Market Timing With Moving Averages PDF

Market Timing with Moving Averages1
Paskalis Glabadanidis2
Accounting and Finance
Business School
University of Adelaide
April 15, 2015
1
I would like to thank Syed Zamin Ali, Tze Chuan ‘Chewie’ Ang, B. Ross Barmish, Jean Canil, Don Chance,
Sudipto Dasgupta, Daisy Doan, Victor Fang, Berowne Hlavaty, Daniel Orlovsky, James Primbs, Bruce Rosser, Vin-
cent Xiang, Takeshi Yamada, Alfred Yawson, Xinwei Zheng, Edward Zychowicz as well as seminar participants at
Deakin University and the University of Adelaide and participants in the 2012 Australasian Finance and Bank-
ing conference in Sydney, the 2014 J.P.Morgan quantitative conference in Sydney and the 2013 Midwest Finance
Association meetings in Chicago. In addition, I would like to express my gratitude to the editor, Huining Cao,
and two anonymous referees for their very detailed and thoughtful comments. Any remaining errors are my own
responsibility.
2
Correspondence: Accounting and Finance, Business School, University of Adelaide, Adelaide SA 5005, Australia,
tel: (+61) (8) 8313 7283, fax: (+61) (8) 8313 0172, e-mail: paskalis.glabadanidis@adelaide.edu.au.
Market Timing with Moving Averages
Abstract
I present evidence that a moving average (MA) trading strategy has a greater average return and skewness as
well as a lower variance compared to buying and holding the underlying asset using monthly returns of value-
weighted US decile portfolios sorted by market size, book-to-market, and momentum, seven international
markets as well as 18,000 individual US stocks. The MA strategy generates risk-adjusted returns of 3% to
7% per year after transaction costs. The performance of the MA strategy is driven largely by the volatility
of stock returns and resembles the payoffs of an at-the-money protective put on the underlying buy-and-
hold return. Conditional factor models with macroeconomic variables, especially the default premium, can
explain some of the abnormal returns. Standard market timing tests reveal ample evidence regarding the
timing ability of the MA strategy.
Key Words: Market timing, security selection, moving average, technical analysis, conditional models.
JEL Classification: G11, G12, G14.

1 Introduction
Technical analysis involves the use of past and current market price, trading volume and, potentially, other
publicly available information to predict future market prices. It is highly popular in practice with plentiful
financial trading advice that is based largely, if not exclusively, on technical indicators. In a perhaps
belated testament to this fact consider the following quote from the New York Times’s issue dated March
11, 1988: “Starting today the New York Times will publish a comprehensive three-column market chart
every Saturday... History has shown that when the S&P index rises decisively above its (moving) average
the market is likely to continue on an upward trend. When it is below the average that is a bearish signal.”
According to Brock, Lakonishok and LeBaron (1992), the moving average in its various implementations,
is the most popular strategy followed by investors who use technical analysis. More formally, Brock,
Lakonishok and LeBaron (1992) find evidence that some technical indicators do have a significant predictive
ability. Blume, Easley and O’Hara (1994) present a theoretical framework using trading volume and price
data leading to technical analysis being a part of a trader’s learning process. A more thorough study of
a large set of technical indicators by Lo, Mamaysky and Wang (2000) also found some predictive ability
especially when moving averages are concerned. Zhu and Zhou (2009) provide a solid theoretical reason
why technical indicators could be a potentially useful state variable in an environment where investors need
to learn over time the fundamental value of the risky asset they invest in. More recently, Neely, Rapach,
Tu and Zhou (2010,2011) find that technical analysis has as much forecasting power over the equity risk
premium as the information provided by economic fundamentals. The practitioners literature also includes
Faber (2007) and Kilgallen (2012) who thoroughly document the risk-adjusted returns to the moving average
strategy using various portfolios, commodities and currencies. In addition, Huang and Zhou (2013) use the
moving average indicator to predict the return on the US stock market while Goh et al. (2013) apply the
same idea to government bond yields and risk premia. Motivated in part by the predictive power of the
moving average indicator, Han and Zhou (2013) and Jiang (2013) construct a trend factor with considerable
cross-sectional explanatory power and substantial historical performance.
The main findings of this study are as follows. First, I present evidence that the returns to a simple
moving average switching strategy dominate in a mean-variance sense the returns to a buy-and-hold strategy
of the underlying portfolio. Second, I demonstrate that the switching strategy involves infrequent trading
1
with relatively long periods when the moving average strategy is invested in the underlying assets and the
break-even transaction costs are on the order of 3% to 7% per transaction. Thirdly, even though there
is overwhelming evidence of imperfect market timing ability of the moving average switching strategy for
a single portfolio or individual stocks, cross-sectional differences remain between the abnormal returns of
different portfolios. These differences persist when controlling for the four-factor Carhart (1997) model for
portfolios formed on past price returns and are mostly driven by differences in the volatility of portfolio
and stock returns. Fourthly, conditional models explain to a certain degree the moving average abnormal
returns but do not completely eliminate them. Fifth, I document the performance of the moving average
strategy using more than 18,000 individual stocks from the Center for Research in Security Prices. Sixth, I
present evidence of the robustness of the performance of the moving average strategy in seven international
stock markets. Seventh, I show that the lagged indicator regarding the switch into or out of the risky
asset has substantial predictive ability over subsequent portfolio returns over and above the predictability
contained in standard instrumental variables, like the default spread, investor sentiment, recession dummy
variable and liquidity risk. Last but not least, the strategy is robust to randomly generated stock returns
and bootstrapped historical returns. Nevertheless, a random switching strategy leads to negative and
statistically significant returns. The inferior performance of random switching is a testament to the market
timing ability of the moving average strategy. Furthermore, random switching generates increasingly poorer
average returns as we investigate its performance with riskier underlying assets. This is also consistent with
the performance being driven by volatility.
This paper is similar in spirit to Han, Yang and Zhou (2013). However, several important differences
stand out. First, I use monthly value-weighted returns of decile portfolios constructed by various character-
istics like size, book-to-market, and momentum.1 Value-weighted portfolios at a monthly frequency should
have a much smaller amount of trading going on inside the portfolio compared to the daily equal-weighted
portfolios investigated by Han, Yang and Zhou (2013). Secondly, the cross-sectional results in this study
are just an artefact of the decile portfolios and not the main focus of this paper while Han, Yang and Zhou
(2013) is mostly concerned with the inability of standard empirical tests to account for the moving average
strategy average returns differences across portfolios. I argue that this is largely due to using the wrong
benchmark pricing model. Using a dynamic market-timing tests and conditional asset pricing models with
1 Further findings using cash-flow-to-price, earnings-to-price, dividend-price, past return, and industry are broadly consistent
with those reported in the text and are available from the author upon request.
2
macroeconomic state variables leads to mostly negative or statistically insignificant risk-adjusted returns for
the moving average strategy. In light of this, my take on the performance of the moving average strategy
is that it is not an anomaly but instead a dynamic trading strategy that exposes investors to potential
upside returns derived from risky assets via its market timing ability. This performance is more pronounced
the more volatile the returns of the underlying risky assets are. A final caveat I need to make is that the
performance of the strategy is investigated using historical returns rather than actually trading in financial
markets. It is likely that in reality there may be adverse price impact of liquidating and initiating large
positions, especially for less liquid assets with lower trading volumes. This possibility is in the spirit of
limits to arbitrage as another potential explanation for the performance of the moving average strategy.
The nature of this empirical study is such that this potential explanation cannot be eliminated.
The highlights of this study are the superior performance of the moving average portfolios relative
to buying and holding the underlying portfolios, the infrequency of trading and the very large break-
even transaction costs, the fact that the switching strategy returns resemble an imperfect at-the-money
protective put, and that cross-sectional differences are not a new anomaly as maintained in Han, Yang and
Zhou (2013) but are due to volatility differences in the underlying portfolios and stocks. An asset with
10% higher standard deviation of returns will experience on average between 2% and 3.5% mean return
improvement between the buy-and-hold and the moving average strategy. The returns of the moving average
strategy relative to the buy-and-hold strategy exhibit a lot of convexity and, hence, will be hard to explain
using standard linear asset pricing models. The anomalous risk-adjusted performance relative to standard
models appears to be largely due to omitting market timing factors in a simple piece-wise linear framework
that captures the moving average strategy’s convexity. Furthermore, the moving average strategy appears
to be antifragile in the sense of Taleb (2012) meaning that for securities with more volatile returns there is
a greater improvement of the moving average returns relative to buy-and-hold returns.
3
2 Moving Average Market Timing Strategies
I use monthly value-weighted2 returns of sets of ten portfolios sorted by market value, book-to-market, and
momentum. The data is readily available from Ken French Data Library. The sample period starts in
January 1960 and ends in December 2011.
The following exposition of the moving average strategy follows closely the presentation in Han, Yang,
and Zhou (2013). Let Rjt be the return on portfolio j at the end of month t and let Pjt be the respective
price level of that portfolio. Define the moving average of portfolio j Ajt,L at time t with length L periods
as follows:
Pjt−L+1 + Pjt−L+2 + · · · + Pjt−1 + Pjt
Ajt,L = , (1)
L
Throughout most of the paper, I use a moving average of length L = 24 months as baseline case. Later
on, in the section dealing with robustness checks I also present results for all sets of portfolios with lags
of 6-months, 12-months, 36-months, 48-months, and 60-months. The way I implement the moving average
strategy in this paper is to compare the closing price Pjt at the end of every month to the running moving
average Ajt,L . If the price is above the moving average this triggers a signal to invest (or stay invested if
already invested at t − 1) in the portfolio in the next month t + 1. If the price is below the moving average
this triggers a signal to leave the risky portfolio (or stay invested in cash if not invested at t − 1) in the
following month t + 1.3 As a proxy for the risk-free rate, I use the return on the 30-day US Treasury Bill.
More formally, the returns of the moving average switching strategy can be expressed as follows:


 Rjt ,
 if Pjt−1 > Ajt−1,L
R̃jt,L = (2)

 rf t ,
 otherwise,
in the absence of any transaction costs imposed on the switches. For the rest of the paper and in all of
the empirical results quoted I consider returns after the imposition of a one-way transaction cost of τ .
2 I use value-weighted portfolio returns to control for the amount of rebalancing trading inside the various portfolios. The
empirical results in this paper are much stronger when equal-weighted portfolios are used. However, this may understate the
break-even transaction costs as equal weighted portfolios require a lot of trading to be replicated. I also use monthly returns
to limit the amount and frequency of trading of the various portfolios. The empirical results using daily portfolio returns are
similar in spirit and generate higher abnormal returns without a disproportionate amount of additional trading.
3 An alternative version of the switching strategy involves investing in the market portfolio instead of the risk-free asset.
This version of the switching strategy has a somewhat inferior performance compared to the baseline case investigated in the
article. Nevertheless, it is an interesting case to consider and I am grateful to an anonymous referee for suggesting this idea to
me.
4
Mathematically, this leads to the following four cases in the post-transaction cost returns:





 Rjt , if Pjt−1 > Ajt−1,L and Pjt−2 > Ajt−2,L ,




 Rjt − τ,
 if Pjt−1 > Ajt−1,L and Pjt−2 < Ajt−2,L ,
R̃jt,L = (3)




 rf t , if Pjt−1 < Ajt−1,L and Pjt−2 < Ajt−2,L ,




 rf t − τ,
 if Pjt−1 < Ajt−1,L and Pjt−2 > Ajt−2,L .
depending on whether the investor switches or not. Note that this imposes a cost on selling and buying
the risky portfolio but no cost is imposed on buying and selling the Treasury bill. This is consistent with
prior studies like Balduzzi and Lynch (1999), Lynch and Balduzzi (2000), and Han (2006), among others.
Regarding the appropriate size of the transaction cost, Balduzi and Lynch (1999) propose using a value
between 1 and 50 basis points. Lynch and Balduzi (2000) use a mid-point value of 25 basis point. In order
to err on the side of caution, I use a value of 50 basis points in all the empirical results presented in the
next section or τ = 0.005.
I construct excess returns as zero-cost portfolios that are long the MA switching strategy and short
the underlying portfolio to determine the relative performance of the moving average strategy against the
buy-and-hold strategy. Denote the resulting difference between the return of the MA strategy for portfolio
j at at the end of month t, R̃jt,L − Rjt , and the return of portfolio j at the end of month t, Rjt , as follows:
MAPjt,L = R̃jt,L − Rjt , j = 1, . . . , N. (4)
The presence of significant abnormal returns can be interpreted as evidence in favor of superiority of the
moving average switching strategy over the buy-and-hold strategy of the underlying portfolio. Naturally,
the moving average switching strategy is a dynamic trading strategy so it is perhaps unfair to compare
its returns to the buy-and-hold returns of being long the underlying portfolio. Nevertheless, I impose
conservatively large transaction costs and later report much larger break-even transaction costs.
5
3 Profitability of Moving Average Portfolios
In this section, I present summary statistics for the underlying portfolios performance, the performance of
the moving average switching strategy, and the excess MAP returns for nine sets of ten portfolios sorted
by market value, book-to-market ratios, and momentum. Next, I present the four-factor Carhart (1997)
regression results for the MAP returns of each set of portfolios. Finally, I discuss the result in light of the
potential reasons for the profitability of the moving average switching strategy.
3.1 Performance
Table 1 reports the first three moments and the Sharpe ratios of the underlying portfolios, the moving
average (MA) switching strategy applied to each portfolio, and the excess return (MAP) of the MA switching
strategy over the buying and holding (BH) of the underlying portfolio. The results are intriguing. First, the
average annualized returns of the MA strategy are substantially higher than the average annualized returns
of the underlying portfolios. Second, this average return difference come with a lower return standard
deviation and, hence, the MA switching strategy appears to dominate the underlying BH portfolio strategy
in a mean-variance sense4 . Third, for the vast majority of portfolios, the underlying BH has a negative
return skewness while the MA strategy in most cases exhibits positive skewness. This feature will make
the MA switching strategy very attractive to investors who have a preference for skewness. Fourth, the
risk-return trade-off is improved substantially resulting in much higher Sharpe ratios of the MA returns
when compared to the Sharpe ratios of the BH returns. Fifth, these results hold for almost all portfolio
across all sorting variables. Furthermore, there appear to be some substantial cross-sectional differences
related to the size effect (Panel A), the value premium (Panel B) as well as momentum premia (Panels C).
Insert Table 1 here.
From the evidence presented in Table 1 it appears that portfolios with higher standard deviations tend
to experience higher average improvements between the buy-and-hold and the moving average strategy
performance or ∆µ. A more formal way to test this is through a cross-sectional regression which is presented
4 Issues related to the statistical significance of the mean return improvement and the return standard deviation reduction
are explored in the next section.
6
next:
∆µ = −2.19 + 0.35σ, (5)
(1.08) (0.06)
where the cross-sectional R2 = 0.5834 and the standard errors are corrected for heteroscedasticity and
reported in parentheses. The slope on σ is highly statistically and economically significant suggesting that
on average ∆µ increases by 3.5% on an annualized basis when the portfolio return volatility increases by
10%.5
The MA strategy clearly performs very well compared to the BH strategy. The next section investigates
more formally the reasons for this performance in the traditional empirical asset pricing framework of factor
models and abnormal returns.
3.2 Abnormal Returns
The asset pricing model I consider in this section is the four-factor Carhart (1997) model:6
MAPjt,L = αj + βj,m rmkt,t + βj,s rsmb,t + βj,h rhml,t + βj,u rumd,t + ǫjt , j = 1, . . . , N, (6)
where rmkt,t is the excess return on the market portfolio at the end of month t, rsmb,t is the return on
the SMB factor at the end of month t, rhml,t is the return on the HML factor at the end of month t, and
rumd,t is the return of the UMD factor at the end of month t. Note that all of the risk-adjusted alphas
are highly statistically statistically significant. Moreover, they are all still quite substantial economically
ranging between 3% and 7% per year. The factor loadings on the market portfolio, SMB, and HML are
largely unchanged across the three sets of decile portfolios while the loadings on the UMD factor are mostly
positive and highly statistically significant (with the exception of momentum deciles 9 and 10). This suggests
that all four factors have a role to play in driving the performance of the MAP returns. Nevertheless, the
average adjusted R2 values indicate that only around half of the return variation can be explained and
5 A quick glance at Table 1 reveals that the Low momentum (extreme loser) portfolio has the highest σ as well as the
highest ∆µ suggesting that it might be an outlier. Dropping this observation from the cross-sectional regression reduces the
magnitude of the slope coefficient to 0.16 but it is still statistically and economically significant.
6 Results for the CAPM and Fama-French three factor models yield very similar and, frequently, stronger than the results
for the Carhart (1997) model. These additional findings are not reported in the paper in the interest of saving space. They
are available from the author upon request.
7
accounted for by the market portfolio return, size, value and momentum. This leaves a large portion of
return variation that cannot be accounted for.
3.3 Explanation
Before making an attempt at explaining the reasons for the profitability of the MA strategies performance,
it is useful to inspect a scatter plot of the MA strategy returns versus the underlying BH strategy returns
for the same portfolio. For ease of exposition I provide a plot for a single portfolio only.7 Figure 1 presents
the scatter plot for the first decile of the market-capitalization sorted deciles.
Insert Figure 1 about here.
The strategy is clearly triggering false positive signals where we are told to stay invested or switch
into the underlying asset with a subsequent negative return (negative quadrant of returns in the figure).
Similarly, there are a few instances of a false negative signal where we switch into the risk-free asset while
the underlying risky asset has a positive excess return in the following period. Nevertheless, the signal is
right about two out of every three times and in those instances the scatter plot resembles the payoff of
an at-the-money put option combined with a long position in the underlying risky asset. This positive
convexity is the driving factor for the relative outperformance of the moving average strategy relative to
the buy-and-hold strategy. Holding the signal success rate constant, risky assets with more volatile returns
will experience a higher average outperformance and this is evidenced in all of the previous tables.
3.4 Individual Stocks
In this subsection I report results on the performance of moving average strategies with individual stocks in
the CRSP database starting in January 1960 until December 2011. This results in 28,685 individual stocks.
I retain only the stocks for which a contiguous block of non-missing 48 monthly returns is available.8 This
leaves a total of 18,397 stocks. Instead of reporting the results in tabular form, I report the key attributes
in Figure 2 as histograms.
7 The scatter plots for the other portfolios sorted on the various characteristics are available from the author upon request.
8 Asa robustness check, I also consider requiring a longer series on non-missing monthly returns of 72 months and 84 returns.
This results in a smaller number of stocks but does not materially change the results presented for the larger set of stocks with
only 48 consecutive non-missing monthly returns. The additional results are available from the author upon request.
8
Insert Figure 2 about here.
The performance of the MA strategy with individual stocks is largely consistent with the performance
of the MA strategy with portfolios. The risk of the MA strategy is uniformly always smaller than the risk
of the underlying stock. The difference in average returns between the MA and BH strategies is positive
for 18,078 or more than 98% of all individual stocks I investigate. The superior performance of the MA
strategy over the BH strategy does not come at the cost of a large number of trades. The MA strategies
of almost 10,000 stocks have between 1 and 10 switches during the sample period under consideration.
The break-even transaction costs of 15,000 stocks are between 0 and 100 basis points. Bear in mind that
the break-even transaction costs are in addition to the 50 basis point one-way transaction cost imposed in
calculating the MA returns. Finally and, most importantly, for the vast majority of individual stocks, the
probability of being on the right side of the market, p1 , is well above 50% with an average value of 72.4%.
Finally, I repeat the cross-sectional regression of ∆µ on σ for all the individual stocks in the sample:
∆µ = 5.09 + 0.20σ, (7)
(0.28) (0.03)
where the cross-sectional R2 = 0.1198 and the standard errors are corrected for heteroscedasticity and
reported in parentheses. The slope on σ is highly statistically and economically significant suggesting that
when σ increases by 10% the mean return of the moving average strategy is 2% higher than the mean return
of the buy-and-hold strategy on an annualized basis. Despite the substantial cross-sectional variation in ∆µ
and σ at the level of individual stocks, the evidence still points towards volatility of the underlying security
as the variable associated with the improvement in the mean return of the spread between MA and BH.
4 Robustness Checks
In this section, I report my findings for several robustness checks performed on the performance of the
MA strategy versus the BH strategy for decile portfolios sorted on market capitalization, book-to-market
ratios and momentum. First, I show evidence of the MA strategy performance in two subperiods of equal
length. Second, I show how the MA strategy performs when various lag length are used. Third, I report
9
the intensity of trading, the break-even transaction costs, the probability of being on the right side of the
market, and the statistical significance of the mean return and standard deviation improvement. Finally, I
also report how the number of trades and the break-even transaction costs vary with alternative lengths of
the moving average.9
4.1 Subperiods
In this robustness check, I split the sample in two when the first half-period starts in January 1960 and ends
in December 1986 while the second half-period starts in January 1987 and ends in December 2011. Overall,
the results reported in previous section are robust with respect to the two sub-periods. The abnormal
returns are a little smaller for size and momentum deciles but most are statistically significant in both
subperiods.
4.2 Alternative Lag Lengths
Next, I investigate the effect of varying lengths of the moving average window on the magnitude of the
average MAP returns for all the sets of portfolios under investigation. Specifically, I investigate the average
MAP returns for moving average windows of 6 months, 12 months, 36 months, 48 months, and 60 months
in length. The average returns are economically and statistically significant with moving average window
lengths of less than 24 months, the baseline window used previously. The significant positive excess returns
persist with moving average window length of 36 months and decrease markedly when I use longer window
lengths of 48 and 60 months. Importantly, significant cross-sectional variation persists for all sets of port-
folios with the exception of book-to-market portfolios. The range of annual MAP returns with a moving
average window of 6 months is between roughly 8% and 21%. The range of annual MAP returns with the
length of the moving average is 12 months is between approximately 5% and 15%. When I increase the
moving average window length to 36 months the range of average annualized MAP returns drops to between
1% and 9%, depending on which sets of deciles I consider.

9 The robustness checks presented here are only a small portion of the total number of robustness checks performed in
preparing this article. Results for equal-weighted portfolio, both daily and monthly returns, double-sorted portfolio sets along
size/book-to-market, volatility and size/past performance show the profitability of the MA switching strategy is robust with
respect the frequency of the data, the portfolio construction and the portfolio composition. These additional results are
available from the author upon request.
10
4.3 Statistical Significance, Trading Intensity and Break-Even Transaction Costs
Table 3 reports the statistical significance in the improvement of the average return ∆µ of the MA portfolio
over the BH portfolio as well as the reduction in the return standard deviation ∆σ. The evidence points
towards a substantial improvement in a mean-variance sense for all sets of portfolios under consideration.
The annualized improvement in the average return ranges from 2% to 10% while the reduction in the
standard deviation is between approximately 3% to 11%. The MA strategy is active more often than not
ranging between 58% to 86% of the sample. Yet, the number of transactions, NT, is never above 60 and
can be as little as 29 for decile 10 of the book-to-market sorted portfolios. In a sample of 600 months this
translates into average holding periods of between 10 and 20 months where the MA strategy is continuously
invested either in the risky asset or the risk-free asset. Next, I report the break-even transaction costs,
BETC, calculated as the level of one-way proportional transaction cost in percent that would eliminate
completely the average MAP portfolio return. The values of the BETC for the various sets of portfolio
range between almost 3% to as high as 7%. This is a very large level of transaction costs which should
more than compensate for the rebalancing costs associated with implementing the value-weighted portfolio
scheme used to construct the portfolio returns. Finally, the last two columns report the fraction of months
that the MA strategy generates a positive return (p1 ) as well as a return that is in excess of the risk-
free rate (p2 ). I report the statistical significance of the null hypothesis that the true fraction of times
is above 50%. With the exception of three momentum, all the observed fractions are highly statistically
significant and range from 55% to 63% success rate of the MA strategy being on the right side of the market.
These are considerably favorable odds and in line with the evidence reported previously about the superior
performance of the MA switching strategy.
The next robustness check I perform is to investigate the intensity of trading and its impact on break-
even transaction costs at various lengths for the moving average window. As expected, the findings are that
shorter window lengths lead to more intensive trading and vice versa. Similarly, the break-even transaction
cost, BETC, decrease when shorter windows are used and increase or stay roughly the same when I increase
the length of the moving average window.
11
The large values of BETC and the relatively small number of transactions NT suggest that the MA
switching strategy is successful at improving the average returns compared to a buy-and-hold investment
strategy. The superior performance is robust with respect to two subperiods, various lag lengths of the
moving average window and persists for between 6 and 60 months with very reasonable intensity of trading
and substantial break-even transaction costs. This suggests that the MA switching strategy will be of use
to not only large institutional investors but will also be of value to individual investors. These findings are
perhaps indicative of the reasons for the wide popularity of the moving average as a technical indicator in
practice.10
5 Drivers of Abnormal Returns
In this section, I investigate the reasons for the superior returns of the MAP portfolios. To this end, I
control the MAP performance for economic expansions and contractions as well as other state contingencies
like the sign of the lagged market return. Furthermore, I investigate the conditional performance of the
MAP returns while controlling for two instrumental variables with documented predictive power over stock
returns and an additional risk factor to control of the possible presence of liquidity risks. Finally, I perform
two simulations using bootstrapped returns and randomly generated returns.
5.1 Market Timing
The first approach towards testing for market timing ability is the quadratic regression of Treynor and
Mazuy (1966):
2
MAPjt,L = αj + βj,m rmkt,t + βj,m2 rmkt,t + ǫjt , j = 1, . . . , N, (8)
where statistically significant evidence of a positive βj,m2 can be interpreted as evidence in favor of market
timing ability. The second approach is to allow for a state-contingent βj,m based on the direction of move
of the market return as in Henriksson and Merton (1981):
MAPjt,L = αj + βj,m rmkt,t + γj,m rmkt,t I{rmkt,t >0} + ǫjt , j = 1, . . . , N, (9)

10 Additional results are available from the author upon request.
12
where I{rmkt,t >0} is an indicator function of the event of a positive market return. A statistically significant
value of γj,m is usually interpreted as evidence of successful market timing ability.
Table 4 presents the results of the two market timing regressions for various sets of value-weighted
decile portfolios. Panel TM presents the empirical results from the Treynor and Mazuy (1966) quadratic
regression while Panel HM presents the results for the state-contingent beta regression of Henriksson and
Merton (1981). In both regressions, both βj,m2 and γj,m are highly statistically significant, indicating there
is strong evidence of market timing ability of the switching moving average strategy. Nevertheless, the
alphas of quite a few decile portfolios are also statistically significant at conventional levels. This suggests
that market timing alone is not the sole driver of the abnormal returns generated by the switching moving
average strategy.
5.2 Business Cycles and Market States
Following Han, Yang and Zhou (2013), I investigate the performance of the MAP portfolio returns in
economic expansions and contractions as well as in up and down markets as defined by the sign of the market
return. Table 5 presents the results for the various sets of portfolio deciles. The evidence overwhelmingly
indicates that MAP abnormal returns are higher during economic contractions and following positive market
factor returns. For portfolios constructed by sorting on past performance (short-term/long-term reversal
and medium-term momentum) there is also evidence of a significant cross-sectional differences between the
High and Low MAP abnormal returns which cannot be accounted for by the four Carhart (1997) factors
and the recession dummy and up market dummy variables. This effect is smaller in magnitude than the
one found by Han, Yang and Zhou (2013). Note, however, that they use daily equal-weighted returns which
could potentially explain the difference in the cross-sectional results between this study and their study.
5.3 Conditional Models with Macroeconomic Variables
Ferson and Schadt (1996) make a strong case for using predetermined variables in controlling for changes
in economic conditions while evaluating investment performance. I augment the four-factor Carhart (1997)
13
model with an intercept that is a linear function of a set of instruments as well as cross-products of the
instrumental variables with the market return to allow for state-dependent betas with the market factor.
I use investor sentiment due to Baker and Wurgler (2006), the aggregate liquidity factor of Pastor and
Stambaugh (2003), and the default spread of Moody’s BAA corporate bond yield over the AAA corporate
bond yield as the instrumental variables Zt in the following regression:
MAPjt,L = αj +βj,m rmkt,t +βj,s rsmb,t +βj,h rhml,t +βj,u rumd,t +βj,Z Zt−1 +γj,Z Zt−1 rmkt,t +ǫjt , j = 1, . . . , N,
(10)
Baker and Wurgler (2006) provide evidence that investor sentiment is associated with expected returns
and risks of the market. When investor sentiment is low, undervalued stocks are likely to be undervalued
more strongly than when investor sentiment is high. Similarly, overvalued stocks are likely to be less
overvalued when investor sentiment is low and more overvalued when investor sentiment is high. Next, I
present evidence regarding the exposure of the MAP returns to changes in investor sentiment.
Table 9 presents the results of the conditional model estimation. Changes in investor sentiment are
important both in increasing conditional alphas but also lead to higher betas with the market factor as
evidenced by the positive coefficient estimate of the cross-product variable ∆S × rm . Increases in the default
spread result in higher conditional alphas but lower conditional betas with the market. The evidence for the
aggregate liquidity factor is a little mixed and there appear to be some cross-sectional differences between
the various decile portfolio returns. However, all the unconditional alphas for all sets of portfolios are highly
statistically and economically significant. This suggests that investor sentiment, liquidity and especially the
default premium can account for the MAP abnormal returns, at least using this particular conditional
specification.
Finally, I put all the instrumental variables along with an NBER recession dummy variable in the same
regression with the four Carhart (1997) factors as well as interactions between the instrumental variables
and the market return. Table 6 presents the results from this conditional model specification. The previous
results vis-a-vis investor sentiment, the default spread, and liquidity largely hold with the same signs albeit
with a smaller degree of statistical significant. The recession indicator emerges as an important driver of
conditional market betas where for all sets of portfolios the interaction term RI × rm is always negative and
highly statistically significant. This suggests that for almost all portfolios betas with the market tend to be
14
significantly lower during economic recessions compared to their values during economic expansions.11
5.4 Simulations
In this subsection, I report the results from two sets of simulations. First, I draw 1000 random samples
designed to match the average historical return and the historical variance-covariance matrix of returns
for each set of portfolios under consideration. Then, I compare the MA versus BH performance for every
random sample and report the averages across all the simulations. Second, I draw randomly and without
replacement 1000 samples from the historical returns. Again, I compare the performance of the MA strategy
over the BH strategy for every bootstrapped sample and report the averages across all the simulations.
5.4.1 Randomly Generated Returns
Table 7 reports the average improvement in mean return and risk as well as the number of switches,
percentage of months the MA strategy is invested in the underlying portfolio, break-even transaction costs,
percentage of months the MA strategy return exceeds zero and the Treynor-Merton and Henriksson-Merton
market timing alphas across 1000 Monte Carlo simulations designed to match the first two moments of the
portfolio returns. Overall, the results are consistent with the results reported in previous sections regarding
the various sets of portfolios. There is a significant improvement in both risk and return when comparing
the moving average strategy over the buy-and-hold strategy. This improvement does not come at the cost of
a lot of trading as the number of switches is between 47 (BM decile 8) and 67 (Momentum decile Low) from
a total of 600 months in the entire sample period. The average break-even transaction costs are of similar
order of magnitude as reported previously and indicate that the MA strategy is superior to the BH strategy
for typical levels of proportional transaction costs available to both institutional and retail investors. Fully
up to 2 out of 3 months the MA strategy delivers a positive return as indicated by the average value of p1 ’s
reported in the table. Interestingly, virtually all of the market timing alphas are statistically significant.
This is an indicator that the simulated returns produce MA returns that are not entirely explained by
11 A further exercise involves including the MA indicator as an additional pre-specified instrumental variable as well as its
interaction with the market return. The parameter estimates for these extra conditioning variables are highly significant and
increase the goodness-of-fit of the predictive regressions. In the interest of saving space, these findings are not reported in the
paper but are instead available from the author upon request. I am very grateful to a referee’s suggestion for trying out this
particular predictive variable.
15
market timing.
What is apparent from these simulations is that what is needed for the superior performance of the MA
strategy over the BH strategy is a time period of sufficient length and a positive drift of the underlying
return. The randomly generated returns are completely independent of each other so all the autocorrelations
and cross-autocorrelations are statistically insignificant from zero. The superior performance of the MA
strategy does not appear to depend on any autocorrelation structure of the underlying returns.
Furthermore, the MA strategy appears to be antifragile in the sense of Taleb (2012). In other words,
simulating returns with larger volatility will tend to sample the extreme tails of the return distribution and
provide an even larger average return improvement. Clearly the improvement is diminishing as we increase
the number of Monte Carlo simulations since the underlying data generation process is iid Gaussian and,
thus, drawing extreme returns is highly unlikely.
5.4.2 Bootstrapped Returns
Table 8 reports average values across 1000 bootstrapped samples from the historical set of portfolio returns
during the same period under consideration used in previous sections. As a starting point, I draw without
replacement one monthly return12 at random from the same sample for every single month and decile
between 1960:01 and 2011:12. I run the moving average strategy and the buy-and-hold strategy for every
simulated sample path and report the average improvement in mean return and standard deviation of return
as well as the average number of switches, the average break-even transaction costs, percentage of positive
returns and the average market timing alphas. The results are broadly consistent with the Monte Carlo
simulation results reported previously as well as the decile portfolio results in Tables 3 and 4.
Note once again the robustness of the results of the performance of the MA strategy vis-a-vis the
BH strategy. Using a bootstrap window of one observation completely removes the autocorrelations from
the bootstrapped sample. The fact that we still observe on improvement over BH suggests that return
autocorrelations are not the main performance driver.

12 In further unreported results I use a block bootstrap with various lengths of the block. The results are not materially
different and are available from the author upon request.
16
5.5 Discussion
The large values of the risk-adjusted abnormal returns presented in the previous subsection demonstrate
the profitability of the MA switching strategy. This raises the question as to what ultimately drives of the
performance of the MA strategy. So far the evidence points towards a strategy that is contrarian, with a
focus on large-cap growth stocks and short the market. However, the goodness-of-fit statistics indicate that
this is at most only half the story. A more fundamental question that arises is how can this strategy survive
in competitive financial markets. A few potential reasons seem plausible.
First, there is ample evidence that stock returns are predictable at various frequencies at least to a
certain degree. This level of predictability is not perfect but is sufficient to improve forecasts of future stock
returns when stock return predictability is ignored. Some of the early evidence presented in Fama and
Schwert (1977) and Campbell (1987) as well as more recent work by Cochrane (2008) clearly demonstrates
that stock return predictability is an important feature that investors should ignore at their own peril.
Evidence regarding the performance of the moving average technical indicator is present in Brock,
Lakonishok and LeBaron (1992) in the context of predicting future moments of the Dow Jones Industrial
Average. Lo, Mamaysky and Wang (2000) provide further evidence using a wide range of technical indicators
with wide popularity among traders showing that this adds value even at the individual stock level over
and above the performance of a stock index. More recently, Neely, Rapach, Tu, and Zhou (2010) provide
evidence in favor of the usefulness of technical analysis in forecasting the stock market risk premium.
Second, early work on the performance of filter rules by Fama and Blume (1966), Jensen and Benington
(1970) concluded that such rules were dominated by buy and hold strategies especially after transaction
costs. Malkiel (1996) makes a forceful and memorable point against technical indicators: “Obviously, I’m
biased against the chartist. This is not only a personal predilection but a professional one as well. Technical
analysis is anathema to the academic world. We love to pick on it. Our bullying tactics are prompted by
two considerations: (1) after paying transaction costs, the method does not do better than a buy-and-hold
strategy for investors, and (2) it’s easy to pick on. And while it may seem a bit unfair to pick on such a
sorry target, just remember: It’s your money we are trying to save.” In a follow up on Brock et al (1992),
Bessembinder and Chan (1998) attribute the forecasting power of technical analysis to measurement errors
arising from non-synchronous trading. Ready (2002) goes even further and claims the results in Brock et al
17
(1992) are spurious and due to data snooping. Formal tests using White’s Reality Check are conducted in
Sullivan, Timmerman and White (1999) confirm that Brock et al (1992) results are robust to data snooping
and perform even better out of sample though there is evidence of time variation in performance across
subperiods. A more recent study using White’s Reality Check and Hansen’s SPA test is Hsu and Kuan
(2005) who find evidence of profitability of technical analysis using relatively “young” markets like the
NASDAQ Composite index and the Russell 2000 both in-sample and out-of-sample.
Furthermore, Treynor and Ferguson (1985) make a strong case in favor of investor’s learning and Bayesian
updating conditional on new information received rationally combining past prices can result in abnormal
profitability. Sweeney (1988) revisits Fama and Blume (1966) and finds that filter rules can be profitable to
floor traders in the 1970–1982 time period. Neftci (1991) presents a formal analysis of Wiener-Kolmogorov
prediction theory which provides optimal linear forecasts. He concludes that if the underlying price processes
are non-linear in nature then technical analysis rules might capture some useful information that is ignored
by the linear prediction rules. More involved and inherently non-linear rules are investigated in the context
of foreign currency exchange rates by Neely, Weller and Dittmar (1997) using a genetic programming
approach. Gencay (1998) goes even further in using non-linear predictors based on simple moving average
rules on the Dow Jones Industrial Average over a long time period between 1897 and 1988. In a similar
vein, Allen and Karjalainen (1999) use genetic algorithms to search for functions of past prices find that
can outperform a simple buy-and-hold strategy and report negative excess returns for most of the strategies
they consider.
Thirdly, it is entirely possible that market prices of financial assets can persistently deviate from funda-
mental values. Those fundamental values themselves are subject to incomplete information and, perhaps,
imperfect understanding of valuation tools as well as dispersion of beliefs and objective and behavioral
biases across the pool of traders and investors who regularly interact in financial markets. When investors’
information is incomplete and they learn continuously over time the true fundamental value, Zhu and Zhou
(2009) show theoretically that the moving average price is a useful state variable that aids in investors’
learning and improves their well-being and utility.
Behavioral and cognitive biases have been proposed in Daniel, Hirshleifer and Subrahmanyam (1998)
and Hong and Stein (1999), among others, as a potential driver of both price under- and over-reaction
in conjunction with the observed price continuation of stock prices. An alternative explanation for price
18
continuation was proposed in Zhang (2006). He argues that investors sub-optimally underweight newly
arriving public information leading to a persistent deviation of the market price from the fundamental
intrinsic value.
Note also that despite the apparent similarity of the MA switching strategy to the momentum strategy,
the four-factor alphas reported previously are statistically significant and of large magnitudes. This is
perhaps not surprising given that the payoff of the MA strategy resembles an at-the-money protective put
strategy. The non-linearity this induces makes the asset pricing task much more difficult when linear models
are used.
6 International Evidence
In this section, I investigate further the performance of the moving average strategy relative to the buy-
and-hold strategy using stock returns from Australia, Canada, France, Germany, Italy, Japan and UK. In
order to avoid the effects of exchange rate changes, I use local currency monthly returns for the entire stock
market of each of the countries I consider as well as portfolio returns sorted on book-to-market, earnings
yield, dividend yield and cash earnings to price ratio.
Table 9 reports the international evidence in favor of the moving average strategy. The findings are
qualitatively and quantitatively similar to the findings reported previously for US portfolios. The MA
strategy clearly outperforms the BH strategy and this outperformance is achieved with less risk. The MA
strategy has a very low trading intensity with between 14 (UK Low DP and Low CEP portfolios) and 48
(Australia Low EP portfolio) switches in a sample of 432 months. Furthermore, the break-even transaction
costs are large and well above realistic one-way transaction costs encountered in practice. Finally, the
outperformance is larger for growth portfolios than for value portfolios. This is consistent with the protective
put option explanation suggested previously since growth stocks are more volatile than value stocks.
For the sake of consistency, I also investigate the performance of the moving average strategy using sets
of six as well as 25 portfolios of international stocks sorted on size and momentum. For the sake of brevity
I do not report these findings here but they are largely consistent with the results reported previously for
19
other international portfolios as well as US portfolios.13 As expected, riskier portfolios like small-caps and
past losers tend to experience a larger return improvement via the moving average strategy compared to
portfolios consisting of large-caps and past winners. Furthermore, the improved performance does not arise
out of a large amount of trading. One notable difference is the slightly lower statistical significance of
∆µ. This is largely driven by the shorter sample of historical international portfolio returns. Nevertheless,
the findings for international portfolios sorted on both size and momentum are consistent with the finding
reported previously in this paper.
Finally, the value of the point estimate of the slope, measuring mean return improvement per unit BH
portfolio risk, is remarkably consistent across the two sets of international portfolios when all regions of the
world are considered jointly suggesting that a portfolio with 10% higher standard deviation will experience
an average return improvement of between 3% to 4% when switching to the MA strategy relative to the
BH strategy. This range of values is quite close to the value reported in (6) for US portfolios and slightly
higher than the value reported in (8) for US individual stocks.
7 Conclusion
In this paper, I report results for a simple moving average switching strategy applied to decile portfolios
sorted by size, book-to-market, and momentum. Further unreported findings for portfolios of stocks sorted
by various measures of yield, past returns and industry classification support the reported findings. There
is overwhelming evidence that the switching moving average strategy dominates in a mean-variance sense
buying and holding any of the decile portfolios. The excess returns of the switching moving average returns
over buying and holding the underlying portfolios are relatively insensitive to the four Carhart (1997) factors
and generate high statistically and economically significant alphas. In addition, abnormal returns for most
deciles decline substantially after controlling for investor sentiment, default, liquidity risks, recessions and
up/down markets. This switching strategy does not involve any heavy trading when implemented with
monthly returns and has very high break-even transaction costs, suggesting that it will be actionable even
for small investors. The findings are robust with respect to portfolio construction, various lag lengths of
the moving average, alternative sets of portfolios, international stock markets, individual stocks, randomly
13 I am grateful to an anonymous referee for making this suggestion. The extended findings for these portfolios of international
stocks are available from the author upon request.
20
generated stock returns and bootstrapped historical returns. Last but not least, the lagged signal indicating
whether the price has crossed the simple moving average has substantial predictive power over the subsequent
index return controlling for standard predictive variables like the default spread, investor sentiment, liquidity
risk and economic conditions. The risk-adjusted performance disappears only in the context of market timing
regressions in the framework of Henriksson-Merton (1981) where the downside market return is included
as an additional factor and empirical asset pricing models with macroeconomic state variables. Hence, it
appears that the success of the moving average strategy does not represent an anomaly and is consistent
with rational asset pricing. In addition, any abnormal returns surviving the previously mentioned tests may
not be actionable in practice due to limits to arbitrage and price impact of trading on illiquid risky assets
with low trading volumes.14
Further work would be necessary to investigate the potential link between the returns of the MA switching
strategy and the payoffs of protective put options on the underlying asset. A more aggressive implementation
will involve selling short the underlying asset in response to a signal to switch instead of shifting the funds
into cash. I conjecture that the payoff of this version of the MA strategy resembles an imperfect at-the-
money straddle. It would also be of use to test more formally whether higher moments like skewness and
kurtosis are improved by the MA strategy over the BH strategy. One potential alternative is to combine
all first four moments using a utility function over them and convert the gains into certainty equivalent
utility gains. Comparing those gain to the break-even transaction costs will provide further evidence into
the superiority of the MA switching strategy.
Considering the vast literature on technical analysis and the numerous technical indicators following by
some traders in practice, this study is just a first step towards investigating the performance and imple-
mentation of one common technical indicator. Future work will determine which other technical indicators
perform well and whether they produce significant abnormal returns over and above the relevant transaction
costs.
14 A variant of the moving strategy using stock futures and interest rate futures (instead of trading the stock and the risk-free
asset) could address this point in practice. I leave the study of this version of the moving average for future investigation.
21
References
[1] Allen, F., Karjalainen, R., 1999, “Using Genetic Algorithms to Find Technical Trading Rules,” Journal
of Financial Economics 51, 245–272.
[2] Baker, M., Wurgler, J., 2006, “Inevestor Sentiment and the Cross-Section of Stock Returns,” Journal
of Finance 61, 259–299.
[3] Baker, M., Wurgler, J., 2007, “Investor Sentiment in the Stock Market,” Journal of Economic Perspec-
tives 21, 129–151.
[4] Balduzzi, P., Lynch A. W., 1999, “Transaction Costs and Predictability: Some Utility Cost Calcula-
tions,” Journal of Financial Economics 52, 47–78.
[5] Barberis, N., Schleifer, A., Vishny, R., 1998, “A Model of Investor Sentiment,” Journal of Financial
Economics 49, 307–343.
[6] Bessembinder, H., Chan, K., 1998, “Market Efficiency and the Returns to Technical Analysis,” Finan-
cial Management 27, 5–17.
[7] Black, F., Scholes, M., 1973, “The Pricing of Options and Corporate Liabilities,” Journal of Political
Economy 81, 637–654.
[8] Blume, L., Easley, D., O’Hara, M., 1994, “Market Statistics and Technical Analysis: The Role of
Volume,” Journal of Finance 49, 153–181.
[9] Brock, W., Lakonishok, J., LeBaron, B., 1992, “Simple Technical Trading Rules and the Stochastic
Properties of Stock Returns,” Journal of Finance 47, 1731-1764.
[10] Brown, D. P., Jennings, R. H., 1989, “On Technical Analysis,” The Review of Financial Studies 2,
527–551.
[11] Campbell, J. Y., 1987, “Stock Returns and the Term Structure,” Journal of Financial Economics 18,
373–399.
[12] Carhart, M. M., 1997, “On Persistence in Mutual Fund Performance,” Journal of Finance 52, 57–82.
22
[13] Cochrane, J. H., 2008, “The Dog that Did Not Bark: A Defense of Return Predictability,” Review of
Financial Studies 21, 1533–1575.
[14] Daniel, K., Hirshleifer, D., Subrahmanyam, A., 1998, “Investor Psychology and Security Market Under-
and Overreactions,” Journal of Finance 53, 1839–1885.
[15] Faber, M. T., 2007, “A Quantitative Approach to Tactical Asset Allocation,” Jouranl fo Wealth Man-
agement 9, 69–79.
[16] Fama, E. F., Blume, M. E., 1966, “Filter Rules and Stock-Market Trading,” Journal of Business 39,
226–241.
[17] Fama, E. F., French, K. R., 1992, “The Cross-Section of Expected Stock Returns,” Journal of Finance
47, 427–465.
[18] Fama, E. F., Schwert, W., 1977, “Asset Returns and Inflation,” Journal of Financial Economics 5,
115–146.
[19] Ferson, W. E., Schadt, R. W., 1996, “Measuring Fund Strategy and Performance in Changing Economic
Conditions,” Journal of Finance 51, 425–461.
[20] Gencay, R., 1998, “The Predictability of Security Returns with Simple Technical Trading Rules,”
Journal of Empirical Finance 5, 347–359.
[21] Goh, J., Jiang, F., Tu, J., Zhou, G., 2012, “Forecasting Government Bond Risk Premia Using Technical
Indicators,” Working paper.
[22] Han, Y., 2006, “Asset Allocation with a High Dimensional Latent Factor Stochastic Volatility Model,”
Review of Financial Studies 19, 237–271.
[23] Han, Y., Yang, K., Zhou, G., 2013, “A New Anomaly: The Cross-Sectional Profitability of Technical
Analysis,” Journal of Financial and Quantitative Analysis 48(5), 1433–1461.
[24] Han, Y., Zhou, G., 2013, “Trend Factor: A New Determinant of Cross-Section Stock Returns,” Working
paper.
[25] Henriksson, R. D., Merton, R. C., 1981, “On Market Timing and Investment Performance. II. Statistical
Procedures for Evaluating Forecasting Skills,” Journal of Business 54, 513–533.
23
[26] Hong, H., Stein, J. C., 1999, “A Unified Theory of Underreaction, Momentum Trading, and Overreation
in Asset Markets,” Journal of Finance 54, 2143–2184.
[27] Hsu, P.-H., Kuan, C.-M., 2005, “Reexamining the Profitability of Technical Analysis with Data Snoop-
ing Checks,” Journal of Financial Econometrics 3, 606–628.
[28] Huang, D., Zhou, G., 2013, “Economic and Market Conditions: Two State Variables that Predict the
Stock Market,” Working paper.
[29] Jensen, M. C., Benington, G. A., 1970, “Random Walks and Technical Theories: Some Additional
Evidence,” Journal of Finance 25, 469–482.
[30] Jiang, F., 2013, “Trend-Based Conditional Asset Pricing: Explaining the Cross-Section of Technical
Analysis Profitability,” Working paper.
[31] Kilgallen, T., 2012, “Testing the Simple Moving Average across Commodities, Global Stock Indices,
and Currencies,” Journal of Wealth Management 15, 82–100.
[32] Lo, A., Mamaysky, H., Wang, J., 2000, “Foundations of Technical Analysis: Computational Algorithms,
Statistical Inference, and Empirical Implementation,” Journal of Finance 55, 1705–1765.
[33] Lynch, A. W., Balduzzi, P., 2000, “Predictability and Transaction Costs: The Impact on Rebalancing
Rules and Behavior,” Journal of Finance 66, 2285–2309.
[34] Malkiel, B. G., 1996, A Random Walk Down Wall Street, W. W. Norton & Company, Inc., New York.
[35] Merton, R. C., 1981, “On Market Timing and Investment Performance. I. An Equilibrium Theory of
Value for Market Forecasts,” Journal of Business 54, 363–406.
[36] Neely, C. J., Rapach, D. E., Tu., J., Zhou, G., 2010, “Out-of-Sample Equity Premium Prediction:
Fundamental vs. Technical Analysis,” Unpublished working paper, Washington University in St. Louis.
[37] Neely, C. J., Rapach, D. E., Tu., J., Zhou, G., 2011, “Forecasting the Equity Risk Premium: The Role
of Technical Indicators,” Unpublished working paper, Federal Reserve Bank of St. Louis.
[38] Neely, C., Weber, P., Dittmar, R., 1997, “Is Technical Analysis in the Foreign Exchange Market
Profitable? A Genetic Programming Approach,” Journal of Financial and Quantitative Analysis 32,
405–426.
24
[39] Neftci, S. N., 1991, “Naive Trading Rules in Financial Markets and Wiener-Kolmogorov Prediction
Theory: A Study of “Technical Analysis”,” Journal of Business 64, 549–571.
[40] Newey, W. K., West, K. D., 1987, “A Simple, Positive Semi-Definite, Heteroscedasticity and Autocor-
relation Consistent Covariance Matrix,” Econometrica 55, 703–708.
[41] Pastor, L., Stambaugh, R., 2003, “Liquidity Risk and Expected Stock Returns,” Journal of Political
Economy 111, 642–685.
[42] Ready, M. J., 2002, “Profits from Technical Trading Rules,” Financial Management 31, 43–61.
[43] Sharpe, W. F., 1964, “Capital asset prices: A theory of market equilibrium under conditions of risk,”
Journal of Finance 19, 425–442.
[44] Sullivan, R., Timmerman, A., White, H., 1999, “Data-Snooping, Technical Trading Rule Performance,
and the Bootstrap,” Journal of Finance 54, 1647–1692.
[45] Sweeney, R. J., 1988, “Some New Filter Rule Tests: Methods and Results,” Journal of Financial and
Quantitative Analysis 23, 285–300.
[46] Taleb, N. N., 2012, Antifragile Things that Gain from Disorder, Random House, Inc., New York.
[47] Treynor, J. L., Ferguson, R., 1985, “In Defense of Technical Analysis,” Journal of Finance 40, 757–773.
[48] Treynor, J. L., Mazuy, K., 1966, “Can Mutual Funds Outguess the Market?”, Harvard Business Review
44, 131–136.
[49] Zhang, F. X., 2006, “Information Uncertainty and Stock Returns,” Journal of Finance 61, 105–136.
[50] Zhu, Y., Zhou, G., 2009, “Technical Analysis: An Asset Allocation Perspective on the Use of Moving
Averages,” Journal of Financial Economics 92, 519–544.
25
Table 1. Summary Statistics.
This table reports summary statistics for the respective buy and hold (BH) portfolio returns, the moving
average (MA) switching strategy portfolio returns and the excess return of MA over BH (MAP) using sets
of 10 portfolios sorted by size, book-to-market and momentum. The sample period covers 1960:01 until
2011:12 with value-weighted portfolio returns. µ is the annualized average return, σ is annualized standard
deviation of returns, s is the annualized skewness, and SR is the annualized Sharpe ratio. The length of
the moving average window is 24 months. A one-way transaction cost of 0.5% has been imposed in the
computation of the MA and MAP returns.
Panel A: Size sorted portfolios.

Portfolio µ σ s SR µ σ s SR µ σ s SR
BH Portfolios MA Portfolios MAP Portfolios
Low 13.57 22.44 -0.14 0.38 17.94 17.04 0.35 0.75 4.37 14.09 0.71 0.31
2 12.92 22.28 -0.22 0.35 17.92 16.84 0.27 0.76 5.00 13.99 0.90 0.36
3 13.62 21.29 -0.41 0.40 17.62 16.25 0.03 0.77 4.00 13.24 1.21 0.30
4 13.01 20.52 -0.46 0.38 17.70 15.17 0.07 0.83 4.69 13.24 1.30 0.35
5 13.30 19.80 -0.48 0.41 17.48 14.66 -0.08 0.84 4.18 12.76 1.07 0.33
6 12.47 18.58 -0.49 0.40 16.59 13.73 0.08 0.84 4.11 11.99 1.38 0.34
7 12.50 18.26 -0.45 0.40 16.41 13.49 0.17 0.84 3.91 11.83 1.32 0.33
8 11.90 17.79 -0.43 0.38 15.91 13.24 0.14 0.82 4.01 11.36 1.34 0.35
9 11.19 16.35 -0.40 0.37 15.17 12.04 0.21 0.84 3.98 10.54 1.30 0.38
High 9.47 14.99 -0.31 0.29 13.11 11.70 -0.19 0.68 3.64 8.92 0.76 0.41
High−Low -4.10 16.95 -0.74 -0.54 -4.83 16.35 -0.71 -0.61 -0.73 12.24 -0.66 -0.06
Panel B: Book-to-market sorted portfolios.

Low 9.13 18.13 -0.19 0.22 14.36 13.28 0.27 0.70 5.23 11.88 0.83 0.44
2 10.21 16.63 -0.43 0.31 14.38 12.67 0.12 0.73 4.17 10.28 1.65 0.41
3 10.95 16.26 -0.45 0.36 15.29 12.39 0.07 0.82 4.34 9.92 1.73 0.44
4 10.89 16.66 -0.46 0.35 14.80 12.14 0.37 0.80 3.91 10.93 1.58 0.36
5 10.76 15.66 -0.39 0.36 14.39 11.84 0.26 0.78 3.63 9.77 1.55 0.37
6 11.68 15.94 -0.40 0.41 15.26 12.17 0.31 0.83 3.58 9.77 1.71 0.37
7 12.31 15.53 -0.08 0.46 15.81 12.76 0.30 0.84 3.50 8.27 1.27 0.42
8 12.97 16.02 -0.44 0.49 15.47 12.42 0.14 0.83 2.50 9.70 1.43 0.26
9 13.93 16.90 -0.29 0.52 17.99 13.02 0.21 0.99 4.06 10.06 1.06 0.40
High 15.29 20.56 0.08 0.49 19.33 16.03 0.55 0.89 4.04 12.29 0.18 0.33
High−Low 6.16 16.17 0.54 0.06 4.97 15.13 0.72 -0.01 -1.19 11.45 -0.58 -0.10
Panel C: Momentum sorted portfolios.

Low 1.23 28.02 0.67 -0.14 11.90 13.53 1.10 0.50 10.68 24.14 -1.00 0.44
2 7.82 21.88 0.24 0.12 13.39 14.19 0.62 0.58 5.57 16.26 -0.43 0.34
3 9.30 18.80 0.33 0.22 13.94 12.33 0.59 0.72 4.64 13.82 -0.79 0.34
4 9.97 16.93 -0.11 0.29 13.77 11.62 0.53 0.74 3.80 11.92 0.34 0.32
5 8.96 15.69 -0.25 0.25 12.89 10.99 0.42 0.71 3.93 10.72 0.83 0.37
6 10.14 15.92 -0.36 0.32 14.27 11.64 0.55 0.79 4.13 10.38 1.62 0.40
7 10.41 15.42 -0.48 0.34 14.23 12.00 0.12 0.76 3.82 9.17 2.14 0.42
8 12.60 15.77 -0.29 0.47 15.25 13.12 -0.08 0.77 2.65 8.31 1.11 0.32
9 13.21 17.05 -0.52 0.47 16.40 14.32 -0.10 0.79 3.19 8.72 2.90 0.37
High 17.62 21.78 -0.39 0.57 21.58 18.69 -0.07 0.88 3.96 10.46 2.57 0.38
High−Low 16.39 24.17 -1.52 0.47 9.68 17.97 -0.41 0.25 -6.72 20.90 1.90 -0.32
26
Table 2. Factor Regressions Results.
This table reports alphas, betas, and adjusted R2 of the regressions of the MAP excess returns on the Carhart
four-factors using portfolios sorted by size, book-to-market and momentum. The alphas are annualized and
in percent. The sample period covers 1960:01 until 2011:12 with value-weighted portfolio returns. The
length of the moving average window is 24 months. A one-way transaction cost of 0.5% has been imposed
in constructing the switching moving average strategy excess returns. Newey and West (1987) standard
errors with 24 lags are used in reporting statistical significance of a two-sided null hypothesis at the 1%,
5%, and 10% level is given by a ∗∗∗ , a ∗∗ , and a ∗ , respectively.

Portfolio α βm βs βh βu R̄2
Low 6.04∗∗∗ -0.46∗∗∗ -0.36∗∗∗ -0.13∗∗∗ 0.27∗∗∗ 53.01
2 6.94∗∗∗ -0.48∗∗∗ -0.33∗∗∗ -0.19∗∗∗ 0.27∗∗∗ 54.23
3 5.41∗∗∗ -0.46∗∗∗ -0.27∗∗∗ -0.11∗∗∗ 0.25∗∗∗ 51.32
4 6.36∗∗∗ -0.47∗∗∗ -0.27∗∗∗ -0.12∗∗∗ 0.24∗∗∗ 52.80
5 4.99∗∗∗ -0.46∗∗∗ -0.21∗∗∗ -0.05 0.27∗∗∗ 54.53
6 5.10∗∗∗ -0.44∗∗∗ -0.17∗∗∗ -0.03 0.22∗∗∗ 52.86
7 5.22∗∗∗ -0.45∗∗∗ -0.14∗∗∗ -0.09∗∗∗ 0.20∗∗∗ 50.85
8 5.20∗∗∗ -0.44∗∗∗ -0.10∗∗∗ -0.08∗∗∗ 0.20∗∗∗ 51.11
9 5.17∗∗∗ -0.43∗∗∗ -0.05∗∗ -0.07∗∗ 0.17∗∗∗ 50.58
High 3.35∗∗∗ -0.33∗∗∗ 0.07∗∗∗ 0.06∗∗∗ 0.17∗∗∗ 46.29
High−Low 2.69∗∗∗ -0.13∗∗∗ -0.43∗∗∗ -0.19∗∗∗ 0.10∗∗∗ 21.83

Low 4.49∗∗∗ -0.45∗∗∗ 0.07∗∗∗ 0.21∗∗∗ 0.21∗∗∗ 51.97
2 4.88∗∗∗ -0.38∗∗∗ -0.03 0.02 0.14∗∗∗ 42.86
3 5.52∗∗∗ -0.39∗∗∗ -0.06∗∗∗ -0.02 0.12∗∗∗ 46.44
4 5.87∗∗∗ -0.45∗∗∗ -0.05∗∗ -0.16∗∗∗ 0.14∗∗∗ 49.35
5 5.22∗∗∗ -0.39∗∗∗ -0.03 -0.14∗∗∗ 0.13∗∗∗ 44.62
6 5.13∗∗∗ -0.38∗∗∗ -0.06∗∗ -0.14∗∗∗ 0.14∗∗∗ 45.83
7 4.41∗∗∗ -0.29∗∗∗ 0.03 -0.13∗∗∗ 0.13∗∗∗ 37.14
8 4.59∗∗∗ -0.35∗∗∗ -0.06∗∗ -0.27∗∗∗ 0.14∗∗∗ 42.04
9 5.75∗∗∗ -0.37∗∗∗ -0.08∗∗∗ -0.21∗∗∗ 0.17∗∗∗ 45.86
High 5.83∗∗∗ -0.42∗∗∗ -0.12∗∗∗ -0.23∗∗∗ 0.20∗∗∗ 40.56
High−Low -1.34 -0.04 0.19∗∗∗ 0.44∗∗∗ 0.00 15.23

Low 7.44∗∗∗ -0.84∗∗∗ -0.26∗∗∗ 0.20∗∗∗ 0.84∗∗∗ 71.75
2 4.75∗∗∗ -0.56∗∗∗ -0.11∗∗∗ -0.08∗ 0.51∗∗∗ 60.02
3 3.13∗∗ -0.45∗∗∗ 0.00 -0.02 0.45∗∗∗ 56.64
4 4.18∗∗∗ -0.45∗∗∗ -0.00 -0.08∗∗∗ 0.27∗∗∗ 50.88
5 5.09∗∗∗ -0.43∗∗∗ -0.00 -0.11∗∗∗ 0.18∗∗∗ 48.14
6 5.25∗∗∗ -0.42∗∗∗ -0.02 -0.05∗ 0.15∗∗∗ 48.84
7 5.19∗∗∗ -0.36∗∗∗ 0.01 -0.04∗ 0.07∗∗∗ 39.95
8 4.01∗∗∗ -0.32∗∗∗ 0.00 -0.06∗∗∗ 0.07∗∗∗ 39.01
9 5.28∗∗∗ -0.34∗∗∗ 0.01 -0.08∗∗∗ -0.00 34.56
High 5.95∗∗∗ -0.39∗∗∗ -0.01 -0.01 0.01 34.30
High−Low 1.49 -0.45∗∗∗ -0.26∗∗∗ 0.21∗∗∗ 0.83∗∗∗ 57.68
27
Table 3. Trading Frequency and Break-Even Transaction Cost.
This table reports the results for the improvement delivered by the MA switching strategy over the buy-and-
hold strategy, the trading frequency as well as the break-even transaction cost using ten decile portfolios
sorted by size, book-to-market and momentum. The sample period covers 1960:01 until 2011:12 with value-
weighted portfolio returns. ∆µ is the annualized improvement in the average in-sample monthly return,
∆σ is the annualized improvement in the return standard deviation, pA is the proportion of months during
which there is a hold signal, NT is the number of transactions (buy or sell) over the entire sample period,
BETC is the break-even one-sided transaction cost in percent, p1 is the proportion of months during which
a buy signal was followed by a positive return of the underlying portfolio and p2 is the proportion of months
during which a buy signal was followed by a portfolio return in excess of the risk-free rate. The length of the
moving average window is 24 months. A one-way transaction cost of 0.5% has been imposed in the reported
∆µ and ∆σ. Statistical significance of the one-sided null hypotheses that ∆µ > 0, ∆σ > 0, p1 > 0.5 and
p2 > 0.5 at the 1%, 5%, and 10% level is given by a ∗∗∗ , a ∗∗ , and a ∗ , respectively.

Portfolio ∆µ ∆σ pA NT BETC p1 p2
Low 4.37∗∗ 5.40∗∗∗ 0.74 39 5.60 0.60∗∗∗ 0.57∗∗∗
2 5.00∗∗∗ 5.44∗∗∗ 0.77 46 5.43 0.57∗∗∗ 0.55∗∗∗
3 4.00∗∗ 5.05∗∗∗ 0.80 44 4.55 0.59∗∗∗ 0.57∗∗∗
4 4.69∗∗∗ 5.35∗∗∗ 0.79 50 4.69 0.59∗∗∗ 0.56∗∗∗
5 4.18∗∗ 5.14∗∗∗ 0.80 44 4.75 0.61∗∗∗ 0.57∗∗∗
6 4.11∗∗∗ 4.85∗∗∗ 0.80 38 5.41 0.60∗∗∗ 0.56∗∗∗
7 3.91∗∗∗ 4.77∗∗∗ 0.81 32 6.11 0.60∗∗∗ 0.57∗∗∗
8 4.01∗∗∗ 4.55∗∗∗ 0.81 42 4.77 0.60∗∗∗ 0.56∗∗∗
9 3.98∗∗∗ 4.31∗∗∗ 0.81 42 4.74 0.59∗∗∗ 0.57∗∗∗
High 3.64∗∗∗ 3.29∗∗∗ 0.80 38 4.79 0.61∗∗∗ 0.57∗∗∗

Low 5.23∗∗∗ 4.86∗∗∗ 0.72 42 6.22 0.57∗∗∗ 0.54∗∗
2 4.17∗∗∗ 3.97∗∗∗ 0.81 44 4.74 0.58∗∗∗ 0.56∗∗∗
3 4.34∗∗∗ 3.86∗∗∗ 0.79 50 4.34 0.60∗∗∗ 0.56∗∗∗
4 3.91∗∗∗ 4.52∗∗∗ 0.80 40 4.88 0.60∗∗∗ 0.58∗∗∗
5 3.63∗∗∗ 3.82∗∗∗ 0.82 30 6.04 0.62∗∗∗ 0.58∗∗∗
6 3.58∗∗∗ 3.78∗∗∗ 0.84 38 4.71 0.61∗∗∗ 0.56∗∗∗
7 3.50∗∗∗ 2.78∗∗∗ 0.84 34 5.15 0.62∗∗∗ 0.58∗∗∗
8 2.50∗∗ 3.60∗∗∗ 0.86 36 3.47 0.62∗∗∗ 0.58∗∗∗
9 4.06∗∗∗ 3.88∗∗∗ 0.84 44 4.61 0.63∗∗∗ 0.59∗∗∗
High 4.04∗∗ 4.53∗∗∗ 0.83 29 6.96 0.62∗∗∗ 0.58∗∗∗

Low 10.68∗∗∗ 14.49∗∗∗ 0.58 69 7.74 0.50 0.46∗∗
2 5.57∗∗∗ 7.69∗∗∗ 0.76 44 6.33 0.55∗∗ 0.51
3 4.64∗∗∗ 6.48∗∗∗ 0.76 44 5.27 0.56∗∗∗ 0.52
4 3.80∗∗ 5.31∗∗∗ 0.80 42 4.52 0.57∗∗∗ 0.53∗
5 3.93∗∗∗ 4.70∗∗∗ 0.80 52 3.78 0.58∗∗∗ 0.55∗∗∗
6 4.13∗∗∗ 4.28∗∗∗ 0.78 38 5.43 0.58∗∗∗ 0.56∗∗∗
7 3.82∗∗∗ 3.42∗∗∗ 0.80 44 4.34 0.61∗∗∗ 0.57∗∗∗
8 2.65∗∗ 2.65∗∗∗ 0.81 38 3.49 0.62∗∗∗ 0.58∗∗∗
9 3.19∗∗∗ 2.73∗∗∗ 0.82 32 4.98 0.63∗∗∗ 0.60∗∗∗
High 3.96∗∗∗ 3.09∗∗∗ 0.81 41 4.83 0.63∗∗∗ 0.60∗∗∗
28
Table 4. Market Timing Regressions: Monthly Decile Portfolios.
This table reports alphas, betas, and adjusted R2 of the market timing regressions of the MAP excess
returns on the market factor using portfolios sorted by size, book-to-market and momentum. The TM
panel reports the results using the Treynor and Mazuy (1966) quadratic regression with the squared market
factor (βm2 ) while the HM panel reports the results using the Henriksson and Merton (1981) regression with
option-like returns on the market (γm ). The sample period covers 1960:01 until 2011:12 with value-weighted
portfolio returns. The length of the moving average window is 24 months. A one-way transaction cost of
0.5% has been imposed in constructing the switching moving average strategy excess returns. Newey and
West (1987) standard errors with 24 lags are used in reporting statistical significance of a two-sided null
hypothesis at the 1%, 5%, and 10% level is given by a ∗∗∗ , a ∗∗ , and a ∗ , respectively.

Portfolio α βm βm2 R̄2 α βm γm R̄2
TM HM
Low 2.19 -0.52∗∗∗ 0.02∗∗∗ 41.20 -1.67 -0.75∗∗∗ 0.41∗∗∗ 39.97
2 2.53∗ -0.52∗∗∗ 0.02∗∗∗ 42.50 -1.86 -0.77∗∗∗ 0.45∗∗∗ 41.27
3 0.89 -0.49∗∗∗ 0.02∗∗∗ 44.08 -4.09∗∗∗ -0.77∗∗∗ 0.50∗∗∗ 42.54
4 1.50 -0.50∗∗∗ 0.02∗∗∗ 46.46 -3.97∗∗ -0.79∗∗∗ 0.54∗∗∗ 45.11
5 1.37 -0.49∗∗∗ 0.02∗∗∗ 47.19 -3.33∗∗ -0.75∗∗∗ 0.48∗∗∗ 45.67
6 1.08 -0.47∗∗∗ 0.02∗∗∗ 49.83 -4.32∗∗∗ -0.75∗∗∗ 0.52∗∗∗ 48.54
7 1.12 -0.46∗∗∗ 0.02∗∗∗ 47.78 -3.81∗∗ -0.72∗∗∗ 0.48∗∗∗ 46.49
8 1.43 -0.44∗∗∗ 0.02∗∗∗ 48.24 -3.47∗∗∗ -0.69∗∗∗ 0.46∗∗∗ 47.19
9 1.59∗ -0.42∗∗∗ 0.02∗∗∗ 49.68 -3.09∗∗ -0.65∗∗∗ 0.44∗∗∗ 48.71
High 4.38∗∗∗ -0.34∗∗∗ 0.00∗∗∗ 38.59 0.19 -0.47∗∗∗ 0.25∗∗∗ 40.16
High−Low -2.20 -0.18∗∗∗ 0.02∗∗∗ 9.20 -1.86 -0.28∗∗∗ 0.17∗∗ 6.82

TM HM
Low 4.42∗∗∗ -0.48∗∗∗ 0.01∗∗∗ 46.44 -0.10 -0.68∗∗∗ 0.37∗∗∗ 46.72
2 1.76∗ -0.38∗∗∗ 0.02∗∗∗ 44.28 -2.81∗∗ -0.61∗∗∗ 0.43∗∗∗ 43.40
3 2.13∗∗∗ -0.39∗∗∗ 0.02∗∗∗ 48.20 -2.78∗∗ -0.62∗∗∗ 0.43∗∗∗ 47.84
4 1.13 -0.43∗∗∗ 0.02∗∗∗ 48.84 -3.51∗∗ -0.67∗∗∗ 0.46∗∗∗ 47.33
5 1.32 -0.36∗∗∗ 0.02∗∗∗ 43.61 -2.33∗ -0.56∗∗∗ 0.37∗∗∗ 42.05
6 0.55 -0.36∗∗∗ 0.02∗∗∗ 45.41 -3.56∗∗∗ -0.59∗∗∗ 0.43∗∗∗ 43.11
7 3.75∗∗∗ -0.27∗∗∗ 0.00∗∗ 28.24 0.29 -0.38∗∗∗ 0.22∗∗∗ 29.46
8 0.78 -0.31∗∗∗ 0.01∗∗∗ 32.00 -2.30∗ -0.48∗∗∗ 0.30∗∗∗ 31.11
9 1.36∗ -0.35∗∗∗ 0.02∗∗∗ 40.03 -3.57∗∗∗ -0.59∗∗∗ 0.45∗∗∗ 39.30
High 1.46 -0.40∗∗∗ 0.02∗∗∗ 33.52 -2.73∗ -0.62∗∗∗ 0.42∗∗∗ 32.38
High−Low 2.97∗ -0.09∗∗∗ -0.01∗ 1.22 2.62 -0.05 -0.05 0.86

TM HM
Low 14.80∗∗∗ -1.03∗∗∗ 0.00 45.82 12.85∗∗∗ -1.11∗∗∗ 0.14 45.85
2 8.80∗∗∗ -0.64∗∗∗ -0.00 37.83 5.66∗∗∗ -0.71∗∗∗ 0.15∗∗ 38.04
3 7.35∗∗∗ -0.51∗∗∗ -0.00 33.38 4.22∗∗∗ -0.57∗∗∗ 0.14∗∗∗ 33.64
4 2.49∗∗ -0.45∗∗∗ 0.01∗∗∗ 41.47 -1.72 -0.65∗∗∗ 0.37∗∗∗ 41.29
5 1.51 -0.40∗∗∗ 0.02∗∗∗ 45.18 -3.17∗∗ -0.64∗∗∗ 0.44∗∗∗ 44.30
6 0.71 -0.40∗∗∗ 0.02∗∗∗ 51.94 -4.89∗∗∗ -0.69∗∗∗ 0.53∗∗∗ 50.39
7 1.24 -0.34∗∗∗ 0.02∗∗∗ 44.76 -3.63∗∗∗ -0.57∗∗∗ 0.44∗∗∗ 44.12
8 1.40∗∗ -0.30∗∗∗ 0.01∗∗∗ 40.19 -1.65 -0.45∗∗∗ 0.28∗∗∗ 39.74
9 -0.10 -0.30∗∗∗ 0.02∗∗∗ 42.90 -4.51∗∗∗ -0.53∗∗∗ 0.44∗∗∗ 40.51
High 1.63∗∗ -0.37∗∗∗ 0.02∗∗∗ 39.19 -3.23∗∗∗ -0.60∗∗∗ 0.43∗∗∗ 38.83
High−Low 13.17∗∗∗ -0.66∗∗∗ -0.01∗∗∗ 24.23 16.08∗∗∗ -0.51∗∗∗ -0.29∗∗∗ 24.06
29
Table 5. Factor Regressions with Business Cycles and Up Markets: Monthly Decile Portfolios.
This table reports alphas, betas, and adjusted R2 of the factor regressions of the MAP excess returns using the Carhart four-factor model with NBER recession
indicator dummy variable (RI) and up market indicators (UP) using portfolios sorted by size, book-to-market and momentum. Alphas are annualized and in
percent. The sample period covers 1960:01 until 2011:12 with value-weighted portfolio returns. The length of the moving average window is 24 months. A
one-way transaction cost of 0.5% has been imposed in constructing the switching moving average strategy excess returns. Newey and West (1987) standard
errors with 24 lags are used in reporting statistical significance of a two-sided null hypothesis at the 1%, 5%, and 10% level is given by a ∗∗∗ , a ∗∗ , and a ∗ ,
respectively.
Portfolio α βm βs βh βu RI R̄2 α βm βs βh βu UP R̄2
Recession Dummy Up Market Dummy
Low 4.95∗∗∗ -0.46∗∗∗ -0.36∗∗∗ -0.13∗∗∗ 0.27∗∗∗ 0.58∗∗ 53.18 -6.11∗∗∗ -0.62∗∗∗ -0.37∗∗∗ -0.13∗∗∗ 0.26∗∗∗ 1.88∗∗∗ 54.97
2 5.94∗∗∗ -0.48∗∗∗ -0.33∗∗∗ -0.18∗∗∗ 0.27∗∗∗ 0.53∗∗ 54.36 -5.47∗∗∗ -0.64∗∗∗ -0.34∗∗∗ -0.19∗∗∗ 0.26∗∗∗ 1.92∗∗∗ 56.32
3 4.51∗∗∗ -0.45∗∗∗ -0.27∗∗∗ -0.11∗∗∗ 0.25∗∗∗ 0.48∗∗ 51.43 -8.20∗∗∗ -0.63∗∗∗ -0.27∗∗∗ -0.11∗∗∗ 0.24∗∗∗ 2.10∗∗∗ 54.17
4 5.63∗∗∗ -0.47∗∗∗ -0.27∗∗∗ -0.12∗∗∗ 0.24∗∗∗ 0.39∗∗ 52.85 -6.57∗∗∗ -0.64∗∗∗ -0.27∗∗∗ -0.12∗∗∗ 0.22∗∗∗ 2.00∗∗∗ 55.37
5 3.85∗∗∗ -0.45∗∗∗ -0.21∗∗∗ -0.05∗ 0.27∗∗∗ 0.60∗∗∗ 54.79 -7.80∗∗∗ -0.62∗∗∗ -0.22∗∗∗ -0.05∗ 0.26∗∗∗ 1.98∗∗∗ 57.25
6 4.26∗∗∗ -0.44∗∗∗ -0.17∗∗∗ -0.03 0.22∗∗∗ 0.44∗∗∗ 52.99 -7.52∗∗∗ -0.61∗∗∗ -0.18∗∗∗ -0.03 0.21∗∗∗ 1.95∗∗∗ 55.86
7 4.12∗∗∗ -0.44∗∗∗ -0.14∗∗∗ -0.08∗∗∗ 0.21∗∗∗ 0.58∗∗∗ 51.13 -7.55∗∗∗ -0.61∗∗∗ -0.15∗∗∗ -0.08∗∗∗ 0.19∗∗∗ 1.97∗∗∗ 54.01
8 4.18∗∗∗ -0.44∗∗∗ -0.10∗∗∗ -0.08∗∗∗ 0.20∗∗∗ 0.53∗∗∗ 51.36 -6.57∗∗∗ -0.59∗∗∗ -0.10∗∗∗ -0.08∗∗∗ 0.19∗∗∗ 1.82∗∗∗ 54.01
9 4.52∗∗∗ -0.42∗∗∗ -0.05∗∗ -0.07∗∗∗ 0.17∗∗∗ 0.35∗ 50.66 -6.40∗∗∗ -0.57∗∗∗ -0.06∗∗ -0.07∗∗∗ 0.16∗∗∗ 1.79∗∗∗ 53.86
High 2.35∗∗∗ -0.33∗∗∗ 0.07∗∗∗ 0.06∗∗∗ 0.17∗∗∗ 0.53∗∗∗ 46.73 -2.80∗∗ -0.41∗∗∗ 0.07∗∗∗ 0.06∗∗∗ 0.16∗∗∗ 0.95∗∗∗ 47.47
High−Low 2.60∗∗ -0.13∗∗∗ -0.43∗∗∗ -0.19∗∗∗ 0.10∗∗∗ 0.05 21.70 -3.31∗ -0.21∗∗∗ -0.44∗∗∗ -0.19∗∗∗ 0.09∗∗∗ 0.93∗∗∗ 22.26

30

Low 3.40∗∗∗ -0.45∗∗∗ 0.07∗∗∗ 0.21∗∗∗ 0.21∗∗∗ 0.57∗∗∗ 52.25 -4.40∗∗ -0.57∗∗∗ 0.07∗∗∗ 0.21∗∗∗ 0.20∗∗∗ 1.37∗∗∗ 53.41
2 3.99∗∗∗ -0.37∗∗∗ -0.03 0.02 0.14∗∗∗ 0.47∗∗ 43.08 -5.46∗∗∗ -0.51∗∗∗ -0.04 0.02 0.13∗∗∗ 1.60∗∗∗ 45.56
3 4.86∗∗∗ -0.39∗∗∗ -0.06∗∗∗ -0.02 0.12∗∗∗ 0.35∗∗ 46.53 -5.06∗∗∗ -0.53∗∗∗ -0.06∗∗∗ -0.02 0.11∗∗∗ 1.63∗∗∗ 49.51
4 5.33∗∗∗ -0.45∗∗∗ -0.05∗∗ -0.16∗∗∗ 0.14∗∗∗ 0.28 49.37 -5.98∗∗∗ -0.61∗∗∗ -0.06∗∗ -0.16∗∗∗ 0.13∗∗∗ 1.83∗∗∗ 52.54
5 4.51∗∗∗ -0.38∗∗∗ -0.03 -0.14∗∗∗ 0.13∗∗∗ 0.37∗∗ 44.75 -4.63∗∗∗ -0.52∗∗∗ -0.03 -0.14∗∗∗ 0.12∗∗∗ 1.52∗∗∗ 47.34
6 3.98∗∗∗ -0.37∗∗∗ -0.06∗∗∗ -0.14∗∗∗ 0.14∗∗∗ 0.61∗∗∗ 46.32 -6.24∗∗∗ -0.53∗∗∗ -0.07∗∗∗ -0.14∗∗∗ 0.13∗∗∗ 1.75∗∗∗ 49.51
7 2.85∗∗∗ -0.28∗∗∗ 0.03 -0.13∗∗∗ 0.13∗∗∗ 0.82∗∗∗ 38.54 0.59 -0.34∗∗∗ 0.03 -0.13∗∗∗ 0.13∗∗∗ 0.59∗∗ 37.54
8 3.81∗∗∗ -0.35∗∗∗ -0.06∗∗∗ -0.27∗∗∗ 0.14∗∗∗ 0.41∗∗ 42.21 -5.01∗∗∗ -0.48∗∗∗ -0.06∗∗∗ -0.27∗∗∗ 0.13∗∗∗ 1.48∗∗∗ 44.64
9 5.02∗∗∗ -0.37∗∗∗ -0.08∗∗∗ -0.21∗∗∗ 0.17∗∗∗ 0.38∗ 45.99 -4.32∗∗ -0.50∗∗∗ -0.09∗∗∗ -0.21∗∗∗ 0.16∗∗∗ 1.55∗∗∗ 48.54
High 4.91∗∗∗ -0.41∗∗∗ -0.12∗∗∗ -0.23∗∗∗ 0.21∗∗∗ 0.49∗ 40.70 -4.19∗ -0.55∗∗∗ -0.13∗∗∗ -0.23∗∗∗ 0.19∗∗∗ 1.55∗∗∗ 42.27
High−Low -1.51 -0.03 0.19∗∗∗ 0.44∗∗∗ 0.00 0.09 15.09 -0.21 -0.02 0.19∗∗∗ 0.44∗∗∗ 0.01 -0.17 14.97

Low 7.33∗∗∗ -0.83∗∗∗ -0.26 ∗∗∗
0.20 ∗∗∗
0.84∗∗∗ 0.06 71.70 -0.55 -0.94∗∗∗ -0.27 ∗∗∗
0.20 ∗∗∗
0.83∗∗∗ 1.23∗∗∗ 71.97
2 3.35∗∗∗ -0.55∗∗∗ -0.11∗∗∗ -0.08∗∗∗ 0.51∗∗∗ 0.74∗∗∗ 60.27 -3.49 -0.67∗∗∗ -0.12∗∗∗ -0.08∗∗ 0.50∗∗∗ 1.27∗∗∗ 60.62
3 1.93∗∗ -0.45∗∗∗ 0.00 -0.02 0.45∗∗∗ 0.63∗∗∗ 56.88 -2.84∗ -0.53∗∗∗ -0.00 -0.02 0.45∗∗∗ 0.92∗∗∗ 57.02
4 3.92∗∗∗ -0.45∗∗∗ -0.00 -0.08∗∗∗ 0.27∗∗∗ 0.13 50.81 -7.51∗∗∗ -0.60∗∗∗ -0.01 -0.08∗∗∗ 0.26∗∗∗ 1.80∗∗∗ 53.46
5 4.66∗∗∗ -0.43∗∗∗ -0.00 -0.11∗∗∗ 0.18∗∗∗ 0.23 48.12 -6.21∗∗∗ -0.57∗∗∗ -0.01 -0.11∗∗∗ 0.17∗∗∗ 1.75∗∗∗ 51.15
6 4.94∗∗∗ -0.42∗∗∗ -0.02 -0.05∗∗ 0.15∗∗∗ 0.17 48.79 -6.81∗∗∗ -0.58∗∗∗ -0.03 -0.05∗ 0.14∗∗∗ 1.86∗∗∗ 52.52
7 4.70∗∗∗ -0.36∗∗∗ 0.01 -0.04∗ 0.07∗∗∗ 0.26 39.96 -5.40∗∗∗ -0.50∗∗∗ 0.00 -0.04 0.06∗∗∗ 1.63∗∗∗ 43.55
8 3.46∗∗∗ -0.32∗∗∗ 0.00 -0.06∗∗∗ 0.07∗∗∗ 0.29∗ 39.09 -5.10∗∗∗ -0.44∗∗∗ -0.00 -0.06∗∗∗ 0.06∗∗∗ 1.41∗∗∗ 42.24
9 4.36∗∗∗ -0.34∗∗∗ 0.01 -0.07∗∗∗ -0.00 0.49∗∗ 34.92 -4.91∗∗∗ -0.47∗∗∗ 0.01 -0.08∗∗∗ -0.01 1.57∗∗∗ 38.24
High 4.88∗∗∗ -0.38∗∗∗ -0.01 -0.01 0.01 0.56∗∗∗ 34.62 -5.42∗∗∗ -0.54∗∗∗ -0.01 -0.01 -0.01 1.75∗∗∗ 37.45
High−Low 2.45 -0.45∗∗∗ -0.25 ∗∗∗
0.21 ∗∗∗
0.83∗∗∗ -0.51 57.70 4.86∗ -0.40∗∗∗ -0.25 ∗∗∗
0.21 ∗∗∗
0.83∗∗∗ -0.52 57.61
Table 6. Conditional Regressions with Investor Sentiment, Default Spread, Liquidity
Factor, and Recession Dummy.
This table reports alphas, betas, and adjusted R2 of the market timing regressions of the MAP excess
returns on the four Carhart factors plus one instrumental variable (change in investor sentiment ∆S from
Baker and Wurgler (2007), default spread D using the difference between Moody’s BAA and AAA corporate
bond yields, liquidity factor L from Pastor and Stambaugh (2003), and a recession dummy RI) as well as
interaction terms of the instrumental variable with the market’s excess return using portfolios sorted by
size, book-to-market and momentum. Alphas are annualized and in percent. The sample period covers
1968:08 until 2010:12. The length of the moving average window is 24 months. A one-way transaction cost
of 0.5% has been imposed in constructing the switching moving average strategy excess returns. Newey and
West (1987) standard errors with 24 lags are used in reporting statistical significance of a two-sided null
hypothesis at the 1%, 5%, and 10% level is given by a ∗∗∗ , a ∗∗ , and a ∗ , respectively.

Portfolio α ∆S D L RI ∆S × rm D × rm L × rm RI × rm R̄2
Low -0.64 0.31∗∗∗ 0.37 -3.76 -0.03 0.07∗∗∗ -0.05 -0.02 -0.30∗∗∗ 57.35
2 1.54 0.36∗∗∗ 0.34 -3.92 -0.15 0.05∗∗∗ 0.01 -0.44 -0.37∗∗∗ 60.12
3 -1.96 0.30∗∗∗ 0.47∗∗ -7.34∗∗∗ -0.19 0.07∗∗∗ -0.03 -0.37 -0.39∗∗∗ 58.49
4 -0.93 0.24∗∗∗ 0.47∗∗ -6.25∗∗∗ -0.35 0.07∗∗∗ 0.05 -0.40 -0.39∗∗∗ 58.67
5 -2.86 0.16∗ 0.48∗∗∗ -5.57∗∗∗ -0.03 0.07∗∗∗ 0.03 -0.60∗ -0.43∗∗∗ 61.97
6 -2.30 0.16∗∗ 0.41∗∗ -6.24∗∗∗ -0.14 0.07∗∗∗ 0.07∗∗ -0.47 -0.42∗∗∗ 59.30
7 -0.44 0.18∗∗ 0.29∗ -7.51∗∗∗ 0.15 0.05∗∗∗ 0.03 -0.09 -0.49∗∗∗ 60.23
8 -2.61 0.19∗∗ 0.46∗∗∗ -6.30∗∗∗ 0.06 0.05∗∗∗ 0.05∗ 0.04 -0.49∗∗∗ 60.35
9 1.23 0.16∗∗ 0.16 -3.01 0.05 0.06∗∗∗ -0.02 0.16 -0.46∗∗∗ 62.76
High -3.39∗ -0.08 0.30∗∗ -0.84 0.21 0.05∗∗∗ -0.01 -0.12 -0.42∗∗∗ 58.34
High−Low 2.76 0.39∗∗∗ 0.07 -2.92 -0.24 0.02 -0.04 0.10 0.12∗∗ 22.34

Low -4.41∗ -0.11 0.48∗∗∗ -3.09 -0.13 0.11∗∗∗ -0.06∗∗ -0.86∗∗ -0.49∗∗∗ 64.67
2 -0.89 0.22∗∗ 0.29 -2.63 0.07 0.06∗∗∗ -0.01 0.12 -0.30∗∗∗ 48.29
3 3.84∗ 0.27∗∗∗ -0.01 -1.06 -0.02 0.05∗∗∗ 0.12∗∗∗ -1.14∗∗∗ -0.45∗∗∗ 55.94
4 1.62 0.19∗∗ 0.17 -3.91∗ 0.06 0.07∗∗∗ -0.15∗∗∗ -0.65∗ -0.24∗∗∗ 55.83
5 2.32 0.09 0.08 -2.20 0.24 0.07∗∗∗ -0.10∗∗∗ -0.66∗ -0.27∗∗∗ 53.03
6 -2.54 0.19∗∗ 0.45∗∗∗ -5.26∗∗∗ 0.33 0.03∗∗∗ -0.11∗∗∗ -0.11 -0.30∗∗∗ 56.00
7 -1.67 0.15∗ 0.26 -0.06 0.79∗∗∗ 0.01 -0.02 1.19∗∗∗ -0.32∗∗∗ 50.62
8 -5.23∗∗ 0.20∗∗ 0.65∗∗∗ -0.43 0.07 0.01 -0.24∗∗∗ -1.08∗∗∗ -0.27∗∗∗ 57.28
9 0.30 0.19∗ 0.34 -4.65∗ 0.10 0.02 -0.07∗ -0.91∗∗ -0.27∗∗∗ 52.65
High -3.43 0.35∗∗ 0.53∗ 5.58∗ 0.39 0.01 -0.31∗∗∗ 1.05∗∗ -0.17∗∗∗ 49.49
High−Low -0.98 -0.46∗∗∗ -0.05 -8.67∗∗∗ -0.51 0.11∗∗∗ 0.24∗∗∗ -1.90∗∗∗ -0.32∗∗∗ 25.98

Low 3.35 -0.44∗∗∗ 0.24 -1.38 -0.50 0.07∗∗∗ -0.11∗∗ -1.12∗∗ -0.48∗∗∗ 75.19
2 0.93 0.10 0.07 -5.19∗∗ 0.50∗ 0.03∗∗ -0.09∗∗ 0.50 -0.69∗∗∗ 70.78
3 0.53 -0.11 0.03 -4.01∗∗ 0.51∗∗ 0.00 0.05∗ 0.62∗∗ -0.64∗∗∗ 66.58
4 -1.10 0.35∗∗∗ 0.26 0.65 -0.17 0.05∗∗∗ -0.02 0.04 -0.45∗∗∗ 58.80
5 3.13 0.21∗∗ 0.00 4.09∗∗ -0.05 0.06∗∗∗ -0.04 -0.79∗∗ -0.42∗∗∗ 59.09
6 4.72∗∗ 0.16∗∗ -0.12 0.71 -0.08 0.09∗∗∗ -0.04 -1.16∗∗∗ -0.37∗∗∗ 59.41
7 2.11 0.17∗∗ 0.02 3.24 0.02 0.09∗∗∗ -0.00 -0.34 -0.33∗∗∗ 46.78
8 -3.12 0.20∗∗∗ 0.43∗∗∗ 0.22 -0.13 0.05∗∗∗ -0.03 0.13 -0.33∗∗∗ 48.83
9 0.13 0.09 0.23 -4.73∗∗ 0.12 0.08∗∗∗ -0.03 -0.35 -0.35∗∗∗ 45.08
High -1.76 0.01 0.36∗∗ -0.67 0.07 0.12∗∗∗ -0.06∗ -0.29 -0.39∗∗∗ 44.92
High−Low 5.12 -0.45∗∗∗ -0.12 -0.71 -0.57 -0.05∗∗ -0.05 -0.83 -0.09 59.32
31
Table 7. Monte Carlo Simulations.
This table reports the results for the improvement delivered by the MA switching strategy over the buy-
and-hold strategy, the trading frequency as well as the break-even transaction cost using 1000 Monte Carlo
simulations with randomly generated returns designed to match the first two moments of ten decile portfolios
sorted by size, book-to-market and momentum. The sample period covers 1960:01 until 2011:12 with value-
weighted portfolio returns. ∆µ is the annualized improvement in the average in-sample monthly return,
∆σ is the annualized improvement in the return standard deviation, pA is the proportion of months during
which there is a hold signal, NT is the number of transactions (buy or sell) over the entire sample period,
BETC is the break-even one-sided transaction cost in percent, and p1 is the proportion of months during
which a buy signal was followed by a positive return of the underlying portfolio. The length of the moving
average window is 24 months. A one-way transaction cost of 0.5% has been imposed in the reported ∆µ
and ∆σ.

Portfolio ∆µ ∆σ pA NT BETC p1
Low 6.233 3.957 0.717 58.550 5.393 0.696
2 6.397 4.054 0.709 59.572 5.447 0.698
3 5.616 3.504 0.737 56.110 5.064 0.691
4 5.483 3.395 0.736 56.769 4.878 0.691
5 5.036 3.099 0.751 54.798 4.657 0.688
6 4.718 2.860 0.756 54.165 4.405 0.686
7 4.515 2.735 0.763 53.271 4.287 0.685
8 4.493 2.701 0.760 53.611 4.247 0.685
9 4.048 2.397 0.769 52.421 3.910 0.683
High 4.094 2.383 0.751 54.835 3.785 0.685

Low 5.873 3.557 0.688 61.167 4.868 0.701
2 4.570 2.711 0.741 55.712 4.155 0.688
3 4.077 2.412 0.766 53.354 3.880 0.683
4 4.347 2.591 0.755 54.129 4.072 0.686
5 3.741 2.202 0.780 50.998 3.709 0.681
6 3.736 2.209 0.784 50.408 3.738 0.681
7 3.324 1.955 0.805 47.599 3.542 0.677
8 3.335 1.981 0.808 47.060 3.575 0.677
9 3.451 2.066 0.809 46.743 3.718 0.677
High 4.770 2.951 0.773 51.940 4.635 0.685

Low 14.519 9.904 0.450 67.526 11.113 0.776
2 8.422 5.366 0.610 65.728 6.548 0.724
3 6.120 3.744 0.683 61.190 5.080 0.703
4 5.015 2.976 0.723 57.578 4.428 0.693
5 4.621 2.708 0.728 57.089 4.099 0.692
6 4.264 2.519 0.751 54.078 3.986 0.687
7 3.986 2.337 0.763 52.913 3.818 0.684
8 3.391 1.986 0.804 47.714 3.576 0.677
9 3.671 2.185 0.800 48.301 3.843 0.679
High 4.419 2.763 0.801 48.140 4.624 0.681
32
Table 8. Bootstrap Simulations.
This table reports the results for the improvement delivered by the MA switching strategy over the buy-
and-hold strategy, the trading frequency as well as the break-even transaction cost using 1000 bootstrap
simulations with randomly drawn returns from the historical returns of ten decile portfolios sorted by
size, book-to-market and momentum. The sample period covers 1960:01 until 2011:12 with value-weighted
portfolio returns. ∆µ is the annualized improvement in the average in-sample monthly return, ∆σ is the
annualized improvement in the return standard deviation, pA is the proportion of months during which
there is a hold signal, NT is the number of transactions (buy or sell) over the entire sample period, BETC
is the break-even one-sided transaction cost in percent, and p1 is the proportion of months during which a
buy signal was followed by a positive return of the underlying portfolio. The length of the moving average
window is 24 months. A one-way transaction cost of 0.5% has been imposed in the reported ∆µ and ∆σ.

Low 5.999 4.355 0.728 53.025 5.810 0.713
2 6.168 4.524 0.718 54.910 5.752 0.705
3 5.531 4.188 0.746 52.878 5.362 0.705
4 5.388 4.093 0.745 53.250 5.197 0.705
5 5.017 3.799 0.759 51.736 4.972 0.709
6 4.691 3.509 0.764 51.186 4.698 0.700
7 4.462 3.351 0.771 49.830 4.601 0.700
8 4.445 3.244 0.767 50.777 4.493 0.698
9 3.999 2.869 0.777 49.728 4.129 0.690
High 4.084 2.764 0.757 51.564 4.067 0.712

Low 5.801 3.890 0.695 58.149 5.121 0.709
2 4.566 3.252 0.748 52.523 4.462 0.697
3 4.098 2.958 0.772 49.318 4.268 0.702
4 4.353 3.191 0.761 49.777 4.504 0.701
5 3.742 2.717 0.785 46.914 4.097 0.711
6 3.720 2.737 0.791 46.224 4.131 0.698
7 3.292 2.170 0.810 44.465 3.794 0.700
8 3.390 2.648 0.814 41.779 4.171 0.695
9 3.506 2.496 0.813 43.614 4.132 0.707
High 4.574 3.386 0.782 46.153 5.088 0.707

Low 14.009 8.645 0.447 62.438 11.525 0.777
2 8.029 5.226 0.616 60.257 6.848 0.724
3 5.769 3.643 0.689 55.939 5.302 0.698
4 4.825 3.219 0.729 53.356 4.643 0.696
5 4.517 3.073 0.737 53.093 4.375 0.702
6 4.236 2.992 0.759 50.717 4.291 0.697
7 3.973 2.881 0.772 49.665 4.109 0.710
8 3.399 2.378 0.810 44.976 3.875 0.702
9 3.678 2.819 0.807 44.727 4.218 0.710
High 4.464 3.455 0.805 45.275 5.044 0.709
33
Table 9. International Evidence of Moving Average Strategies Performance.
This table reports the results for the improvement delivered by the MA switching strategy over the buy-and-
hold strategy, the trading frequency as well as the break-even transaction cost using local currency value-
weighted returns of the market portfolios and portfolios sorted by several variables in seven countries. ∆µ
is the annualized improvement in the average in-sample monthly return, ∆σ is the annualized improvement
in the return standard deviation, pA is the proportion of months during which there is a hold signal, NT is
the number of transactions (buy or sell) over the entire sample period, BETC is the break-even one-sided
transaction cost in percent, and p1 is the proportion of months during which a buy signal was followed by
a positive return of the underlying portfolio. The length of the moving average window is 24 months. A
one-way transaction cost of 0.5% has been imposed in the reported ∆µ and ∆σ.
Panel A: Australian portfolios between 1975:01 and 2010:12.

MKT 4.253 3.711 0.828 30 4.820 0.711
High BM 2.770 3.800 0.880 20 4.710 0.716
Low BM 5.647 5.607 0.740 38 5.052 0.686
High EP 2.537 3.325 0.900 18 4.793 0.725
Low EP 6.457 5.674 0.723 48 4.574 0.701
High CEP 1.685 1.708 0.897 24 2.387 0.681
Low CEP 7.608 6.423 0.676 42 6.159 0.708
High DP 1.726 1.778 0.904 16 3.667 0.708
Low DP 6.178 5.721 0.757 36 5.835 0.686
Zero DP 11.449 10.351 0.564 42 9.268 0.735
Panel B: Canadian portfolios between 1977:01 and 2010:12.

MKT 3.409 3.959 0.805 26 4.195 0.706
High BM 4.189 3.714 0.818 26 5.155 0.690
Low BM 5.881 5.300 0.737 40 4.704 0.698
High EP 2.943 3.727 0.865 22 4.281 0.698
Low EP 6.450 5.017 0.693 36 5.733 0.706
High CEP 2.770 2.576 0.878 22 4.029 0.690
Low CEP 6.492 5.846 0.682 30 6.925 0.734
High DP 3.059 2.957 0.898 22 4.449 0.708
Low DP 5.691 4.082 0.737 34 5.356 0.711
Zero DP 11.222 10.956 0.557 40 8.978 0.763
Panel C: French portfolios between 1975:01 and 2010:12.

MKT 4.552 5.184 0.723 31 4.993 0.735
High BM 4.809 6.290 0.748 31 5.275 0.713
Low BM 4.943 5.001 0.730 31 5.422 0.748
High EP 5.240 6.796 0.745 31 5.747 0.738
Low EP 5.391 4.608 0.713 29 6.321 0.733
High CEP 4.682 5.734 0.772 33 4.824 0.699
Low CEP 4.794 5.027 0.767 25 6.520 0.735
High DP 3.631 4.374 0.811 25 4.938 0.721
Low DP 6.012 6.632 0.676 31 6.594 0.752
Zero DP 6.781 6.723 0.662 25 9.223 0.708
34
Table 9 Continued.
Panel D: German portfolios between 1975:01 and 2010:12.

MKT 4.272 5.191 0.755 32 4.539 0.723
High BM 3.426 4.973 0.804 28 4.160 0.713
Low BM 5.872 6.532 0.706 29 6.884 0.699
High EP 3.914 6.543 0.738 22 6.048 0.730
Low EP 5.456 5.889 0.723 33 5.622 0.699
High CEP 3.377 4.242 0.831 34 3.377 0.696
Low CEP 5.410 6.327 0.689 25 7.357 0.725
High DP 3.045 4.885 0.811 26 3.982 0.733
Low DP 5.262 6.896 0.718 31 5.771 0.699
Zero DP 11.785 10.579 0.566 28 14.311 0.750
Panel E: Italian portfolios between 1975:01 and 2010:12.

MKT 6.373 4.863 0.662 29 7.471 0.716
High BM 7.849 6.499 0.623 36 7.413 0.706
Low BM 6.214 4.560 0.667 31 6.815 0.701
High EP 4.391 5.077 0.694 33 4.524 0.718
Low EP 7.273 5.099 0.618 32 7.727 0.713
High CEP 5.776 4.714 0.716 32 6.137 0.684
Low CEP 7.413 5.599 0.593 21 12.002 0.745
High DP 4.266 5.180 0.706 29 5.002 0.708
Low DP 8.315 5.636 0.593 41 6.895 0.730
Zero DP 9.773 8.406 0.588 29 11.458 0.738
Panel F: Japanese portfolios between 1975:01 and 2010:12.

MKT 6.792 6.082 0.642 22 10.496 0.745
High BM 5.314 5.705 0.735 24 7.529 0.728
Low BM 9.756 7.033 0.583 32 10.366 0.735
High EP 4.792 4.853 0.730 20 8.146 0.691
Low EP 9.004 7.647 0.608 26 11.775 0.723
High CEP 4.307 4.073 0.730 16 9.152 0.696
Low CEP 8.933 7.524 0.561 32 9.491 0.755
High DP 5.743 4.843 0.703 22 8.876 0.708
Low DP 9.647 6.979 0.576 34 9.647 0.735
Zero DP 9.601 9.626 0.593 28 11.659 0.733
Panel G: UK portfolios between 1975:01 and 2010:12.

MKT 2.277 2.030 0.860 20 3.871 0.713
High BM 3.406 3.723 0.860 16 7.237 0.696
Low BM 3.200 2.684 0.811 22 4.945 0.716
High EP 3.653 2.866 0.890 24 5.175 0.672
Low EP 3.866 2.937 0.828 20 6.571 0.725
High CEP 2.750 2.701 0.902 20 4.675 0.676
Low CEP 3.432 2.785 0.809 14 8.335 0.735
High DP 2.773 3.760 0.877 26 3.626 0.679
Low DP 3.495 3.370 0.821 14 8.487 0.716
35
Figure 1. Scatter Plot of Buy-and-Hold returns versus the Moving Average returns:
High ME decile portfolio
High ME
20
15
10
5
36
rMA
−5
−10
−15
−20
−20 −15 −10 −5 0 5 10 15 20
rBH
Notes: Figure 1 presents a scatter plot of the returns of the High ME decile buy-and-hold portfolio returns versus the moving average strategy returns. The
sample contains 624 monthly observations and the data covers the 1960:01 until 2011:12 period.
Figure 2. Performance of MA strategy with individual stocks
∆µ ∆σ p
A
10000 12000 3000
8000 10000 2500
8000 2000
6000
6000 1500
4000
4000 1000
2000 2000 500
0 0 0
−100 0 100 0 200 400 0 0.5 1
37
NT BETC p1
5000 8000 5000
4000 4000
6000
3000 3000
4000
2000 2000
2000
1000 1000
0 0 0
0 50 100 −500 0 500 1000 0 0.5 1
Notes: Figure 2 presents histograms of the annualized percentage change improvement of MA over BH (∆µ), the annualized percentage change improvement
in standard deviation of return (∆σ), the percentage active (pA ), the number of trades (NT), break-even transaction cost (BETC) and the percentage of
times the MA return exceeds the risk-free rate (p1 ) for the entire sample of stock in the CRSP database for which there is at least 48 contiguous non-missing
monthly returns available during the 1960:01 until 2011:12 period.

Market Timing With Moving Averages PDF

Uploaded by

Copyright:

Available Formats

Market Timing With Moving Averages PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Market Timing With Moving Averages PDF

Uploaded by

Copyright:

Available Formats

Market Timing with Moving Averages1

Accounting and Finance

April 15, 2015

timing ability of the MA strategy.

JEL Classification: G11, G12, G14.

cross-sectional explanatory power and substantial historical performance.

the performance being driven by volatility.

a greater improvement of the moving average returns relative to buy-and-hold returns.

January 1960 and ends in December 2011.

next section or τ = 0.005.

MAPjt,L = R̃jt,L − Rjt , j = 1, . . . , N. (4)

Insert Table 1 here.

are explored in the next section.

∆µ = −2.19 + 0.35σ, (5)

models and abnormal returns.

3.2 Abnormal Returns

return variation that cannot be accounted for.

Insert Table 2 here.

Insert Figure 1 about here.

3.4 Individual Stocks

∆µ = 5.09 + 0.20σ, (7)

the moving average.9

4.2 Alternative Lag Lengths

1% and 9%, depending on which sets of deciles I consider.

performance of the MA switching strategy.

Insert Table 3 here.

the length of the moving average window.

5 Drivers of Abnormal Returns

two simulations using bootstrapped returns and randomly generated returns.

5.1 Market Timing

of the market return as in Henriksson and Merton (1981):

MAPjt,L = αj + βj,m rmkt,t + γj,m rmkt,t I{rmkt,t >0} + ǫjt , j = 1, . . . , N, (9)

value of γj,m is usually interpreted as evidence of successful market timing ability.

Insert Table 4 here.

5.2 Business Cycles and Market States

Insert Table 5 here.

5.3 Conditional Models with Macroeconomic Variables

bond yield as the instrumental variables Zt in the following regression:

Insert Table 6 here.

5.4.1 Randomly Generated Returns

Insert Table 7 here.

thus, drawing extreme returns is highly unlikely.

5.4.2 Bootstrapped Returns

Insert Table 8 here.

autocorrelations are not the main performance driver.

different and are available from the author upon request.

in competitive financial markets. A few potential reasons seem plausible.

learning and improves their well-being and utility.

yield, dividend yield and cash earnings to price ratio.

Insert Table 9 here.

reported previously in this paper.

higher than the value reported in (8) for US individual stocks.

stocks are available from the author upon request.

with low trading volumes.14

the superiority of the MA switching strategy.

of Financial Economics 51, 245–272.

of Finance 61, 259–299.

tives 21, 129–151.

tions,” Journal of Financial Economics 52, 47–78.