results of the test are
provided in the Appendix 2.1.
Testing proved that the best performing system was the one with
50 days used for the Buy and Sell rules.
2.
01/01/1984 - 01/01/1989
Money Allocation
System
1.
Time Frame
Results
System
NS-50BS-EQ
Time Frame
01/01/1984 - 01/01/1989
Money Allocation
Equal
System
Index
7.70%
9.00%
Sharpe Ratio
4.49%
0.42
Maximum Loss
0.40
3.85%
0.34
0.58
Still the performance of the above system is not very impressive, so lets consider further research. Lets modify Rule 1 from
the system definition and replace it by following rule:
1A. Money management rule - Proportional allocation rule
Use $100,000 of capital into one hundred S&P 100 Index stocks,
allocating money into each stock according to its percentage
participation in the index at the starting date of the testing period (January 1, 1984).
Summer-Fall 2000
13
1989 and January 1, 1994). Here is the comparison of market statistics between the Index and the NS-50B-P-S25 system in the time
frame from January 1, 1989 to January 1, 1994.
System Code: NS-xBS-P (No Shorts, x Days for Buy and Sell
Rules, Proportionally Allocate Capital)
The new system consists of Rule 1A and Rules 2-6.
The performance of the NS-xBS-P system where x is the number of days for Buy and Sell rules was tested for x between 10 days
and 90 days. Results of the test are provided in the Appendix 2.2.
Testing proved that the best-performing system was one with 50
days used for Buy and Sell rules.
NS-50BS-P
Time Frame
01/01/1984 - 01/01/1989
Money Allocation
Proportional
50
System
Index
10.57%
9.00%
Sharpe Ratio
4.44%
3.85%
0.51
0.34
Maximum Loss
0.48
0.58
The last system outperforms the S&P100 Buy-and-Hold Strategy but lets consider further research. So far modifications were
limited to systems with different number of days for the Buy and
Sell rule and for using different initial allocations of the capital
equal and proportional. Lets consider the following hybrid of the
original NS-20BS-EQ system by replacing Rule 3 with following
new rule:
3B. Money management Stop Loss Rule - Sell losing positions
Sell a stock if it is losing more than y% of its buy price, where y
is a system parameter
Lets name this system:
NS-50B-P-S25
01/01/1984 - 01/01/1989
Money
Allocation
Days used
for Buy
Proportional
System
50
9.00%
0.52
0.34
Maximum Loss
0.64
0.58
0.59
0.49
0.40
System
Index
50,000
40,000
30,000
20,000
10,000
0
Jan-84
Jan-86
Jan-88
Jan-90
Jan-92
Jan-94
Jan-96
Some may argue that the good performance of the NS-50B-PS25 system is the result of a continuous Bull market between 1984
and 1997. So, at such conditions, buying low and not selling unless
the stock loses significant percent of its value will always result in a
winning strategy. However, such a system will perform very poorly
in the bear market.
To investigate this claim, lets test performance of the NS-50BP-S25 over the time period from August 25, 1987 to August 25,
1992. August 25, 1987 was chosen as the new start date became it is
the high of the S&P 100 market before 1987 crash. The next
graph presents the performance of this test.
CONCLUSION
The last system which is the result of several cycles of modifications to the initial 4WR outperforms the S&P100 Buy-and-Hold
Strategy by a good margin. To find out how time-stable the above
system was, a blind test was conducted.
MTA JOURNAL
0.68
Maximum Loss
60,000
Index
3.85%
70,000
25%
4.48%
6.92%
System NS-50B-P-S25
50 days 25%
% Loss used
in Rule 3B
14.78%
5.47%
Results
Time Frame
10.00%
GRAPHS
Index
13.88%
Comparing these values, we see that the system still outperformed the Index in respect to the Average Annual Compounded
Return (by 40%) and the Return Retracement Ratio, but marginally under-performed with the other two statistics. This can be explained by comparing the monthly DMI readings in the time period of 1984-1989 and 1989-1994. In the first time period, the standard 14-month Directional Movement Index (DMI) was well above
25 (a strong trending market); however, in the second time period, the DMI was only marginally above 25 (a weak trending market). So in such a time period, a trend following system doesnt
display as impressive results.
Results
System
System
Summer-Fall 2000
14
systematic traders favor backtesting, analyzing patterns and eliminating emotions. According to Barclay Trading Group Ltd., over
the last ten years systematic traders have yielded higher annual returns than discretionary traders six times. iii
Optimization of the trading system: The process of finding the
best performing parameter set for a given system. The underlying
premise of optimization is that the parameter set must work not
only in its initial time frame but any time frame. Almost any mechanical system can be optimized in a way that it will show positive
results in any given period of time. ii
Parameter: A value that can be freely assigned in the trading system in order to var y the timing of signals. ii
Parameter Set: Any combination of parameter values. ii
Parameter Stability: The goal of optimization is to find broad regions of parameter values with good system performance, instead
only one parameter which can represent an isolated set of market
conditions. ii
Time Stability: In the case of positive performance of the mechanical system in a specific time frame, it should be analyzed in different time frames to make sure the good performance is not dependent only on the initial time frame. ii
Blind Simulation: This is the test of an optimized parameter set in
a different time frame to see if the good results reoccur.
Average Parameter Set Performance: The complete universe of
parameter sets is defined before any simulation. Simulations are
then run for all the selected parameter sets, and the average of
these is used as an indication of the systems potential performance.ii
Average Annual Compounded Return: R = exp(1/N(ln(E) - ln(S)) - 1
S - starting equity
E - ending equity
N - number of years ii
Return Retracement Ratio: (RRR) = R/AMR
R - average annual compounded return
AMR - average maximum retracement for each data point. Using
drawdowns (the worst at each given point in time) to measure
risk, the risk component of RRR (AMR) comes closer to describing risk than standard deviation.
n
AMR=1/n( MRi)
i=1
MRi=max(MRPPi,MRSLi)
4WR_NS_B50_P_M25
50 days 25%
20,000
System
Index
15,000
10,000
5,000
0
Aug-87
Aug-88
Aug-89
Aug-90
Aug-91
BIBLIOGRAPHY
i John. J. Murphy, Technical Analysis of the Futures Markets, New
York Institute of Finance,1986.
ii Jack D. Schwager, Schwager on Futures - Technical Analysis, John
Wiley & Sons, Inc., 1996.
iii Carla Cavaletti, Trading Style Wars, Futures, July 1997.
MRPPi=(PEi - Ei)/Pei
MRSLi=(Ei - MEi)/Ei-1
Ei - equity at the end of month i,
PEi - peak equity on or prior to month i,
Ei-1 - equity at the end of month prior to month i,
MEi - minimum equity on or subsequent to month i.
APPENDIX 1
Glossary of Terms:
Mechanical Trading System: A set of rules that can be used to generate trade signals and trading performed according to the rules
of mechanical system. Primary benefits of mechanical trading systems are elimination of emotions from trading, and consistency of
approach and risk management. Mechanical trading systems can
be classified as Trend-Following (initiating a position with the
trend), and Counter-Trend (initiating a position in the opposite
direction to the trend). Trend following systems can be divided
into fast and slow. Fast a more sensitive system responds quickly
to signs of trend reversal and will tend to maximize profit on valid
signals, but also generate far more false signals. A good trend following system should not be too fast or too slow. ii
Trading according to signals generated by mechanical trading
system is called systematic trading which is opposite to discretionary trading. Discretionary traders claim that emotions, which are
excluded from systems trading, offer an edge. On the contrary,
MTA JOURNAL
Summer-Fall 2000
15
APPENDIX 2.1
APPENDIX 3
Number of
Average
Days for Buys
Annual
and Sell
Compounded
Rules
Return (%)
Sharpe
Return
Ratio Retracement
in %
Ratio
0.30
3.93
2.82
96.20
20
7.58
4.15
0.45
296.53
1.81
36
30
8.46
4.27
0.48
471.96
2.05
40
40
8.04
4.36
0.44
559.21
2.09
40
50
7.70
4.49
60
7.84
4.50
0.43
770.62
638.58
1.34
2.13
40
40
70
7.76
4.60
0.42
859.30
2.41
41
80
7.59
4.64
0.40
930.93
2.49
41
90
7.56
4.64
0.39
1011.86
2.59
42
10%
End
01/01/1984
01/01/1989
15%
20%
25%
30%
35% 40%
50%
60%
70%
14.82 14.95
14.92 14.95
14.81 14.94
14.03
14.36
14.56 14.68
14.56 14.68
14.38 14.33
70 13.59
14.08 14.38
14.35 14.39
80 13.30
14.08 14.18
14.16 14.22
14.07 14.17
This table shows the Sharpe Ratio in %. The columns are the
percentages used in the Sell Losing Positions Rule and the rows are
the number of days used in the Buy rule.
APPENDIX 2.2
Results for NS-xBS-P where x represents number of days
used in Buy and Sell rules, x>=10 and x<=90
Sharpe
Return
Ratio Retracement
in %
Ratio
Start
No
40 13.83
The optimal the system with 50 days used for Buy and Sell rules.
Number of
Average
Days for Buys
Annual
and Sell
Compounded
Rules
Return (%)
Sell Rule
Proportional
26
2.30
Money Allocation
NS-xB-P-Sy
10
0.42
System
10%
15%
20%
25%
30%
50%
60%
70%
10 4.32
4.29
4.38
4.38
4.41
35% 40%
4.42
4.43
4.44
4.43
4.43
20 4.22
4.34
30 4.33
4.45
40 4.29
4.37
4.41
50 4.31
4.37
4.43
4.37
4.43
4.42
4.45
4.47
4.48
4.49
4.49
4.49 4.49
4.46
4.49
4.49
4.53
4.53
4.53
4.53
4.46
4.46
4.47
4.47
4.47
4.46 4.46
4.47
4.47
4.48
4.48
4.48
4.47
4.45
4.45
4.47
4.47
4.45
4.45
4.53
4.47
10
4.48
3.07
0.30
109.72
1.40
29
60 4.25
20
8.85
4.18
0.48
361.66
2.04
40
70 4.33
4.37
4.43
4.43
4.43
4.45
4.45
4.43
4.43
4.45
4.43
4.36
4.39
4.39
4.39
4.40
4.40
4.39
4.39
4.39
30
10.05
4.37
0.52
594.16
2.44
44
80 4.28
40
10.15
4.37
0.52
763.29
2.73
46
90 4.32
50
10.57
4.44
60
10.90
4.38
70
10.96
4.55
0.51
1206.84
3.85
49
80
10.62
4.60
0.48
1479.84
3.91
50
10%
15%
20%
25%
30%
35%
40%
90
10.30
4.61
0.47
1585.23
4.19
50
10 0.49
0.48
0.50
0.49
0.49
0.50
0.50
0.50
0.50
0.50
20 0.49
0.49
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
30 0.51
0.51
0.51
0.53
979.46
1206.84
2.95
3.43
48
4.39
4.39
4.39
4.40
4.40
4.38
4.38
4.38
48
The optimal is the system with 50 days for Buy and Sell rules.
0.51
40 0.51
0.50
0.51
50 0.51
0.51
0.51
60 0.51
0.51
0.51
0.50
0.50
0.51
0.51
0.51
0.51
0.52
0.51
0.51
0.51
50%
0.51
60%
0.51
0.51
0.51
0.51
0.51
0.51
0.52
0.52
0.52
0.51
0.51
0.52
0.52
70 0.52
0.53
0.53
0.53
0.53
0.53
0.53
0.54
0.54
80 0.51
0.53
0.53
0.53
0.53
0.53
0.53
0.54
0.54
90 0.51
MTA JOURNAL
4.38
0.53
0.53
Summer-Fall 2000
0.53
0.53
0.53
0.53
0.54
0.54
70%
0.51
0.51
0.52
0.52
0.54
0.54
0.54
16
This table shows the Expected Net Profit Per Trade (ENPPT)
in $. The columns are the percentages used in the Sell Losing Positions Rule and the rows are the number of days used in the Buy rule.
10%
15%
20%
25%
30%
6597.08 7739.21
7252.84 7716.12
35% 40%
50%
60%
70%
10%
10 9.76
15%
10.04
20%
14.62
25%
16.25
30%
19.87
35% 40%
25.01
50%
27.07
60%
39.85
70%
62.18
81.81
21.44 22.65
7009.01 7807.08
19.61 22.47
7140.84 7725.92
20.79 23.31
7345.42 7969.82
7492.27 7966.95
9961.49
7384.87 8043.70
9830.62
70 11.45
16.57 23.84
7355.29 8109.93
9619.90
80 11.08
19.83 23.73
7275.61 8127.05
9544.71
15%
20%
25% 30%
35% 40%
50%
60%
0.66
0.66
0.66
0.66
0.66
0.67
0.67
0.67
0.67
20 0.63
0.65
0.65
0.66
0.66
0.66
0.66
0.66
0.67
0.67
30 0.62
0.64
0.64
0.65
0.65
0.65
0.66
0.66
0.66
0.66
40 0.61
0.62
0.63
0.64
0.64
0.64
0.65
0.65
0.65
0.65
50 0.61
60 0.60
0.62
0.61
0.63
0.62
0.63
0.63
0.64
0.63
0.64
0.63
0.64
0.63
0.64
0.64
0.64
0.64
0.61
0.61
0.62
0.62
0.62
0.62
0.62
0.62
0.62
0.61
0.61
0.61
0.61
0.61
0.61
0.61
0.61
0.61
90 0.59
0.60
0.61
0.61
0.61
0.61
0.61
0.61
0.61
24.85 29.52
24.23 31.79
33.52 38.51
61.27
93.06
0.64
80 0.59
24.46 27.89
24.91
0.64
70 0.60
24.92 25.00
The best performing is the system with 50 days used for Buy
rule and 25% Sell Losing Position rule.
70%
10 0.65
21.31 24.21
0.61
MTA JOURNAL
Summer-Fall 2000
17
MTA JOURNAL
Summer-Fall 2000
18
Olsens study shows that the more analysts are wrong, which is
another source of stress, the more their herding behavior increases.11
How can seemingly rational professionals be so utterly seduced
by the opinion of their peers that they will not only hold, but change
opinions collectively? Recall that the neocortex is to a significant
degree functionally disassociated from the limbic system. This
means not only that feelings of conviction may attach to utterly
contradictory ideas in different people, but that they can do so in
the same person at different times. In other words, the same brain can
support opposite views with equally intense emotion, depending upon
the demands of survival perceived by the limbic system. This fact
relates directly to the behavior of financial market participants, who
can be flushed with confidence one day and in a state of utter panic
the next. As Yale economist Robert Schiller puts it, You would
think enlightened people would not have firm opinions about markets, but they do, and it changes all the time.12 Throughout the herding process, whether the markets are real or simulated, and whether
the participants are novices or professionals, the general conviction of the rightness of stock valuation at each price level is powerful, emotional and impervious to argument.
Falling into line with others for self-preservation involves not
only the pursuit of positive values but also the avoidance of negative values, in which case the reinforcing emotions are even stronger. Reptiles and birds harass strangers. A flock of poultry will peck
to death any individual bird that has wounds or blemishes. Likewise, humans can be a threat to each other if there are perceived
differences between them. It is an advantage to survival, then, to
avoid rejection by revealing your sameness. D.C. Gajdusek researched a
long-hidden Stone Age tribe that had never seen Western people
and soon noticed that they mimicked his behavior; whenever he
scratched his head or put his hand on his hip, the whole tribe did
the same thing.13 Says MacLean, It has been suggested that such
imitation may have some protective value by signifying, I am like you.
He adds, This form of behavior is phylogenetically deeply ingrained.14
The limbic system bluntly assumes that all expressions of I am
not like you are infused with danger. Thus, herding and mimicking are preservative behavior. They are powerful because they are
impelled, regardless of reasoning, by a primitive system of mentation that, however uninformed, is trying to save your life.
As with so many useful paleomentational tools, herding behavior is counterproductive with respect to success in the world of modern financial speculation. If a financial market is soaring or crashing, the limbic system senses an opportunity or a threat and orders
you to join the herd so that your chances for success or survival will
improve. The limbic system produces emotions that support those
impulses, including hope, euphoria, cautiousness and panic. The
actions thus impelled lead one inevitably to the opposite of survival
and success, which is why the vast majority of people lose when
they speculate.15 In a great number of situations, hoping and herding can contribute to your well-being. Not in financial markets. In
many cases, panicking and fleeing when others do cuts your risk.
Not in financial markets. The important point with respect to this
aspect of financial markets is that for many people, repeated failure
does little to deter the behavior. If repeated loss and agony cannot overcome the limbic system's impulses, then it certainly must have free
rein in comparatively benign social settings.
Regardless of their inappropriateness to financial markets, these
impulses are not irrational because they have a purpose, no matter
how ill-applied in modern life. Yet neither are they rational, as they
Figure 2
Stock Mutual Funds Cash/Assets Ratio vs. Aggregate Stock Prices
Monthly Data 12/31/65 - 12/31/98 (log scale)
MTA JOURNAL
Summer-Fall 2000
19
are within mens unconscious minds, i.e., their basal ganglia and
limbic system, which are equipped to operate without and to override the conscious input of reason. These impulses, then, serve
rational general goals but are irrationally applied to too many specific situations.
MTA JOURNAL
Summer-Fall 2000
20
CONCLUSION
R.N. Elliott discovered before any of the above was known, that the
form of mankinds evaluation of his own productive enterprise,
i.e., the stock market, has Fibonacci properties. These studies and
statistics say that the mechanism that generates the Wave Principle,
mans unconscious mind, has countless Fibonacci-related properties. These findings are compatible with Elliotts hypothesis.
NOTES
1. MacLean, P. (1990). The triune brain in evolution: role in
paleocerebral functions. New York: Plenum Press.
2. Scuoteguazza, H. (1997, September/October). Handling emotional intelligence. The Objective American.
3. MacLean, P. (1990). The triune brain in evolution, p. 17.
4. Chapters 15 through 19 of The Wave Principle of Human Social
Behavior explore this point further.
5. Wright, K. (1997, October). Babies, bonds and brains. Discover, p. 78.
6. Pigou, A.C. (1927). Industrial fluctuations. London: F. Cass.
7. Pigou, A.C. (1920). The economics of welfare. London: F. Cass.
8. Among others, such measures include put and call volume ratios, cash holdings by institutions, index futures premiums, the
activity of margined investors, and reports of market opinion
from brokers, traders, newsletter writers and investors.
9. Bishop, J.E. (1987, November 17). Stock market experiment
suggests inevitability of booms and busts. The Wall Street Journal.
10. Olsen, R. (1996, July/August). Implications of herding behavior Financial Analysts Journal, pp. 37-41.
11. Just about any source of stress can induce a herding response.
MacLean humorously references the tendency of governments
and universities to respond to tension by forming ad hoc committees.
12. Passell, P. (1989, August 25). Dow and reason: distant cousins? The New York Times.
This phenomenon of time is the same as the one that R.N. Elliott described for price swings in the 1930-1939 period recounted
in Chapter 5 of The Wave Principle of Human Social Behavior.
In the past three years, modern researchers have conducted experiments that further demonstrate Elliotts observation that phi
and the stock market are connected. The October 1997 New Scientist reports on a study that concludes that the stock markets Hurst
exponent,28 which characterizes its fractal dimension, is 0.65.29 This
number is quite close to the Fibonacci ratio. However, since that
time, the figure for financial auction-market activity has gotten even
closer. Europhysics Letters has just published the results of a market
simulation study by European physicists Caldarelli, Marsili and
Zhang. Although the simulation involves only a dozen or so subjects at a time trading a supposed currency relationship, the resulting price fluctuations mimic those in the stock market. Upon measuring the fractal persistence of those patterns, the authors come
to this conclusion:
The scaling behavior of the price returns...is very similar to that
observed in a real economy. These distributions [of price differences] satisfy the scaling hypothesis...with an exponent of H =
0.62.30
The Hurst exponent of this group dynamic, then, is 0.62. Although the authors do not mention the fact, this is the Fibonacci
ratio. Recall that the fractal dimension of our neurons is phi. These
two studies show that the fractal dimension of the stock market is
related to phi. The stock market, then, has the same fractal dimensional
factor as our neurons, and both of them are the Fibonacci ratio. This is
MTA JOURNAL
Summer-Fall 2000
21
13. Gajdusek, D.C. (1970). Physiological and psychological characteristics of stone age man. Symposium on Biological Bases of
Human Behavior, Eng. Sci. 33, pp. 26-33, 56-62.
14. MacLean, P. (1990). The triune brain in evolution.
15. There is a myth, held by nearly all people outside of back-office employees of brokerage firms and the IRS, that many
people do well in financial speculation. Actually, almost everyone loses at the game eventually. The head of a futures brokerage firm once confided to me that never in the firms history
had customers in the aggregate had a winning year. Even in
the stock market, when the public or even most professionals
win, it is a temporary, albeit sometimes prolonged, phenomenon. The next big bear market usually wipes them out if they
live long enough, and if they do not, it wipes out their successors. This is true regardless of todays accepted wisdom that
the stock market always goes to new highs eventually and that
todays investors are wise. Aside from the fact that the new
highs forever conviction is false (Where was the Roman stock
market during the Dark Ages?), what counts is when people
act, and that is what ruins them.
16. Kelly, G.A. (1955). The psychology of personal constructs, Vols. 1
and 2.
17. Osgood, C.E., and M.M. Richards (1973). Language, 49, pp.
380-412; Shalit, B. (1960). British Journal of Psychology, 71, pp.
39-42; Rapoport, A. and A.M. Chammah (1965). Prisoners dilemma. University of Michigan Press.
18. Poulton, E.C., Simmonds, D.C.V. and Warren, R.M. (1968). Response bias in very first judgments of the reflectance of grays:
numerical versus linear estimates. Perception and Psychophysics,
Vol. 3, pp. 112-114.
19. Adams-Webber, J. and Benjafield, J. (1973). The relation between lexical marking and rating extremity in interpersonal
judgment. Canadian Journal of Behavioral Science, Vol. 5, pp.
234-241.
20. Adams-Webber, J. (1997, Winter). Self-reflexion in evaluating
others. American Journal of Psychology, Vol. 110, No. 4, pp. 527541.
21. McGraw, K.M. (1985). Subjective probabilities and moral judgments. Journal of Experimental and Biological Structures, #10, pp.
501-518.
22. Washburn, J. (1993, March 31). The human equation. The
Los Angeles Times.
23. Lefebvre, V.A. (1987, October). The fundamental structures
of human reflexion. The Journal of Social Biological Structure,
Vol. 10, pp. 129-175.
24. Lefebvre, V.A. (1992). A psychological theory of bipolarity and reflexivity. Lewinston, NY: The Edwin Mellen Press. And Lefebvre,
V.A. (1997). The cosmic subject. Moscow: Russian Academy of
Sciences Institute of Psychology Press.
25. Butler, D. and Ranney, A. (1978). Referendums Washington,
D.C., American Enterprise Institute for Public Policy Research.
26. Rhea, R. (1934). The story of the averages: a retrospective study of
the forecasting value of Dows theory as applied to the daily movements
of the Dow-Jones industrial & railroad stock averages. Republished
January 1990. Omnigraphi. (See discussion in Chapter 4 of
Elliott Wave Principle by Frost and Prechter.)
27. Sornette, D., Johansen, A., and Bouchaud, J.P. (1996). Stock
market crashes, precursors and replicas. Journal de Physique I
France 6, No.1, pp. 167-175.
MTA JOURNAL
28. The Hurst exponent (H), named for its developer, Harold
Edwin Hurst [ref: Hurst, H.E., et al. (1951). Long term storage:
an experimental study] is related to the fractal, or Hausdorff dimension (D) by the following formula, where E is the embedding Euclidean dimension (2 in the case of a plane, 3 in the
case of a space): D = E - H. It may also be stated as D = E + 1 - H
if E is the generating Euclidean dimension (1 in the case of a
line, 2 in the case of a plane). Thus, if the Hurst exponent of a
line graph is .38, or /-2, then the fractal dimension is 1.62, or
; if the Hurst exponent is .62, or -1, then the fractal dimension is 1.38, or 1 + -2. [source: Schroeder, M. (1991). Fractals,
chaos, power laws: minutes from an infinite paradise. New York: W.H.
Freeman & Co.] Thus, if H is related to , so is D.
29. Brooks, M. (1997, October 18). Boom to bust. New Scientist.
30. Caldarelli, G., et al. (1997). A prototype model of stock exchange. Europhysics Letters, 40 (5), pp. 479-484.
31. Lefebvre, V.A. (1998, August 18-20). Sketch of reflexive game
theory, from the proceedings of The Workshop on Multi-Reflexive Models of Agent Behavior conducted by the Army Research
Laboratory.
32. Caldarelli, G., et al. (1997, December 1). A prototype model
of stock exchange. Europhysics Letters, 40 (5), pp. 479-484.
Summer-Fall 2000
22
INTRODUCTION
filtering method to new high/low data taken from publicly available sources and applied to the TDHLI.
The purpose of this paper shall be to evaluate the efficacy of
the TDHLI using four years of historical prices to calculate new
high/low data on the largest 5000 stocks on the NYSE, Amex and
NASDAQ from January 31, 1996 to December 31, 1999. The conclusions of my earlier paper were that the traditional rules suggested by both Messrs. Cohen and Redegeld did not perform particularly well. The dynamic rules that I suggested improved performance and flexibility, but did little to explore the impact of several
factors like market cap on the performance and predictive power
of indicators like the TDHLI. The aim of the current study is to
explore the merits of calculating new high/low data and parsing it
by market cap to provide a deeper look into the technicals of the
stock market.
The attempt at compiling a relatively large database of stock
prices necessary to create the capability to generate new high/low
data nearly undid my project to submit for the 2000 Charles Dow
Award. Despite my best efforts over several months and a surprisingly ineffectual collection of state-of-the-art PC hardware, the entire enterprise nearly collapsed from the strain on memory and
processing speed available today. The final, successful effort required four PCs with Pentium 500-600 megahertz processors with
a combined 704 Megabytes of RAM running in parallel on small
subsets of data that had to be reorganized and recompiled after
the initial run. To say that the additional capabilities made possible by using proprietary data comes at a cost is an understatement. Still, the benefits of flexibility may yet prove to be worth the
ultimate effort.
My suggested approach, as before, uses percentage filters held
to a tolerance of two standard deviations from the mean for past
signals. Filter percentages varied as one might expect with the
market cap and volatility of the stocks. When the value of the indicator changes, for example, by 20% from its most recent high or
low, a buy or sell is signaled. The percentage hurdle was derived by
taking the percent move that correctly captured 95% (or 2 standard deviations from the mean) of the historic index moves. The
mid cap data required a 21% filter for the period studied while
small caps needed a 25% band to meet the criterion. Totals for all
the new high/low data, perhaps as a result of the time period utilized and the relative numbers of small cap stocks in the data, required a 23% filter. The benefit of dynamic rules appears to be
supported by my experience with both studies conducted to date.
Based on comments by fellow MTA members shortly after submission of my 1999 CMT paper, Testing the Efficacy of New High/
New Low Data, I began to ponder how I might explore further the
possibilities of using new high/new low data as a stock market indicator. The NYSE new highs and new lows have been used in technical analysis and by market watchers for many years. The theory is
that the stocks reaching new 52-week highs or lows represent significant events relative to the market and its sectors. If possible,
the study of the actual prices underlying new high/low data might
suggest intriguing new ways to explore different applications of
indicators like the 10-Day High/Low Index (referred to herein after as TDHLI) for predicting market action.
There are a number of ways to use new high and new low data.
In order to test the efficacy of new high/new low indicators using
proprietary data, we will employ the same 10-day moving average
of the percentage of new highs over new highs plus new lows that I
employed last year to test publicly available new high/new low data.
The traditional rules, by way of review, were introduced to me by
Jack Redegeld, the head of technical research at Scudder, Stevens
& Clark in 1986. A definition of terms from my 1999 CMT paper
follows below:
What differentiates the TDHLI from other indicators that
use new high and new low data is that it tracks the oscillation
from 0 to 1 of the net new highs (new highs/(new highs + new
lows)) and then uses a 10-day simple moving average to smooth
the results. The TDHLI signals a buy when the indicator rises
above 0.3 (or 30% of the range), and indicates a sell when it
falls below 0.7 (or 70% of the range). The origin of the 70/30
filter is from Jack Redegeld's work over time. A.W. Cohen published an approach in 1968 using similar rules to Jack Redegeld's
application of the TDHLI while at Scudder (1961-1989):
The extreme percentages on this chart are above 90%
(occasionally above 80%) and below 10% (occasionally below 20%). Intermediate down moves and bear markets usually end when this percentage is below the 10% level. The
best time to go long is when the percentage is below the 10%
level and turns up. This is a bull alert signal. Short positions
should be covered and long positions taken. A rise in the
percentage above a previous top or above the 50% level is a
bull confirmed signal. The best time to sell short is when this
percentage is above the 90% level and turns down. This is a
bear alert signal. Long positions should be closed out and
short positions established. A drop in the percentage below
a previous bottom or below the 50% signals a bear confirmed
market.
In my CMT paper, I suggested a dynamic filtering method to
improve performance of the TDHLI. The technique is simply to
apply a percentage filter based on two standard deviations from
the mean of the data to the TDHLI. The net effect of the dynamic
filter method is to capture a greater portion of the market's move,
but at the cost of higher transaction costs and more frequent signals (some of which will be false signals that incur loses further
increasing the cost of doing business). In my prior study, the data
showed substantial performance gains from employing the dynamic
MTA JOURNAL
Table 1
TDHLI Standard Deviations
Large Cap
Mid Cap
Small Cap
Total
4 year
20.67%
20.87%
25.00%
23.95%
3 year
21.77%
21.95%
26.16%
25.17%
2 year
23.65%
22.80%
25.58%
25.22%
1 year
14.29%
14.12%
19.55%
18.47%
Summer-Fall 2000
23
METHODOLOGY
New highs are defined as stocks reaching a new high over the
previous year of daily prices. New lows conversely are stocks descending below the lowest price over the prior year. Stocks were
selected based on the 5000 largest market capitalizations of the
three main US exchanges and may represent biased data to the
extent that prior performance influenced the market caps at the
time that the data was sorted (the end date). Spreadsheets were
then constructed to reflect new highs/lows from yearly databases
of daily closing prices.
The tables below show buys and sells in the first column, the
trade date in the second, the index value in the third, the return
for each trade in decimal format next and the final column shows
cumulative results (1.04 = 4% gain, 0.99 = 1% loss) for all trades in
sequence. At the bottom of each table the total index and indicator returns can be found.
MTA JOURNAL
Summer-Fall 2000
24
Table 2
Large Cap Rates of Return
Date
SPX
Return
b 01/31/96
636.02
s 03/08/96
633.5
b 03/22/96
650.62
s 06/26/96
664.39 2.12%
b 07/05/96
657.44
s 07/12/96
646.19 -1.71%
b 07/25/96
631.17
s 09/05/96
649.44 2.89%
b 09/19/96
683
s 12/13/96
728.64 6.68%
b 12/30/96
753.85
s 01/27/97
765.02 1.48%
b 02/07/97
789.56
s 03/20/97
782.65 -0.88%
b 04/21/97
760.37
s 08/21/97
925.05 21.66%
b 08/29/97
899.47
s 10/23/97
950.69 5.69%
b 11/10/97
921.13
s 12/15/97
963.39 4.59%
b 01/06/98
966.58
-0.40%
Table 3
Index
Date
SPX
100.000
s 03/12/96
Return
Index
Date
215.67
100
b 07/19/96
221.66
100
228.51 5.95%
105.9535
s 12/18/96
249.87 12.73%
112.7267
s 12/17/96
726.04 13.45%
112.9456
b 03/20/96
230.28
b 04/16/97
251.47
s 06/17/96
237.87 3.30%
s 10/27/97
308.39 22.63%
b 07/02/96
236.98
b 09/16/98
308.5
s 07/15/96
218.6
s 02/12/99
367.09 18.99%
164.4971
Total
27.68%
MID
65.61%
101.7118
b 08/24/98 1088.14
127.6837
99.97137
102.8652
221.66
SPX
93.09%
s 09/03/96
231.21 4.31%
b 09/16/96
238.76
s 12/16/96
247.16 3.52%
b 01/02/97
252.11
s 03/13/97
263.14 4.38%
b 04/08/97
255.84
s 04/10/97
253.76 -0.81%
b 04/16/97
251.47
s 10/24/97
329.48 31.02%
b 11/06/97
325.13
s 12/19/97
320.26 -1.50%
b 01/05/98
333.39
s 05/26/98
363.22 8.95%
b 06/26/98
356.97
s 07/24/98
357.91 0.26%
b 08/17/98
329.85
s 08/31/98
299.86 -9.09%
b 09/08/98
291.65
s 10/07/98
289.31 -0.80%
b 10/16/98
305.41
s 02/08/99
367.23 20.24%
b 03/10/99
364.25
s 03/29/99
363.05 -0.33%
b 04/08/99
362.46
s 06/01/99
395.87 9.22%
b 06/16/99
399.73
s 07/26/99
415.37 3.91%
b 08/19/99
400.3
s 09/16/99
401.09 0.20%
b 10/06/99
384.82
s 10/19/99
369.5
b 10/29/99
391.2
s 12/15/99
411.75 5.25%
134.2985
141.9461
148.4583
166.6635
149.4369
b 10/14/98 1005.5
182.8158
b 03/01/99 1236.2
189.9501
0.23%
190.3942
b 06/08/99 1317.3
b 09/02/99 1319.1
s 09/24/99 1277.4
-3.17%
184.3682
-4.82%
175.4817
b 10/07/99 1317.6
s 10/18/99 1254.1
b 10/28/99 1342.4
s 12/31/99 1469.3
9.45%
105.3069
b 07/19/96
110.3903
3.90%
100.9572
27.68%
111.3649
192.0581
109.4458
-7.76%
Total
109.7389
155.3455
s 08/31/99 1320.4
Index
s 03/04/96
994.24 -8.25%
s 05/25/99 1284.4
Return
b 01/25/96
169.3186
MID
99.55309
1.59%
970.68 -3.80%
Date
100.000
b 09/11/98 1009.1
s 10/07/98
Index
639.95 -0.45%
b 08/17/98 1083.7
s 09/01/98
Return
637.09
b 06/25/98 1129.3
s 07/27/98 1147.3
MID
99.60379 b 07/31/96
-3.98%
109.0117
113.7811
112.856
147.8658
145.651
158.683
159.1009
144.6354
143.475
172.5167
171.9483
187.7978
195.1456
195.5307
187.7465
197.609
Total
92.06%
Total
97.61%
SPX
131.01%
MID
90.92%
MTA JOURNAL
Summer-Fall 2000
138.2423
25
Table 4
Small Cap Rates of Return
Date SML
Return
Table 5
Index
Date SML
b 02/02/96
122.95
100
b 08/02/96
128.73
s 06/18/96
134.96 9.77%
109.7682
s 10/29/96
136.1
b 07/30/96
123.87
s 11/01/96
136.48 10.18%
b 11/13/96
139.67
s 03/05/97
146.21 4.68%
b 04/16/97
137.27
s 10/27/97
174
Return
b 07/31/96
639.95
100
s 11/01/96
703.77 9.97%
109.9727
b 07/05/96
657.44
b 04/14/97
743.73
135.2556
s 07/15/96
629.8
-4.20%
99.3368
s 10/27/97
876.99 17.92%
b 07/30/96
635.26
s 11/04/96
706.73 11.25%
110.5127
b 11/13/96
731.13
s 12/18/96
731.54 0.06%
b 01/02/97
737.01
s 03/17/97
795.71 7.96%
b 04/21/97
760.37
190.86
b 03/04/99
161.45
s 05/24/99
177.79 10.12%
b 08/16/99
179.48
s 11/22/99
185.95 3.60%
200.67 9.46%
172.4588
188.79
s 07/24/98
183.18 -2.97%
b 08/07/98
175.83
s 08/26/98
160.54 -8.70%
b 09/08/98
152.35
s 10/07/98
133.73 -12.22%
b 10/15/98
139.4
s 12/14/98
161.26 15.68%
b 12/29/98
171.51
s 01/27/99
172.17 0.38%
b 03/01/99
160.47
s 03/24/99
155.23 -3.27%
b 04/07/99
158.75
s 05/25/99
175.68 10.66%
b 06/28/99
182.07
s 07/28/99
183.64 0.86%
b 08/17/99
180.22
s 09/22/99
175.96 -2.36%
b 10/07/99
176.07
s 10/18/99
168.96 -4.04%
b 10/28/99
174.11
s 11/30/99
182.97 5.09%
b 12/29/99
195.47
Index
100.0000
164.75 -13.68%
b 06/25/98
Return
103.6964
b 06/08/98
s 05/06/98
SPX
662.06 3.70%
s 12/10/98
157.5566
Date
638.46
126.6057
176.1 -1.82%
183.33
Index
s 06/18/96
178.17 27.93%
179.37
Return
b 02/01/96
139.27
s 01/16/98
b 02/04/98
SPX
100
s 10/28/97
b 12/03/97
Date
105.7252
b 04/08/97
160.4822
Index
5.73%
120.9426
26.76%
116.7524
128.5687
133.2034
s 06/24/98 1132.88
s 07/21/98 1165.07 2.84%
133.362
b 02/23/99 1271.18
110.5747
s 05/25/99 1284.4
1.04%
134.749
b 10/05/99 1301.35
Total
33.20%
s 10/27/97
876.99 15.34%
SML
44.45%
b 11/11/97
923.78
167.3341
129.6773
119.3815
146.7323
137.6914
Total
46.73%
SPX
121.44%
163.233
b 06/24/98 1132.88
152.7829
s 07/24/98 1140.8
0.70%
164.3741
164.3696
b 08/12/98 1084.22
134.11
b 09/08/98 1023.46
155.1405
s 10/07/98
970.68 -5.16%
155.893
b 10/15/98 1047.49
155.7375
s 12/14/98 1141.2
8.95%
169.8394
173.0617
b 12/29/98 1241.81
150.652
b 02/24/99 1253.41
166.7184
175.1576
b 04/07/99 1326.89
168.156
s 05/25/99 1284.4
-3.20%
169.5487
b 06/28/99 1331.35
164.1812
173.559
b 08/17/99 1344.16
157.5513
168.8357
b 10/06/99 1325.4
165.5686
160.6729
b 10/28/99 1342.44
Total
65.57%
SML
58.98%
b 12/29/99 1463.46
167.2892
Total
67.29%
SPX
118.92%
Table 6
Trades Detailed
Cumulative Loses
Traditional Rules
Dynamic Rules
Largest Loss
Cumulative Loses
Traditional Rules
Dynamic Rules
Largest Loss
MTA JOURNAL
Summer-Fall 2000
Large Cap
Mid Cap
Small Cap
-0.45%
0%
-13.68%
Total Cap
0%
-23.03%
-24.26%
-35.38%
-20.11%
-8.25%
-9.09%
-12.22%
-5.16%
Large Cap
Mid Cap
Small Cap
Total Cap
28.13%
27.68%
46.88%
46.73%
115.09%
121.87%
100.95%
87.40%
22.34%
31.02%
26.76%
18.55%
26
OBSERVATIONS
MTA JOURNAL
CONCLUSION
The traditional TDHLI and, in the current period, the dynamic
TDHLI as well, failed to keep pace with the market despite posting
strong results. The utilization of proprietary data made the TDHLI
considerably more flexible and provided new and interesting dimensions to the indicator. While it provided useful sell and buy
signals under most conditions, the TDHLI even with the added
functionality of proprietary data did not perform well enough and
robustly enough to be considered an effective indicator solely on
its own. The ability to predict market direction, to provide reasonably competitive performance particularly considering results from
my prior study and to track the market along the lines of market
capitalization may prove over time to be a worthwhile addition to
the art of technical analysis as embodied by the TDHLI.
ATTRIBUTION
Joseph Redegeld, The Ten Day New High/New Low Index, 1986.
A.W. Cohen, Three-Point Reversal Method of Point & Figure Stock
Market Trading, 8th Edition, 1984, Chartcraft, Inc., pg 91.
2Data was provided by Factset, Inc.
Summer-Fall 2000
27
BIRTH OF A CANDLESTICK
Using Genetic Algorithms to Identify Useful Candlestick Reversal Patterns
Jonathan T. Lin, CMT
INTRODUCTION
MTA JOURNAL
Summer-Fall 2000
28
effective candlestick pattern, these points should all be considered integral parts of the pattern definition.
[Authors Note: Although volume and open interest accompanying
the candlestick patterns could be used as confirmation, the author has decided not to include either as part of the pattern definition for two reasons:
1) A stock can rally or decline without increasing volume. Volume tends to
be more evident around structural breakouts and breakdowns, but not always so around early reversal points. This is especially true for thinlytraded stocks where the price can move either way quickly without much
volume. Depending on volume as confirmation is impractical at times. 2)
The pattern the author plans to find should be rather universal just like the
other candlestick patterns. Shooting stars are not unique to crude oil futures; nor is the bearish engulfing pattern only designed for Microsoft stock.
Since some market data, such as spot currencies and interest rates, do not
contain volume nor open interest information, a candlestick pattern with
volume or open interest as an integral part of it will not be universally
useful.]
MTA JOURNAL
Summer-Fall 2000
29
really know when or when not to eat maybe some are eating all
the time while some simply do not move at all. As some of them
over ate or starved to death, the smarter ones knew how to eat
right, correctly spotting potential for profit. As only the smart
ones survive, they begin to preserve only the smart genes. As these
smarter ones mate and reproduce, some of the next generation
may contain, by chance, some even better combination of genes,
resulting in even smarter candlesharks. As the process continues,
the candlesharks should evolve to be pretty smart eaters. Once in a
while, some genes will mutate, creating candlesharks unlike their
parents. Whether the newly injected genes will be included in the
gene pool will depend on the success of these mutated creatures in
adapting to their environment.
One thing that should be pointed out is that the late generation candlesharks will probably swim better in bond futures pool
better than in a spot gold pool or equity market pool which
they have never been in before. Survival of the fittest is more like
survival of the curve-fittest here. It should be understood that
the candlestick pattern found here has evolved within the bond
environment and thus is best-fit for it. If thrown into the spot
gold pool, these candlesharks might die like the dinosaurs did
when the cold wind blew as the Ice Age hit them so unexpectedly,
as one of the many theories goes. As in many cases in life, the finer
a design with one purpose in mind, whether natural or artificial,
the less adaptive it will be when used for other purposes. For example, a 16-gauge wire stripper, while great with 16-gauge wires,
will probably do a lousy job stripping 12-gauge wires even when
compared to a basic pair of scissors, which can strip any wire though
slowly.
The number of digits needed within each gene can be calculated. The size of a real body for a bond contract can not be larger
than 96 ticks since the daily trading limit is set to three points. A
seven-bit binary number is capable of handing numbers up to 128
and therefore needed for a gene like RB-m of C2. The 40-day trading could be no larger than 96 ticks times 40, or 3840 ticks. (Besides the fact that it seems very unlikely that the bond contract
would go up, or down, 3 points day after day for 40 days.) A 12-bit
number, capable of handing a decimal number up to 4096 is needed
here.
As the reader might have noticed after referencing Exhibit 1, a
number of trading ranges are used. As previously mentioned, the
size of the recent trading range and the candle patterns position
within the range are crucial elements that might make or break
the pattern. How does one define recent though? It then seemed
obvious to me that a multitude of day ranges is needed here, much
like the popular practice of using a number of moving averages to
access the crosscurrents of shorter and longer-term trends. I have
chosen to include 5-day, 10-day, 20-day and 40-day trading ranges,
which are more or less one, two, four, and eight trading weeks, in
my study. The advantage of using a multitude of day ranges could
be demonstrated using two examples. Let us say a morning star, a
bullish candlestick reversal pattern, was found near when the price
is at the bottom of both the five-day trading range and the 40-day
trading range, as it would if it was just making a new reaction low.
A morning star around this level is probably less useful since that
pattern is more indicative of a short-term bounce. While this morning star, a one-day pattern, could be signaling a possible turn of
the trend of the last five days, it is unconvincing that this one-day
pattern could signal an end to the trend that lasted at least 40 days.
Let us now say that the same morning star was found near the
bottom of the five-day trading range, but near the top of the 40-day
trading range, as the price of a stock would if it experienced a shortterm pullback after breaking out of a longer-term trading range.
This morning star now could be a signal for the investor to go long
the stock. The morning stars in both examples could be of the
same size, both found near the bottom of the five-day trading range,
but would have significantly different implications just because they
appear at the different points of the 40-day trading range. It should
be clear now that the inclusion of multiple trading ranges in the
decision-making process could be very beneficial.
Since each gene is a binary number, a string of 0s and 1s, each
will have a MSB (most significant bit, the leftmost digit, like 3 in
39854) and LSB (least significant bit, the rightmost digit.) Since
the MSB obviously has more impact on the candlesharks behavior,
the common state of the more significant bits among the candlesharks
are what we should closely examine after the program has let the
candlesharks breed for a while. The candlesharks main features
should be similar after a while. That is, if we do have a nice batch
of candlesharks to harvest, they should all have the same sets of 0s
and 1s among the more significant bits within each gene. The 0s
and 1s among the trailing bits are less significant by comparison,
much like the curvature of an athletes forehead should have less
to do with his speed than his torso structure. The trailing bits are
in the genes and do make a difference, but are basically not significant enough for us to worry about.
When we are ready to harvest our candleshark catches as the
performance improvement from one generation to the next decreased to a very small level, we could reverse-engineer the gene to
find the criteria that trigger their bullish signals. For instance, if
all eight candlesharks have -00011 as the first five digits in RB-m of
C2 (that means RB-m is between -00011000 and -00011111, or -24
MTA JOURNAL
Summer-Fall 2000
30
EVALUATION OF RESULTS
and -31) and 00000 as the first five digits of RB+ of C2 (that
means RB+ is between 0000000 and 0000011, or 0 and 3), these
candlesharks would only give bullish signals when the real body of
two days ago is black, and between size of -21 and -31. (The minimum = -31 + 0 = -31; the maximum = -24 + 3 = -21)
MTA JOURNAL
Summer-Fall 2000
31
and to ensure that the reproduction process was performed correctly. As it turned out, this feature had an extra benefit.
Looking through pages and pages of printouts, I every so often
spotted one candleshark with excellent performance. Since a genetic algorithm is based on the principle that nature has no memory,
this candlesharks excellent gene sequence could not be preserved.
The only traces of the sequence is in its children which might not
be as good a predictor as it was. This does occur in nature as well.
Einsteins being an incredible physicist did not imply that his children would have been, too. Even if some of his genes would live
on within his children, none might be as great as he was. A number of Johann Sebastian Bachs sons were outstanding composers,
but none as great as he was.
With the printouts in my hands, though, I could reconstruct
that one unique gene sequence. I could examine the organism by
itself and let it reproduce several times to see if it could be improved upon. Better yet, I could find two great performing
candlesharks who did not have to be of the same generation, and let
them mate, something impossible in nature. Imagine the possibility of seeing the children of J.S. Bach and Clara Schumann if they
were ever married, or turning Da Vinci into a female to mate with
Picasso. One might hesitate doing so in nature, but I face no moral
dilemma moving a few 0s and 1s all over the place.
FUTURE POSSIBILITIES
A Wonderful Example
MTA JOURNAL
Summer-Fall 2000
32
MTA JOURNAL
Summer-Fall 2000
33
gene pool one starts with, very different organisms from the
one we found might evolve. Much like different species in nature,
of which each excels in its habitat and its own way, different candlestick patterns can be found to be effective under different situations. Change the gene pool a little and let the computer go at it.
One day, your machine might find something to surprise you.
As the computers become faster and faster, one day I should be
able to include a large number of organisms. Using a multitude of
selection/evaluation criteria, several species might evolve at the
same time, just like an ecosystem in nature. I might be able to find,
after numerous iterations, one pattern particularly good for a fiveday forecast while another one is found to have an excellent oneday forecasting ability. The possibilities are endless.
Exhibit 1
Definition of Genes and Chromosomes
Chrom
Gene
Name
Name
Description
C2
RB-m
C1
C0
R5D
Format
RB+
8-bit binary
US-m
7-bit binary
US+
7-bit binary
LS-m
7-bit binary
LS+
7-bit binary
RB-m
RB+
8-bit binary
US-m
7-bit binary
US+
7-bit binary
LS-m
7-bit binary
LS+
7-bit binary
CH-m
CH+
8-bit binary
RB-m
RB+
8-bit binary
US-m
7-bit binary
US+
7-bit binary
LS-m
7-bit binary
LS+
7-bit binary
CH-m
CH+
8-bit binary
TR-m
9-bit binary
TR+
9-bit binary
RP-m
9-bit binary
RP+
9-bit binary
10-bit binary
TR+
10-bit binary
RP-m
10-bit binary
RP+
10-bit binary
11-bit binary
TR+
11-bit binary
RP-m
11-bit binary
RP+
11-bit binary
R10D TR-m
R20D TR-m
R40D TR-m
12-bit binary
TR+
12-bit binary
RP-m
12-bit binary
RP+
12-bit binary
Note: All the numbers in the genes are in ticks. A tick = 1/32
move in bond futures. Bond futures has a limit of daily movement
of no more than 3 points, or 96 ticks.
MTA JOURNAL
Summer-Fall 2000
34
Exhibit 2
Test Bed: Sample Bond Futures Data Spreadsheet
Exhibit 3
Gene Pool: Sample of Binary Bits
Exhibit 4
Gene Pool: Sample of Gene in Decimals for Printing & Reading Purpose
MTA JOURNAL
Summer-Fall 2000
35
Exhibit 5
Excel Macro Visual Basic Program:
EVOLUTION
Const pop = 8, num_gene = 38, mutation_level = 0.1, crossover_level
= 0.1
Const TB_sheet = Test Bed, GP_sheet = Gene Pool, SR_sheet =
Signal Results, GPD_sheet = GP in Dec.
Const GP_row1 = 5, GP_col1 = 6
Const SR_row1 = 4, SR_col1 = 1
Const TB_row1 = 43, TB_col1 = 7
Const generation_rep = 5, TB_sample_size = 1500, days_per_signal
= 5, min_num_signal = 20
Dim gene_signed(num_gene), gene(pop, num_gene),
top_genes(2, num_gene) As String
Dim gene_len(num_gene), limits(num_gene / 2, 2) As Integer
Sub EVOLUTION()
Set GP = Worksheets(GP_sheet)
Set GPD = Worksheets(GPD_sheet)
Set SR = Worksheets(SR_sheet)
Set TB = Worksheets(TB_sheet)
For j = 1 To num_gene
gene_len(j) = GP.Cells(3, GP_col1 + j - 1).Value
gene_signed(j) = GP.Cells(4, GP_col1 + j - 1).Value
Next j
gen_num = GP.Cells(1, 2).Value
For generation = gen_num To gen_num + generation_rep - 1
SR_rowcount = SR_row1
SR.Select
Clear enough area for new Signal Results coming in
Rows(SR_row1 & : & pop * (Int(TB_sample_size /
days_per_signal) +
2)).Select
Selection.ClearContents
GPD.Select
GPD.Cells(1, 2).Value = GP.Cells(1, 2).Value
For org = 1 To pop Evaluation each organism within population
utimate_gain = 0
utimate_loss = 0
total_gain = 0
total_loss = 0
num_signal = 0
GPD.Cells(GP_row1 + org - 1, 1).Value = GP.Cells(GP_row1 +
org - 1,
1).Value
For i = 1 To num_gene Convert binary gene bits to number
limits
gene(org, i) = GP.Cells(GP_row1 + org - 1, GP_col1 + i - 1).Value
If gene_signed(i) = (signed) Then
If Left(GP.Cells(GP_row1 + org - 1, GP_col1 + i - 1).Value,
1) =
+ Then mult = 1 Else mult = -1
shift = 1
Else
mult = 1
shift = 0
End If
num = 0
For n = shift + gene_len(i) To shift + 1 Step -1
If Mid(gene(org, i), n, 1) = 1 Then
num = num + 2 ^ (gene_len(i) - n + shift)
End If
Next n
pointer = Int((i + 1) / 2)
If i - (pointer - 1) * 2 = 1 Then
limits(pointer, 1) = mult * num
Else
limits(pointer, 2) = limits(pointer, 1) + mult * num
End If
Next i
For i = 1 To num_gene / 2
GPD.Cells(GP_row1 + org - 1, GP_col1 + (i - 1) * 2).Value =
limits(i, 1)
GPD.Cells(GP_row1 + org - 1, GP_col1 + (i - 1) * 2 + 1).Value
= limits(i, 2)
Next i
Start looking for signals
signal = NO
For i = 1 To TB_sample_size
row_num = TB_row1 + i - 1
curr_date = TB.Cells(row_num, 2).Value
num_gene 1)).Select
Selection.Sort Key1:=range(E + Trim(Str(GP_row1))),
Order1:=xlDescending, Header:= _
xlGuess, OrderCustom:=1, MatchCase:=False, Orientation:=
_ xlTopToBottom
GPD.Select
range(Cells(GP_row1, 1), Cells(GP_row1 + pop - 1, GP_col1 +
num_gene 1)).Select
Selection.Sort Key1:=range(E + Trim(Str(GP_row1))),
Order1:=xlDescending, Header:= _
xlGuess, OrderCustom:=1, MatchCase:=False, Orientation:=
_ xlTopToBottom
GPD.PrintOut Copies:=1
For org = 1 To 2
For i = 1 To num_gene
top_genes(org, i) = GP.Cells(GP_row1 + org - 1, GP_col1 + i 1).Value
Next
Next
MIXING GENES FOR THE NEXT GENERATION
GP.Cells(1, 2).Value = generation + 1
For org = 1 To pop
GP.Cells(GP_row1 + org - 1, 1).Value = Application.Text(generation
+ 1, 0000) & x & Application.Text(org, 00)
For i = 1 To num_gene - 1 Step 2
Randomize
which_parent = Int(Rnd * 2) + 1
gene_string1 = top_genes(which_parent, i)
gene_string2 = top_genes(which_parent, i + 1)
If Rnd < mutation_level Then
gene_mutate = Int(Rnd * (gene_len(i) + gene_len(i + 1))) + 1
If gene_mutate <= gene_len(i) Then
If gene_signed(i) = (signed) Then shift = 1 Else shift = 0
gene_bit = Mid(gene_string1, gene_mutate + shift, 1)
If gene_bit = 0 Then
gene_bit = 1
Else
gene_bit = 0
End If
gene_string1 = Left(gene_string1, gene_mutate - 1 + shift) &
gene_bit & Right(gene_string1, gene_len(i) - gene_mutate)
Else
gene_mutate = gene_mutate - gene_len(i)
gene_bit = Mid(gene_string2, gene_mutate, 1)
If gene_bit = 0 Then
gene_bit = 1
Else
gene_bit = 0
End If
gene_string2 = Left(gene_string2, gene_mutate - 1) &
gene_bit & Right(gene_string2, gene_len(i + 1) - gene_mutate)
End If
End If
GP.Cells(GP_row1 + org - 1, GP_col1 + i - 1).Value = &
gene_string1
GP.Cells(GP_row1 + org - 1, GP_col1 + i).Value = &
gene_string2
Next
Next
crossover = Rnd
If crossover < crossover_level Then
Do
child1 = Int(Rnd * pop) + 1
child2 = Int(Rnd * pop) + 1
Loop While child1 = child2
XO_point = Int(Rnd * num_gene / 2) * 2 + 1
For j = 1 To XO_point
new_gene = GP.Cells(GP_row1 + child1 - 1, GP_col1 + j 1).Value
GP.Cells(GP_row1 + child1 - 1, GP_col1 + j - 1).Value =
GP.Cells(GP_row1 + child2 - 1, GP_col1 + j - 1).Value
GP.Cells(GP_row1 + child2 - 1, GP_col1 + j - 1).Value =
new_gene
Next
End If
Next generation
End Sub
MTA JOURNAL
Summer-Fall 2000
36
Exhibit 6
Sample of Signal Results
BIBLIOGRAPHY
Nison, Steve. Japanese Candlestick Charting Techniques: A Contemporary Guide to the Ancient Investment Technique of the Far East. New
York, N.Y.: New York Institute of Finance, 1991.
Deboeck, J. Guido, Editor. Trading on the Edge: Neural, Genetic
and Fuzzy Systems for Chaotic Financial Markets. New York, N.Y.:
John Wiley & Sons, Inc., 1994. Chapter 8 by Laurence Davis.
Genetic Algorithm and Financial Applications.
Note: All gene definition as described in Exhibit V-1 and the
Excel Macro program Evolution written in Visual Basic are of
original work, and therefore no further reference is given. The
author drew on the knowledge and experience as an electrical
engineering and computer science major during undergraduate years and seven years as an programmer/analyst to develop
the program.
MTA JOURNAL
Summer-Fall 2000
37