Public Economics Lectures Business Taxes, Education, and Income Transfers

Transcription

1 Public Economics Lectures Business Taxes, Education, and Income Transfers John Karl Scholz University of Wisconsin Madison Fall 2010 JK Scholz Dividends -() Human Capital 1 / 19

2 I Saw a Set of Interesting Papers Last Week The 2009 "Cash for Clunkers" program (Atif Mian, Berkeley, and Amir Sufi, Chicago Booth School). Use cross-community variation in the fraction of "clunkers" driven in a MSA (CBASA). Analyze car sales conditioning on X s and the fraction of clunkers. Saw a big, positive correlation, so that the program increased car sales by 360,000 in July-August, The effects of the program appeared to be completed reversed by March 2010 (7 months latter). There was no discernible effect on employment, house prices, or household default rates, except, perhaps, in communities that made automobiles. JK Scholz Dividends -() Human Capital 2 / 19

3 Interesting Papers Last Week, #2 "Do Expiring Budgets Lead to Wasteful Year-End Spending? Evidence from Federal Procurement" (Jeff Liebman and Neale Mahoney). Spending in the last week of the fiscal year is 4.9 times higher than the rest-of-year weekly average. Quality scores of the IT spending in the last week are 2.2 to 5.6 times more likely to be below the central value. You can do a study like this! See and/or contracts at and/or Amazing data resources for the entrepreneurial student. Think about collaboration! JK Scholz Dividends -() Human Capital 3 / 19

4 Interesting Papers Last Week, #3 "The Price Effects of Cash Versus In-Kind Transfers" (Jesse Cunha, Naval Postgraduate School; Giacomc De Giorgi, Stanford; and Seema Jayachandran, Stanford). Uses interesting data from a field experiment in Mexico. But it shows an important difference between cash and in-kind transfers of food. Cash transfers increase prices (by increasing demand for products). In-kind will result in lower prices, since they increase supply, which shifts some surplus from producers to consumers. In their case study, they find these price effects are large (the price effects of the in-kind program increase net transfers by 12 percent; the price effects for the cash program dissipates 11 percent of the transfer). JK Scholz Dividends -() Human Capital 4 / 19

5 Start with Chetty and Saez, August 2005 QJE Dividend taxes on the highest MTR households fell from 35 percent to 15 percent. Proposed on 1/7/03. to 1/1/2003. Signed into law 5/28/03, but made retroactive This paper is an event study, examining how dividend payout policy is affected by the change in the tax treatment of dividends. The dividend puzzle: why do firms pay dividends when they could repurchase shares (and this give shareholders capital gains treatment on their income)? Bernheim and Wantz (AER signalling explanation) it s still something of a puzzle. JK Scholz Dividends -() Human Capital 2 / 16

6 Theories of Dividend Taxation "Old" view Dividend taxes reduce the net return on investment and hence the supply of saving (assuming SEs outweigh IEs). Dividend tax cuts increase saving, investment, profits, and dividend payouts. New (but really not new), tax capitalization (or trapped equity) view. Marginal investments are entirely financed by retained earnings (rather than by new share issues, which may have "lemon s issues"). Dividend taxes are then capitalized into the value of the firm (equity is trapped). Hence, reductions in dividend taxes will not affect corporate decisions. Rather, they simply lower tax burdens on individuals who receive them. JK Scholz Dividends -() Human Capital 3 / 16

7 Data CRSP: Center for Research on Security Prices Dividend, stock price, and volume information on all companies on the NY, Amex, and NASDAQ stock exchanges. The looks at 80:Q1 to 04:Q2, dropping utilities and financial companies, both of which have odd dividend payout policies. Huge fluctuations in the number of firms in the data, so they also develop a "constant number of firms" sample. Merge in balance sheet information from Compustat, Execucomp, and data on institutional ownership. JK Scholz Dividends -() Human Capital 4 / 16

8 JK Scholz Dividends -() Human Capital 5 / 16

9 Figure 1 (dollars) Can t really tell much. There s a post-reform bump, but a time series regression that includes asset and earnings doesn t reveal a significant change. Why the heck is this a QJE paper??? Does entry and exit (sample churning) distort time series patterns? JK Scholz Dividends -() Human Capital 6 / 16

10 JK Scholz Dividends -() Human Capital 7 / 16

11 Figure 2 (initiations) A substantial, statistically significant number of firms started paying dividends following the 2003 change. The result is robust (and identical) in a specification that conditions on many factors thought to influence dividend payouts. Figure 3 (by month) shows the same thing. Initiations went from roughly $13 million per quarter (prior to the tax change) to $205 million per quarter after the reform. The 6 largest quarterly initiation amounts since 1990 took place after the tax change. JK Scholz Dividends -() Human Capital 8 / 16

12 JK Scholz Dividends -() Human Capital 9 / 16

13 Figure 5 There was a sharp uptick in the number of firms that increased their dividends by 20 percent or more. This is evidence on intensive margin changes One can t help but conclude that something happened. cut the causal force behind the dividend changes? Was the tax JK Scholz Dividends -() Human Capital 10 / 16

14 Did Tax Cuts Cause the Dividend Changes? There was a truckload of corporate scandals in , largely involving accounting fraud. Perhaps this lead to shareholders being worried about management, and therefore forcing firms to get money quickly out of the firm (via dividends). It turns out that only dividends income distributed to individuals through non-tax-favored accounts was affected by the reform. Dividends distributed to pensions, for example, didn t change. Use Thomson financial data on institutional ownership. Look at companies controlled by insurance companies, pensions, non-profits, etc. In Table 3, diff-in-diff estimates show that the significant increase in initiations only apply to the companies not controlled by "unaffected entities." JK Scholz Dividends -() Human Capital 11 / 16

15 Which firms responded? An executive who holds a large stake in the company experiences a large change in personal tax payment from a dividend payout. An executive with no shares doesn t experience any change. Figure 7 breaks firms into quintiles of share ownership: sure enough, dividend changes are largest for companies with the largest share ownership. Executives with large stock options are hurt by dividend payouts, since they will lower the value of the firm and hence make it less likely that the options will be valuable. JK Scholz Dividends -() Human Capital 12 / 16

16 JK Scholz Dividends -() Human Capital 13 / 16

17 Agents and Principals A literature argues that there is an association between the presence of a large individual or institutional shareholder and the degree to which firm behave in a value-maximizing fashion. The literature looks at a) institutional ownership and b) the existence of a large, unaffi liated director. Figure 8 shows institutional ownership matters. But either a large institutional owner or an independent director is suffi cient. "Our results show that principal-agent issues play a first-order role in determining behavioral responses to taxation and should be explicitly included in models of optimal dividend taxation." There is no obvious substitution of dividends for share repurchases, though the counterfactual (would share repurchases have exploded) cannot be ruled out. JK Scholz Dividends -() Human Capital 14 / 16

18 JK Scholz Dividends -() Human Capital 15 / 16

19 Conclusions Total dividends by nonfinancial, nonutility companies increased by 20 percent within 6 quarters of the tax change. No conclusive proof for either the old view or the new view, though the results are probably closer to the old view, in that taxes do seem to matter (though the response is very fast). They call for a theory of dividend behavior that incorporates principal-agent relationships. Perhaps one could get leverage from comparisons of C- and S-Corps, since S-Corps are unaffected. Moreover, one might be able to look at other margins, like investment. How do the dividend tax changes affect investment? JK Scholz Dividends -() Human Capital 16 / 16

20 A New Test of Borrowing Constraints for Education Meta Brown John Karl Scholz Ananth Seshadri New York Fed (Brown) and the University of Wisconsin - Madison December 11, 2009

21 What is this paper about? Going back at least to Becker (1967), economists have worried that borrowing constraints may impede efficient human capital investment. But the large literature on the topic does not focus on the family s expected family contribution (EFC) the difference between the cost of attending college and what federal formulas determine is the family s adjusted available income for college. The EFC, however, is neither legally guaranteed nor universally offered. Children whose parents refuse or are unable to make their EFC may face financial constraints in attending college.

22 Anecdotal Evidence From the Becker-Posner blog Currently if you are under 25 and not in graduate school you are considered dependent on your parents income and have to include their income on your FAFSA which will count against you when figuring your expected family contribution. For those of us who did not receive any financial support from parents other than cosigning loans this is a real kick in the ass. Not only is my family lower middle-class and unable to contribute to my education, but the government will tell me that they expected them to contribute and will punish me by lowering my available loans total. Posted by Diana on January 11, 2005 at 11:10am in response to Government s Role in Student Loans-BECKER

23 Our New Approach Key assumptions It is difficult to study the EFC directly. One problem, for example, is that it is difficult to determine what a parent would have contributed to a child that does not go to college. For the EFC to matter, there must be some scope for disagreement between parents are children. Parents with college-age children typically have extensive access to credit. With unitary household preferences, contracts could be made between parents and children to repay loans made by parents. There is considerable evidence against key implications of the unitary model of intergenerational relationships (for example, Altonji, Hayashi, and Kotlikoff, 1992). If children and parents can write legally binding contracts, where the parent pays for the child s college and the child repays the loan with interest, our paper is less interesting.

24 Our New Approach The world we re living in: children and parents may not always agree If parents and children are independent decision-makers, we know college-age children are likely to be constrained since they generally cannot borrow against future human capital. A relevant question in this context is whether parents choose to relieve the constraint? HRS parents report that 33% of their children who went to college did so without their financial backing. 25 percent (16 percent) of children whose parents had $200,000 to $400,000 (over $400,000) of net worth in 2000 received no parental financial support for college.

25 Our New Approach Financial aid policy assumes parents make their EFC U.S. financial aid policy (like the prior academic literature on borrowing constraints for education) assumes that parents who are able to pay for college are willing to pay for college: Students access to federal loans and grants is contingent on parents income and assets. But parents are under no legal obligation to make their Expected Family Contributions (EFCs). The standard for students to be independent is strict: Namely, the child needs to be an orphan, veteran, parent, 24 or older, a grad student, or married. Do borrowing constraints affect the educational attainment of students whose parents do not meet their EFCs?

26 Our New Approach A simple analytic model points to a new way to examine this issue We model interactions between parents and children as a dynamic non-cooperative game, following the intuition of Bruce and Waldman (1991). Parents care about their own consumption and their children. Children, however, care only about their own consumption ( one-sided altruistic preferences).

27 Our New Approach A simple analytic model points to a new way to examine this issue, part 2 There are two regions to the equilibrium of our model. In one, parents make post-schooling cash transfers. Children achieve the efficient level of education. Financial aid will have no effect on educational attainment. In the other, parents do not make post-schooling cash transfers. Children do not (or are less likely) to achieve the efficient level of education. Financial aid will, therefore, affect educational attainment. The model tells us how to split the data and where to look for evidence for credit constraints. As I will describe, there is considerable empirical evidence that we think is consistent with the presence of borrowing constraints for children in families where parents are unwilling or unable to make post-college transfers.

28 Model Objectives, constraints and timing Preferences U k (c1 k, c2 k = u(c1 k ) + βu(c2 k ( ) U p { ) c i t = u(c }i=p,k; p t=1,2 1 ) + βu(cp 2 ) + αuk (c1 k, c2 k ) Physical capital: a i invested in 1 returns Ra i in 2 for i = p, k Human capital: e p & e k returns h(e p + e k + τ) to the child in pd 2 Choices, constraints & timing: In 1 Parent moves, choosing c p 1 + g 1 + e p + a p x p ; g 1, e p 0. Child chooses c1 k + e k + a k g 1 ; a k, e k 0. In 2 Parent chooses c p 2 + g 2 Ra p ; Child consumes c2 k = Ra k + h(e) + g 2.

29 Period 2 There is a discontinuity in optimal second period transfers Parent { max u(ra p g 2 ) + αu(ra k + h(e) + g 2 ) } g 2 0 g 2 (Ra P, Ra k + h(e)) = g 2 s.t. u (Ra p g 2 ) = αu (Ra k + h(e) + g 2 ) where u (Ra p ) < αu (Ra k + h(e)) 0 otherwise (1) Second period transfers are compensatory, since they decrease with the child s income and assets.

30 Period 1 Children may overconsume and credit constraints for college may arise rationally Child s Euler equation u (c k 1 ) = β max { R, h (e) } ( ) g 2 1+ (Ra k u (c k 2 + h(e)) ) (2) We show e k = a k = 0. There are two cases to consider. When g 2 > 0 u (c p 1 ) = αu (c k 1 ); u (c p 1 ) = βru (c p 2 ); (3) If g 2 0 binds, however h (e) = R; u (c p 2 ) = αu (c k 2 ) u (c p 1 ) = αu (c k 1 ); u (c p 1 ) = βru (c p 2 ); u (c k 1 ) = βh (e)u (c k 2 ); h (e) > R; u (c p 2 ) > αu (c k 2 )

31 Description of equilibrium Two regions based on whether second period transfers are made Partition the x p α h space into 2 regions Region 1 x p and/or α large and/or h ( ) to R quickly g 2 > 0, h (e) = R Strategic concerns; efficient HC investment Region 2 x p and/or α small and/or h ( ) to R slowly g 2 = 0, h (e) > R No strategic concerns; but underinvestment in HC

32 When does financial aid matter? Our central Proposition Proposition 2: In any equilibrium in which g 2 > 0, (ep +e k ) τ = 1. Financial aid does not influence total educational attainment. In any equilibrium in which g 2 0 binds, (e p +e k ) τ > 1. Financial aid increases total educational attainment.

33 Empirical strategy Our model calls for data on parent-child pairs, g 2, and financial aid The model implies the response of educational attainment to financial aid should differ for g 2 > 0 and g 2 = 0 families. No data set we know contains full financial aid information, realized educational attainment, and post-schooling transfers. However, we can go to the HRS for educational attainment and post-schooling transfers, if we can find an informative financial aid proxy. A consistent feature of U.S. federal aid formulas generates substantial aid variation with family structure: EFC(I,A) calculated at the parent level At the student level, Aid = COA - EFC(I,A) N Therefore a student s federal aid can vary substantially with the number of siblings he or she has in college. We use birth spacing as a proxy for financial aid.

34 Is Birth Spacing a Reasonable Proxy for Financial Aid? Evidence from the NLSY-97 suggests it is Appendix Table 2: OLS Estimates of Financial Aid, NLSY-97 Parameter Independent variable (Std error) Sibling-years of overlap in first term of college ** (179.52) Parent's 1997 income, 1000s *** (6.38) Parent's 1997 income squared, 10000s 7.14*** (2.33) Parent's 1997 net worth, 1000s -2.19** (0.98) Parent's 1997 net worth squared, 10000s 0.06* (0.03) AFQT percentile * (16.69) AFQT percentile squared 0.63*** (0.15) Constant 3,152.17*** (463.69) Observations 2608 R-squared 0.06 * significant at 10%; ** significant at 5%; *** significant at 1%

35 The Data Used to Examine Our Central Proposition The Health and Retirement Study U.S. national panel study initiated in Cohorts: HRS (born between ); in 1998 additional cohorts were added, including the AHEAD (born before 1923); CODA ( ); and War babies ( ). For much of the analysis we start with 13,091 families in the 1998 HRS. We restrict the sample by requiring: Those with complete information on the child s DOB, education, gender, gift information, and relationship to the HRS respondent. Children who are 24 or older in 2000 with at least one sibling The main sample is reduced to 9,471 families with 34,593 children. The module sample (see next slide) goes from 427 families to 334, with 1,262 children.

36 HRS Measures of Post-College Transfers The HRS poses two specific, gift questions that are useful to this study We rely on transfers reported by parents over the period in response to the Waves 4, 5, 6 & 7 questions: Including help with education but not shared housing or shared food (or any deed to a house), in the last 2 years did [the Respondent or Spouse] give financial help totaling $500 or more to any of their children or grandchildren? An even more ideal question was asked of a smaller group of respondents in the 1994 Wave 2 Transfers Module: Other than contributions toward education expenses, have you ever given substantial gifts to your grown children?

37 Central Empirical Model Many observable and unobservable factors influence schooling differences between two arbitrarily chosen individuals, including parental attitudes and investments in their children, and heritable components of aptitude. We account for time invariant family-specific factors by estimating the following model. e is = ω i + X is β + γo is + ε is, where i = 1,..., N families s i = 1,..., S i children of family i e is education of child s of family i ω i family i fixed effect X is exogenous characteristics of child s of family i o is years of overlap in college ages of child s with family i sibs

38 Central Results Birth spacing affects schooling, but only for the no-gift samples Table 2: Family Fixed Effect Estimates of Years of Schooling, HRS, Gift v. No Gift Gifts to Children Transfer Module Gifts to Children Gifts No Gifts Gifts No Gifts Parameter Parameter Parameter Parameter Independent variable (Std error) (Std error) (Std error) (Std error) Child gender, male= *** ** (0.087) (0.088) (0.195) (0.134) Child age *** *** (0.013) (0.012) (0.029) (0.018) Oldest child indicator ** (0.117) (0.121) (0.258) (0.186) Youngest child indicator (0.124) (0.126) (0.265) (0.194) Sibling-years of overlap *** ** in college ages (0.031) (0.030) (0.064) (0.046) Number of Children 16,892 17, Number of Families R-squared Adjusted R-squared * indicates significance at the 10 percent, ** at the 5 percent, and *** at the 1 percent level.

39 Table 2: Family Fixed Effect Estimates of Years of Schooling, HRS, Gift v. No Gift Gifts to Children Transfer Module Gifts to Children Gifts No Gifts Gifts No Gifts Parameter Parameter Parameter Parameter Independent variable (Std error) (Std error) (Std error) (Std error) Child gender, male= *** ** (0.087) (0.088) (0.195) (0.134) Child age *** *** (0.013) (0.012) (0.029) (0.018) Oldest child indicator ** (0.117) (0.121) (0.258) (0.186) Youngest child indicator (0.124) (0.126) (0.265) (0.194) Sibling-years of overlap *** ** in college ages (0.031) (0.030) (0.064) (0.046) Number of Children 16,892 17, Number of Families R-squared Adjusted R-squared * indicates significance at the 10 percent, ** at the 5 percent, and *** at the 1 percent level.

40 The Results and Next Steps A student with a sibling 5 years (or more) younger or older, will get roughly one semester less schooling than an otherwise identical twin. Themes and variation Overlap won t matter for low-income families (who get full financial aid) or high income families. Use net worth terciles to distinguish groups. Make use of historical changes in financial aid. The margin should be college: it is. Identification in the fixed effects models come from families with three or more children. This, in a sense, throws away a lot of information. Are results robust to random effects? Split the sample by altruism.

41 Cross-Child Financial Aid Differences are Likely Small in Low- and High-Income Families The EFC for high-income families is likely the full cost of college. Therefore, birth spacing will generate no financial aid differences for children in these families, since financial aid will be $0 regardless of spacing. The EFC for low-income parents is $0. Therefore, birth spacing will also generate no financial aid differences for children in these families, since financial aid will be complete regardless of spacing. We don t know parental income or net worth at the time the child attended college. Instead, we separate the overlap coefficients for each parental net worth tercile. We expect overlap to matter most (or only matter) in the middle tercile of the no-gift sample.

42 Sensitivity Analyses The effects hold in the middle net worth tercile of the no-gift sample Table 3: Family Fixed Effect Estimates of Years of Schooling, HRS Gifts to Children Transfer Module Gifts to Children Gifts No Gifts Gifts No Gifts Parameter Parameter Parameter Parameter Independent variable (Std error) (Std error) (Std error) (Std error) Child gender, male= ** * (0.0878) (0.089) (0.0204) (0.153) Child age ** ** (0.013) (0.012) (0.029) (0.023) Oldest child indicator * (0.117) (0.121) (0.273) (0.216) Youngest child indicator (0.125) (0.127) (0.286) (0.224) Sibling-years of overlap in college ages*tercile 1 (0.045) (0.046) (0.103) (0.063) Sibling-years of overlap ** * in college ages*tercile 2 (0.051) (0.047) (0.083) (0.082) Sibling-years of overlap in college ages*tercile 3 (0.055) (0.054) (0.128) (0.114) Number of Children 16,824 17, Number of Families R-squared Adjusted R-squared * significant at 5%; ** significant at 1%

43 Table 3: Family Fixed Effect Estimates of Years of Schooling, HRS Gifts to Children Transfer Module Gifts to Children Gifts No Gifts Gifts No Gifts Parameter Parameter Parameter Parameter Independent variable (Std error) (Std error) (Std error) (Std error) Child gender, male= ** * (0.0878) (0.089) (0.0204) (0.153) Child age ** ** (0.013) (0.012) (0.029) (0.023) Oldest child indicator * (0.117) (0.121) (0.273) (0.216) Youngest child indicator (0.125) (0.127) (0.286) (0.224) Sibling-years of overlap in college ages*tercile 1 (0.045) (0.046) (0.103) (0.063) Sibling-years of overlap ** * in college ages*tercile 2 (0.051) (0.047) (0.083) (0.082) Sibling-years of overlap in college ages*tercile 3 (0.055) (0.054) (0.128) (0.114) Number of Children 16,824 17, Number of Families R-squared Adjusted R-squared * significant at 5%; ** significant at 1%

44 Results Should Be Larger When There is More Available Financial Aid The Middle Income Student Assistance Act of 1978 substantially increased financial aid to middle and higher income families. We split the sample that includes only students who reached age 18 by 1978; and those who reached 18 after We expect overlap in the post-misaa sample to larger than in the pre-misaa sample for the middle tercile of the no-gift sample.

45 Sensitivity Analyses Financial aid got much more generous in results are larger then Table 4: Family Fixed Effect Estimates of Years of Schooling, HRS, Around Reform College Entry Pre-reform Post-reform Gifts to Children Gifts No Gifts Gifts No Gifts Parameter Parameter Parameter Parameter Independent variable (Std Error) (Std Error) (Std Error) (Std Error) Child gender, male= ** (0.113) (0.121) (0.142) (0.157) Child age ** (0.0191) (0.0182) (0.0296) (0.0327) Oldest child indicator * (0.145) (0.159) (0.201) (0.242) Youngest child indicator (0.172) (0.187) (0.196) (0.215) Sibling-years of overlap ** 0.239** in college ages*tercile 1 (0.0688) (0.0684) (0.0656) (0.0829) Sibling-years of overlap * ** in college ages*tercile 2 (0.0716) (0.0670) (0.0828) (0.0874) Sibling-years of overlap in college ages*tercile 3 (0.0745) (0.0758) (0.0929) (0.0974) Number of Children , Number of Families R-squared Adjusted R-squared * significant at 5%; ** significant at 1%

46 Table 4: Family Fixed Effect Estimates of Years of Schooling, HRS, Around Reform College Entry Pre-reform Post-reform Gifts to Children Gifts No Gifts Gifts No Gifts Parameter Parameter Parameter Parameter Independent variable (Std Error) (Std Error) (Std Error) (Std Error) Child gender, male= ** (0.113) (0.121) (0.142) (0.157) Child age ** (0.0191) (0.0182) (0.0296) (0.0327) Oldest child indicator * (0.145) (0.159) (0.201) (0.242) Youngest child indicator (0.172) (0.187) (0.196) (0.215) Sibling-years of overlap ** 0.239** in college ages*tercile 1 (0.0688) (0.0684) (0.0656) (0.0829) Sibling-years of overlap * ** in college ages*tercile 2 (0.0716) (0.0670) (0.0828) (0.0874) Sibling-years of overlap in college ages*tercile 3 (0.0745) (0.0758) (0.0929) (0.0974) Number of Children , Number of Families R-squared Adjusted R-squared * significant at 5%; ** significant at 1%

47 Greater Education Should Be College and Not High School Overlap is insignificant in the g 2 > 0 and g 2 = 0 sample if both are restricted to those children with high school or less education. Not surprisingly, the overlap coefficients get substantially larger if high school dropouts are excluded.

48 Fixed or Random Effects? The fixed effect estimates are identified off spacing differences between children in families with three of more children. Children in 2-child families will have identical years of overlap, and hence will be captured by the fixed effect. Random effects allow us estimate the overlap coefficient using families with two or more children.

49 Sensitivity Analyses Results are robust using cross-sectional variation: the RE estimates Table 5: Family Random Effect Estimates of Years of Schooling, HRS Gifts to Children: Gifts No gifts Independent variable Parameter (SE) Parameter (SE) Sibling-years of overlap ** in college ages (0.03) (0.03) Number of children ** (0.12) (0.11) Number of children squared (0.01) (0.01) Child gender, male=1-0.25*** (0.08) (0.08) Child age 0.02*** (0.01) (0.01) Oldest child indicator (0.10) (0.11) Youngest child indicator (0.11) (0.12) Parent's 2000 income in 100,000s (0.15) (0.42) Income squared in billions (0.00) (0.01) Parent's 2000 net worth in millions 0.37*** 0.39 (0.13) (0.26) Net worth squared in 100 billions -2.24** (0.92) (0.76) Black 0.44** 0.63*** (0.20) (0.23) Hispanic *** (0.30) (0.27) Parent's education less than HS -0.75*** -0.75*** (0.18) (0.20) Parent some college 0.70*** 0.91*** (0.17) (0.25) Parent college graduate 1.27*** 0.98** (0.22) (0.39) Parent post graduate education 1.72*** 1.79*** (0.23) (0.44) Mean family effect 13.02*** 14.74*** (0.43) (0.50) Total number of children Number of families * indicates significance at the 10 percent, ** at the 5 percent, and *** at the 1 percent level.

50 Table 5: Family Random Effect Estimates of Years of Schooling, HRS Gifts to Children: Gifts No gifts Independent variable Parameter (SE) Parameter (SE) Sibling-years of overlap ** in college ages (0.03) (0.03) Number of children ** (0.12) (0.11) Number of children squared (0.01) (0.01) Child gender, male=1-0.25*** (0.08) (0.08) Child age 0.02*** (0.01) (0.01) Oldest child indicator (0.10) (0.11) Youngest child indicator (0.11) (0.12) Parent's 2000 income in 100,000s (0.15) (0.42) Income squared in billions (0.00) (0.01) Parent's 2000 net worth in millions 0.37*** 0.39 (0.13) (0.26) Net worth squared in 100 billions -2.24** (0.92) (0.76) Black 0.44** 0.63*** (0.20) (0.23) Hispanic *** (0.30) (0.27) Parent's education less than HS -0.75*** -0.75*** (0.18) (0.20) Parent some college 0.70*** 0.91*** (0.17) (0.25) Parent college graduate 1.27*** 0.98** (0.22) (0.39) Parent post graduate education 1.72*** 1.79*** (0.23) (0.44) Mean family effect 13.02*** 14.74*** (0.43) (0.50) Total number of children Number of families * indicates significance at the 10 percent, ** at the 5 percent, and *** at the 1 percent level.

51 Further Evidence: split the sample on a direct altruism question Would parents give 5 percent of their income to a child with a third, half, or three quarters of the parent s income? 62 percent of the subsample (914 parents with 3,292 children have useable data) answered yes if their child had three-quarters of the parents income. These parents were most altruistic. We expect financial aid (and hence overlap) should not matter (or matter less) for the most altruistic families. We interact the overlap coefficient with three dummies that separate the sample by self-reported altruism.

52 Sensitivity Analyses Altruism: financial aid matters less for more altruistic families Table 6: Family Fixed Effect Estimates of Years of Schooling, HRS 2000 Economic Altruism Module Altruism Module Parameter Independent variable (Std error) Child gender, male= (0.203) Child age (0.0313) Oldest child indicator 0.581* (0.276) Youngest child indicator (0.286) Sibling-years of overlap in college ages*give 3/4 (0.0894) Sibling-years of overlap 0.644** in college ages*give 1/2 (0.109) Sibling-years of overlap 0.398* in college ages*never give (0.157) Number of Children 3292 Number of Families 914 R-squared Adjusted R-squared * significant at 5%; ** significant at 1%

53 Table 6: Family Fixed Effect Estimates of Years of Schooling, HRS 2000 Economic Altruism Module Altruism Module Parameter Independent variable (Std error) Child gender, male= (0.203) Child age (0.0313) Oldest child indicator 0.581* (0.276) Youngest child indicator (0.286) Sibling-years of overlap in college ages*give 3/4 (0.0894) Sibling-years of overlap 0.644** in college ages*give 1/2 (0.109) Sibling-years of overlap 0.398* in college ages*never give (0.157) Number of Children 3292 Number of Families 914 R-squared Adjusted R-squared * significant at 5%; ** significant at 1%

54 Table 6: Family Fixed Effect Estimates of Years of Schooling, Bequest Measures from HRS 1994 Economic Altruism Module & HRS 2000 Core Altruism Module HRS 2000 Core Parameter Parameter Independent variable (Std error) (Std error) Child gender, male= ** (0.247) (0.0755) Child age * (0.0367) (0.0111) Oldest child indicator ** (0.335) (0.102) Youngest child indicator (0.354) (0.107) Sibling-years of overlap in college ages*bequest (0.165) very important Sibling-years of overlap 0.186** -- in college ages*bequest (0.0868) somewhat or not at all important Sibling-years of overlap in college ages*100,000 (0.0347) bequest Pr >= 50% Sibling-years of overlap ** in college ages*100,000 (0.0336) bequest Pr < 50% Number of Children ,326 Number of Families R-squared Adjusted R-squared * significant at 5%; ** significant at 1%

55 Table 7: Probit Estimates (Marginal Effects) of College Completion, NLSY-97 Respondents Whose Parent(s) Don't Pay for College, EFC>0 Independent variable Parameter (SE) Parent's 1997 income, 1000s * (0.001) Parent's 1997 income squared, 10000s 0.001** (0.000) AFQT percentile 0.002*** (0.000) Mother's education <HS (0.038) Mother HS grad (0.023) Number of siblings (0.007) Female 0.069*** (0.023) Black (0.029) Hispanic (0.036) Broken home (0.022) Urban (0.029) South (0.023) 12 years old in 1997 wave *** (0.019) 13 years old in 1997 wave *** (0.021) 14 years old in 1997 wave (0.024) 15 years old in 1997 wave (0.027) Observations 511 Pseudo R-squared * significant at 10%; ** significant at 5%; *** significant at 1%

56 Table 8: OLS Estimates of Highest Grade Completed, NLSY-97 Parameter Independent variable (Std error) Parent's 1997 income, 1000s 0.010*** (0.002) Parent's 1997 income squared, 10000s *** (0.001) Parent's 1997 net worth, 1000s 0.000* (0.000) AFQT percentile 0.018*** (0.001) Mother's education <HS (0.129) Mother HS grad (0.080) Number of siblings ** (0.021) Female 0.283*** (0.066) Black 0.432*** (0.107) Hispanic (0.108) Broken home *** (0.078) Urban (0.076) South (0.071) 12 years old in 1997 wave *** (0.123) 13 years old in 1997 wave *** (0.126) 14 years old in 1997 wave * (0.131) 15 years old in 1997 wave (0.145) Funding gap in first term of college, 100s *** (0.005) Constant *** (0.206) Observations 1318 R-squared * significant at 10%; ** significant at 5%; *** significant at 1%

57 Conclusions As we mention at the outset, many prior papers discuss borrowing constraints for higher education. In our framework, all else equal, educational attainment will vary inversely with parental resources for g 2 = 0 families. EFC rises with parental resources. Hence, as income increases, children must finance a greater share of education costs if parents are unwilling or unable to contribute. The relationship between education and parental income is difficult to interpret when studies mix together families willing to meet and unwilling to meet their EFCs.

58 Conclusions, continued We develop a theory of human capital investment that recognizes the distinct roles of parents and children. The model reveals an identifiable category of students who are likely to be meaningfully borrowing constrained. Estimates using HRS and NLSY97 data are consistent with the existence of such constraints. Hence, we think the literature suggesting there are no important financial barriers to college is mistaken. Aid policy is incomplete to the extent that it leaves children of noncontributors under-funded. Policy prescriptions based on these results are unclear: Though increasing financial aid is would move the HC of noncontributors children toward the efficient level, many children of contributors would receive inframarginal subsidies.

59 THE EARNED INCOME TAX CREDIT AND LABOR MARKET PARTICIPATION OF FAMILIES ON WELFARE V. Joseph Hotz, UCLA & NBER Charles H. Mullin, Bates & White John Karl Scholz, Wisconsin & NBER

60 What is the Federal EITC? It is a refundable tax credit directed primarily at low-income working families. There is a small credit available to childless taxpayers. There are three ranges to the credit: the phase-in (or subsidy); flat; and phase-out (or clawback) ranges. EITC is a large program: $31.5 billion in FY2000 (larger than the combined federal spending on food stamps and TANF).

61

62 Coincident Trends: Are They Related? Expansion of the Earned Income Tax Credit Between 1990 and 1999 Real EITC spending increased from $9.6 billion to $31.9 billion (in 1999 dollars)

63 Spending on Cash and Near-Cash Means Tested Transfers, 1999 Dollars $35,000 $30,000 $25,000 Millions $20,000 $15,000 $10, $5,000 $0 Year AFDC/TANF EITC SSI Food Stamps

64 Coincident Trends (cont.) Employment Rates of Single Women with Children Between 3/1990 and 3/2000 Employment rates of single women with children rose from 55.2% to 73.9%. Rates for female-headed households on AFDC/TANF in previous year also went up during 1990s (see graph)

65 70 Percentage of Low-Income Female Heads Employed by Year, California Welfare Sample

66 Coincident Trends (cont.) Standard labor-leisure model: Expansions of (wage) subsidy like EITC should generate increases in employment of low-wage workers. Question: Did EITC play in substantial role in increases in employment of single women with children?

67 Previous Work on Relationship between EITC & Employment Fairly Large Number of Papers on this Issue Dickert, Houser, Scholz (1995, TPE); Keane & Moffitt (1998, IER). Eissa & Liebman (1996, QJE); Ellwood (2000, NTJ); Meyer & Rosenbaum (2000 NTJ; 2001 QJE); Grogger (2003, ReStat) Hoynes and Eissa (2001, wp) examine EITC employment effects for 2-parent families. All find Positive, Large EITC effects on employment. Employment elasticities with respect to net income of 0.69 to (See Hotz & Scholz, 2003 for full survey of these results.) All but first two papers use Diff-in-Diff approach. Use episodic expansions in EITC and compare changes between groups who were eligible and not eligible for EITC (e.g., single mothers vs. single women).

68 Potential Concerns about Inferences drawn from Previous EITC Employment Studies A. Use of national episodic expansions of EITC to explain national trends vulnerable to possibility that t other things changed. Secular Changes in Welfare programs (AFDC/TANF, Food Stamps, Child Care subsidies) Aggregate labor market conditions could be driving changes in employment rates.

69 Spending on Cash and Near-Cash Means Tested Transfers, 1999 Dollars $35,000 $30, $25,000 s Million $20, $15,000 $10,000 $5,000 $0 Year AFDC/TANF EITC SSI Food Stamps

70 Annual Real Earnings per Worker in Service Sector in California, $60 $ s of $ $40 $30 $ Bay Area Counties Central & Southern Farm Counties Northern and Mountain Counties Central Valley Counties Los Angeles County Southern Calif. Counties, Other than LA All Counties

71 Annual Employment to Population Ratios in California, Bay Area Counties Central lvalley Counties Central & Southern Farm Counties Northern and Mountain Counties Los Angeles County Southern Calif. Counties, Other than LA All Counties

72 Potential Concerns about Inferences (cont.) B. Use of Second Diff in Diff-in-Diff strategy requires composition of comparison group doesn t change over time. Previous studies use Repeated Cross-Sectional data typically from CPS in Diff-in-Diff analyses. Population we analyze single mothers on welfare in California during 1990s Sizeable changes in racial composition, family structure and other characteristics. ti

73 Fraction of Cases that are White Fraction of Cases that are Black Fraction of Cases that are Hispanic Fraction of Cases that are Asian Number of Kids in Hshld. Fraction Fraction of of Hsehlds Children with 2+ Less than Kids Age 6 Year % Chge., % 17.2% 69.1% -18.1% -2.9% -6.6% 11.5%

74 Potential Concerns about Inferences (cont.) C. If EITC expansion truly caused increases in employment rates of single mothers, should see similar systematic changes in rates of EITC take-up, i.e., claiming EITC on tax returns Analogous to studies of effects of welfare & other social programs on employment Look for changes in program participation to corroborate program effects on employment. Systematic examination of relationship btwn. EITC expansions & differences in EITC take-up rates btwn. treatment & comparison groups has not been done.

75 Contributions of this Paper 1. We use data from a single state (California) to mitigate influence of secular changes in social policies & local labor market conditions. Over period we examine, low-income populations subject to limited set of policy changes. Better able to control for changes in state t policy, some of which vary at county level. Also control for detailed set of county-level measures of labor market conditions to capture local conditions more accurately.

76 Contributions of Paper (cont.) 2. We exploit longitudinal data on households and focus on temporal within household changes to control for potential composition bias problem in Diff-in-Diff estimation strategy. Use longitudinal data on households in estimation.

77 Contributions of Paper (cont.) 3. Use different Diff-in-Diff identification strategy than in most previous work. Compare differential behavior of families with 2+ children vs. 1-child families before & after EITC expansion in 1990s. EITC expansion in 1994 substantially increased generosity of EITC for 2+ children vs. 1-child households.

78 Table 1: Earned Income Tax Credit Parameters, (in nominal dollars) Year Phase- In Rate (%) Phase-In Range Max Credit Diff. in Max Credit: Child Phase-Out Rate (%) Phase-Out Range , ,920 15, , ,840 18, , ,240 19, , ,730 20, a 1992 a ,140 1,192 1, ,520 1,324 1, ,250 21,250 11,250 21,250 11,840 22,370 11,840 22, a , , , , , ,200 23, , , , , , , , , , , , , , , , , , , , , ,610 2,038 2, , ,110 1, ,152 3, , , ,656 1, ,271 3, ,485 2,312 3,816 1, ,353 3, , ,000 23,755 11,000 25,296 5,000 9,000 11,290 24,396 11,290 26,673 5,130 9,230 11,610 25,078 11,610 28,495 5,280 9,500 11,930 25,750 11,930 29,290 5,430 9,770 12,260 26,473 12,260 30,095 5,570 10,030 12,460 26,928 12,460 30,580 5,670 10,200 12,690 27,413 12,690 31,152 5,770 10,380

79 Contributions of Paper (cont.) 3. We use a different Diff-in-Diff identification strategy t (cont.) t) We systematically assess the validity of implication of this identification strategy EITC policy treated all households with 2 or more children the same (i.e., same credit). So, we should expect to see no difference in outcomes of interest (e.g., employment) between 2+ and 3+ child households.

80 Contributions of Paper (cont.) 4. Focus on effects of EITC on employment for important population. Female-Headed households on welfare sometime during 1990s. Look at behavior of these households both on & off of welfare. Estimating effects of EITC for this population is particularly relevant from public policy perspective.

81 Contributions of Paper (cont.) 5. Most Novel Feature of Paper: Examine differential effects (2+ vs. 1-Kid households) of EITC on incidence of claiming EITC. Exploit access to data on federal tax returns for households in sample over 1990s. If our Diff-in-Diff identification strategy is isolating EITC effects on employment, should see differential rates of EITC claiming by 2+ vs. 1-Kid households before & after expansion.

82 Our Data Combine Several Administrative i ti Sources Monthly AFDC/TANF case records. Demographic information and benefit receipt. Prior information starting in 1987 on benefits come from Medicaid data. Quarterly data from UI system. Measure employment starting in 1986 Federal tax return information. i Data from CA Franchise Tax Board beginning in 1990.

83 Our Data Combine Several Administrative Sources County-level (local) local labor market data County-level policy data (from county welfare & training gprogram implementation) Sample exclusions: Child-only cases Cases with more than 2 adults in household We focus most of analyses on AFDC-FG cases.

84 Sampling Start random stratified sample of all assistance units on Welfare in California between 1987 & Drawn by Rand for another evaluation. Sample includes ~ 50% of all cases. Define a sampling date 4 th quarter of a household s s spells on welfare. We determine number & ages of children when household on welfare. All cases, on and off welfare, are treated symmetrically. Avoid overweighting long-term welfare recipients in our sample.

85 Sampling (cont.) Utilize two samples of households in our analyses Cross sectional: Employment in year following sampling date. This sample mimics repeated cross-section data used in previous studies. Longitudinal: Employment in periods -3, -2, -1, and 0 (sampling). Longitudinal data on households allow control for householdspecific fixed effects, so focus on within household changes to identify EITC effects.

86 Table 2a: Employment Rates (in Percentages) by Family Size, , Cross-Sectional Sample Year All Cases Cases with One Child Cases with 2+ Children Difference (2+ - One) [27,568] [37,999] Diff-in-Diff (199x Average) [24,433] [32,354] [23,098] [29,532] [24,005] [30,402] (0.66) [25,500] [31,164] (0.66) [25,457] [30,256] (0.67) * [24,794] [28,650] (0.68) *** [22,479] [26,174] (0.69) *** [19,066] [21,973] (0.74) *** [19,731] [22,490] (0.74)

87 Table 2b: EITC Claiming (in Percentages) by Family Size, , Cross-Sectional Sample Year All Cases Cases with One Child Cases with 2+ Children Difference (2+ - One) [27,568] [37,999] Diff-in-Diff (199x Average) [24,433] [32,354] [23,098] [29,532] [24,005] [30,402] (0.59) [25,500] [31,164] (0.61) *** [25,457] [30,256] (0.63) * [24,794] [28,650] (0.66) *** [22,479] [26,174] (0.69) *** [19,066] [21,973] (0.75) *** [19,731] [22,490] (0.75)

88 Econometric Specifications (Diff-in-Diff) Y = φ + αyear_ s + β[2 + Kids Year_ s ] + δkids _ j + ψkidsage _ k ict i S t s ict t j ict k itc s= 1992 s= 1994 j= 1 k= 1 + κ X + λ W + η L + θ County_ m + ε 55 ic ct ct m c ict m= 1 where Y ict th 1, if at least 1 adult in i household in county c is employed in year t Emp ict = (1) 0, otherwise th 1, if i household in county cfiles tax return and claims EITC in year t ClaimEITC ict = 0, otherwise 2+Kids ict is 2+ kids indicator variable; Kids_j ict is j children in household indicator variable; KidsAge_k ict = # of children in the household age k; X ic time-invariant demographics of household; W ct county-level measures of California s welfare caseload and policies; L ct time-varying measures of county-level labor market conditions; County_m ict indicator variable for residing in county m

89 Econometric Specifications (Diff-in-Diff) (Cont.) In some regressions for testing we also include an (exactly) 2 children indicator variable: 0, if Kids it < 2 or Kids it 3 + it + it = 1, if Kidsit = 2 [ 2+ Kids 3+ Kids ]

90 Covariates Demographic characteristics Number of kids & number of kids by age. Local labor markets Year dummies, employment share by sector, avg. income by sector Wlf Welfare rules Proportion of population in GAIN program Time-invariant i i covariates (in OLS Cross-Sectional Models). Race/ethnicity county dummies gender age timing of entry Race/ethnicity, county dummies, gender, age, timing of entry onto welfare.