Discussion of

Corporate Scandals and Household Stock Market Participation by M. Giannetti and T.Y. Wang

Stefan Nagel University of Michigan, NBER, CEPR

March 2014

Stefan Nagel

Discussion of Corp. Scandals and HH Stock Mkt. Participation

Stock market participation: Conventional framework

State uncertainty of stock investment payoffs: Nature determines outcome Participation decision trades off participation costs with return/risk rewards from participation Perception of higher risk or greater aversion to risk can lead to non-participation

Stefan Nagel

Discussion of Corp. Scandals and HH Stock Mkt. Participation

Stock market participation: Role of trust

Strategic uncertainty: Outcome partly under control of other player(s) Trust = belief that opponent will not cheat Possibility: Strategic uncertainty just adds an additional layer of risk. Aversion to it captured by convenetional risk aversion. However, evidence (Bohnet and Zeckhauser 2004): Betrayal aversion over and above risk aversion. Evaluation of prospective returns from stock market participation not only relative to state-uncertainty risks but also betrayal risk: Non-participation if perceived risk of betrayal is high = lack of trust.

Stefan Nagel

Discussion of Corp. Scandals and HH Stock Mkt. Participation

Identifying the causal effect of trust Prior evidence on cross-sectional determinants of differences in trust, effect on economic outcomes Problem: Endogeneity of trust, correlation with unobservables Instruments, e.g., Religion (Guiso et al. 2009) Ethno-linguistic homogeneity (Knack and Keefer 1997)

but correlation with unobservable omitted variables hard to rule out. Fehr (2009): “The most convincing strategy seems to be to induce optimistic or pessimistic beliefs about other people’s trustworthiness exogenously and observe whether this leads to interesting and lasting changes in behaviors and economic outcomes.”

Stefan Nagel

Discussion of Corp. Scandals and HH Stock Mkt. Participation

Experience of fraud as driver of time-variation in trust

This paper: Experience of fraud as shifter of beliefs about trustworthiness Identification: Variation in local corporate fraud experiences Main concern: Correlation with local economic conditions, but Controls for local economic conditions Arthur Andersen natural experiment Within-state variation in adult-life fraud experiences

Overall, evidence is quite compelling.

Table 1: Summary Statistics Panel A: Household Data This table presents the main household characteristics. The unit of observation is the household-year. All variable definitions are in the Appendix. Stefan of Corp. Scandals HH Stock Mkt. Variable N Nagel MeanDiscussionS.D. 25th p. andMedian 75th p.Participation Equity Participation 66615 0.218 0.413 0 0 0 Equity Participation (IRA) 66615 0.298 0.457 0 0 1 Equity Value 66574 24,203 153,424 0 0 0 Net Equity Purchase 65540 7,719 87,441 0 0 0 Equity-Wealth Ratio 66556 0.043 0.134 0 0 0 Bonds & Fixed Income Securities 17142 0.130 0.336 0 0 0 Age (household head) 66615 45.07 16.20 32 42 55 Married 66615 0.554 0.497 0 1 1 Family Size 66615 2.730 1.497 2 2 4 Family Income (in thousands) 66115 54.1 78.4 19.4 38.1 67.6 Wealth (excl. equity, in thousands) 66594 130.9 929.3 0.3 10.2 60 Years in School 64720 12.734 2.766 12 12 15

Comment on analysis 1: Highly skewed fraud variables

Panel B: Distribution of Fraud Revelation by State State

# of State # of State # of State # of State # of State # of Frauds Frauds Frauds Frauds Frauds Frauds AL 5 FL 68 LA 2 NC 8 OK 5 TX 71 AR 1 GA 22 MA 29 NE 2 OR 3 UT 12 AZ 10 IA 1 MD 8 NH 2 PA 24 VA 14 CA 127 ID 1 MI 12 NJ 29 PR 4 WA 6 CO 16 IL 19 MN 13 NM 2 RI 1 WI 3 CT 18 IN 8 MO 8 NV 13 SC 5 WV 1 Intensity DC 2 KSPanel10C: Cumulative MS 2 Fraud NYRevelation 84 SD 3 WY 1 “Cumulative is the total1 number DE # of Frauds” 2 KY MTof frauds 1 revealed OH in a state 25 in the TNpast four 6 years. We also present the

alternative proxies for fraud revelation intensity in the state over the past four years, defined as described in the Appendix. “FraudYear in Industry” fraud intensity in a 3-digit # of measures Year # of revelation Year # of Year # ofSIC code Year industry # of in the past four Frauds Frauds Frauds years. All fraud intensity Frauds measures have been multiplied by Frauds 100. 1980 1981 1982 1983 1984 1985

4 1986 12 6 1987 23 Variable 11 1988 14 Cumulative # of Frauds 12 1989 12 Fraud 17 in State 1990 16 Fraud 2 11 in State 1991 21

1992 1993 N 1994 1402 1995 1402 1996 1402 1997

24 27 40 26 32 23

1998 1999 Mean 2000 2.430 2001 0.010 2002 0.011 2003

29 29 51 38 59 25

2004 S.D.2005 2006 0.048 2007 0.024 2008 0.025 2009

27 40 42 17 9 11

Fraud in State 3 1402 0.013 0.029 Fraud in State 4 1402 0.019 0.044 Fraudnot in State 5 728 0.008 0.021to keep in Skewness necessarily a problem, but useful Fraud in Industry 7975 0.012 0.046 mind that there arein aState) few highly influential observations (Earnings Mgmt. 1402 0.000 0.058

Panel D: State and Firm Level Control Variables Stefan Nagel Discussion of Corp. Scandals and HH Stock Mkt.

Participation

This table presents descriptive statistics for the main variables. The unit of observation is the state-year for state level variables and the firm-year for firm level variables. Median 75th p. Variable N Mean S.D. 25th p.

market participation (including IRA). In column 3, we exclude observations from the 1984 and any equity held in the IRA. The sample period is 1984-2009. All variables are defined in th ordinary least squares. All regressions include a constant term, which we do not report. Stand corrected for heteroskedasticity. ***, **, * denote significance at 1%, 5%, and 10% levels, estimated coefficients on “Fraud in State”, standardized to make the coefficients comparabl sample mean from the variable and by dividing by 100 times the sample standard deviation.

Comment on analysis 2: Magnitude of 2SLS point estimate OLS: ∆Fraud of 2pp → 0.72pp decline in prob(participation)

Panel A: Probability of Participation Dependent Variable: Equity Participation

(3) (4) Excl. 1984 & With IRA 1989 Fraud in State -0.289** -0.363** -0.386** -0.327*** Panel C: IV Estimates (0.146) (0.165) (0.173) We present 2SLS estimates for household equity participation. All variables are defined in the Appendix.(0.119) The [-0.851] sample period is 1994-2005. All regressions include a constant term, which we do not report. Standard errors are Fraud in and Statecorrected 2 clustered at the household level for heteroskedasticity. ***, **, * denote significance at 1%, 5%, and 10% levels, respectively. 2SLS: ∆Fraud of 2pp → 18pp decline in prob(participation)? Fraud in State 3 Fraud in State 4 Instrumental Variable AA Shock Fraud in State 5 Endogenous Variable Fraud in State Log(Age) Control Variables Log(Age) the dependent Married

Is variable the equity participation? Married Log(Family Size)

Stefan Nagel

Log(Family Size) Log(Family Income) Log(Family Income) Log(Wealth) Log(Wealth)

(1)

First Stage Fraud in State

Second Stage Equity Participation

-8.927*** (0.917) 0.091*** 0.345*** 0.289*** (0.006) (0.041) 0.047** log0.001 equity value0.017** rather than(0.056) 0.059*** (0.001) (0.023) 0.006 (0.005) (0.007) 0.000 0.003 (0.008) -0.048*** 0.004 (0.010) 0.000 (0.000) Discussion of Corp. Scandals and HH Stock Mkt. Participation (0.004) (0.005) -0.000 0.011 (0.006) 0.032*** 0.009*** 0.007*** (0.000) (0.008) (0.002) (0.001) 0.007***(0.001) -0.000 0.021*** 0.010*** (0.002)0.009*** (0.000) (0.000) (0.000) 0.010***(0.001) -0.000

Comment on analysis 3: Exclusion restriction State Stock Return Household F.E. Year F.E. Observations

(0.001) 0.985*** (0.199) -0.004 (0.010) x x 37,579

40

(1) Exclusion restriction: AA instrument uncorrelated with factors affecting stock market participation post-2001 (2) Table 3, panel B: AA instrument is uncorrelated with factors affecting stock market participation pre-2001 (2) implies (1) only for time-constant factors (1) is not directly testable. Label for Table 3, Panel B as “test of exclusion restriction” somewhat misleading. But nevertheless plausible that it holds.

Stefan Nagel

-0. (0 [-0

0.029*** (0.002)

(0.000) 0.088*** (0.011) -0.002** (0.001) x x 37,579

State GDP Growth

(2)

Discussion of Corp. Scandals and HH Stock Mkt. Participation

0.389*** (0.038) 0.016** (0.006) -0.003 (0.005) 0.007*** (0.001) 0.009*** (0.000)

0.3 (0 0. (0 0 (0 0.0 (0 0.0 (0

Comment on analysis 4: Weighting of experiences of fraud

Finding: Average state-level fraud intensity since max(birthyear + 18, 1980) correlated with stock market partcipation Paper: Unlike experience effects in Malmendier and Nagel (2011), fraud experiences in the distant past weighted equally strongly as recent experiences. But: approach in the paper also forces some (complex) overweighting of recent experience.

Stefan Nagel

Discussion of Corp. Scandals and HH Stock Mkt. Participation

Comment on analysis 4: Weighting of experiences of fraud First, starting at age 18 implies step-function weighting of life-time experience with zero weight on pre-18 experience. Second, fraud data not available before 1980, so for all but the youngest individuals this induces further overweighting of recent data 60-year old in 1994: only data from age 46 to 60 is used 60-year old in 2005: only data from age 35 to 60 is used

Not clear how different this implicit weighting scheme is from the MN weights that are slowly decaying from current period back to birth. Not clear whether there is sufficiently long history of data to precisely estimate whether there is a slow decay, as in MN (2011), or not.

Stefan Nagel

Discussion of Corp. Scandals and HH Stock Mkt. Participation

Potential additional analyses Disentangle effect of fraud experiences on risk aversion from effects on betrayal aversion? Perhaps: PSID 1996 supplement on (job-choice related) risk tolerance Is betrayal aversion effect specific to stocks, or does it affect propensity to invest in mutual funds? in index funds? Possible to analyze with brokerage data? How big are the effects on participation in aggregate? Based on the cross-sectional estimates, how much aggregate change in the participation rate would the point estimates imply? Applications to betrayal risk in delegated portfolio management, e.g., Hedge funds: Madoff Mutual funds: Market timing scandal

Stefan Nagel

Discussion of Corp. Scandals and HH Stock Mkt. Participation

Potential broader implications

Fraud revelations concentrated in recessions: Part of the reason why empirically observed risk premia rise in recessions? Local price impact documented in this paper is relatively easy to ”arbitrage”, aggregate effect is not

Recent literature on “safety” premium of Treasuries, etc. Not clear exactly what safety attribute is about. Is it lack of betrayal risk? Welfare Desirability of privatized, self-directed retirement saving? Trust in companies, fund managers, government? Assessment of damage of fraud: Not only direct costs, but also wide-ranging indirect effects on individuals’ portfolio choices through betrayal aversion channel

Stefan Nagel

Discussion of Corp. Scandals and HH Stock Mkt. Participation