Quantitative Stock Selectioncharvey/Teaching/...If market is deemed “cheap,” (as many asset...
Transcript of Quantitative Stock Selectioncharvey/Teaching/...If market is deemed “cheap,” (as many asset...
1
Quantitative Stock SelectionQuantitative Stock Selection
Campbell R. Harvey
Global Asset Allocation and Stock Selection
Quantitative Stock Selection1. Introduction
Research coauthored with• Dana Achour• Greg Hopkins• Clive Lang
2
Quantitative Stock Selection1. Introduction
IssueTwo decisions are important:• Asset Allocation (country picks)• Asset Selection (equity picks)
Quantitative Stock Selection1. Introduction
Issue• Considerable research on the asset
allocation side• Research has paid off in that many models
avoided “overvalued” Asian markets inmid-1990s
• Many models began overweighing after theonset of the Asia Crisis
3
Quantitative Stock Selection1. Introduction
Issue• Little research on the stock selection side.
Why?– Sparse data on individual stocks– Information asymmetries among local and
global investors– Extremely high transactions costs
Quantitative Stock Selection1. Introduction
With recent plummet in emerging markets,stock selection is important.
If market is deemed “cheap,” (as manyasset allocation models would now suggest),which stocks do we select?
4
Quantitative Stock Selection2. Stock Selection Metrics
Ingredients for success:• Identify stable relationships• Attempt to model unstable relationships• Use predictor variables that reflect the
future, not necessarily the past• Do not overfit• Validate in up-markets as well as down• Tailor to country characteristics in emerging
markets
Quantitative Stock Selection2. Stock Selection Metrics
Methodologies:• Cross-sectional regression• Sorting• Hybrids
5
Quantitative Stock Selection2. Stock Selection Metrics
Cross-sectional regression:For country j, estimate:
wherei denotes firm i;A is a firm specific attribute (could be multiple)γ are common regression coefficients
tititi AR ,1,10, εγγ ++= −
Quantitative Stock Selection2. Stock Selection Metrics
Cross-sectional regression:• Used in developed market stock selection• Problem with unstable coefficients• Bigger problem given noisy emerging market
returns
6
Quantitative Stock Selection2. Stock Selection Metrics
Sorting:• Used in developed market stock selection• Potentially similar in stability problems• Can be cast in regression framework
– (a regression on ranks, or a multinomial probitregression)
• Rank regression may have advantages giventhe high variance (high noise) in emergingequity returns
Quantitative Stock Selection2. Stock Selection Metrics
Sorting:• Simple methodology that provides a good
starting point to investigate stock selection
7
Quantitative Stock Selection2. Stock Selection Metrics
Hybrid:• Create portfolios based on stocks sorted by
attributes• Use regression or optimization to weight
portfolios• Produces a flexible, highly nonlinear way to
select stocks
Quantitative Stock Selection3. Our methodology
Focus on three emerging markets:• Malaysia (representative of Asia)• Mexico (indicative of Latin America)• South Africa (unique situation)
8
Quantitative Stock Selection3. Our methodology
Specify exhaustive list of firm specific factors• Includes many traditional factors• Extra emphasis on expectations factors
Specific a number of diagnostic variables• Includes factors that reflect the type of firm we
are selecting
Quantitative Stock Selection3. Our methodology
Identify the best stocks and the worst stocks• Do not impose the constraints of a tracking
error methodology [Tracking error can be dealt with at a later
stage of the analysis]
9
Quantitative Stock Selection3. Our methodology
Steps:
1. Specify list of factors2. Univariate screens (in sample)3. Bivariate diagnostic screens4. Battery of additional diagnostics emphasizing performance through time5. Bivariate selection screens
Quantitative Stock Selection3. Our methodology
Steps:
6. Optimize to form “scoring screen” (in sample)7. Run scoring screen on out-of-sample period8. Diagnostics on scoring screen9. Form “buy list” and “sell lists”10. Purge “buy list” of stocks that are identified
by predetermined set of “knock out criteria”
10
Quantitative Stock Selection3. Our methodology
Steps:
11. Investigate turnover of portfolio– various holding periods analyzed
Quantitative Stock Selection4. Past research
Very few papers:• Rouwenhorst (JF) looks at IFC data• Claessens, Dasgupta and Glen (EMQ) look at
IFC data• Fama and French (JF) look at IFC data• Achour, Harvey, Hopkins, Lang (1998, 1999,
2000)
11
Quantitative Stock Selection4. Past research
What we offer:• No one has merged IFC, MSCI, Worldscope,
and IBES data• First paper to look at comprehensive list of
firm attributes• First paper to look at expectational attributes
Quantitative Stock Selection4. Factors
Fundamental factors• Dividend yield• Earnings yield• Book to price ratio• Cash earnings to price yield• Change in return on equity• Revenue growth• Rate of re-investment• Return on equity
12
Quantitative Stock Selection4. Factors
Expectational• Change in consensus FY1 estimate - last 3
or 6 months• Consensus FY2 to FY1 estimate change• Consensus forecast earnings estimate
revision ratio• 12 months prospective earnings growth rate• 3 year prospective earnings growth rate• 12 month prospective earnings yield
Quantitative Stock Selection4. Factors
Momentum• One month/ 1 year price momentum• One year historical earnings
growth/momentum• Three year historical earnings growth rate
13
Quantitative Stock Selection4. Factors
Diagnostic• Market capitalization• Debt to common equity ratio
Quantitative Stock Selection5. Diagnostics
• Average return• Average excess return• Standard deviation• T-stat (hypothesis that excess return=0)• Beta (against benchmark index)• Alpha• R2
14
Quantitative Stock Selection5. Diagnostics
• Average capitalization• % periods > market index (hit rate)• % periods > market index in up markets• % periods > market index in down markets• Max number of consecutive benchmark
outperformances
Quantitative Stock Selection5. Diagnostics
• Max observed excess return• Min observed excess return• Max number of consecutive negative returns• Max number of consecutive positive returns• Year by year returns
15
Quantitative Stock Selection5. Diagnostics
• Factor average for constructed portfolio• Factor median• Factor standard deviation
050
100150200250300350400
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Malaysia IFC US$ Malaysia FX
Quantitative Stock Selection6. Summary Statistics: Malaysia Benchmark
87% drop
Data through January 2001
16
0100200300400500600700800900
1000
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
Mexico IFC Mexico FX
Quantitative Stock Selection6. Summary Statistics: Mexico Benchmark
68% drop
Data through January 2001
0
50
100
150
200
250
300
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000
South Africa IFC US$ South Africa FX
Quantitative Stock Selection6. Summary Statistics: South Africa Benchmark
55% drop
Data through January 2001
17
-30
-25
-20
-15
-10
-5
0
5
10
15
CapD ROI
D/E Div
1 Yr E
arn M
om
3 Yr E
arn M
om E/P
D FYI 3m
o
D FYI 6m
o
FY1 t
o FY2
Rev Ratio B/P
CE/P
1 mo M
om
1 yr M
om
Prop E
/P
Prop 3
yr D Ea
rn
12 m
o Prop
E/P
24 m
o Prop
E/P
D Rev
Reinves
tROE
Index
return
Top Bottom
Quantitative Stock Selection6. Malaysia: Factor returns
-5
0
5
10
15
20
25
30
35
CapD ROI
D/E Div
1 Yr E
arn M
om
3 Yr E
arn M
om E/P
D FYI 3m
o
D FYI 6m
o
FY1 t
o FY2
Rev Ratio B/P
CE/P
1 mo M
om
1 yr M
om
Prop E
/P
Prop 3
yr D Ea
rn
12 m
o Prop
E/P
24 m
o Prop
E/P
D Rev
Reinves
tROE
Index
return
Top Bottom
Quantitative Stock Selection6. Mexico: Factor returns
18
0
5
10
15
20
25
30
CapD ROI
D/E Div
1 Yr E
arn M
om
3 Yr E
arn M
om E/P
D FYI 3m
o
D FYI 6m
o
FY1 t
o FY2
Rev Ratio B/P
CE/P
1 mo M
om
1 yr M
om
Prop E
/P
Prop 3
yr D Ea
rn
12 m
o Prop
E/P
24 m
o Prop
E/P
D Rev
Reinves
tROE
Index
return
Top Bottom
Quantitative Stock Selection6. South Africa: Factor returns
0
10
20
30
40
50
60
70
CapD ROI
D/E Div
1 Yr E
arn M
om
3 Yr E
arn M
om E/P
D FYI 3m
o
D FYI 6m
o
FY1 t
o FY2
Rev Ratio B/P
CE/P
1 mo M
om
1 yr M
om
Prop E
/P
Prop 3
yr D Ea
rn
12 m
o Prop
E/P
24 m
o Prop
E/P
D Rev
Reinves
tROE
Top Bottom
Quantitative Stock Selection6. Malaysia: % Periods Benchmark Outperformance
19
0
10
20
30
40
50
60
70
CapD ROI
D/E Div
1 Yr E
arn M
om
3 Yr E
arn M
om E/P
D FYI 3m
o
D FYI 6m
o
FY1 t
o FY2
Rev Ratio B/P
CE/P
1 mo M
om
1 yr M
om
Prop E
/P
Prop 3
yr D Ea
rn
12 m
o Prop
E/P
24 m
o Prop
E/P
D Rev
Reinves
tROE
Top Bottom
Quantitative Stock Selection6. Mexico: % Periods Benchmark Outperformance
0
10
20
30
40
50
60
70
CapD ROI
D/E Div
1 Yr E
arn M
om
3 Yr E
arn M
om E/P
D FYI 3m
o
D FYI 6m
o
FY1 t
o FY2
Rev Ratio B/P
CE/P
1 mo M
om
1 yr M
om
Prop E
/P
Prop 3
yr D Ea
rn
12 m
o Prop
E/P
24 m
o Prop
E/P
D Rev
Reinves
tROE
Top Bottom
Quantitative Stock Selection6. South Africa: % Periods Benchmark Outperformance
20
0
50
100
150
200
250
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Top Benchmark Bottom
Quantitative Stock Selection6. Malaysia: Dividend Yield Screen: Index=100 each year
0
50
100
150
200
250
300
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Top Benchmark Bottom
Quantitative Stock Selection6. Mexico: Historical Earnings Momentum Screen: Index=100 each year
21
0
20
40
6080
100
120140
160
180
200
1993
1994
1995
1996
1997
1998
Top Benchmark Bottom
Quantitative Stock Selection6. South Africa: Change in Consensus FY1-3 mo. Screen: Index=100 each year
-50-40-30-20-10
01020304050
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Mal
aysia
Mex
ico
Sout
h A
fric
aQuantitative Stock Selection6. Book to Price: Low-High Spread
22
-50-40-30-20-10
01020304050
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Mal
aysia
Mex
ico
Sout
h A
fric
a
Quantitative Stock Selection6. IBES Revision Ratio: Low-High Spread
-50-40-30-20-10
01020304050
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Mal
aysia
Mex
ico
Sout
h A
fric
aQuantitative Stock Selection6. IBES 12-month Prospective Earnings Yield: L-H Spread
23
-50-40-30-20-10
01020304050
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Mal
aysia
Mex
ico
Sout
h A
fric
a
Quantitative Stock Selection6. One-year Momentum: Low-High Spread
-50-40-30-20-10
01020304050
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Mal
aysia
Mex
ico
Sout
h A
fric
aQuantitative Stock Selection6. Size Effect: Low-High Spread
24
-20
-15
-10
-5
0
5
10
15
Monthly
Quarterl
y
Semian
nual
Semian
nual w
/KOMark
et
Top Bottom
Quantitative Stock Selection6. Malaysia: Scoring Screen Various Holding Periods
0
5
10
15
20
25
30
35
Monthly
Quarterl
y
Semian
nual
Market
Top Bottom
Quantitative Stock Selection6. Mexico: Scoring Screen Various Holding Periods
25
-10
-5
0
5
10
15
20
Monthly
Quarterl
y
Semian
nual
Market
Top Bottom
Quantitative Stock Selection6. South Africa: Scoring Screen Various Holding Periods
0102030405060708090
100
Monthly
Quarterl
y
Semian
nual
Semian
nual w
/KO
Top Bottom
Quantitative Stock Selection6. Malaysia: Scoring Screen % Periods Benchmark Outperformance
26
0102030405060708090
100
Monthly
Quarterl
y
Semian
nual
Top Bottom
Quantitative Stock Selection6. Mexico: Scoring Screen % Periods Benchmark Outperformance
0102030405060708090
100
Monthly
Quarterl
y
Semian
nual
Top Bottom
Quantitative Stock Selection6. South Africa: Scoring Screen % Periods Benchmark Outperformance
27
0
50
100
150
200
250
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Top Bottom
Quantitative Stock Selection6. Malaysia: Scoring Screen: Index=100 each year
0
50
100
150
200
250
300
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
Top Bottom
Quantitative Stock Selection6. Mexico: Scoring Screen: Index=100 each year
28
0
2040
60
80100
120140
160
180200
1993
1994
1995
1996
1997
1998
Top Bottom
Quantitative Stock Selection6. South Africa: Scoring Screen: Index=100 each year
0.00
100.00
200.00
300.00
400.00
500.00
600.00
700.00
800.00
900.00
12/31/88 12/31/89 12/31/90 12/31/91 12/31/92 12/31/93 12/31/94 12/31/95 12/31/96 12/31/97
CU
MU
LATI
VE
RET
UR
NS
- IN
SA
MPL
E
0.00
20.00
40.00
60.00
80.00
100.00
120.00
140.00
160.00
CU
MU
LATI
VE
RET
UR
NS
- OU
T O
F SA
MPL
E
IFCG MALAYSIA
IN SAMPLE OUT OF SAMPLE
TOP FR
IFCG MALAYSIA
BOTTOM
IN SAMPLE OUT OF SAMPLE
IBES DATA ADDED
Quantitative Stock Selection6. Malaysia: Scoring Screen
29
Quantitative Stock Selection6. Mexico: Scoring Screen
0.00
100.00
200.00
300.00
400.00
500.00
600.00
700.00
800.00
900.00
1000.00
1100.00
1200.00
1300.00
1400.00
1500.00
1600.00
1700.00
1800.00
1900.00
2000.00
2100.00
12/31/88 12/31/89 12/31/90 12/31/91 12/31/92 12/31/93 12/31/94 12/31/95 12/31/96 12/31/97
CU
MU
LATI
VE
RET
UR
NS
- IN
SA
MPL
E
0.00
50.00
100.00
150.00
200.00
250.00
CU
MU
LATI
VE
RET
UR
NS
- OU
T O
F SA
MPL
E
IN SAMPLE OUT OF SAMPLE
TOP
IFCG MEXICO
BOTTOM
IN SAMPLE OUT OF SAMPLE
Quantitative Stock Selection6. South Africa: Scoring Screen
0.00
50.00
100.00
150.00
200.00
250.00
300.00
350.00
12/31/92 12/31/93 12/31/94 12/31/95 12/31/96 12/31/97
CU
MU
LATI
VE
RET
UR
NS
- IN
SA
MPL
E
0.00
20.00
40.00
60.00
80.00
100.00
120.00
CU
MU
LATI
VE
RET
UR
NS
- OU
T O
F SA
MPL
E
TOP
BOTTOM
IFCG SOUTH AFRICA
IN SAMPLE OUT OF SAMPLE
30
Quantitative Stock Selection7. Research Directions
1) Comparison of regression method andmultivariate screening process– Panel multinomial probit models– How do we reduce the noise in emerging market
equity returns?
Quantitative Stock Selection7. Research Directions
2) What are the characteristics of countries thatmake some factors work and other not work?– Stage of market integration process– Industrial mix– Openness of economy– Microstructure factors
31
Quantitative Stock Selection7. Research Directions
3) What causes the shifting importance of factorsthrough time, e.g. value versus growth?– Can the cross-section of many stock returns help
us identify when a factor is likely to work?
Quantitative Stock Selection7. Research Directions
4) Can the country selection process be mergedwith the stock selection exercise?– Should “buy” portfolios be used in top-down
optimizations?– Does country-specific tracking error really matter
in global asset allocation?
32
Quantitative Stock Selection7. Research Directions
5) Should we expand our view of risk in both thestock selection and country selection exercises?– Mean, variance, skewness?– What are the driving forces of changing variance?– What are the determinants of skewness?