See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228131626

The Profitability of Style Rotation Strategies in the United Kingdom

Article  in  The Journal of Portfolio Management · May 2005

DOI: 10.3905/jpm.1999.319770

CITATIONS READS
62 589

2 authors, including:

Mario Levis
City, University of London
34 PUBLICATIONS   1,143 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Rights Issues View project

All content following this page was uploaded by Mario Levis on 23 September 2014.

The user has requested enhancement of the downloaded file.

 The Profitability of
Style Rotation Strate gies
in the United Kingdom
Depends on temporal volatility of underlying return spread between styles.

Mario Levis and Manolis Liodakis

erformance differentials between small/large

P and value/growth stocks are not exclusively

characteristics of the U.S. market. Capaul,
Rowley, and Sharpe [1993], Arshanapalli,
Coggin, and Doukas [1998], and Fama and French
[1998] report that value stocks, defined in terms of
book-to-market, have outperformed growth stocks in
different countries over relatively long time periods.
Hawawini and Keim [1999], in their review of interna-
tional evidence, suggest broadly similar results for small-
capitalization stocks.1
Style consistency, however, is not necessarily an
optimal strategy. As the asset mix drift creates a need for
active asset allocation, the apparent style drift creates a
need for style rotation strategies. For the active portfo-
lio manager, rotation among different style segments,
may provide opportunities for portfolio performance
enhancement. Kahn [1996], for example, reports that
most funds do not systematically follow a value or
growth stock orientation, but instead tend to either
MARIO LEVIS is head of the ac- shift between one and the other, or adopt a blend. In
counting and finance department at addition, although half of the equity funds studied
the City University Business School stayed within their target size category, a few moved
in London (EC2Y 8HB). across portfolios of small and large stocks. At the same
time, Indro et al. [1998] report that funds that institut-
MANOLIS LIODAKIS is a Ph.D. ed both a change in their value/growth as well as
student in the accounting and small/large-capitalization stock allocation strategy were
finance department at the City the worst-performing group of actively managed funds.
University Business School in We assess the potential profitability of style rota-
London.. tion strategies based on value/growth and small/large-

FALL 1999 THE JOURNAL OF PORTFOLIO MANAGEMENT 1

cap segments of the market. First, we explore the it relatively easier to pass along price increases in infla-
potential of such strategies in the U.K. during the peri- tionary times. He also shows that the yield curve is pos-
od 1968-1997 assuming both perfect forecasting and a itively related to the relative performance of small over
range of intermediate levels of forecasting ability. Using large stocks.
a Monte Carlo simulation, we assess the average gains Ragsdale, Rao, and Fochtman [1995] argue that
from rotation after adjusting for transaction costs. future earnings growth is the decisive factor in predict-
Second, we develop and test a style rotation model ing the performance of small- versus large-cap stocks.
based on a set of macroeconomic factors selected for One of the factors that is likely to affect earnings is
their ability to predict the direction of the style spread exposure to foreign markets. Thus, large stocks are
at a given month. Our out-of-sample tests provide penalized when the exchange rate is highly volatile.
strong support for small versus large but not value ver- At a more general level, Macedo [1995] main-
sus growth style rotation strategies. tains that the equity risk premium is the strongest dis-
Sharpe [1975] was the first to examine the effi- criminator for future style performance. A high equity
cacy of market timing between cash and equities and to risk premium f avors riskier portfolios. Value stocks are
highlight the potential benefits of such strategies. Jeffrey perceived to be more risky, and so tend to do well
[1984], using the same framework, casts some doubt on when the equity risk premium is high.
the benefits of market timing. He argues that the incre- In short, there are good fundamental reasons and
mental rewards from such strategies do not necessarily considerable empirical evidence to suggest that both
match the associated incremental risks. the size and value spreads are associated with econom-
Kester [1990] expands the scope of market-tim- ic fundamentals.
ing strategies by including small firms, while Case and
Cusimano [1995] apply the same principles on value DATA AND INDEX CONSTRUCTION
and growth indexes. They report some profit enhance-
ment depending on transaction costs and the frequen- Our customized style indexes cover the period
cy of portfolio revision. Unfortunately, however, they 1968 through 1997, and use a total of 3,868 firms that
offer very limited guidance in terms of the level of were listed on the Exchange at some point during the
forecasting ability required and the potential of achiev- period under consideration, ensuring a wide stock
ing this level. coverage.2 To construct the size and value portfolios
Coggin [1998] tests the random walk and the we independently rank companies based on their
long-term memory hypothesis for a number of U.S. end-of-June market value (MV) and book-to-market
equity style indexes and style spreads, and finds evi- ratio (BM). 3
dence in support of the former. He concludes that style We then assign stocks to nine portfolios accord-
indexes cannot be predicted using only the time series ing to criteria as follows. We divide stocks into three
of returns as information variables; rather, forecasts market value and three book-to-market groups.
should be conditioned on outside information, such as Companies whose MV makes up the bottom 10% of
the business cycle and interest rates. the equity capitalization in our sample constitute the
A number of studies have indeed attempted to small-size portfolio, while companies in the top 80% of
link the performance of style portfolios to various the total MV are allocated to the large portfolio. All the
macroeconomic factors. Fama and French [1993] sug- other stocks belong to the midcap segment.
gest that book-to-market and size are proxies for dis- Stocks are also divided into three BM categories.
tress, and that distressed firms may be more sensitive to The top 30% of the companies with the highest B/M
certain business cycle factors. Jensen, Johnson, and are in the value portfolio, the middle 40% the middle
Mercer [1998], for example, find that size and book-to- portfolio, and the bottom 30% the growth portfolio.
market depend on the monetary environment. Nine size-BM portfolios are then created from the
Sorensen and Lazzara [1995] find a positive relationship intersection of the three MV and the three BM groups.
between the growth in industrial production and inter- Companies with low MV and a high BM ratio
est rates and the value/g rowth return spread. constitute the small-value segment, while companies
Anderson [1997] reports that small stocks bene- with low MV and a low BM ratio are in the small-
fit from inflation, perhaps because small companies find growth portfolio. This ranking procedure is repeated

2 THE PROFITABILITY OF STYLE ROTATION STRATEGIES IN THE UNITED KINGDOM FALL 1999
every year (twenty-nine times), and the portfolios’ with the highest P/E ratio (21.01) and small-value the
composition changes annually. Following Fama and portfolio with the highest cash flow yield (0.38).
French [1995], we estimate portfolio returns starting in The last column shows the debt-to-equity ratio.
July of each sample year. We use total simple monthly Note the substantial difference in leverage between size
returns from the London Share Price Database (LSPD) portfolios with the same MB. It is also worth noting
and accounting data from Datastream. 4 that value stocks appear to be more leveraged than their
growth counterparts. The average debt-to-equity ratio
FUNDAMENTALS AND PERFORMANCE for value stocks is 0.24, slightly higher than that of
OF SIZE AND VALUE PORTFOLIOS growth stocks.
Exhibit 2 reports average annual equal- and
For each one of the nine portfolios and for every value-weighted returns for the entire sample period
year in the sample period, we calculate the average (July 1968-June 1997) and for three subsample periods
market value and median market-to-book ratio, divi- (July 1968-June 1978, July 1978-June 1988, and July
dend yield, price-earnings, cash flow yield, and total 1988-June 1997). The average annual return of the
debt-to-equity ratio. We also estimate the values of value index is 23.58% (22.31% on a value-weighted
these ratios for the aggregate small-cap, large-cap, value, basis), over 11 percentage points higher than the returns
and growth portfolios. of the growth index.
In Exhibit 1, we report descriptive statistics for What is even more striking, however, is that this
all portfolios. The first two columns show the market difference seems to persist for all three subperiods and
value and market-to-book ratio for every style portfo- across different size categories. The average annual
lio. Given our portfolio formation procedure, small- value-growth spread (in percentage points) is 12.46
and large-cap portfolios have substantially different MV (15.30) for the first decade; 12.37 (10.39) for the sec-
but the same MB ratios on average, while value and ond; and 9.80 (7.12) for the third, indicating a high
growth portfolios differ significantly on MB but not so persistence in performance for value stocks.
much on their MV. It is interesting that this difference in perfor-
Value portfolios have higher dividend and cash mance is apparent whether we focus on the small-cap,
flow yields and lower price-earnings ratios than growth midcap, or large-cap segment. These results confirm the
portfolios. Size portfolios show no clear differences on findings of previous studies, which document a positive
any of these measures. Large-value stocks appear to have relation between book-to-market and stock returns.
the highest dividend yield (5.99), while middle-growth The average annual return on an equally weight-
have the lowest (3.62). Large-growth is the portfolio ed basis for the small-cap index is 17.51% for the whole

EXHIBIT 1
FUNDAMENTAL CHARACTERISTICS OF SIZE AND VALUE PORTFOLIOS

Market Value ) Market- Dividend Price- Cash Flow Debt/

(million £ to-Book Yield Earnings Yield Equity

Small-Cap Value 12.17 0.63 5.55 15.11 0.38 0.10

Small-Cap Medium 20.24 1.32 5.44 14.69 0.31 0.07
Small-Cap Growth 24.67 4.41 4.14 18.40 0.21 0.07
Midcap Value 168.95 0.68 5.71 18.83 0.32 0.27
Midcap Medium 164.79 1.36 5.37 14.12 0.28 0.20
Midcap Growth 163.84 4.30 3.62 18.25 0.19 0.16
Large-Cap Value 1142.92 0.70 5.99 16.88 0.29 0.35
Large-Cap Medium 1346.31 1.36 5.37 14.66 0.28 0.32
Large-Cap Growth 1317.13 4.21 5.71 21.01 0.18 0.30
Small-Cap 19.03 2.12 5.04 16.07 0.30 0.08
Large-Cap 1268.78 2.09 5.09 15.73 0.25 0.32
Value 441.34 0.67 5.75 15.60 0.33 0.24
Growth 501.88 4.31 3.89 19.22 0.19 0.18

FALL 1999 THE JOURNAL OF PORTFOLIO MANAGEMENT 3

EXHIBIT 2
AVERAGE ANNUAL PORTFOLIO RETURNS (%)

1968-1997 1968-1977 1978-1987 1988-1997

E.W. V.W. E.W. V.W. E.W. V.W. E.W. V.W.

Small-Cap Value 25.75 21.83 24.44 22.90 35.91 29.58 15.92 12.03
Small-Cap Medium 16.51 14.98 14.59 13.76 25.63 22.85 8.52 7.59
Small-Cap Growth 10.25 10.07 9.68 8.29 17.78 16.84 2.52 4.53
Midcap Value 24.41 23.68 25.58 24.69 30.31 29.95 16.57 15.59
Midcap Medium 15.94 15.95 12.59 13.64 23.16 22.24 11.65 11.53
Midcap Growth 12.15 11.74 10.48 9.87 19.50 18.44 5.85 6.37
Large-Cap Value 20.58 21.41 18.37 22.45 26.03 24.28 16.99 17.06
Large-Cap Medium 15.92 15.90 13.20 11.97 22.96 21.90 11.11 13.58
Large-Cap Growth 13.54 11.90 10.83 5.98 17.87 17.38 11.74 12.40
Small-Cap 17.51 15.63 16.24 14.98 26.44 23.09 8.98 8.05
Large-Cap 16.68 16.40 14.13 13.47 22.29 21.18 13.28 14.35
Value 23.58 22.31 22.79 23.35 30.75 27.94 16.50 14.89
Growth 11.98 11.24 10.33 8.05 18.38 17.55 6.70 7.77

E.W. stands for equal-weighted returns and V.W. for value-weighted returns.

sample period, which is just 83 basis points higher than Apart from the cyclical movements of the size
the return of the large-cap index.5 Although this result spread, the rationale for style rotation is also evident
may appear surprising at first, given the size effect that from higher frequency variations. Out of 348 months
has been observed in the U.K. market for many years, in our sample, small-caps performed better in 183
it can be explained by looking at the last subperiod. months or 53% of the time, while large-caps were bet-
From July 1988 to June 1997, the large-cap index ter off the rest of the 348 months (47% of the time).
earned an average equal-weighted return of 13.28%, Although the sign variation in the case of the
while the small-cap index had an average annual per- value/growth spread is not as apparent as in the case of
formance of 8.98%. size spread, there are still some periods when growth
The size spread, however, is not consistent across stocks g ive better returns compared to value stocks. In
different subportfolios. Comparing the performance of 232 months (67% of the total period), the value/growth
small-cap and large-cap indexes of the same MB cate- spread was positive; in 116 (33%), it was negative. It is
gory, it is worth noting that small-caps with low MB clear therefore that, even in this case, an investor who
ratios perform better than large-caps with low MB could predict the reversals between value and growth
ratios. Conversely, we observe a monotonic increasing could outperform a passive buy-and-hold strategy.
pattern in performance as we move from small-cap Effective implementation of style rotation strate-
growth to large-cap g rowth. gy requires a realistic assessment of the manager’s degree
of forecasting ability. Exhibit 4 shows the maximum
LIKELY GAINS FROM STYLE R OTATION and minimum possible profit for investors with differ-
ent levels of forecasting skills, rotating between small
Exhibit 3 shows the twelve-month moving aver- and large stocks. The allocation decision is assumed to
age of the return spread (on an equally weighted basis) occur at monthly intervals on the basis of the corre-
between the small- and large-cap index and between sponding index performance. A month is assumed to
value and growth portfolios. It is clear that different have been predicted correctly when the hypothetical
times favor different types of stocks. The small-cap investor has chosen to allocate all its funds in the best-
investment strategy, for example, has two good cycles in performing index.
the first half of our sample, each one lasting about 2.5 to Exhibit 4 first illustrates the two extreme paths of
3.0 years (1971-1973 and 1977-1980). Large-caps, on style rotation results that connect the ultimate points of
the other hand, are more profitable from 1988 to 1992. being totally right (at the top left) and totally wrong (at

4 THE PROFITABILITY OF STYLE ROTATION STRATEGIES IN THE UNITED KINGDOM FALL 1999
EXHIBIT 3
TWELVE-MONTH MOVING AVERAGE OF STYLE SPREADS

the lower right). There are 348 months in our sample, tion costs exceed the potential benefits of rotation.
from July 1968 to June 1997. Someone who is able to Exhibit 4 shows the impact of forecasting accu-
predict every single one of the 348 months correctly racy rates on the net average annual returns. The upper
and choose the right style to invest in could have earned curve cor responds to the best case scenario, the lower
a 33.81% average annual return, or 17.47 percentage curve to the worst case scenario for the same accuracy
points above the average total return of the FT All Share level. The slope of the upper best case curve falls grad-
during the past thirty years. Assuming 100 basis points ually at the top but becomes steeper as the timing accu-
transaction costs for every switch, the perfect foresight racy rate diminishes. Exactly the reverse applies to the
strategy results in 160 switches, which reduces the aver- lower worst case curve; the downward slope is steepest
age returns to 28.29%. At the other extreme, an investor at the beginning and flattens at the end.
who is consistently wrong on the direction of the spread It is interesting to note that even with a modest
would generate an average gross annual return of 0.38% 35% forecasting ability the highest possible profits from
or –5.14% after deducting transaction costs. rotation exceed the profits of the FT All Share index
We assume that an investor can succeed in fore- and break even with the performance of the small-cap
casting only a given percentage of these 348 months. To index. An investor who picks the wrong months (worst
assess the robustness of style rotation, we calculate the case scenario) needs more than a 90% accuracy rate to
highest and the lowest possible profits under different have an advantage over passive strategies.
forecasting accuracy levels starting from 5% to 95%. We also experiment with transaction costs of
The best case scenario for a 60% accuracy rate, for 200 basis points for every switch. The higher costs
example, corresponds to the case when an investor obviously have a negative effect on the rotation profits.
manages to capture the best 209 (60%) months with the The net average annual return for a perfect foresight
highest absolute return spread. The worst case scenario strategy is now reduced to 22.77%. In addition, at least
on the other hand is when an investor misses the 139 a 55% accuracy rate is required to outperform the
best months (40%). In this case, of course, the transac- small-cap passive benchmark.

FALL 1999 THE JOURNAL OF PORTFOLIO MANAGEMENT 5

EXHIBIT 4
IMPACT OF STYLE TIMING ACCURACY RATES ON NET ANNUAL RETURNS
(SMALL VERSUS LARGE ROTATION)

Consider the same type of analysis in the case of gy are even more disappointing when applying a 200
value/growth rotation. A perfect foresight rotation bp transaction cost. Even if someone could forecast the
strategy earns a 29.10% gross average annual return or sign of the style spread in all months correctly, the
24.51% net of 100 bp transaction costs; this is just 93 bp investor would not be able to outperform the value
more than the return of a value buy-and-hold strategy. buy-and-hold strategy.
The downside risk, however, is also less than in the case The analysis shows that being generally accurate
of small/large rotation. The gross returns of a rotation in style rotation is important, but not as important as
strategy that has been totally wrong in forecasting the being accurate at certain periods in time. Jeffrey [1984]
sign of the value/growth spread are 6.47%, or 1.88% shows that high forecasting accuracy may not necessar-
after adjusting for transaction costs. It is obvious that ily correspond to high returns if one misses some of the
the smaller variation of the value spread makes rotation few months when the absolute spread is very wide. A
less profitable but less risky as well. skillful market timer, therefore, is the one who not only
Exhibit 5 shows the minimum and maximum manages to forecast with high accuracy rate but also can
possible gains for value/growth rotations corresponding be right on the months that count most.
to different levels of forecasting accuracy. The shapes of So far, we have assumed that an investor makes
the cur ves look the same as in Exhibit 4, but the dis- either the best or the worst possible choice for a given
tance between them is now narrower, indicating the level of forecasting accuracy. For each accuracy rate,
lower risk that is involved with value/growth rotation. however, there is a whole distribution of profits. To
In this case, however, even if investors always pick the assess the average profitability of a style rotation strategy,
right month, they need a minimum of 75% accuracy to one has to estimate the entire distribution of the prof-
exceed the value buy-and-hold strategy. it/loss function at different levels of forecasting skill.
The results for the value/growth rotation strate- Thus, we conduct a simulation experiment in which

6 THE PROFITABILITY OF STYLE ROTATION STRATEGIES IN THE UNITED KINGDOM FALL 1999
EXHIBIT 5
IMPACT OF STYLE TIMING ACCURACY RATES ON NET ANNUAL RETURNS
(VALUE VERSUS GROWTH ROTATION)

each iteration corresponds to a different combination of age points. Even a 70% accuracy ratio this time is not
months assumed to be correctly predicted for a certain very likely to outperform the small-cap index. The
probability rate, and consequently produces a different average net returns for a 70% accuracy rate is 14.03%
gross average annual profit and loss schedule. We run a when transaction costs of 200 basis points are assumed.
simulation with 10,000 iterations and estimate the dis- We again apply the same simulation exercise to
tribution of the resulting profits for four different fore- the value/growth rotation scheme. Exhibit 7 gives the
casting accuracy levels — 50%, 60%, 70%, and 80%. distributions of annual net returns from rotation for
Exhibit 6 shows the simulated cumulative distri- different accuracy rates. The means of the simulated
butions of the net annual returns that result from a distributions after the deduction of 100 basis points
monthly rotation strategy with small- and large-cap transaction costs are 12.85%, 15.17%, 17.61%, and
stocks. Each curve corresponds to a different level of 20.14% for the four accuracy rates. These results show
forecasting accuracy. The straight line denotes the aver- that even with a rotation scheme with an 80% accura-
age annual return performance of the small-cap index, cy rate it is almost impossible to beat the value buy-
which is the target threshold. and-hold strategy. The returns are significantly reduced
The impact of accuracy rates on the distribution when we repeat the simulations with 200 basis points
of rotation profits is clear from the graph. The whole transaction costs.
distribution is shifted to the right as the rate of fore- In short, our results suggest that a successful
casting accuracy increases. The simulated mean annual value/growth rotation strategy requires considerable
return after adjusting for 100 basis points transaction levels of forecasting skill. More specifically, after con-
costs is 12.15% with a 50% accuracy rate; 15.51% with trolling for transaction costs, one needs to be correct
60%; 18.90% with 70%; and 22.34% with 80%. more than 80% of the time to exceed the performance
Applying 200 basis points every time a switch is of a buy-and-hold value strategy. Thus, the long-term
made reduces the net annual returns about 4 percent- consistency of value stocks makes any attempt to take

FALL 1999 THE JOURNAL OF PORTFOLIO MANAGEMENT 7

EXHIBIT 6
SIMULATED NET ANNUAL RETURN DISTRIBUTIONS (SMALL VERSUS LARGE ROTATION)

100 basis points are deducted every time a switch is made. The vertical line is the average annual return of the small-cap buy-and-hold
strategy, which is the target threshold.

EXHIBIT 7
SIMULATED NET ANNUAL RETURN DISTRIBUTIONS (VALUE VERSUS GROWTH ROTATION)

100 basis points are deducted every time a switch is made. The vertical line is the average annual return of the small-cap buy-and-hold
strategy, which is the target threshold.

8 THE PROFITABILITY OF STYLE ROTATION STRATEGIES IN THE UNITED KINGDOM FALL 1999
advantage of its monthly variations rather risky with ordinary least squares, and logit estimation on the
anything less than really superior forecasting skills. small/large spread for the whole sample period (July 1968-
The results are more promising for size style rota- June 1997).7 The first column shows the results of univari-
tion. Further evidence suggests that a 65% to 70% accu- ate OLS regression and the second of multiple OLS. All
racy rate is not beyond the scope of a relatively simple OLS regressions have been corrected for heteroscedastici-
forecasting model based on economic fundamentals. ty using the method suggested by White [1980]. The last
column in Exhibit 8 gives coefficients and t-statistics from
DETERMINANTS OF SIZE a logit regression estimation, which uses as a dependent
AND VALUE SPREADS variable a binary variable that takes the value of one if the
spread is positive and zero if the spread is negative.
An effective implementation of a style rotation In the univariate regression, the inflation and the
strategy requires the development of some type of equity risk premium are highly significant, while the
model to predict the relevant style spreads. We test the term structure and the dividend yield ratio only
sensitivity of the size and value spreads to variables as marginally pass the significance test. All the variables
follows: the annual percentage change in the coincident retain their sign in OLS and logit estimation, and some
indicator; the monthly change in the three-month of them become more significant. The adjusted R2 in
Treasury bill yield; a term structure variable defined as the multiple OLS regressions is 12.33%, and the value of
the monthly yield difference between the twenty-year the log likelihood function 642.05. The particular selec-
government gilt and the three-month Treasury bill; and tion of variables gives the highest adjusted R2 and the
an inflation variable derived from the monthly logarith- lowest Akaike information criterion. The results of the
mic change in the consumer price index. We also test regressions point out that small-cap returns benefit from
the monthly equity risk premium, the monthly change rising interest rates and equity risk premium, widening
in the pound/dollar exchange rate, and a dividend yield of the yield cur ve, and lower rates of inflation.
ratio variable that corresponds to the average dividend Exhibit 9 shows the coefficients and the t-statis-
yield of one index over the average dividend yield of the tics from the regressions on the value/growth spread. In
other. All the variables that are used are lagged one contrast to the size spread regressions, the term structure,
month to ensure that the tests are predictive in nature.6 the equity risk premium, and the dividend yield ratio are
Exhibit 8 reports results from univariate, multiple now insignificant in the univariate OLS regressions, and

EXHIBIT 8
DETERMINANTS OF SMALL/LARGE SPREAD

Univariate OLS OLS Logit

Constant –0.0140 –0.5839

(–1.3578) (0.8993)
Inflation –1.1769∗∗ -1.4503∗∗ -51.4201∗∗
(–2.6860) (–2.9284) (–2.3330)
Term Structure 0.0021∗ 0.0008 0.1290∗∗
(1.7828) (0.7403) (2.1570)
Annual Change in Coincident Indicator 0.0548 0.0461 2.2591
(1.5102) (1.2543) (0.9720)
Change in Three-Month T-Bill Yield 0.0294 0.0411∗∗ 2.3729
(1.2440) (2.1989) (1.5426)
Small/Large Dividend Yield Ratio 0.0283∗ 0.0238∗ 0.9051
(1.6927) (1.9067) (1.2059)
Equity Risk Premium 0.1996∗∗ 0.0411∗∗ 11.9202∗∗
(2.9033) (2.1989) (3.7066)

Monthly difference between returns of low market value index and high market value index (equal-weighted and same market-to-book
ratio). All independent variables are lagged one month. T-statistics are reported in parentheses.

FALL 1999 THE JOURNAL OF PORTFOLIO MANAGEMENT 9

are not included in the other estimation procedures. market characteristics. Thus, forecasting the sign of the
Instead, the one-month lagged value/growth spread style spread may be sufficient for a successful style rota-
together with the inflation variable seems to be the most tion strategy. We classify each month as 1 or 0 based on
important explanatory factors. The inflation sign is con- the sign of the style spread. If in a particular month
sistent with the U.S. findings and indicates that rising small-caps (value stocks) perform better than large-caps
inflation hurts value stocks more than growth stocks, (growth stocks), we classify this month as 1; otherwise
causing the next-month return spread to be negative. we set it to 0. We use the variables described previous-
The change in the short-term interest rate and ly as predictors.
the annual change in the coincident indicator that seem Fitting the two logit models, we get the form:
to be marginally significant in the univariate case
become insignificant when they are regressed together
Pˆ t
with other variables. The monthly change in the £/$ log = aˆ + bˆ X t
exchange rate exceeds the 95% significance level in all 1 − Pˆ t
estimation cases. The sign of the variable indicates that
a rise in the monthly £/$ exchange rate benefits The conditional probability P̂t gives the likeli-
growth more than value securities. The adjusted R2 of hood that the next month will be a value (growth) or a
the model is lower than in the case of the small/large small-cap (large-cap) month and is clearly the focus of
spread but still at a satisfactory level, 9.56%. this exercise. We use a recursive forecasting technique
and a holdout sample of 276 months for out-of-sample
STYLE R OTATION BASED ON THE evaluation. An initial estimation period of seventy-two
LOGIT REGRESSION MODEL months from July 1968 through June 1974 is used to
generate probabilities for the next twelve months start-
The logit regressions suggest that the sign of the ing from July 1974 through June 1975. Then the pre-
style spreads is related to a number of economic and vious twelve months are added to the estimation peri-

EXHIBIT 9
DETERMINANTS OF VALUE/GROWTH SPREAD

Univariate OLS OLS Logit

Constant 0.0111∗∗ 0.8363∗∗

(6.6317) (4.9181)
[Value-Growth] (–1) 0.2140∗∗ 0.2059∗∗ 18.0020∗∗
(3.7413) (3.6142) (3.1944)
Annual Change in Coincident Indicator 0.0404∗ 0.0218 1.7330
(1.9410) (1.0520) (0.7789)
Inflation –0.5777∗∗ -0.5355∗∗ –42.6208∗∗
(–3.3918) (–3.1417) (–2.5598)
Change in Three-Month T-Bill Yield –0.0131∗ -0.0071 –0.7569
(–1.9596) (–0.5158) (–0.5141)
Term Structure 0.0004
(0.9204)
Monthly Change in £/$ Exchange Rate –0.0850∗∗ -0.0926∗∗ –4.8142∗∗
(–2.1185) (–2.4233) (–2.4177)
Equity Risk Premium 0.0298
(1.0776)
Value/Growth Dividend Yield Ratio –0.0001
(–0.0298)

Monthly difference between returns of low MB index and high MB index (equal-weighted and same market value). All independent vari-
ables are lagged one month. T-statistics are reported in parentheses.

10 THE PROFITABILITY OF STYLE ROTATION STRATEGIES IN THE UNITED KINGDOM FALL 1999
od, and the model is reestimated. Using the new coef- months (17.39%) when the probabilities signal a
ficients, we generate probability estimates for the fol- growth month (P̂t < 0.5).
lowing twelve months, starting from July 1975 through Exhibit 10 provides the results of different trad-
June 1976. The procedure is repeated twenty-three ing strategies that can be developed based on the esti-
times, and a time series of logit probabilities for both mated logit probabilities, and compares their perfor-
spreads is generated. mance with the performance of various passive strate-
A probability value above 0.5 indicates that a gies. Strategy 1 invests 100% in small-cap securities
month that favors small-caps is likely to occur, while a whenever the logit model signals a small-cap month
probability value below 0.5 indicates a preference for (probability greater than 0.5), and moves to 100% large
large-cap stocks. To evaluate the model, we assume stocks, whenever the logit model signals an upcoming
that if in a particular month the spread is positive and large-cap month (probability less than 0.5).
the probability is above 0.5, or if the spread is negative The drawback of strategy 1 is that it classifies
and the probability is below 0.5, the forecast has been each month as either small-cap or large-cap favorite,
successful; otherwise, the forecast has failed. regardless of the magnitude of the probability. To min-
According to this criterion, we find that the first imize this limitation and to make the allocation strate-
model results in 60.14% of correct predictions. The gy more effective, we test two other trading rules.
model for the small/large spread gives 175 months Strategy 2 defines the probability range of
(63.41%) when the logit probability is higher than 0.5 0.45–0.55 as neutral, and in this case simply allocates
and 101 months (36.59%) when the estimated proba- 100% of the funds to the same equity class as in the pre-
bility is lower than 0.5. vious month. Strategy 3 assumes that a two-month
The second model for the value/growth spread trend (sequential signal) in the predicted probabilities
gives a 68.84% accuracy rate, when employing the will give a better indication of the likelihood of differ-
50% cutoff point criterion. According to the estimat- ences in equity class returns in subsequent time periods.
ed probabilities, there are 228 months (82.61%) when Therefore, it requires the predicted probabilities to be
the logit probability is higher than 0.5 and just 48 higher (lower) than 0.5 not just for the current month

EXHIBIT 10
RESULTS OF LOGIT FORECASTING MODEL FOR SMALL/LARGE SPREAD (1974-1997)

Strategy Strategy Strategy Perfect Small- Large-

1 2 3 Foresight Cap Cap FTALL

Average Annual Returns (%) 25.20 25.02 24.634 37.46 20.58 20.21 20.07
net of trans costs (100 bp) (23.24) (23.93) (23.40) (32.11)
net of trans costs (150 bp) (22.26) (23.39) (22.79) (29.44)
net of trans costs (200 bp) (21.28) (22.85) (22.18) (26.76)
End of Period Wealth £ 21,405 £ 20,724 £ 18,698 £ 310,429 £ 7,795 £ 5,864 £ 5,848
net of trans costs (100 bp) (£ 13,479) (£ 16,068) (£ 13,998) (£ 93,311)
net of trans costs (150 bp) (£ 10,677) (£ 14,135) (£ 12,105) (£ 50,932)
net of trans costs (200 bp) (£ 8,448) (£ 12,426) (£ 10,465) (£ 27,716)
Break-Even Transaction Costs
(benchmark: small-cap index) 217 bp 379 bp 301 bp
Standard Deviation 18.18 17.95 17.93 20.59 17.18 22.13 21.86
Sharpe Ratio 1.39 1.39 1.37 1.82 1.20 0.91 0.92
Recommended Switches 47 26 59 123
% of Correct Predictions 60.14% 60.87%
% of Small-Cap Predictions 63.41% 62.68% 54.71% 51.09%
% of Large-Cap Predictions 36.59% 37.32% 27.90% 48.91%
% of Neutral Positions 17.39%

Break-even transaction costs for each strategy are transaction costs that give the same end-of-period wealth as the small-cap passive index.

FALL 1999 THE JOURNAL OF PORTFOLIO MANAGEMENT 11

but for the previous month as well, before it signals a and-hold strategies, they do not involve higher risk.
100% allocation of funds to small-caps (large-caps). If The standard deviation is about 18% for all three rota-
this condition is not met, then a 50/50 fixed allocation tion strategies, which results in Sharpe ratios before
is preferred. Strategies 2 and 3 both result in reducing transaction costs between 1.39 and 1.37. Not surpris-
the amount of monthly switches and therefore the ingly, the third rotation strategy is the one with the
transaction costs. lowest historical volatility, 17.93%.
In Exhibit 10 are the average annual returns and For each timing strategy, we calculate the level
the end-of-period wealth that corresponds to an initial of transaction costs that gives the same end-of-period
investment of £100 for each timing strategy, assuming wealth with the small-cap buy-and-hold strategy.
different levels of transaction costs. Exhibit 10 also Rotation strategy 1 can be advantageous with transac-
reports the annualized standard deviation and the tion costs up to 217 basis points, while the second strat-
Sharpe ratio as well as the number of recommended egy gives a winning edge with transaction costs less
switches for each strategy. For comparison purposes, we than 379 basis points. Institutional investors that can
also give the relevant figures for the perfect foresight switch from one equity class to another and pay less
strategy and three passive buy-and-hold strategies than 301 basis points every time can also make profits
(small-cap, large-cap, and All Share). by following the last rotation strategy.
It is clear that all three timing strategies perform Although the model can predict with an accura-
much better than the buy-and-hold strategies even after cy rate of nearly 60%, the trading strategies developed
adjusting for high levels of transaction costs. Strategy 1 from the model easily outperform the passive indexes.
seems to be the most profitable when no transaction This may indicate the ability of the probability model
costs are taken into account. When 100, 150, or 200 to capture the “good” months in our sample period.
basis points are deducted every time a switch is made, The potential gains from a perfect foresight strategy
the second strategy appears to be preferable. remain far from reach, implying that there is room for
It is interesting to note that although the three further improvement.
rotation strategies perform much better than the buy- The results of the value/growth spread model

EXHIBIT 11
RESULTS OF LOGIT FORECASTING MODEL FOR VALUE/GROWTH SPREAD (1974-1997)

Strategy Strategy Strategy Perfect

1 2 3 Foresight Value Growth FTALL

Average Annual Returns (%) 27.47 27.12 26.84 32.28 26.67 15.27 20.07
net of trans costs (100 bp) (26.18) (26.29) (25.47) (27.84)
net of trans costs (150 bp) (25.53) (25.87) (24.84) (25.63)
net of trans costs (200 bp) (24.89) (25.454) (24.18) (23.41)
End of Period Wealth £ 33,128 £ 30,651 £ 28,859 £ 98,271 £ 27,506 £ 2,122 £ 5,848
net of trans costs (100 bp) (£ 24,411) (£ 25,085) (£ 24,494) (£ 36,330)
net of trans costs (150 bp) (£ 20,931) (£ 22,676) (£ 22,750) (£ 22,008)
net of trans costs (200 bp) (£ 17,933) (£ 20,487) (£ 21,007) (£ 13,299)
Break-Even Transaction Costs
(benchmark: value index) 61 bp 54 bp 29 bp
Standard Deviation 20.15 20.13 20.07 20.04 20.32 19.66 21.86
Sharpe Ratio 1.36 1.35 1.34 1.61 1.31 0.78 0.92
No. of Recommended Switches 31 20 33 102
% of correct predictions 68.84% 67.39%
% of Value predictions 82.61% 85.28% 83.33%
% of Growth predictions 17.39% 14.72% 5.43%
% of neutral positions 11.23%

Break-even transaction costs for each strategy are transaction costs that give the same end-of-period wealth as the value passive index.

12 THE PROFITABILITY OF STYLE ROTATION STRATEGIES IN THE UNITED KINGDOM FALL 1999
confirm the message of our simulation experiment and value strategy, however, is markedly more difficult; it
are in line with the notion that the value buy-and-hold requires more than an 80% forecasting ability.
strategy is superior to value/growth rotation. We eval- We identify a number of macroeconomic and
uate three value/growth strategies following the same market factors that appear to predict the direction of
principles as in Exhibit 10, and present the results in the next month’s style spread. More specifically, our
Exhibit 11. Ignoring transaction costs, all three strate- logit regressions suggest a significant relation between
gies perform slightly better than the passive value buy- economic activity, or more generally the stage of the
and-hold index. When transaction costs are taken into business cycle, and the equity style spread dummy.
account, the profits from the rotation strategies are sig- Using the fitted logit probabilities, we develop and test
nificantly reduced. Increasing the level of transaction three alternative trading rules. Our results suggest that
costs and calculating the end-of-period wealth net of while style rotation strategies based on small and large
100, 150, and 200 basis points, we cannot find any rota- firms can be highly rewarding, they are only marginal-
tion strategy that is superior to the value buy-and-hold. ly successful in the case of value and growth stocks.
The volatilities and the Sharpe ratios indicate The fundamental implication of our findings is
that almost all the strategies have the same volatility, but that the profitability of style rotation strategies depends
their Sharpe ratios are slightly better than the value entirely on the temporal volatility of the underlying
index — between 1.36 and 1.34, compared to 1.31 for return spread between the styles that the manager is fol-
the passive value buy-and-hold strategy. lowing. Thus, the ongoing debate among professional
The solution to the linear programming prob- fund managers about style consistency and market per-
lem to calculate the amount of transaction costs to formance is fundamentally an empirical question.
break even leads to the same conclusion. The first trad- Style consistency is a prudent strategy for
ing rule can be advantageous for institutional investors investors with very long investment horizons and
that can make trades without losing more than 61 basis strong views on the performance of the targeted style.
points, while the two neutral strategies can make prof- In all other cases, controlled style rotation strategies
its relative to the value index with transaction costs of based on the underlying fundamental characteristics of
up to 54 and 29 basis points, respectively. The actual the relevant style indexes can be value-enhancing.
transaction costs of the rotation strategies, however, are
very unlikely to be that low as the specific trades (move ENDNOTES
from 100% value to 100% growth) impose very high
turnover rates. The authors are grateful to Arie Melnik, Maurizio
These results suggest that our model for the Murgia, Nick Tessaromatis, and an anonymous referee for useful
value/growth rotation can only marginally outperform suggestions and comments. They also thank seminar participants at
City University Business School and at the Joint INQUIRE Europe
the passive index alternative, and when transaction costs and INQUIRE U.K. conference in Lausanne in March 1998 for
are assumed no real benefits can be gained. Even though valuable comments.
the model can predict the monthly style trend more 1For early evidence on the size effect in the U.K. see

accurately than the size spread, the relative advantage Levis [1985], and for recent European evidence see Levis and
that it offers is much more modest. A very high accura- Steliaros [1999].
2To avoid the problems of accounting definitions we
cy rate and precision is needed for a market timer to out-
exclude from our sample banks, insurance companies, investment
perform the value buy-and-hold strategy. trusts, and property companies.
3We calculate the ratio as follows:

CONCLUSION BV (Equity Capital + Reserves) − (Good Will and Other Intangibles)

=
MV Market Value of Equity
During the thirty-year period 1968 through We deduct the goodwill and the other intangibles from the numer-
1997, value and small-cap stocks in the U.K. outper- ator to make the ratio less subject to manipulation. Some companies
formed their growth and large-cap counterparts by an write intangibles in their balance sheets and therefore report high
book values of equity, while others exclude them and report lower
average of 1,160 and 80 bp per year. Our simulation
book values. To ensure that companies calculate the book value of
results suggest that forecasting the size spread with a equity in the same way, we deduct this term from the numerator.
65%-70% accuracy rate may be sufficient to outperform We confine the analysis to companies with positive BM
a long-term small-cap strategy. Beating a long-term ratios. The companies in our sample with negative book values

FALL 1999 THE JOURNAL OF PORTFOLIO MANAGEMENT 13

 range from thirteen in 1968 to 101 in 1996. ——. “Value versus Growth: The International Evidence.” Journal
4Most companies in the U.K. have fiscal year-ends of of Finance, 1998, pp. 1975-1999.
December 31, but a number of firms use the last day of March.
Practically all companies, however, publish their financial state- Hawawini, Gabriel, and Donald Keim. “The Cross Section of
ments within three months of the end of the fiscal year. For this rea- Common Stock Returns: A Synthesis of the Evidence and
son, we decide to form portfolios and measure their performance at Explanations.” In Donald Keim and William Ziemba, eds., Security
the end of June, so that our tests would be predictive in nature both Market Imperfections in World Wide Equity Markets. New York:
for companies with December and March fiscal year-ends, and so Cambridge University Press, 1999.
that we do not use accounting information that is not actually avail-
able to the investor at the time of portfolio formation. We thus Indro, Daniel K., Christine X. Jiang, Michael Y. Hu, and Wayne
avoid a possible look-ahead bias (see Banz and Breen [1986]). Y. Lee. “Mutual Fund Performance: A Question of Style.” Journal
5Our results are consistent with the average annual of Investing, Summer 1998, pp. 46-53.
returns of Hoare Govett, the well-established U.K. small-cap index,
over the same period. The total returns of the Hoare Govett Jeffrey, Robert. “The Folly of Stock Market Timing.” Harvard
Smaller Companies index from 1968 through 1997 was 17.9% (see Business Review, July-August 1984, pp. 689-706.
Dimson and Marsh [1998]).
6There is no severe multicollinearity problem in using all Jensen, Gerald R., Robert R. Johnson, and Jeffrey M. Mercer.
these variables in the regression. The highest correlation is 0.31 “The Inconsistency of the Small-Firm and Value Stock Premiums.”
between the term structure variable and the dividend yield ratio. Journal of Portfolio Management, Winter 1998, pp. 27-36.
7We also use a GMM (generalized method of moments)

estimation that yields results that are robust to both heteroscedastic- Kahn, Virginia. “A Question of Style: Must Consistency Equal
ity and serial correlation, but the results were not different, and are Mediocrity in Mutual Funds?” Financial World, July 8, 1996, pp. 70-
not reported. 75.

REFERENCES Kester, George W. “Market Timing with Small vs. Large Firm
Stocks: Potential Gains and Required Predictive Ability.” Financial
Analysts Journal, September/October 1990, pp. 63-69.
Anderson, Richard. “A Large versus Small Capitalization Relative
Performance Model.” Market Timing Models. Burr Ridge, IL: Irwin,
Levis, Mario. “Are Small Firms Big Performers?” The Investment
1997.
Analyst, 76 (April 1985).
Arshanapalli, Bala, T. Daniel Coggin, and John Doukas.
Levis, Mario, and Michael Steliaros. “The Relative Performance
“Multifactor Asset Pricing Analysis of International Value
and Characteristics of European Small Companies.” Working
Investment Strategies.” Journal of Portfolio Management, 1998, pp. 10-
paper, City University Business School, February 1999.
23.
Macedo, Rosemary. “Value, Relative Strength, and Volatility in
Banz-Breen, 1986. AUTHOR PLEASE SUPPLY
Global Equity Country Selection.” Financial Analysts Journal,
March/April 1995, pp. 70-78.
Capaul, Carlo, Ian Rowley, and William F. Sharpe. “International
Value and Growth Stock Returns.” Financial Analysts Journal,
Ragsdale, Easton E.K., Gita R. Rao, and Leo Fochtman. “Small
January/February 1993, pp. 27-36.
versus Large Cap Stocks: Quantifying the Fundamental Reasons
Behind Relative Market Performance.” In R. Klein and J.
Case, Douglas W., and Steven Cusimano. “Historical Tendencies
Lederman, eds., Small Cap Stocks: Investment and Portfolio Strategies for
of Equity Style Returns and the Prospects for Tactical Style
the Institutional Investor. Chicago: Probus Publishing Company,
Allocation.” In R. Klein and J. Lederman, eds., Equity Style
1995.
Management. Burr Ridge, IL: Irwin, 1995
Sharpe, William F. “Likely Gains from Market Timing.” Financial
Coggin, T. Daniel. “Long-Term Memory in Equity Style
Analysts Journal, March/April 1975, pp. 60-69.
Indexes.” Journal of Portfolio Management, Winter 1998, pp. 37-46.
Sorensen, Eric, and Craig Lazzara. “Equity Style Management: The
Dimson, Elroy, and Paul Marsh. “Hoare Govett Smaller
Case of Growth and Value.” In R. Klein and J. Lederman, eds.,
Companies Index 1955-1997.” ABN AMRO Hoare Govett Limited,
Equity Style Management. Burr Ridge, IL: Irwin, 1995.
January 1998.
White, 1980. AUTHOR PLEASE SUPPLY
Fama, Eugene F., and Kenneth R. French. “Size and Book-to-
Market Factors in Earnings and Returns.” Journal of Finance 50
(1995), pp. 131-155.

14 THE PROFITABILITY OF STYLE ROTATION STRATEGIES IN THE UNITED KINGDOM FALL 1999

View publication stats