Presentation

advertisement
Text-Based Product Characteristics, Competition
and Dividends
Presented at 2011 UBC Winter Conference
Gerard Hoberg
Gordon Phillips
Nagpurnanand Prabhala
Robert H. Smith School of Business
University of Maryland, College Park
Research Question
 How the structure and evolution of a firm’s product space
shapes its payout policy.
 Text to characterize product space
Fluidity, competition, and product customer type
 Several payout decisions
Propensity to pay dividends
+ initiations, omissions, changes.
Repurchases
Dividend-repurchase substitution
2
Motivation: Fluidity
Brav, Graham, Harvey, Michaely (2005)
“Sustainability and stability of future earnings are the
most important determinants of payout” policy.”
Sustainable earnings less likely in fluid product markets in
flux due to rivals.
3
Blackberry
Nexus S
Smartphone
Dell Streak
Blackberry
Playbook
Sony S1
Tablets
Toshiba
Motorola Xoom
Samsung Galaxy
Acer Iconia
LG Slate
Motivation: Fluidity
We construct a new metric of fluidity from product text. Why
should this matter?
 Fluidity measures ex-ante threats. This can be quite
different from measured ex-post cash flow risk and both
can matter.
 Power. Product text is (a) voluminous; (b) timely. Thus, it
contains detailed, forward looking information about
product markets as seen by senior managers.,
AAPL
“Music”
“Phone”
# words
SIC
1999
0
0
867
3571
2009
24
35
3592
3571
6
Hypotheses
H1: Fluidity
- Firms facing fluid product markets are less likely to make
payouts. Especially via dividends.
H2: Competition and Differentiation
- Firms with differentiated products and in more protected
markets should be more likely to pay dividends.
H3: Business (not investor) clientele
- Business customers may value long-term stability to ensure
stable supply chain.
- If yes, then firms with more business (non-retail) clientele
should prefer dividends over repurchases..
7
Related Literature
 Product life cycle (Abernathy-Utterback, 1978)
- Stable products and dominant designs late in life cycle
favor payouts.
 Life cycle + agency (DeAngelo et al., 2009)
- Mature firms pay out to avoid agency problems.
- Competitive threat from product market fluidity
makes disciplining dividends less necessary.
 Firm maturity: DeAngelo et al., 2006; Grullon et al, 2002
- Maturity is gradual ageing over life cycle.
- But old, mature firms can also face fresh threats in
the product market. We can pick these up.
- Both can matter for payout.
8
Related literature
Payout Policy
 Reviews by Allen and Michaely (2005), DeAngelo,
DeAngelo, and Skinner (2009).
 CEO/CFO surveys by Lintner (1956), and Brav et al.
(2005).
 Investor-driven clientele hypothesis finds weak support.
Brav et al. (2005); Grinstein and Michaely (2005), Jain
(2007).
 Dividends vs. Repurchases: Fama-French-2001,
Grullon-Michaely-2002, or Jagannathan, Stephens, and
Weisbach 2002. Choice boils down to managers’ view
about permanence and stability of cash flows.
9
Related literature
Text-Based Analysis
 Asset pricing applications are in Tetlock-2007, TetlockTsechanksy-Macskassy-2008, and Loughran-McDonald2010. Studies relate word content to stock price
movements.
 The roots of this paper are in Hoberg-Phillips (2010a,b)
- HP introduce product text to the corporate literature
- We build on their work by introducing new metrics of
fluidity and dynamics of product space
10
Data
The guts of our sample is from HP
- 49,000+ 10-Ks from 1997 to 2005 for product text
- 10-Ks from 1996 only used for starting lagged variables.
- 95%+ of eligible COMPUSTAT/CRSP sample
Linguistics: Converting Text to
Quantitative Mappings
1. Union of all words in product descriptions (87,385 in 1997)
minus words in more than 5% of all 10-Ks (3027 in 1997)
a. Optional: throw out words that are not part of “cliques.”
b. Remaining “local” words are “industry vocabularies”
2. Form boolean vectors for all word vectors (1=word used,
0=not used). Normalize to unit length.
3. Compute “cosine similarities” or dot products of these
84,000 element vectors.
13
Cosine Similarity
 Hoberg and Phillips (2010a, 2010b) introduce the notion
of cosine “similarity.”
- Similarity is the dot product between a firm’s word list
and another vector of words.
- HP analyze firm-to-firm similarity for each year and
reconstruct industry pairs.
http://www.rhsmith.umd.edu/industrydata/index.html
14
Similarity in HP
Firm 1: “They sell cabinet products.”
Firm 2: “They operate in the cabinet industry.”
 Step 1) Drop words "they", "the", "and", "in" (common words).
 Step 2) 5 elements: "sell" "operate", "cabinet", "products", "industry"
P1 = (1,0,1,1,0)
P2 = (0,1,1,0,1)
Vi 
Pi
 Step 3) Normalize vector to have unit length of 1:
P .P

V1 = (.577,0,.577,.577,0)
V2 = (0,.577,.577,0,.577)
i
i
 Step 4) Compute document similarity V1 • V2 = .33333
 Document similarity is bounded between (0,1)
15
Similarity for “Fluidity”
We develop HP ideas of similarity along new directions.
1. Product Fluidity Cosine similarity between own word
vector and a vector of word changes.
- Local fluidity – based on close competitors
- Broad fluidity – aggregate word change list. Note that
it excludes 5% common words.
2. Self Product Fluidity 1- Cos(qit,it-1) (similarity of 10K to
last years 10K)
3. Business Clientele Similarity See next slide
16
“Similarity” for Clientele
Cosine similarity of own words to words in input-output
matrix of industries that sell over 90% of their products
to non-retail customers.
PLASTICS RUBBER PULP PAPER PAPERBOARD TRANSPORTATION SUPPORT
AGRICULTURE CONSTRUCTION MINING MACHINERY ACCOUNTING
BOOKKEEPING SERVICES ADMINISTRATIVE SUPPORT SERVICES MOTOR
VEHICLE BODIES TRAILERS PARTS HVAC COMMERCIAL REFRIGERATION
EQUIPMENT CHEMICAL PRODUCTS INDUSTRIAL MACHINERY NONMETALLIC
MINERAL GENERAL PURPOSE MACHINERY AGRICULTURAL CHEMICALS YARN
FABRICS TEXTILE MILL PAINTS COATINGS ADHESIVES MAGNETIC MEDIA
PRINTED ANIMAL AGRICULTURE FORESTRY SUPPORT SERVICES PIPELINE
TRANSPORTATION TURBINE POWER TRANSMISSION EQUIPMENT AEROSPACE
PARTS FABRICATED METAL WOOD WAREHOUSING STORAGE MANAGEMENT
TECHNICAL CONSULTING SERVICES FORGINGS STAMPINGS EMPLOYMENT
SERVICES PRIMARY FERROUS METAL ELECTRICAL EQUIPMENT BOILERS
TANKS SHIPPING CONTAINERS METALWORKING MACHINERY BASIC
CHEMICALS ADVERTISING RELATED SERVICES SEMICONDUCTORS
ELECTRONIC COMPONENTS COAL NONMETALLIC MINERALS MACHINERY
EQUIPMENT RENTAL LEASING ARCHITECTURAL STRUCTURAL METAL PRIMARY
NONFERROUS METAL FOUNDRY
17
Fluidity
Some Familiar Examples
 Microsoft (started dividends)
 Adobe Systems (stopped dividends)
 Apple Computers (never paid and still does not)
18
Microsoft
Dividends and Fluidity
2.5
2
1.5
Dividend Payer
Fluidity
1
0.5
0
1997 1998
1999 2000
2001 2002
2003
2004
2005
2006
Fluidity
Dividend Payer
2007
2008
Adobe
Dividends and Fluidity
1.8
1.6
1.4
1.2
1
Dividend Payer
Fluidity
0.8
0.6
0.4
0.2
0
1998 1999
2000 2001
2002
2003
2004
2005
2006
Fluidity
Dividend Payer
2007
2008
Apple Computers
Marches To Its Own Beat
2
1.5
1
0.5
Payer
Fluidity
0
1
-0.5
-1
2
3
4
5
6
7
8
9
10
Fluidity
Payer
11
12
Dividends, Fluidity, and Clientele
0.7
0.6
0.5
Bus Q1
0.4
Bus Q2
Bus Q3
0.3
Bus Q4
Bus Q5
0.2
Bus Q5
Bus Q4
0.1
Bus Q3
0
Bus Q2
Fluidity Q1
Fluidity Q2
Bus Q1
Fluidity Q3
Fluidity Q4
Fluidity Q5
Payer/Non-Payer Fluidity
Propensity Matched Firms
10
9
8
7
6
5
Payer
Non-Payer
4
3
2
1
Non-Payer
0
FF PTP Q1
Payer
FF PTP Q2
FF PTP Q3
FF PTP Q4
Payer/Non-Payer Clientele
Propensity Matched Firms
1.2
1
0.8
Payer
Not Payer
0.6
0.4
0.2
Not Payer
0
FF PTP Q1
Payer
FF PTP Q2
FF PTP Q3
FF PTP Q4
Table V: Dividend Payer Likelihood
Regressions include controls for firm risk, firm age, M/B, Asset Growth,
Profitability, Firm Size, R&D, and Negative Earnings Dummy.
25
Conclude: Fluidity (-), Clientele (+), Concentration (+)
Smorgasboard of Controls
26
Table VIII: Economic Magnitude
27
Table VII: Repurchaser Likelihood
Conclude: Product characteristics impact repurchase likelihood.
29
Opposite effect of Business Clientele Similarity.
Dividends (upper panel) and Repurchases
Payer/Non-Payer Fluidity
Propensity Matched Firms
10
8
6
Payer
Non-Payer
4
2
Non-Payer
0
FF PTP Q1
FF PTP Q2
Payer
FF PTP Q3
FF PTP Q4
Repurchaser Fluidity
Propensity Matched Firms
10
Repurchaser
Not Repurchaser
5
Not Repurchaser
0
FF PTR 1
FF PTR 2
Repurchaser
FF PTR 3
FF PTR 4
31
Payer/Non-Payer Clientele
Propensity Matched Firms
1.2
1
0.8
Payer
Not Payer
0.6
0.4
0.2
Not Payer
0
FF PTP Q1
FF PTP Q2
Payer
FF PTP Q3
FF PTP Q4
Repurchaser/Non-repurchaser Clientele
Propensity Matched Firms
1
0.8
0.6
0.4
0.2
0
Not Repurchaser
FF PTR
FF PTR 2
1
FF PTR 3
Repurchaser
Not Repurchaser
Repurchaser
FF PTR 4
32
Tables XI to XIII: Initiations/Omissions
Dividend Initiation Policy
Conclude: Fluidity significantly impacts initiations. Historically hard to
33
explain in the data.
Tables XII Dividend Omissions
34
Conclusions
“Product Characteristics and competition matter”
http://www.rhsmith.umd.edu/industrydata/index.html
Download