The Making of an Investment Banker Macro Economic Shocks Career Choice and Lifetime Earnings Updated 2009

8/14/2019 The Making of an Investment Banker Macro Economic Shocks Career Choice and Lifetime Earnings Updated 2009

1/35

The Making of an Investment Banker: Macroeconomic Shocks,Career Choice, and Lifetime Income

Paul Oyer

July 31, 2006

Abstract

New graduates of elite MBA programs ock to Wall Street during bull markets and starttheir careers elsewhere when the stock market is weak. Given the transferability of MBA skills,it seems likely that any eff ect of stock returns on MBA placement would be short-lived. In thispaper, I use a survey of Stanford MBAs from the classes of 1960 through 1997 to analyze therelationship between the state of nancial market at graduation, initial job placement, and long-term labor market outcomes. Using stock market conditions at graduation as an instrument for rst job, I show that there is a strong causal eff ect of initial placement in investment bankingon the likelihood of working on Wall Street anywhere from three to twenty years later. I thenmeasure the investment banking compensation premium relative to other jobs and estimate theadditional income generated by an MBA cohort where a higher fraction starts in higher-paid jobsrelative to a cohort that starts in lower-paid areas. The results lead to several conclusions. First,random factors play an important role in determining the industries and incomes of members of this high-skill group. Second, there is a deep pool of potential investment bankers in any givenStanford MBA class. During the time these people are in school, factors beyond their controlsort them into or out of banking upon graduation. Finally, industry-speci c or task-speci chuman capital appears to be important for young investment bankers.

I thank Ken Corts, Vicente Cunat, Liran Einav, Eric Forister, Dan Kessler, Kathryn Shaw, Andy Skrzypacz,Till von Wachter, Jeff Zwiebel and seminar participants at Berkeley, Dartmouth, IZA/SOLE, and Gerzensee forcomments. I thank Ed Lazear for both sharing the MBA survey data and useful suggestions. I am also gratefulto Stanfords Vic Menen and Andy Chan and to Whartons Christopher Morris and Jennifer Sheffl er for providinghistorical placement information for their schools and to Kenneth Wong for research assistance.

Stanford University Graduate School of Business and NBER, 518 Memorial Way, Stanford, CA 94305-5015.e-mail: [email protected]. Phone: (650) 736-1047.


2/35

Back in January 1987... Wall Street was booming... When the job off ers rolled in,students played one house against another. They were the supply, and the demand wasstrong... After the crash, the receptions that had once played to packed houses weredrawing a few dozen students. Out went the tenderloin on toast and the shrimp; incame the dips and the hot dogs on toothpicks. The school placement office sent outa memo suggesting career exibility for nance majors like me; we should look intoopportunities in manufacturing and consulting. Brown (1988)

1 Introduction

New MBAs take jobs on Wall Street in large numbers during bull markets. When stock marketconditions are less favorable, a combination of supply and demand factors make new MBAs less

likely to start their careers in the nancial sector. Given the transferability of MBA skills and thenew MBAs reputation for job hopping, one might expect people to move in and out of nance jobssuch that the long-term labor market outcomes of MBA classes would be similar. In this paper,I analyze the degree to which MBAs rst jobs are driven by the stock market and whether theeff ects of initial conditions do, in fact, fade quickly with labor market experience.

I use data from a 1996-1998 survey of several thousand Stanford MBAs. The Stanford GraduateSchool of Business (GSB) alumni provided details on jobs between graduation and the survey, aswell as jobs before studying at Stanford. I rst show that stock returns while GSB students are

in school have a large eff ect on the fraction of any graduating cohort that starts their careers ininvestment banking (which, throughout the paper, I de ne as working for an investment bank, ininvestment management, or for a venture capital rm.) This eff ect is substantial for the class asa whole and for those who did not work in investment banking before getting an MBA. But theeff ect is particularly large for those who worked in investment banking before getting an MBA.

I then show that this eff ect of conditions at the time of graduation is very persistent. Speci cally,I instrument for taking a job in investment banking after graduation with stock returns and WallStreet activity while the person was attending Stanford. Then I estimate the causal eff ect of starting

in investment banking on working in that industry later in the persons career. I nd that startingin investment banking increases the probability of working there at any given point three to twentyyears after graduation by 50-80%. Though it is difficult to be sure that the additional investmentbankers created by bull markets would not otherwise have taken jobs similar to those taken byother GSB graduates, I show suggestive evidence that bull markets cause GSB graduates to be lesslikely to become entrepreneurs or consultants. I also show evidence that the additional investment

1


3/35

bankers created by bull markets are no less interested in nance and no less likely to stay on WallStreet than other investment bankers. Rather than attracting people with marginal skill or interestto Wall Street, bull markets push people towards a career they are as well suited to as those whoneed less encouragement to work there. That is, bull marketscreate long-term investment bankers.

These ndings are consistent with a model where there is a large pool of MBAs in any givenStanford class that would make good investment bankers and they are drawn into that eld whenconditions are favorable. However, they only take the position due to the compensating diff erentialand would be nearly as happy taking a position in another industry. The results also suggest thatMBAs develop nance-speci c human capital while in school and in their initial years on Wall Street,which explains the persistence of initial sector choice. The empirical results are not consistent witha model where there is signi cant heterogeneity in MBAs propensities to be investment bankersand banks have to dig much deeper into this pool when bull markets increase their demand forMBAs.

The nal empirical step shows that these eff ects on career choice have large eff ects on lifetimelabor income. I estimate the eff ect of macroeconomic cohort eff ects on the labor market income of various GSB individuals and classes. Though the ability to estimate income diff erences precisely islimited by the survey design, it is clear that investment bankers earn a substantial premium relativeto other GSB alumni. The premium varies from about 60% for a new MBA on Wall Street relativeto one in management consulting to over 300% for an investment banker fteen years after leavingStanford relative to an average alumnus with the same amount of experience in any other industry.Depending on assumptions about discount rates, length of career, and what other jobs people wouldtake if they had not gone to Wall Street, I estimate that a new MBA that goes to Wall Street canexpect between $1 million and $3 million in discounted additional lifetime income (using $1996)relative to what he would earn if he took a job elsewhere. This premium can be interpreted invarious ways, but may simply re ect the disutility associated with working in investment bankingrelative to other elds.

An example of the experiment in this paper may help motivate the exercise. Members of theStanford MBA classes of 1988 and 1989, all of whom were in school during the stock market crash of the Fall of 1987, were unlikely to take jobs on Wall Street relative to other classes. During the longbull market that started a few years later, new MBAs returned to Wall Street in large numbers.But few members of the classes of 1988 and 1989 moved into investment banking. Also, thoughunusually large numbers of Stanford MBAs from mid-1980s classes went to Wall Street, thereis no evidence that these graduates were less interested in (or committed to) Wall Street careersthan those smaller groups that went to Wall Street during the bear market. This suggests that

2


4/35

many members of the classes of 1988 and 1989 had signi cantly lower lifetime income (though notnecessarily lower utility) due to the timing of their graduation. Depending again on assumptionsabout career length and the like, I estimate that stock return patterns between 1984 and 1989transferred $60 million to $150 million of discounted lifetime labor market income from the Stanfordclasses of 1988 and 1989 to the classes of 1986 and 1987.

There have been numerous prior studies of how macroeconomic conditions when people enterthe labor market aff ect their long-term outcomes. Typically, these papers study large samplesof people across a large set of occupations. Two potential weaknesses of these studies are thatmany employees transition slowly into the labor force, making it difficult to isolate a single entrypoint, and that the average eff ects in these large samples may mask interesting diff erences acrossprofessions. I try to get around these issues by looking at a group of graduates who enter thelabor market at a predictable (and, as I argue in the analysis, exogenous) time. While StanfordMBAs are surely not representative of the economy as a whole, they provide a useful example of high-skilled knowledge workers.

Two recent examples from the literature on cohort eff ects using larger samples are Oreopoulos,von Wachter and Heisz (2006) and Kahn (2005). Oreopoulos et al. (2006) look at the early careersof Canadian college graduates. They nd that graduating in a recession lowers wages early inemployees careers, though it does not have a noticeable eff ect on employment. In their sample,wage diff erences fade over the rst ten years or so after graduation as those who graduated inrecessions move from small, low-paying rms to larger rms (and rms in higher-pay industries)when the economy recovers. Kahn (2005) uses the National Longitudinal Survey of Youth to studyU.S. college graduates in the classes of 1979-1988. She nds that macroeconomic conditions haveimportant wage eff ects (which are more persistent than those found by Oreopoulos et al. (2006)).Consistent with the ndings in this paper, she attributes the long-term eff ects to the types of jobs(occupations) where graduates start.

Other papers that look at the relationship between pay and macroeconomic conditions when jobsstart include Beaudry and DiNardo (1991), Devereaux (2004), and Baker, Gibbs and Holmstrom(1994). Beaudry and DiNardo (1991) nd that macroeconomic conditions at the time an employeestarts a job are related to wages, but that this relationship goes away when controlling for themost favorable economic conditions (that is, the lowest unemployment rate) during the job spell.Devereaux (2004) nds that workers who accept low-paying jobs are stuck with low wages for atleast several years, relative to observably equivalent workers. Those who switch jobs undo someof the ill eff ects of starting a job at the wrong time, but not all. Baker et al. (1994) examine asingle rm. They nd that pay diff ers for cohorts hired at diff erent times and that these diff erences

3


5/35

are persistent.Oyer (2006) looks at the importance of getting a good rst job for another set of highly skilled

professionals PhD economists. He shows that the sector of initial job placement (top researchinstitution, teaching institution, private sector, etc.) has a causal eff ect on the sector of later jobplacement. He also shows rst placement at higher ranked research institutions leads people tobe more productive in terms of research output. Oyer (2006) and this paper both use the stateof the economy when professionals initially enter the labor market as an instrument for initial jobplacement. Together, the papers lead to the conclusion that random events (or luck) early inones career has important long-term eff ects on the sectors where many high-skill and highly paidpeople work, even in labor markets where one might expect people to move relatively freely between rms .

Though they comprise a large and growing segment of the labor market, there are surprisinglyfew studies using micro-data focused on MBAs. Lazear (2005) and Hvide (2005), both of whomuse the same MBA dataset I use, focus on the decision to become an entrepreneur. Reder (1978)and Tracy and Waldfogel (1997) study the determinants of starting salaries for newly graduatedMBAs. I know of no prior academic studies that study compensation of MBAs beyond startingsalaries.1

The rest of the paper proceeds as follows. The next section lays out the theoretical backgroundfor why initial placement might have long-term implications. Section 3 describes the data andSection 4 analyzes how initial MBA placement is aff ected by stock returns. Section 5 documentsa causal eff ect of initial MBA placement on Wall Street on the likelihood of working there as thepersons career develops. Section 6 estimates the amount of discounted lifetime labor market incomethat exogenous shifts into or out of Wall Street careers create for aff ected individuals and for MBAcohorts as a whole. Section 7 concludes with a summary and suggestions for future research.

2 Theoretical Background

Consider a labor market with two sectors, the Financial (f ) sector and the General (g) sector.

Assume that, subject to expending some search eff ort, any MBA can nd employment in either of

1There have been several academic studies of careers in a segment of what I de ne to be investment bankers mutual fund managers. Chevalier and Ellison (1999), for example, show that long-term career concerns aff ectmutual fund manager behavior. Brown, Harlow and Starks (1996) and Chevalier and Ellison (1997) analyze potentialcon icts of interest between investors and mutual fund managers. See Chen and Ritter (1995), Gompers and Lerner(1999), Lin and McNichols (1998), and Michaely and Womack (1999), among others, for studies of the fees charged byinvestment rms and potential con icts of interest within these rms. These papers do not study pay of individuals,however. See Stewart (1993) for an interesting discussion of the allocation of compensation at one investment bank.

4


6/35

these two sectors immediately after graduation. Then, as the person graduates, he compares theexpected utility streams from each of these sectors over the course of his future career. Letu f (w0f )be the expected utility, as of career year0 (that is, upon graduation), of a career that starts in the nance sector. The function captures the persons disutility of eff ort in the nancial sector. Thew term captures the income stream he can expect from a career that starts in that sector.2 Theperson may have various options (either off ers from multiple rms or a choice among the variousfunctions within an investment bank) within the nancial sector, but this utility function capturesexpectations about the job he believes to have the highest expected utility of his options in thissector. Similarly, letu g(w0g ) be the expected utility from his best option in the general sector.

Naturally, the person will start in the nancial sector if u f (w0f ) > u g (w0g ). If all graduating

MBAs were alike in terms of skill and preferences, then either all workers would go to one sector orthe wage diff erential would make employees exactly indiff erent between the two sectors and peoplewould sort randomly. In reality, skills and preferences diff er and people sort between the two sectorsbased on this heterogeneity (as in Jovanovic (1979), Miller (1984), or McCall (1990).) As a resultof this heterogeneity, though the marginal graduate is indiff erent between the two sectors, some(perhaps nearly all) graduates expect to strictly prefer the sector they choose.

Though expected income streams in the two sectors may be correlated for any given cohort of MBA graduates, the state of the stock market is likely to have a larger marginal eff ect on expec-tations about w0f than income in the general sector. This is because favorable conditions on WallStreet will increase demand for labor and, at least, short-term expected pay. Also, under the stan-dard assumption that stock returns follow a random walk, any short-term change in stock marketconditions should increase long-term expectations about the level of stock prices.3 Therefore, giventhat a bull market will increaseu f (w0f ) relative to u g (w

0g ) for some MBAs (and not decrease it

for any), it would not be a surprise to nd that more people choose jobs in the nancial sector inclasses that graduate when stock prices and returns are relatively high.

The question of interest, however, is whether this initial eff ect of a bull market on industrychoice is persistent. At yeart , a person who took a job in the nancial sector upon graduationfaces expected utility from staying in the nancial sector of u f (w tf ). He can also switch to the generalsector where he can expect utility of u g(w tg). There are reasons to expect that, if u f (w

0f ) > u

g(w0g ),then u f (w tf ) will be greater than u g(w tg). That is, people who show an initial preference for the

2The person can change sectors. So,w re ects the income in both sectors and the persons expected probabilityof working in each sector at any given time in the future.

3Also, if MBAs make career decisions assuming momentum in stock prices (which would be consistent with theretirement allocations studied by Benartzi (2001)), then high stock returns would encourage them to be more inclinedto take a job on Wall Street.

5


7/35

nancial sector are likely to nd the work there relatively pleasant and one would expect that tobe the case later. There are three underlying models (or classes of model) that would predict thosewho start in the nancial sector are more likely to work there later on, each of which has distinctempirical predictions:

Model 1 Worker Heterogeneity : Suppose employees diff er in type, as in the Spence (1973)signaling model. Speci cally, suppose that there are two types of people who are interested instarting their careers in investment banking. The rst type, bankers, will be highly productiveinvestment bankers because their skills match the production function well. Non-bankers havea high marginal utility for money (and so seek the highest paying job possible no matter theirskills.) When times are lean on Wall Street, the second type shows less interest in working there(that is, the expected value of wf is lower so they consider alternatives.) When conditions improve,banking rms are reluctant to hire those who did not start their careers on Wall Street because theyhave revealed themselves to be unproductive investment bankers. But, when hiring new MBAs,they have no method for separating the productive bankers from the non-bankers. After some timeworking on Wall Street, the non-bankers are revealed (after a period of enjoying a high income)and they are either red or choose to move to the general sector. This model predicts that bankersend up in banking and non-bankers do not, no matter when they enter the market. Therefore,though it implies that there would be acorrelation between starting in banking and working theresubsequently, there is nocausal eff ect of rst job on subsequent jobs.

Model 2 Homogeneous workers with Investment-Banking Speci c Human Capital : Supposethere is a large pool of MBAs that would be productive investment bankers.4 Much of this poolis nearly indiff erent between the two sectors, given the expected income diff erences over time.However, anticipating opportunities in banking, those who go to school during bull markets developWall-Street-speci c human capital both in school and at the beginning of their post-graduationcareers.

To be a little more concrete, consider the model in Gibbons and Waldman (2006). They modeltask-speci c human capital and show that it can lead to long-term eff ects of initial job placementon the types of jobs workers hold. In their model, those hired under favorable conditions areinitially given high value tasks and develop more valuable human capital that persists throughouttheir careers. A similar idea, industry-speci c human capital, has been shown to be important in thelabor market generally (see, for example, Parent (2000)) and it is often cited as a primary reasonthat wage losses can be quite high for displaced workers (see Neal (1995).) Therefore, nance-

4This group need not be the entire MBA class, but enough to meet hiring demands during bull markets.

6


8/35

speci c skill may widen the gap betweenw tf and w tg as a career in the nancial sector continues(that is, as t increases.)5

Finance-speci c human capital would lead those who go to Wall Street to be relatively produc-tive there and would lead to a causal link between starting a career on Wall Street and workingthere later on. Homogeneity implies that those who go to Wall Street during bull markets wouldnot be noticeably diff erent from those who go to Wall Street during bear markets. As a result,even though the entering pool of bankers would be larger in bull markets, they would not be anyless prone to success in banking than those who choose to go to Wall Street during a bear market(in stark contrast to Model 1.) This would lead to the empirical prediction that those hired duringbear markets would be as likely to stay in investment banking as those hired in bull markets andthat those hired in bull markets would be no less able (in terms of nance training and interest)than those hired in bear markets.

Model 3 Some Heterogeneity and Some Speci c Human Capital (Hybrid Model) : It seemsunlikely that the world is as stark as either of the two models just sketched. So consider a worldwhere there is heterogeneity in how well MBAs are suited to work in banking and where bankersdevelop Wall-Street-speci c capital. During bull markets, increased demand for MBAs would leadto the marginal hire being less t for a career in banking. However, because these lower- t bankersdevelop nance-speci c human capital, at least some of them stay because their speci c capitalmakes them better bankers than new graduates who are naturally better ts. That is, like Model2, the hybrid model makes the empirical prediction that speci c human capital will cause thosewho start their careers on Wall Street to be more likely to work there later on, even controlling forability or t. However, unlike Model 2, given sufficient heterogeneity, banks would replace the worst ts hired in a bull market with the best ts available during a bear market. This distinction leadsto the prediction that new MBAs who go to Wall Street during bull markets will be, on average,less t for careers in banking than new bankers who graduate in bear markets.

In the sections that follow, I investigate the predictions of these three models. First, I showthat new MBAs are more likely to go to Wall Street during bull markets, which is an importantassumption of each model. Second, I show that those who go from Stanford Business School directlyto Wall Street are more likely to work on Wall Street later in their careers, which is consistent with

5While I will discuss speci c human capital as though it is a productivity investment, it could simply be the resultof lower transaction costs. For example, models where incumbent rms have more information about an individualthan other potential employers (such as Akerlof (1970) and Waldman (1984)) or pure search cost models would leadto stickiness in choice of industry. The cost of search, any cost of switching industries, or aversion to the risk of unknown features of the general sector will loweru gt (w g ) for any employee in the nancial sector in the same waythat speci c nance skills raiseu f t (w f ). Another related alternative with the same implications is that, as workersget accustomed to a job, the disutility of eff ort may decline.

7


9/35

each of the models. Then, using Wall Street conditions while MBAs are in school as an instrumentfor rst job, I show that the link between initial placement and later employment on Wall Streetis causal in the sense that an MBA who starts on Wall Street is more likely to work there laterbecause he started his career there . I also show evidence that suggests that those who take jobson Wall Street after graduating in bull markets are no less interested or tied to Wall Street. As Iexplain in Section 5.2, the evidence suggests that random factors play an important long-term rolein MBA careers and that model 2 most closely matches the patterns in the data. I then go on tomeasure the magnitude of the eff ects of these random shocks.

3 Data

The data is based on a mail survey of Stanford Graduate School of Business (GSB) Alumni. The

survey was conducted in 1996 and 1998 and had a response rate of approximately 40%. Surveyrespondents provided detailed job histories, including jobs before they entered Stanfords MBAprogram. I use information gathered from members of the GSB classes of 1960 through 1995. Idropped any job where the person worked less than half time. If the person reported two jobssimultaneously, I use the one which he reports working more hours. If the hours are the same, I usethe job where salary information is available.

Table 1 provides summary statistics. Column 1 shows averages for all post-graduation per-son/year observations, column 2 provides details on each persons job the year after graduation,

and column 3 summarizes the persons job at the time of the survey. Observations in this tableand throughout the analysis are a snapshot of the persons job as of the end of January of eachyear.6 As noted above, this group is not representative of the broader economy (even those withgraduate degrees.) Based on analyses of the 2001 American Community Survey and 1996 respon-dents to the NLSY, it is clear that Stanford MBAs are higher paid, much more likely to work ininvestment banking or consulting, and slightly less likely to switch jobs than other people who workin for-pro t businesses and hold masters degrees.7 The goal of the following analysis is to use thedetailed information about Stanford MBAs, despite the fact that they are a fairly elite group, to

isolate possible mechanisms that lead to cohort eff ects. These mechanisms may or may not apply

6Columns 1 and 4 include all relevant person/year observations for a given person while the middle columnsinclude at most one observation per person. Because older people have, on average, more years of data, the data incolumns 1 and 4 are weighted towards earlier graduates. Column 2 does not include people who were unemployed inthe January after graduation and Column 3 does not include those who were unemployed (usually due to retirement)at the time of the survey.

7Further details on the similarities and diff erences between Stanford MBAs and those in these alternative datasetsare available from the author.

8


10/35

Total First Job Survey Job I-bank JobsFemale 11.6% 19.3% 19.0% 10.4%Work in USA 86.1% 83.2% 83.2% 86.6%Minority 7.3% 12.2% 11.9% 6.8%Investment Banking 14.5% 14.2% 18.3% 100%Consulting 10.7% 18.6% 13.6% 0%High Technology 10.6% 10.9% 12.0% 0%Partner/Owner 24.9% 7.6% 31.4% 33.8%Founder 11.4% 2.9% 15.9% 13.6%Employees (median) 1,000 2,000 450 500Salary > $50,000 77.8% 41.4% 93.4% 89.5%Salary > $100,000 47.8% 5.6% 71.3% 76.1%Salary > $500,000 9.0% 0.1% 13.7% 31.5%Graduation Year 1973.5 1980.4 1980.1 1975.5Age 39.6 29.4 44.4 39.1Total Person/years 62,115 3,782 3,886 8,844

Table 1: MBA Sample Summary Statistics. First Job is the job the person held in the Januaryafter graduating. Survey Job is the job held when answering the survey in 1996 or 1998. I-bankJobs is the subset of column 1 person/years where the respondent was employed for an investmentbank, a money manager, or a a venture capital rm. Employees is the number of employees atthe rm where the respondent worked.

to broader samples.Respondents also provided details on the industries where they worked. I de ne investment

banking (or, in some tables and gures I-bank) broadly to include investment banking, investmentmanagement, and venture capital. The nal column of Table 1 provides information about allperson/year observations within this industry. Men and non-minorities are slightly over-representedin this group. Investment banking has become more common over time. However, as will becomeclear below, the fraction in investment banking does not change noticeably with age or years of experience (unlike, for example, management consulting, which is much more likely to be a rst jobthan a job held later in a career.) The investment bankers as a whole are somewhat more likely tohave founded the business where they work than the rest of the sample. This is because venturecapitalists and money managers are much more likely to be founders, while those who work forinvestment banks are less than half as likely as the rest of the sample to be founders.

The usefulness of the income data is limited in three ways. First, the survey asked people theirsalaries. Individuals may have interpreted this question diff erently, with some including bonuses andthe value of equity. But the reported numbers are likely understatements of labor market earningsas a whole. Second, the survey asked for the beginning and ending (or current, if the person holdsthe job at the time of the survey) salary on each job. Therefore, I cannot create a dataset with

9


11/35

person/year income information. I primarily rely on the cross-section of income information atthe time of the survey. Finally, the survey provided categorical answers to the income questions.Respondents could either say that the relevant salary was under $50,000, between $50K and $75K,between $75K and $100K, between $100K and $150K, between $150K and $200K, between $200Kand $300K, between $300K and $400K, between $400K and $500K, between $500K and $750K,between $750K and $1 million, between $1 million and $2 million, and over $2 million. In theanalysis that follows, I assume the persons income is the mid-point of the reported range and thatit is $3 million if the person reports income greater than $2 million.

Despite these limitations, there are two indications that the data are reasonably accurate. Firstof all, the average starting salaries for the class of 1995 reported by the Stanford GSB career officeis approximately equal to the average I calculated from the survey. Retrospective salary data maynot be as accurate, but I only use the wages reported at the time of the survey. Second, the fractionof each class that the GSB career office reported taking an initial job in investment banking closelytracks the fraction of each class that I calculated using retrospective job information. For the GSBclasses of 1976-1994 (the classes for which I have information from both the career office and thesurvey), the correlation between the fraction of the class starting in investment banking based onmy calculations and on surveys by the career office at the time of graduation is 0.84. Given thatboth surveys were voluntary and the industry choices varied from year-to-year in the career officesurvey, this is a very high correlation.

To get a sense of the dynamics of movement between industries for the respondents, Table2 shows transitions into and out of investment banking for the rst 14 years after graduating.Graduates are listed as moving into investment banking in year 1 if they worked in the industryright after graduation but had never worked in that industry before getting an MBA. Those whomove out of the industry in year 1 worked there at some point before school but did not take a jobin the industry upon receiving an MBA. In subsequent years, movement in or out simply indicatesthat the person switched industries during the year.

In the rst few years after receiving MBAs, 5-10% of the people in investment banking leavethe industry each year. Movement slows after about year ve. Movement into the industry issomewhat higher than exit in the rst few years after graduation, but the fraction of a typicalgraduating class that works in investment banking does not change sharply as time progresses.The table also suggests that investment bankers have stronger industry-speci c attachments than rm-speci c attachments. For every investment banker that leaves the industry, about three switch jobs within the industry.

Aside from the GSB survey, I also use stock market data and data on the initial placement

10


12/35

Post-MBA Year Sample I-bank I-bank in I-bank out I-bank within1 3,754 533 362 173 N/A2 3,613 512 32 18 563 3,459 493 46 26 854 3,326 489 46 17 815 3,143 486 52 28 866 3,005 471 27 21 647 2,877 445 27 23 628 2,723 425 24 15 529 2,584 398 20 20 4510 2,475 380 19 13 5411 2,371 356 13 16 3612 2,264 333 16 15 3413 2,156 305 15 11 3014 2,061 291 17 6 35

Table 2: MBA Transitions. I-Bank is the total number of people working at an investment bank,money manager, or venture capital rm. The in column lists the number of people who movedto the relevant type of rm a given number of years after receiving MBAs and out column liststhe number who moved out of the type of rm. Year 1 in/out refers to change relative to pre-MBA job. The within column lists the number who changed employers but were investment bankersat both jobs. Sample size changes because the last observation for a given person is the year of thesurvey (1996 or 1998).

of MBAs graduating from the University of Pennsylvanias Wharton School. I de ne the 2-yearS&P return for a given MBA class as the percentage change in the S&P 500 in the two-year period

through the end of June when the person graduates. This measure has the nice feature that,with very few exceptions, it is fully determined during the period after the person has decided toenter Stanfords MBA program. Though it is currently common for MBA students to accept off erswell before the actual graduation date, I focus on classes graduating in 1995 and earlier when therecruiting season ran closer to graduation. As another measure of demand for labor by investmentbanks, I use Mergerstat LLCs measure of all announced mergers and acquisitions (M&A) activityduring each calendar year involving US rms as either buyer or seller. Details of how Mergerstatcalculates this measure are available on their website. I relate the fraction of a Stanford MBA class

that takes a job on Wall Street to M&A activity in the calendar year before they graduate. For theWharton classes of 1973 and later, I de ne a variable that is the fraction of each class that wentinto investment banking. This information is based on surveys of each graduating class done byWhartons career office.8

8Because Wharton changed the way it reported (and, perhaps, the way it calculated) the fraction going intoinvestment banking starting with the class of 1984, I include a class of 1984 or later indicator variable in any

11


13/35

- . 5

0

. 5

1

1 . 5

1970 1975 1980 1985 1990 1995MBA Class

Inv. Ban k j ob 2 -yea r S&P 50 0 Retur n

Figure 1: Stock Returns During School and Investment Banking Job Placement

4 Initial Job Placement

Figure 1 shows how the fraction of graduates whose initial placement is at an investment bank(normalized to one for the class of 1994) rises and falls with the two-year return on the S&P 500as of June of the year of graduation. The graph shows that the fraction of graduates taking jobson Wall Street is at least somewhat responsive to recent stock market returns. The graph showsthat graduates went to Wall Street in large numbers as the market boomed in the mid-1980s.But, after the market crash of 1987, there was a noticeable drop in the fraction of graduates goingto Wall Street. While the swings in the fraction of the class going into investment banking weremost noticeable around the 1987 crash, the relationship between investment banking and S&Preturns is strong throughout and the results are not sensitive to dropping the graduating classes of 1986-1989.9

Figure 2 shows the relationship between MBA placement and M&A activity for US rms in thecalendar year before the person graduates from Stanford. Throughout the paper, I use the log of the value of M&A activity, adjusted for in ation. The graph shows a linear transformation of thisvariable so that the possible values coincide with those for the investment bank placement variable.

analysis where I use the Wharton career data.9Details on the initial placement of Stanford MBAs from the classes of 1997-2005, including industry and com-

pensation details, can be found at http://www.gsb.stanford.edu/cmc/reports/index.html.

12


14/35

0

. 5

1

1 . 5

1970 1975 1980 1985 1990 1995MBA Class

Inv. Bank job M&A Activity

Figure 2: M&A Activity During School and Investment Banking Job Placement

The graph shows that the relationship between prior year M&A activity and investment bankingplacement is very strong and that M&A activity is a useful predictor of MBA placement.

As hinted at by the prior discussion, one potential problem with interpreting the gures wouldbe if people who wanted to work in investment banking somehow timed the market. Also, if therewere predictable cycles in Wall Street hiring and the market, then one might worry that a cohortsinterest is correlated with stock returns rather than their rst position being driven by it. If thestock market is anywhere close to perfect and a random walk, then this concern is not an issue inFigure 1. Unlike stock returns while the person is in school, an argument could be made that theM&A variable will be predictable to a potential student before entering Stanford and, therefore,may aff ect his decision about whether to attend. I would expect, however, that this would dampenthe relationship between this variable and post-MBA investment banking jobs. This is because, if a person who is interested in nance anticipates a good year is about to take place in investmentbanking, he might be inclined to delay his entrance into business school until a time when theopportunity cost would be lower. If this were the case, then those who graduate after good M&Ayears would be less interested in nance than those who graduate after slow M&A years.

To insure timing of the market by students is not an issue, I control for pre-MBA investmentbank experience throughout the analysis when looking at post-MBA job selection. Also, considerFigure 3, which shows the relationship between the fraction of an MBA cohort that worked ininvestment banking before school and stock returns while the cohort was in school. The graph

13


15/35

- . 5

0

. 5

1

1970 1975 1980 1985 1990 1995MBA Class

Pre -MBA I nv. Bank job 2-yea r S&P500 Re tu rn

Figure 3: Stock Returns and Pre-MBA Employment

shows that there has been a gradual trend up in terms of the fraction of pre-MBA investmentbankers. But the year-to-year changes in the two lines are generally unrelated, with the possibleexception that many investment bankers returned to school when the market crashed in 1987. Moreformally, I analyzed the relationship between going into investment banking upon MBA graduation

and the S&P return in the two years prior to enrolling at Stanford. As I show below,S&P returns while at Stanford are an important predictor of starting ones post-MBA career on Wall Street.However,S&P returns in the period before enrollment always have a small, insigni cant, and fairlyprecisely estimated relationship with the likelihood of being a post-MBA investment banker.

While these gures provide evidence that there is a relationship between stock returns whilestudents are in school and their rst job, I will now be more precise in investigating this relationshipas it forms the rst-stage of the instrumental variables analyses that follow. De ne F it to be anindicator for whether personi who enters the job market in yeart starts his career in investment

banking. Following the notation in Section 2,F it = 1 if uf 0 (wf ) > u

g0 (wg). F it is observable in the

survey data, so I estimate linear probability regressions of the form

F it = t + X it + it (1)

where t is a measure of demand for MBAs in investment banking in yeart , X is a vector of observable characteristics (linear, quadratic, and third-power time trends, gender, ethnicity, and

14


16/35

whether the person ever worked as an investment banker before entering Stanfords MBA program),and it includes unobservable individual characteristics that aff ect the demand by investment banksfor the persons services and the persons preferences for working in investment banking relative toother industries.

Measures of market demand () include the two-year S&P 500 return through the end of Juneof the year the person graduates, the Mergerstat index, and the fraction of the relevant graduatingclass from Wharton that initially placed in investment banking.10

The results are shown in Table 3.11 Panel A focuses on the S&P 500 to establish the basicrelationship between stock returns and MBA placement while Panel B includes the M&A andWharton measures. Column 1 of Panel A establishes the basic relationship between stock returnsand MBA placement. It shows that, in a year when the S&P 500 increases by 18.5% (one standarddeviation) relative to another year, a typical Stanford graduates probability of entering investmentbanking increases by about two percentage points. Given a base probability of 14%, this means thata one standard deviation increase in stock returns increases initial investment bank employmentlikelihood by about one seventh. While the state of the stock market is certainly not the only factorthat determines whether a person works in investment banking or not, it is an important predictor.

Column 2 shows that those who worked in banking before getting an MBA are much morelikely than other students to work in investment banking immediately after graduating but thatcontrolling for pre-MBA industry does not change the relationship between stock returns and rst job. Column 3 limits the sample to those who did not work in banking before getting an MBA andshows a similar eff ect of stock returns on rst job. This is not surprising, given that over 90% of the sample in columns 1 and 2 is also in the column 3 sample.

Columns 4 and 5 limit the sample to groups that have already shown some interest in nance bythe time they attend Stanford. Column 4 includes the eighteen percent of the sample that workedin any type of nance job before getting an MBA (including investment or commercial banking,insurance, real estate, accounting, or other nancial services) while column 5 limits the samplefurther to the nine percent that worked for an investment bank before entering the Stanford GSB.The estimated eff ect of stock returns on these samples is noticeably larger than for the broadersample. A one standard deviation increase in the S&P 500 return increases the probability of

10 The measures of do not vary within a graduating class, so all standard errors are clustered at the class level. Ialso tried the analysis taking the more conservative approach of averaging at the graduating class level. This did notalter any of the conclusions that follow.

11 Table 3 displays the results of linear probability (OLS) regressions that are the rst-stage regressions in IVanalyses below. The results (in terms of the signi cance of the estimates and the marginal eff ects of the coefficients)are nearly identical when using logit or probit speci cations.

15


17/35

(1) (2) (3) (4) (5)

Panel A: S&P 500 Only2-year S&P return 0.1034 0.1052 0.0896 0.2063 0.2836

(0.0345) (0.0363) (0.0386) (0.0692) (0.1017)Pre-MBA I-bank 0.3712 Dropped 0.3506 Others

(0.0254) (0.0330) DroppedSample (Pre-MBA) All All Non-IB Finance IBR 2 0.0353 0.1236 0.0172 0.1891 0.0567N (People) 3,547 3,547 3,230 624 317

Panel B: Other Instruments2-year S&P return 0.1062 0.0081 0.0148

(0.0417) (0.0508) (0.0559)Log($ M&A) 0.0718 0.0705 0.0137

(0.0186) (0.0153) (0.0201)Wharton I-Bank 0.6548 0.6450 0.5888

(0.1799) (0.1794) (0.2077)Pre-MBA I-bank 0.3653 0.3660 0.3697 0.3697 0.3694

(0.0257) (0.0256) (0.0265) (0.0264) (0.0268)Sample (Pre-MBA) All All All All AllR 2 0.1277 0.1303 0.1363 0.1363 0.1365N (people) 2,943 2,943 2,410 2,410 2,410

Table 3: Initial Placement in Investment Banking. Coefficients are linear probability estimates(OLS) where the dependent variables are indicators for the person being employed in investmentbanking (including money management and venture capital) as of the January after graduation.Each regression also controls for gender, ethnicity (through indicators for Black, Hispanic, andAsian), year, year squared, and year to the third. Pre-MBA I-bank equals one if, before startingMBA studies, the person ever worked in investment banking. Log($ M&A) is the log of the realvalue of M&A transactions involving US rms in the calendar year before the person graduatedfrom Stanford. Wharton I-bank is the fraction of graduating Wharton MBAs that took jobs ininvestment banking in a the year the Stanford MBA graduated. The M&A and Wharton variables

are only available for certain years, so the sample size is smaller. Unemployment is the nationalaverage unemployment rate as of June of the year the person graduated. The S&P returns arethrough June of the year the person graduated. Standard errors (in parentheses) are adjusted forany correlation within graduating class.

16


18/35

one of these graduates entering (or returning to) investment banking by four to six percentagepoints. This diff erence compared to the full sample is to be expected for at least two reasons.First, the unconditional probability of these groups going to investment banking immediately aftergraduation about a third (column 4) and about a half (column 5). So the increase of four to sixpercentage points is about the same proportionately as the two percentage points for the largersample. Second, a typical Stanford GSB class includes a large fraction of people that would notseriously consider seeking a job in investment banking. For these people,u f (w0f ) is always less thanu g(w0g ), no matter what the level of t . These people attenuate any eff ect of in columns 1-3 of Table 3.

Panel B shows the relationship between Stanford MBAs going to Wall Street and other measuresof investment banking labor demand. The sample size is smaller in Panel B than in Panel A becausethe M&A variable starts in 1969 and the Wharton variable starts in 1975. Column 1 shows thatan increase of the M&A measure by one standard deviation (0.65) is associated with nearly anadditional ve percent of Stanford graduates going into investment banking. Column 2 shows thatthe S&P 500 and M&A eff ects are not changed much when included in the same regression.12

Column 3 shows a strong correlation between the fraction of graduating MBAs from Stanford andWharton that go to Wall Street. As one might expect, when there is more Wall Street demandfor Stanford MBAs and/or Stanford MBAs are more interested in Wall Street, the same holds forWharton MBAs. Columns 4 and 5 show that the Wall Street measures off er no explanatory powerwhen the Wharton variable is included.

In addition to the explanatory variables in Table 3, I will also instrument for initial job choicewith a set of indicator variables for each graduating class. This allows the rst stage to be estimatedless parametrically than the estimates in Table 3, though it introduces other issues that I discussbelow.

Given that the fraction of a class that goes to Wall Street uctuates signi cantly from year toyear, it might seem reasonable to think that there are important diff erences in the types of MBAsthat go to Wall Street in good times and in bad times. Bull markets attract many extra studentsto Wall Street and one might expect these marginal graduates to be less interested in nancethan those who go to Wall Street when times are leaner. That is, assuming bull markets raise allstudents estimates of u f (w0f ), both models 1 and 3 in Section 2 suggest that the marginal studentfor whomu f (w0f ) roughly equalsu g (w

0g ) will be less of a natural t for a Wall Street career. To

12 In unreported regressions, I also found that the value of initial public off erings (as estimated byInvestors Dealers Digest magazine) in a calendar year is correlated with the fraction of Stanford graduates the following year that goto work in investment banking. However, this relationship becomes small and insigni cant when controlling for theM&A variable, so I do not use it in the analysis.

17


19/35

investigate this idea, I matched survey responses by members of the classes of 1984-1995 with theclasses they took as students at Stanford GSB. Given that the available data only includes twelveyears, the macroeconomic variation is not as great as one might hope and I will not present formalanalyses. However, it appears that students who went to school during strong stock markets tookmore nance classes and that this is especially true among those who went on to be investmentbankers. Finance enrollments dropped dramatically after the Fall of 1987 stock market crash.While the data do not allow a great deal of statistical precision, it is clearly NOT the case thatthose who went to Wall Street during the bull markets of the mid-1980s and early 1990s were lessprepared for nance careers than those that went to Wall Street in the bear markets of 1988-1989and 1993-1994.

In summary, stock returns while Stanford MBAs are in school have a statistically and econom-ically signi cant eff ect on the likelihood that they work in investment banking immediately aftergraduating. That is, exogenous shocks aff ect the initial career choices of this sample. In the rest of the paper, I examine how long these shocks go on aff ecting the graduates and whether it has anyeff ects on their incomes.

5 Initial Conditions and Long-Term Outcomes

5.1 Persistence in Investment Banking

Figure 4 provides an initial look at how rst job after MBA graduation is related to jobs heldlater. The graph shows the fraction of each graduating class that initially takes a job in investmentbanking and then what fraction of the class works in banking for up to ten years after graduation.As the graph shows, classes where a relatively large set of people go into banking still have ahigh fraction in banking at any given year over this rst post-graduation decade. For example,among those classes where there was a substantial drop in people entering investment banking inthe late 1980s after the crash of 1987, representation on Wall Street remained low over the entireavailable sample. While this suggests that an exogenous shock has long-term eff ects on humancapital investments and careers, I now consider this issue more formally.

I model MBAis industry as of yeart by updating equation (1) to

F it = t + X it + F 0i + it (2)

where F 0i is an indicator for whether the person worked in investment banking in the rst yearafter graduation. OLS will not reveal thecausal eff ect of F 0i on F it because an individual with an

18


20/35

0

. 1

. 2

. 3

F r a c

t i o n

i n I n v e s

t m e n

t B a n

k i n g

70 75 80 85 90 95MBA Class

First Year Pos t-Grad uati on Fourth Ye arSeventh Year Tenth Year

Figure 4: Fraction of Class in Investment Banking 1-10 Years After MBA

appropriate set of skills and/or tastes for a given industry will be more likely to both start in andeventually work in that industry. That is, bothF 0i and F it will be correlated with unobserved tasteand ability captured by , so that I would expect OLS estimates of to be biased upwards.

However, to establish the basic relationship between initial and long-term investment bank

employment that is predicted by all three models discussed in Section 2, I start by studying the re-lationships between long-term investment banking attachment, initial investment bank placement,and stock returns while in school. This provides a useful benchmark to compare with the IV esti-mates below and allows me to see how the basic relationship between initial and later employment(F 0i and F it ) varies with the state of the market at graduation (t ). I run OLS regressions where anobservation is a person/year at least two and a half years after the person graduates from Stanford.The dependent variable is one if the person is an investment banker at the time of the observationand zero otherwise. Results are in Table 4.

As expected, there is a strong relationship betweenF 0

i and F it . The probability that a personwho starts in investment banking will work there in a later year is about eighty percentage pointshigher than someone who starts elsewhere. Controlling for starting in investment banking afterbusiness school, the relationship between working in investment banking before business school andworking there later is small.

I repeat the analysis dividing the sample into groups that were in school when returns wereabove the sample average (bull markets), when returns were below an approximation of the risk

19


21/35

(1) (2) (3) (4)Initially I-Bank 0.7881 0.8072 0.7628 0.7794

(0.0157) (0.0279) (0.0242) (0.0227)

I-Bank Pre-MBA 0.0454 0.0607 0.0346 0.0257

(0.0178) (0.0176) (0.0450) (0.0486)State of Market at Graduation All Years Bull Average BearN (observations) 49,920 21,087 10,641 18,192N (people) 3,313 1,787 511 1,015

Table 4: Industry of Longer-Term Job. All columns are results of linear probability regressions. Thedependent variable, which is based on a persons job as of the end of January in a year at least twoand a half years after graduation from Stanford GSB, equals one if the person works in investmentbanking (including money management or venture capital.) Initially I-Bank equals one if theperson was working in investment banking in the January after graduation. I-Bank Pre-MBAequals one if the person worked in investment banking before studying at Stanford GSB. Bull

market indicates the two-year average annual return of the S&P 500 was greater than 13% whilethe person attended Stanford GSB. Bear market indicates annual return of less than 5%. Eachregression includes indicator variables for gender, Black, Hispanic, Asian, year of observation, andyears since graduation. Standard errors (in parentheses) are adjusted for any correlation within agraduating class.

free rate (bear markets), and other years (average markets.) The most noteworthy result in Table4 is the consistency of the relationship between starting in investment banking and working therelater. Columns (2)-(4) show that the eighty percentage point diff erence holds no matter what the

conditions of the market when the person graduated. To the extent that there is any diff erence atall, it appears that those who go to Wall Street during bull markets are the most attached while thesorting model would predict the opposite. Combined with the suggestive evidence on nance classenrollments in the last section, this indicates that there is no evidence that bull markets attractless quali ed or less interested candidates.

I now estimate the causal eff ect of starting in investment banking (F 0i ) on working there later(F it ) by using instruments forF 0i . The return on the S&P 500 has the ideal features of a validinstrument. It aff ects initial placement of MBAs, as shown in Table 3, but I see no reason it would

aff

ect where MBAs work later except through the eff

ect on initial placement. I would expect theM&A variable to be a similarly valid instrument, though a better case can be made that M&Aactivity can be predicted ahead of time than stock returns. As noted above, to the extent thatM&A activity can be predicted, I would expect it to be negatively related to unobserved nancetaste and to dampen IV estimates of when estimating equation (2).

As mentioned above, year indicators are more powerful instruments. If this is because the classindicators are more exible and direct measures of demand for MBAs in a given year, then they

20


22/35

constitute a valuable instrument. That is, if classes of Stanford MBAs are generally similar inskills and interests, at least controlling for pre-MBA industry, then the class indicators are validinstruments because they capture diff erences in career prospects and demand by employment sector.However, if there are important diff erences from year to year in the types of people admitted toStanford, then this may not be a valid approach. The fact that all the MBA results are not generallysensitive to controlling for pre-MBA experience and that the results are generally similar (thoughless precisely estimated) when using stock return and unemployment as instruments is somewhatcomforting, but it is still important to bear in mind the underlying assumptions when using theclass indicators as instruments.

I also use the fraction of MBAs graduating from the University of Pennsylvanias WhartonSchool that went into investment banking as an instrument. This gets around some of the potentialproblems with using class indicators as instruments, while hopefully capturing supply and demandfeatures of the MBA/investment bank match in a given year that are not captured by the stockmarket variables. Unless Wharton and Stanford changed their admissions and recruiting policiesin a similar way or the types of people that applied to top MBA programs changed systematically(neither of which is impossible), initial Wharton placement should be correlated with initial Stan-ford placement but not longer-term Stanford career choice. Because the Wharton information isonly available after the class of 1973, the sample size is reduced when using this instrument.

Two-stage least squares (that is, linear probability with instrumental variables) estimates of equation (2) with instruments forF 0i are displayed in Table 5.13 Panel A uses the one-year andtwo-year S&P returns while the person is at Stanford as instruments for rst job after graduation.14

The point estimate in Column 1, which includes all available person/years, indicates that a personwho takes a job on Wall Street upon graduation has about a 36% higher probability of working onWall Street in a later year than someone whose rst job is elsewhere. Column 2 shows that thepoint estimate drops noticeably when focusing on those who did not work on Wall Street beforegoing to Stanford. Neither of these results is statistically diff erent from zero nor from the higherand more precise estimates in the rest of the table. Columns 3 and 4 of Panel A show that the eff ectis more precisely estimated, and stronger, for those who worked in nance (column 3) or on WallStreet (column 4) before studying at Stanford than on the rest of the sample. For the pre-MBA nance sample, going to work on Wall Street right after graduating from Stanford increases the

13 The linear probability speci cation is relatively simple to implement and keeps the interpretation straight-forward.Angrist (2001) argues that linear probability is an appropriate empirical approach in contexts such as this.

14 Stock returns during the second year at Stanford have a slightly higher eff ect on initial placement than rst yearreturns, so I include them individually as instruments.

21


23/35

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

Panel A: S&P 500 InstrumentsInitially I-Bank 0.3572 0.1300 0.6573 0.7963

(0.2953) (0.4966) (0.2803) (0.1397)

Sample (Pre-MBA) All Non-IB Finance IBN (observations) 49,920 47,860 5,114 2,060N (people) 3,313 3,052 536 261

Panel B: S&P 500 and M&A InstrumentsInitially I-Bank 0.5160 0.5395 0.6981 0.3988

(0.2189) (0.2484) (0.1584) (0.3340)


Panel C: Wharton InstrumentInitially I-Bank 0.7594 0.7605 0.5689 0.7820

(0.1269) (0.2174) (0.1476) (0.1347)


Panel D: Class InstrumentsInitially I-Bank 0.7959 0.8372 0.7903 0.9654

(0.0771) (0.0941) (0.0964) (0.0537)


Table 5: Industry of Longer-Term Job. All results are based on two-stage least squares linearprobability regressions. The dependent variable, which is based on a persons job as of the endof January in a year at least two and a half years after graduation from Stanford GSB, equalsone if the person works in investment banking (including money management or venture capital.)Initially I-Bank equals one if the person was working in investment banking in the January aftergraduation. The Non-IB (IB) Pre-MBA sample is limited to people that did not (did) work ininvestment banking before studying at Stanford GSB. The Finance Pre-MBA sample is limitedto people that worked in investment banking, accounting, commercial banking, insurance, realestate nance, or other nancial services before studying at Stanford GSB. S&P instruments forInitially I-Bank, which are measured as of time of MBA graduation, are 1-year and 2-year S&Preturn. M&A Instrument is log of value of M&A activity in the calendar year before the persongraduated. Wharton instrument is the fraction of new Wharton graduates that took investmentbanking jobs in the year the Stanford MBA graduated. Class instruments is a set of graduatingyear indicator variables. The Wharton and M&A instruments are not available for all classes, sothe sample size is smaller. Standard errors (in parentheses) are adjusted for any correlation withina graduating class.

22


24/35

likelihood of working there in later years by 66%. The eff ect is 78% for those who return to WallStreet right after going to Stanford. Overall, Panel A indicates that there is a strong causal eff ectof initial Wall Street employment on longer-term Wall Street employment among the subset of theclass that had nance experience before getting an MBA.

Panel B adds the M&A variable to the instrument set. Assuming potential entrants to MBAprograms cannot anticipate M&A activity six months or more in advance (or, alternatively, thatthey can anticipate it but it does not aff ect their MBA attendance decision), this speci cation ispreferred to Panel A because the rst stage regression is more precisely estimated. Columns 1 and2 show that, even in the broad and non-banker samples, there is now a strong and signi cant causaleff ect of starting on Wall Street on working there later. An MBA that goes to Wall Street uponleaving Stanford has a little over a 50% higher probability of working there at any given year laterin his career. The eff ect is somewhat stronger for the pre-MBA nance group. However, in thisspeci cation, the eff ect of initial job on later job is not signi cant for the pre-MBA Wall Streetsample.

Panel C repeats the analysis adding the Wharton placement instrument for rst jobs. Theseestimates are slightly larger and more precise than those that use only the stock market instruments.Though the coefficients are somewhat higher than the prior estimates, the economic conclusionsare unchanged. However, if Wharton placement is a valid instrument, the causal eff ect of initialinvestment bank employment on later investment bank employment is strongly signi cant for thesample as a whole and all subsamples. Finally, Panel D shows the results when using the classindicators as instruments. The coefficients and signi cance levels are very similar to the Panel Cresults.

Overall, Table 5 provides strong evidence that getting a job in investment banking has a strongcausal eff ect of working in investment banking later among the subset of MBAs that has alreadyshown an interest in a nance career. The eff ect for the rest of the class ranges anywhere from zeroto the same as for the pre-MBA investment bankers, depending on ones con dence in using M&Aactivity, Whartons placement, or the class indicator variables to instrument for taking an initialWall Street job.

While Table 5 makes it clear that initial placement in investment banking is sticky, the eff ectsin the table are averaged over all career years for the sample. To see how this eff ect varies overtime, I ran a series of IV regressions similar to those in column 1 of Table 5 with each regressionlimiting the sample to person/years a speci c number of years after graduation. Figure 5 graphsthe IV coefficient on the Initially I-Bank variable for each year using the speci cation in Column1 of Panel B of Table 5. For example, when running this regression, but limiting the sample

23


25/35

0

0.2

0.4

0.6

0.8

1

1.2

1.4

4 5 6 7 8 9 10 11

Years Since Graduation

2SLS Coefficient for "Initially I-bank"where sample is limited to those who

are a given number of years post-MBA

Dashed lines show bounds of 95% confidenceinterval for each year's coefficient

Figure 5: Year-by-Year IV Estimates of Eff ect of Initial I-Bank Placement on I-Bank Employment

to person/year observations as of the fourth January after a given person receives his MBA, thecoefficient is 0.811 and the 95% con dence interval for this coefficient spans from 0.630 to 0.992.

Given the nature of the sample, the sample size gets smaller as the number of years since graduationincreases. Therefore, the estimates get less precise moving to the right on the Figure. However, the gure makes clear that, for the rst decade or more after graduation, those who go from Stanfordto Wall Street are much more likely to work there than those who start in another industry. Theestimate drops off a bit after the rst few years and then stabilizes at about the level in regressionin the table. The coefficients for years twelve and higher are quite similar (all but one are at least0.35 and most are near one), but noisy.15

The overall message from Tables 3 and 5, as well as Figure 5, is clear. Stock market returns

while Stanford MBAs are in school have an important eff ect on whether or not many of them gofrom Stanford to a job on Wall Street. If they do go to Wall Street, even for random reasonsdriven by stock market conditions, they are much more likely to work there at any given point laterin their careers than if they do not go to Wall Street. While people move in and out of investment

15 Graphs similar to Figure 5, but based on regressions with Wharton placement or class indicators as instruments,generally have coefficients that are closer to one and precisely estimated.

24


26/35

banking after they enter the labor force, where they start matters a great deal. It appears thatthis eff ect is stronger for those most likely to be interested in working on Wall Street (that is, thosewho worked there before going to Stanford.)

5.2 Interpretation

The key empirical results so far can be summarized as follows. High stock returns while an MBAis in school have a sizeable eff ect on the likelihood that the MBA will go to Wall Street upongraduation. MBAs who start their career on Wall Street are more likely to work there later on.This relationship does not vary with the state of the market at graduation, so those who go toWall Street during bull markets are not less attached to Wall Street than those who go during bearmarkets. The relationship is causal, in that those who go to Wall Street right after graduation aremore likely to work there laterbecause they started their careers on Wall Street . The relationshipis particularly strong (or at least particularly precisely estimated) for those who worked in the nancial services industry before pursuing an MBA.

The combination of these results suggests that the pool of potential investment bankers ina typical Stanford MBA class is relatively homogeneous and that those who go to Wall Streetmake important nance-speci c investments. That is, the patterns in the data most closely matchModel 2 presented in Section 2. The data are consistent with a labor market where a largenumber of Stanford MBAs could be successful investment bankers, Wall Street rms demand morepeople when the stock market is doing well, and the wage diff erence between investment bankingand other jobs is a compensating diff erential that roughly off sets the unpleasant parts of being aninvestment banker. This would explain the ndings that the relationship between initially workingon Wall Street and working there later is not dependent on the state of the stock market whenMBAs graduate and that MBAs who go to Wall Street during bull markets are no less interestedor successful in nance-related MBA classes than those who go during bear markets. That is, Ifound no evidence that the lucrative off ers during bull markets attract those who are less able orless interested in investment banking to start their careers on Wall Street.

Given the causal relationship between initial Wall Street jobs and long-term Wall Street jobs,the patterns in the data also indicate that Stanford MBAs build up signi cant task-speci c humancapital (as in Gibbons and Waldman (2006)) while in school and very quickly after leaving school.That is, people who go to school during bull markets invest in nance classes at Stanford andin valuable on-the-job training shortly after graduation. This increases the wage diff erential theyenjoy in banking relative to other elds and may also lower some of the disutility of investmentbanking work. This model is supported by the informal analysis of classes discussed in Section 4.

25


27/35

It appears that students who were at Stanford during bull markets built nance-speci c humancapital that got them started in nance careers. The long-term career results in Section 5 suggestthat they continued building nance-related human capital upon starting work on Wall Street.

The fact that rst jobs are so closely related to macroeconomic conditions appears, at rstinspection, to also be consistent with a simple signaling story (Model 1 in Section 2.) In thatmodel, there are two types of people (bankers and non-bankers) that are interested in startingtheir careers in investment banking one because they are good at the work and one because theylike the money. Bear markets separate the two types during the job application process while bullmarkets force investment banks to sort between the two types after hiring them. This model wouldimply that stickiness in banking should be stronger for new MBAs hired in lean times, rather thanin good times. In an extreme form where there are only two types of MBAs, there would be nocausal eff ect of rst jobs in banking on later employment in that sector. It also implies that thoseMBAs who go to Wall Street when times are good should be less interested in nance, which mightshow up by their taking fewer nance classes at Stanford. Given that there is a causal link between rst jobs on Wall Street and later jobs there, that stickiness in banking is not related to conditionswhen MBAs are in school, and that the larger group that goes to Wall Street during bull marketstakes more nance classes, this model is not consistent with the data.

The ndings are also not consistent with a hybrid model (Model 3 in Section 2) whereinvestment banks value nance-speci c investments but also focus on MBAs who are most suitedto work on Wall Street. As I discussed above, such a model would lead to a causal link betweeninitial Wall Street jobs and later Wall Street employment. However, it would also lead to theaverage Wall Street hire during bull markets being less interested and attached to banking than theaverage bear market hire. This is simply not consistent with the ndings. So, while there is surelysome heterogeneity in a typical Stanford MBA class in how well they t in investment banking,there appears to be a sufficient supply of those who are able and interested in the eld to meet thebanks demand without compromising signi cantly in terms of t.

It appears that a diff erent form of sorting and hybrid model may explain the patterns in thedata. The data are consistent with the following slightly more speci c version of Model 2.Suppose there are two types of MBAs those who have an interest in nance and those whodo not. The nance-disposed group is largely indiff erent (given the wage diff erential) between nance and other opportunities when starting their careers. Given sufficiently strong beliefs aboutWall Street conditions, they will start their careers in nance and they will make nance-speci cinvestments while at Stanford and shortly thereafter. This group of homogeneous MBAs includesthose who worked in nance before getting MBAs and some unobservable subset of the rest of the

26


28/35

class. This variant of the model with homogeneous workers that make nance-speci c investmentsis consistent with all the ndings above, including the fact that all results are somewhat strongerfor those who worked in nance before getting MBAs.

6 Measuring The Financial Impact

6.1 What Would They Do if Not Investment Banking?

Stock market conditions at graduation lead to careers on Wall Street. But, in order to estimate the nancial rami cations of this eff ect, I need to make assumptions about what these people wouldhave done if they had not gone to work on Wall Street. I address that issue in this section bylooking at the eff ect of initial placement in investment banking on the probability of working inother industries in the longer term. To do this, I change equation (2) to

T it = t + X it + F 0i + it (3)

where T it is an indicator for whether the person works in some other industry in yeart . By thesame logic as in the last section, I cannot use OLS or a standard logit to estimate (3) becauseunobserved ability and taste factors are likely to be correlated with whether the person rst goesto Wall Street and whether or not he works in another industry later on. I therefore use the sameinstruments as in the last section.

Table 6 displays results whereT is an indicator for being an entrepreneur (that is, working ata rm that he founded) or working in the management consulting industry. I should note that theresults here need to be interpreted cautiously because these are the two outcome variables whereI found a relationship with initially being an investment banker. I found no such relationship forworking in high technology or working for a large manufacturing rm. One interpretation of this isthat the additional investment bankers were more likely to become entrepreneurs or consultantsthan to work in these other areas. But another possibility is that, by looking at a bunch of possibleoutcomes, some are likely to appear to have a signi cant relationship with initial investment banking

placement. While I therefore prefer to cautiously interpret the evidence here as only suggestive,choosing between these interpretations will be important when analyzing the eff ects of initial jobon income in the next section.

Columns 1 and 2 show that there is mixed evidence that initial jobs on Wall Street lead GSBgraduates to start fewer businesses. The coefficients is essentially zero when using the stock marketinstruments, but negative and signi cant when using the Wharton instrument. If the Wharton

27


29/35

Dependent Variable Founder Founder Consultant Consultant(1) (2) (3) (4)

Initially I-Bank 0.0478 -0.3531 -0.1795 -0.4565(0.2146) (0.1542) (0.2939) (0.1673)

Instruments S&P and M&A Wharton S&P and M&A WhartonTable 6: Industry of Longer-Term Job. All columns are results of two-stage least squares linearprobability regressions. Observations are based on a persons job as of the end of January atleast two and a half years after graduation from Stanford GSB. Initially I-Bank equals one if theperson was working in investment banking (including money management or venture capital) in theJanuary after graduation. Founder equals one if the person founded the company where he/sheworks at the time of the observation. Consult equals one if the person works for a managementconsulting rm. Instruments for Initially I-Bank are explained in Table 5. Sample sizes are thesame as in column 1 of Table 5 for each instrument. Standard errors (in parentheses) are adjustedfor any correlation within a graduating class.

variable is a valid instrument, it suggests that starting in investment banking lowers the probabilityof being an entrepreneur by about 35%. The consulting estimates (columns 3 and 4) are somewhatmore consistent across the two speci cations and the estimates using the Wharton instrument isagain negative and signi cant. Overall, Table 6 provides suggestive evidence that the Wall Streetcareers generated by stock returns while students attend Stanford come at the expense of careersas consultants and as entrepreneurs.

6.2 How Much Wealth is Transferred by Initial Conditions?I now turn to the question of how much money is involved in the random movement of MBAs in andout of investment banking careers. As mentioned above, the data are not perfect for this purpose.Because people only report beginning and ending salary for each job, I cannot directly estimate theeff ects on an individual of starting on Wall Street by tting wage regression equivalents of equation(2). However, because respondents provided income information as of the date of the survey, Ican use this cross-section to estimate wage pro les in investment banking and other elds over thecourse of MBA careers. I then discount these pro les over various career lengths to estimate the

lifetime labor income gained by those who become investment bankers.I rst break the sample into four sectors investment bankers, consultants, entrepreneurs (those

who founded a non-investment-banking business at which they work), and other. I also divide thesample into groups based on the number of years since graduation at the time of the survey. Com-bining these divisions, I create sector/years-since-graduation cells and calculate the average wage ineach cell. Each sectors average wages for years one through fteen after graduation are displayed in

28


30/35

0

200

400

600

800

1000

1200

1400

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15Years Since Graduation

L o g

W a g e

( n e w

" o t h e r

" g r a

d u a t e =

0 )

OtherInv-BankConsultantEntrepreneur

Figure 6: Career Wage Pro les by Type of Job

Figure 6.16 Two clear patterns emerge. First, investment bankers earn a substantial premium to theother groups throughout the rst fteen years after graduation. Second, management consultants

earn a premium over others, but not as much as investment bankers.The income diff erence that I want to calculate, however, is not the simple diff erence between

an investment banker at a given point in his career and someone at a similar point in a career inanother eld. I want to estimate the eff ect of starting in investment banking. Therefore, I calculatethe expected income in career yeart for a person who starts his career in investment banking as

E (wF 0 t ) = Pr( F t |F 0 )wF t + (1 Pr( F t |F 0 ))wGt (4)

where wF t

and wGt

are expected income in career yeart in investment banking and an alternative job, respectively, andPr( F t |F 0 ) is the probability the person will be in investment banking in yeart , conditional on starting in investment banking.Pr( F t |F 0 ) for a givent is the yearly coefficient

16 The average wage in each cell is estimated imprecisely because of the small numbers of people in some categories.To smooth out some of this imprecision, the graph actually plots, for each cell, the average salary for that cell, thecell from the same sector with one more year since graduation, and the cell from the same sector with one fewer yearsince graduation. For the rst year, I average the cell and the second year cell.

29


31/35

(1) (2) (3)Alternative Job Other Consult EntrepreneurWage Diff erence Estimates:

Year 1 Wage Diff . ($000) $97.5 $71.2 $114.9Year 1 Wage Diff . (%) 115.4% 64.3% 170.1%Year 7 Wage Diff . ($000) $504.0 $308.1 $321.6Year 7 Wage Diff . (%) 331.6% 88.6% 96.2%Year 15 Wage Diff . ($000) $937.2 $578.4 $1,014.5Year 15 Wage Diff . (%) 327.4% 89.7% 485.6%

Lifetime Income Diff erence Estimates (Discount Rate = 5%):10 Year Diff erence ($000) $1,867 $948 $1,54610 Year Diff erence (%) 150.8% 43.9% 99.2%20 Year Diff erence ($000) $3,818 $1,430 $3,11720 Year Diff erence (%) 152.8% 29.3% 97.4%

Lifetime Income Diff erence Estimates (Discount Rate = 10%):10 Year Diff erence ($000) $1,543 $839 $1,310

10 Year Diff erence (%) 158.0% 34.5% 108.3%20 Year Diff erence ($000) $2,564 $1,075 $2,14120 Year Diff erence (%) 157.2% 34.5% 104.3%

Table 7: Income Diff erences Between Investment Bankers and Others. All calculations are basedon salary averages for sector and years since graduation from the cross-section of 2,598 surveyrespondents. Investment banker wage estimates are adjusted for the likelihood that they will stillbe investment bankers at each year after graduation. See text for details.

from estimating equation (2), as displayed in Figure 5.

To summarize the underlying logic, I generate expected year-by-year income levels for MBAswho become management consultants, entrepreneurs, or any other non-investment-banker based onthe cross-section of wages as of the time of the survey. I generate expected year-by-year income levelsfor MBAs that start as investment bankers by taking the weighted average of the investment bankerincome and income in the other jobs. The weights for this calculation are based on the estimatedcausal eff ect of still being an investment banker in yeart if the person went into investment bankingright after getting his MBA at Stanford.

Based on this process for estimating experience/wage pro les in each of the four types of jobs,

Table 7 presents estimates of the pay diff erences between investment banking and other jobs atvarious points in a persons career. It also shows the cumulative present value of the incomediff erence between investment banking and other jobs over the rst ten and twenty years aftergraduation, assuming either a 5% or 10% discount rate on future income. While the wage diff erencebetween elds and over a career pro le will surely change over time, these estimates are what anew graduate might expect in 1996 looking forward.

The rst few rows of Table 7 show the individual year expected wage diff erences between some-

30


32/35

one who starts as an investment banker and someone who works continuously in one of the otherareas. These estimates vary from a 64% diff erence between investment bankers and consultantsright after graduation to cases where the investment banker can expect to earn three to six timeswhat people in the other groups can expect. For example, members of the other category earn anaverage of $286K after fteen years while someone who starts as an investment banker can expectto earn $1.2 million at that point. At this same point, investment bankers can expect to earn almostdouble the $645K earned by management consultants.

Discounting these gures over a ten or twenty year career leads to substantial absolute incomediff e

Documents

The Making of an Investment Banker Macro Economic Shocks Career Choice and Lifetime Earnings Updated 2009