Essays in economic and financial decisions of households

(1)

Tilburg University

Essays in economic and financial decisions of households Jerphanion, Emiel DOI: 10.26116/center-lis-2006 Publication date: 2020 Document Version

Publisher's PDF, also known as Version of record

Link to publication in Tilburg University Research Portal

Citation for published version (APA):

Jerphanion, E. (2020). Essays in economic and financial decisions of households. CentER, Center for Economic Research. https://doi.org/10.26116/center-lis-2006

General rights

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain

• You may freely distribute the URL identifying the publication in the public portal

Take down policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

(2)

Essays in Economic and

Financial decisions of Households

(3)

Essays in Economic and Financial Decisions of Households

Proefschrift ter verkrijging van de graad van doctor aan Tilburg University, op gezag van de rector magnificus, prof. dr. K. Sijtsma, in het openbaar te verdedigen ten overstaan van een door het college voor promoties aangewezen

commissie aan Tilburg University op donderdag 2 juli 2020 om 16.00 uur

(4)

promotores: prof. dr. F. Braggion

prof. dr. F.C.J.M. de Jong

copromotor: dr. R.G.P. Frehen

(5)

Acknowledgments

This dissertation is the tangible output of some very exiting, challenging, rewarding and en-joyable years of my life. In the writing process I have received tremendous support from many different people, and I would like to take this opportunity to express my deepest gratitude to all those people.

First, I want to thank my supervisors, Fabio Braggion and Rik Frehen for their incredible mentoring and guidance. I deeply appreciate their honesty, patience and sincere interest in my well-being, both professionally and personally. In our joint work I benefited directly from their contagious ambition, dedication, integrity and it showed me the importance of laughing together. Fabio was instrumental in motivating me to consider a career in academia, always stimulated me to pursue my own research interests, aim for big economic questions and pushed me to think about the larger picture. Rik taught me to be a ‘grote speler’, to be explicit, and always forced me to think deep and hard about the contribution of my research.

I am very grateful to Frank de Jong for his engagement and all the work he did in his role as my promotor. I would also like to thank Rob Alessie, Lieven Baele, Joost Driessen and Jasmin Gider for accepting to take part in my dissertation committee. I have greatly benefited from their feedback, deep questions and inspiring ideas.

(6)

sharpening the writing in my job market package and sharing his job market wisdom.

I would also like to thank fellow Ph.D. candidates who have been extremely valuable col-leagues and have become my friends. I specifically want to thank Pranav, Martijn, Kristy, Peter, Bart and Haikun. Throughout the years we have shared dinners, laughs, struggles and I appreciate all the conversations we had about new research ideas and numerous non-research topics.

I was also very fortunate to have many friends outside academia that kept both of my feets on the ground, provided me with incentives to continue and were important in my personal development. I am very grateful for our friendship Jan-Willem, Marjella, Ruud, Jolanda, Jurgen, Jonatan, Ronald and all members of rood12 (you made it possible to climb this ‘berk’). I want to deeply thank my family for their unconditional love, encouragement, support and helping me to place all academic things in perspective. I want to thank my parents, Rob and Lucy, who have always supported me during my studies, encouraged me to follow my dreams and emphasized that they love me irrespective of any accomplishments. I want to thank my (little) brothers, Wilrik and Jesse, for all interest they showed in my work and I want to say to them that I am very proud of them. I want to express my heartfelt gratitude to my wife, Harmienke, who supported me throughout, married me even though I sometimes talk too much about work, and was my safe haven when I felt frustrated or insecure.

Finally, I want to praise God. He is my faithful, loving father and He guides me in the right paths to show that He is good.

(7)

Introduction

“Consumption -to repeat the obvious- is the sole end and object of all economic activity” — John Maynard Keynes

This dissertation is a collection of three independent chapters that aim to improve our un-derstanding of economic behavior of households. Many decisions households make are complex and potentially have long-lasting effects on the consumption of the family. For instance, the decision of parents to save for the future education expenses of their children may directly affect the family’s current consumption (postpone buying a new car), future consumption of the parents (potentially sacrifice retirement savings), and the future consumption of the child as college education increases his/her life-time earnings. I will focus on two important areas where households make investment decisions, namely in financial markets and obtaining ed-ucation. Like all investments, education and stocks create costs in the present but hold the potential to increase future consumption as households can receive higher capital gains or la-bor income. A prominent feature in most of my research is understanding how credit provision affects household’s financial choices since this potentially explains the observed heterogene-ity in households’ saving and portfolio decisions, and more broadly, is critically important in assessing the financial welfare of people.

(8)

atten-dance. Since student aid is typically not sufficient to cover all expenses, the parents increase their savings after a credit expansion because they face a higher probability of financing college cost of their children in the future. Consistent with this interpretation, I also find that college attendance disproportionately increases for families affected by the reform. Furthermore, my results show that affected parents shift the allocation of wealth towards riskier assets. A key contribution of my paper is demonstrating the causal effect of student loan supply on parental saving behavior, since most of the literature has focused on the effects of educational debt on students’ future consumption and investment decisions.

(9)

sion of matrimonial property rights for women generates a shift in intra-household bargaining power. This increase in decision making power provides incentives to invest in female education since it reduces the risk of expropriation of joint property by the husband and allows women to alter the allocation of household resources. I use the introduction of a 1988 South African law to identify a positive shock in expected decision-making power of black married women. I exploit the presence of tribes that are split by a border, to use foreign black women from the same tribe as the control group. The central result is that women incorporate these expected marital gains in the human capital investments since women extent their schooling by 8-12 months. This increase in female schooling is driven by women that extend their schooling vis-`a-vis women that already left the educational system at the time of the legal expansion of matrimonial property right.

(10)

Chapter 1 Student Loan Supply, Parental Saving

& Portfolio Allocation

∗

∗_{I am very grateful to my advisors Fabio Braggion and Rik Frehen for invaluable comments and support.}

I also like to thank Pat Akey, Julio Crego, Adeline Delavande, Jens Kvaerner, Wilbert van der Klaauw and Nicola Pavanini as well as all seminar participants at Alliance Manchester Business School, Bristol University, CSEF - University of Naples Federico II, KU Leuven, Nova SBE, Tilburg University, VU Amsterdam and Young Finance Scholar Conference. I am responsible for all remaining errors or omissions. Please sent correspondence to e.jerphanion@tilburguniversity.edu.

(14)

CHAPTER 1. STUDENT LOAN SUPPLY, PARENTAL SAVING & PORTFOLIO ALLOCATION

Abstract

(15)

1.1 Introduction

The swift rise in student aid in recent years made educational debt the largest non-mortgage liability for U.S. households (Brown et al., 2014). From 2000-2001 to 2016-2017 the total (av-erage per student) aid grew from $108.5B to $240.1B ($9,762 to $16,343).1 _{This exponential}

growth has attracted the interest of economists and policymakers, as high levels of educational debt may adversely affect students’ future consumption, investment and personal default deci-sions (Dearden et al., 2008; Rothstein and Rouse, 2011; Chakrabarti et al., 2020; Krishnan and Wang, 2019; Goodman et al., 2019; Mueller and Yannelis, 2019; Mezza et al., 2020). While a growing literature studies the relationship between student aid and graduate outcomes, much less is known about potential effects on families’ intertemporal choices. This paper is the first to demonstrate the causal effect of student aid supply on parental saving behavior. The parental saving response to the rise in student loans potentially has important implications for the al-location of assets within households, and more broadly, as total household saving corresponds to 48.9% of U.S. national saving, to the distribution of wealth in the economy.2

Parental saving is intimately linked to the provision of educational financing as 70% of parents accumulate financial wealth, using both saving instruments and financial markets, to finance college expenses (Fidelity, 2018). The parental saving decision is characterized by a trade-off between consumption smoothing and expected college attendance of their offspring. Economic theory provides two contrasting mechanisms through which student aid levels di-rectly affect parental saving decisions. On the one hand, the supply of student aid reduces parental savings since family wealth is a substitute for student loans in alleviating credit con-straints of students. On the other hand, the provision of student aid lowers the entry barrier of marginal college entrants. Since the effective costs of obtaining the college premium are reduced, marginal students, who would have previously not attended university, enroll for col-lege. As college attainment becomes a positive NPV investment for marginal students, parents increase their savings to cover the remaining unmet financing needs.3 _{Similarly, attendance}

1_{These figures are from the annual reports ‘Trends in Student Aid’ published by CollegeBoard, based on}

administrative data from the U.S. department of education and income tax returns. All figures are expressed in 2017 Dollars.

2_{This percentage is calculated using data of the U.S. Bureau of Economic Analysis on gross private and}

public saving for the second quarter in 2019

3_{Long and Riley (2007) show that these remaining unmet needs are substantial. They estimate an average}

(16)

of a superior quality college becomes a positive NPV investment and induces parents to save more. Ultimately, it is an empirical question which of these effects dominates.

(17)

enrollment disproportionately increases in families that are more affected by the reform. I esti-mate that college enrollment increases by 12 percentage points. This result provides evidence that the expansion of student aid programs succeeded in its primary goal to promote access to post-secondary education for students that would otherwise be unable to attend college.

The results on positive parental saving responses are robust to using an alternative identifi-cation strategy that exploits the notion that expected student aid amount sharply increases if siblings are likely to attend college simultaneously. A placebo test validates that the saving re-sponse to student aid supply is absent in families without children. Cross sectional tests reveal that the effects are largest among lower- and middle-income families and for college educated parents. Furthermore, I find more substantial saving responses for parents that identify them-selves as savers. Documenting this new relationship between student aid supply and families’ portfolio decisions improves our understanding of the intergenerational effects of student aid supply, and more broadly, contributes to explaining the observed heterogeneity in households’ saving and portfolio decisions.

(18)

While I do not directly test these models, the empirical results in this paper are consistent with the theoretical prediction that parents incorporate expected college expenses of their children in their savings decision. Finally, the mechanism I identify in this paper is directly relevant to the current policy debate regarding the optimal design of federal loan programs to stimulate college enrollment while minimizing the consequences for consumption smoothing (Lochner and Monge-Naranjo, 2011; Hanushek et al., 2014; Lochner et al., 2018; Abbott et al., 2019). I document a new spillover effect of student aid supply on parental wealth accumulation and asset allocation. This finding is relevant for a nascent literature that studies the relationship between household saving, asset allocation and wealth inequality (De Nardi and Fella, 2017; Bach et al., 2018; Fagereng et al., 2020).

The remainder of this paper is organized as follows. Sections 1.2 and 1.3 describe the policy environment and data sources respectively. Section 1.4 discusses the identification strategy. Section 1.5 present the empirical results and section 1.6 concludes.

1.2 Institutional Background

(19)

To access federal aid, prospective college students must fill out an application form (FAFSA) that collects detailed information on household income, assets and family composition. The FAFSA form asks market values of all assets an family owns and for the family income. These inputs are used by the U.S. Department of Education to estimate the dollar amount a family can pay out of pocket to cover college expenses, called the expected family contribution (EFC). The intuition is that high income families and households with large asset holdings are able to cover more expenses. Important to note is that the EFC formula also includes demographic factors like family size, age of parents and other family members’ enrollment in post-secondary education. For instance, the EFC sharply decreases as the number of college-going family members increases (Brown et al., 2011). The eligibility for subsidized loans is determined by the difference between EFC and the cost of attendance, with an annual limit that caps the amount that students are allowed to borrow. The student loans are distributed through the student aid offices at the college a student applied for. The college receives the amount of student aid awarded to the student and first subtracts their tuition fee, and whatever amount is left is transfered to the student.

(20)

This reform constituted a significant increase in student balances of federal student loans as average annual amount of student loans received rapidly grew from$4,421 (1992-1993) to $5,365 (1995-1996) (Wei et al., 2004). Obviously, families on the margin of receiving aid experienced the largest increase in student aid receipt. Dynarski (2003b) estimates that before HEA, each dollar of home equity reduced the federal aid eligibility by three to six cents for these marginal families. In a simple back of the envelope calculation for an average family with $50,000 home equity this corresponds to an increase in annual student aid of $1,500-$3,000.

[Figure 1.1 about here.]

1.3 Data

The primary data come from the Panel Study of Income Dynamics (PSID) for the years 1984 to 1999. The PSID is an annual panel data survey, which contains detailed information on family income, housing, family structure and other demographics. In additional wealth supplements, households are asked questions on their net worth and financial asset holdings for the years 1984, 1989, 1994, and 1999. These wealth supplements allow me to construct measures of household saving and stock market participation. The PSID is particularly well suited for this analysis since it enables me to track individual households over time and exploit within-household variation in saving behavior. This allows me to include within-household fixed effects that capture unobservable risk preferences, beliefs and other time-invariant characteristics.

(21)

the literature.5 In several cases households report a very low house value, therefore I follow the suggestion of Gerardi et al. (2010) and eliminate all observations for which the reported house value is below $5,000. I exclude a household-year observation if the household head is retired in that particular year. To eliminate gross outliers from the sample I follow Juster et al. (2006) and trim the top and bottom percentile of each wealth component, income and home equity.6

After deleting observations with missing values for income, wealth or demographic variables the baseline sample consists of 3,111 observations, in 1,207 unique households.

1.3.1 Household Saving Measures

Households’ saving is calculated following the active saving approach of Juster et al. (2006). This measure captures the change in total household wealth minus capital gains for housing and financial assets, inheritances and gifts received plus the value of debt repayment. The active saving approach is particularly well suited to measure changes in saving behavior, because capital gains (passive saving) are not included. For example, household wealth accumulation may reflect revaluation of assets that are independent from an active saving decision. Since my analysis focuses on changes in actual parental saving, I eliminate these capital gains to obtain a more precise measure of the true saving intention of a household (Dynan et al., 2004). Naturally, higher-income households may have the ability to save more, therefore I normalize total household saving by the total family income. More formally, I define a saving rate for household i at time t: SavingRatei,t = PJ j=1ActiveSaving i,j t−1,t Incomei t−1,t (1.1)

where the sum of accumulated wealth in all assets (j) over the years t − 1 to t is divided by total income of household i over the same period. I consider a wide range financial and real assets7_{, however I exclude home equity as saving vehicle because I use variation in home}

5_{The exclusion of these households addresses the concern that a change in household head affects the}

unobserved heterogeneity in saving behavior. In total there are 155 households in which the head changes, however the main results are not sensitive to their exclusion.

6_{All results remain with winsorizing the top and bottom percentile.}

7_{The PSID contains information on real estate other than home equity, a farm or private business, vehicles,}

(22)

equity to define treatment exposure. The measurement of active saving of household i in asset j (ActiveSaving_t−1,ti,j ) depends on the presence of potential capital gains in that particular asset. The exact method is described below, however the main intuition is to exclude potential capital gains by measuring net flows. Since this wealth data was gathered in 5 year intervals, the household saving rate (SavingRatei,t) is defined for the periods 1984 to 1989, 1989 to 1994

and 1994 to 1999.

For assets where capital gains do not play a major role according to the PSID classification, I define active saving as the difference between the market value in period t and its value in period t − 1. More specifically, I compute equation 1.2 using reported values of households’ saving and checking accounts, bond holdings, vehicle values and consumer debt.

ActiveSaving_t−1,ti,j = V_ti,j− V_t−1i,j (1.2)

AactiveSaving_t−1,ti,j represents the active saving of household i in asset j over time period t − (t − 1) where V_ti,j is simply the reported value of asset j in time t.

(23)

money received from a full or partial sell of the households’ interest in a private business or real estate.

To validate the main results I also use an alternative measure of household saving behavior proposed by Cronqvist and Siegel (2015). They measure savings as the change in a household’s total non-housing wealth, and scale this amount by the disposable income over the same period. Therefore the saving rate of household i is defined as:

SavingRatei,t = ∆N etW orthi t− ∆HomeV alueit Incomei t−1,t (1.4)

where N etW orthi

t is the sum of the wealth value in all asset classes at year t minus total

debt. HomeV aluei_t−1,t measures capital gains in housing value over a 5 year period, excluding households that moved between two consecutive periods.

1.3.2 Descriptive Statistics

(24)

[Table 1.1 about here.] [Figure 1.2 about here.]

Furthermore, Table 1.1 shows that the average household consists of roughly 2 children, a 43 year old head and an average (median) annual family income of $57.48k ($51.36k). The table reports few marital transitions, however this is at the household-year level. At the household level these are substantially larger as 8.8% of the families gets married and 5.6% experiences a divorce during the sample period.

1.4 Empirical Strategy

While the legal expansion of student aid constitutes an economy-wide shock, I propose to isolate its effect on family finances by studying differential post reform changes across households. The removal of home equity in the federal aid formula induces variation between households in expected student aid. I construct a household-specific treatment intensity based on the share of household wealth represented by home equity shortly before the reform. This empirical approach is similar to Lucca et al. (2018), who study the effect of federal student aid expansion on tuition fees using variation in treatment intensities among universities. This differences-in-differences specification eliminates the potential concern that a trend in tuition fees increased students’ reliance on federal loans over time (Lochner and Monge-Naranjo, 2011). Equation 1.5 shows the baseline regression I estimate using the PSID panel data.

SavingRatei,t = αi+ αst+ βI[HEA]t×

HomeEquity1989 i

N etW orth1989 i

+ λCi,t+ i,t (1.5)

(25)

to cover college expenses before the children go to university, this coefficient can be interpreted as the savings response of families to expected student loan supply available to their children. Furthermore, Ci,t is a vector that contains time-varying household level characteristics that

affect savings decisions. These controls include income volatility, household size, age of the head, age squared and three indicator variables that equal one if the head owns a business in the previous year, got married or got divorced. By including household fixed effects (αi) I

effectively study the effect of HEA on within-household saving behavior. Hence, it is unlikely that my results are driven by unobserved heterogeneity in time-invariant characteristics that correlate with saving behavior and demand for debt. For instance, religious households tend to save more, while they borrow less (Guiso et al., 2003). In this specification, I absorb any state-level shocks by including state × year fixed effects (αst). The mean effect on the population over

the sample period is absorbed by the inclusion of αst. All regressions cluster standard errors

at the household level, since observations are unlikely to be independent within households. A crucial assumption for this estimator to be valid is that treatment and control groups follow parallel trends in absence of the reform. Unfortunately PSID only started collecting wealth data in 1984, therefore I am unable to extent the saving rate measure to periods before 1989. In order to examine the similarity between saving trends of the treatment and control groups I rely on a different PSID question that asks respondents whether the household has any savings. Although this is a crude measure of saving behavior, an indicator for household savings is used more often (e.g. Puri and Robinson (2007)). In Figure 1.3, I plot the average share of households with positive savings from 1971 to 2003. I split the sample by the median level (0.62) of the treatment indicator HomeEquity1989i

N etW orth1989

i . The graph shows that the two groups

follow parallel trends prior to the reform. After the introduction of HEA, the propensity to save of affected families exponentially increases relative to the control group.

[Figure 1.3 about here.]

(26)

are substantial wealth differences between both groups. This finding is intuitive as wealthier families typically hold wealth in multiple assets, whereas poorer households primarily rely on housing wealth (Fagereng et al., 2019).

[Table 1.2 about here.]

1.5 Results

1.5.1 Savings Response

This section studies the effect of expected student loans on parental saving. Table 1.3 suggests a strong causal effect of student aid expansion on parents’ savings. The point estimate of the interaction term is consistently positive and statistically significant at the 1% level. The magnitude of the estimates remain stable if I saturate the model with fixed effects. The point estimate drops somewhat when I include household-fixed effects, which suggests that households with more housing wealth also have higher saving rates. Overall, the effect is economically sizable: a one-standard-deviation increase in expected student aid HomeEquity1989i

N etW orth1989 i

(27)

Consistent with the documented decline in household saving rates in the 1990s (Parker, 1999; Skinner et al., 2001), I find a strong negative effect in the post HEA period. Also notable is the sizable positive effect of entrepreneurship on household saving rates. This result confirms previous findings that entrepreneurial risk increases households’ savings (Quadrini, 2000; Gentry and Hubbard, 2004). The coefficients of age are negligible since there is little variation in age is left after including time trends.

1.5.2 College Enrollment

(28)

enrollment as having completed more than 12 grades of schooling.

I follow the empirical specification of Lovenheim (2011) and estimate the following linear probability model:

Enrollmenti,k,t = αt+ αs+ βI[HEA]t×

HomeEquity1989 i

N etW orth1989 i

+ λCi,k,t+ i,k,t (1.6)

where Enrollmenti,k,t is a dummy variable equal to one if child k in household i enrolls in

college in year t. Important to note is that the unit of observation changes from the family to the student since a household can have multiple children. I[HEA]t is a dummy that equals

one if child k enrolls in college after the introduction of HEA, therefore the parents could still adjust their savings in response to the reform, and zero otherwise. HomeEquityi1989

N etW orth1989 i

measures the household-variation in exposure to the reform as the fraction of housing equity in total wealth before the reform. The coefficient of interest is β, which captures the effect of student aid supply on the college enrollment of children. The specification also includes a vector of student and household characteristics (Ci,k,t). I include gender and ethnicity dummies to control for

student characteristics. Furthermore, I control for total family income and the number of children in the household in the year of college enrollment and a dummy that equals one if the household head has a college degree. Finally, I include college year fixed effects (αt) and state

fixed effects (αs). In this specification I cannot include household fixed effects since there are

only few households that contain both children who enroll before and after the reform.

(29)

1.5.3 Alternative Identification Strategy

To provide additional evidence for the positive saving response I use an alternative identification strategy that relies on the notion that student aid amount sharply increases if there are multiple college-going family members (Brown et al., 2011). To be more specific, I define a treatment indicator that equals one if a household contains at least two siblings with a birth spacing less than 4 years and zero if the household contains no ‘overlapping’ children. Since most college degrees require 4 years, the overlapping sibling indicator proxies for higher (per student) expected student aid. My alternative specification is then as follows:

SavingRatei,t = αi+ αst+ βI[HEA]t× I[SiblingOverlap]i+ λCi,t+ i,t (1.7)

The main coefficient of interest remains β, the effect of sibling overlap (which proxies for expected financial aid) on saving behavior after student aid expansion. Panel A in Table 1.5 presents the results. Unless otherwise mentioned, I suppress all control variables for brevity.

The number of observations slightly grows since I can include families where the home equity-wealth ratio is not properly defined (at least one of the inputs is negative). The results show that the main result, a positive saving response, remains unchanged using this alternative identification, albeit somewhat smaller as in Table 1.3.

(30)

1.5.4 Home Equity Adjustments

Since the HEA provides households with an incentive to shift more wealth into housing, as the HEA removes an implicit wealth tax for this asset category, economic theory would suggest that parents would first adjust their home equity levels before adjusting their saving behavior. Therefore, any saving response that I find is in excess of any home equity adjustment that households might do. However, typically households are unable to adjust the amount of wealth invested in home equity fully. Furthermore, since accessing home equity wealth often entails large costs it may not be optimal to shift all wealth into home equity (Hurst and Stafford, 2004). In table 1.6 I directly test whether households adjust their home equity levels actively. The results are mixed. I find no adjustment in the dollar values of mortgages but there is an slight increase in probability of moving to a more expensive house.

1.5.5 Heterogeneity and Robustness

Treatment Heterogeneity

(31)

college costs, increases the expected return of the child’s college graduation for more educated parents.

Cross sectional Robustness Checks

This section tests important cross sectional implications that naturally follow from the hy-pothesized relationship. The effect of student aid supply on parental savings should not affect all households equally. If the provision of student aid induces parents to save more because it raises the expected marginal college returns, the relationship should be absent in families with no children. I test this implication by estimating equation 1.5 on a sample of childless families. The first two columns of Panel B in Table 1.8 report the results. I find that families without children that have a high share of wealth in home equity clearly do not adjust their saving behavior because they have no exposure to the student aid reform. This finding also mitigates the potential concern that a differential exposure to the housing market might drive the results in the previous sections.

The results in the previous sections suggest that parents increase their saving because they face a higher probability of financing a part of their children’s college expenses because of the increased probability of college enrollment due to the student loan expansion. If parents incorporate the expected future college costs in their saving decision, we should observe a more pronounced saving response by families that live in areas with high average college expenses. In the first two columns of Panel A I split the sample between states with above-median levels of average college costs and states with low costs of attending college (below median). Consistent with the previous findings I find that parents that life in states with high college costs increase their saving by more compared to families that life in states with low college costs. The difference in saving responses is statistically significant at the 1% level.

(32)

survey that asks respondents whether they prefer to ‘save for the future’ or ‘spent money to-day’.8 _{The responses are coded as a five-point Likert scale. I follow Knowles and Postlewaite}

(2005) and focus on households that indicated a clear preference (disregard indifferent respon-dents). Since this measure is only available for households that were already in the PSID sample in 1972, the total number of observations shrinks to approximately 500. In the final two columns of Panel A in Table 1.8 I split this sample by saving attitudes. The results show that the saving response is driven by families with a positive saving attitude.

To learn about the timing of saving response of parents, I examine whether the age of children present in the household affects the saving response of parents. The final 2 columns in Panel B shows that the positive saving response is primarily driven by families with secondary school-aged children. This finding is consistent with the notion that parents typically learn about the child’s ability during secondary school.

1.5.6 Wealth Allocation

The results in previous sections consistently show that parental saving increases after a positive shock to student aid supply. While the ultimate objective is to explore the dynamics of wealth accumulation, recently the attention in the literature has shifted more towards the allocation of savings because it introduces heterogeneity in rates of return on household savings (e.g. Campbell, 2016; Bach et al., 2018; Fagereng et al., 2019).

(33)

and Giannetti and Wang (2016)) I measure equity market participation using an indicator variable that equals one if the household holds a any stocks at a given time. This includes both directly held stocks, and indirect equity holdings via investment trusts, mutual funds and retirement accounts. The linear probability model estimates in Column 2 show that equity market participation among affected parents increases. In the final two columns I examine the proportion of the liquid financial portfolio invested in equity of the full sample (column 3) and conditional on equity participation in the previous period (column 4). This ratio is a common measure of financial risk taking in household finance (e.g. Calvet and Sodini (2014)). I find that affected parents tilt their portfolio towards risky assets. These findings suggest that the provision of student aid have an additional impact on household wealth accumulation through household portfolio returns induced by a change in allocation of assets.

A potential explanation for tilting the parental portfolio towards riskier assets is that stu-dent loans provide liquidity to invest when it is optimal. Since the stustu-dent loan expansion enables families to borrow against the child’s human capital, the demand for equity increases as this mitigates the concern of being liquidity constrained when it is optimal to invest in the child’s human capital (Roussanov, 2010). Moreover, Roussanov (2010) predicts that this portfolio adjustment is more pronounced for less wealthy families. This is consistent with the finding that lower- and middle-income families are the main beneficiaries of the student loan expansion.

1.6 Conclusion

(34)

one-CHAPTER 1. STUDENT LOAN SUPPLY, PARENTAL SAVING & PORTFOLIO ALLOCATION

standard deviation increase in exposure to student aid yields a 2.2 percentage point increase in the fraction of income saved by affected families. The mechanism that drives this result is the anticipation by parents of the positive effect of student aid on college enrollment of their children, i.e. the college investment NPV becomes positive for students on the margin of college attendance. Parents increase their savings to cover the remaining unmet financing needs in college expenses after receiving student aid. Consistent with this interpretation, I show that college attendance disproportionately increases for families affected by the reform. Furthermore, I find that student credit expansion shifts parental wealth allocation towards riskier assets.

My findings point to a previously undocumented and non-trivial intergenerational impact of student credit. Most policy discussions have largely ignored the interaction between student loans and parental wealth as the two most important sources of college financing in the United States. My analysis suggests that the overall effect of providing subsidized student loans is likely to be higher since parental savings act as an complement to this subsidy. This implication challenges the common belief that student aid provision leads to substitution effects like lower student earnings during college (e.g. Evans and Nguyen, 2019).

(35)

(36)

FIGURES

Figure 1.1: Total Annual Federal Student Loan and Grant Volume

(37)

Figure 1.2: Distribution of Home Equity Share

This histogram shows the distribution of HomeEquity1989i

N etW orthi 1989

(38)

FIGURES

Figure 1.3: Percentage of Households with Positive Savings (1994=1)

1

1.05

1.1

0.95

Share of Households with Positive Savings

1971 1975 1980 1984 1989 1994 1999 2003

Year

Low Equity Share High Equity Share

This graph plots the share of households that report having any savings, such as savings accounts or government bonds. I split the sample by 0.62 (median) of the treatment indicator HomeEquityi1989

N etW orth1989 i

(39)

Table 1.1: Descriptive Statistics

This table reports descriptive statistics for my main variables of interest. I report mean, median 10th percentile and 90th percentile for all observations at the household-year level. SavingRatei,tis total annual savings divided

by income as defined in section 1.3.1. ‘No Saving (d)’ is a dummy variable that equals one if the household has non-positive savings in a given year. Similarly ‘Equity Participation (d)’ is an indicator variable that equals one if the household holds any stocks in publicly held corporations or mutual funds in a given year, including equity in IRAs. HomeEquity1989i

N etW orthi 1989

is the fraction of home equity wealth of the total wealth in 1989, before the reform. ‘Annual Family Income’ is defined as the 5 year average income and ‘5yr Income Volatility’ is the volatility of annual income over these 5 year periods. Furthermore, I include the number of children in the household (‘Number of Children’) and age of the head of the household (‘Age (years)’). ‘Entrepreneur (d)’ and ‘College Degree (d)’ are dummies that equal one if the head of the household own a business or holds a college degree at a given year respectively. Marital transitions within the household are defined as dummies that equal one if the head of the household got married (‘Married (d)’) or got divorced (‘Divorced (d)’) in a given year. Finally, a dummy equals one if the head of the household is of black ethnicity (‘Black (d)’).

Mean SD p10 p50 p90 Obs. HomeEquityi 1989 N etW orthi 1989 0.596 0.249 0.249 0.620 0.935 3,111 SavingRatei,t 0.049 0.151 -0.073 0.028 0.189 3,111 No Saving (d) 0.346 0.476 0 0 1 3,111 Equity Participation (d) 0.464 0.498 0 0 1 3,111

Annual Family Income ($k) 57.48 30.08 25.32 51.36 96.19 3,111 5yr Income Volatility ($k) 12.48 13.54 3.09 8.36 24.54 3,111

(40)

TABLES

Table 1.2: Balancedness

This table describes the characteristics of parents with above median levels of home equity (HomeEquityi1989

N etW orth1989 i

) and below-median parents. ‘Age (years)’ is the age of the head of the household and ‘College Degree (d)’ is a dummy that equals one if the head of the household holds a college degree. ‘Number of Children’ is defined as the number of under-aged children residing in the household in a given year. Finally, I measure the total wealth of a household by summing the reported values for all assets a household owns (where the top and bottom percentile of each wealth component are trimmed). The final column reports t-tests comparing the means of above- and below-median equity shares.

High Equity Low Equity

(41)

(42)

TABLES

Table 1.4: College Enrollment

This table reports the results of the test whether affected families experience a disproportional increase in college enrollment. The level of observation is the enrollment decision of the child in the household. I estimate equation 1.6 on the college enrollment decision of children given the household-variation in treatment exposure. I[HEA]t is a dummy that equals one if the child enrolls in college after the introduction of HEA and zero

otherwise. HomeEquity1989i

N etW orth1989 i

measures the household’s exposure to the reform as the fraction of housing equity to total wealth before the reform. I control for number of children and family income in the year of college enrollment. In the first two columns I include college year fixed effects (αt) and state fixed effects (αs), while in

the final two columns I include the more stringent state times college year fixed effects (State×CollegeY earF E) and wealth quartile fixed effects (Wealth Quartile FE ). All standard errors are clustered at the household level and reported in the parentheses. Finally, I also report the number of observations (N ).

Enrollmenti,k,t Enrollmenti,k,t Enrollmenti,k,t Enrollmenti,k,t

I[HEA]t×HomeEquity 1989 i N etW orth1989 i 0.103∗ 0.115∗ 0.218∗∗ 0.173∗ (0.062) (0.065) (0.098) (0.092)

Controls Yes Yes Yes Yes

College Year FE No Yes No No

State FE No Yes No No

State × College Year FE No No Yes Yes

Wealth Quartile FE No No No Yes

(43)

Table 1.5: Alternative Specifications

This table reports alternative empirical specifications that validate the results of Table 1.3. Panel A shows the results from an alternative identification strategy that relies on the notion that student aid amount granted sharply increases if there are multiple college-going family members. More specifically, I estimate a difference-in-difference regression specification with a treatment indicator that equals one if a household contains at least two siblings with a birth spacing less than 4 years and zero if the household contains no ‘overlapping’ children (I[SiblingOverlap]i). The treatment indicator is interacted with ‘I[HEA]t’, that takes the value of one after

the introduction of HEA. Panel B reports the results for the difference-in-difference regression specification of equation 1.5 with alternative saving rate measures. In the first two columns the dependent variable is the measure described in section 1.3.1. This measure captures the change in non-housing wealth between two periods (includes capital gains) and is scaled by the total income of the family over the same period. In the final column the dependent variable is a saving rate that includes saving in housing (‘TotalSavingRatei,t’). The

number of observation drops since saving in housing wealth is only defined for households with information on recent moving activity. More specifically, saving in housing is defined as the change in home equity between two survey waves if the household moved and the change in total outstanding mortgage debt in the household did not move (Juster et al., 2006). In both panels I control for number of children, income volatility, age of the households head, age squared, and dummies that equal one if the household hold owns a business, holds a college degree, got married or divorced. Furthermore, I include household fixed effects (HouseholdF E) and state times year fixed effects (State × Y earF E). All standard errors are clustered at the household level and reported in the parentheses. Finally, I also report the number of observations (N ).

Panel A: Alternative Identification

SavingRatei,t SavingRatei,t SavingRatei,t

I[HEA]t× I[SiblingOverlap]i 0.040∗∗∗ 0.030∗ 0.038∗∗

(0.015) (0.016) (0.016)

Controls Yes Yes Yes

Household FE No Yes Yes

State × Year FE No No Yes

N 3,261 3,122 3,119

Panel B: Alternative Saving Measures

∆N etW orthi Incomei t−1,t ∆N etW orthi Incomei t−1,t TotalSavingRatei,t I[HEA]t× HomeEquity1989 i N etW orth1989 i 0.095 ∗∗ _0.167∗∗∗ _0.086∗∗ (0.044) (0.047) (0.040)

Household FE Yes Yes Yes

State × Year FE No Yes Yes

(44)

TABLES

Table 1.6: Home Equity Adjustments

This table reports the results of the effect of HEA on home equity adjustments of households, with standard errors clustered at the household level in the parentheses. Home equity adjustment decisions are measured by examining the effect on the total outstanding dollar value of the mortgage (‘Total Mortgage Valuei,t’), a

dummy variable that equals one if the family moved to another house and zero otherwise (‘Moving (d)’) and measure of the dollar value of the houses for the sample where there is information on moving (‘Value of the House’). Similar to the main specification in equation 1.5, I[HEA]ttakes the value of one after the introduction

of HEA, and HomeEquityi1989

N etW orth1989 i

measures the household’s exposure to the reform as the fraction of housing equity wealth of total wealth before the reform. I control for time-varying household characteristics as the number of children, age of the households head, age squared, and dummies that equal one if the household hold owns a business, holds a college degree, got married or divorced. Furthermore, I control for income shocks by including 5 year income volatility. I control for household fixed effects (HouseholdF E), state times year fixed effects (State × Y earF E) and wealth quartile fixed effects (Wealth Quartile FE). Finally, I also report the number of observations (N ).

Total Mortgage Valuei,t Moving (d) Value of the House

I[HEA]t× HomeEquity1989 i N etW orth1989 i 1056.22 0.124* 5388.76** (5673.57) (0.071) (2717.73)

Household FE Yes Yes Yes

State × Year FE No No Yes

Wealth Quartile FE Yes Yes Yes

(45)

(46)

TABLES

Table 1.8: Cross Sectional Robustness Checks

This table reports the results of testing cross sectional implications that naturally follow from the hypothesized relationship. In the first two rows of Panel A, I split the sample between states with above-median levels of average college costs and states with low costs of attending college (below median). I retrieve the average state level college expenses for four year public universities for the academic year 1993-1994 from Snyder and Hoffman (1995, p. 184). In the final two columns I split the sample by different saving attitudes. Column 3 limits the sample to families where the household head indicated in a 1972 survey that he/she rather ‘save for the future’ than ‘spent money today’. In column 3 I limit the sample to households that prefer to spent. In Panel B, I examine how family composition affects the positive saving response of the parents. In the first two columns I estimate the difference-in-difference regression specification of equation 1.5 on a sample of families without children, that should not be affected by HEA. In the final two columns I split the sample by the age of the children to test whether parents exhibit a differential saving response if the household contains children in secondary school. I[HEA]t

takes the value of one after the introduction of HEA, and HomeEquity1989i

N etW orth1989 i

measures the household’s exposure to the reform as the fraction of housing equity wealth of total wealth before the reform. I control for number of children, income volatility, age of the households head, age squared, and dummies that equal one if the household hold owns a business, got married or divorced. Furthermore, I include household fixed effects (HouseholdF E) and state times year fixed effects (State × Y earF E). All standard errors are clustered at the household level and reported in the parentheses. Finally, I also report the number of observations (N ). The difference between states with high college costs and low costs is statistically different at 1% level.

Panel A: College Expenses & Saving Preferences

SavingRatei,t SavingRatei,t SavingRatei,t SavingRatei,t

High College Low College Saving Spending

Costs States Costs States Preference Preference I[HEA]t× HomeEquity1989 i N etW orth1989 i 0.119∗∗∗ 0.073∗∗ 0.148∗∗∗ -0.055 (0.044) (0.073) (0.058) (0.045)

Household FE Yes Yes Yes Yes

State × Year FE Yes Yes No No

N 1,796 1,173 245 239

Panel B: Family Composition

SavingRatei,t SavingRatei,t SavingRatei,t SavingRatei,t

Families with Children in Children in

no Children Ages of 1-10 Ages of 11-17

I[HEA]t×

HomeEquity1989 i

(47)

Table 1.9: Asset Allocation

This table reports the average change in asset allocation after the introduction of HEA. In the first column I consider the ratio of household debt over non-housing wealth. The second column examines equity participation using a dummy that equals one if the household is active in the stock market. The final two columns examines the average change in proportion of equity in the household’s portfolio of cash, bonds and equity (‘Risky Share’). The empirical specification of the difference-in-difference regression is similar as in Table 1.3. I[HEA]t takes

the value of one after the introduction of HEA, and HomeEquity1989i

N etW orth1989 i

measures the household’s exposure to the reform as the fraction of housing equity wealth of total wealth before the reform. I control for number of children, income volatility, amount of non-housing wealth, and dummies that equal one if the household hold owns a business, holds a college degree, got married or divorced. Furthermore, I include household fixed effects (HouseholdF E) and state times year fixed effects (State × Y earF E). All standard errors are clustered at the household level and reported in the parentheses. Finally, I also report the number of observations (N ).

(1) (2) (3) (4)

Household Leverage Equity Risky Risky

over total Wealth Participation Share Share

I[HEA]t×HomeEquity 1989 i N etW orth1989 i -1.409∗∗∗ 0.259∗∗∗ 0.111∗∗ 0.272∗ (0.452) (0.076) (0.055) (0.141)

Household FE Yes Yes Yes Yes

State × Year FE Yes Yes Yes Yes

(48)

Chapter 2 Does credit affect stock trading?

Evidence from the South Sea Bubble

∗

(49)

Abstract

(50)

CHAPTER 2. DOES CREDIT AFFECT STOCK TRADING?

2.1 Introduction

The credit boom preceding the 2008 financial meltdown has sparked economists’ interest in the relation between credit supply, trading decisions and asset prices. Recent empirical work has shown that easy access to credit is at the root of the 2000s housing price boom (Mian and Sufi, 2009; Favara and Imbs, 2015; Di Maggio and Kermani, 2017). Similarly, Jord`a et al. (2015) and Brunnermeier and Schnabel (2016) have related credit provision to equity market booms and busts. While the positive relation between credit and asset prices has been widely documented, the channel through which credit fuels prices is still unclear.

Economic theory provides a wide array of channels with different implications for stock trading, asset valuation and wealth transfers across investors. On the one end, the discount rate channel, predicts that cheap credit reduces the cost of capital without trading, deviations from fundamental values or wealth transfers among traders.1 _{On the other extreme, the extrapolation}

channel predicts the opposite: naive extrapolators lose money because they use credit to ride (and thereby fuel) the bubble unsuccessfully (Fischer, 1933; Galbraith, 1955, pp. 46-50, Kindleberger, 1978, Barberis et al., 2018). In between the extremes, we find channels with mixed implications, for example, loan holders could successfully ride the bubble as in Abreu and Brunnermeier (2003), or borrow from traders with more pessimistic priors about the security’s payoffs as in Geanakoplos (2003, 2010) or Simsek (2013).

In order to identify the channel, an ideal test would compare trading strategies and real-ized returns of a loan holder vis-`a-vis another trader without a loan. Such a test would also show whether loan holders’ trading behavior is in line with some particular bubble theories. However, an in-depth analysis of leverage, trading strategies and realized returns is empirically challenging for three reasons.

(51)

trading data for a large fraction of the market.

We tackle these issues by studying margin loan provisions in the London equity market during the 1720 South Sea episode, a financial boom and crash that is widely considered as a classical example of a bubble. Similar to other bubble episodes, there was an increase in leverage preceding the asset bubble (e.g. margin loans in the 1929 stock market bubble (Rappoport and White, 1993) and the 2014-2015 Chinese equity bubble (Bian et al., 2019b); and mortgage credit in the housing bubble early 2000s (Favara and Imbs, 2015)). In the early run up of the bubble, the Bank of England opens a facility allowing its shareholders to borrow money by collateralizing their shares in the Bank.2 _{For three major British companies, we}

hand-collect every single equity transaction and margin loan with unique buyer, seller and borrower identities. The three companies represent about 50% of the market in terms of pre-bubble capitalization. We link the trading and loan data to the complete list of subscribers of new share offerings initiated by highly overvalued companies. Since our information also covers five years prior to the bubble, we can control for pre-bubble trading strategies and performance.

Our data has the scope and level of detail to address each of the three empirical challenges outlined above. First, our main data source consists of trader-specific ledger accounts which record trader and borrower identities. Since loans and trades are recorded in the same ledger account, we precisely observe each trader’s daily loan position and share trades. Moreover, as we observe the complete universe of investors’ holdings and trades, we are able to accurately measure wealth transfers between loan holders and other investors. Second, we observe each trader’s behavior across firms and hence we can control for company-specific events and changes in the macroeconomic environment. Third, the large and representative market coverage allows us to make general statements about the relationship between debt, trading strategies and asset prices.

Before we study the trading behavior and realized returns of loan holders, we study who takes a loan. We find that physical proximity to the market, trading experience and trading frequency are major determinants of the propensity to take a loan. In particular, investors who do not hold any shares before 1720 and investors who are in the top percentile of the trading frequency distribution, are more likely to take a loan. We also find that shareholders living 2_{Shortly before, the South Sea Company opens a similar facility. We collect loan information for both}

(52)

close to the market were significantly more likely to collateralize their shares.

When we analyze trading behavior, we find that investors with margin loans behave as extrapolators, that is, they are more likely to buy (sell) following days of high (low) share returns. For instance, in the spring of 1720, loan holders are approximately 65% more likely to buy shares vis-`a-vis other traders. Consistently with these results, margin loan holders are also twice as likely to subscribe to new share offerings when these shares trade at peak prices (six `a eight times pre-bubble quotes). Even without taking returns on these share subscription positions into account, loan holders incur large trading losses. A margin loan holder realizes a 14 to 23 percentage point lower return than the average investor. In additional tests, we show that our findings cannot be explained by asymmetric information among investors (Brennan and Cao (1994, 1997); Brennan et al. (2005)), investors’ portfolio rebalancing (see Bian et al. (2019a)) or destabilizing short selling (Lamont and Stein (2004); Hong et al. (2012)).

Our paper makes three contributions. First, we contribute to the literature on leverage, margin loans and the behavior of asset prices. A large number of papers has studied whether credit provision destabilized financial markets by increasing stock price volatility (Salinger, 1989; Schwert, 1989; Hardouvelis, 1990; Hardouvelis and Peristiani, 1992; Hardouvelis and Theodossiou, 2002; Kahraman and Tookes, 2017; Hansman et al., 2019). Most of this literature is based on time series evidence where it is hard to disentangle reverse causality and it is difficult to understand whether the findings are driven by margin loan supply or other factors such as monetary policy or economic growth forecasts.3 _{Since we observe investors’ daily trades across}

(53)

Second, we contribute to the literature on the determinants of trading strategies during financial bubbles. In particular, we highlight that margin loan holders behave as extrapolators as predicted by various bubble theories. We also show how debt can fuel a bubble as predicted by a number of theoretical models (Allen and Gorton, 1993; Allen and Gale, 2000; Geanakoplos, 2003, 2010; Simsek, 2013; Scheinkman, 2014). Furthermore, we show that the extrapolative strategies employed by margin loan holders lead to poor performance. This finding adds to the literature on trading strategies during bubble periods and subsequent investor performance (Brunnermeier and Nagel, 2004; Temin and Voth, 2004; Dass et al., 2008; Greenwood and Nagel, 2009; Griffin et al., 2011; Xiong and Yu, 2011; Barberis et al., 2018; An et al., 2019).

Third, our paper also contributes to the literature that relates the use of margin loan accounts to trading behavior (Barber et al., 2019; Bian et al., 2019a,b; Heimer and Imas, 2020; Heimer and Simsek, 2019). In contrast with these studies, we do not focus on a particular type of investor. As a result, we can identify wealth transfers between loan holders and other traders.

The remainder of the paper is organized as follows. Section 2.2 discusses the 1720 London securities market and the historical setting of the margin loan facility. In sections 2.3 and 2.4 we describe the data and discuss the empirical results respectively. In section 2.5 we test the robustness of our results. We conclude in section 2.6.

2.2 Historical setting

2.2.1 The South Sea Scheme

The South Sea Company is established in 1711, as a result of the Peace of Utrecht and is granted the trade monopoly between Britain and South America. Rather than being involved in international trade, the directors move the South Sea Company’s business into finance and in particular sovereign lending. During the war campaigns of the early eighteenth century, the British government accumulates a large amount of debt. The government pays relatively high yields because its debt is illiquid.4 _{The South Sea Company proposes to swap government debt}

with South Sea Company shares and bonds. In theory, such a scheme would make everybody 4_{The illiquidity is caused by large denominations and because annuities are assigned to a particular person}

(54)

better off. The government would profit as it would pay a lower yield to the South Sea Company; the public would hold a more liquid and standard financial asset; and the South Sea Company would earn a spread between the yield received by the government and the yield paid to its bondholders. Two swaps of limited size work out successfully in 1711 and 1719, and the company proposes a more ambitious scheme in 1720. The new plan considers the swap of almost the entire British government debt with South Sea Company claims.

The Bank of England also bids for a similar scheme, and it is believed that the competition between the two companies leads the South Sea Company to overpay in order to be granted the swap (Dale, 2004, p. 75). In the final agreement, the company pays the government a fixed amount of £7.6 million to receive the right to exchange government annuities for South Sea Company equity and bonds. The terms of the agreement give strong incentives to the South Sea Company to raise share prices. The debtholder trades her annuity for shares valued at market prices: a higher market price implies that the company can purchase the outstanding government debt with fewer shares. The government further allows the company to raise£31.5 million of nominal capital to finance the debt acquisition. If the South Sea Company needs less than £31.5 million to swap the annuities, it can use the leftover to issue new shares in the market against (high) prices. Government debtholders respond enthusiastically to the swap.

(55)

numerous new equity issues. The new offerings of the South Sea Company more than doubles the amount of its outstanding shares. Moreover, Mackay (1852) lists that at least 86 new stock listed companies were formed in 1720. Similarly, Ofek and Richardson (2003) document that during the unfolding of the Tech bubble the asset float of Internet stocks increased by more than 200 billion dollars.

2.2.2 Bank of England and South Sea margin loan facilities

On May 10th 1720, the Bank of England creates a margin loan facility that undercuts prevailing interest rates.5 Dickson (1967, pp. 192-193) argues that the bank’s initiative is a direct response to the opening of a margin loan facility by the South Sea Company on April 25th. However, the court of director minutes remain vague about the motives for providing cheaper credit: “it may be for the service of this bank to lend money to the proprietors upon this bank stock”. Shareholders can borrow cash by depositing Bank of England shares against 5% interest per year (lowered to 4% on July 14th, 1720). The Bank gives an amount of cash equal to the nominal value of the shares deposited as collateral.6 In our sample period, the Bank of England’s market price is always at least 30% higher than the par value of the shares. As a consequence, the loan to value ratio is significantly smaller than one, which mitigates potential moral hazard concerns.7

In early September 1720, the South Sea Company is in financial difficulties and needs immediate funding to finance its operations. Perhaps giving in to political pressure, the Bank of England decides to assist in resolving the South Sea Company issue by agreeing to buy subscription shares for a total of £3.75 million against a pre-determined price (Neal, 1993, p. 115). Shortly after, the financial problems of the South Sea Company spill over to the Bank of England.8 The South Sea Company’s prime bank (Sword Blade Bank) defaults on its payments 5_{Temin and Voth (2004) report that in April 1720 interest rates on collateralized loans were 10% per month}

and became 1% per day thereafter.

6_{Since the Bank of England always lends an amount equal to the nominal value of the shares deposited}

as collateral, the investors’ borrowing capacity declines when share prices increase. If any, this feature should prevent investors from buying overvalued assets.

7_{We find only two Bank of England loan holders that default on their loans in October 1720. The Bank}

sells their shares and uses the proceeds to redeem the loan and pay transaction fees. After all these payments, there is still a small amount left which is returned to the borrower.

8_{Scott (1912) Volume III - p. 327 reports that there were rumors that Robert Walpole, a prominent member}

(56)

and the South Sea Company share prices dive well below the Bank of England’s purchase price, generating large losses for the Bank. The Bank of England’s directors address these financial difficulties by recalling all outstanding Bank of England loans and offer an interest rate discount to margin loan holders who repay on a short notice (Neal, 1993, p. 113). On October 6th, the Bank officially announces the full annulment of the loan facility: “no loans to be made upon bank stock until further order”.

In April 1720, few weeks before the Bank of England, the South Sea Company opens a loan facility designed to lend money to its shareholders using South Sea Company shares as collateral. The terms were more generous than those established by the Bank. A South Sea shareholder could borrow £2.5 for each £1 of nominal South Sea capital. The initial interest rate was set at 5%, but then reduced to 4% in May 1720.9

All in all, the Bank of England and South Sea Company margin loans create a substantial influx of capital. Between May 10th and October 6th, 1720 the Bank of England lends a total of £1,476,350 in 958 loans to 659 different shareholder, while it has been estimated that the South Sea Company lent about£9 million on the security of its stock (Dickson, 1967, p. 143).

2.2.3 London Financial Market in the Early Eighteenth Century

(57)

2.3 Data

2.3.1 Stock ledgers and Transfer Files

Our main data sources are the Bank of England (henceforth: BOE) and East India Company (henceforth: EIC) stock ledgers which consist of trader-specific accounts recording buys, sells and BOE share loans for every trader (see Figure 2.1a). Each account is linked to an index containing trader names and characteristics (titles, street address and occupation, see Figure 2.1b). In addition to trader-specific accounts, transactions are also signed by both buyers and sellers in so-called transfer files, which correspond to the actual shares’ sale agreement. BOE transfer files often contain more trader information than ledger books (and indexes) and we use the extra information to enrich our database. We combine the universe of transactions in BOE and EIC with all 1720 trades in a major new share issue of the Royal African Company (henceforth: RAC). Unfortunately, the RAC ledger books have not survived and we have to retrieve names and compute holdings from the RAC transfer files.

We present an example of how share transactions were recorded in the BOE ledger book in Figure 2.1a. Figure 2.1a shows the sale side John Myster’s BOE ledger account. The first entry records a share sale for £500 nominal to Henry Dobson on March 3rd, 1719 under transaction number 8,607. This transaction is recorded as a buy in Dobson’s account on page 5,509 with the same transaction details. Figure 2.1b displays Myster’s ledger index entry and shows that he is a goldsmith living at Charterhouse Square in London. We hand-collect every BOE transaction between 1st August, 1715 and 29th September, 1725 and every East India transaction between 24th June, 1715 and 25th March, 1723. This implies that we observe every transaction for each individual trader in our sample period with buyer and seller identities and trader characteristics. Moreover, we compute holdings and trading gains for each trader on a daily basis. Figure 2.3 shows that our sample consists of 4,657 BOE traders, 1,982 EIC traders and 1,814 RAC traders. The average trader holds £819 (see Table 2.1) and Figure 2.3 shows that there is a fair amount of overlap across companies since 1,519 traders are active in multiple firms and 134 traders hold shares in all three companies. Our sample covers about 50% of the market in terms of January 1720 market capitalization (see Anderson (1787, pp. 104-107)).

Essays in economic and financial decisions of households

Essays in Economic and

Financial decisions of Households

Essays in Economic and Financial Decisions of Households

Acknowledgments

Introduction

Contents

Chapter 1

Student Loan Supply, Parental Saving

& Portfolio Allocation

∗

1.1

Introduction

1.2

Institutional Background

1.3

Data

1.3.1

Household Saving Measures

1.3.2

Descriptive Statistics

1.4

Empirical Strategy

1.5

Results

1.5.1

Savings Response

1.5.2

College Enrollment

1.5.3

Alternative Identification Strategy

1.5.4

Home Equity Adjustments

1.5.5

Heterogeneity and Robustness

1.5.6

Wealth Allocation

1.6

Conclusion

Chapter 2

Does credit affect stock trading?

Evidence from the South Sea Bubble

∗

2.1

Introduction

2.2

Historical setting

2.2.1

The South Sea Scheme

2.2.2

Bank of England and South Sea margin loan facilities

2.2.3

London Financial Market in the Early Eighteenth Century

2.3

Data

2.3.1

Stock ledgers and Transfer Files