Tackle Big Questions. Use Big Data

 What inspired you to dig deep into the fields of inequality and different risk factors in household workers and firms?

That's a big question. Let me just think for a moment… … Some of it started with my interest in macroeconomics, and, in particular, in how financial markets interact with the macro economy. The first thing that you come across when you study financial markets, is the wealth inequality. There’s a very small fraction of the population that owns a substantial share of the stock market. For example, 1 percent of the population in the U.S. holds about half of all stocks. The traditional models on asset price dynamics were based on the representative-agent construct. You are basically assuming that the aggregate economy functions as if it's one person— whereas in reality, this 1 percent might really be very different than the remaining 99 percent. And they are different along certain lines. In my thesis, I argued that to understand certain financial markets, and how they interact with the macro economy, we need to account for the fact that there is immense heterogeneity and the people who are at different parts of distribution are also different along key dimensions.

And so it went from there, because you realize that for some questions from macro the issue of wealth inequality is very important. Another literature says that for many questions it's not that important. Then, I started thinking about particular questions where heterogeneity is first-order, like a key differentiator, which brought me into income- and consumption inequality because— at the end of the day— we don't consume our wealth but rather our consumption. This was the entry point of understanding macro but then realizing that heterogeneity inequality for some questions is really important.

This year, your lectures are focused on the latest insights from big data inequality and risk. Could you give us a more specific overview on two issues: first, how big are these data versus these small ones that we're talking about, and second, what conclusions have changed substantially that basically alter our view about inequality and risk?

Big data is very big. For the questions I studied, income risk and income inequality, we went from survey data sets with a few thousand households in panel data to data sets where it's in tens of millions— and in case of some countries it's the entire population. So you have data on everyone, and you follow them for very long periods of time.

In terms of what insights have changed, the key conclusions that have changed will be discussed during all three days of the lectures. One conclusion has to do with whether or not there is more income uncertainty in the world today compared to the 1970s. The previous conclusion— the conventional wisdom, almost unquestioned in the literature— was that there is a lot more uncertainty today, based on analyses of survey data. In most of the U.S. administrative data we have found the opposite: there is less and less income volatility, incomes are less unstable. This is a big change because uncertainty is an input into many other questions we analyze. When we want to understand government policy, when we want to understand how the world changed, oftentimes, as an input, we said, “Oh, there's more uncertainty today, so people behave differently. There's more uncertainty today, so the policy should be this way”. And now, if it did not go up, it went down, we have to re-evaluate a number of the conclusions that we have reached.

Another insight has to do with the nature of income risk. We had small data sets, and a lot of our empirical analyses were extremely restrictive: we used parsimonious parametric econometric models combined with small data that had their own limitations: the sample wasn’t always representative, there were measurement errors and so forth. This combination led us to many conclusions that turned out to be very different from what we now think. One of these is a simplification that assumes that income shocks are Gaussian. It sounds like a technical issue but it turns out to matter for many quantitative conclusions that we draw. We found that actually many income shocks are very different from Gaussian— in fact, they are not even close to being Gaussian. We thought that the income shocks become more dispersed in recessions, which was, again, a conclusion from a parametric model. Now we find that dispersion— or the variance— hardly changes at all over the business cycle. But, another aspect of risk changes, which is the downside risk, in recession: when people have a negative income change, it is much bigger than when it's a positive income change, whereas the average change is stable over the business cycle. It's different than overall income shocks becoming more variable. It's that one-sided thing that's actually becoming much worse.

Given these changes in inequality and risk because of these new insights, and taking into account the rapid rise of AI and robotization of jobs, what do you think about the income shock to the life cycle of the low-income group in the population? What kind of policies could governments adopt, for instance, in the context of European Union?

That's a good question. This area is highly important, and we need to understand it much better. At the moment, the research in this area is in its infancy. We have this old traditional line of thinking that since the college premium, or education premium, is rising, we should educate people more. There is a huge shortcoming to this argument: the college premium is a statement about the average wage of college workers relative to the average wage of high school workers. However, there’s a lot of dispersion in wages within each group. So, for example, suppose we look at what fraction of college graduates don't earn an income in the top quartile of the high school income distribution. That is, you go to college, and after four years you graduate, and the job that you get doesn't pay what you would have gotten or what people get in the top quartile of the high school distribution. That fraction is very high: about 40-45 percent. So, in countries like the U.S., you spend four years paying a huge tuition, and after all of your effort you’re not even making what the top high school graduates make. What is really interesting is that, even though the college premium went through the roof in the last fifty years, this fraction has remained almost constant. So the chance that you won't make it is still the same. How is that possible? Well, the averages are diverging but within each group, there is more inequality in wages. In that sense, simplistically saying we should educate people is not the right answer, because there are too many people who go to college and don't have college graduate jobs; they still do the jobs that high school graduates do. I think it should be a lot more based on specific skills and requires more focus on the curriculum. What do you teach these workers? The system should also be more flexible because the world is changing. I am thinking about continuing education: perhaps instead of teaching very specific skills in college, we should focus more on teaching students how to learn new skills. In that way, down the road, when some of their skills become obsolete, being taken over by robots or trade, they can “re-tool”. Therefore, the government has a responsibility to provide these people the opportunities to re-train. I think it's a clear question with a pretty complicated answer. There is no silver bullet with which you can solve the problem, yet I want to argue against the point that you hear rather often: that once you go to college you will be fine.

This also has to do with the kinds of majors people choose in college. There’s been a massive shift towards STEM majors, and more funding is being invested in these majors than the social science majors. What is your opinion about this?

For a lot of students, especially in the U.S., there is a general problem of going into majors with high unemployment rates. These majors have low employment rates and low wages, and students don't realize that this is the case. English is an example, as well as many other Humanities disciplines. When you talk to students who graduate from these majors, you discover that they had no idea about this, they actually thought English was a good major in terms of employment. Yet, there are few jobs that require an English major. In that sense, teaching them some broad skills or cognitive ability would be more useful. Some recent research by Altonji, Kahn and Speer (AER, 2014) looks at wage inequality among college graduates that has increased substantially; they show that, by some measures, inequality among college graduates increased as much as the increase in the overall population. About two-thirds of that increase within college is actually across majors! In other words, some majors make a lot more than other majors. Sometimes it's difficult to predict. For example, among doctors, while neurologists were some of the highest paid specialists in the ’70s, they are now among the lowest paid. Dermatologists used to be some of the lowest paid, and now they are among the highest paid. So when Altonji, Kahn and Speer look at what explains the growing gap across majors, it is again going back to basic skills. The wages of majors that require a lot of cognitive ability have gone up. Fundamentally, it's like a return on the harder skills, the more complex tasks that cannot be automated. As for STEM, my only concern is that for some people it’s too difficult, so they may just give up. Other than that, I think it is useful to teach solid cognitive skills because they're widely applicable in every field.

There is an interesting related fact. Take the human capital theory, which was a field that won several Nobel prizes, in terms of being an important determinant of wages, earnings and so on. The paper that was perhaps the first in that field was written in the 1950s by Theodore Schultz, and was inspired by his conversations with farmers. He had realized that, even in farming technology, adoption depended a great deal on the education of the farmer. The more educated the farmer was, the more quickly they adopted the technology, became more productive and thereby became richer. So Schultz thought, if this is something that matters even for something like agriculture, it should matter for everything else.

In that sense, cognitive skills are important across the board. I don't know if you should go to STEM for that, or expand the set of classes that emphasize analytical ability and related hard skills.

Speaking about cognitive ability, here’s a question related to intergenerational mobility. A recent genoeconomic paper by Papageorge & Thom (NBER, 2018) uses polygenic scores to predict educational attainment. Essentially, they conclude that it is better to be born rich than gifted: on average, a very gifted kid born into a low-income family graduates from college at a much lower rate compared to a kid with average ability from a high-income family. What is your take on this?

College graduation is one measure. Another thing that others also have explored in their own research, is that for the ability to earn a good living, soft skills are just as important as hard skills. We just spoke about cognitive ability. There are also characteristics such as self-confidence, trust in other people, self-esteem, positivity— and there are measures of these things. When you run a wage regression, in addition to your test scores in math, in verbal and all the hard skills, social skills turn out to be really important. So where income enters, or wealth enters, is that it might help you develop those skills even by observing, for instance, your parents and other people you interact with, which may serve you well in the long run.

As a side anecdote, a paper by Athey, Katz, Krueger, Levitt and Poterba (AER P&P, 2007) looked at PhD students in economics in the top five departments. The paper is written by five professors, one from each of those departments. They looked at all the information that the departments had when the students applied for a PhD. So they knew how they ranked their candidates before they came in (they had the students’ grades and the qualifying exam scores) and then they looked at the placement of these students. They ask what qualities or attributes predict placement success. They find a lot of obvious stuff, such as having a high GRE score or high grades. One factor that they couldn’t explain, which stood out, is this: if you had a top liberal arts college education in your undergrad, controlling for everything else, you did better in the job market. And they speculated that, in a way, a liberal arts college education is geared toward the individual: it gives you confidence, the ability to interact with others, to carry yourself. These skills are valuable in the labor market. In that sense, I agree it is important— but still, you cannot discount cognitive ability.

A paper of yours looks at the gender gap in income among the top earners (Guvenen, Kaplan and Song, NBER, 2014). Would you say that some of the factors that we just discussed also play into this gap?

I think they do. A much harder question is whether we can quantify it, or how much each contributes. Climbing to the top is like climbing the peak of a mountain. The closer to the top the more narrow it gets, the more competition there is. Sometimes you put somebody down, and everybody wants to get to the top. Certain personality traits actually matter more. Muriel Niederle at Stanford carried out a number of experiments with men and women, in which they play games and have to compete.[1] A common difference that she documents is competitiveness. I don’t think it is good or bad, it is just a personality trait, which also differs hugely among men, and also among women. At the very top those things matter more: how much you are willing to fight, or to be more precise, how much you are willing to give up in the rest of your life to actually get to the very top. For example, when we think of a promotion and higher wages, we often forget that, at the same time those higher level jobs often come with a lot more stress and a lot more responsibility. While you may not necessarily work more hours, the buck stops at you, so if there is a screw-up then you are the manager in charge— and not everybody wants that.

There is one piece of research about the Chicago Booth MBA graduates,[2] and there is another paper that looks at physicians.[3] If we look first at the jobs that MBA graduates at Chicago Booth do, the starting salary for men and women is pretty much the same. If anything, women may have slightly higher earnings around age 25. Later, in their late 20s or early 30s, when women want to have kids and take time off to do so, the nature of the jobs is such that it's very hard to go back in: you lose your network, lose your connections. Since the hours are extremely long, it is almost impossible for a woman to have kids and work part-time. So, for example, 40 hours in law or finance is like part time. Now, even at a young age, people literally can be working 70, 80 or 90 hours. Contrast this with medicine, where the profession has adapted itself, especially for pediatricians and gynecologists. There are no individual practices. In the U.S., if you have a pediatrician and your kid gets sick, you don't just call the pediatrician and go to see her. You go to the practice, where there are five or six doctors, and you see whoever is available at the time. The authors of the paper document that this change makes it possible for women to work fewer hours while raising kids and still stay attached; then, when their kids reach a certain age, they can return to full time employment at the same level. I think there will need to be some changes made in the production function of occupations to adapt to a world in which women, especially at child-bearing ages, need more time— and that need for more time doesn't have a permanent effect on their life. 

It is interesting to examine the life cycle income trajectories of men and women. I'm not aware, actually, of any paper that has done this before, so we looked at this recently. For men it is a very hump-shaped trajectory, meaning that incomes of men grow rapidly in their 20s and 30s, and the trajectory reaches its peak early. For women, it is almost like a straight line that grows much more slowly early on. This slow start is due partly to the time-out- and the time-off issues I mentioned. But women catch up almost completely by age 55. That is, the total increase in wages from age 25 to 55 in the U.S. is very similar for men and women, but the shapes are very different. (The levels are different of course).  If you look at employment, you actually see the same thing: women work more than men in their 40s and 50s (conditional on working similar total hours). So I don't think we have really nailed it down about how much each of these factors contributes. Cognitive ability is something very similar, by every measure, but there are other aspects that also feed into this.

What sort of policies from governments and companies could help to address these inequalities?

In general, not just about gender, we should be quite careful in designing policies to deal with inequality. Sometimes they can backfire. If we don't really know what we are doing, it may turn out that the group that we really want to help is actually the one that is hurt the most. There are many examples of this. In a somewhat different context, a well-known example is the Ban the Box regulation. In the U.S., some states passed a law that prohibited employers asking job applicants whether they had a criminal record. Their reasoning was along the lines of, ‘somebody made a mistake, they went to jail for three months but you do not want that history to be with you all your life.’ They thought employers were going to discriminate against these applicants, so they banned this question. This law was drafted with some consideration given to African-American males because of their high incarceration rate. And what you find, actually, is that when employers cannot tell who has a criminal record, they start to spread such probability more evenly to all applicants. That actually hurt qualified applicants who did not have a criminal record, so it did not help the very people it aimed to help.

Another related example: in Norway, they passed a law saying that in every company greater than a certain size, 30 percent of the board need to be female. But such changes are sometimes very sudden. When you pass this law, are there enough women that are close to that threshold? Otherwise, you are essentially trying to grow that plant by pulling the leaves. It's much more important to grow them from the bottom, encouraging them a lot more at the college level, making those choices more attractive to them, the first job. It takes longer, but then you literally grow everything from the bottom up rather than imposing from the top, as it did in the case of Norway. The studies show that those companies had lower performance. And then people say, ‘oh, let's not do any of that’. Perhaps you want to go a bit more slowly and solve it in a more permanent way.

Competition matters for some of this because even in academia it's taking time. And a good question is why the MBA finance-type professions haven't adapted when doctors did. In general, I don't have a good specific policy on this; with regard to any inequality policy, my sense is that you have to be extremely careful to avoid the measure backfiring on exactly on that group you wanted to help. I hope somebody comes up with something good to implement. Maybe you will.

So, to close, as a very successful and influential researcher, what advice would you give to the PhD students who are looking into the exciting research areas of inequality and risk?

I always believe in challenging every assumption we have. Sometimes, when we are young we are a bit too deferential. We think, oh, all these people are so amazing, they have shown all of this, how can they be wrong? Well, we are all human beings, and progress is made one step at a time.

The best research is done by taking a second look at what people have found or what they believed in. And when it comes to inequality, it's a lot more complicated. That's actually one frustration I have with the general academic and non-academic debate. What I'm trying to teach here in class is that distributions are incredibly complex objects— consider 300 million people in America, 300 million points on that density, which means that statements like ‘inequality increases or decreases …’ are too simplistic. There are many types of inequality: the gender gap, the racial gap, the seniority gap. There is also the gap between workers at different firms, across different locations, the gap between top 1 percent and the rest, and the gap between the bottom 10 percent. These different measures often go in different directions. Now as we have the data, we need to dig one level deeper especially into big data. I would question every assumption, everything that we think we know. Grad students and young economists have the advantage of not being encumbered with old ideas. If you ask people who have done really influential work, many say that they did what they did because they didn't know that everybody thought it couldn't be done. Because they didn't know that, they just tried and sometimes it worked. That's the greatest advantage young people have. So I would just stick to that: be brave, be daring, work hard, and good things happen.

 In this case, ignorance is bliss.

In a way, yes. If you are not afraid, sometimes you open a door, and there is treasure there.

Fatih Guvenen is the Curtis L. Carlson professor of Economics at the University of Minnesota, United States, and a research associate in the NBER’s Economic Fluctuations and Growth Program.

 Notes

[1] See, for instance, Gneezy, Niederle, Rustichini, QJE, 2003.

[2] Bertrand, Goldin and Katz, AEJ: Applied Economics, 2010.

[3] Goldin and Katz, JLE, 2016.

Relevant papers throughout the talk, in order of mention:

 Altonji, Joseph G., Lisa B. Kahn and Jamin D. Speer. "Trends in earnings differentials across college majors and the changing task composition of jobs." American Economic Review 104.5: 387-93, 2014.

 Schultz, Theodore W. “Reflections on Agricultural Production, Output and Supply”. Journal of Farm Economics. 38 (3):748-762, 1956.

 Papageorge, Nicholas W. and Kevin Thom. Genes, education, and labor market outcomes: evidence from the health and retirement study. No. w25114. National Bureau of Economic Research, 2018.

 Athey, Susan, Lawrence F. Katz, Alan B. Krueger, Steven Levitt and James Poterba. “What Does Performance in Graduate School Predict? Graduate Economics Education and Student Outcomes.” American Economic Review Papers and Proceedings. 97 (2):512-18, 2007.

 Guvenen, Fatih, Greg Kaplan and Jae Song. The glass ceiling and the paper floor: Gender differences among top earners, 1981-2012. No. w20560. National Bureau of Economic Research, 2014.

 Gneezy, Uri, Muriel Niederle and Aldo Rustichini, “Performance in Competitive Environments: Gender Differences”, Quarterly Journal of Economics, CXVIII, 1049 – 1074, August 2003.

 Bertrand, Marianne, Claudia Goldin and Lawrence F. Katz. "Dynamics of the gender gap for young professionals in the financial and corporate sectors." American Economic Journal: Applied Economics 2.3: 228-55, 2010.

 Goldin, Claudia and Lawrence F. Katz. "A most egalitarian profession: pharmacy and the evolution of a family-friendly occupation." Journal of Labor Economics 34.3: 705-746, 2016.