Mngng rsk mmntm

Electronic copy available at: http://ssrn.com/abstract=2041429Electronic copy available at: http://ssrn.com/abstract=2041429
Managing the Risk of Momentum
Pedro Barrosoy
and Pedro Santa-Claraz
This version: April 2012
Abstract
Compared to the market, value or size risk factors, momentum has of-
fered investors the highest Sharpe ratio. However, momentum has also had
the worst crashes, making the strategy unappealing to investors with rea-
sonable risk aversion. We …nd that the risk of momentum is highly variable
over time and quite predictable. The major source of predictability does not
come from systematic risk but from speci…c risk. Managing this time-varying
risk virtually eliminates crashes and nearly doubles the Sharpe ratio of the
momentum strategy. Risk-managed momentum is a much greater puzzle
than the original version.
We thank Eduardo Schwartz for helpful comments.
y
PhD student at Nova School of Business and Economics. E-mail: pmsb@novasbe.pt.
z
Millennium Chair in Finance. Nova School of Business and Economics and NBER and
CEPR. Campus de Campolide, 1099-032 Lisboa, Portugal. Phone +351-21-3801600. E-mail:
psc@novasbe.pt.
1

Electronic copy available at: http://ssrn.com/abstract=2041429Electronic copy available at: http://ssrn.com/abstract=2041429
1 Introduction
Momentum is a pervasive anomaly in asset prices. Jegadeesh and Titman (1993)
…nd that previous winners in the US stock market outperform previous losers by
as much as 1.49 percent a month. The Sharpe ratio of this strategy exceeds the
Sharpe ratio of the market itself, as well as the size and value anomalies. Mo-
mentum returns are even more of a puzzle since they are negatively related to the
market and the value risk factors. From 1927 to 2011, momentum had a monthly
excess return of 1.75 percent per month controlling for the Fama-French factors.
This result has led researchers to use momentum as an additional risk factor.1
Mo-
mentum is not just a US stock market anomaly. Momentum has been documented
in European equities, emerging markets, country stock indices, industry portfolios,
currency markets, commodities and across asset classes.2
Grinblatt and Titman
(1989,1993) found most mutual fund managers incorporate momentum of some
sort in their investment decisions, so relative strength strategies are widespread
among practitioners.
But the remarkable performance of momentum comes with occasional large
crashes.3
In 1932, the winners-minus-losers strategy delivered a -91.59 percent
return in just two months. In 2009 momentum experienced another signi…cant
crash of -73.42 percent over three months. Even the large returns of momentum
do not compensate an investor with reasonable risk version for these sudden crashes
that take decades to recover from.
The two most expressive momentum crashes occurred as the market rebounded
following large previous declines. One explanation for this pattern is the time-
varying systematic risk of the momentum strategy. Grundy and Martin (2001)
show that momentum has signi…cant negative beta following bear markets.4
They
1
Carhart (1997)
2
See Rouwenhorst (1998,1999) for international evidence, Asness, Liew, and Stevens (1997)
for country stock indices, Moskowitz and Grinblatt (1999) for industry portfolios, Okunev and
White (2003) and Menkhof, Sarno, Schmeling, and Schrimpf (2012) for currency markets, Erb
and Harvey (2006) for commodities and Asness, Moskowitz, and Pederson (2011) for momentum
across asset classes.
3
Daniel and Moskowitz (2011)
4
Following negative returns for the overall market, winners tend to be low-beta stocks and
the reverse for losers. Therefore the winner-minus-losers strategy will have a negative beta.
2

Electronic copy available at: http://ssrn.com/abstract=2041429
argue that hedging this time-varying market exposure produces stable momentum
returns but Daniel and Moskowitz (2011) show that using betas in real time does
not avoid the crashes.
In this work we propose a di¤erent method to manage momentum risk. We
estimate the risk of momentum by the realized variance of daily returns and …nd
that it is highly predictable. An auto-regression of monthly realized variances
yields an out-of-sample (OOS) R-square of 57.82 percent. This is 19.01 percentage
points (p.p.) more than a similar auto-regression for the variance of the market
portfolio which is already famously predictable.5
Making use of this predictability in risk management leads to substantial eco-
nomic gains. We scale the long-short portfolio by its realized volatility in the
previous 6 months, thereby obtaining a strategy with constant volatility. The
Sharpe ratio improves from 0.53 for unmanaged momentum to 0.97 for its risk-
managed version. But the most important bene…t comes from a reduction in crash
risk. The excess kurtosis drops from 18.24 to 2.68 and the left skew improves from
-2.47 to -0.42. The minimum one-month return for momentum is -78.96 percent
while for risk-managed momentum is -28.40 percent. The maximum drawdown of
momentum was -96.69 percent versus -45.20 percent for its risk-managed version.
To assess the economic signi…cance of our results, we evaluate the bene…ts of
risk management for a risk-averse investor using a power utility function with Con-
stant Relative Risk Aversion (CRRA) of four. The representative investor holding
the market portfolio has an annual certainty equivalent of 0.14 percent. Adding
momentum to the portfolio reduces this certainty equivalent to -5.46 percent. How-
ever, combining the market with risk-managed momentum already achieves a cer-
tainty equivalent of 13.54 percent. We …nd that the main bene…t of risk manage-
ment comes in the form of smaller crash risk. This alone provides a gain of 14.96
p.p. in annual certainty equivalent.
One pertinent question is why managing risk with realized variances works
while using time-varying betas does not. To answer this question we decompose
the risk of momentum into systematic risk (from time-varying exposure to the
market) and speci…c risk. We …nd that the systematic component is only 23
percent of total risk on average, so most of the risk of momentum is speci…c. This
5
Engle and Bollerslev (1986), Schwert (1989).
3

speci…c risk is more persistent and predictable than the systematic component.
The OOS R-square of the speci…c component is 47 percent versus 21 percent for
the systematic component. This is why hedging with time-varying betas fails: it
focuses on the smaller part of risk and also the less predictable one.
The work that is more closely related to ours is Grundy and Martin (2001)
and Daniel and Moskowitz (2011). But their work focuses on the time-varying
systematic risk of momentum, while we focus on momentum’s speci…c risk. Our
results have the distinct advantage of o¤ering investors using momentum strategies
an e¤ective way to manage risk without forward-looking bias. The resulting risk-
managed strategy deepens the puzzle of momentum.
This paper is structured as follows. Section 2 discusses the long-run properties
of momentum returns and its exposure to crashes. Section 3 shows that momentum
risk varies substantially over time in a highly predictable manner. We analyze the
implications of such predictability for risk management in Section 4. Section 5
decomposes the gains of risk management according to the moments of returns. In
Section 6 we decompose the risk of momentum and study the persistence of each
of its components. Finally, Section 7 presents our conclusions.
2 Momentum in the long run
We compare momentum to the Fama-French factors using a long sample of 85
years of monthly returns from July 1926 to December 2011.6
This is the same
sample period as in Daniel and Moskowitz (2011).
Figure 1 presents the cumulative returns of each factor. As each factor consists
of a long-short strategy, we construct the series of returns assuming the investor
puts $1 in the risk-free asset at the beginning of the sample, buys $1 worth of
the long portfolio and sells the same amount of the short portfolio. Then in
each subsequent month, the strategy fully reinvests the accumulated wealth in the
risk-free asset, assuming a position of this same size in the long and short leg
of the portfolio. The winners-minus-losers (WML) strategy o¤ered an impressive
performance for investors. One dollar fully reinvested in the momentum strategy
6
See annex for a description of the data.
4

grew to $68,741 by the end of the sample. This compares to the cumulative return
of $2,136 from simply holding the market portfolio.
There would be nothing puzzling about momentum’s returns if they corre-
sponded to a very high exposure to risk. However, running an OLS regression of
the WML on the Fama-French factors gives (t-stats in parenthesis):
rWML;t = 1:752 0:378 rRMRF;t 0:249 rSMB;t 0:677 rHML;t
(7:93) ( 8:72) ( 3:58) ( 10:76)
so momentum provided abnormal returns of 1.75 percent per month with neg-
ative exposure to the Fama and French (1992) risk factors. This amounts to a 21
percent per year abnormal return and the negative loadings on other risk factors
imply momentum actually diversi…ed risk in this extended sample.
Table 1 compares descriptive statistics for momentum in the long-run with
the Fama-French factors. Buying winners and shorting losers has provided large
returns of 14.46 percent per year, with a Sharpe ratio higher than the market.
The impressive excess returns of momentum, its high Sharpe ratio and negative
relation to other risk factors, particularly the value premium, make it look like a
free lunch for investors.
But as Daniel and Moskowitz (2011) show there is a dark side to momentum. Its
large gains come at the expense of a very high excess kurtosis of 18.24 combined
with a pronounced left-skew of -2.47. These two features of the distribution of
returns of the momentum strategy imply a very fat left tail – signi…cant crash
risk. The apparent free lunch of momentum returns can very rapidly turn into a
free fall, wiping out decades of returns.
Figure 2 shows the performance of momentum in the two most turbulent
decades for the strategy: the 1930’s and the 2000’s.
In July and August 1932, momentum had a cumulative return of -91.59 percent.
From March to May 2009, momentum had another large crash of -73.42 percent.
These short periods leave an enduring impact on cumulative returns. For example,
someone investing one dollar in the WML strategy in July 1932 would only recover
it in April 1963, 31 years after and with considerably less real value. This puts
the risk to momentum investing in an adequate long-run perspective.
Both in 1932 and in 2009, the crashes happened as the market rebounded after
5

experiencing large losses.7
This leads to the question of whether investors could
predict the crashes in real time and hedge them away.
Grundy and Martin (2001) show that momentum has a substantial time-
varying loading on stock market risk. The strategy ranks stocks according to
returns during a formation period, for example the previous 12 months. If the
stock market performed well in the formation period, winners tend to be high-
beta stocks and losers low-beta stocks. So the momentum strategy, by shorting
losers to buy winners, has by construction a signi…cant time-varying beta: positive
after bull markets and negative after bear markets. They argue that hedging this
time-varying risk produces stable returns, even in pre-WWII data, when momen-
tum performed poorly. In particular, the hedging strategy would be long in the
market portfolio whenever momentum has negative betas, hence mitigating the
e¤ects of rebounds following bear markets, which is when momentum experiences
the worst returns. But the hedging strategy in Grundy and Martin (2001) uses
forward looking betas, estimated with information investors did not have in real
time. Daniel and Moskowitz (2011) show that using betas estimated solely on
ex-ante information does not avoid the crashes and portfolios hedged in real time
perform even worse than unhedged momentum.
3 The time-varying risk of momentum
One possible cause for the excess kurtosis of momentum is time-varying risk.8
The
very high excess kurtosis of 18.24 of the momentum strategy (more than twice the
market portfolio) leads us to study the dynamics of its risk and compare it with
the market (RMRF), value (HML) and size (SMB) risk factors.
For each month, we compute the realized variance RVt from daily returns in
the previous 21 sessions. Let frdgD
d=1 be the daily returns and fdtgT
t=1 the time
series of the dates of the last trading sessions of each month. Then the realized
variance of factor i in month t is:9
7
Daniel and Moskowitz (2011) argue this is due to the option-like payo¤s of distressed …rms
in bear markets.
8
See, for example, Engle (1982) and Bollerslev (1987).
9
Correcting for serial correlation of daily returns does not change the results signi…cantly.
6

RVi;t =
20X
j=0
r2
dt j (1)
Figure 4 shows the monthly realized volatility of momentum. This varies substan-
tially over time, from a minimum of 3.04 percent (annualized) to a maximum of
127.87 percent.
Table 2 shows the results of AR (1) regressions with the realized variances of
the WML, RMRF, SMB and HML:
RVi;t = + RVi;t 1 + "t (2)
Panel A presents the results for RMRF and WML, for which we have daily
data available from 1927:03 to 2011:12. Panel B adds the results for HML and
SMB, for which daily data is available from 1963:07 onwards.
Momentum returns are the most volatile. From 1927:03 to 2011:12, the average
realized volatility of momentum was 15.03, more than the 12.81 of the market port-
folio. For the 1963:07 onwards sample, the average realized volatility of momentum
was 16.40, the highest when compared to RMRF, SMB and HML.
In the full sample period, the standard deviation of monthly realized volatil-
ities is higher for momentum (12.26) than the market (7.82). Panel B con…rms
this result for the other factors in the 1963:07 onwards sample. So the risk of
momentum is the most variable.
The risk of momentum is also the most persistent. The AR(1) coe¢ cient of
momentum in the 1963:07 sample is 0.77, which is 0.19 more than the market for
the same sample period and higher than the estimate for SMB and HML.
To check the out-of-sample (OOS) predictability of risk, we use a training sam-
ple of 240 months to run an initial AR(1) and then use the estimated coe¢ cients
and last observation of realized variance to forecast the realized variance in the
following month. Then each month we use an expanding window of observations to
produce OOS forecasts and compare these to the accuracy of the historical mean
RV i;t. As a measure of goodness of …t we estimate the OOS R-square as:
7

R2
i;OOS = 1
T 1P
t=S
(bt + btRVi;t RVi;t+1)2
T 1P
t=S
(RV i;t RVi;t+1)2
(3)
where S is the initial training sample, bt;bt and RV i;t are estimated with informa-
tion available only up to time t.
The last column of table 2 shows the OOS R-squares of each auto-regression.
The AR(1) of the realized variance of momentum has an OOS R-square of 57.82
percent (full sample), which is 50 percent more than the market. For the period
from 1963:07 to 2011:12, the OOS predictability of momentum-risk is twice that
of the market. Hence more than half of the risk of momentum is predictable, the
highest level among risk factors. In the next section we explore the potential of
this predictability for risk management.
4 Risk-managed momentum
We use estimates of momentum risk to scale the exposure to the strategy in order
to have constant risk over time.
For each month we compute a variance forecast ^2
t from daily returns in the
previous 6 months.10
Let frWML;tgT
t=1 be the monthly returns of momentum and
frWML;dgD
d=1; fdtgT
t=1 be, as above, the daily returns and the time series of the
dates of the last trading sessions of each month.
The variance forecast is:
^2
t = 21
125X
j=0
r2
dt 1 j=126 (4)
As WML is a zero-investment and self-…nancing strategy we can scale it with-
out constraints. We use the forecasted variance to scale the returns:
10
We also used one-month and three-month realized variances as well as exponentially-weighted
moving average (EWMA) with half-lifes of 1, 3 and 6 months. All worked well with nearly
identical results.
8

rWML ;t =
target
^t
rWML;t (5)
where rWML;t is the unscaled or plain momentum, rWML ;t is the scaled or risk-
managed momentum, and target is a constant corresponding to the target level of
volatility. Scaling corresponds to having a weight in the long and short legs that
is di¤erent from one and varies over time, but it is still a self-…nancing strategy,
so the choice of the constant is arbitrary. We pick a target corresponding to an
annualized volatility of 12 percent.11
Figure 3 shows the cumulative returns of risk-managed momentum compared
to plain momentum. The risk-managed momentum strategy achieves a higher
cumulative return with less risk. So there are economic gains to risk-management
of momentum. The scaled strategy bene…ts from the large momentum returns
when it performs well and e¤ectively shuts it o¤in turbulent times, thus mitigating
momentum crashes. As a result, one dollar invested in risk-managed momentum
grows to $6,140,075 by the end of the sample, nearly 90 times more than the plain
momentum strategy.12
Also, the risk-managed strategy no longer has variable and
persistent risk, so risk management indeed works.13
Table 3 provides a summary of the economic performance of WML in 1927-
2011. The risk-managed strategy has a higher average return, with a gain of 2.04
(p.p. per year), with substantially less standard deviation (less 10.58 p.p. per
year). As a result, the Sharpe ratio of risk-managed portfolios almost doubles
from 0.53 to 0.97. The most important gains of risk-management show up in the
improvement in the higher order moments. Managing the risk of momentum drops
the excess kurtosis from a very high value of 18.24 to just 2.68 and reduces the left
11
The annualized standard deviation from monthly returns will be higher than 12% as volatil-
ities at daily frequency are not directly comparable to those at lower frequencies due to small
positive autocorrelation of daily returns.
12
This di¤erence in cumulative returns is fundamentally due to risk management successfully
avoiding the two momentum crashes. But in the post-war period from 1946 to 2007, not including
the crashes, the Sharpe ratio of momentum was 0.86, versus 1.15 for risk-managed momentum.
So risk management also contributes to performance in not so turbulent times.
13
The AR(1) coe¢ cient of monthly squared returns is only 0.14 for the scaled momentum
versus 0.40 for the original momentum. Besides, the auto-correlation of momentum is signi…cant
up to 15 lags while only 1 lag for risk-managed momentum. So persistence in risk is much smaller
for the risk-managed strategy.
9

skew from -2.47 to -0.42. This practically eliminates the crash risk of momentum.
Figure 5 shows the density function of momentum and its risk-managed version.
Momentum has a very long left tail which is much reduced in its risk-managed
version.
The bene…ts of risk-management are specially important in turbulent times.
Figure 6 shows the performance of risk-managed momentum in the decades with
the most impressive crashes. The scaled momentum manages to preserve the
investment in the 1930’s. This compares very favorably to the pure momentum
strategy which loses 90 percent in the same period. In the 2000’s simple momentum
lost 28 percent of wealth, because of the crash in 2009. Risk-managed momentum
ends the decade up 88 percent as it not only avoids the crash but also captures
part of the positive returns of 2007-2008.
Figure 7 shows the weights of the scaled momentum strategy over time –in-
terpreted as the dollar amount in the long or short leg. These range between the
values of 0.13 and 2.00, reaching the most signi…cant lows in the early 1930’s, in
2000-02, and in 2008-09. On average, the weight is just 0.90, slightly less than full
exposure to momentum. As these weights depend only on ex ante information this
strategy could actually be implemented in real time.
5 Economic Signi…cance: An Investor Perspec-
tive
Momentum o¤ers a trade-o¤ between an appealing Sharpe ratio, obtained from
the …rst two moments of its distribution, and less appealing higher order moments,
such as high kurtosis and left skewness. An economic criterion is needed to assess
whether this trade-o¤ is interesting. Risk management o¤ers improvements to
momentum across the board, higher expected returns, lower standard deviation
and crash risk. Still it is pertinent to evaluate the relative economic importance
of each of these contributions.
We use an utility-based approach to discuss the appeal of momentum to a
representative investor. We adopt the power utility function as it has the advantage
of taking into consideration higher order moments instead of focusing merely on the
10

mean and standard deviation of returns. This is particularly suitable as momentum
has a distribution far from normal. The utility of returns is:
U(r) =
(1 + r)1
1
(6)
where is the Constant Coe¢ cient of Relative risk aversion (CRRA). Bliss and
Panigirtzoglou (2004) estimate empirically from risk-aversion implicit in one-
month options on the S&P and the FTSE and …nd a value very close to 4. This is
a more plausible value for CRRA than previous estimates featured in the equity
premium puzzle literature using utility over consumption. So we adopt this value
for CRRA. We obtain the certainty equivalent from the utility of returns:
CE(r) = f(1 )E[U(r)]g
1
1 1 (7)
This states the welfare a series of returns o¤ers the investor in terms of an equiv-
alent risk-free annual return, expressed in a convenient unit of percentage points
(p.p.) per year.
For an economic measure of the importance of the mean return, variance and
higher order moments, we use a Taylor series approximation to expected utility
around the mean:
E[U(r)] = U(r) +
1
2
U00
(r)E(r r)2
+ 3(r) (8)
where 3 is the Lagrangian rest corresponding to the utility from moments with
order greater than 2. From this we obtain the certainty equivalent due to each
moment:
CE( 1) = f(1 )U(r)g
1
1 1 (9)
CE( 2) = (1 )[U(r) +
1
2
U00
(r)E(r r)2
]
1
1
CE( 1) 1 (10)
CE( i>2) = CE(r) CE( 1) CE( 2) (11)
We compute the certainty equivalent from annual overlapping returns, an ade-
11

quate horizon from an investor perspective. Also, one-year horizons capture better
the occasional large drawdowns of momentum documented in Section 2.
Table 4 shows the decomposition of the certainty equivalent for the represen-
tative investor holding the market portfolio. It also assesses whether it is optimal
to deviate from the market portfolio to include (risk-managed) momentum.
The …rst row shows the results for holding only the market portfolio. The mean
return had a positive contribution for the certainty equivalent of 11.72 percent per
year, but the variance of returns reduces this by 7.39 p.p. Higher order moments
diminuish the certainty equivalent by a further 4.18 p.p. As a result the certainty
equivalent of the market portfolio was only 0.14 percent per year.
Adding momentum to the market portfolio increases returns. As a result the
certainty equivalent of the mean return increases from 11.72 percent per year to
28.51 percent. The higher standard deviation partially o¤sets this gain by reducing
the certainty equivalent by 6.51 p.p. Still, looking only at the …rst two moments
of the combined portfolio leads to the conclusion that the investor is better o¤
including momentum.
But the increase in higher order risk –the momentum crashes –reduces the
certainty equivalent by 15.89 p.p. per year. As a result, including momentum
actually dampens the economic performance of the market portfolio. The cer-
tainty equivalent of the market plus momentum is -5.46 percent per year versus
0.14 percent of the market only. The high Sharpe ratio of momentum does not
compensate the investor for bearing the increased crash risk. So, in spite of the
impressive cumulative returns of momentum in the long-run, crash risk is so high
that a reasonable risk-averse investor would rather just hold the market portfolio.
This illustrates with an economic measure how far the distribution of momen-
tum is from normality. Indeed, momentum has a distribution with many small
gains and few, but extreme, large losses. Taking this into account the momentum
puzzle of Jegadeesh and Titman (1993) is substantially diminished. Our discus-
sion in terms of utility shows that rare observations with large losses are signi…cant
enough to change the interest of momentum for a risk-averse investor holding the
market.
In contrast, risk-managed momentum produces large economic gains. These
come from higher returns when compared to the market (a 19.92 p.p. gain) and
12

less crash risk than the market with plain momentum (a 14.96 p.p. gain). As a
result, the annual certainty equivalent of the market with risk-managed momentum
is 13.54 percent, which compares very favorably to the 0.14 percent of the market
alone and even more so with the -5.46 percent of the market combined with simple
momentum.
Comparing the strategies with plain and scaled momentum, risk management
produces a 3.13 p.p. gain in returns and a 15.86 p.p. gain from reduction of
risk. So the bulk of the gains comes from less risk, especially high-order moments.
Essentially, scaling momentum eliminates non-normal risk without sacri…cing re-
turns.
6 Anatomy of momentum risk
A well documented result in the momentum literature is that momentum has time-
varying market risk (Grundy and Martin (2001)). This is an intuitive …nding since
after bear markets winners tend to be low-beta stocks and the inverse for losers.
But Daniel and Moskowitz (2011) show that using betas to hedge risk in real time
does not work. This contrasts with our …nding that the risk of momentum is highly
predictable and managing it o¤ers strong gains. Why is scaling with forecasted
variances so di¤erent from hedging with market betas? We show it is because
time-varying betas are not the main source of predictability in momentum risk.
We use a CAPM regression to decompose the risk of momentum into systematic
and speci…c risk:
RVwml;t = 2
t RVrmrf;t + 2
e;t (12)
The realized variances and betas are estimated with 6-months of daily returns.
On average, the systematic component 2
t RVrmrf;t accounts for only 23 percent of
the total risk of momentum. Almost 80 percent of the momentum risk is speci…c.
Also, the di¤erent components do not have the same predictability. Table 5 shows
the results of an AR (1) on each component of risk.
Either in-sample or out-of-sample (OOS), 2
t is the least predictable component
of momentum risk. Its OOS R-square is only 5 percent. The realized variance of
13

the market also has a small OOS R-square of 7 percent. When combined, both
elements of systematic risk component already show more predictability (OOS R-
square of 21 percent) but still less than the realized variance of momentum (OOS
R-square of 44 percent). The most predictable component of momentum variance
is the speci…c risk with an OOS R-square of 47 percent, more than double the
predictability of the systematic risk.
Hedging with betas alone, as in Daniel and Moskowitz (2011) fails because
most of the risk is left out.
7 Conclusion
Unconditional momentum has a distribution that is far from normal, with huge
crash risk. We …nd that taking this crash risk into consideration, momentum is
not appealing for a risk-averse investor.
Yet, the risk of momentum is highly predictable. Managing this risk eliminates
exposure to crashes and increases the Sharpe ratio of the strategy substantially.
As a result, the momentum puzzle shows up stronger in its risk-managed version
and the case for a peso explanation of momentum returns is severely weakened.
References
[1] Asness, C. S., J. M. Liew, and R. L. Stevens (1997). “Parallels between the
cross-sectional predictability of stock and country returns”, Journal of Port-
folio Management 23, pp. 79-87.
[2] Asness, C. S., T. Moskowitz, and L. Pedersen (2011). “Value and momentum
everywhere”, University of Chicago, working paper.
[3] Bliss, R., and N. Panigirzoglou (2004). “Option-Implied Risk Aversion Esti-
mates”, Journal of Finance, Vol LIX, 1, pp. 407-446.
[4] Bollerslev, T. (1987), “A Conditional Heteroskedastic Time Series Model for
Speculative Prices and Rates of Return”, Review of Economics and Statistics,
69, pp. 542-547.
14

[5] Carhart, M. (1997). “On Persistence in Mutual Fund Performance”, The Jour-
nal of Finance, vol. 52, no. 1, pp. 57-82.
[6] Daniel, K., and T. Moskowitz (2011).“Momentum Crashes”, Working paper.
[7] Engle, R. (1982). “Autoregressive Conditional Heteroscedasticity with Esti-
mates of the Variance of U.K. In‡ation”, Econometrica 50, pp. 987-1008.
[8] Engle, R., and T. Bollerslev (1986). “Modeling the Persistence of Conditional
Variances”, Econometric Reviews, 5, pp. 1-50.
[9] Erb, C. B., and C. R. Harvey (2006). “The strategic and tactical value of
commodity futures”, Financial Analysts Journal 62, pp. 69-97.
[10] Fama, E. and K. French (1992). “The Cross-Section of Expected Stock Re-
turns”, The Journal of Finance, vol. 47, no. 2, pp. 427-465.
[11] Fama, E. and K. French (1996). “Multifactor Explanations of Asset Pricing
Anomalies”, The Journal of Finance, vol. 51, no. 1, pp. 55-84.
[12] Grinblatt, M., and S. Titman (1989). “Mutual fund performance: An analysis
of quarterly portfolio holdings”, Journal of Business, 62, pp. 394-415
[13] Grinblatt, M., and S. Titman (1993). “Performance measurement without
benchmarks: An examination of mutual fund returns”, Journal of Business,
66, pp. 47-68.
[14] Grundy, B., and J. S. Martin (2001). “Understanding the nature of the risks
and the source of the rewards to momentum investing”, Review of Financial
Studies, 14, pp. 29-78.
[15] Jegadeesh, N., and S. Titman (1993). “Returns to Buying Winners and Selling
Losers: Implications for Stock Market E¢ ciency”, The Journal of Finance,
vol. 48, no
1, pp. 65-91.
[16] Menkho¤, L., L. Sarno, M. Schmeling, and A. Schrimpf (2012). “Currency
Momentum Strategies”, Journal of Financial Economics, forthcoming
15

[17] Moskowitz, T., and M. Grinblatt (1999). “Do Industries Explain Momen-
tum?”, Journal of Finance, vol. 54, no. 4, pp. 1249-1290.
[18] Okunev, J., and D. White (2003). “Do momentum-based strategies still work
in foreign currency markets?”, Journal of Financial and Quantitative Analysis
38, pp. 425-447.
[19] Schwert, G. (1989). “Why Does Stock Market Volatility Change over Time?”,
Journal of Finance, 44, pp. 1115-1154.
[20] Rouwenhorst, K. (1998). “International momentum strategies”, Journal of
Finance, 53, pp. 267-284.
[21] Rouwenhorst, K. (1999). “Local return factors and turnover in emerging stock
markets”, Journal of Finance 54, pp. 1439-1464.
16

Annex: Data sources
We obtain daily and monthly returns for the market portfolio, the high-minus-low,
the small-minus-big, the ten momentum-sorted portfolios and the risk-free (one-
month Treasury-bill return) from Kenneth French’s data library. The monthly
data is from July 1926 to December 2011 and the daily data is from July 1963 to
December 2011.
For the period from July 1926 to June 1963, we use daily excess returns on
the market portfolio (the value-weighted return of all …rms on NYSE, AMEX
and Nasdaq) from the Center for Research in Security Prices (CRSP). We also
have daily returns for ten value-weighted portfolios sorted on previous momentum
from Daniel and Moskowitz (2011). This allows us to work with a long sample
of daily returns for the winner-minus-losers (WML) strategy from August 1926 to
December 2011. We use these daily returns to calculate the realized variances in
the previous 21, 63 and 126 sessions at the end of each month.
For the momentum portfolios, all stocks in the NYSE, AMEX and Nasdaq
universe are ranked according to returns from month t-12 to t-2, then classi…ed
into deciles according to NYSE cuto¤s. So there is an equal number of NYSE …rms
in each bin. The WML strategy consists on shorting the lowest (loser) decile and
a long position in the highest (winner) decile. Individual …rms are value weighted
in each decile. Following the convention in the literature, the formation period for
month t excludes the returns in the preceding month. See Daniel and Moskowitz
(2011) for a more detailed description of how they build momentum portfolios. The
procedures (and results) are very similar to those of the Fama-French momentum
portfolios for the 1963:07-2011:12 sample.
17

Tables
Max Min Mean STD KURT SKEW SR
RMRF 38.27 -29.04 7.33 18.96 7.35 0.17 0.39
SMB 39.04 -16.62 2.99 11.52 21.99 2.17 0.26
HML 35.48 -13.45 4.50 12.38 15.63 1.84 0.36
WML 26.18 -78.96 14.46 27.53 18.24 -2.47 0.53
Table 1. The long-run performance of momentum compared to the Fama-French
factors. ‘KURT’stands for excess kurtosis and ‘SR’for Sharpe ratio. The sample
returns are from 1927:03 to 2011:12.
t-stat t-stat R2
OOS R2
Panel A: 1927:03 to 2011:12
RMRF 0.0010 6.86 0.60 23.92 36.03 38.81 12.81 7.82
WML 0.0012 5.21 0.70 31.31 49.10 57.82 15.03 12.26
Panel B: 1963:07 to 2011:12
RMRF 0.0009 5.65 0.58 17.10 33.55 25.46 13.76 8.48
SMB 0.0004 8.01 0.33 8.32 10.68 -8.41 7.36 3.87
HML 0.0001 4.88 0.73 25.84 53.55 53.37 6.68 4.29
WML 0.0009 3.00 0.77 29.29 59.71 55.26 16.40 13.77
Table 2. AR (1) of 1-month realized variances. The realized variances are the
sum of squared daily returns in each month. The AR (1) regresses each realized
variance on its lag and a constant. The OOS R-squared uses the …rst 240 months
to run an initial regression so producing an OOS forecast. Then uses an expanding
window of observations till the end of the sample. In panel A the sample period
is from 1927:03 to 2011:12. In panel B we repeat the regressions for RMRF and
WML and add the same information for the HML and SMB. The last two columns
show, respectively, the average realized volatility and its standard deviation.
18

Max Min Mean STD KURT SKEW SR
WML 26.18 -78.96 14.46 27.53 18.24 -2.47 0.53
WML 21.95 -28.40 16.50 16.95 2.68 -0.42 0.97
Table 3. Momentum and the economic gains from scaling. The …rst row presents
as a benchmark the economic performance of plain momentum from 1927:03 to
2011:12. The second row presents the performance of risk-managed momentum.
The risk-managed momentum uses the realized variance in the previous 6 months.
CE( 1) CE( 2) CE( i>2) CE(r)
RM 11.72 -7.39 -4.18 0.14
RM+WML 28.51 -13.90 -20.07 -5.46
RM+WML* 31.64 -13.00 -5.11 13.54
Table 4. The economic performance of momentum for a representative investor.
The …rst row shows the performance of the market portfolio. The second row
combines the market portfolio with momentum and the third one with scaled
momentum. The …rst column shows the certainty equivalent of the mean return
of each strategy. The second and third columns present the contribution to the
certainty equivalent of standard deviation and higher moments, respectively. The
last column shows the certainty equivalent obtained from annual non-overlapping
returns. The returns are from 1927:03 to 2011:12. The decomposition uses a Taylor
expansion of the utility function around the mean return of the portfolio with a
CRRA of 4.
t-stat t-stat R2
R2
OOS
2
wml 0.0012 2.59 0.70 12.58 0.49 0.44
2
rmrf 0.0012 4.29 0.50 7.37 0.25 0.07
2
0.3544 6.05 0.21 2.83 0.05 0.05
2 2
rmrf 0.0007 2.73 0.47 6.80 0.22 0.21
2
" 0.0007 2.69 0.72 13.51 0.52 0.47
Table 5. Risk decomposition of momentum risk. Each row shows the results of
an AR (1) for 3-month, non-overlapping periods. The …rst row is for the realized
variance of the WML and the second one the realized variance of the market. The
third row is squared beta, estimated as a simple regression of 63 daily returns of the
WML on RMRF. The fourth row is the systematic component of momentum risk
and the last row the speci…c component. The OOS R-squares use an expanding
window of observations after an initial in-sample period of 20 years.
19

Figures
30 35 40 45 50 55 60 65 70 75 80 85 90 95 00 05 10
10
0
10
1
10
2
10
3
10
4
10
5
Mom and FF factors: 1927:03-2011:12
cumulativereturns(logscale)
$2,135
$145
$475
$68,741
RMRF
SMB
HML
WML
Figure 1. The long-run cumulative returns of momentum compared to the Fama-
French factors. Each strategy consists on investing $1 at the beginning of the
sample in the risk-free rate and combine it with the respective long-short portfolio.
The proceeds are fully reinvested till the end of the sample. On the right is the
terminal value of each strategy.
20

30 31 32 33 34 35 36 37 38 39
0
0.5
1
1.5
2
2.5
Mom and the market: 1930:01 to 1939:12
cumulativereturns
$0.98
$0.1
00 01 02 03 04 05 06 07 08 09
0
0.5
1
1.5
2
2.5
3
3.5
Mom and the market: 2000:01 to 2009:12
cumulativereturns
$1.02
$0.72
RMRF WML
Figure 2. Momentum crashes. The …gure plots the cumulative return and ter-
minal value of the momentum and market portfolio strategies in its two most
turbulent periods: the 1930’s and the 2000’s.
21

30 35 40 45 50 55 60 65 70 75 80 85 90 95 00 05 10
10
0
10
1
10
2
10
3
10
4
10
5
10
6
Risk-managed momentum: 1927:03-2011:12
cumulativelogreturns
$68,741
$6,140,075
WML
WML*
Figure 3. The long-run performance of risk-managed momentum. The risk-
managed momentum (WML*) scales the exposure to momentum using the realized
variance in the previous 6-months. In the beginning of the sample the strategy
invests $1 in the risk-free asset and combines it with the long-short portfolio. The
proceeds are fully reinvested till the end of the sample. On the right is the terminal
value of the strategy.
22

30 35 40 45 50 55 60 65 70 75 80 85 90 95 00 05 10
20
40
60
80
100
120
The monthly realized volatility of momentum
annualizedinp.p.
Figure 4. The realized volatility of momentum obtained from daily returns in
each month from 1927:03 to 2011:12.
23

-80 -60 -40 -20 0 20 40
0
2
4
6
8
10
Density of WML and WML* monthly returns
-80 -70 -60 -50 -40 -30 -20 -10
0
0.2
0.4
0.6
0.8
1
Density of WML and WML* for returns lower than -10 p.p.
WML
WML*
WML
WML*
Figure 5. The density of plain momentum (WML) and risk-managed momentum
(WML*). The risk-managed momentum uses the realized variance in the previous
6 months to scale the exposure to momentum. The returns are from 1927:03 to
2011:12.
24

30 31 32 33 34 35 36 37 38 39
10
-1
10
0
Risk-managed momentum: 1930:01 to 1939:12
$0.1
$1
00 01 02 03 04 05 06 07 08 09
10
0
Risk-managed momentum: 2000:01 to 2009:12
$0.72
$1.88
WML WML*
Figure 6. The bene…ts of risk-management in the 1930’s and the 00’s. The risk-
managed momentum (WML*) uses the realized variance in the previous 6 months
to scale the exposure to momentum.
25

30 35 40 45 50 55 60 65 70 75 80 85 90 95 00 05 10
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
The weights of the strategy: 1927:03-2011:12
weightonWML
Figure 7. Weights of the scaled momentum. The risk-managed momentum uses
the realized variance in the previous 6 months to scale the exposure to momentum.
26

Mngng rsk mmntm

More Related Content

What's hot (19)

Viewers also liked (7)

Similar to Mngng rsk mmntm (20)

More from ivan weinel (6)

Recently uploaded (20)

Mngng rsk mmntm