Title: Follow the Leader: Strategic Interaction in Corporate Capital Structure Mark T' Leary Cornell Univer
1Follow the Leader Strategic Interaction in
Corporate Capital StructureMark T.
LearyCornell University, The Johnson School
Michael R. Roberts University of Pennsylvania,
The Wharton School September 8, 2009
2Motivation
- Virtually all theory and empirics assume firms
set financial policy independently of each other - Theory suggests that firms financing decisions
may not be made in isolation - Product market feedback (Brander Lewis, 1986)
- Learning (Conlisk, 1980)
- Signaling (Ross, 1977)
- Empirically industry median leverage is the most
important observable determinant of capital
structure - Survey evidence also suggests competitor behavior
is important factor (Graham Harvey, 2001)
3Our Study
- Problem impact of peer firm financial policy
does not have a unique interpretation because of
reflection problem (Manski, 1993) - Three potential explanations
- Proxy Effect Firms in same industry have similar
characteristics (e.g., production technologies,
investment opportunities) - Contextual Effect Firms are responding to
characteristics of other firms in industry (e.g.,
liquidation values a la Shliefer Vishny, 1992) - Peer Effect Firms are responding to the
financing decisions of their peers - Goal Disentangle these explanations identify
underlying mechanism
4Key Findings and Messages
- Peer effects exist and are economically large
- 1 SD change in peer firm leverage ? 10 change in
own-firm leverage - 40 larger partial effect than next biggest
determinant - Peer effect in debt-equity choice drives leverage
result - Young, less successful firms mimic financial
policies of more mature, successful firms - But not the reverse!
- Product market competition not responsible
- Mixed support for signaling-herding story
- Firms do not operate in a vacuum with respect to
financial policy - Theory and empirics should recognize this
5Data
- Intersection of CRSP-Compustat 1965-2006
- Exclude
- Financials SIC in 6000,6999,
- Utilities SIC in 4900, 4999, and
- Government entities SIC in 9000,9999
- Firms undergoing significant acquisitions (aftnt1
AB) - Observations with missing data
- Industries with fewer than 10 firms
- Define industry by 3-digit SIC code
6Summary Stats
7Leverage and Industry Leverage
- ?stat signif at 1 , ? stat signif at 5,
Coefficients scaled by SD(X), Book leverage
results similar
8Empirical Model Identification Problem
- Population Regression Function
- Mean Regression of y on X and µg
- Solve for equilibrium, E(y µg)
9Identification Problem Intuition
- This is a classic simultaneity problem
- Firm is financial policy affects firm js and
vice versa - is endogenous because it is
simultaneously determined with yigt - We need an instrument z
- z affects firms js leverage (relevance) but
does not affect firm is leverage (exclusion) - We use the lagged idiosyncratic component of
other firms equity returns
10Condition 1Instrument Relevance
- Financial policy linked to stock prices via
information asymmetry (Myers and Majluf, 1984) - Empirically strong link (Loughran and Ritter
(1995), Baker and Wurgler (2002), Welch (2004)) - Unclear whether idiosyncratic component is linked
but this is testable
11Condition 2The Exclusion Restriction
- Consider reduced form model
- Identification threat must come from variable
that is - Correlated with instrument,
- Correlated with firm is financial policy, and
- Uncorrelated with firm is idiosyncratic return,
systematic return, and other controls - E.g., must argue that other firms idiosyncratic
stock returns are better measures of firm is
investment opps, default risk, etc., than firm
is firm characteristics.
12Instrument Construction
- Firm-specific rolling regressions using 2-5 years
of monthly data - Coefficients are constant within years
- E.g., Idiosyncratic and expected returns for 1/90
12/90 for IBM computed by - Estimation using data form 1/85 to 12/89
- Use estimated coefficients realized factor
returns to estimate E(r) ? - Annualize via monthly compounding
13Return Regression Summary Statistics
14Exclusion Restrictions
- We cant test this (and neither can anyone else!)
- But, we can look at the correlation of our
instrument and firm is characteristics. Should
be low if not zero, otherwise concern - No significant correlations between instrument,
, and any firm characteristics, Xigt-1, - Regression R2 0
- Correlation between instrument and firm i
idiosyncratic return is less than 0.05. - Recall we include firm is idiosyncratic return
in regression
152SLS Leverage Results
Contextual effects estimates omitted
162SLS Policy Results
Contextual effects estimates omitted
172SLS Leverage Policy ResultsIndustry-Size
Groupings
Contextual effects estimates omitted
18What is the Mechanism Behind the Peer Effects?
- Theory
- Learning May be optimal for firms to mimic
others if otherwise costly to optimize (Conlisk
(1980)) - Product Market Feedback Brander and Lewis
(1986), duopoly model in which both firms choose
high debt levels to protect themselves from
aggressive behavior of other. If price predation
(Bolton and Scharfstein, 1990) or
under-investment (Chevalier Scharfstein, 1996)
is severe enough, high levered firms will mimic
less levered counterparts - Signaling Pooling equilibrium to avoid costly of
separation (Ross, 1977, Yilmaz et al, 2009) - Empirical Strategy Heterogeneity in peer effect
19LearningLeader-Follower
- Do some firms follow others?
20Product Market Competition
- Use industry level avgs to sort industries by
year
21Signaling
- Sort firms within an industry-year
22Robustness Tests
23Conclusions
- Firms dont make financing decisions in a vacuum
- Decisions of peers has a profound influence on
capital structure - Initial evidence suggestive of learning or
leader-follower behavior - Theory and empirics need to recognize this
interaction to understand capital structure - Better understand precisely how and why firms
interact