Quant Signals: Performance
and Promising Ideas
Narasimhan Jegadeesh
Emory University and NBER
Nomura Global Quantitative Equity Conference
London, 2010 1
Outline
Traditional Signals – Recent performance
Price momentum and Earnings momentum
Sell-side Analysts’ recommendation
Valuation ratios: Book-to-market and Cash-flow to
price
Lessons from 2009
Promising Ideas
Exploit Deviation from fundamentals
Combine Momentum with fundamentals
Exploit biases in analysts’ forecasts
2
Price Momentum
Sample:
US stocks
Exclude stocks priced < $5
Exclude stocks with market cap in the 20th
percentile of NYSE stocks
Strategy
Ranks stocks based on returns in months t-12 to
t-1
Buy Winner decile – sell loser decile. Hold for
month t.
3
Performance : Momentum (-12 to -2)
1990- 1990- 2000- 2009
2009 1999 2008
Mean 14.7 25.2 13.9 -83.7
(% P.A.)
SD 29.2 18.8 34.8
(% P.A.)
Sharpe .50 1.34 .40
Ratio
4
What was different about 2009?
Beta (winner minus loser portfolio)
1990-2008 : -.01
2009 : -.65
Losers were high beta stocks that were beaten
down with the market
Beta neutral portfolio increases Sharpe
ratio by 10%
Reduces 2009 loss from 83% to 63%
5
Valuation Ratio
Cash flow to price: Ratio of previous 12
month cash flow to recent price
CF-to-Price (winner minus loser portfolio)
1990-2008 : -3.9%
2009 : -13.8%
Losers were priced cheaply in 2009 – On average CF-to-Price
for Losers was 22.5% in 2009; Losers beaten down a lot more
than justified by cash flows.
Momentum profit (%) = 2.48+.29 CF-to-P difference (%)
(t stat 2.64)
6
Sell-Side Analysts’ Recommendations for
Past Winners and Losers
Winners Losers
1994-2008 2.4 2.0
2009 2.5 2.3
7
Some Lessons
Momentum strategies are highly volatile
because of a low correlation between
winners and losers
Any strategy based on technical signals
should also pay attention to
fundamentals and valuation
8
Earnings Momentum
Many variations of earnings momentum
strategy have been proposed in the
literature; e.g. Analyst forecast revision,
Standardized Unexpected Earnings etc.
Preferred Measure:
Actual Consensus Forecast the Previous Month
Earnings Surprise=
Std.Dev of Analysts' Forecast
9
Performance : Earnings Momentum
1990- 1990- 2000- 2009
2009 1999 2008
Mean 10.6 14.1 9.6 -18.8
(% P.A.)
SD 7.8 5.7 8.7
(% P.A.)
Sharpe 1.3 2.4 1.1
Ratio
Why is the risk of the earnings momentum strategy so much lower?10
Characteristics in 2009
Beta (Positive minus negative surprise
portfolio)
1990-2008 : -.13
2009 : -.14
Beta not very different in 2009
Yet, beta neutral portfolio increases Sharpe
ratio by 10%
Reduces 2009 loss from 19% to 10%
11
Valuation Ratio
Cash flow to price: Ratio of previous 12
month cash flow to recent price
CF-to-Price (Positive minus negative surprise portfolio)
1990-2008 : .5%
2009 : -1.2%
Earnings Mom profit (%) = .8+.19 CF-to-P difference (%)
(t stat 2.26)
12
Sell-Side Analysts’ Recommendations
# of upgrades - # of downgrades
Buy positive Fraction up and Fraction up=
# of upgrades + # of downgrades
sell negative over the
previous month
Hold for one month
13
Performance : Upgrades minus
Downgrades
1994- 1994- 2000- 2009
2009 1999 2008
Mean 4.5 7.9 1.9 6.9
(% P.A.)
SD 5.8 4.2 6.4
(% P.A.)
Sharpe .78 1.88 .30
Ratio
14
Valuation Ratios
Book-to-Price
Cash Flow-to-Price
Earnings-to-price – not as effective as
cash flow to price
15
Performance: Book-to-price
1990- 1990- 2000- 2009
2009 1999 2008
Mean 2.1 -5.1 7.4 33.6
(% P.A.)
SD 22.7 16.7 27.1
(% P.A.)
Sharpe .09 -.3 .27
Ratio
16
Characteristics: 2009
Beta (value minus growth portfolio)
1990-2008 : -.36
2009 : .34
Beta neutral portfolio increases Sharpe
ratio from .09 to .28
17
Valuation Ratio
Cash flow to price: Ratio of previous 12
month cash flow to recent price
CF-to-Price (value minus growth portfolio)
1990-2008 : 14%
2009 : 20%
Value minus Growth(%) = -1.4+.11 CF-to-P difference (%)
(t stat 1.15)
18
Performance: CF-to-price
1990- 1990- 2000- 2009
2009 1999 2008
Mean 15.4 10.4 19.8 35.6
(% P.A.)
SD 22.9 15.9 28.1
(% P.A.)
Sharpe .67 .65 .71
Ratio
19
Characteristics: 2009
Beta
1990-2008 : -.57
2009 : .45
Beta neutral portfolio increases Sharpe
ratio from .67 to .96
20
Valuation Ratio
Cash flow to price: Ratio of previous 12
month cash flow to recent price
CF-to-Price (high minus low portfolio)
1990-2008 : 31%
2009 : 42%
Value minus Growth(%) = -.59+.6 CF-to-P difference (%)
(t stat .99)
21
Correlation
Price Earnings Book-to-
Momentum Momentum price
Earnings .53
Momentum
Book-to- -.72 -.21
price
CF-to-Price -.41 .02 .82
22
Recent Ideas
Examine divergence between past returns and changes in
fundamentals (Dha et al., 2010, “Decomposing the Short-Term
Return Reversal”)
Buy winners with strong fundamentals and sell losers with weak
fundamentals (Lee and Shih, 2010, “Technical, Fundamental,
and Combined Information for Separating Winners from
Losers”)
Exploit biases in analysts’ forecasts (Green et al., 2010,
“Inferring Investor Sentiment From Analyst Forecasts”)
23
Past Returns and Changes in
Fundamentals
One-month change in fundamental value: Present value of
changes in cash flows implied by changes in analysts’ one- and
two-year ahead earnings forecast and long term growth
(Earnings growth rate assumed to linearly decline from the LTG
forecast to steady state over years +5 to +10)
Sort stocks based on previous month returns minus change in
fundamental value (Diff)
High Diff indicates returns too high to be justified by changes in
fundamentals and Low DIFF indicates returns too low relative to
change in fundamentals.
Low Diff portfolios should outperform high Diff portfolio
24
Strategy
Short-Horizon Return reversals:
Buy Decile of stocks with the lowest return
in the previous month and sell the highest
return stocks (Jegadeesh, JF 1990)
Diff Reversal
Buy Decile of stocks with the smallest (or
negative) Diff in the previous month and
sell the highest Diff stocks
25
Performance: Reversal and Diff
(1982-2008)
Reversal Diff
Mean 8.0 18.9
(% P.A.)
SD 14.5 9.3
(% P.A.)
Sharpe .56 2.1
Ratio
Second month profit for Diff strategy is 3.6% 26
Momentum and fundamentals
Lee and Shih (2010)
12-month momentum
Covariance between returns and abnormal trading
volume of the previous 12-months (Intuition –
larger the covariance, larger is the informed
trading)
Fundamentals based on Financial statements
F-Score for value firms (Piotroski, JAE, 2000)
G-Score for growth firms (Mohanram, RAS 2005)
27
Piotroski (JAE, 2000): F-score
Sum of the following indicator variables
1 if ROA>0; 0 otherwise
1 if cash flow > 0
1 if change in ROA>1
1 if accrual (Earnings-Cash flow)<0
1 if change in leverage <0
1 if change in current ratio >0
1 if No new equity issue in the last 12 months
1 if year-over-year increase in gross margin
1 if year-over-year increase in asset turnover
Large F-Score indicates strong and improving fundamentals
F-Scores predict returns for value firms (quintile of firms with largest
book-to-price ratio) High score minus low score portfolio earns about
10% per year over the sample period 1976-1996
28
G-Score (Mohanram, RAS 2005)
Sum of the following indicator variables
1 of ROA greater than median ROA for growth firms in the same
industry; and 0 otherwise
1 of Cash Flow ROA greater than industry median
1 if cash flow > earnings (negative accruals)
1 if earnings variability is less than industry median
1 if sales growth variability is less than industry median
1 if R&D/assets greater than industry median
1 if Capex/assets
1 if advertisement/sales greater than industry median
Large G-Score indicates better fundamentals than industry peers
G-Scores predict returns for Growth firms (quintile of firms with
smallest book-to-price ratio)
High score minus low score portfolio earns about 18% per year over
the sample period 1978-2001
29
Performance (1982-2007)
Signals Growth Value
Momentum Quintiles (-6 to -2) .91% .60%
(winner – Loser)
Momentum .84% .91%
+ Cov (unexpected Vol., ret)
Momentum 3.3% 1.78%
+ Cov (unexpected Vol., ret)
+G-Score/F-Score
Abnormal returns persist for up to 6 months 30
Strategy to directly exploit biases in
analysts’ expectations
Numerous papers document that analysts’ earnings
forecasts are biased
It is likely that stock for which analysts are most
favorably biased are overpriced and stocks for which
analysts are most unfavorably biased are underpriced
How would we identify analyst biases?
Green et al. (2010, Emory) measure bias as Analysts’
forecasts minus Statistical Forecast soon after
earnings announcements and construct a trading
strategy
31
Statistical Forecasts
Estimate statistical forecast for year-ahead EPS using
the following independent variables soon after fourth
quarter earnings announcements for December year-
end firms:
Past annual EPS
F-Score
Accrual
Earnings Volatility
EPS t 1 0.145 0.719 * EPS t 0.0126 * F _ SCORE t 0.3304 * ACCRUALS t 0.022 * ( Earningst / t 4 ) * EPS t
32
Timing
Form deciles for
December fiscal year
end firms by (AF-SF)
Returns from June year t
to May year t+1
Dec June Dec May June Dec May Dec
Compute mean AF and SF
Calendar year t-1 Calendar year t Calendar year t+1
Error=Analysts Forecast - Statistical forecast
Analysts consensus forecasts measured as the average of the first forecast
33
by each analyst after earnings announcement
Performance: Subperiods
Subperiods, Portfolio (AF-SF) ranking
D1 (Low AF-SF) D2 D3 D4 D5 D6 D7 D8 D9 D10 (High AF-SF) D1-D10
PanelA: 3 factor alpha
1981-1987 0.47** 0.47*** 0.66*** 0.44** 0.42*** 0.36 0.05 0.04 0.01 -0.20 0.67***
t-stat (2.65) (4.25) (4.10) (2.60) (3.85) (1.67) (0.22) (0.47) (0.07) (-0.83) (3.44)
1988-1994 0.49 0.46 0.52 0.33 0.24 0.03 -0.06 -0.23 (0.11) -0.39 0.88*
t-stat (1.63) (3.47) (6.30) (2.26) (1.66) (0.31) (-0.34) (-1.71) (0.67) (-1.92) (2.36)
1995-2001 0.31 0.49 0.21 0.01 0.01 0.01 -0.02 -0.28* -0.34 -0.71*** 1.01***
t-stat (1.11) (1.85) (1.41) (0.06) (0.06) (0.05) (-0.07) (-2.19) (-1.81) (-6.47) (3.72)
2002-2008 1.02*** 0.37 0.36 0.77*** 0.45 0.16 0.12 0.12 -0.18 -0.33 1.35***
t-stat (2.93) (1.75) (1.38) (3.62) (1.82) (1.24) (1.01) (0.99) (-0.92) (-1.35) (2.89)
PanelB: 4 factor alpha
1981-1987 0.64*** 0.72*** 1.03*** 0.68*** 0.75*** 0.49*** 0.41*** 0.19 -0.03 -0.04 0.68**
t-stat (3.71) (5.65) (6.11) (4.63) (5.25) (3.02) (3.07) (1.57) (-0.13) (-0.15) (2.69)
1988-1994 0.67** 0.57*** 0.55*** 0.43*** 0.38*** 0.06 -0.04 -0.21* -0.18 -0.29** 0.96**
t-stat (2.30) (3.60) (3.09) (5.56) (2.97) (0.55) (-0.81) (-2.03) (-1.93) (-2.28) (2.82)
1995-2001 1.10** 1.04** 0.76 0.41 0.50 0.44 0.18 -0.08 0.18 -0.16 1.26***
t-stat (2.37) (2.73) (1.57) (1.46) (1.32) (1.30) (0.74) (-0.22) (0.54) (-0.35) (5.30)
2002-2008 0.82** 0.47*** 0.48** 0.78*** 0.48*** 0.28 -0.04 -0.02 -0.27 -0.40 1.22**
t-stat (2.64) (5.42) (2.15) (3.64) (3.03) (1.78) (-0.72) (-0.18) (-1.75) (-1.96) (2.70)
• Raw return difference is of the same magnitude
as 3-factor alpha
•Average return for 1981 to 2008: 12%
34
•Sharpe Ratio: 1.17
Annual Returns
35
Exploiting Analysts’ Biases
Profits increase over time; investors
seem more focused on analysts’
forecasts in recent periods
Strategy could be improved when
complimented with momentum and
fundamental signals
36
Conclusion
Performance of momentum strategies weaker over
the last decade compared with the ’90s
Performance of Value strategies stronger over the
last decade compared with the ’90s
Value and momentum strategies are negatively
correlated
More money likely chasing momentum after the
strong performance in the ’90s and the poor
performance of value strategies
Important to pay attention to valuation even when
applying momentum strategies
37
Conclusion
Evidence on some strategies that combine
past returns with measures of valuation is
promising
Short and intermediate horizon strategies
Biases in analysts’ earnings forecasts lead to
mispricing
Longer horizon strategies
More precise estimates of forecast biases
coupled with momentum and fundamental
signals could lead to improved strategies
38
