FORMULA RESEARCH TM

Quantitative Treatment of the Financial Markets

Volume VII, No. 2 September 10, 2003 Nelson F. Freeburg, Editor

Notes & Comments T HE M ONITOR S ERIES : L EADING M ARKET P ROFESSIONALS S HAR E U NCOMMON I NSIGHTS

{ Don't Miss Part II Synergy and Collaboration: The Asset Management

Team of Tom McClellan and Roger Kliminski, Part I
As sometimes happens, the material developed
in this study proved to be richer and more Two Analysts with Bold Vision Achieve Market-Beating Returns
extensive than a single report can accommo-
date. To do justice to the findings I had to
divide the study into two parts. You will receive
the second installment in a couple of weeks.
O ver thirty years ago market analyst Sherman McClellan and his
There we present some of the most profitable
and risk-averse timing methods we have ever mathematician wife Marian introduced a revolutionary market
featured. timing tool. Today the McClellan Oscillator stands as one of the most
{ Help for Market Professionals... popular and effective technical indicators of all time.
I am privileged to serve money managers
of distinction in 26 countries. It's been a
The McClellans went on to create a host of advanced technical
pleasure to come to know many of you methods. Ten years ago their son Tom joined the research effort. A
personally, even if by the exquisitely West Point graduate who served as an Army helicopter pilot, Tom retired
remote medium of email.
from active duty as a captain. In 1995 Tom and Sherman, after much
As you know, over the past few years preparation and research, founded a first-rate market advisory service. I
there has been a lot of consolidation and
cost-cutting in the U.S. financial industry. closely follow the McClellan Market Report and its companion service, the
Due to the retrenchment, many skilled Daily Edition. Tom presides over both publications in consultation with
young professionals have suddenly found
themselves without jobs. Some of these
Sherman.
bright and energetic individuals have
sought out my help. Given the family's signature contribution to the field, one aspect of
My pitch: If you are a money manager or the McClellan approach to technical analysis may come as a surprise. The
investment advisor in a position to hire, at McClellan Oscillator is by no means the centerpiece of the analytic effort.
any given time I can put you in touch with
a small roster of highly qualified market Yes, the eponymous indicator plays a featured role.
analysts and traders. Please let me know (Continued on Page Two)
if I can help in any way.

Zweig Bond Timing Model--March 1993

In his 1987 book Winning with the New IRAs,

Martin Zweig presented a simple but power-
ful fixed-income timing model. The model
(which was developed by Ned Davis
Research) combines trend-following
elements with fundamental inputs.

We featured the Zweig bond model in our

March 1993 issue, where we added a few
refinements. All versions of the model have
continued to perform well since then.

Going back to 1970, our variant returned

12.9% a year compared to 10.2% for the
Dow Jones Corporate Bond index. Maximum
drawdown was held to 4% compared to 21%
for buy-and-hold. The chart at left shows
recent signals applied to the Fidelity Invest-
ment Grade Bond fund (FBNDX).

Copyright 2003, Reprinted by McClellan Financial Publications by special permission. All Other Rights Reserved
FORMULA RESEARCH, Inc., 4646 Poplar Ave, Suite 401, Memphis, TN 38117, A twelve-issue subscription is $295. A six-issue trial is $175. Overseas surcharge: 20%. (901) 756-8607
9710
9
 Volume VII, No. 2 F O R M U L A R E S E A R C HTM Page 2
Quantitative Treatment of the Financial Markets

But Tom's commentary ranges far devoted to timing interest rates, precious
{ Philadelphia
Area Subscribers
beyond, tapping a rich array of diverse metals and the stock market. We'll
timing methods and technical tools. It is present systematic strategies for all three
John Durham of
Villanova, PA would
tempting to think of Tom as a one-man sectors based on Tom (and Sherman's)
like to contact Ned Davis Research--a dogged individual original research.
subscribers in that doing the work of an entire team of first-
area to exchange
views, programming rate institutional analysts. Be advised that Tom himself is not a
tips and related strictly mechanical trader. While he relies
ideas. You can
reach John at (610)
Of course, Tom continues to benefit on rigorous historical testing, Tom blends
519-0867. Email: from his father's insights. (For one thing, his findings with personal insights from
pd23@
mindspring.com.
Sherman's hawk-eyed text editing helps dozens of timing tools, most of them
make Tom's prose some of the most liter- unorthodox.
ate and polished in the field.) Meanwhile,
several years ago Tom entered into a This mix of quantitative research and
money management partnership with intuitive judgment has served Tom well as
Roger Kliminski. a market analyst. According to Timer
Digest, a ratings service that tracks over
Roger is a gifted market analyst, an 100 market advisory programs, Tom ranks
experienced portfolio manager and a No. 2 for long-term stock market
longtime friend of Formula Research. I forecasting over the past five years and
have seen the independent performance No. 3 for intermediate timing. Tom ranks
rankings and Global Investment No. 1 in precious metals timing for the
Solutions, the partners' money manage- same period.
ment arm, has consistently beaten the

For information on market with low levels of risk. Tom's body of research ranges so
the McClellans twice
monthly newsletter and
widely we can't begin to feature every
Daily Edition, visit their We'll touch on Roger's contribution unique indicator here. If you explore his
web site at
www.mcoscillator.com
at the end of this report and in greater work in depth, you'll be treated to a fasci-
detail in Part II of this two-part study. In nating array of innovative timing methods.
or call (800) 872-3737 particular, we'll highlight a key feature of Interested in precious metals? Follow
or (253) 581-4889. If
you prefer to write, the price behavior that counts as an authentic Tom and plot the difference between
address is: market discovery. Building on these 1-month and 12-month gold lease rates, as
McClellan Financial
Publications, Inc.
findings, we'll develop two timing models reported by the London Bullion Market
P.O. Box 39779 that show gains of better
Lakewood, WA 98439
The fax number is
than 25% a year. z
(253) 584-8194.
25%
$600
Cash Gold
Offset 15 Months Forward
The McClellan 20% (Scale Right)

Research Imperative: 15%

$400

Quest for Innovation

10%

Higher $200
For now we focus 5% Rates
on the McClellan side of 90-Day T-Bill Yield ???
(Scale Left)
the partnership. Tom's 0% $0
principal efforts are
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
Volume VII, No. 2 F O R M U L A R E S E A R C HTM Page 3
Quantitative Treatment of the Financial Markets

Association. You'll find

the spread revealing. 16%
10%
U.S. Unemployment Rate Lower
14%
(Scale Right) Unemployment
Trade fixed income? 12% ??? 8%

Take a leaf from Tom's 10% 6%

book and track Commit- 8% CPI Inflation Rate
ment of Traders data for 6%
Offset Forward 2 Years
(Scale Left)
4%

copper futures. When 2%

commercial copper
4%

0%
traders get heavily net 2%

long or short, bond 0% -2%

1983
1984

1985
1986
1987

1988
1989

1990
1991

1992

1993
1994

1996

1997
1998

1999
2000

2001

2002
2003
2004
prices tend to go in the
opposite direction. I
could cite many other by about two years. The fit is not perfect
examples of offbeat but inspired but the overall symmetry is impressive. z
indicators.

One favored McClellan timing

method is intermarket correlation. The The Goldollar Index
idea is to select a market with concrete
forecasting significance for a second
market, which tends to follow with a Intermarket pioneers like John
predictable lag. You can then synchronize Murphy and Martin Pring have long
the parallel swings by realigning the time pointed out that precious metals tend to
scales. Tom did not invent this method move opposite the U.S. dollar. A strong
of analysis, but I can affirm he is an expert dollar is generally bearish for gold and
practitioner. silver while a weak dollar is bullish.
Trends in precious metals can sometimes
The chart on the previous page is an depend more on currency dynamics than
illustration. You can see that trends in fundamentals unique to that sector.
gold are echoed in short-term interest
rates about 15 months later. Based in part A rally when gold is priced in dollars
on this association Tom correctly forecast may appear as a sideways trading range or
the recent surge in T-bill rates. even a pullback when gold is priced in
another currency. To factor out the
Of course, such intermarket align- effect of exchange rate flux, Tom multi-
ments don't always hold up. Sometimes plies the price of gold by the U.S. Dollar
extraneous events can confound the Index. The result is what he calls the
projection. In other cases the market Goldollar Index, a reflection of the trend
rhythm abruptly fades out, which is why in gold isolated from movement in the
Tom relies on confirmation from multiple dollar. On a long-term basis, Tom has
indicators. found that trends in gold are often
foreshadowed months in advance by the
The chart above shows another one Goldollar Index.
of Tom's interesting patterns, the ten-
dency of inflation to lead unemployment
 Volume VII, No. 2 F O R M U L A R E S E A R C HTM
Quantitative Treatment of the Financial Markets
The Goldollar Index is useful in
short-term trading as well. The chart
above compares cash gold with the
Goldollar Index on a daily basis. You
can see that divergences between the two
almost always lead to a tradable move in
the direction indicated by the Goldollar
Woops...after I index.
prepared the chart I
realized that "Goldollar"
is spelled with just one I came up with a system for trading
d. gold futures that exploits the leading
properties of the Goldollar Index. The
rules are simple. You go long when
Comex gold is below its 3-day simple
The PMI, published
monthly by the Institute moving average (SMA) and the Goldollar
for Supply Management, Index is above its 3-day SMA. You go
typically ranges between
45 and 60 (75% of the short when gold is above its 3-day SMA
data since 1948). A and the Goldollar Index is below its 3-day
reading above 50
smoothing. Enter on the next day's open.
suggests the economy
is expanding while a If long, exit on a stop at the lowest low of
reading below 50 the past 15 days. If short, exit on a stop
suggests economic
contraction. at the highest high of the past 15 days.
Since 1975 this system grossed
$161,540 per contract trading gold futures.
Drawdown was $15,180. With Comex
silver, profits soared to $317,959 per
contract with $23,000 drawdown.
The results exclude slippage and
commissions, which would dampen
performance. But the sizable gains and
manageable drawdown underscore the
Page 4
potential of the Goldollar index in a
range of applications. z
Forecasting Short-Term
Interest Rates
As we saw earlier, gold
prices tend to lead T-bill rates by
about 15 months. One of Tom's
companion tools for predicting
short-term interest rates is keyed to
the Purchasing Managers Index (PMI).
Tom notes that almost all initial Fed rate
hikes come after the PMI climbs above
56. A drop in the PMI below 50 usually
leads to an easing move by the Fed.
I tested this pattern back to 1950.
Tom would never use a simple formula
like this in isolation, but the theoretical
results are instructive. In our test the
benchmark for short-term interest rates is
the yield on 90-day commercial paper.
We "go long" interest rates when the PMI
rises above 56. Our expectation is that
commercial paper yields will rise.
Stay bullish on yields until the PMI
drops below 56, at which point our model
becomes neutral. When the PMI falls
below 50, we "sell short." The expecta-
tion is that commercial paper rates will
decline. Remain bearish on yields until
the PMI climbs back above 50, when we
return to a neutral stance.
In January 1950 the yield on 3-month
commercial paper was 1.31%. Today the
yield is 1.19%. The net change in rates
after 53 years is a decline of 12 basis
points. Contrast that with the startling
results from our PMI model. If you
"traded" commercial paper according to
the rules outlined above, you would have
Volume VII, No. 2 F O R M U L A R E S E A R C HTM Page 5
Quantitative Treatment of the Financial Markets

gained a phenomenal 6,398

basis points. Forty-nine of 58
market calls were profitable
long and short, a batting
average of 84%.

Obviously, the PMI indica-

tor contains useful information.
The findings could be the
nucleus of a promising
Eurodollar trading system
among other applications.

To an economist or a
fundamental analyst more
familiar with government statis-
tics, the stunning performance
edge might not seem so aston-
ishing. But speaking for myself,
I would never have known
about this improbably accurate
forecasting tool but for the
depth of Tom's research.

By the way, in experiment-

ing with the PMI I stumbled on
an even simpler use of the data.
Signals are given when the PMI
merely crosses above and below the 50
level. On a rise above 50, look for higher
commercial paper yields. On a drop
below 50, look for lower rates. Applying
this model would have produced 7,311
basis points since 1950 compared to 12
basis points for "buy and hold."
The two charts at right show all the
signals generated by this simple method
since 1964. The formula captures small
and large trends, often identifying turning
points in interest rates with remarkable
accuracy. z
Sector Strength and
Market Returns: I
The broad stock market is more likely
to advance when the NASDAQ Compos-
ite is the strongest of the major market
averages. To my mind it was Gerald
Appel who first pointed out this
phenomenon, but many analysts have
cited the tendency. In his treatment of
the pattern, Tom takes the ratio of the
NASDAQ to the NYSE Composite. He
then compares this ratio to its 44-day
exponential moving average. When the
ratio is above its smoothing, the broad
market as represented by the S&P 500
tends to rally. When the OTC ratio is
below its smoothing, weakness looms.
 Volume VII, No. 2 F O R M U L A R E S E A R C HTM
Quantitative Treatment of the Financial Markets
I tested this relative strength observa-
tion back to 1971, as far as OTC data
Actually, it was allow. Results were convincing. When
convenient to substitute the NASDAQ index was the relative
the S&P 500 for the
NYSE index in the
strength leader, the S&P 500 produced an
calculations. The two annualized return of 14.6%. By contrast,
are highly correlated. when the OTC market lagged in relative
strength, the annualized gain fell by more
than half to 7.1%. The S&P 500 itself
returned 11.1% a year on a buy-and-hold
basis.
Tom cites a variation on the theme
that I have not seen reported elsewhere.
He notes that a similar pattern applies to
the OEX index. When the S&P 100 is
stronger than the S&P 500, stocks in
general tend to advance. When the OEX
lags in relative strength, the broad market
suffers.
I tested this idea since 1976, when
OEX data begin. I used the same relative
strength formula cited above, this time
substituting the S&P 100 for the
NASDAQ. Again results were positive.
When the OEX was dominant the S&P
500 returned an annualized 15.1%. When
the OEX lagged in relative strength, the
annualized gain dropped to 10.9%. The
S&P 500 itself returned 12.5% a year since
1976.
I wondered what would happen when
both the NASDAQ and the OEX give
For information on
similar signals, either positive or negative.
Global Investment
Solutions call Roger As you might expect, the contrast
Kliminski at (800)
between bullish and bearish performance
440-7283 or (949)
660-7960. If you prefer is even more striking. When both sectors
to write, the address is
1300 Bristol Street
are positive, the S&P returns an annual-
North, Suite 208, ized 16.2%. When both sectors lag in
Newport Beach, CA relative strength, the annualized return
92660. The fax number
is (949) 660-7945. drops to 7.3%.
The lesson is simple. When the OEX
and NASDAQ are dominant, you want to
Page 6
be quick to buy and slow to sell. When
both sectors are weak, you want to be
slow to buy and quick to sell. z
Sector Strength and
Market Returns: II
It was Roger Kliminski who originally
identified our next price pattern. Tom
later refined the observations. The impli-
cations of these findings are among the
most riveting market tendencies I know
of. We'll fully explore a variety of applica-
tions here and in Part II of this study, but
first some background.
Roger Kliminski has been a successful
money manager for almost two decades.
When I first came to know him several
years ago, Roger was working with his
colleague and close friend, Peter Mauthe.
In 1997 an opportunity opened up and
Peter left to manage the commodity
trading funds of Market Wizard Tom
Basso.
Soon Roger forged a new alliance
with Tom. For both it was the start of a
rewarding association. Their money
management firm, Global Investment
Solutions, benefits from a distinct synergy,
reflecting the like-minded but independ-
ent perspectives of two veteran analysts.
Since Roger and Tom joined forces,
Global Investment Solutions has posted
superb returns for its money management
clients.
Now let's discuss the distinctive
market pattern that Roger and Tom
uncovered. As most of us know, the
Russell 2000 is a popular index of small-
cap stocks while the Russell 1000 covers
large-caps. The Frank Russell Company in
 Volume VII, No. 2 F O R M U L A R E S E A R C HTM Page 7
Quantitative Treatment of the Financial Markets

turn divides each parent index into two
sub-groups, Growth and Value. The
upshot is an array of four sectors: Russell
1000 Growth, Russell 1000 Value, Russell
2000 Growth and Russell 2000 Value.
The relative strength of each sector
has far-reaching implications for price
behavior. Roger and Tom have worked
out a specific technique to assess this
sector strength and exploit it in trading.
Since their winning track
record depends in part on a $100,000
Some interesting findings emerge
once you identify the dominant sector in
this fashion. Suppose you simply switch
among the four Russell segments accord-
ing to which index currently ranks highest.
Since 1995 this method would have
returned 22.5% compounded annually.
By contrast, the S&P 500 gained just 9.2%
a year over the same period. Note that all
comparisons to the S&P 500 refer to total
return, with dividends reinvested. The
Timing Four Russell Sectors vs.
S&P 500 Buy-and-Hold: Growth of $10,000

proprietary calculation, we will Russell Timing:

work with an alternate specifi- $56,000

$10,000 Grows to
$48,300

cation here. Don't worry, our

proxy formula offers outstand-
ing results.
$32,000

Before describing the strat- $18,000

egy, be aware that Tom and

S&P Total Return:
$10,000 Grows to
$20,300

Roger do not apply their $10,000

May-97

May-98

Apr-99
Jan-96

Jun-96

Mar-00

Mar-01

Feb-02

Feb-03
Jul-95

Dec-96

Nov-97

Sep-00

Aug-01

Aug-02

Jul-03
Oct-98

Oct-99
methods mechanically. The
managers can and do overrule
the signals based on experience
and judgment. For this reason, and chart below shows the comparative equity
because our treatment varies from the curves.
original, the present exercise may be one Breaking down the price data into
of those cases where real-world returns four discrete segments yields surprising
The Frank Russell
website posts the surpass simulated results, especially in the new insights, information unavailable in
necessary data back to
area of risk control. the parent indices. Suppose you applied a
1995. Go to
www.russell.com and
similar strategy to the Russell 1000 and the
follow the links. Here's how to implement our version Russell 2000. Switching only between
You will have an option of the strategy. We start with daily closing these larger aggregates, the annual gain
to download nominal prices of the four Russell sub-indices. drops to just 11.8%, not much better than
price values, which are
Now compute the percentage change in the S&P's return of 9.2%.
not adjusted for
dividends, or total return each Russell sub-group over four different
values, which are. I
time frames. In this case we calculate the Roger and Tom's sector research has
used total return. Later I
found that Roger and percent gain or loss over the preceding 5, enormous potential, and we'll present
Tom use unadjusted
15, 25, and 35 days. Next, average the other compelling findings in Part II. First
prices.
results for all time frames into a single we have to address three constraints
The practical difference
reading. The Russell sector with the forthrightly.
is minimal. Perform-
ance results are equally highest composite score is deemed to be
strong. Among the four the relative strength leader. Item number one is the question of
market segments, only risk. This switching strategy offers
the Russell 1000 Value
index would be much extraordinary gains but at a cost.
affected by dividends in
any case.
Volume VII, No. 2 F O R M U L A R E S E A R C HTM
Quantitative Treatment of the Financial Markets
Maximum drawdown was a steep 33%.
This beats the 48% drawdown suffered by
the S&P 500, but volatility is high.
One solution to risk control is to use
discretion and be selective about which
trades to take. This is Tom and Roger's
practice. For these savvy professionals,
personal judgment works well. Their
managed accounts show far lower levels
of risk than that cited above. For the rest
of us, we'll take up more systematic
solutions to risk management in the
second part of our study.
A second concern pertains to issues
of real-world execution. How can you
trade an abstract basket of stocks like the
four Russell sectors? As it happens, there
are exchange traded funds (ETFs) for
each of the four Russell components.
The funds are part of the Barclays iShares
family, traded on the Amex. The
downside is that some of the Russell
ETFs are thinly traded. Bid-ask spreads
can be punishing. Due to these
constraints, Tom and Roger have at times
had to scramble to find surrogate trading
instruments to implement their timing
strategies.
Liquidity issues are apt to affect
money managers more than individual
investors. At least for larger accounts,
here's a tactic that may be worth explor-
ing. The Russell indices are weighted by
market capitalization. With cap-weighted
averages you can often identify a handful
of stocks which have a dominant influ-
ence on performance. You could then go
to the marketplace and trade the individ-
ual shares of this representative but
manageable sample.
NOTE: Hypothetical testing such as that reported here is not as accurate and dependable a measure of profitability as
actual trading results. Even if simulated historical testing were completely reliable, which is not the case, past levels of
performance cannot be assumed to prevail in the future. It is not our intention to state, suggest or imply that any
technique or treatment found in FORMULA RESEARCH can guarantee profitable investment results. Trading should be
undertaken only by those well aware of the many risks.
Page 8
We'll work around liquidity concerns
in the next installment by shifting the
focus to the S&P 500, an index traded in
ample depth. Frankly, no ready remedy is
available for a third acknowledged
weakness in our treatment. You may ask
why historical testing for this study goes
back only to 1995. Normally our analysis
reaches back decades.
The fact is, there is very limited price
history available for the four Russell
sectors. I checked the best-known
commercial data vendors plus numerous
Internet sites. No source I could find
offers a fuller archive than the Frank
Russell website, where prices are posted
back to 1995. If you want Russell sector
data in greater historical depth, apparently
the only solution is to build the price
history yourself.
Perhaps the news isn't so bad after all.
While our data sample is limited in scope,
it spans a desirably wide range of price
behavior. The observations include a
historic bull market, a monumental bear
market, lengthy trading ranges and
assorted airpockets, blow-offs and
meltdowns. In other words, our database
may lack historical reach but it is reassur-
ingly rich in contrast and texture.
Our next report will look even deeper
into Russell sector performance. We will
develop a family of timing models that
offer a range of investment returns--from
conservative to aggressive--while still
containing risk. For now, we are very
grateful to Tom and Roger for sharing
their discerning market perspectives.
 FORMULA RESEARCH TM

Quantitative Treatment of the Financial Markets

Volume VII, No. 3 December 24, 2003 Nelson F. Freeburg, Editor

Notes & Comments T HE M ONITOR S ERIES : L EADING M ARKET P ROFESSIONALS S HAR E U NCOMMON I NSIGHTS

Synergy and Collaboration: The Asset Management

{ Part III Enclosed! Team of Tom McClellan and Roger Kliminski, Part II
Well, it happened again. Initially this study was
The Unique Sector Work of Two Discerning Market Analysts
to take the form of one issue devoted to the
masterly technical work of Tom McClellan and
Roger Kliminski. As you know, the research
proved to be more extensive than a single
report could accommodate. So I split the study
into two installments.
W hen I said last time that you would receive the concluding part of
our study in a couple of weeks, I really meant two or three weeks. I
didn't mean two or three months. When no follow-up report material-
Well, a similar adjustment proved necessary in
the course of preparing Part II. An opportunity ized by mid-October, some of our newer subscribers could only question
to work with newly available data led to unfore- the state of the mails. Longtime subscribers knew better. It was obvious
seen findings and new applications. So I have
divided the remaining section of our study into
the punctually-impaired editor got absorbed in other tasks.
two further installments. Both new sections are
complete and are included in this mailing.
One unexpected detour took the form of an impromptu speaking
engagement in San Francisco, a welcome opportunity to address Dr.
{ Annual Holiday Sale Hank Pruden's graduate finance class at Golden Gate University. The
trip proved rewarding, but I admit I spent more time exploring San Fran-
Here's a chance to acquire our back issues
and spreadsheet models at favorable prices.
cisco's restaurant life than sharing insights into price behavior. Still, the
Our annual sale will last through January 31, occasion will be fondly remembered, especially as I got to share the
2004. The price for back issues will be
reduced by $5 per report. For spreadsheets,
podium for the first time in too many years with my friend Linda
the savings are $10 per diskette. Pricing Raschke.
details and order forms are enclosed. We
will gladly send you a digest which describes There was another reason for the delay in getting out the report, a
all of our models and spreadsheets. You can
receive it by email or regular mail. Email:
development more pertinent to our present inquiry. You may remember
sigma20@midsouth.rr.com. Or phone us at the findings cited at the close of Part I. Building on the work of Roger
800 720-1080 or 901 756-8607.
(Continued on Page Two)
Happy Holidays to All!

The K-Ratio Gold Timing Model--

October 1994
The K-Ratio is so named because market
analyst and commodity fund manager Jay
Kaeppel invented the indicator, which
compares the price of gold stocks to the
price of gold bullion. The K-Ratio can be
used to trade gold and silver (cash and
futures) as well as precious metals mutual
funds.
In our model, we calculate two sets of
adaptive bands around the K-Ratio. A
countertrend channel helps pick tops and
bottoms, entering on strong evidence of a
reversal. A trend-following channel kicks in
on signs of market impulse.
Trading gold since 1976, the K-Ratio model
returned 17.7% a year. Drawdown was
under 15%. Gold itself returned 3.84%
annually since 1976 with 70% drawdown.

Copyright 2003, All Rights Reserved, FORMULA RESEARCH, Inc., 4646 Poplar Ave, Suite 401, Memphis, TN 38117,
A twelve-issue subscription is $295. A six-issue trial is $175. Overseas surcharge: 20%. (800) 720-1080 or (901) 756-8607. 9711
 TM

Volume VII, No. 3 FORMULA RESEARCH Page 2

Quantitative Treatment of the Financial Markets

Kliminski and Tom McClellan, we mix of price shocks, trading ranges and
Last time I told you showed that systematic switching among other diverse price behavior.
Tom McClellan edits
one the finest finan-
four Russell stock sectors offered striking
cial newsletters avail- gains. Specifically, by rotating among the Then something unexpected
able, but I failed to four Russell sub-groups--1000 Growth, happened. Shortly after publishing Part I,
furnish contact infor-
mation. 1000 Value, 2000 Growth and 2000 I got an email from Patrick Cunningham,
Value--we achieved annual returns of 22% a longtime friend of FORMULA RESEARCH.
Tom and his father
Sherman publish the since 1995. Patrick is a portfolio analyst at Gardner-
McClellan Market Lewis Asset Management in Westchester,
Report, a twice-
monthly advisory
As you may recall, there were three PA, a respected institutional investment
publication. Tom problems with the switching strategy. advisor.
also edits the First was the question of risk. Maximum
companion Daily
Edition for short-term equity drawdown was high at 33%. We Patrick advised me that he had access
updates. promised to introduce new risk-control to additional Russell price data. The
Visit the McClellan tactics in Part II of the study. You'll read history consists of two extra years of daily
web site at www. about our proposed solution in what is Russell index data going back to 1993 and
mcoscillator. com.
Phone numbers are
now Part III of the study. over 20 years of monthly Russell data
(800) 872-3737 or going back to 1979. The data came from
(253) 581-4889. The A second problem pertained to issues Bloomberg.
fax number is (253)
584-8194. of trade execution. There are Exchange
Traded Funds available for each of the I jumped at the chance to test my
The postal address is
McClellan Financial four Russell sectors, part of the Barclays already developed findings on unseen
Publications, Inc., iShares family listed on the Amex. But data. This would be the ultimate exercise
P.O. Box 39779
Lakewood, WA
some of these ETFs are thinly traded, in out-of-sample testing. The only
98439. with crippling bid-ask spreads. To problem is that I had already prepared
If you are interested
address liquidity issues, we proposed to most of Part II including myriad test
in the investment shift the focus of trading to the S&P 500. results, charts and the lion's share of the
services of Global Again, you'll see our suggested treatment text. All of that would now have to be
Investment Solutions,
the partners' money in Part III. redone. The readjustment is a second
management arm, reason for the delay in getting the report
call Roger Kliminski
at (800) 440-7283 or
A third problem appeared to be the out.
(949) 660-7960. The most challenging of all. Our testing of
fax number is (949)
660-7945.
the Russell switching strategy was unusu- Because of the new perspective intro-
ally restricted in scope, reaching back only duced by the expanded data, especially the
The mailing address to 1995. After checking many data monthly Russell price history, I present
is 1300 Bristol Street
North, Suite 208, sources, I simply could not find Russell the results of this study not in the
Newport Beach, CA price history in greater depth. sequence I developed them, but in what I
92660.
hope is a fitting logical progression.
As a result, our findings were
supported by an uncomfortably narrow Meanwhile, you're probably wonder-
range of data, probably the most finite ing about how our prior findings held up
data sample of any of our published when tested on the new data. Results
studies. Fortunately, that price history were broadly consistent with previous
was reassuringly broad, encompassing findings. While I did adjust a minor point
historic bull and bear markets and a rich (cited in a footnote in Part III), the new
data actually boost confidence in our key

Volume VII, No. 3 FORMULA RESEARCH
Quantitative Treatment of the Financial Markets
timing strategies. Now, let's discuss those
trading methods. z
The Four-Sector
Switching Strategy
It was Roger Kliminski who initially
developed the Russell switching strategy
cited above. Later Tom McClellan
The Frank Russell
website posts the enhanced and refined the logic. The
necessary data back to switching rules we describe here differ
1995. Go to www.
russell.com and follow
slightly from the original. Rest assured,
the links. You'll have a our alternative treatment offers outstand-
choice between using
total return values (divi-
ing returns.
dends reinvested) or
unadjusted values. I Let's review our version of the strat-
used total return data
throughout this study. egy and update performance with the new
data. We start with daily closing prices of
the four Russell sectors. Now compute
the percentage change in each Russell
The slight pullback
can be attributed to
sub-group over four distinct time frames.
lower volatility in the
1993-1995 years along
with the sheer nature of
In this case we calculate the percent
compounding. Due to gain or loss over the preceding 5, 15, 25,
the exponential dynam- and 35 days. Next, average the results for
ics, as you test further
back in time, it becomes all time frames into a single reading. The
progressively harder to Russell sector with the highest composite
sustain high rates of
return. score is deemed to be the relative strength
leader. You invest in that sector and
remain there until another Russell index
claims the number one spot.
Let's illustrate the mechanics with
some calculations from a recent trading
session. Last December 9, Russell 1000
Value closed at 487.45. This is -0.16%
below its close of five days earlier (2-Dec);
+2.53% above its close of 15 days earlier
(17-Nov); +1.36% above its close of 25
days earlier (3-Nov); and +3.30% above
its close of 35 days earlier (20-Oct). Add
up each of these figures and divide by
four to get a simple average, in this case
1.76%. That score proved to be the
TM
Page 3
highest reading for this session, prompt-
ing a switch into Russell 1000 Value.
We originally tested this switching
method over a roughly eight-year period,
from June 1995 to August 2003. The
strategy produced an annual return of
22.5%. Now we make two changes in the
data. We extend testing back another two
years to June 1993. And we bring results
forward to mid-December 2003. Over
this expanded 10.5-year period our annual
return was 20.1%, a tick below earlier
levels but still strong.
Contrast this 20%-plus return with
the performance of key benchmarks.
Since mid-1993 the Russell 1000 Growth
index gained 9.1% a year, 1000 Value
gained 11.3% a year, 2000 Growth gained
6.0% a year, and 2000 Value gained 12.7%
a year. For its part, the S&P 500 returned
10.7% since 1993.
Not only was the switching strategy
more profitable, it was less risky.
Maximum drawdown was 33%, high in
absolute terms. But this compares with
66% drawdown for Russell 1000 Growth
and 68% drawdown for Russell 2000
Growth. The other three benchmarks
also exhibited higher risk, though the gap
was not as dramatic.
The chart on the next page shows the
comparative equity curves for the switch-
ing strategy and the S&P 500. It is an
updated version of a similar chart seen at
the end of the last report. The shaded
area reflects the new Russell data incorpo-
rated into the analysis since Part I was
published.
You can see that the two equity
curves are very similar until about the year
2000, when the bear market struck. At
 TM

Volume VII, No. 3 FORMULA RESEARCH Page 4

Quantitative Treatment of the Financial Markets

Russe ll Switc h M o d e l vs. S&P 500:

G ro w t h o f $ 1 0 , 0 0 0
components. Russell 1000 Growth
$80,000 returned 11.7% a year since 1980,
Switching $70,000 1000 Value returned 13.9%, 2000
S&P 500 $60,000 Growth returned 8.5% and 2000
$50,000
$40,000
Value returned 14.7%.
$30,000
$20,000 The S&P 500 itself
$10,000 returned 13.2% annually over the
$0
same time frame. At this rate of
Jul-93

Jul-94

Jul-95

Jul-96

Jul-97

Jul-98

Jul-99

Jul-00

Jul-01

Jul-02

Jul-03
return, an initial $10,000 stake would
now be worth $191,000, less than
that point our switching model starts to half the dollar gains from our switch-
pull away convincingly. Eventually the ing model.
strategy would produce more than double
the dollar gains of the S&P 500. Patrick's variant beat all benchmarks,
despite key differences in the historical
I am the first to admit that this reach and very structure of the data. This
performance edge is based on a rather is reassuring confirmation of our switch-
skimpy stream of data. It would be better ing approach. z
if we had access to Russell price history
going back decades. Actually, in a way we
do. Thanks to Patrick Cunningham, we The Two-Sector
have monthly Russell index data going Switching Strategy
back to 1979. Suppose we adapt a similar
switching strategy for testing on monthly
Now we return to the work of Roger
data. If results were clearly positive it
and Tom, again using daily data from
would reinforce confidence in the earlier
1993 to 2003. This perceptive team went
findings.
on to discover an even more compelling
tendency in the Russell sector data. They
Patrick himself came up with a simple
found that two of the Russell sectors have
monthly variant on our switching model.
special forecasting significance--Russell
Here's how his formula works. For each
2000 Growth and 2000 Value. By
of the four Russell sectors, track the
restricting the analytical universe to these
percentage change over three periods of
two exceptional sectors, powerful new
time. In this case calculate the 3-month,
findings emerge.
6-month and 12-month rates of change.
Then, much as before, you average the
I have no idea how Roger and Tom
readings into a composite. Switch to the
came up with this offbeat but riveting
Russell sector with the top combined
insight, so let me just pass on what they
score on a month-end basis.
uncovered. In this case you track only
Russell 2000 Growth and 2000 Value.
Patrick's monthly model gained an
Ignore the other two Russell sectors. As
impressive 16.8% annually from 1980 to
before, the aim is to pick the relative
2003. A $10,000 investment would now
strength leader. Use the same ranking for-
be worth $408,000. Compare that to the
mula cited earlier. Calculate 5-, 15-, 25-
buy-and-hold returns of the four Russell

Volume VII, No. 3 FORMULA RESEARCH
Quantitative Treatment of the Financial Markets
and 35-day rates-of-change for Russell
2000 Growth and 2000 Value. Average
the results. The sector with the highest
composite score has the edge.
What Roger and Tom found out is
that when Russell 2000 Growth is the
relative strength leader, the entire stock
market gets a lift. By contrast, when
Russell 2000 Value is dominant, profit
suffers and risk increases. Nor are we
talking about subtle differences in
performance.
For example, when Russell 2000
Growth is the relative strength leader, the
S&P 500 appreciates at a 17.4% annual-
ized rate. On the other hand, when 2000
Value is dominant, the annualized gain
drops to 4.5%.
A similar stark contrast is seen in the
area of risk. When Russell 2000 Growth
is dominant, maximum drawdown for the
S&P 500 is 21%. When Russell 2000
Value takes the lead, drawdown more
than doubles to 50%. The S&P 500 itself
returned 10.7% a year since 1993 with
47% drawdown.
The table below offers another per-
spective. The table breaks down perfor-
mance of the four Russell sectors
according to current leadership, Russell
2000 Growth versus 2000 Value.
Consider the left-most column, "R1K
Annualized Returns and Drawdown: 1995 - 2003
Relative Strength Leader
R2K Growth R2K Value
ARR % DD ARR %
R1K Growth 20.6% 25% -1.3%
R1K Value 15.1% 22% 7.2%
R2K Growth 40.4% 20% -20.4%
R2K Value 24.3% 17% 1.7%
TM
Page 5
Growth." The numbers show how
Russell 1000 Growth fared depending on
whether Russell 2000 Growth or 2000
Value was the relative strength leader.
"ARR %" stands for annualized rate of
return while "DD" stands for maximum
equity drawdown.
To illustrate, when Russell 2000
Growth was the relative strength leader,
1000 Growth returned an annualized
20.6%. Drawdown was 25%. By
contrast, when Russell 2000 Value had the
edge, 1000 Growth suffered an annualized
loss of -1.3%. Drawdown soared to 65%.
You can see similar contrasting results
across all four sectors.
If you study the table, you'll note a
developing pattern we'll see again later on.
When Russell 2000 Growth is dominant,
virtually every area of the stock market
shows higher returns and lower risk.
When 2000 Value is dominant, those
same sectors exhibit lower returns and
higher risk.
Perhaps the most telling statistic
appears in the last row of the table. This
entry shows performance of the Russell
2000 Value index. Paradoxically, Russell
2000 Value is far more profitable and risk-
averse when 2000 Growth is dominant.
When Russell 2000 Growth has the
edge, 2000 Value appreciates at an annual-
ized rate of 24.3%. Drawdown is 17%.
When Russell 2000 Value is dominant, the
annual return drops to 1.7%. Drawdown
surges to 44%. Russell 2000 Value is
curiously contrarian. It performs much
better when its own antithesis in sector
DD composition is dominant in relative
65% strength.
40% The performance gap is magnified
78% when we turn to high-beta stocks. Con-
44% sider the volatile Nasdaq Composite.
 TM

Volume VII, No. 3 FORMULA RESEARCH Page 6

Quantitative Treatment of the Financial Markets

When Russell 2000 Growth is leading the is favored. But those funds would have
market, OTC stocks climb at an annual- been limited in price history, idiosyncratic
ized rate of 41.9%. Drawdown is 22%. in style or otherwise unrepresentative.
By contrast, when Russell 2000 Value is Without a doubt, the average Fidelity
dominant, the annualized return drops to Select fund shows higher gains and lower
-14.0%. Drawdown explodes to 80%. risk when Russell 2000 Growth is the
The disparity is all the more striking as relative strength leader.
the two sectors have shared the top rating
for relative strength almost equal 15 Fidelity Sector Funds:
amounts of time since 1993. Returns and Drawdown, 1993 - 2003
Relative Strength Leader
I was so impressed by Roger R2K Growth R2K Value
and Tom's findings I undertook a ARR % DD ARR% DD
challenging test. I looked at 15 Air Transport 31.9% 22% -5.0% 63%
Fidelity sector funds representing Auto 18.2% 29% -3.7% 44%
a diverse mix of industry groups. Biotech 32.4% 38% -6.2% 72%
Much as before, I compared risk Brokerage 41.4% 30% -4.3% 66%
and return according to which Cyclical 21.8% 22% -6.7% 47%
sector was dominant, Russell Devel. Commun. 55.3% 28% -22.6% 86%
2000 Growth or 2000 Value. The Electronics 46.6% 32% -2.4% 83%
results appear in the table at right. Energy Service 29.9% 32% -6.1% 56%
Be warned, the findings compress Indust. Equip. 25.0% 22% -3.1% 54%
a lot of numeric data into a small Leisure 35.4% 23% -6.7% 54%
space. Multimedia 36.8% 18% -5.6% 57%
Technology 46.2% 26% -10.5% 85%
Let me cut to the key point. Software 39.8% 22% -3.6% 73%
When Russell 2000 Growth is Telecom 38.4% 27% -21.0% 82%
dominant, all 15 sector funds are Util. Growth 16.2% 24% -4.3% 63%
profitable. When Russell 2000
Average 34.4% 26% -7.5% 66%
Value is dominant, all 15 sectors
show losses. You can see the As final proof, I built a composite
average results for both scenarios in the price index that tracks as a group the 34
bottom row. Fidelity sector funds that have been active
since 1993. When Russell 2000 Growth
When Russell 2000 Growth is was the market leader, this portfolio
dominant, the average annualized gain is posted annual gains of 21.4%.
34.4%. Average drawdown is 26%. On Drawdown was moderate at 16%. But
the other hand, when Russell 2000 Value when Russell 2000 Value was dominant,
is dominant, the 15 sectors show annual- the annualized gain dropped to 4.7%.
ized losses of -7.5%. Average drawdown Drawdown was a punishing 42%.
climbs to 66%.
We will build on these findings in the
In truth, I prepared this table with a final installment of our study, which will
view to dramatic effect. I might have feature two high-performance switching
found one or two counterexamples which strategies, one aggressive, the other
perform better when Russell 2000 Value balanced with manageable risk.
 FORMULA RESEARCH
Quantitative Treatment of the Financial Markets
Volume VII, No. 4
Notes & Comments
{ Part III Enclosed [Cont.]
This is the third and final installment in our
series devoted to the technical work of Tom
McClellan and Roger Kliminski. In this
concluding section we present two timing
models based on the unique research of this
talented pair. Speaking for all of us, let me
express our gratitude to Roger and Tom for
sharing their ground-breaking findings.
{ Custom FORMULA RESEARCH Timing
Models for Investment Managers
Over the past several years we have
prepared proprietary stock market timing
models for money managers and other
investment professionals. The great
advantage of these composite models is
that they combine the individual signals
from a variety of components into a struc-
tured whole. The consensus of indicators
includes a diverse mix of data-- market
breadth, sentiment, monetary and trend-
sensitive inputs. All of our custom models
have outperformed the S&P 500 in real
time in terms of risk and reward.
We recently developed one such model for
a client in Hong Kong. In testing back to
1970, the model produced four times the
dollar gain of the S&P 500 with one-third
the risk. For information on our proprietary
research, please email me:
sigma20@midsouth.rr.com.
Copyright 2004, All Rights Reserved, FORMULA RESEARCH, Inc., 4646 Poplar Ave, Suite 401, Memphis, TN 38117,
A twelve-issue subscription is $295. A six-issue trial is $175. Overseas surcharge: 20%. (800) 720-1080 or (901) 756-8607.
TM
December 30, 2003 Nelson F. Freeburg, Editor
T HE M ONITOR S ERIES : L EADING M ARKET P ROFESSIONALS S HAR E U NCOMMON I NSIGHTS
Synergy and Collaboration: The Asset Management
Team of Tom McClellan and Roger Kliminski, Part III
Drawing on their Insights, We Build Two High-Performance Timing Models
L et's take stock of our Russell sector work so far. We started by
analyzing four sub-indices--Russell 1000 Growth, 1000 Value, 2000
Growth and 2000 Value. We found that rotation among these sectors
consistently outperformed any individual Russell index on a buy-and-hold
basis. The same strategy also beat the S&P 500, offering both higher
returns and lower risk. Our switching method performed well on daily
data since 1993 and, suitably modified, on a monthly basis back to 1980.
Next we explored one of the most fascinating market patterns I have
ever seen. Roger and Tom found that two Russell sectors--2000 Growth
and 2000 Value--are especially rich in forecasting significance. Following
their lead, we confined analysis to the two small-cap sectors. With re-
markable consistency, the entire stock market performed very differently
depending on which Russell sub-index was leading in relative strength.
Now let's pick up where we left off. Recall our composite index of
34 Fidelity sector funds. That index tracks the collective performance of
all Select funds in operation as far back as 1993. We found that when
(Continued on Page Two)
The Hendrickson Stock Market
Timing Model--January 1995
Harland Hendrickson presides over a thriv-
ing portfolio management firm. Ten years
ago he generously allowed us to feature his
simple but effective method for timing
stocks. You can see the trading signals
since early 1995 have been quite accurate.
Since 1970 the Hendrickson model has
returned 13.0% a year compared to 11.1%
for the S&P 500. Drawdown was 13.3%
versus 46% for the S&P. Ten out of ten buy
signals were profitable, including all three
trades since publication.
Our sale of back issues and spreadsheets
lasts through January, 2004. This is an
opportunity to acquire the Hendrickson
report and spreadsheet as well as other
interesting material at an attractive price.
9712
 FORMULA RESEARCH
Volume VII, No. 4
Quantitative Treatment of the Financial Markets
Russell 2000 Growth was dominant in
relative strength, the Fidelity composite
showed high returns and low risk. When
Russell 2000 Value was dominant, the
composite showed low returns and high
risk.
The chart below is an attempt to
graphically capture the performance gap
just cited. The chart shows the Fidelity
composite during a revealing period, the
steep bear market decline from March
2001 to November 2002.
Notice the black dots clustered at
points along the equity curve. The dots
often appear when prices are falling.
That's not a coincidence. These dots
signify periods when the Russell 2000
Value index is leading in relative strength.
Now note the portions of the equity
curve with no dots. These smooth
segments show when Russell 2000 Growth
is leading in relative strength. The graphic
realization is not perfect, but you can
easily see the implications. Black dots
dominate throughout much of this bear
market action. But on the few occasions
when prices rally, it is when 2000 Russell
Growth leads in relative strength. z
TM
Page 2
Switch Fund Strategy I:
The Mekros Method
Before discussing the two timing
models we will feature in this final section
of the study, let me sound a note of
caution. We often report annualized gains
and losses as one benchmark of financial
performance. Annualized returns offer
useful snapshots of price behavior but
they don't tell the full story.
Let me give you a
concrete example.
We noted in Part II
that the S&P 500
posts strong annual-
ized gains of 17.4%
when Russell 2000
Growth leads in
relative strength.
When 2000 Value is
dominant the annual-
ized return drops to
4.5%, a notable fall-
off in performance.
It would seem an easy matter to build
a winning switch fund model based on
these sharply opposing tendencies. You
would simply buy an S&P index fund
when Russell 2000 Growth is favored and
exit to the money market when leadership
shifts to 2000 Value.
But translating this theoretical edge
into actual gains is more of a challenge.
Had you traded such a strategy since
1993, you would earned a modest 10.7% a
year. This is almost identical to the S&P
500 return over that period. To be sure,
drawdown was much lower, 20%
compared to 47% for S&P buy-and-hold.
But our task is to produce higher returns
 FORMULA RESEARCH
Volume VII, No. 4
Quantitative Treatment of the Financial Markets
and lower risk. In this case the profit side
of the ledger fell short.
Here's why. That alluring 17.4%
annualized return is available only about
half the time, namely when Russell 2000
Growth is dominant in relative strength.
Otherwise you would be on the sidelines
earning money market interest. Let me go
deeper.
Our practice in these studies is to peg
money market interest at 90% of the yield
on 90-day commercial paper. By this
measure, hypothetical money market rates
since 1993 ranged from 6.0% to 0.8%,
averaging about 4%. This means that for
half of the test period you would have
been earning only single-digit returns.
It is this fuller, more realistic reckon-
ing that accounts for the lower than
expected gains. We will devote the rest of
this report to developing two timing
strategies I feel have merit both theoreti-
cally and in practical application, taking
into account the real-world impact of
switching.
With that as background, our first
timing model is designed to trade the
Rydex Mekros fund (RYMKX). Mekros
is the oldest and most popular index fund
keyed to the Russell 2000. Note I said the
Russell 2000, not its Growth or Value
sub-groups. One reason I selected
Mekros, apart from its ample liquidity, is
its leverage. Mekros is designed to
magnify daily movement in the Russell
2000 by a factor of 1.5.
As the Mekros fund was introduced
in November 2000, our initial testing
begins there. First, the theoretical
results. When Russell 2000 Growth is the
relative strength leader, Mekros shows an
TM
Page 3
annualized gain of 90.5% (yes). When
Russell 2000 Value is dominant, Mekros
shows an annualized loss of -38.0%. This
is the kind of performance gap that can
serve as the nucleus of an effective switch
fund strategy.
Suppose you purchased the Mekros
fund when Russell 2000 Growth is
dominant and switched to cash when
2000 Value takes the lead. You would
have made an impressive 30.2% a year
since late 2000. Drawdown was a mild
16%. By contrast, the Mekros fund itself
lost -0.4% a year with 54% drawdown.
Over the same period the S&P 500 lost
-7.4% annually with 44% drawdown.
You might wonder what would
happen if we reversed the logic. Here you
go long when Russell 2000 Value is
dominant and switch to cash when 2000
Growth takes the lead. In this case you
would have suffered losses of -20.4% a
year. Drawdown was a punishing 70%.
Another intriguing scenario would
include the option to go short. For this
analysis we'll continue with the Mekros
fund even though it is traded from the
long side only. In the real world there are
bearish index funds that move opposite
the Russell 2000, including some funds
with leverage. We stick with the Mekros
fund because it has more available price
history.
As before, here you buy the Mekros
fund when Russell 2000 Growth leads in
relative strength. But when leadership
switches to Russell 2000 Value, you sell
Mekros short rather than switch to the
money market. I admit this is a theoreti-
cal exercise but consider the results.
Suddenly the return climbs to a remark-
able 47.1% a year since late 2000.
 FORMULA RESEARCH
TM

Volume VII, No. 4 Page 4

Quantitative Treatment of the Financial Markets

is hard to systematically exploit the

Two other variants on our Mekros pullbacks by selling short.
model offer additional insights. First, let's
again reverse the sector logic. You go With their sector work, Roger and
long when Russell 2000 Value is dominant Tom bring new understanding to the
and go short when 2000 Growth is forces that govern strength and weakness
This footnote is not
especially relevant to
favored. In this case you would have lost in the stock market. As a result, short
anything you may have -43.0% a year. selling may become feasible over a wider
just read. I could have
inserted it anywhere in
range of applications.
the text. Or I could Now let's return to the original sector
have omitted it
altogether--except that
preference but focus exclusively on short- Now, getting back, it is true this
I promised earlier to selling. As before, we sell Mekros short short-selling exercise rests on a very
own up to the one when Russell 2000 Value is the relative narrow span of data, just three years. And
adjustment I had to
make after taking into strength leader. But when 2000 Growth much of that period was consumed by a
account the new is dominant, instead of going long, we vicious bear market. If you applied the
Russell sector data
(1993 to 1995). It's a park our money in cash. This way we same strategy to a wider, more representa-
small point, but in the isolate performance on the short side. tive range of data, short-side performance
cause of full disclosure,
here goes. might not be so convincing.
In this case you would have made an
This study has stressed
that the broad market impressive 17.5% a year since November To alleviate this concern and to
gets a lift when Russell 2000. That contrasts with the -0.4% loss satisfy natural curiosity, I simulated
2000 Growth outper-
forms 2000 Value. I of the Mekros fund and the -7.5% loss of performance of the same switching
had previously thought the S&P 500 over the same period. method back to 1993. To do this I
the market performs
best when Russell
constructed a price index that replicates
2000 Growth beats all Admittedly, stocks were generally action of the Russell 2000 index but with
four Russell
sub-groups, including
weak through this period. Perhaps any 150% exposure. In other words, I built a
Russell 1000 Growth short-selling strategy would have proxy Mekros fund.
and 1000 Value.
performed well through that phase of
Without going into liquidation. Then again, there were Then I re-tested the switching strat-
detail, this point is not several furious bear market rallies during egy over this more inclusive data sample.
as conclusive as I once
thought. At times the the period (as the chart on page two We'll test performance under the same
market can post explo- makes clear). We'll follow up on this issue conditions cited above: 1) long only; 2)
sive gains when any of
the Russell sectors is in detail shortly. long and short; and 3) short only.
dominant.
Here's my point with respect to short First, consider the theoretical annual-
What has not changed
is the central finding. selling. If you have ever tried to develop ized returns from 1993. When Russell
For timing purposes, a stock market timing model with the 2000 Growth is the relative strength
the most powerful
perspective comes option to go short, and if you tested that leader, our Mekros proxy fund offers
from comparing Russell model going back many decades, you annualized gains of 44.8%. When 2000
2000 Growth to 2000
Value. The new sector know how hard it is to capitalize on Value is dominant, the result is an annual-
data confirm and periodic corrections in the market. ized loss of -18.4%.
reinforce this key point.

As a student of the market, you know This is a promising start. But those
the sell-offs are coming. But due to the are abstract returns which mathematically
long-term bullish trend in stock prices, it distill price action into one of two
contrasting modes. To assess a real-world
 FORMULA RESEARCH
Volume VII, No. 4
Quantitative Treatment of the Financial Markets
switching strategy you have to factor in
the totality of the data, allowing for the
logistics of switching in and out of the
market.
On this more demanding basis, and
trading from the long side only, our
switching strategy gained 22.8% a year
since 1993. Drawdown was 25%.
Though risk is high, this is a fine showing.
If you had turned the logic on its
head, reversing the roles of 2000 Growth
and Value, you would have lost -7.2% a
year. Drawdown was 77%. Our proxy
Mekros fund itself gained 8.5% annually
since 1993 with 67% drawdown.
Let's look at what happens when the
same model includes a provision to go
short. In this case long entries remain the
same, but exits are treated as signals to go
short. Since 1993 the compound annual
return climbs to 26.8%. Drawdown was
39%.
Compare results when you reverse
the logic, interchanging the two Russell
sectors. Trading long and short since
1993, you would have lost -28.9% a year.
Drawdown surges to a monumental 98%,
a new record high for this observer after
many years of model building. To
recover from a 98% equity dip requires an
Olympian gain of 4,900%.
Now let's consider short-sale
performance in isolation, the scenario that
prompted this simulation in the first
place. Here sell signals are treated as
short entries while buy signals prompt a
switch into cash. Most conventional
stock models would show devastating
losses operating only from the short side.
But our switching method showed a
TM
Page 5
healthy gain of 8.4% a year since 1993.
Drawdown was 33%.
Recall that the S&P 500 gained 10.7%
annually over the same period with 47%
drawdown. Though our results come
exclusively from the short side, they
approach the S&P's gains while entailing
less risk. This is fine countertrend
performance against a pronounced
bullish tendency in the data. z
Switch Fund
Strategy II: S&P 500
As promised, we will close the study
by developing a switch fund strategy that
addresses problems of risk and liquidity
encountered earlier. First some
background. As I have stressed, treating
performance in terms of annualized
returns offers a compressed, somewhat
artificial portrait of price behavior.
Let me give you a final illustration of
these constraints, an example very
relevant to our developing model. In Part
I we noted Tom McClellan's findings with
respect to the Nasdaq Composite. When
the Nasdaq outperforms the S&P 500, it
is bullish for stocks as a whole.
As reported, when the OTC market
leads in relative strength, the S&P posts
annualized gains of 14.6%. When Russell
2000 Value is dominant, the annualized
return drops to 7.1%. We cited similar
findings regarding OEX relative strength.
But those annualized gains portray a
static, hypothetical edge. The picture
changes when we take into account the
real-world impact of switching. On this
basis the OTC strategy returned a modest

Volume VII, No. 4
NOTE: Hypothetical
testing such as that
reported here is not as
accurate and depend-
able a measure of profit-
ability as actual trading
results. Even if simulated
historical testing were
completely reliable,
which is not the case,
past levels of perform-
ance cannot be
assumed to prevail in the
future. It is not our inten-
tion to state, suggest or
imply that any technique
or treatment found in
FORMULA RESEARCH can
guarantee profitable
investment results.
Trading should be under-
taken only by those well
aware of the many risks.
FORMULA RESEARCH
Quantitative Treatment of the Financial Markets
11.8% a year. This is no better than the
S&P 500's return over the same period.
What seemed like a sure way to beat the
market wound up falling short.
Despite the apparent letdown, I was
convinced there was real merit in the
OTC and OEX findings. I wrestled with
the relative strength tendencies for weeks
in an effort to realize their full potential.
That was when I tried combining
three of this study's featured price
patterns into a larger whole. No single
indicator could do the job in isolation.
But when I treated the patterns in interac-
tion, results proved compelling. The
exact trading rules appear below. They
build on three of Tom and Roger's key
findings.
First, take the ratio of the Nasdaq
Composite to the S&P 500 on a daily
basis. When this ratio is above its 50-day
simple moving average, the indicator is
bullish. When the ratio is below its
50-day smoothing, the indicator is bearish.
Repeat with the OEX index. Take
the ratio of the OEX to the S&P 500.
When that ratio is above its 50-day simple
moving average, count the second indica-
tor bullish. When the ratio is below its
50-day smoothing, rate it
bearish.
Finally, compare Russell
2000 Growth to Russell
2000 Value. Use the same
relative strength ranking
formula cited throughout
the study. When Russell
2000 Growth is dominant,
count this third indicator
Sep-93
Sep-94
Sep-95
TM
Page 6
bullish. When 2000 Value leads, rate the
indicator bearish.
When all three components are
bullish, purchase an index fund keyed to
the S&P 500. This benchmark is traded
in ample depth, thereby boosting liquidity.
Good choices include the Amex Spider
(SPY) and the Vanguard Index 500
(VFINX). Meanwhile, when all three
components are bearish, exit to the
money market.
Since mid-1993 this model returned
15.3% compounded annually. That beats
every FORMULA RESEARCH stock market
timing model over the same time frame.
Of 17 long entries, 14 were profitable, an
82% batting average.
As for risk, drawdown was held to a
moderate 11.7%. That also beats all of
our published timing models over the test
period. Meanwhile, the S&P 500 returned
10.7% and suffered 47% drawdown since
mid-1993.
This powerful switching strategy
draws on the unique research of Tom
McClellan and Roger Kliminski. We are
grateful to these gifted analysts and
money managers, who combine scholarly
inquiry and rich practical experience.
Switch Fund M odelvs.S&P 500:
Com parative Equity Curves,1993 -2003
$45,000
$40,000
$35,000
$30,000
$25,000
$20,000
$15,000
$10,000
Sep-96
Sep-97
Sep-98
Sep-99
Sep-00
Sep-01
Sep-02
Sep-03