Tag Archives: econometrics

Even More on Youth Co-residence

I have been trying to understand how to setup a panel on youth college financing, particularly a monthly panel as the yearly panel cannot give information on youth co-residence, at least I don’t think it can. Also, because the results were rubbish it is important to get the work right.

I ran into a job market paper (chapter 3) discussing almost exactly the same research but instead controls for different variables. I take into consideration family aid, grants and GPA.

I am still unsure about the estimation design using NLYS97. In particular, in Kaplan (2012) the discussion on sample design notes that co-residency question ends in 2002, at which point none of the cohorts from the survey were eligible to enter college. To replace this between 2003 and 2011 the respondents give the date of their first time moving out and first time they moved back with their parents. All these spells had to be at least three months, as noted by the question given to respondents. However, once I used these variables and followed a person for 1 year from their graduation I got incredibly imprecise results. I got no conclusions. Partly I think it’s due to the fact the once you restrict the data to a subset of college graduates you do end up with a lot of skipped responses (for valid reasons) coupled with plain missing values. Now with a better understanding of the survey I can formulate a proper panel, yearly or monthly depending on what I can actually use. It is a little difficult dealing with semester or term based variables (which are my main concern). I have to make assumption about the type of schools to make the work doable and efficient.

The job market paper is interesting. Isolating to just the effects of parental co-residence and student loans, one sees similar results to the papers discussed in the last post. What this says about using parents to smoothing consumption choices isn’t clear. Part of that is a discussion of expectations and human capital. Though the empirical evidence can be seen as an explanation of debt aversiveness or response to new budget constraints, which aren’t totally mutually exclusive from the argument of smoothing yet.


More on Parental Co-residence and Student Loan Debt

In my last blog post I had some doubts over the popular economic punditry’s claim that more young adults living at home is due to the increasing reliance on student loans. There are large scale changes in marital status and demographics that also important. I do think that reframing the question to whether student debt causes parental co-residence is valid. In fact, I think parental co-residence is part of the consumption smoothing used by college graduates who face borrowing constraints. Outside standard life-cycle model, I think credit constraints and debt averse behaviour are likely contributing factors.

There are some interesting papers on the subject. Looney and Yannelis (2015) at the Brookings Institute investigate the characteristics that drive default rates. In their analysis of course is the discussion on the burden 2-year community colleges and for-profit colleges share for the increased default rate since 2000. This isn’t really on topic but in Avery and Turner (2012) and a quick look through the Department of Education’s student loan data for default rates shows that the schools classified as for-profit tend to have exceptionally high default rates. Most statistical evidence shows public and private 4 year universities have a substantially smaller default rate. While the popular discussion ignores the difference amongst schools, it is very important for any future policy goal to understand difference amongst institutions. As an investment, attending a 4 year program (as discussed in the last post) remains a “good” deal (leaving this vague due to heterogeneity amongst students).

Back to the question at had, do student loans (or rather the debt) cause parental co-residence post graduation? Bleemer et al. (2015) find evidence of causality using the Federal Reserve Bank of New York’s Consumer Credit Panel. They use the panel and local factors such as housing price, employment rate and student debt in local college. Another paper by Dettling and Hsu (2016) is similar but for the consideration of credit risk and payment delinquency as part of the inquiry in place of considering geographical factors for individuals after a period of independent living. Dettling and Hsu are really interested in the consumption smoothing of young adults post-graduation. Dettling and Hsu find that change in debt portfolios of young adults contributed 32 percent of increases in co-residence with parents between 2005 and 2013.

This leads to me to question my results from my last post. There will be an addendum to them. I believe I already mentioned they were in hast. It’s a good time to remind any readers, that if anything, this blog is me thinking out loud. Nonetheless, the information above, particularly Bleemer et al (2015), is very important and I recommend reading it. On economic situations and living arrangements I recommend Kaplan (2012).



Living Arrangments and Student Loans

The discussion surrounding student loans has always been focused on a supposed bubble. College students with debt in the U.S. are thought to be borrowing too much. The debt in popular media is characterized as extremely harmful to their living standards after they graduate. As well, college education is itself questioned when the heavy debt burdens are considered. Large-scale demographic changes are also linked to student loans at times.

Living arrangements broadly speaking are of concern. The number of adults who live with their parents has steadily increased for 50 years (figure 1). With increases in total student loan debt for college-aged individuals growing as well, many link the two to the trend in figure 1.


For young adults, there is also a similar trend (figure 2). For what we assume to be surely out of college population, there is an upward change in the number of young adults that co-reside with parents. For the in-college cohort there is less significant amount of variation (figure 3).

There could be many reasons for these living arrangements. This could represent the heavy burden of student loans. Students may not understand the true costs of attending college. The may in fact borrow too much, at least in comparison to the future earnings their major brings. Not only are they borrowing too much but they could also be substantially more debt averse than believed to be. The immediate impact of a heavy debt burden may influence their cohabitation preferences if they want to pay off the loans quickly.

There is some evidence to the debt averse behaviours of college students, one that doesn’t reflect life-cycle model predictions of smoothing. In Rothstein and Rouse (2011)  empirical investigation into an anonymous university’s “no-loans” policy revealed a natural experiment which shows that students do not follow standard model assumptions of consumption smoothing. Regular model assumptions would mean that student loans would impact career and major choices in a very limited manner. By looking at career choices in “public-interest” jobs and gifts to the university, they find the prevalence of student loan debt to have a negative effect on both. To counter their large debt burden, students chose jobs in finance amongst others. Moreover, Rothstein and Rouse argue that credit constraints are the main culprit behind the break down of the lifecycle model due to the effect of debt on defaults in pledges of alumni.

This brings into question whether or not all students have it in their best interest to take up the costs of college. For a while now giving college a try was considered to be a good investment for anyone. However, if college tends to just push students towards well paying jobs in the face of surprising debt tolls then maybe college is harmful to those who cannot make it into finance or other well-paying “knowledge” private sector jobs.

Avery,Turner 175

Avery and Turner (2012) produced the figure above assuming returns to college begin four years after entrance into a college. The opportunity cost in this case is very high. Of course, there is uncertainty these gains. Some, due to heterogeneity among students, earn a greater return to their education. Even among institutions, wages and graduation rates differ tremendously. Avery and Turner note the following:

If individuals can make accurate predictions about whether they will complete college individuals can make accurate predictions about whether they will complete college and what they would earn conditional on attaining a college degree, then most and what they would earn conditional on attaining a college degree, then most of the variation in lifetime earnings outcomes can be attributed to heterogeneity

In fact, Avery and Turner in a footnote reference Chen (2008) that attributes heterogeneity to the wage inequality amongst college graduates. So, one may say that the difference amongst graduates (and for that matter those who actually graduate or drop out) to be determined by aptitude in college. Thus, Avery and Turner conclude that there is no “student loan bubble” due to the gains possible from attending college (with 2-year community colleges and for-profit institutions being the exception) for those who graduate in good standing.

As part of a quick investigation, I looked through the National Longitudinal Survey of Youth from 1997 from around 2003 to 2010. Looking at college financing and living arrangements. The estimated the following latent model:

Screen Shot 2016-06-11 at 3.28.43 PM

By panel random effects logit, for R > 0. The time variant variables are owed college loans, total income, college grants, college family aid and the dependent variable for residence with parents. The time constant variables are race, phd, masters, professional degree and female. From this I found no significant effect, part of that might due to programming error since I am observing an unbalanced panel with limited information on durations.

Demographic choices are far more significant, and even then they aren’t very precise. I did have some difficulty navigating the NLS database to find some way to noting if an individual lived with their parents. I used the date they moved back in with parents but in the dataset there was some issues as it was unclear how long they moved back in and when they moved out. Many respondents validly and invalidly skipped the query. I do not think this result is well identified but my hunch is better data cleaning and choices would not change much. Neither would controlling for macroeconomic effects and marital status.

On living arrangements Pew Research Centre‘s investigation by Richard Fry (2016) points to demographic changes as the leading cause of increased parental residency. I find in general that result to be very complementing to the debt averseness story. There is more to the dynamics of human capital investments than is commonly believed in other discussions. Here, one can see that debt averseness can curb the assumed consumption smoothing but this is masked as young adults unable to get a start in the world. In fact, as has already been discussed, a college education has great returns on investment. For the most part, one must attend college to get a start in the economy.

Students do not enter college knowing exactly what their aims are or have perfect information about their chances at success. This may be the market failure in student loan market, students may not have perfect information. Nonetheless, college education as investment can be viewed as an option to some degree, thus the assumed relationship between living arrangements and student debt implicates student behaviours much more than the life cycle model. It is unlikely that students are so debt averse and ignorant of information they receive while attending college that they are saddled to living with their parents in droves, so much that they are responsible for the trend in the first figures from the Current Population Survey.

This is a very interesting subject. I will be revisiting it as I acclimate myself with NLSY97 dataset. Specifically, I think one could find better data or more information on specifications of the survey’s questions.


Stata data file here

CPS data used here , graphs here and here.


Update 2