Journal Editor & Reviewers

New Relative Strength Concept - An Updated Version of an Old Indicator
Nicholas Daxelhoffer

A Review of Three Risk Control Methods for the Stock and Futures Markets
Leopold A. Hauser, IV

Lead Time Analysis of Standard & Poor’s Groups for Market Peaks and Troughs
Robert C. Schuster, CMT

The Ross Multiple Moving Averages Method - Using Separate Moving Averages for
Buy & Sell Signals in Equity Markets
Donald M. Ross, MS, CFP

The OEX vs. Equity-Only Put/Call Ratio
W. Lawson McWhorter

A Comparison of Japanese Kagi Charting with Point & Figure Charting
Julia E. Bussie, CMT

Trading at the Extreme
Christopher P. Hendrix, CMT

Editor

Henry 0. Pruden, Ph.D.
Golden Gate University
San Francisco, California

Associate Editor

David L. Upshaw, CFA, CMT
Lake Quivira, Kansas

1. Introduction

1.1 Why I Chose This Subject

The point of departure for this paper was frustration.

Consider the following example: In 1986, Microsoft (MSFT) went public at an adjusted price of $0.625. Ten vears on, the stock trades at around $190 - some 300 times its original price. Compare this to the S&P 500, a common standard benchmark, which increased only fivefold during the same period - a relatively meager performance. Back in 1986, Microsoft and its spiritual leader Bill Gates were already well-known to the world. The “frivolous” notion of buying shares in MSFI occurred to me late in 1989, but my acting on this idea was blocked in a matter of seconds b: an analyst’s magical comments: “The share price has already gone up dramatically this year. It trades at an estimated PE of 38 times FY ‘91 earnings, well above the market. I think current expectations are much, much too high... We could possibly consider the stock later when the price drops back a bit. But have a look at the Baby Bells; they look terrific...!”

Microsoft never did drop - at least not enough to make it “cheap” based on traditional valuation criteria. Instead, it continued to surprise the market. I never bought that stock. Recently, someone mentioned that the market capitalization of MSFI accounts for 1.6% of the Morgan Stanley Capital International N’orld Index. And I never had a piece of that action....

1.2 The link to My Profession as an Institutional Portfolio Manager

Today a professional fund manager or institutional portfolio manager is usually measured against a benchmark. In the institutional business, this responsibility is often shared by a group of portfolio managers. They act as a group of specialists, be it for a specific asset class, e.g. “global equities,” or for asset allocation decisions. Nhen it comes to investments, the number of ideas and principles grows at least proportionallv to the number of members on the investment committee. In addition, for whatever reason, not everyone wants to be a technician. If one is a technician, however, one might follow a different method: one’s own! The need for a simple selection method - one which not only makes sense to technicians nwf fundamentalists but moreover doesn’t blatantly offend believers in the efficient market theon - seems obvious.

Essentially, what we would all like to do is to identify and purchase securities whose performance potential is strong, while selling those with weaker potential. In technical analysis (TA), this concept is very well-known - it is called relative strength (RS) - and it perfectly suits the need of an investment manager who is graded on relative performance. I wish to elaborate on why it is sensible, as well as profitable, to support decisions based on a modified and somewhat extended version of this classic TA concept, relative strength (RS).

The development of a global financial market has created the opportunity to apply the concept of relative strength between different markets and asset classes. One might say that all assets belong to the same, global market, but I believe we should apply the concept sensibl! and make sure that immtors sharp the vim that the assets belong 10 fhP same market. The concept of RS should only be applied when the assets in question have a sufftcient number of risk factors in common. From the TA point of view, past experience plus visual inspection - especially based on inter-market analysis - help establish the relationship between the assets. For example, an interest-sensithe group of stocks such as utilities has a sufficient number of risk factors in common with the bond market and the stock market. Based on the findings of Modern Portfolio Theorv (MPT), this subjective definition can be quantified by determining a meaningful correlation between the items in statistical terms.

2. The Concept of Relative Strength (RS)
2.1 Basic RS Indicator

Relative strength (RS) measures the relationship between two financial series. It compares the price movements of two financial assets to answer the question ‘Which one is performing better?” RS is calculated bp dividing the price of one item by another. It is most widely used by comparing an item lvith an index, usually “the market,” as measured by a stock market a\-erage (table 2.1). RS is therefore a powerful tool for selecting stocks that maIoutperform a given benchmark.

2.2 Critique and Shortcomings

Comments author Richard Brealey on RS: “In bull markets it is the risti);stocks that generally have the highest returns; in bear markets it ts the safest stocks. Therefore, by following the relative strength rule, an investor will tend to invest in risky stocks after a market rise and in safe stocks after a market fall.” (Richard A. Breale); p. 14f).

One of the most significant disadvantages is probably the fact that the indicator does not take into consideration the’different risk levels of the assets to be compared. In other words, the danger exists that in a selection decision based on relative strength, preference will be given to riskier or defensive assets. This depends on the development of the market as a whole, of course. Expressed differently it is possible to confuse the higher return of an asset compared to the market with the higher (or lower) risk which one is undertaking.

2.3 What is Expected of an RS Indicator

Based on the anahsis of the weaknesses inherent in the familiar RS indicator, a ne\k indicator must fulfill various requirements.

1. Any newly-formulated Relative Strength Indicator should overcome this primary criticism insofar as it should also take risk into consideration (Beta in the Slodern Portfolio Theon sense).

2. It is generally recognized that er-en Beta is not always stable over time. The observed Beta is only an estimation of the future expected volatility For this reason, the new indicator should be supplemented by information which is independent of Beta.

3. The indicator should make it possible to recognize broad price movements which heavily influence the indicator itself.

4. Beyond these requirements. it should be possible to compare the effect of erratic relative price movements, often caused by news which the market suddenly becomes aware of and is often registered in “price spikes, ” with the effect of continuous, “normal” price movements. The objective is to recognize the signs of relative strength which signal continuous accumulation.

3. Integrating the Capital Asset Pricing Model (CAPM) into Relative Strength/Technical Analysis

3.1 Completely Different Assumptions

If we compare their basic tenets, Technical Analysis and Modern Portfolio Theory (SIPT) are as alike as fire and water: one believes in price trends, while for the other, it’s a random walk.

In the purely; academic sense of Modern Portfolio Theorv, integration of a concept based on technical analysis is not possible. But markets “must be made efficient.” Someone in the market will act, and these actions will leave “footprints” for the technical analyst to discover and study. On the other hand, the technical analyst should not deny that different stocks have different risk characteristics which should not be overlooked. If one could integrate a risk factor (= volatility) into the relative strength indicator, the result would be a more modern concept of RS, because it would take into account one of the most important findings in finance: the relation between risk and return.

3.2 CAPM: The Concept of Risk and Return

Security returns are correlated because of their common response to market changes. A useful measure of this correlation is obtained by relating the return on a stock to the return on a stock market index, which is a key assumption of the (CAPhl) Capital Asset Pricing Model. See End Note for the CVM equation

This relationship describes the expected return for all assets, and portfolios of assets, in the economy The c$mf/ewebd wtum on cry two mvts or portfolios can bv &ted simj$ to the differewe in their B&z coQ?rients. The higher the Beta (“risk”) is for any security the higher must be its equilibrium return. Furthermore, the relationship between Beta and expected return is linear.

3.3 Beta-Adjusted New Relative Strength (NRS)

Given the formula for RS:

where: R(X) is the return on asset .\ for a period, and the formula for the return on stock X according to the security market line in the CAPRI:

where the expression ( 1 -beta (A) ) * ( R(M) - R(F) ) represents the return of stock X due to its Beta, the market return, and the risk-free rate, it is possible to combine the terms as follows:

or, transformed into the Beta-adjusted or New Relative Strengthformula:

This formula allows one to calculate the relative return of a stock A relative to the market hl (whose Beta = 1) adjusted for the stock-specific risk Beta and the risk-free rate. This addresses one of the core problems regarding the application of the relatil-e strength concept in technical analysis.

3.4 New Relative Strength is a Concept

In order to resolve the conflicting efforts to find a way to represent relative price changes as trends, on the one hand, and as individual& occurring price spikes on the other hand, additional depictions are required. It is not possible to summarize all the manifestations of relative strength in a single indicator. For this reason, it is reasonable to call this a “relative strength concept.”

4. Problem of Price Spikes
4.1 Analysis of Price Spikes

As already mentioned, RS indicators have the disadvantage of being heavily influenced b!- significant price movements. which means that the evaluation of the result is made more difficult. As a rule, such relative price movemen& arise when news becomes known that substantially deviates from the expectations of market participants. Typical examples are the announcement of quarterly results, merger intentions. etc. - in other words, e\-entswhich are of a non-systematic nature and do not follow the development of a trend. Severtheless, it is interesting to obsene that positive quarterly results, for example, lead to an increase in price and as a consequence begin to improve the relative strength. In this case, it may be assumed that investors are surprised b!- the results, and, as a consequence, analysts proceed to revise their earnings estimates upward.

Because it would be too costly to determine and mark significant price movements indi\iduall!; a procedure should be applied to this end which, on the one hand, can deal with processing large amounts of data - such as used to produce charts - and, on the other hand, can determine and depict significant price movements according to an objective statistical process.

The procedure used to determine significant price movements is based on standard distribution, whereby in this case, the signal to noise ratio can be pre-selected bv inputting the sensitivity. For the calculation of spikes, it is not significant whether Beta-ad&ted price changes are used or not. To work with the data further, however, adjusted values are required. X sensitivit!- of 0.2 means that 20% of the price movements - i.e. the largest 10% and the smallest 10% - are represented.

The application and interpretation of the “spikes” indicator series are described as follows: The application of the frequencl depends on the universe. If liquid stocks are compared to large universes, a value of 20% is recommended as a rule. If, within the framework of a particular selection, similar stocks are being compared with a specific universe, this value should not be varied.

As has already been mentioned, spikes are of interest because they make it possible to observe the reaction of the market before and after. A spike acts as a signal, a challenge to market participants to deal intensively with the stock. The reaction to an announcement of news, and ;he development shortly thereafter, provide valuable clues as to the positioning of active participants in the market, and thus to the possible further development asso ciated with it.

4.2 Cumulative NRS of “Spikes” versus “Non-Spikes”

By depicting, in cumulative form, periods with price spikes on the one hand and “normal price” movements on the other, certain conclusions may be drawn.

Price spikes occur when:

• important new information about a company or its business environment becomes known.

• a company is “discovered” by the media, causing most investors to react unprofessionally as a rule, i.e. orders are given to buy without specif$ng price limits.

• market-technical factors are unfavorable, for example liquiditv levels are low. Price spikes are signs of large imbalances bktween supply and demand, and can be the trigger for other market participants to move their position or revise their attitude or opinion.

In evaluating such price spikes, we base our judgment on the assumption that professional investors avoid price spikes as a rule.

The representation of cumulative normal price movements allows certain conclusions to be made concerning constancy, continuity, and quality, which can be used as the basis on wh’ich to build up or reduce market positions. A steady flow of funds into a stock is termed “accumulation” and is a sign that:

• professional investors and fund managers, etc. are building up (or reducing) their positions.

• the majority of investors, influenced by the flow of information emanating from the media and encouraged by the trend, behave in the same way.

Observing the combination of these two effects provides clues to shifts in equilibrium between supply and demand, and/or to shifts in the balance of opinion of market participants. We can sume that:

• in the formation of a floor for a stock, negative price spikes that are triggered bp bad news gradually yield to accumulation.

• during the phase in which prices are rising, positive price spikes are at first accompanied bp an increase in accumulation.

• in the topping-out phase, price spikes are no longer confirmed from continuous price movements.

• at the turning point, and during the downward phase, negative price spikes gain prevalence and are accompanied bp an increase in continuous negative price movements.

The depiction of these two groups of price movements (spikes and normal movements) lends itself to an in-depth analysis of the quality of relative strength, in particular in connection with the behavior of Beta-independent relative price movements, described in the next section.

4.3 Independent of Beta: The Frequency/ Persistency of Strength

So far, the indicators that comprise the new relative strength conept (NRS) are based solely on relative price momentum.

A further indication of relative price strength can be expected by observing the frequency of occurrence. This could give some additional insight about the quality and persistence of a stock displaying relative strength.

A common measure in T-1 is the number of advancing issues minus declining issues for stocks. Based on this concept, the formula applied to measure the number of "up" day to “down” days relative to the market is:

The formula does not take into account days unchanged.

The interpretation of this indicator follows the general principles of TA. One can make the following assumptions:

• A stock in a bullish phase will rise more often relative to the market than it will fall. The reason for this, according to technicians, is that a stock under accumulation and in a rising trend should display lengthy periods of successive rise in price and rather short periods of a corrective nature. This price movement is called “right-sided,” and after each correction the data series should end higher than at the beginning of the correction.

• A stock in a bearish phase will therefore tend to fall more often, while rising strongly only for short time periods. The reaction (pull back) will end lower, and the price movement is “left-sided.”

• The frequency observation, together with “price spikes” and “continuous price action,” hints of the stock’s relative strength behavior. For example, a stock displaying good relative strength based on continuous price movements and a rising number of up periods will be rated superior to a stock displaying good NRS based only on significant price spikes but which lacks follow-through.

Based on different time horizons, and together with RS, NRS, and price-spike analysis, the frequency analysis should offer additional clues about the future price and rela&e strength action of a stock. The different analytical methods combined should be subjected to divergence analysis.

5. Putting It All Together

5.1 An Elementary Presentation of the NM and Related Concepts

Based on the reflections we have examined thus far, the Sew Relative Strength concept can be expressed using four different charts.

Graph 51.1 plots RS and Beta-adjusted RS based on the formula spelled out in this paper. To calculate and display Beta-adjusted SRS, the risk-free rate (i.e. T-Bill yield) and the Beta are both required variables.

Graph 5.1.2 contains what I term “price spikes,” a function that allows the inspection of the relative price movement between two assets on a day-to-day basis. The program tracking these data allows the technical analyst to predetermine the “intensity” of the spikes he/she would like to see, based on the assumption that relative price movements follow a normal pattern of distribution. A setting of 0.2, in other words, will select only those relative price movements that are among the most extreme 20% of all such movements during the period under review.

Graph 51.3 depicts the number of “up” periods in relation to the number of “down” periods, without taking into consideration the actual amount gained or lost. It is based purely on the direction of one variable (e.g. a stock) in relation to the other (e.g. the market). This observation is independent of Beta, as it treats highly volatile stocks exactly the same as stocks with little volatile BARRA risk factor model. Purely statistical calculations such as those one can lind on Bloomberg (function MSFT Beta) are much less reliable.

Graph 5.1.4 tracks the cumulative return on all “spike” days compared to all other days. This should provide additional insight into the quality of Relative Strength, since it explodes the assumption that RS based on “spikes” - i.e. with no clear follow-through - indicates a lack of trend and persistence. Here, a clear trend based on “non-spike” days can also be seen, a valid trend which might be exploited.

Input for Chart Construction

The aforementioned graphs are presented for one year, using daily data (264 trading days, from August 18, 1996 to August 18, 1995).

Monthly expected Beta coefficients were taken from BARRA. These are about the best Betas one can find because these values are calculated based on the BARRA risk factor model. Purely statistical calculations such as those one can lind on Bloomberg (function MSFT Beta) are much less reliable.

Computations and graphic presentations were made by the author using the MS Excel 5.0 spreadsheet program. Data were downloaded from DataStream into a spreadsheet.* it$ ,i forward-shifted llday moving average is also calculated (smooth line).

5.2 Selected Chart Examples
(With Interpretative Comments)
(See pages 14-18)

6. Conclusion

In Section 2.4 of this paper, I enumerated four essential requirements that should be fulfilled by the New Relative Strength indicator, in order for it to be deemed successful. Let us examine the extent to which the new concept addresses these points:

First and foremost, by finding a Icay of combining the all-important notion of risk (as defined in the Capital Asset Pricing Model) with the traditional Relative Strength indicator, the primary criticism against applying RS in the process of selecting securities has been swept away. Examples given here show clearly that a new indicator can be constructed and applied, but that, as with any other technical indicator, it should not be used in isolation.

Several observations concerning the risk-adjusted Relative Strength indicator can be made, and the following conclusions drawn.

A positive Beta, i.e. one which is higher than the market, is a serious challenge to the Relative Strength of a stock, on a risk-adjusted basis. The NRS therefore shows that one would have been better off to have bought “the market” on a leveraged basis (e.g. the S&P 500 Futures contract) than to have invested in the stock.

As has been seen, Beta is not stable over time. The value of Beta is very much dependent on the source chosen and the length of the time period during ivhich measurements are taken, and the frequency of the measurements. Fixing even slightly different starting dates can have a significant impact on the outcome of Beta. Selecting the “right” Beta therefore becomes the crud challenge, especially berause XRS is extremely sensitive to changes in Beta, whereas the level of the risk-free rate has only a minor influence on the outcome.

In academia, it is a well-known fact that high-Beta stocks tend to perform less well than one would expect, given their stock-specific risk; for low-Beta stocks, the opposite is true. Risk-oriented investors tend to overpay in order to participate in the growth they expect from high-risk stocks.

It has been my observation that stocks with a very low Beta appear to have “lost contact” with the market. Actively traded stocks with very low Beta coefficienh often fall into the categoy of “fallen angels,” i.e. distressed situations. Their low Beta should serve as a warning signal. Furthermore, if they show up positively using SRS, then the traditional RS indicator should be carefully reviewed as well!

With respect to the application of Beta in technical analysis in the first instance, the findings mentioned above have a very important implication, viz. that the risk-adjusted Relative Strength indicator should be applied with greater confidence for a group of stocks (industry, sector) that share common characteristics than for an individual stock.

Given the many problems associated with Beta, it is obvious that this new indicator - as interesting as it may be - will still not be the Holy Grail of technical analysis, and that it is best presented together with other indicators, all of which can point to the existence of Relative Strength. Of these, traditional RS is surely one, since at the end of the day we wish to outperform the market on an absolute basis. Frequency or the number of days advancing relative to the market, is another important variable to keep in mind because it is only dependent on the relative direction, yet independent of the size of the movement.

These additional indicators/charts, such as “spikes” and “Cumulative spike & non-spike return” should not only, help to detect movements but more importantly make it possible to separate erratic price movements from persistent price action.

Although a large amount of data is presented here, detail is maintained where necessary and aggregation is made where possible, without clouding the detail. The different aspects of Relative Strength are well segregated, yet visually connected - this allows the analyst to do trend and momentum analysis and convergence/divergence analysis.

In sum, the New Relative Strength Concept allows the technical analyst a deeper and broader insight into Relative Strength acti\itl\; since it highlights man! of the different aspects of Relative Strength. Today, when being able to ascertain relative performance is so crucial, no other standard indicator or graphical representation presently known offers such a unique possibility.

Reference

• Richard A. Breale!; An Introduction to Risk and Return from Common Stocks, Second edition, Basil Blackwell, Oxford Great Britian, 1983
  

This equation merely breaks down the return on a stock into
three components: the part due to the risk-free rate, the part due
to the market, and the part which is independent of the market.
Beta is therefore the measure of how sensitive a stock’s return is
to the return on the market as a whole. X Beta of 2 means that
the stock’s return is expected to change br 27/eweb when the market
is changing by 1%.

Biography

Nicholas Daxelhoffer was born in Zurich in 1955. He studied at the University of Zurich and was graduated with a master’s degree in business science. Before joining Pictet 8: Cie he worked at Lbntobel &: Cie as a portfolio manager and analyst in the research department, where he was responsible for the beverage and brewen sector. ;Zt Pictet 8c Cie, he became a senior portfolio manager for institutional clients and currently manages LSD $500 million in Swiss pension fund money from various institutions. Since 1993 he has been a permanent member of the Asset Allocation Committee of the Institutional Department.

*Nicholas Daxelhoffer can be reached at ndaxelhoffer @pictet.com and is willing to send you a chart package which is very interesting to this study

In his book Market Wizards, Jack Schwager mentions that risk control is one of the most important concepts of successful investing. Dr. Van Tharp has stated, both in his seminar and witten materials, that risk control could be even more important than the selection or timing method itself.

Why is risk control so important? First, remember that investing can be viewed as nothing more than a statistical game with a consistent set of components.

Some of these components include percent winners, percent losers, the size of the average win and loss, and the ratio of profitable to unprofitable trades.

In addition, the size of your position will have a direct effect on the results of vour system. For example, if your system is 50%’ correct in a series of ten trades, you should expect to have fire winners and five losers. If you then decide to risk $20,000 of a starting $100,000 account on each trade and the first fire trades are losers - you are wiped out. Nothing happened to your systern. The statistical components did not change. It was vour risk control or lack of it that caused the result. I realize this’is a simplistic example, but it happens every day to investors who have not looked at ways to control risk in their accounts.

Risk control is critical to trading success. Therefore, I have chosen three examples of risk control from three different sources as mv fi/ewebcus. This is not a definitive view of any of the three metgods. It is intended to review the topic and to give the reader a flavor of three different approaches to the same problem.

Protect Yourself and Your Account

William J. O’Neil, in his book How to Make Monet in Stocks,and in his seminars, outlines a step-by-step approach for both en- try and exit of a position. He does this bp purchasing only stocks that are breaking out of basing patterns that are at least six to eight weeks long. In the simplest sense, the highest point of the recent base is the entry point, or as hlr. O’Neil calls it, the “pivot point.” He cautions that you should not buy more than 5% above the “pivot point.” This prevents you from chasing a stock too far above its base. As a confirmation, he also looks for at least a 50% increase in average daily trading volume on the day of the break out.

Mr. O’Neil also has strict rules for his position size. Although he suggests a total of between four to five positions for a $100,000 account, he does not recommend putting the whole position on at one time. In his seminars, he discusses the technique of breaking your buying decision into two or three different purchases. For example, you could enter onlv 30% of your position on the first buy. Only if the position coniinues in your direction do you add another 30%. Then, if the stock still looks good, add the final 20%. This entry approach has a couple of distinct advantages. First, you reduce the risk of the first purchase. This has great monetary and psychological advantages. Second, by making your first purchase the largest, and by not buying more than five percent beyond the pivot point, you keep your average cost low. Finally, because all stocks will not follow through for the second and third buys, more money is forced into the strongest stocks.

Mr. O’Neil’s sell rules show the same precision and forethought. His first and most important rule is to always cut losses at 7-S%. Therefore, in a five-position account you are only risking 1.6% of the total portfolio value per stock. A great number of your trades will have less risk exposure because YOU are never able to build, through three consecutive buy, up to a full position. You have the advantage of a highly concentrated portfolio, with very manageable risk.

In reviewing the entry and exit criteria, you will notice an interesting relationship. You never buy more than five percent past the “pivot point” or breakout, and you cut your losses at eight percent. In most cases, you would expect resistance to become support. So if you are buying upside breakouts out of consolidations you would expect the strong stocks to never reenter the base. Because you are not allowed to chase a stock more than 5% after a breakout, cutting your losses at 8% means the stock would have to reenter the base by 3% to hit that stop point. Therefore, reew tering the base becomes an obvious caution sign.

He also makes use of a time stop in his rules. If neither an 8%’ loss or a 20% gain takes place, sell the entire position in 13 weeks. This forces you to have patience and allows the move to unfold. Again, like a number of his techniques, this one forces the mane! to stay in only the strongest stocks.

Once you have a profit, how do you handle it? Mr. O’Neil supplies us with rules here, too. His first concept is a profit objective. If you are fortunate enough to have a 20-25% gain - take it. This allows TOU to be wrong three times and right only once and still break even. He later added an exception to this rule. It states, “If the stock was so strong that it vaulted 20% in less than eight weeks, the stock had to be held at least eight weeks.” You should reexamine the stock for a longer, stronger more. This technique forces the money to stay in the strongest stocks.

Let us take a moment to look at the performance possibilities of these rules. Since actual stock selection includes criteria that are beyond the scope of this article, let us make the following assumptions. First, let us assume that one third of your trades are winners. This is in line with most breakout or trend-following systems. Let us further assume that one third of your trades are losers and the final one third break even. Finallj, these illustrations do not account for slippage or commissions. In the first scenario, we will presume your winners have a 25% gain and your losers an 8%’ loss. Your expected return would be 5.67% per quarter, or an annual compounded return of 24.67%. The second scenario is more conservative, with one third of the trades posting a 20% profit, one third posting an 8% loss and one third breaking even. That equates to a 4% gain per quarter, or a 16.99% annual compounded return.

The advantages of this approach to risk are many. Vrtually any system in which the size of the winners is three times the size of the losers greatly increases the chances of overall success. The rules are very co&se and easy to follow. There is verv little subjectivity in the rules. Additionally, there is a rule for virtuallv evet-y scenario - win, lose or break even. Patience is built intd the rules. If vou do not reach a 20% gain or an 8% loss, you need to stay with the trade for at least 13 weeks. The rules follow the old trading saying “cut your loses short, let your profits run.” Finally the incremental position building, price objective, price objective exception and time stop components all force more mane! into winning positions.

However, there are a couple of disadvantages that the reader should be aware of. Incremental position building, although safer, increases commission costs. And because the stops are placed at fixed percentages, they do not take into consideration the normal volatility of a stock, and you could be stopped out in a “normal” reaction.

The One Percent Rule

The one percent rule is referred to a number of times in Jack Schwager’s books Market M’izards and The New Market Wizards. It is a fairly simple concept that allows you to equalize risk exposure across a portfolio, adjust position size based on account equity and monitor market volatilitv.

And all this can be done with B simple two-part equation:

A) Entry - Exit = Spread

B) Dollar Risk / Spread = Number of Shares

To begin, let me define the components of the equation. The first is what I will refer to as the spread, which is simply the entry point minus the exit point. This number will force you to know where you will get in and out before you place the trade. The advantage here is you can let the market tell you where to get in and out. You can use volatility, support and resistance, or whatever tools you presently use to define these points. For example, Lou want to buy if there is a breakout above resistance at 20, and if the price goes below- support at 18, you will sell. The first part in our equation is the buy point (20) minus the sell point, (18) which gives us a spread of 2.

20 (buy point) - 18 (sell point) = 2 (spread)

The second part of the equation determines the size of your position and the number of shares to purchase. To start, take the value of your account, which we will refer to as equity Equity can be calculated in two ways. The more conservative is to use only closed equity defined as closed or cash positions. The other, more aggressive way is to use current equity and include open positions in your calculation. This means vow equity would fluctuate on a daily if not hourly basis. You c&d moniior this fluctuation to fine tune your risk exposure.

Once you have determined your equity, you simply take that amount times a fixed percent to determine dollar risk. The traders profiled in Market U’izards recommend no more than l-2%. For example, using the 1% figure and a fixed equity of $100,000, your dollar risk would be .Ol x $100,000 = $1,000.

.Ol (fixed percent) x $100,000 (account equity) = $1,000 (dollar risk)

This percentage can be adjusted down as your account size grows. Ifyou find it too difficult or too stressful to trade a SlOO,OOO account that has grown into $l,OOO,OOO, simply reduce the percentage to a position size that is comfortable. However, I have never seen an increase in risk exposure recommended. This is especially true during in a losing streak. That would be the equivalent of lowering a stop on a losing long position - a cardinal sin in successful trading. This brings us to another benefit of the technique. It \+ill automatically adjust your position size down during a losing period and up after the winners.

As your equity declines, so does your dollar risk. If, for example, your account drops from $100,000 to $90,000, vour dollar risk would now be .Ol x $90,000 = $900. Your position’is adjusted for winning trades as well. If, after a winning streak, the accomit is up from $100,000 to $110,000, your risk size would increase to ’ .Ol x $110,000 = $1,100.

The final step is to divide the dollar risk by the spread. This will give you the number of shares you will want to buy for the position. To use the above example $1,000/2 = 500 shares.

$1.000 (dollar risk) / 2 (spread) = 500 (number of shares)

Please note that your position size and dollar risk are two completely different numbers. Some systems start with position size and use a fixed percentage to determine dollar risk or stop placement. For instance, the O’Neil rules stated above use this technique. The 1% rule method approaches the problem from the opposite direction. It starts with dollar risk and uses spread to determine position size. 500 shares at a buy stop of $20 gives us a position size of $10,000.

$20 (buy point) x 500 (shares) = $10.000 (position size)

And with a protective stop at $18, you are only risking a $2 spread per share, or a total dollar risk of $2 x 500 = $1,000 on this position.

$2 (spread) x 500 (shares) = $1,000 (dollar risk)

There are many advantages to the 1% rule. It can help tremendously with psychological balance. “If I onlv lose 1% of my account value, do I care?” There is a great degree of comfort in knowing that the most risk you are exposing yourself to, on any trade, is 1% of your account value. The result is, over an infinite series of trades, the account value will theoreticallvnever hit zero. This situation is not adjusted for slippage and commissions, but it is a comforting thought none the less. Finally, no one position can wipe vou out. Granted, this method ma!; at times, force VOLI to take pdsitions smaller than you normallv would. All that mdans is that it may take Lou a little longer to get to your goal. And vow chances of reaching that goal are increased because vou have a greater likelihood of being able to continue to trade. ’

Beyond the psychological advantages, this technique. by its yen, nature, measures and adjusts for market volatility. This can help /ewebOLI in your market analysis. Accumulation periods are general11 associated with quiet markets and narrow price spreads. This lo& ered volatilih would lead to a larger position. On the other hand. distribution is generally associated with active markets and wide price spreads. Markets tend to be more volatile at tops. The increased volatility would lead to smaller positions. Monitoring the change in spread size or volatility could help to determine what stage a market is in.

The adjustment in position based on volatility helps to build a balanced portfolio. The portfolio is based on individual position volatility and risk. The simple calculation and application of this technique results in a risk-balanced portfolio. Ltid, as stated before, there is a definite comfort in knowing that no one position can wipe you out.

Your position size is adjusted, both LIP and down, with your equity. This allows you to increase !-our position size during winning streaks, and, more importantl!; decrease your size during losing streaks. The result is a mechanical method which tells you when to he more or less aggressive with your positions.

For all of its good points, there are some disadvantages to the 1% system. You might find your position sizes too small at times, which will lower your overall rate of return. But, as mentioned earlier, smaller positions simply mean that you will still reach your goal; it will just take a little longer. One definite disadvantage is higher proportional commission charges for the possibly smaller positions.

Diversification

The final way we will look at controlling risk is through diversification. Standard diversification is accomplished by spreading vour positions amongst different stocks in different industry groups. Another type of diversification is that of system diversification.

In system diversification, you can design selection models with different criteria. For example, you can have a growth stock model and a value stock model, and divide your capital between the groups of stocks. Another option would be to build two market timing models. You could design one trend-following model using moving averages and one counter-trend model using oscillators, and, as with the previous example, divide your capital behveen the two models. The idea is that one view of the market will not work all the time. By dividing your approach to the market, you may stand a better chance of being consistently profitable.

Another view of system diversification is to use the same systern, but different sources of data. You could use intramarket or intermarket data and compare them to the market vou are trying to analyze. Intramarket analysis would be based on’internal data. The price of the market itself, advance/decline, new high and low figures would all be examples of internal or intramarket data.

Intermarket analysis, on the other hand, would be based on external market data and its relationship to the market vou are tyi,ng to analyze. One interesting advantage of intermarket analysis IS that it’s built on at least hvo different data sources or price series. Most intramarket analysis tools are built of all the same price data. So, an aberration in prices would effect all the indicators or tools based on those data. By using hvo different data sources, you can help to insulate yourself against such an event. A common example of intermarket analysis is the study of interest rates and the stock market.

To demonstrate an intermarket model, I will use a simple trendfollowing model using the long-term Treasury bond yield and the S&P 500. The benchmark system is the S&P 500 and a 40-week simple moving average. The rules are quite familiar: if the weekly close of the S&P 500 is above its 40-week simple moving average, buy at the next Monday’s close. The sell rules are inverse. If the weekly S&P 500 closes below its 40-week simple moving average, then sell at the next Monday’s close. The calculations of the systern contain no adjustment for slippage or commissions. M’hen the system is bullish, dividends are reinvested. When the system is bearish, we get the interest on go-day commercial paper, adjusted to reflect quarterly compounding. The results of the Benchmark System and a Buy-and-Hold comparison from August 15, 1942 to January 23, 1998 are as follows:

The intermarket analysis diversified system rules add the longterm Treasury bond vield as an entry and exit filter to the benchmark system presented above. The filters are based on the observation that declining interest rates are generally positive for the stock market. If interest rates are declining, let the long-term direction of the S&P 500 alone determine your market position by following the buy and sell rules of the benchmark system. Rising interest rates are normally considered negative for the stock market. In an unfavorable interest rate em-ironment, do not to enter any new positions and sell any existing positions at the first sign of intermediate-term weakness in the S&P 500.

The buy rules are these: if the weekly long-term Treasury bond yield closes below its 40-week simple moving average and the weekly S&P 500 closes above its 40-week simple moving average, buy the S&P 500 at next Slonday’s close. If the weekly long-term Treasury bond yield closes above its 40-week simple moving arerage, do not enter any new positions.

The sell rules have two distinct conditions based on the direction of the long-term Treasury bond yield. First, in the benchmark system, a close of the S&P 500 below its 40-week simple moving average is always negative for the stock market, regardless of the direction of’interest rates. Therefore, if the close of the weekly long-term Treasury bond yield is below its 40week simple moving average and the weekly S&P 500 closes below its 40-week simple mating ayerage, then sell the S&P 500 at next Monday’s close. Second, if the close of the weekl!: long-term Treasury bond yield is above its 40-week simple mobmg average and the weekly S&P 500 demonstrates intermediate-term weakness b) closing below its lo-week simple moving average, then sell the S&P 500 at next Monday’s close. The calculations of the system contain no adjustment for slippage or commissions. When the system is bullish, dividends are reinvested. When the svstem is dearish, we get the interest on go-day commercial paper, adjusted to reflect quarterly compounding. The results of the Intermarket System and a By-and-Hold comparison from August 15, 1942 to January 23, 1998 are as follows:

The addition of the long-term Treasury bond yield entry and exit filters to the benchmark system lowered the number of trades from 79 to 43. In addition, itincreased the percentage of profitable trades from 53% to 71% and reduced maximum equity draw down from 32.4% to 9.8%. It accomplished all of this while only lowering the percent return of the benchmark ystem from 12.6% to 12.3%.

You can use system diversification in a number of ways in overall system design. One way to approach the problem may be to use the “stoplight” analogv mentioned in Market \Vizards. If all of your subsystems say “gd” then you have a “green light” for action. If some subsystems say “go“ while others say “stop,” you have a “yellow light” and should proceed with caution. And finally if all of your subsystems say “stop” then you hare a “red light” and should take no action. The “stoplight” approach can be incorporated into both buy and sell rules, as well as position size rules.

The “stoplight” interpretation is used in the following manner in the model above. If the S&P 500 is above its 40-week simple moving average and the long-term Treasun bond yield is below its 40-week simple moving average, then both are “go.” If both subsystems are a “go,” the system has a “green light” and you can take all trade signals.

If the S&P 500 is above its 40-week simple moving average, it is a buy or “go.” If the long-term Treasury bond yield is above its 40-week simple moving average, then it is a sell or “stop.” If one subsystem is a “go” and the other is a “stop,” then the system has a “yellow light” and you should proceed with caution. M’hen the above model has a “yellow light,” you would not initiate any new positions, and you would liquidate if the S&P 500 crosses below its lo-week simple moving average and generates a sell signal.

Finally if the S&P 500 is below its 40-week simple moving average and the long-term Treasury bond yield is above its 40-week simple moving average, then both are a sell or “stop.” \\hen both subsystems are a “stop,” then the system has a “red light” and you should liquidate all positions.

The “stoplight” analogy can also be used to interpret position size. \+71en the system has a “green light,” you invest 100% of Tour position size. If the system has a “yellow light,” you could mrest 50% of your position size on the bu? signals from the S&P 500 and keep 50% in cash or Treasury bills. Finally, in a “red light” situation you would not invest in the S&P 500. Instead, you would keep 100% of vow- position in cash or Treasury bills.

System divers&a&on has many advantages as demonstrated above. These include risk reduction and reduction in the number of trades. Additionally, because of the reduced number of trades, you will also see lower overall commission expenses. The disadvantage of system diversification as we have used it here is quite obvious. You need all the model subsystems to confirm before you enter or exit a position. This may cause you to enter a market late or turn cautious too early in a moye.

The top traders agree that risk control is very important. The three examples presented here are br no means the only approaches available. They were selected for their various attributes and can be used independently combined, or as a framework for additional research. The O’Neil model is very complete. His methodology provides an answer for just about eve,?; scenario. The 1% rule is very interesting because of its simphclty. There are not many market solutions that are so easy to calculate and offer so many benefits. Finally the use of diversification can force you to look at the market in a number of different ways. This change of perspective can possibly offer you a more complete view of the market you are analyzing.

The key to using any form of risk control is to understand that it may not give you the highest returns on any given trade, but it will allow you to keep trading and stay in the game. I found this attitude best summarized in the Berkshire Hathaway Inc. 1991 Annual Report: “As one of the Indianapolis ‘500’ winners said: ‘To finish first, you must first finish.“’

Bibliography

• Arnold, Curtis, M., Curtis Arnold’s PPS Tradinp System : A Proven Method for Consistently Beatine the Market, Irwin Professional Publishing, 1995

• Buffett, MBrren E., Berkshire Hathawav Inc.. 1991 Annual Report, FVarren E. Buffett, 1992

• O’Neil, M’illiam, J., How to Make Monet in Stocks: A Winning System in Good Times or Bad, McGraw-Hill 1988, 1991

• V’illiam O’Neil & Co., Seminar -Advanced Investment Workshop, Investor’s Business Daily Inc., 1993

• Schwaper, Jack, D., Market Wizards: Interviews with Top Traders, NYIF Corp., 1989

• Schwager, Jack, D., The Sew Market Wizards: Conversations with America’s Tar, Traders, HarperCollins Publishers, Inc., 1992

• Tharp, Van K., Ph.D., The Investment Psvcholog) Guides, Invest ment Psycholop Consulting, 1984, 1989

• Tharp, Van K., Ph.D., Seminar - Del-eloDinc a bYinning Investing/Trading System that Fits You, International, Institute of Trading hlastery Inc., 1996, 1997

• Tharp, Van K., Ph.D., Course Update ReDort 29: Understanding Money Management, I.I.T.M., Inc., 1995

• Tharp, Van K., Ph.D., Course UDdate ReDort 30: Money Management: Part II, I.I.T.M., Inc., 1995

Historical Testing

All historical simulation and testing for the Benchmark System, Intermarket System, and Buy-and-Hold models were performed under contract to Formula Research, Inc. (901/75&8607).

Biography

Leopold A. Hauser, F is a Senior Registered Representative for Vanguard Brokerage Services. He has used and researched technical analysis for over ten years. He has completed the CFP Professional Education Program, and received his Bachelor of Science from the University of Minnesota

Introduction

Historical patterns in the stock market can shed light on current conditions as well as increase the possibiliy of recognizing future conditions. An overbought/oversold oscillator identifies tendencies (or turning points) in market behavior when the oscillator reaches certain parameters. In the same rein, this paper looks at historical tendencies of various Standard & Poor’s stock groups around stock market peaks and troughs. The purpose is to find Standard & Poor’s (S&P) stock groups that have tended to peak prior to market peaks and to bottom prior to market troughs. If such groups exist, then one can pay closer attention to their price action when it is diverging from the market’s price action. Also explored is group price action around recession-related and non-recession-related bull and bear markets, to see if different groups may have leading tendencies in stock market cycles within different economic backgrounds. Also tested is a composite of the best groups, called a Leading Group Index, to see if the leading tendencies of a composite index improved compared to the individual groups (based on the percent of market peaks or troughs led and/or lower variability of results).

The broker and bank groups are mentioned quite often in market shoptalk and financial literature as having leading tendencies. Does the S&P Brokerage group historically have a leading tendency? Since 1978, the S&P Brokerage group led market peaks in three of five cases, or 60% of the time. Since 1978, at market troughs the S&P Brokerage group led once, was coincident once and lagged four times; however, each of the lags had been bp only one week (see table in Example of Output section). These results did not render the S&P Brokerage group as one of the better leading groups tested using the criteria established for this paper (see Discussion of Criteria section).

What about the bank groups? Since 1966, the S&P Maney Center Bank group led market peaks in seven of ten cases and market troughs in six of ten cases. Also since 1966, S&P Major Regional Bank group led market peaks in four of ten cases and market troughs in five of ten cases, with the variability of lead/lag weeks being very low. Money Center Banks, therefore, have consistently led market peaks and have had a higher tendency to lead market turns than the brokers. But there are better groups to follow then either the banks or the brokers. The number of observations (discussed in detail later in this paper) may be shy of that generally used for statistical significance testing. Nevertheless, the price action of these two groups is often mentioned in the financial media as having led major market turning points.

The objective of this paper is not to establish a statistical correlation, but provide evidence for factual historical tendencies, as to which S&P stock groups led major stock market peaks and troughs using data that are currently available.

Summary of Results

Thus the premise of this paper is that certain S&P stock groups have tended to peak and/or trough prior to the general stock market (as gauged by the DJIA). The first objective is to provide evidence as to which groups had leading tendencies, thus showing the reality of the stock market as opposed to the myths of the stock market. The stock market’s peaks and troughs are broken down into three categories each: all, recession-related and non recession-related. The second objective is to create an index that encompasses the best leading groups. This index, called the Leading Group Index, is tested to see if an index of good leading groups provided superior results versus to an indil-idual group.

The results of the analysis showed that there are S&P groups that have consistently led market peak and troughs. Although no group led both “All Market Peaks” and “All Market Troughs,” Electric Ctilities did make the cutoff in four of the six categories. Home-building had the highest percentage lead for “.211 Market Peaks,” as it led 86% of the peaks. Beverage Soft Drink had the highest percentage lead for “All Market Troughs” at 90%. The results also disproved the zuiddyhdd beliefthat the Brokerage group has historically been one of the better leading groups. The results of the Leading Group Index were mixed, as combining the best groups that identified market peaks into an index did not provide superior results; however, combining the best groups that identified market troughs did provide excellent results. The trough composite index led 89% of all market troughs, which is better than any indi\-idual group tested.

So little Data From Which to Choose

S&P group indices are chosen for several reasons (see appendix A for groups used in this analysis). The price history of the S&P groups provides more observations (market peaks and troughs) then other data sources. S&P data are widely available, widely followed and have a good reputation for accuracy The number of groups on which S&P provides price histories allows for vey specific testing. Morgan Stanley and the CBOE have a few indices, but they are considered sector indices, not group indices. X sector index is composed of several groups. The hlorgan Stanley Consumer Index (a sector index) consists of Cosmetics, Household Products, Retail Apparel, Retail Department Stores, Retail General Merchandise, Retail Specialty Shoes, and Toys. S&P has separate group indices for each of these components.

Another option would be to use mutual fund price data. Fidelity offers 36 mutual funds that invest in different sectors of the market. Prices can be found daily in most newspapers, so data are widely available. Nevertheless, two obvious problems stand out. First, most of these funds have fewer than 10 years of price history, which only allows for the testing of a few market peaks and troughs. Second, turnover in the funds is so high that there is not a consistent base of stocks to follow through time, and the weightings of the stocks in the funds is not constant. It is true that the stocks that make up the S&P groups will change over time, but this is minor compared to the turnover in mutual funds.

S&P provides various frequencies for its group indices - daily, weekl! and monthly. There are pros and cons to each of these, but this paper tested weekly S&P group price data. Daily data will provide a more precise reading of leading tendencies. However, daily prices are only available for four market peaks and troughs, as the data only go back to 1979. Therefore the lack of obsewations made the daily data series inadequate when compared to weekly data, especially when the analysis broke the market peaks and troughs into recession and expansion - related peaks and troughs. X bull or bear market is considered recession-related if a recession occurs during the bull or bear market cycle. If the economy was in an expansion during the entire bull or bear market, the cycle was categorized as a non-recession related bull or bear market. It is very important to realize there are external differences in each bear and bull market. Market analysis should include periods of war and peace, inflation and deflation, tax increases and tax decreases, and Republican and Democrat presidents. The group that consistently led market troughs and peaks through all of these circumstances may have a better chance of doing the same in the future.

Because the majority of weekly S&P group price histories went back to 1969, weekly data provided twice as many market peaks and troughs as daily data. The weekly data series are clean and have been verified for accuracy. There are many data sources that do not provide accurate data, and consequently some analysis done in this industrv is inaccurate. It is the author’s strong opinion that verified clean data are an important criterion when choosing a data source. Some precision in the lead time of groups is lost using weekly data instead of daily data; however, a hvo- or three-day difference in lead times is insignificant when twice as many observations can be obtained. Monthly S&P group price data do provide a greater number of bull and bear markets to test, and they therefore provide more testing of recession and non-recession-related market peaks and troughs. Severtheless, precision is significantly reduced in monthly data relatire to weekly data. In addition, the number of groups that can be tested using monthly data is significantly less than if weekly data are used. There are onl! 41 monthly groups that had a price history farther back than 1945, but there are 61 weekly groups that had price history back to at least 1970.

The conclusion was to use weekly S&P group price data in testng lead times in this analysis. ireeklp data provided nearly 50% more groups to test than did monthly data. Various types of bull and bear markets can also be tested using weekly data without losing significant precision in the results.

The criterion for distinguishing bull and bear markets is based on all 15% price changes in the DJLA on a daily basis since June 1966. T\VO exceptions are made to this rule: encompassed within the lengthy 1973 to 1974 bear market were two 15% rallies (8/ 22/73-10/26/73 and 10/4/7411/5/74) that lasted two months or less. Because we are focusing on group price movement around major market peaks and troughs, these contra-trend rallies are not used in the analysis.

The dates that the criterion above produces are daily dates and the analysis is done in a weekly frequency In order to get the weekly trough dates used in the analysis, the Dow Jones Industrial Average Friday closing low that associated with the daily bear market low (using the 15% price change criterion mentioned in the preceding paragraph) is used. To illustrate, the daily market trough in October of 1987 was 10/19/87 (Monday) - the corresponding DJW weekly Friday closing low was 10/23/87. This analysis therefore used 10/23/87 as the trough date. Dates used in this analysis are shown in appendix B. Bull market peaks and bear market troughs are also broken down into recession- and expansion-related peaks and troughs (see chart below), to see if any cyclical characteristics can be identified.

The expansion and recession dates used in this analysis are those defined by the National Bureau of Economic Research. Dates of economic recessions are shown in appendix C. The dates of market peaks and troughs associated with economic expansions and recessions are shown in appendix D. (Chart courtesy of Xed Davis fisenrch.)

Discussion Of Criteria

There are several criteria used for testing the leading tendencies of the S&P groups.

First, the group price history has to encompass at least half of the bull and bear markets used for any particular category. For example, when testing all 10 market peak and trough dates, the price history has to encompass at least five peak and trough dates. Once the peak and trough dates are established, and the price history of the group is sufficient, the analysis runs each group’s price pattern from 50 weeks prior to the peak (trough) through the end of the bear (bull) market. If the bear (bull) market is greater than one year, a maximum of one year after the peak (trough) is used (these are the adjusted dates used in the shading on the chart shown above). For bear markets, the date of the S&P group price low for each time frame is compared to the stock market trough date. The number of weeks between the two dates indicates the lead/lag time.

From these results, the mean number of weeks is calculated to find the average lead/lag time. The standard deviation of the results (column four in the Results of Analysis table) is used to gauge the variability of the results (see Appendix E for the calculation of the mean and standard deviation). The key criterion calculated is the number of leading observations as a percentage of total observations. The kev cut off level for this criterion was 70%. If a group led market troughs (peaks) at least 70% of the time, it is considered a priority group. Once priority groups are chosen, they are ranked according to the standard deviation of results. The lower the standard deviation, the more reliable the leading tendency.

The S&P group indices are tested in six categories:

1. All Bull Market peaks

2. Bull Market peaks associated with recessions

3. Bull Market peaks not associated with recessions

4. All Bear Market troughs

5. Bear Market troughs associated with expansions

6. Bear -Market troughs not associated with expansions

The idea behind testing recession and non-recession market peaks and troughs is to identiQ the group action at market turning points given various underlving economic conditions. Since 1968, recession-related bear markets have produced, on average a 428-da): 28.5% decline, while non-recession related bear markets have produced, on average, a 243day 23.7% decline. Recession-related bear markets have lasted nearly twice as long as nonrecession related bear markets.

(Click To Enlarge)

In the table labeled SUP Electric Llility Croup at Bull ,Market Peaks (Bear Markets), each bull market peak date is shown in the left hand column, and it corresponds with the Lead Date column which shows the peak date for the S&P Electric Utility group. The number of weeks the group led or lagged (indicated by a negative number) bull market peaks or bear market troughs 1s shown in the last column. The lead/lag number of weeks is then used to calculate the mean lead time in weeks and the standard deviation (variability of results). The percentage of leading occurrences is calculated by dividing the number of leads (number of peak dates that the group index actually led the market peak date) by the number of total observations. (Chart above courtesy of A?d Dauis Research Inc.)

Results of Analysis

The following tables include those groups that met the cutoff criterion (percentage of leading occurrences equals at least 70%). The title of the S&P group is listed, followed by the percentage of times the group led the market peak or trough date. The mean (average) number of weeks the group led is followed by the standard deviation of results. The groups listed in each category are in order of lowest standard deviation of results to highest. Lead times that show the lowest deviation from the mean are considered to have had the most consistent lead time results. An average lead time of 17 weeks indicates that the price of the group index peaked or troughed 1’7 weeks prior to the market’s peak or trough. From these results, the following groups are the best S&P group indices for each corresponding peak and trough category over the time tested:

Summary and Conclusion of Results

This study not only provides evidence that there are groups that have consistently led market peaks and troughs, but it also identifies those groups. It also illustrates that during different cvcles (peaks, troughs, recession and non-recession market conditions) in the stock market, there have been tendencies in the price action of certain groups which may provides clues to future market action.

There are no groups that met the “best groups” cutoff in all six categories, nor are there any groups that met the cutoff for both “All Market Troughs” and “All Market Peaks.” Nevertheless, Electric Utilities met the cutoff in four categories and was tested back to 1966; therefore, this group is tested in all 10 peaks and troughs used for this study (increasing the significance of the resuits). In addition, Beverage Soft Drinks led “all Market Troughs” 90% of the time (again tested back to 1966) and Home Building led “All Market Peaks” 86% of the time. Several groups met the “best groups” cutoff in three categories: Household Products, Tobacco, Health Care Drugs and Home Building.

There are three groups that led “Market Troughs Associated with a Recession” 100% of the time - Air Freight, Hospital Management and Household Products. In addition, three groups led “Market Troughs not Associated with a Recession” 100% of the time - Broadcast Media, Beverage Soft Drink and Pollution Control. The two groups that led “Market Peaks Associated with a Recession” 100% of the time are Aerospace and Auto. And the four groups that led “Market Peaks not Associated with a Recession” are Home Building (with a very low \-ariability of results), Money Center Banks, Household Products and Toys.

Although one does not know officially if a recession has occurred until well after the fact, watching particular groups diverge from the market ma7; provide a clue as to the qpe of a bear market. Knowing if an ensuing bear market is recession-related can be valuable, as it can help determine the severity of the bear market. As previously mentioned, recession-related bear markets have lasted nearly twice as long as non-recession-related bear markets and on average were more severe.

One noticeable difference in the leading tendencies for troughs and peaks is the number of groups that led troughs versus the number that led peaks. There were 45% more groups that led “All Market Troughs” than “All Market Peaks.” There were also numerous groups that lagged “All Market Troughs” by only a week or so. The difference here may be that bottoms form much faster than tops. At bottoms, all groups begin to participate in the bull market quickly. At tops, groups start peaking in a rolling fashion, and therefore don’t peak consistently around the top.

Leading Group Index

The next step is to analyze a combination of S&P groups as a composite index. First, a geometric (unweighted) index is created using all seven of the S&P groups which made the cutoff in the category “All Market Peaks.” The peak dates in the unweighted index are then compared against the bull market peak dates. The same statistics are used as before. Unfortunately, the results did not fare well, as the Leading Group Peak Index only led 44.4% of all market peaks, had a mean lead time of 3.3 weeks, and a standard deviation of 2’i.6 weeks. However, the Leading Group Trough Index, comprised of the 11 groups that made the cutoff for leading “All Market Troughs” fared much better. It led 89% of the market troughs (tested in nine troughs) with an average lead time of 25 weeks and a standard deviation of 19.3 weeks.

Data and Observations

There are limited observations for this analysis, if one bases the analysis on a statistical significance test. Generally 30 observations are used as a criterion. However, there are only 15 post World War II bear markets (still below that considered statistically significant). As mentioned previously, accurate, reliable financial price data such as the Standard and Poor’s data generally only include price history back to the late sixties. Fewer obsemations do not invalidate the analysis, they simply reduce one’s confidence in the results. Major cycle analysis should not be dismissed on the grounds of the number of observations. There have been countless studies of the Kondratieff L+‘ave, but how many cycles can we analyze using accurate data? Maybe one. This does not mean, however, that the subject isn’t worth exploring.

Another problem with limiting this analysis based on observations is that through time, new groups emerge to the forefront of economic leadership. Computer and Information System groups should be tested now, and tested again in the future to see if time has changed its significance as a leading group. The S&P Brokerage group only has data back to 1978, and it did not make our cutoff as one of the best leading groups; however, it is still mentioned quite often in the financial media as a leading group. It may not have 30 observations, but it does have historical results in which analysts, money managers and the financial media are interested. We should notjust throw out these test results due to a lack of observations simpl!: because complex and extensive financial price data are in an infant stage. Rather, we should use these results as a stepping stone for further analysis in the future. For most analysis the number of observations should be an important factor; nevertheless, in major cycle analysis it should be understood that a significant number of observations might be impossible to obtain.

Possible Future Analysis

The analysis done here basically took a snap shot in time around major market peaks and troughs. Nevertheless, there are various other ways to test for leading tendencies. Future analysis can be done by testing the monthly S&P group indices for confirmation of the weekly results stated above (at least for those groups that haye both weekly and monthlv data). Another approach is to move the price data for the group indices forward by a certain number of weeks or months and to analyze the statistical correla- tion between a general market index such as the DJIA or the S&P 500 and the forwarded group price data.

Shorter time frames can also be analyzed. One can also test 12Oday new lows in a market index versus confirmed or non-confirmed new lows in a group index. A market index making a new low, while a group has already made a new low, would indicate a leading tendency. Other analysis can be done by producing a diffusion index: counting the number of groups that have fallen from their 26-week highs as a percentage of all groups.

As time passes, the results could change because of changes to the economy. New groups for which S&P has not yet created indices will become more relevant to the overall economy and may provide good leading tendencies in the future. An example is new groups within the dynamic world of information systems.

Lead time analysis can take a step into economic analysis. For example, one can test chain store sales (a retail economic indicator) with the price action of the S&P Retail Department Store Index. Although the Leading Group Index was mixed, another option would be to combine like groups into a sector index (e.g. combining health care-related groups into a Health Care Sector Index) and to then test that sector index for leading tendencies.

For any market analyst, money manager orjournalist, a goal is to identify the major trend of the stock market. Following the price action of important leading groups, such as Electric Utilities, Beverage Soft drinks and Home-Building (to name a few), may provide clues to the stock market’s major trend. Using price divergences between these groups and the general market ma! provide confirmation to other technical divergences that may be occurring, such as volume or the advance decline line. Watching the price patterns of good leading groups is a technique that should be added to the technician’s analytical arsenal.

Bibliography

• Levin, Richard I., Statistics For Management, Third Edition, Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 1984, Pg. 62-63, 115-21.

• Standard & Poor’s Statistical Service, Standard SC Poor’s, 25 Broadway New York, NY 10004

Biography

Robert C. Schuster is Director of Customized Research for Ned Davis Research Inc. He provides analytical research to institutional investors, developing and testing customized indicators and conducting scenario analysis. He also develops and supervises the development of large customized multi-indicator models for institutions.

Robert was graduated with honors from the University of Illinois with a BS degree in finance. Prior to joining Ned Davis Research in 1989, he traded limited partnerships at NAPEX. He is currently pursuing his CMT designation and the CFA charter.

I Get By With a little Help From My Trendline

The Ross Multiple Moving Averages hlethod (RMMAM) is based on one of the most universally-accepted doctrines in technical analysis: the principles of Dow Theory. Specifically rn!; work is based on the key component of Dow Theory, the assumption of trend.’ To begin at the beginning, for me, requires a quote from Messrs. Edwards & Magee in their great work, Technical Analysis of Stock Trends: “The fact is that the real value of a share....is determined at any time solely, definitively and inexorably bp supply and demand, which are accurately reflected in the transactions consummated in the floor of the New York Stock Exchange.“’

Edwards & Magee go on to make a point that is so obvious to technicians that many technical analysts may from time to time actually lose sight of its importance. That is, the mere knowledge that supply and demand is the sole determinant of price is of little significance, were it not accepted bp every technician as a fact of life that prices move in trends, and that trends stay in place until there is a change in supply 8: demand. Of the twelve basic tenets of Dow Theory as described by Edwards 85 Magee, nine pertain directly to the determination of trend, trendlines, and trend change. Trends are what tell the ston:

In Technical Analvsis of Stock Trends, a w&k of over 600 pages, the authors devote one 6-page chapter to what they refer to as the “Automated Trendline”: the moving average. To quote again from their book: “The moving average is a fascinating tool, and it has real value in showing the trend of an irregular series of figures (like a fluctuating market) more clearly.” The moving average is a fascinating tool for evaluating trend and anticipating changes in trend.

Is a Bear Just a Bull on its Back?

The RVMXM is based on the foregoing core principle of technical analysis: the trend. The “Automated Trendline” had to be updated and plotted by hand in the 1940s. Today moving averages can be oscillated, smoothed, and enveloped to where they must be presented separately from the price chart in order for each to be read clearly For the technician in the 199Os, the simple moving average as an indicator has become as mundane and archaic as it was fascinating for Edwards & Magee 30 years ago.

As “worked over” as the moving average has been, I was intrigued by the idea of using it to quantify the proposition that stock prices fall differently than they rise. Certain chart formations are more common at tops than at bottoms, or I-ice versa (the rounding bottom, for example). Obsen-ation has led to the generally accepted principle and proverb that it takes volume to push stocks up, but they can fall of their own weight. It is also generally-accepted that stocks fall more sharply than they rise, generally reasoned on the premise that fear is a more powerful emotion than greed. I was also interested in tying to discern, through the use of moving averages, whether or not stocks in various industry groups behaved in any sort of distinctive manner.

The purpose of this research is to:

1. Determine if the RMhLXU historically produced consistently higher profits than a single moving average, and to determine the number of trades signaled compared to a single moving average.

2. Analyze the difference in weeks between a buy moving average and a sell moving average in an optimized combination, and how that relates to the overall performance of the stock price for the period tested.

3. Investigate the data to look for relationships within stock groups.

4. Determine if there is an!; predictive value in an optimized moving average combination by forward averaging.

5. Determine whether there is, in fact, any validitv to the widelyheld practice of using an optimized moving a&age for an in- dex or average as an indicator for individual stocks. Specifically; I tested the 40-week moving average, which is commonly used as an effective stock price indicator, mainlv because it is widely accepted as an optimum indicator for the Dow Jones Industrial Average and the S&P 500.

My endeavor is intended to help equity portfolio managers who generally do not day trade or sell short and may not be particularly proficient at chart reading, but who are looking for more from technical analysis than the guidance of a somewhat arbitrary moving average.

The Universe

The RMMAM is the result of research using weekly data on NI’SE listed equity securities for the period 1973-1996. Because the research is security and industry specific, the testing was performed for two overlapping 15-year periods, to provide more meaningful results. The first period tested was from 1973 to 1987; the second period was from 1982 to 1996. One hundred stocks were tested for each period.’ The stocks selected represent almost all of the industry groups with equities trading on the WSE.

The multivariate model (RMMAM) used for back-testing consisted of two simple moving averages, one to provide a buy signal (closing price above) and one for a sell (out) signal (closing price below). Short-selling was not tested. The variable parameters were 20 to 56 weeks for both moving averages (thus enclosing Fibonacci numbers, with hopes of gaining possible Elliott 1ZBve insights). Testing for profit percentages do not take commissions into consideration; so the results should be read in light of the number of trades, which may slightly reduce or increase the significance of differences in profit percentages.

As suggested by Edwards & Magee, semi-logarithmic-scaled charts, as opposed to arithmetic-formatted charts, are yen‘ useful in security price analysis. Because simple moving averages are exclusively & specific, optimized results will not be affected by the format used for the visual evaluation of the data.

In a market of stocks, the range of profitabiliq. varies widely within an? given time period. Since one of the objectires of the research 1s to validate, or not, the practice of applying an optimized moling average for a market index to individual stocks, the use of standard deviation, or any other method, to weight or performance-adjust the data would skew the results away from an objective evaluation.

Optimizing for this multivariate model required 1369 tests for each security and for each time period. The top 70 (5%) moi-ing average combinations were recorded and the 9Yh percentile combination is shown (see Appendix 1 and Appendix 2) to help discern any inconsistency of results.

Optimization

One of the risks of using optimization to test a trading method is that optimized results will obscure underlying randomness. For example, in testing for a particular security the most profitable mo\ing average combination may differ substantially on either the buy or sell side, or both, from other combinations that were only slightly less profitable. In order for optimized results to by meaningful, they should be checked for consistency, or degree 01 stationarity.”

For each security tested there are 1,369 possible moling average combinations in a range of 20 to 56 weeks. As a test of the optimization results for consistency I calculated the differences in weeks, between the most profitable making average combina tion and other moving average combinations that were in the toI 1% of profitability for each security, for each time period. Tht variances in weeks are plotted on histograms (See Graph G. 1.A& B, and Graphs G.2A & B). The graphs contain the data from the 14 most profitable mo\-ing average combinations for each of the 100 stocks tested for each time period.

A significant number of random moring averages in the to 11% of profitable mo\ing average combinations would plot a large shaded area across the entire graph, with only weak emphasis a the zero point (the single most profitable moving average combi nation). On the other hand, if there were few random result among the most profitable combinations for a particular secu rity, the shaded area of the graph will appear as a pronouncec peak at the center (zero point). This would strengthen the valid ity of the optimization results, because there would be a demon strably close relationship between the most profitable mo\inf average combinations for a particular security, and the number of weeks in the moving average combinations that produced those profits. The graphs show a high lerel of consistency, or stationarity my test results. See graphs G.lA & B and G.2A & B.

Results - Stocks Tested 1973-1987
Profit Percentage

For the period 1973-1987 (Appendix l), the RMMLM outperformed a single optimized moving average crossover, a 40-week single moving average crossover and Buy/Hold for the average of all stocks tested. The 70L” most profitable multiple moving average combination (9Yh percentile) was the second most profitable strategy a single optimized moving average crossover provided the third highest total profit percentage, followed by Buy/ Hold. The 40-week moving average crossover came in 5’h, with only 78% of the total profit of Buy/Hold. Of the 100 stocks tested, the RMMAM was the most profitable indicator 87 times, and Buy/ Hold was the most profitable 13 times (Table T.lA).

The numbers of trades for each of the market averages are indicated in Table T.lD.

Results - Stocks Tested 1973-1987

Number of Weeks Difference Between Buy & Sell
Moving Averages in Optimized Combination

With five exceptions out of 100 stocks tested, if one of the moving averages (either buv or sell) in anv optimized RMMAM combination was 25 weeks or longer, the difference between the buy and sell moving average for a particular stock was 5 weeks or longer. Three of the fiye exceptions had a difference between the buy & sell moving average of four weeks. X scattergram, (Chart G.lC) shows this “Interval Gap” and the five exceptions. Even though the vast majoritv of individual stocks had a difference between the buy & sell ‘moving average of five weeks or more, when the results were averaged for the 100 stocks tested, there was only a two week difference between the average length of optimized buy & sell signal moving averages in combination (35 weeks buy/33 weeks sell).

Additionally, a full 25 of the optimized moving average combinations in the RMMAM contained a 20-week moving average and 6 of the combinations contained a ?&week moving average (one of the optimized moving average combinations was 20 weeks buy/ 56 weeks sell). For the 100 individual stocks tested. 58 had a longer buy signal MA and 42 had a shorter buy signal MA.

As Table T.lE shows, stocks with a shorter bup signal moving average in the optimized combination averaged a higher Buy / Hold return than stocks with a longer buy signal moving average. This would suggest using a combination of moving averages with a shorter buy signal moving average for stocks that are in an uptrend and outperforming the market.

Results Market Indexes 1973-1987

Number of Weeks Difference Between Buy & Sell
Moving Averages in Optimized Combination

For both the NYSE Composite and S&P 500, the buy signal MA was 15 weeks longer than the sell signal MA (42 weeks buy/ 27 weeks sell and 41 weeks by/26 weeks sell respectively). This is in contrast to the average for the 100 WSE stocks tested where there was only a 2 week difference between the average length of optimized buv and sell signal MA in combination.

Results - Industry Groupings 1973-l 987

The RMMAM provided the best profit percentage in all of the macroeconomic groups” represented. In the consumer cyclical group, RMMAM outperformed a single optimized mo\-ing average crossover b,: more than 50%, and outperformed Buy/Hold by more than 230%. For cyclical stocks, where trend-following techniques such as moving average crossovers would be anticipated to provide the most effective intermediate and long term signals, the RMMAM was substantiallv more profitable than a single optimized mo\ing average crossover, a 40-week moving average crossover, and Buy/Hold. In addition to providing a greater profit percentage for consumer cyclicals than a single moving average crossover, the RMMAM averaged almost 30% fewer trades (22 trades vs. 31).

Similar results occurred in other macro-groups that are cyclical in nature. Table T.lF shows how the ItMMUI performed for the macroeconomic groups, shown as the average return for stocks tested within those groups. The table demonstrates, by comparing the performance of the RMMAM to Buy/Hold, how moving average crossovers are more effective with cyclical stocks than with more defensive issues.

Results - Stocks Tested - 1982-l 996
Profit Percentage

For the period 1982-1996 (Appendix 2), Buy/Hold outperformed RMMAM, a single optimized moving average crossover, and the 40-week moving average crossover for the total of all stocks tested. The RMM;LM was the second most profitable strategy. The 70’h most profitable multiple moving average combination (95%) provided the third highest total profit percentage, followed bp a single optimized moving average crossover. The 40-week moving average crossover came in a distant ?irh, with onlv 25% of the total profit of Buy/Hold (Table T.2A). For the period 1982-1996, of the 100 stocks tested, the RMMAM was the most profitable indicator 48 times, and Buy/Hold was the most profitable 52 times.

Number Of Trades

As in the earlier period tested, compared to the other moving average methods tested, the RMMAM provided, on average, higher profits with less trading. The RMMAM averaged 35% fewer trades than a single optimized moving average crossover, and 31% fewer trades than a 40-week moving average crossover (Table T.2B).

Results - Market Indexes - 1982-1996
Profit Percentage & Number of Trades

For both the NYSE Composite and the S&P 500, during the period 1982 to 1996, the RMMAM outperformed all other methods tested. For the hYSE Composite, the 701h most profitable multiple moving average combination (95’” percentile) was the second most profitable, followed by Buy/Hold, a single optimized moving average crossover, and a 40-week moving average cross-over. For the S&P 500, Buy/Hold was the second most profitable strategy followed by the 70* most profitable multiple moving average combination, a single optimized moving average crossover, and a 40-week moving average crossover. (Table T.2C)

The number of trades for each of the market averages are indicated in Table T.2D

Results - Stocks Tested - 1982-l 996

Number of Weeks Difference Between Buy & Sell
Moving Averages in Optimized Combination

With only 3 exceptions out of 100 stocks tested, if one of the moving averages (either buy or sell) in any optimized RMMAM combination was 25 weeks or longer, the difference behveen the optimized buy and sell moving average for a particular stock was 5 weeks or longer. Two of the three exceptions had a difference behveen the bu? & sell MA of four weeks. X scattergram, (Chart G.2C) shows this “Interval Gap” and the three exceptions.

For individual stocks tested from 1982-1996, the average optimized moving average combination for the 100 stocks tested was 29-week buy/40-week sell, compared to only a P-week difference (35/ewebweek buy/33week sell), for the average of stocks tested for 1973-1987.

One-third (34) of the optimized moving average combinations in the RMMAM contained either a 20-week or a X-week MA, or both. Twenty six of the 100 stocks tested contained a 20-week moving average and 12 of the combinations contained a X-week moving average (three of the optimized moving average combi- nations were 20-week buy/56+eek sell, and one was 56-week buy/ 20week sell).

For the period 1982-1996, 26 of the stocks tested had a longer buy signal moving average and 74 had a shorter buy signal moving average in the most profitable combination. This differs from the 1973-1987 period, where the majority of stocks had a longer buy signal moving average. (Of the 26 stocks with a longer buy signal moving average in the 1982-1996 period, 21 also had a longer buy signal moving average in the 19731987 period).

As in the earlier period tested, stocks with a shorter buy signal moving average in the optimized combination averaged a higher Buy/Hold return than stocks with a longer buy signal moving average (Table T.2E). Also, almost 75% of the optimized moving average combinations had a shorter by signal moving average component, a much higher number than in the earlier period tested. The average Buy/Hold return for the stocks tested during 1982-1996 was 255% of the return for the same stocks during 19731987. These results add greater weight to the idea of using a combination of moving averages with a shorter buy signal moving average for stocks that are in an uptrend and outperforming the market.

Results - Market Indexes - 1982-l 996

Number of Weeks Difference Between Buy & Sell
Moving Averages in Optimized Combination

In analyzing the difference between the buy and sell MA in optimized combinations, I discovered that there was a marked contrast behveen the two time periods tested with regard to the market indexes. For 1982-1996, the R’SE Composite was the most profitable with a 20-week buy/45-week sell moving average combination. This is almost the reverse of the 19731987 period, where the AYSE was most profitable with 42-week bm/27-lveek sell moving average combination. Likewise, for 198i-1996, the S&P 500 had a 20-week buy/54-week sell combination using the RMRWM, compared to a 41-week buy/26-week sell for the 19i3-1987 period.

Results - Macro-Economic Groups
1982-1996

For the 1982-1996 period, RMMAM outperformed a single optimized moving average crossover and Buy/Hold for five of the nine macroeconomic groups. During this period, in contrast to the 1973-198i period, Buy/Hold had a higher profit percentage for the average of the 100 stocks tested than any of the moving average methods tested. Table T.2F shows how the RMMv\I performed for the macroeconomic groups, shown as the average return for stocks tested within those groups. With the exception of Utilities and Financials, the table does show how moving average crossovers tend to be more effective when applied to cyclical stocks.

Looking Over The Edge

As indicated in Chart G.lC, for the 1973-1987 test period, 30% (all triangles and squares) of the optimized multiple moving average combinations “pushed the envelope,” i.e., had a compo nent moving average at the parameter of 20 or 56 weeks. The majority of these (25 of the 30 or 83% - all triangles) contained ; component 20-week moving average in the combination. For the 1973-1987 period, in analyzing the 30 stocks where 20 or 56 week: was a component of the optimized moving average combination I expected to find a pattern of relative out-performance or un der-performance compared to the averages of BUT/Hold profit ability for the stocks tested for that period. I would expect to set relative under-performance for those stocks where the parameter were apparently too narrow. on the buy or sell side, to encompas a moving average combination that was not compromised by the limits I had set. These stocks, then, would not show the profit ability of stocks whose moving average combination fell withi] the 20- to 56-week parameters, and therefore were not encum bered by optimization limits.

Alternatively for a stock where a 20-week moving average with the buy signal, I expected Buy/Hold to show the greatest prof percentage. I would have expected this because 20 weeks is tE shortest buy moving average the optimization allowed, and tE closest to a one week buy moiing average, which is essentially Buy/Hold result. Thus, I expected a 20-week buy moving aYe age to correlate to the most profitable stocks of those tested fc that time period. However, for only 5 of the 30 moving arerag combinations (I 7%) where 20 or 56 weeks was a component, rJi uy/Hold hare the highest profit percentage VS. 13 out of 100 for le universe tested.

Likewise, during the period 1982-1996, 34 of the 100 stocks sted had a component moving average at the parameter of 20 or 56 weeks (see Chart G.2C, all triangles and squares’). Buy/ old had the highest profit percentage for 22 or 65% of these 34 ocks. For the total universe of 100 stocks tested, Buy/Hold outperformed all other moling average methods for 52 (52%) of the stocks.

I found little correlation between stocks whose moving averge components were at the parameters and stocks with Buy/Hold s the most profitable strategy However, using less strict criteria D find relationships, I did obtain results that quantified my intution regarding the length of modng averages and stock perfornance. As shown on Tables T.lG and T.2G, stocks with a 20-week buy moving average or a X-week sell moving average component n the optimized combination were more profitable (as measured by Buy/Hold) on average than stocks with a S&week buy mo\ing yerage or a 20-week sell moring average component. Although ome of the sample sizes were very small, the data taken as a whole upport the proposition of using a short bupsignal moving average and a long sell-signal mo\ing average m combination with to&s that are in uptrends. Conversely a long buy-signal moving n-erage and a short sell-signal moving average in combination could be more effective with stocks that are in a downtrend or tre underperforming the market.

Broader Horizons
1973-1987

I was curious to see whether widening the parameters would significantly reduce the number of optimized moving average combinations with component moring averages touching the boundaries. To investigate, I tested the 100 subject securities for the 1973 to 1987 period, using 10 weeks to 75 weeks as the new parameters (Appendix 3). The wider parameters required 4,356 tests for each securit?. As a check on the integrity of the data, I also tested for a single optimized moving average using the new parameters, and confirmed the results where the single moving average fell within the narrower boundaries in both cases.

The results of the tests using wider parameters were surprising. Of the 100 stocks tested for 1973-1987,25 had an optimized moving average combination with either 10 weeks or 75 weeks as a component moving average. This is very close numerically to the results using a 20-56 week range where 30 stocks had 20 weeks or 56 weeks in the most profitable moving average combination, with one result being 20.week buy/56-week sell. However, there was a low correlation between the specific stocks that tested with an optimized moving average at the parameter. Of the 25 stocks that resulted with a lo-week or 7Sweek component moving average in an optimized combination, only 6 (24%) corresponded to equities with a 20- or 56-week component moving average when the narrower parameters were used.

These interesting results clearly indicate that a 20-56 week range was not more restrictive in the optimization process than a much broader lo-75 week range.

Tables T.lG, T.2G and Table T.3A lend support to the idea of varying the length of moving averages in combination depending on the trend of the stock.

Fibonacci

Looking for Fibonacci numbers when testing under wider moving average parameters for 1973-1987, I found that only 34 (17%) of the 200 optimized moving averages in combination were Fibonacci or adjacent numbers (12,13,14 ; 20,21,22 ; 33,34,35 ; 54,55,56). Statistically 18% of the moving average possibilities are Fibonacci or adjacent numbers (12 out of 66 possible results in the lo-75 week range is equal to 18%). I was somewhat disappointed, but indeed it appears that Fibonacci numbers are not golden when testing for optimized moving averages in a multivariate model.

Forward Testing

Because there was a six-year overlap of the two ISyear periods, a forward testing of the optimized moving average combinations for each stock might reveal some predictive value.

During the 1982-1996 period, the buy/hold profit percentage was higher for the average of the 100 stocks tested than any of the moving average methods. However, applying the 1973-1987 optimized RMM!M results to their respective stocks in the 1982-1996 period provided a substantially higher average profit percentage than the 40-week moving average, with fewer trades. In addition, the 1973-1987 optimum RMMAhl, when applied to the 1982-1996 period, provided 85% of the return of a single moving average optimized for the 1982-1996 data, with 26% fewer trades (Appendix 4). Because the single optimized moving average was calculated after the fact, the results of forward testing did reveal some predictive value for RMMAM.

Conclusions
The Power of Two

The RMhMM produced higher profits than a single optimized moving average for every stock tested for both time periods. In addition, for 93 out of the 100 stocks tested from 1973-1987, and for 98 out of the 100 stocks tested from 1982-1996, the RMMAM had the same or fewer number of trades than a single optimized moving average crossover.

For the MSE composite and the S&P 500, the RMMAM also produced a higher profit percentage for both periods. M’ith regard to the indices, while the number of trades for 1973-1987 did not differ remarkably between the RMMxRiI and a single moving average crossover, trading the indices using the RMMAM from 1982 to 1996 outperformed a single optimized moving average crossover with less than half of the number of trades. In that same 1982-1996 period, the stock market’s total return was five times that of the 1973-1982 period. The significantly fewer trades in the later period using the RMMAM on the S&P 500 and the M’SE composite suggests that a multivariate model using separate moving averages for buy and sell signals would produce higher profits with more efficient trading than a single moving average crossover, especially during markets that are trending strongly.

The results of the forward testing of optimized RMMAM results to a later period do indicate a predictive value to this method.

The Interval Gap

The discovery of an “Interval Gap” of at least 5 weeks in over 95% of the tests of individual stocks using the RMMAM indicates that, indeed, an individual stock trends up differently than it trends down. The time differences in moving averages in a multivariate model are apparent when testing individual stocks; however, the results are obscured when that model is applied to market indexes (NYSE Composite and S&P 500).

It is not apparent why the gap was at least 5 weeks for nearly all of the stocks tested, and during two quite different market periods. I think that this phenomenon is in actuality a quantification of a recurring aspect of investor behavior: the difference in investors’ collective response to an uptrending vs. a downtrending stock. Experienced brokers and traders know that greed and fear are quite different emotions but not exact opposites. There is a difference in the process by which stocks go up based on greed (demand) and the way they go down based on fear (supply). I think the phenomenon of the interval gap demonstrates this. Although establishing empirical proof is beyond the scope of this research, I think that the marked consistencv of this gap for a large sample and over two lengthy market periods is further evidence that investors, collectively, are emotional rather than rational. Therefore, technical analysis is the most effective method for anticipating change in the markets, and in individual stocks.

Difference in Lengths of Moving Average Components of RMMAM and Stocks Relative Performance

For both time periods, stocks with better-than-average-performance tended to have a shorter buy signal moving average in the most profitable RMMAM. It is apparent from the results of several analyses of the data that for an uptrending stock, applying the RMMX/eweb with a shorter buy moving average component will, more often than not, provide more profitable trading signals than if a longer buv moving average component is used.

For overall market direction, it is somewhat intuitive to postulate that a shorter buy signal moving average would perform best in a bull market, and that a shorter sell signal moving average should be used in a bear market (See Table T.4). The length of optimized moving averages in the RMhLW for the NYSE and the S&P 500 compared to the Buy/Hold profit percentage for the respective time periods lend support to this assumption.

Macro Economic Groups

The data indicate, with regard to stock groups, that moving average crossovers tend to be more effective when applied to stocks with a cyclical nature than to noncyclicals (See Tables T.lF and T.2F). However, the small sample sizes for industry groups preclude anything more than a general observation.

The Folly of 40 Weeks

A final conclusion that I have drawn from this research is that there is very little correlation between an optimized moving average crossover for a market index and for individual stocks for the same period. For example, for the 1973-1987 period, a 40 week moving average crossover provided about hvice the profit percentage of Buy/Hold for the NYSE Composite and S&P 500, and was close, in both number of weeks and profit percentage, to a single optimized moving average crossover for these indexes. However, for the 100 WSE stocks tested for the same period, a 40-week moving average crossover averaged only 78% of the return of Buy/ Hold. hloreover, of the 100 stocks tested for the 1973-1987 time period, the 40-week moving average crossover outperformed Buy/ Hold for only 34 of the stocks. As an even stronger indication of the insignificance of a 40-week moving average crossover for the 100 NYSE stocks tested, a single optimized moving average crossover was within 10% of 40 weeks (36-44 weeks) less than 20% of the time (18 of the 100 stocks tested), even though that range (8 weeks) is 22% of the range for which stocks were tested (37 weeks).

For the period 1982-1996, in contrast, a 40-week moving average crossover provided less than 60% of the return of Buy/Hold for the market indexes, and was about 20% longer than the optimized single moving average crossover for the period. Also during that period, only the R/ewebJMXV/eweb outperformed Buy/Hold for both the WSE Composite and the S&P 500, while for 1973-1987, all three moving average methods tested outperformed Buy/Hold by from 90% to 280%. Again, with the stock market from 1982- 1996 having 500% of the return of 19i3-198i, it is apparent that it’s much harder to beat Buy/Hold in a strong secular bull market. For the 100 NOSE stocks tested for the period 1982-1996, a 40-week moving average crossover averaged only 25% of the return of Buy/Hold, and outperformed Buy/Hold for only 10 of the 100 stocks tested. For 1982-1996, a single optimized moving average crossover was within 10% of 40 weeks (36-44 weeks) less than 10% of the time (9 out of 100 stocks tested).

Based on the aforementioned data, it seems that, at least for any one of the 100 NYSE stocks tested, a 200-day - or 40-week - simple moving average on a stock chart would have had no more predictive significance as a signal to change position than an) other simple moving average from 20 to 56 weeks.

John Magee wrote a work published in 19% titled, The General Semantics of Wall Street. It is a book about behavioral finance, written 40 years ago. In it, he describes the constant human tendency to use mental “maps” to simpli? real world situations. People abstract, or generalize, a conclusion based on various information and then apply, that conclusion to other venues without investigating to determine if the new situation supports this “cut and paste” conclusion. Much work has been done in technical analysis using stock market indices as surrogates for stocks. 1 have concluded that the use of a stock index (which is an abstraction) in moving average studies does not necessarily provide results that can be applied to individual stocks with the same degree of reliability

Footnotes

1. Edwards, Robert D & dlagee, John, Technical Analysis of Stock ws, Boston, Mass, John Magee Inc., 1992, pp. 15-28

2. Edwards, Robert D & dlagee, John, Technical Analysis of Stock ms, Boston, Mass, John Magee Inc., 1992, pp. 6.

3. Edwards, Robert D & Magee, John, Technical Analysis of Stock MS, Boston, Mass, John Magee Inc., 1992, pp. 48?.

4. . Current parameters required 1369 tests for each security for each time pemod using a Pentium 133 computec based on the downloading of 24 Jears of week4 high-low-close and uolume history Also, the stocks tested needed to be in existence &ring the entire 24 rear time period (19i3-1996), thereby restricting the number of securities auailable for testing. Faster computers and greater memory capacity will make expanded research in multivariate analysis much more feasible.

5. Hight, Gregory 1V., “Checking for Stationa+, ‘I Technical Analysis of Stocks @ Commodities, Volume 15, Sumber 6,June 199% pp. ii- 83.

6. The classification of conglomerates as a macroeconomicgroup, as well as the word itself, has become somewhat archaic afier three decades of LBOs and corporate diversification. For purposes of consistent!, I chose to deft to Dow Jones’judgement and not fit GE and TXT into another economic group.

7. A number of stocks tested for this period had iden tical optimized moving average components; thus some of the data points represent results for more than one security

Bibliography

• Achelis, Steven B., Technical Analvsis From A to Z, Chicago, Irwin Professional Publishing, 1995

• Arms, Jr., Richard W., Trading M’ithout Fear - Eliminating Emotional Decisions with Arms Tradinp Strategies, NewYork, John Wiley & Sons, Inc., 1996

• Cohen, Jerome B., Zinbarg, Edward D., and Zeikel, Arthur, Investment Analvsis and Portfolio Management, Boston, Mass., Richard D. Irwin, Inc., 1987

• Colby, Robert M’. and Meyers, Thomas A., The EncrcloDedia of Technical Market Indicators, New York, New York, Irwin Professional Publishing, 1988

• Edwards, Robert D. and Magee, John, Technical Analysis of Stock Trends, Boston, Mass., John Magee Inc., 1992

• Frost, A.J. and Prechter, Jr., Robert R., Elliott U’ave PrincitAe - Kev to Stock Market Profits, NewYork, Haddon Craftsmen Inc., 1990

• Hight, Gregory S., “Checking for Stationarity,” Technical Analysis of Stocks and Commodities, Volume 13, Number 6, June 1997

• Magee, John, The General Semantics of Wall Street, Springfield, Mass., John Magee, 1958

• Murphy John J., Technical Analvsis of the Futures Market - A Comnrehensire Guide to Trading Methods and Applications, New York, New York, W Institute of Finance Apprentice-Hall Co., 1986

• Pring, Martin J., Technical Analvsis Explained, NewYork, McGrawHill Inc., 1991

Biography

Don Ross is a First \Tce President - Investments with Prudential Securities, Inc. and has been an investment broker since 1979. His current focus in technical analysis is on equity portfolios utilizing volume/price indicators and chart analysis. Don has a bachelor’s degree in accounting from Susquehanna University and has a master’s degree from the College for Financial Planning. He has been a CFP since 1984 and is an affiliate of the Market Technicians Association. Don also is an instructor for adult evening school programs at local high schools.

The put/call ratio, developed by Dr. Martin Zweig, and introduced in a 1971 Barron’s article,’ has enjoyed immense popularity becoming one of the standard sentiment gauges used by traders and market analysts. It is even calculated and distributed intraday on quote machines, much like the Arms Index.’ The put/ call ratio is calculated simpl! by dividing the volume of puts, which represents the number of bearish bets, by the volume of calls, which represents the number of bullish bets. The resulting quotient is usually smoothed with a moving ayerage to eliminate some of the short-term noise and expose the underlying trends. Readings above 1 indicate more puts were traded relative to calls, while readings below 1 indicate more calls were traded relative to puts. In the intervening years since the put/call ratio’s introduction, substantial and lasting structural changes have taken place in the U.S. financial markets. The options markets in particular have been the birthplace of many of these innovations. Listed call options first began trading at the Chicago Board Options Exchange (CBOE) in April 1973, followed by listed put options four years later in June 1977. The next major milestone occurred in March 1983, with the advent of index options on the Standard and Poor’s 100 Index, more commonly known as the OEX.” Since then, we have seen the introduction of options on interest rates, Long-Term Equity Appreciation Securities or LEAPS,4 and sector options on any group imaginable.

Nor has the put/call ratio been immune from these changes. In fact, many would argue that these changes hare rendered put/ call ratios ineffective. This certainly wouldn’t be the first time structural changes have overtaken an indicator. In the mid-1980s, the odd lot short sales indicator was “ruined” by program traders shorting 99.share lots to skirt the uptick rule.> Options themselves have impacted other technical indicators such as the public and specialist short sale ratios, as investors began to favor put options in lieu of short sales.

This paper will examine the effectiveness of put/call ratios in light of the modern marketplace. N’e will attempt to determine if put/call ratios are still reliable sentiment indicators, and more importantly, which ratio is the best measure of investor sentiment - the OEX put/call ratio or the CBOE equity-only put/call ratio.

Like most sentiment indicators, the put/call ratlo is interpreted in a contrary fashion. Extreme readings suggest that fear and greed have overwhelmed rationality. It is precisely these times, when the lure of the crowd is irresistible and investors are clamoring to jump on board or cover their widening losses, that it usually: pays to take a stance opposite the majority. Although often difficult to implement, a contrary stratepworks for the simple fact that bullish investors have already bought, leaving fewer marginal buvers to push prices up; conversely bearish investors have already sold, leaving fewer marginal sellers to push prices down.

There are several inherent advantages to put/call ratios. Unlike many indicators in the technician’s repertoire, the put/call ratio has the distinct advantage of not including price directly in its calculations. Chande and Kroll have shown that there is roughly a 70% to 90% correlation between popular indicators such as momentum, RSI, and Stochastics.” A cursory observation will likewise reveal that there is little functional difference between moving averages of similar lengths, such as the commonly used 200- day simple moving average versus a 15Oday moving average. Each of these indicators presents essentially the same information with slight but ultimately inconsequential variances. The put/call ratio, however, is exogenous, and therefore avoids much of the colineararity that plagues other technical indicators.

Many other popular sentiment indicators today are modeled after the venerable Investors Intelligence Survey of Investment Advisors, which polls newsletter writers on their market views, categorizing them as either bullish, bearish, or anticipating a correction.’ However, this stratification doesn’t adequately reflect the realities most investors face. Investors are seldom simply bullish or bearish, but instead fit somewhere on a continuum between the hvo extremes. Moreo\-er, there can often be significant divergences between opinions and actions. Portfolio managers have been known on occasion to profess to being bearish, while at the same time, whether by fiat or competitive pressures, remaining nearly fully invested. Can someone really be considered bearish if they still hare significant long positions? The put/call ratio recognizes that action is the best measure of bullishness or bearishness. After all, opinion alone has never moved the market, but buying and selling based on an opinion certainly has.

The put/call ratio also enjoys an immediacy lacking in other sentiment indicators. The Investors Intellipence Survey suffers from unavoidable delays due to the time involved in receiving the letters, categorizing the advisors, and releasing the results. Other sentiment indicators like short sales or mutual fund cash levels are released with a delay ranging from weeks to months. On the other hand, the put/call ratio is market-generated and can be calculated over any time frame desired, from intra-day to monthly In a rapidly changing market environment, this timeliness can be critical.

However, the put/call ratio is not without its weaknesses. As an indicator designed to capture extremes in fear and greed, the put/call ratio should work best on data that measure pure speculative activity. 14’hile the options market can undoubtedly be used as a speculative vehicle, its primary purpose is actually to reduce risk. Buying options to hedge a portfolio of existing securities is a very different strategy than buying out-of-the-money options betting on market direction. yet both are weighted equally in the put/call ratio. It is impossible to determine what portion of open interest is hedged or outright speculation. Is someone bearish because they are buving puts, or are they bullish because they actually own the underliing stocks, or are they neutral? Unfortunately there is reallv no wav to satisfactorily resolve these ambiguities.

Unlike many technical indicators which have a fixed range by design, the p&/call ratio can range from near zero to infinity Obviously, there is a big difference behveen zero to one - the optimistic zone - and one to infinity - the pessimistic zone. Pessimistic readings can therefore theoretically extend much further relative to the neutral zone than similar optimistic readings. As the levels of sentiment extremes change from turning point to turning point, the lack of a fixed range, and consequently definitive buy and sell zones, poses some additional challenges. U’hile in actual practice this asymmetrical design will create few prob lems, it is important to be aware of the put/call ratio’s various nuances nonetheless.

The put/call ratio is probably most commonly calculated on the OEX. This capitalization-weighted average of 100 large blue-chip stocks tracks the S&P 500 very closelv. The OEX offers a simple, liquid vehicle for making a direct bet on the U.S. equic market. Its popularitv is confirmed by the nearly 6 million contracts that traded injanuary 1996 alone.’ This depth not only appeals to the individual speculator, whose motives are generalh clear and conducive to contrarv analysis, but also to institutions and arbitrageurs, whose objectives, as we will see, are often less clear. The other common put/call ratio is calculated on the combined volume of all equity puts and equity calls traded at the CBOE. Just as the S&P 500 is realh the sum of many smaller parts, any large enough group of indhidual equities taken together becomes the “market.” As a corollary the volume of puts and calls on individual equities should cancel out individual idioyncrasies and become a proxy for the market. So, although put and call options on individual equities are not necessarily a direct bet on the direction of the market, the aggregate volume should yield useful clues about overall market sentiment.

In order to compare the performance of the OEX and equity only put/call ratios, we need an objective methodology. It is easy to spot significant peaks and troughs in the put/call ratio using hindsight; however, the future isn’t nearly as obliging. One of the biggest difficulties in interpreting the put/call ratio is that the levels of sentiment extremes change from turning point to turning point. Moreover, the normal levels for the equity-only put/call ratio differ dramatically from the OEX put/call ratio. From 1986 through 1995, the average reading for the equity-only put/call ratio was 0.38 versus 1.05 for the OEX. Rather than us ing fixed levels and potentially missing some important signals, we will use Bollinger Bands that compensate for the differing norms and adjust to the underlying ratio, expanding and contracting with the volatility.” The entry signal for the buy side is triggered when the put/call ratio crosses above the upper Bollinger Band and then turns down. Conversely, sell signals are triggered when the put/call ratio crosses below the lower band and then turns up. When performing contrary analysis you want to be opposite the crowd; however, the crowd can often be right for an extended period of time. Much like a rubber band, when sentiment is stretched to an extreme, it tends to reverse quickly The hook in the put/call ratio outside of a band ensures that sentiment is not onlr at an extreme, but is also reversing. ;Ul positions were initiated at the next dav’s opening price, which better reflects the realities of implementing strategies based on daily data than the common practice of buying or selling on the close. Figure 1 shows the familiar Bollinger Bands superimposed on the put/call ratio, along with buy and sell setups marked with up and down arrows respectively.

Now that we have an objective entry signal, we need an objective exit. This is quite easily solved - after a position is established, it will automaticallv be closed a fixed number of days in the’future. All intervening signals are ignored until the holding period has elapsed. Given that it is the most widely referenced equity market index, we will use the cash Standarb & Poor’s 500 Composite Index as our benchmark. 1z’e will test the entire 10 year period from 1986 through 1995. In addition, we will isolate the put/call ratio for individual years to see how the performance has changed over time.

Our results lvill be highly dependent on the parameters we choose for the put/call ratio smoothing, the Bollinger Bands, and the holding period. Therefore, to achieve the most unbiased estimate of performance possible, we will test our methodology over a wide range of parameters, using the average percentage change in the S&P 500 over all parameter variations as our comparison benchmark. Our aim is not to isolate the definitil-e timing model with these optimizations, but rather to compare the general effectiveness of the two put/call ratios.

The put/call ratio itself was smoothed with a simple moving average ranging from 5 to 20 days in 5-day increments. The length of the Bollinger Band varied between 30 and 100 days in steps of 10. In addition, the number of standard deviations of the Bollinger Band was tested from 1 to 2.5 in steps of 0.25. X holding period corresponding to 1, 2, 3, and 3 weeks was also tested in one week increments. This relativelv short-term time horizon was chosen to minimize as much as pdssible the overriding influence of the secular bull trend in place during the entire study period. Finally, each individual vear was tested, as well as the total data stream. Although these optimizations required considerable computer time, we can have more confidence in the results knowing they are unlikeI!- to be skewed by an arbitrary choice of parameters.

Table 1 shows our findings. The results are broken down first b! buy or sell signal, second b? the equity-only or OEX put/call ratio, and finally bp year. These numbers represent the alerage percentage gain or loss of the S&P 500 for all possible parameter combinations. The “All Years” row shows the average performance of the equity-only and OEX put/call ratios o\-er the entire 10 year time span.

Over the entire test period, the CBOE equity-only put/call ratic outperformed the OEX on the long side with an average gain a 0.95% versus 0.85% for the OEX. Furthermore, the equityonly put/call ratio beat the OEX in 6 out of 10 years. Its worst perfor mance was an average -2.17% loss in 1990, versus a correspond ing -3.68% loss for the OEX. However, in 1991, the OEX turnec in its best average one year performance at 3.37% versus a 2.199 gain for the equity-only put/call ratio. But on balance, the equity-only put/call ratio has the edge.

The CBOE equity-only also outperforms the OEX on the short side with a loss of -0.33%’ versus an even worse loss of -0.77%. Or a year-by-year basis, the equity-only outperformed the OEX 609 of the time. The worst loss was -2.49% versus -2.48% for the OEX The best gain was 2.51% versus a paltry 0.06% gain for the OEX Again it would appear that the equity-only put/call ratio is the better indicator.

The results on the short side don’t appear very impressive until you take into account the testing period, which encompasse one of the great bull markets of this century Any strategy employing short sales is swimming against a rapidly moving current. II fact, the holding period for our optimizations averages two and one half weeks, or roughly 13 days. The average 13-day period over our 10 year test period produces a 0.61% gain for the S&P 500.“’ Perhaps a more realistic way to view these results is relative the 13-da! S&P benchmark. From this perspective, the CBOE equity-only put/call ratio outperforms the benchmark by 34 basis points on the long side and 28 basis points on the short side.” In other words, a buy signal from the equiy-only put/call ratio would give a performance boost of 34 basis pomts, on average, above the typical 13day S&P gain of 0.61%. On the short side, you would get a performance boost of 28 basis points after a sell signal, losing -0.33% over 13 days, on average, versus a -0.61% loss as the norm (Remember the secular bull trend. ;Umost any strategy that buys and holds for 13 days will be profitable.) The OEX put/call ratio on the other hand, outperforms the benchmark by only 24 basis points on the long side, and actually underperforms by 16 basis points on the short side. The mixed results of the OEX put/ call ratio relative to the S&P benchmark further suggest that the equity-only put/call is the more reliable and accurate indicator.

One of the common criticisms one hears leveled against put/ call ratios is that “they don’t work anymore.” M’hile our findings demonstrate that put/call ratios do indeed work, there ma!; some evidence that performance has deteriorated over time. Figures 2a through 2d show the data from Table 1 presented in graphical format. The bars depict the average yearly performance along with a simple regression line fitted to show the overall trend. In as much as 10 observations are far too few to establish statistical significance, all four regression lines have negative slopes, suggesting that there may be some truth to the accusations that the put/call ratio’s effectiveness has declined. Next we will explore what could account for the outperformance of the equity-only put/call ratio and the general trend of declining effectiveness present in all four combinations.

Sentiment indicators thrive on data which capture the emotions and irrationality of the crowd. Although there are certainly exceptions, in general, small investors lack the sophistication of their professional counterparts. Their emotional swings between fear and greed are more pronounced and conducive to contrary analysis. And there are several reasons why the small investor - the modern day “odd-lotter” - gravitates toward individual equity options over index options.

Although people may have an opinion on the market, the): usually want to participate by buying indwidual securities. It is the “story,” and the proverbial cocktail conversation, that captures the imagination. Equi$; options allow the small investor to partlclpate in the movements of a stock of which he may not be able to afford round lots. In addiiion, the premiums are generally far less for equity options than index options. Much like a lottery ticket, the lure of cheap out-of-the-money options is too strong for many investors to resist. Equiq options provide a convenient, leveraged vehicle for exposure to the equitv market with limited risk. These characteiistics appeal to the small speculator in particular. On the other hand, the institutional investor tends to employ more sophisticated strategies, buying puts for portfolio protection, writing calls for income, and the like, rather than option purchases for bullish or bearish market bets. Compared to individual investors, it can be argued that institutions don’t do that much outright speculation. Mhile there is a limited retail presence in the index options markets, it is by and large the domain of institutional investors - the proverbial “smart money.” Their activities will tend to be more subtle, and their objec&es less clear than individual investors. Consequently contrary analysis of the index options market may be less rewarding.

In addition to portfolio hedging, index arbitrage is another common strateg? that employs index options. Previously we mentioned the dilemma posed by portfolio hedging: is a put buyer bearish, bullish, or neutral because of underlying stock? Arbitrage activity introduces a whole other set of distortions. Arbitrageurs look for mis-pricings among options and their underlying cash and futures markets, whereby they can buy one market and simultaneously sell the other, locking in a profit. These activities are generally independent of overall market direction, but will have an impact on the put/call ratio nonetheless. For example, someone using OEX options to create a synthetic long position for arbitrage purposes would buy calls at a specific strike price and simultaneously sell an equal number of puts. Assume this hypothetical transaction involves 100 calls and 100 puts. If the put/call ratio had previously been 1 .j (1300 puts/ 1000 calls), this transaction would cause the ratio to drop to 1.43 (1600 puts/ 1100 calls). The put/call ratio is affected even though this hypethetical arbitrage strategy has no implications for market direction. From this exaggerated example, vou can easily see how arbitrage actiriw can obscure valuable sentiment information cow tained in the index options volume.

Equity options are much less affected by these extraneous factors, which helps explain the superior performance of the equityonly put/call ratio \-is-a-vis the OEX put/call ratio. Furthermore, advances in technology hare played a large role. Sophisticated options pricing models and computer assisted execution are employed to arbitrage the slightest inefficiencies in the index options markets. U%ile equity options are also subject to arbitrage, the methods used to pick stocks haven’t changed dramaticall) since Edwards and Magee (or for that matter, Graham and Dodd).

Figure 3 shows a major divergence behveen the hvo put/call ratios during the summer of 1996. In Jul!; the equity-only put/ call ratio surged above the upper band to le\-els not seen since January 1991. The OEX managed to climb briefly abol-e its upper band, but was essentially in a trading range for two months. \\%ile it is impossible to tell if arbitrage activity in the OEX was responsible for this divergence, anyone living through the decline in Julwould probably attest that it felt more like the picture show by thk equitv-only indicator than the OEX.

As technicians, we know that in the absence of other factors, trends tend to persist. I would anticipate that the OEX put/call ratio will continue to exhibit declining performance against both the equitv-only put/call ratio and the market in general. The OEX putjcall ratio is likely a victim of structural changes in the marketplace, the growth of institutions and arbitrage desks, and the relentless march of technology In the perverse isay of the market, its own success probably also contributed to its downfall, as more and more investors watched its signals.

However, the equity-only put/call ratio still appears to be a useful sentiment tool. It is less widely watched than the OEX put/call ratio. This may extend its longevity, despite the downward trend in performance. It is a testament to the elegant design and logic that 25 years later, through bull and bear markets, and unbelievable changes in the marketplace, this indicator is still a valuable addition to the market analyst’s toolbox.

Footnotes

1. Martin E Zweig, “Option to Sell: Sew Puts/Cnlls Ratio Is Signaling an Intermediate Market Decline, ” Barron ‘s, 10 MUJ 1971, 9.

2. The indexput/call ratio is usually carried on quote machines as .PCI or some combination thereof: The equity put/call ratio is usuall,y . PCE.

3. “CBOE Produrt Listing Dates, ” The Chicago Board Options Exchange, 400 S. LaSalle, Chicago, Illinois, 60605.

4. LEAPS are essentia@ long-term options with issued with maturities of two Jears or more.

5. Conversation with Dennis E. Jarrett, CM?; Jarrett Investment Research, February 1996.

6. Tushar S. Chande and Stanley Kroll, The AV/ewebj Technical Trader: Boost Your Profit h Plupting Into the Latest Indicators (AVpw York: John M’ilrq’ tJ Sons, 1994/, 7-9

7. Investors Intelligence, 30 Church Street, Xew Rochelle, .\Y 10801.

8. Historical Comparisons of ,Monthly Iblume and Open Interest, ” The Chicago Board Options Exchange, 400 S. LnSalle, Chicago, Illinois, 60605.

9. John Bollinger, Bollinger Capital Management, i?O. Box 3358, Manhattan Beach, CA 90266.

10. This benchmark was calculated bJ finding the 13 da? rute of change for euq day over the lO?ear test period and calculatzng an average.

11. Indicator Performance - Xormal StiP Performance = Indicator Outpelformance (0.95% -0.61% = 0.34%).

Selected Bibliography

• Hines, Ray “Hines Ratio,” Technical Analvsis of Stocks & Commodities, April 1989.

• Martin, James P., “Options ratios for sentiment,” Technical Analvsis of Stocks SC Commodities, June 1990.

• Leonard, Brent L., “Heed the \‘.O.I.C.E,” Market Technicians Association Tournal 41 (Summer 1993): 12-13.

• Lund, Carsten, “Skewness: An Options Based Indicator to Measure Sentiment” Market Technicians Association Journal 43 (Summer/Fall 1994) : 29-38.

• McMillan, Lawrence G., Ootions As X Stratezic Investment, New York: New York Institute of Finance, 1993.

• Mchlillan, Lawrence G., “Put-Call Ratios,” Technical Analvsis of Stocks & Commodities, October, 1995, 98.

• Neill, Humphrey B., The Art of Contrarv Thinking, Caldwell: Caxton Printers, 1976.

• Prechter, Robert R. and Dal-id A. Allman, “Put-call sentiment indicator,” Technical Analysis of Stocks 8: Commodities, January 1990.

• Pring, Martin J., Investment Psvcholow Explained: Classic Strategies to Beat the Markets. Sew York: John L$‘iley & Sons, 1995.

• Schaeffer, Bernie, “Timing’s the Thing: Bernie Schaeffer,” Interview by Thorn Hartle, Technical Analysis of Stocks & Commodity, November 1997, 72.

Biography

Lawson McLVhorter is a proprietary trader at C.E. Unterberg, Towbin. Prior to Ilnterberg, he was a proprietar) equity and options trader at Genesis Capital Management, L.P., a hedge fund based in New York City Lawson got his start on Wall Street as a research assistant to Dennis Jarrett, CMT, the Chief Market Analyst at Kidder, Peabody Inc., and later followed Dennis to his own firm, Jarrett Investment Research, Inc. Lawson graduated with a degree in economics from Dal-idson College, in Davidson, North Carolina

Introduction

As the world shrinks to become one global community with borders and language barriers disappearing, individuals become more receptive to the ideas, methods and techniques of other cultures. Such has been the case in recent years in the C.S. concerning market analysis methods historically used by Japanese investors. Thanks in most part to the efforts of analyst and author Steve Nison, CMT, who researchedJapanese methods of technical analysis for his books Jananese Candlestick Chartirw Techniques and Beyond Candlesticks, such terms as “candlesticks,” “kagi” and “renko” are familiar to American technicians. Candlesticks are now included in most computer programs for technical analysis. These methods have similarities with certain techniques that have been in use in the IVest for manv vears, but each has its own unique flavor providing differing vantage points for viewing markets. Of particular interest is kagi charting and its comparison to M’estern-style point-and-figure chart analysis. Both methods belong to the older, chart-reading analvtical school as opposed to the more modern mathematicallv-derived technical anah-sis. .4.s a consequence, they engender a ckrtain amount of subjec&ity in the interpretation of trading signals. In comparing the two techniques, this paper presents a brief description of how kagi and point-and-figure charts are created and the strengths and weaknesses of both. Using a compilation of simple and complex primary trading signals, the effectiveness of each method is reviewed. Of necessity, only the basics are studied in an attempt to delineate an objective contrast between the two techniques, providing the analyst with a clearer choice of tools.

Kagi and point-and-figure charting originated at approximately the same time. Although no definite date or inventor of either method is known, both began to appear in the 1870s. In the IVest, the writings of Charles Dow contained the theories preceding the development of point-and-figure charting. In the East, it is interesting to note that Japan had a functioning forward rice market since the 16OOs, and technical analysis was already in use. In 1868, an imperial government was restored in Japan and the Emperor invited foreign advisors into the country to help modernize the near-feudal society A stock market opened for trading in the 1870s. The increased commingling of East and West facilitated the transfer of information and ideas on market analysis, furnishing a common breeding ground for both analytical techniques.

The two charting techniques share an ideology that seeks to present a clear visualization of the forces of supply and demand and bullish and bearish trends while excluding irrelevant price fluctuations. Neither time nor volume are factors in the creation or analysis of either method. Recurrent chart patterns are used to infer market action.

Kagi

Kagi charts are drawn with straight lines, the thickness of the line depending on whether the market is in a bullish or bearish mode. A bullish or rising trend in prices is designated by a thick kagi line called the “>-ang” line. The thin kagi line, called the “yin” line, is used when prices are falling in a bearish trend. The line changes from thick to thin or pang to yin (bullish to bearish) and lice versa when the trend changes, allowing a quick and clear reading of whether a market is in a bullish or bearish trend. The proportion of a line that is yang versus yin (thick I-ersus thin) indicates the relative strength or weakness of the market. ‘Kn” and ‘Yang” are ancient Chinese philosophic and religious terms that fbrmed the basis for Chinese medical practices as far back as the 8th century B.C., and it is likely through medicine that the\ were introduced into Japanese culture. They signify opposing forces that are naturally seeking equilibrium and a dynamic balance. In market terms, they characterize the struggle between supple and demand.

Selecting a “reversal size” is the first step in constructing a kagi chart. The “reversal size” is the amount market prices must reverse before a change is made in the direction of the line on the chart. This dictates the sensitivity of the chart and the magnitude of the trend sought. A small reversal size can lead to many minor fluctuations that obscure the trend. Conversely a larger reversal size reveals only major trends. \hlatility and the general price range of a market should be considered when determining reversal amounts. Commonl>; the reversal size on a kagi chart is states as a fixed number of pomts. However, it is sometimes calculated as a percentage of the price. offering advantages similar to a semilogarithmic bar chart. By reducing all price movements to the same relative significance, a percentage reversal size allows a direct comparison between high-and-lowpriced charts. For example, a chart of a Sl price change when prices are at $10 will look identical to a chart of a $10 price change when prices are at $100. This is most helpful in analyzing a long-term trend that displays a significant price increase or decrease.

Price is marked on the vertical axis with only a column number (optional) on the horizontal axis. If prices are rising, a thick, or yang line, is drawn from the beginning price until a change in price equal to or greater than the reversal size occurs. At this point, a horizontal line is drawn to the next column where the line is then drawn down to the new lower price. The horizontal line is called an “inflection point.” The downward line is continued until the next reversal occurs and another inflection point is drawn. An inflection point is called a “shoulder” when it turns the line from up to down and a “waist” when it turns the line from down to up. Breaking the previous waist engenders a change in the line from vang (thick) to yin (thin). This denotes a trend change from bullish to bearish. \2’hen a preceding shoulder is broken, the line changes from yin to yang or thin to thick indicating a change from bearish to bullish. ,\ change in thickness of the line (breaking a previous shoulder or waist) is an elementaly buy or sell signal.

Illustrations A-D show four patterns presenting important trading opportunities. The names reflect the images symbolized by the patterns. The “multi-level break” pattern (Illus. A) is an amplification of the basic buy/sell signal. Instead of acting on the initial signal, some traders look for verification bp waiting for a second or even third shoulder (or waist in a bear market) to be broken, hence the name “multi-level break.” The inflection points being broken should be close to each other in price. The “three-Buddha” and “reverse three-Buddha” patterns (Illus. B) are similar to the standard head-and-shoulders patterns on IVestern bar charts and are topping and bottoming formations. A “two-level break” pattern is created when the “neckline” i.e., the highest shoulder in the pattern, is exceeded. The “double-windows” formation is an extension of the three-Buddha. At a double-windows top, (Illus. C) shoulders bracket one or more waists followine an uptrend. LVhat differentiates this pattern from a multile;el-bre’ak pattern is that neither shoulder is as high as the waist(s), leaving a small “window )( on either side. The reverse is true for a double-windows bottom. These patterns give clear by and sell signals on breakouts.

Several techniques allow the analyst to more objectively assess the strength of the current trend. The length of the yang portion of a line versus the yin portion of a line indicates whether bulls or bears are the stronger force. The midpoint of a kagi line offers significant support or resistance, much as do 30% corrections on bar charts. If a correction halts before the 30% level is reached and then reverses to break a previous shoulder or wAst, that break becomes a stronger buy or sell signal. h grouping of waists or shoulders at the same price level indicates important support or resistance.

As on most charts, a series of higher highs and higher lows or lower highs and lower lows creates an image of the trend. Trendlines are useful as secondan tools. X trendline break often occurs after a kagi reversal, pro&ing confirmation of the earlier signal. On a kagi chart, a series of nine (not necessarily consecu tive) higher shoulders or lower waists is called a “record session” 1 (Illus. D) and indicates that the trend may be reversing soon. Previous areas of support become resistance after a market rt verses and vice versa. Adherents of kagi charting recognize thl importance of double tops and double bottoms, referring to then as “tweezer tops” and “tweezer bottoms.” Taken together, these features strengthen the analyst’s bullish or bearish interpretation and help with stoploss placement.

Point-and-Figure

In point-and-figure charting, displaying price movement us ing Xs and OS to fill in squares or boxes on graph paper as OF posed to a straight line going from point to point gives the char a different look from the kagi chart. A column of Xs is made a prices rise and boxes are filled in. Declining prices are shown b a column of OS. As on kagi charts, prices must change by a prc selected reversal size before a column of Xs is ended and a co umn of OS is begun and vice versa, but in point-and-figure char ing, reversal size is the number of boxes that the market mw reverse rather than the number of points. Each box or square i assigned a point value and a new X or 0 is made on the chal only when the price has moved in an increment of a whole bor Fractions of a box are ignored. For example, if the box size is designated at two points and the reversal size is three, the markt must move three boxes or six points in the opposite direction before filling in the next column, i.e., a stock price that trades as high as 60 1.i/eweb in an uptrending market must decline to 54 (thre full boxes) before a column of OS are started on the chart. Th high at 60 l/2 is regarded as filling the box at 60, so anythin above 60 is ignored until the market reaches 62 (another bo higher). A larger box size condenses the chart and is useful fc high-priced markets and on long-term (weekly or monthly) char1 to display the primaT trend. The combination of different bo sizes and reversal sizes can reduce the chart impact of wide pric swings in a volatile market and filter frequent minor reversals c “noise” from the chart, leaving major trend changes more e\ dent. Three-box reversal sizes are common in the volatile future markets, but the reversal size can be whatever is considered opt ma1 and may take some trial and error to determine. Because box is filled’onlv on a price change equal to the full box size c more, absolute highs and lows that are less than the full box size do not appear on a point-and-figure chart. Loss of these data is drawback but does not affect interpretation of chart patterns.

Point-and-figure patterns de\-elop, as in kagi charting, that are onsistent and classifiable, providing guidelines for judging fuire market activity In earlvwritings, the point-and-figure method ‘as often described as “sc/ewebentif? or “mechanical” because it is nalogous with the physical principle of leverage utilizing a price ongestion as a “fulcrum” and the invisible forces of supply and emand as the “lever.” As those forces shift, a “catapult point” is eached and the market “springs” away in a new direction. Contruction of the chart ignores gaps and uses the closing prices of :t time periods, as does the kagi. Those time periods can be of ny length. Short-term traders often track every price change. A ample buy signal comes when a column of Xs exceeds the high f the pre\ious X column. h downward column of OS creates a -11 signal when it breaks the low of a previous 0 column. Simple uy and sell signals occur frequently without tradeable Alowthrough. Complex patterns often provide signals for more extended price mores offering an improved chance of profit.

All patterns are a combination or expansion of the three elementam fulcrum or basing patterns: ideal or full, broad, and recoil. “Catapult points” or “breakout points” also come in three styles: true, false, and semi-catapult. The “ideal fulcrum” pattern (Illus. E) is similar to a head-and-shoulders pattern and the three-Buddha kagi pattern, which develop when supply and demand are in balance. Prices mol-e up and down but remain in a range. The “catapult point” occurs when a column of Xs exceeds the tops of previous X columns. The reverse occurs on the down-side when a column of OS exceeds the lows of previous 0 columns, and prices continue lower, confirming an “inverted ideal fulcrum.” If no followthrough develops from the catapult point and prices return to the area of the fulcrum, the ideal or full fulcrum becomes a “broad fulcrum” (Illus. F). The “recoil fulcrum” (Illus. G) develops after a sharp move. At the bottom (or top) of an extended column, a group of columns develops with each succeeding column shorter than the one before, creating the recoil fulcrum which generally has a triangular shape. Breakouts from a fulcrum typically occur in conjunction with chart patterns such as “triples,” three X columns with the tops at the same level or three 0 columns with lows at the same level, where multiple columns are broken simultaneously (Illus. I) or “triangles,” called “signal” formations (Illus. J). The latter are differentiated from recoil fulcrums because they are part of larger formations rather than reversals at major highs or lows.

Webster’s Colleze Dictionan’s definition of “catapult” is “to move quickly suddenl? or forcibly,” as the market is expected to do when a catapult point is reached. A true catapult point leads to the development of profits quickly with little or no correction from the breakout. A “false catapult” is a failure, usually indicated by a greater than 50 R correction of the breakout column and warns that the consolidation may not have been a fulcrum. A “semi-catapult” (Illus. H) develops during a pause in a price move after prices have traded sideways and the chart shows one or more reversals in a narrow range. The “secondary signal” or “semi-catapult” occurs when the original move resumes. “False semi-catapults” fail to follow through as expected.

These are the basic trading patterns involved in point-and-figure charting. Because each pattern has many variations, specific patterns are sometimes difficult to recognize, but the underlying signal to buy or sell always occurs on a breakout.

Trendlines offer supplemental information on point-and-figure charts, also. They are drawn as on a normal bar chart connecting a series of highs or lows. An exception is on a three-box reversal chart, where trendlines are most frequently drawn at a 45 degree angle up from a low point or down from a high point. Although these display the trend, they are more accurately described as support and resistance lines. In either method of trendline construction, penetration of a trendline needs accompanying buy or sell signals to indicate a valid trend change.

A feature unique to point-and-figure charts is the ability to arrive at price objectives through the use of horizontal or vertical counts, an aid in determining whether the potential reward of following a trading signal is sufficient given the measured risk. The sides of a fulcrum provide the parameters for horizontal counts with the total then added to the low of the fulcrum. A vertical count can be calculated by multiplying the number of boxes in the first column of a new trend by the reversal size. No comparable technique exists on kagi charts. Consequently this feature, while a definite advantage for point-and-figure charting, is not applicable in comparing it with kagi charting.

Test

A comparison of trading signals on daily kagi and point-and-figure charts in several markets over an l&month time span between October 1993 and July 1995 was undertaken to determine if one method led to more profitable trading signals than the other. The chosen markets had high-I-olume trade activitv but were from different market segments. Daily charts and three-box l-e\-ersals were selected to facilitate comparison between the two methods and to provide a miiform test. This is not to be construed as a preference for a three-box reversal size or that these methods work best on dailv charts. Optimization of reversal sizes for specific markets is a topic for another study. Each chart was analyzed using the most rudimentary buy and sell signals and again using more complex, clearly visible patterns. The first procedure was in the market at all times, exiting and reversing positions at the same time. The latter method utilized previous simple signals as exit points to take profits and limit losses. Restricting the second study to only the most obvious formations minimized subjectivitv. Specialists in these analytical techniques may see more esoteric trading signals that were either missed or ignored in this studv. More aggressive traders may also use different rules for stop-loss placement. No consideraiion was given for transaction costs. This study is meant as a guideline for the analyst or trader who is considering adding kagi or point-and-figure analysis to his/ her blend of technical methods.

Results

Using simple signals. no advantage was found using either kagi or point-and-figure as both resulted in losses, although losses using kagi techniques were less than with point-and-figure. In both cases, severe losses in the soybean market were a primal? factor in the negative results. This was likely due to the volatile nature of soybean price movement and mav be representative of futures markets in general, requiring optimized reversal and/or box sizes to improve profitability, but again, that is a topic for another study. Disregarding the results of the soybean test improved the overail outcome and gave the edge in profitability to the point-and-figure charts. In addition, a lower level of relative risk was indicated with point-and-figure, but the standard deviation of both methods still indicated a dispersion of results that is too wide to be relied upon.

In every market studied, the test results clearly illustrate the advantages of exclusivel: following complex patterns. Both methods were profitable. Filtering out smaller moves and ignoring the myriad simple bw/sell signals focuses on larger trends, greatly enhancing profitability per trade. The contrast between using simple and complex trading signals vas speciallv apparent in the sol-bean market. A side benefit was fewer trades resulting in lower to;al commission costs. Complex signals reduced the number of trades to an average of one trade every fiye months I-ersus an alerage of one trade every two months following simple signals.

Kagi charts offered a performance edge over point-and-figure at this level, returning a total of 38% more in the five markets with a slightly larger average profit per trade. This superior performance was primarily due to earlier entry signals and more profitable exit signals directl: resulting from point-and-figure’s use of whole box sizes and the consequent loss of accuracy. However. kagi charts had a standard deviation of 89.6% versus 52.7% for point and figure, indicating that point-and-figure, while not as profitable, was significantly less risky.

Conclusion

Both kagi and point-and-figure charts provide excellent illustrations of trend and offer alternatives to other trend indicators. They are similar and yet distinctive. The kagi chart presents a clearer picture of the strength or weakness of a market and is easier to draw. Its simplicity and clarity make it perfect for the novice trader while its profitabilit? is attractive to all traders. Unfortunately, the popularity of kagi charts is limited because only a few U.S. computer charting programs offer them. However, the lower risk of point-and-figure charting makes this method yen attractive, particularly for the trader with a lower-risk profile: When other considerations are factored in, such as the wide availabilitv of point-and-figure charting and the ability to calculate price’objectives, point-and-figure holds a decided advantage over kagi charting.

lnterpretation of signals on kagi or point-and-figure charts comes from “reading” the charts, and is dependent on the analyst’s abilir): to recognize and identie those signals when they occur. Famlharization with the basic patterns helps to develop an “artistic eye” that quickly recognizes complex patterns when they develop, sometimes anticipating a formation in progress and preparing for action. Several refinements and advanced tools are available in both methods to further enhance profitability Trendlines (or support and resistance lines) and 30% corrections can add confidence to trading signals, provide caution flags to uncertain signals and enhance stop placements. The profitability of complex trading signals in both techniques makes it worthwhile to initiate a system of periodic searches for these patterns in a variety of markets no matter lvhich analytical technique is preferred.

The usefulness of kagi and point-and-figure charting has been confirmed by over a century of testing. Choosing one over the other is a personal decision to fit each analyst’s trading style. The time and effort spent learning to understand both of these techniques will be rewarded with greater profits and a better comprehension of the ebb and flow of bullish and bearish forces in the markets.

Bibliography

• Robert D. Edwards and John hlagee, Technical Analrsis of Stock Trends. Fifth edition. Boston, RL-\: John Rlagee Inc. 1966

• Steve Sison, Beyond Candlesticks-More Tapanese Charting Techniques. New York, !KY John iVile!- 8: Sons, Inc. 1994

• Steve Nison, T/v 1771 nrzd lirngo/Tding, Futures Slapazine, Oster Communications, November 1994

• \‘ictor de \‘illiers and Owen Taylor, The Point and Figure Method of Anticipating Stock Price Rlovement, Basic Princinles - Book 1. New York, AY distributed by Morgan, Rogers and Roberts, Inc. 1931

• .12: Cohen, How to Use the Three Point Reversal Method of Point and Fiwre Stock Market Trading. Sew Rochelle, AY published by Chartcraft, Inc. 198’7

• Alexander H. IVheelan, Study HelDs in Point and Fizure Tech. New York. PUY Morgan. Rogers and Roberts, Inc. 19.51

• The Sew Grolier hlultimedia EncycloDedia. Release 6, Groliers Inc. 1993, Online Computer Systems Inc. 1987-1993

• Charts from Metastock b!- Equis International. version 45 RT, 19851994

• Data from Signal by Data Broadcasting Corporation, San Mateo. CA

Biography

Julia Bussie is currently a private trader in southern California concentrating on the futures markets. Prior to this, she was an analyst with A.G. Edwards in Chicago. She was a member of the Chicago Mercantile Exchange for eleven !-ears and a member of the Chicago Board of Trade for ten years. She has been involved in many aspects of the futures industry over the past twenty-five years and has written numerable market commentaries, newsletters and long-term outlooks with a more fundamental bias, but is now focusing almost entirely on technical analyis. Julia is the Admissions Chairperson’for the Market Technicians ,Association, Inc. and can be reached at Jebussie@aol.com.

(Click To Enlarge)

(Click To Enlarge)

(Click To Enlarge)

Having an edge, that additional skill or strategy, has been sought after by investors and traders since the first stock or commodity was traded. Man!: believe that investors such as George Lane, Gerald Rappel, Martin Pring, and others fomid edges when they identified and helped popularize momentum indicators. In order to add to the body of knowledge regarding “momentum investing,” this paper details a market occurrence in which momentum is so extreme, that, historically there has been a tendency for non-average market behavior. Observations over the last several years led me to believe that when the S&P 500 closed on its extreme high or low for the da!; there seemed to be immediate follow-through. In this paper, I describe my research as I attempted to veri? my follow-through assumptions through observing the daily high, low, and close of the fiye davs subsequent to closing on the extreme.

Background

In its most basic form, momentum measures the rate of price ) change. Furthermore, changes in momentum often lead to changes in trend. Thus, focusing on points of extreme momentum may allow investors to take advantage of the lag period between a potential momentum peak and a trend change. Market hlakers and Specialists adrust the bid and ask prices based on changes in the incoming stream of orders. The relevant information is deciphered, analyzed, and acted upon at different speeds, and the net effect often results in an ebb and flow type of price movement. Therefore, once the buying and selling pressure begins to swell, it tends to not stop abruptly. The closing bell, however, can act as an artificial mechanism that temporarily suspends the buy/sell equation. ,%n analogy can be made to a person who pauses a videotape, and then assumes that the scene will continue upon release of the pause.

Methodology

In order to examine any links between a close on the extreme and immediate continuation, the following stud! observes the S&P 500 from l/1/82 to 12/ 31/97. Data were obtained from Pinnacle Data (l/82-2/95) and Telescan (2/95-12/97). The beginning date was selected as January 1982 because, prior to that time, theoretical (rather than true) high and low data were maintained b!: the database. In this way, the Closing at the Extreme study contams the entire population of true close on high and close on low information. During this 16-vear period, the market was mostly in a major uptrend, though bearish periods were experienced: mid-1983 to 1983, late 1987, mid-1989 to 1991, and 1994. Therefore, the test incorporates bullish as well as bearish periods.

The high and close for the subsequent fire days are compared to the Close on High (COH) day and the low and close are compared to the Close on Low (COL) day for the same period. Between 1982 and the last trading day in 1997. the S&P 500 closed on its high for the day 4li times, or 10.31% of the time, compared to closing on its low only 94 times. or 2.32% of the time. Rather than focusing on the cause of such a discrepancy the purpose of the stufl, is to observe the events subsequent to the occurrences. As addltlonal background, there were a total of 404.5 trading days during the l&year period during which the S&P 500 rose 847.i9 points. The average dail!- return, therefore, is 0.21 points or 0.1956% -the benchmark to which the study will be compared.

Results

The results of this study can be examined from two perspectires. The first is through observing the historical tendency. The second incorporates basic statistics in order to determine whether or not the observances are meaningful.

Figure la displays the results comparing the Close on High day (COH) versus the subsequent five trading days. Sotice the pattern of the subsequent days’ highs: the greatest average in-crease (percent) is the day following the COH day and the lowest is on the fifth day. Another observation is that the percentage of highs greater than the COH day stayed over 50%. for each of the subsequent days. The ratio that compared the subsequent closes to the COH day also stayed above 50%. These obsewations support the hypothesis that there is a tendency for subsequent highs and closes to stay above the COH day as follow-through action takes place.

Upon changing the focus from frequency of occurrence to average price movement, however, a different picture comes to light. Figure 2a display the average point change data from Figure la and compares the data versus the average point increase (0.21 pts) of the S&P 500 during the l&year test period. A for price changes, the highs of the subsequent five days stayed between gains of 1.75 and 2.12 points; this is consistentlv above the previously mentioned S&P 500 average increase of 6.12 points. The closes of the subsequent five days stayed between up 0.20 and 0.56 points, also above the S&P’s average. Plotting the average S&P 500 gain with each subsequent day 0.21 points higher than the previous day yields an important tendency. Each dar’s average close was less than the S&P average gain. An additional observation is that the gain made by each subsequent day’s high averaged just under ten times that of the average S&P 500 gain. Such action displays a series of failed intraday rallies. Based on these observations, traders may want to use a COH day as a signal that long positions should be closed during the next trading session (COH tl) in order to avoid the choppy trading sessions that have historically followed the extreme move. It is proposed that the observations should not be used as a stand-alone svstem, but rather as an additional piece of information that could be incorporated into an existing system

The historical tendency suggested that action should be taken: specifically, closing the position during the next trading session. This conclusion led to the possibility that there exists an ideal exit point. Through the use of basic statistics, the first step \vas to determine the mean (average) price change. This step was the focus in the Figure 2a observation. The second step was to determine the standard deviation from the mean. Figure la displays the standard deviations for the highs and closes for the five trading sessions subsequent to the COH day. The standard del-iations ranged from 2.20 to 6.49. Such high standard deviation values suggest that the subsequent highs and closes fell within a wide zone. M’hen the standard deviation is subtracted from the mean or even the median, prices below the COH dav would have been included for all five of the subsequent five days. The statis- tical observation, therefore, suggests that an ideal exit point cannot be determined with enough accuracy to be significant.

Figure lb display the comparison between the Close on Low day (COL) and low and closes of the subsequent five days. During the 1982-1997 period, the S&P 500 closed on its low onlr 92 times versus closing on its high 417 times. Refer to Figure 26.

The average low of the subsequent five days consistently sta!-ed below the COL day with the lowest points being the lows of the first three days. Such action displays the tendenc!- for the S&P 500 to have downside follow-through. However, the only average close that staved under the COL dal- was on COL tl. On days COL t3, t4, ‘and t5, the average cl&e was not only above the COL da); but above the S&P 500 average gain. The historical tendency therefore, is that the S&P 500 attempts to strongh- extend the selloff intraday during the first two days, but then experiences a rebound that results in an above average close tendencl during the following three days. Based on this tendency investors may want to close short positions following a COL occurrence. during the first two days Aggressive investors could establish long positions in an attempt to take advantage of the rebound tendency that occurs on davs COL t3, A, and t5. Given that this strategy ‘is based on tendency, actual exit and entr: points could be based on the investor’s own system.

The COL statistical analysis had similar results to that of the COH observations in that the range of standard deviations (2.9’7 to 8.18) was large. Thus, an attempt to predict the likely turning points within an acceptable range appears to be futile.

Any time a market tendency is identified. its real-Jvorld application often brings out its weaknesses and limitations. In the situations presented, profit would be highly dependent on the investor‘s fills which could experience a high degree of slippage due to the increased volatility surrounding the COH/COL days, including gap openings. The choice of trading vehicle may also limit the usefulness of the COH/COL tendencies. S&P futures could be traded though their accompanying put/call options and would likely haye many other factors needing consideration. As for the study itself, the data have been limited to the subsequent five days. Perhaps the COH/COL tendencies are longer lasting, though the data have been limited in an attempt to focus on the most immediate affects of the COH/ COL days. -bother limitation may arise from the definition of a COH/COL day as being out two decimal places. Therefore, the S&P 500 may close at 431.32 after being as high as 431.33 and be missed by the system, even though the noted tendencies may still exist.

The limitations of this’study seem to point out the direction in which additional research may be needed. That this COH/COL study focuses just on all of the COH/COL days by the S&P 500 from 1982-1997 brings up several questions. Is the tendency maintained during prolonged bear markets? Are the tendencies the same for other indices, individual issues, or even commodities? Does the degree of upside/downside bias vat-y based on where in the trend that COH/COL day occurs? Would the probability of follow-through be higher if the COH/COL day was confirmed bu technical indicators? As the questions are answered, investors could perhaps integrate the COH/COL tendencies in an attempt to further refine their systems.

Several conclusions can be derived from the Closing on the Extreme stude First, there was a tendency for the S&P 500 to show strong follow-through during the intra-day action on the day following the extreme move (for both cases). Perhaps more importantly than the first, if the S&P 500 closes on its extreme, look for contractions to occur by the closing bell in the subsequent trading sessions. W’hile this study does not reveal the anticipated clear-cut trading edge, at least the path has been taken, one that may allow investors to take one step closer to that elusive trading edge.

Bibliography

• Murphy John J., Technical Analysis of the Futures Market, New York, hjy: New York Institute of Finance, 1986.

• Lane, George, The Theorv of Momentum Indicators and Lane’s Stochastic, Dow Jones Telerate Seminar, Las \‘egas 1995.

• Nison, Steve, JaDanese Candlestick Chartine Techniaues: a Contemporary Guide to the Ancient Investment Techniaue of the Far East, Sew York, NY: New York Institute of Finance, 1991.

• Pring, Martin J., Technical Analysis ExDlained, New York, NY McGraw Hill, Inc., 1991.

• Pring, Martin J., Martin Pring on Market Momentum, International Institute for Economic Research, Inc. 1993.

Biography

Chris is currently employed by Coastal Trading Management, LLC in U’ilmington, North Carolina as Senior Trader/ Portfolio Manager for the Low Volatility Fund, LP. In 1993 he cofounded the technical analysis department for Olde Discount Corporation and served as Senior Technical Analyst until May 1998. \4%ile at Olde, Chris led many seminars instructing brokers and traders in using technical analysis. Prior experience also includes the position of registered rep resentative for Olde as well as Raymond James and Associates. In 1989 and again in 1990 while pursuing a Bachelor of Science degree from Florida State Cniversit!; Chris placed in the top 1% in the AT&T/USA Today National Collegiate Investment Challenge. Chris dedicates this paper to his children, all three of which were born during its development.