Asset Management and Investment Strategies

Lecture 4

Andrea Buraschi

Spring 2015
Road Map

I Timing the Market: Out of Sample Evidence

I From a Long-Only Approach to a Long-Short Approach
I Empirical methods and results
I The 3-Factor models (1992)
I The 5-Factor model (2015)
 Market Timing:
Out-of-Sample Evidence
Can We Time the Market?

I Lettau and Ludvigson (2001, p.842) “It is now widely accepted that
excess returns are predictable by variables such as dividend-price
ratios, earnings-price ratios, dividend-earnings ratios, and an
assortment of other …nancial indicators.”
I The typical speci…cation is based on regressing some lagged
predictor on the stock market rate of return or the equity premium.

Rt +T = a + bXt + εt +T

I This may suggest that if you were an asset manager, these variables
are potent state variables to form portfolios to exploit time-series
predictability.
I Goyal and Welch (2008) is a comprehensive study of the empirical
evidence of each of these variables using the same methods,
time-periods, and estimation frequencies.
Can We Time the Market?

I Many tests in the literature may su¤er of “look-ahead” bias: the

parameters at time-t are estimated using advance knowledge about
future data, not using exclusively data available prior to time-t.
I Goyal and Welch compare the IS (in-sample) and OOS
(out-of-sample) performance
I Main punchline: For many models, any earlier apparent statistical
signi…cance was often based exclusively on years up to and especially
on the years of the Oil Shock of 1973–1975.
I Most models have poor out-of-sample (OOS) performance.
Can We Time the Market?
Can We Time the Market?
Can We Time the Market?

The role of oil shock period

Conclusion

• dp (and many other models) perform well IS not OOS

• The Oil Shock recession period of 1973–1975 makes
the difference. During his period, the S&P500
dropped from 108.29 in October 1973 to 63.54 in
September 1974— and its recovery back to 95.19 in
June 1975.

• Many models depend on it for their apparent

forecasting ability, often both IS and OOS. (And none
performs well thereafter.)

• Term spread: tms has positive OOS performance only

if forecasting begins in 1965. Using 1927–2005 data
and starting forecasts in 1947, the OOS RMSE and R2
are negative.

Explanation: These figures plot the IS and OOS performance of annual predictive
regressions. Specifically, these are the cumulative squared prediction errors of the NULL
minus the cumulative squared prediction error of the ALTERNATIVE. The ALTERNATIVE is a
model that relies on predictive variables noted in each graph. The NULL is the prevailing
equity premium mean for the OOS graph, and the full-period equity premium mean for the
IS graph. The IS prediction relative performance is dotted (and usually above), the OOS
prediction relative perfomance is solid. An increase in a line indicates better performance of
the named model; a decrease in a line indicates better performance of the NULL. The blue
band is the equivalent of 95% two-sided levels, based on MSE-T critical values from
McCracken (2004). (MSE-T is the Diebold and Mariano (1995) t-statistic modified by Harvey,
Leybourne, and Newbold (1998)). The right axis shifts the zero point to 1965. The Oil Shock
is marked by a red vertical line.
A Possible Answer to Goyal and Welch

I They focus on mean forecasts. Not on the combination of mean and

variance. The use of dp ratios may reduce the variance and increase
Sharpe ratios, or utility equivalent. Indeed Campbell and Thompson
show that even if OOS some models have R 2 < 0, they produce
higher utility.
I The OOS test experiment is throwing away a lot of data. The
parameters are estimated only once, in the very early part of the
year, then the OOS period starts later on. This is not what an asset
manager would do either.
The Asset Management Industry

I If you believe in the CAPM and no market timing: Buy the Market
and just wait.
I If you believe in market timing : Long-Only approach (change your
βm with discretionary or systematic approach)
I If you do not believe in market timing: Long-Short approach (limit
your βm ).
... Long-Short Strategies are the topic of this lecture..
Market Timing: Systematic CTA
I Substantial increase in AuM in both Systematic CTA (market predictability)
Market Timing: Long-Short Equity
I and also in Equity Long-Short (cross-sectional predictability)
The CAPM
I At the beginning there was the CAPM (1964) ...
I E¢ cient frontier: upper edge of the risk-return space. If no risk free
rate (i.e. zero standard deviation) is available, portfolios on the
e¢ cient frontier dominate all other portfolios
I Capital market line: set of the dominating portfolios when a risk free
securitiy can be traded
Capital Asset Pricing Model (CAPM)

I Reference equilibrium pricing model in the literature

I Security market line (SML):

E [ Ri ] = Rf + β i [ E [ RM ] Rf ]

where
Cov [Ri , RM ]
βi =
σ2M
I Beta: sensitivity of the asset returns to systematic risk
I Since all agents diversify, only non diversi…able risk matters
Empirical tests of the CAPM

I Lintner (1965) tests the model by plotting average returns against

the betas of individual stocks
I Issues:
1. The beta of a single …rm changes over time (owing to variation in
size, business model, leverage etc.)
2. Betas are estimated with measurement error: this biases regression
estimates
3. Individual stock returns are so volatile that we cannot reject the
hypothesis that all the average returns are the same
I In the 1970s researchers address the issue by sorting stocks into
portfolios: the sensitivity of portfolios to market wide risks is more
stable than that of single stocks, and di¤erence among portfolios’
returns are statistically signi…cant
To address the previous concerns the robust and rigorous approach is:
I Sort stocks into portfolios based on some characteristics which could
be associated with average returns, and check if there is a di¤erence
in average returns between portfolios
I Compute betas for the portfolios, and check whether the average
return spread is explained by the spread in betas
I If not, there is an anomaly, and we need to consider multiple betas
(i.e. additional risk factors)
What does the …gure in the previous slide tell us?

I CAPM is not doing a bad job in explaining the overall risk-return

features of available asset classes. In particular, the long term and
corporate bonds have mean returns in line with their low betas,
despite their standard deviations nearly as high as those of stocks

I ...BUT: the smallest …rms seem to earn an average return a few

percent higher than predicted by CAPM. This is the famous
"small-…rm e¤ect"! (Banz, 1981)
Traditional Asset pricing anomalies

Empirical …ndings: patterns in average stock return that cannot be

explained by the CAPM
Anomalies / Explanations
I BE/ME and Size e¤ects (Banz, 1981; Rosenberg, Reid, and
Lanstein, 1981): (a) Size of a …rm and its return are inversely
related; (b) stocks with high book-to-market (value stocks) ratios
earn higher returns than stocks with low book-to-market ratios
(growth stocks).
I Reversal ( DeBondt and Thaler, 1985): the long-term past losers
bounce back, whereas the long-term past winners earn lower
subsequent returns.
More Recent Findings

I Quality and Pro…tability (Novy-Marx (2013); Haugen and Baker,

1996 and Cohen, Gompers, and Vuolteenaho, 2002): more pro…table
…rms have higher average stock returns
I Investments and Capital Expenditure: low capex and low
investments …rms earn extra returns not explained by CAPM
I Low Volatility Strategy and Low Beta Strategies: (a) Ang, Hodrick
and Zing, Thang, “High Idiosyncratic Volatility and Low Returns:
International and Further U.S. Evidence,” Journal of Portfolio
Management, Fall 2007; (b) Frazzini, A., and L. Pedersen, 2013,
“Betting against Beta”)
I Net stock issues (Daniel and Titman, 2006 and Ponti¤ and
Woodgate, 2008): there is a negative relation between net stock
issues and average returns
I Accruals (Sloan, 1996): higher accruals predict lower stock returns
Multifactor models: Fama-French (1992) and Carhart

I Do these anomalies point to irrationality or to risks which are not

captured by the covariance with marginal utility of wealth?
I To account for asset pricing anomalies, multifactor-models were
developed to incorporate additional risk factors
I Fama and French (1993) 3-factor model:

Ri Rf = ai + bi (RM Rf ) + si SMB + hi HML + ei

I We will discuss a further extension later, the Carhart (1997) 4-factor

model:

Ri Rf = ai + bi (RM Rf ) + si SMB + hi HML + pi PR1YR + ei

The size and book portfolios

I Fama and French (1996) form 25 portfolios based on size and B/M
ratio. The summary statistics (table 1) are:

I Small stocks tend to have higher returns than big stocks, and high
book/market stocks have higher returns than the low book/market
stocks!
How to detect an anomaly

I The standard approach is to sort returns based on the “anomaly

variables”, then ask if the standard model (say CAPM) can explain
the dispersion in average excess returns.
I For instance: Can the beta on the market explain alone the
di¤erence in excess returns of small vs. large …rms?
I Sorts help in highlighting the presence of a non-linear functional form
in the relation between average returns and the anomaly variable.
I However, they are not a substitute for a multiple regression slopes,
which provides direct estimates of marginal e¤ect (Fama MacBeth
(1973))
Empirical …ndings based on sorts
I Average returns vs. market beta for Fama-French 25 portfolios
(Cochrane 2004, Figure 20.9)

I CAPM is a disaster when confronted with these portfolios!

I There is a lot of variation which is not explained by beta (e.g.,
around beta 1.2 the ratio of highest to lowest return is around 3)
I This is the "value-size puzzle": value (growth) stocks systematically
overperform (underperform) the CAPM prediction
 Let’s connect the portfolios that have di¤erent size
within the same book/market category (Figure 20.10):

45o
small

Growth stocks

large
Growth stocks

Excess Returns should be on the 45 degres line: proportional to Beta on

market; Size produces a variation in average returns that is positively
related to variation in market betas.
 Let’s connects the portfolios that have di¤erent book/market
within the same size category

45o

High B/M

small

Low B/M
large

I Variation in B/M produces a variation in average returns that is

negatively related to variation in market betas: the red arrow in the
…gure shows the trend (from low to high)
I Conclusion: the value e¤ect is the main factor which leads to the
Empirical …ndings based on regressions: Fama and French

I TS regression model:

Ri ,t Rf ,t = ai + bi (RM ,t Rf ,t ) + si SMBt + hi HMLt + ei ,t

I New regressors: SMBt and HMLt

I Assumption: forming portfolios periodically on size and BE/ME
results in loadings on the three factors that are roughly constant
I If the three factor model describes expected returns, the regression
intercepts should be close to 0
I The F test of Gibbons, Ross, and Shanken (GRS 1989) is employed
to test the hypothesis that the regression intercepts for a set of
portfolios are all 0
Figure: Three-factor regressions for monthly excess returns of the 25
size-value portfolios (Fama and French, 1996)
I The model captures most of the variation in the average returns: the
average R 2 is 0.93, and the average absolute intercept is only 0.093
I However, the estimated intercepts shows that the model leaves a
large negative unexplained return for the portfolio of stocks in the
smallest size and lowest BE/ME quintiles, and a large positive
unexplained return for the portfolio of stocks in the largest size and
lowest BE/ME quintiles
 45o

Growth

Predicted

45o

Big

Predicted

The …t is much better; the worst case is for he Growt stocks in the pic
above
Time-Series Anomalies
I Two important articles:
I Fama and French (1996) "Multifactor explanations of asset pricing
anomalies";
I Fama and French (2008), “Dissecting Anomalies”
I Are there properties in stock returns that the Three-factor model
cannot explain? If so, these would be called “anomalies”
I “Long-term Reversal”
I "Short-term Momentum”
.. But now they ask “Can these e¤ect be explained simply by the
di¤erent exposure to HmL and SmB”?
I Reversal: Long-term past losers load more on SMB and HML, so
they behave more like small distressed stocks, and the model predicts
that the long-term past losers will have higher average returns in the
future: the reversal e¤ect can be explained by this model
I Momentum: The model however fails to account for the
momentum e¤ect. Short-term past losers also load more on SMB
and HML than short-term past winners: hence, the three-factor
model predicts they should have high average returns. Their model
cannot explain momentum.
Latest Developments in
Factor Models
Charlie Munger - Berkshire Hathaway Vice-Chairman
I In 1934, Ben Graham advocated buying stocks at a steep discount to their
intrinsic value, giving birth to “value investing.”
I Warren Bu¤ett — who took Graham’s class at Columbia in 1951 — saw the
approach had merit and used it to become a billionaire.
I Charlie Munger (Berkshire Vice-Chairman) improves Bu¤ett’s approach to
quality companies at fair prices. American Express (AMEX) and Coca-Cola (KO)
were prime examples.
I Quality + Value then became the two pillars of Berkshire Hathaway.
I Example: Berkshire’s $44 billion acquisition of Burlington Northern Santa Fe.
P/E was not “cheap” but the compan
Munger and Novy–Marx

I Looking at NYSE …rms between 1963 and 2010 and international …rms between
1990 and 2009 (ex-…nancials), Novy-Marx discovered that a company’s gross
pro…tability did as good a job at predicting its future returns as conventional
value metrics like book-to-market.
I More pro…table companies today tend to be more pro…table companies
tomorrow. Although it gets re‡ected in their future stock prices, the market
systematically underestimates this today, making their shares a relative bargain –
diamonds in the rough.
I Gross pro…tability = (total revenues - costs of goods or services it sells) / Assets.
I Wait a minute, you object: this leaves out half the income statement. What
about all the expenditures — shouldn’t we take these into account to determine
the business’s true intrinsic pro…tability? The answer is “no,” and that’s
Novy-Marx’s insight
I The further down the income statement we go, the more the mind becomes
clouded by Jedi accounting tricks that occlude rather than reveal true value,
until …nally the bottom-line ‘net’pro…tability becomes completely useless as a
predictor. The top line is the bottom line.
I Novy-Marx found that $1 in July 1973 grew to over $80 by 2011, that same
dollar invested in value + pro…table stocks grew to $572.
Quality
Novy-Marx: Investing in Quality

I Form portfolio based on 2 screens: Quality + Value.

I Quality screen: don’t buy cheap value stocks that are cheap because
they are bad
I Value screen: don’t overpay for the quality stocks that are expensive.
I What is the performance of using both screens at the same time.
Investing in Quality
Investing in Quality
Investing in Quality
Fama French: The 5 Factor model

I In the beginning (1964), there was the One-Factor Model, also known as the
Capital Asset Pricing Model. That factor was called beta. Beta was the measure
of how much each stock moved in relation to the stock market as a whole.
I The latest empirical model adds 2 new factors: (a) Pro…tability (Robert
Novy-Marx) and (b) Investment.
Investment

I Imagine that a company announces they are going to invest a lot of money in
some new project. Is this good news or bad news? Should you buy or sell?
I Titman, Wie and Xie (2004) controlled for the relevant variables and found that
…rms that signi…cantly increase capital investment tend to achieve sub-par
subsequent returns.
I Most managers become capital destroyers.
I By the new model, the highest expected returns can be expected from
companies that are small, value (high book-to-market, for example), pro…table,
AND are not embarking on major growth initiatives.
I What is a recent example of the opposite? Amazon.com (big investments)
Fama and French (2015)
Fama and French (2015)
What about HmL?

I HML (i.e Value): a redundant factor?

I “The …ve-factor model never improves the description of average returns from
the four-factor model that drops HML (Table 5). The explanation is interesting.
The average HML return is captured by the exposures of HML to other factors.
Thus, in the …ve-factor model, HML is redundant for describing average returns,
at least in U.S. data for 1963-2013.”
What Else? Betting against beta

Frazzini and Pederson (2013) …nd empirical evidence that:

1. Since constrained investors bid up high-beta assets, high beta is
associated with low alpha, as we …nd empirically for U.S. equities,
20 international equity markets, Treasury bonds, corporate bonds,
and futures;
2. A betting-against-beta (BAB) factor, which is long leveraged
low-beta assets and short high-beta assets, produces signi…cant
positive risk-adjusted returns;
3. When funding constraints tighten, the return of the BAB factor is
low;
4. Increased funding liquidity risk compresses betas toward one;
5. More constrained investors hold riskier assets.
They present a model with leverage and margin constraints that vary
across investors and time that is consistent with these …ve empirical
results.
Thanks!