Close

This page had been redirected to a new URL, please update any bookmarks.

Font Size: A A A

macroblog

« March 2014 | Main | May 2014 »

April 28, 2014

New Data Sources: A Conversation with Google's Hal Varian

New Data Sources: A Conversation with Google's Hal Varian

In recent years, there has been an explosion of new data coming from places like Google, Facebook, and Twitter. Economists and central bankers have begun to realize that these data may provide valuable insights into the economy that inform and improve the decisions made by policy makers.

Photo of Hal VarianAs chief economist at Google and emeritus professor at UC Berkeley, Hal Varian is uniquely qualified to discuss the issues surrounding these new data sources. Last week he was kind enough to take some time out of his schedule to answer a few questions about these data, the benefits of using them, and their limitations.

Mark Curtis: You've argued that new data sources from Google can improve our ability to "nowcast." Can you describe what this means and how the exorbitant amount of data that Google collects can be used to better understand the present?
Hal Varian: The simplest definition of "nowcasting" is "contemporaneous forecasting," though I do agree with David Hendry that this definition is probably too simple. Over the past decade or so, firms have spent billions of dollars to set up real-time data warehouses that track business metrics on a daily level. These metrics could include retail sales (like Wal-Mart and Target), package delivery (UPS and FedEx), credit card expenditure (MasterCard's SpendingPulse), employment (Intuit's small business employment index), and many other economically relevant measures. We have worked primarily with Google data, because it's what we have available, but there are lots of other sources.

Curtis: The ability to "nowcast" is also crucially important to the Fed. In his December press conference, former Fed Chairman Ben Bernanke stated that the Fed may have been slow to acknowledge the crisis in part due to deficient real-time information. Do you believe that new data sources such as Google search data might be able to improve the Fed's understanding of where the economy is and where it is going?
Varian: Yes, I think that this is definitely a possibility. The real-time data sources mentioned above are a good starting point. Google data seems to be helpful in getting real-time estimates of initial claims for unemployment benefits, housing sales, and loan modification, among other things.

Curtis: Janet Yellen stated in her first press conference as Fed Chair that the Fed should use other labor market indicators beyond the unemployment rate when measuring the health of labor markets. (The Atlanta Fed publishes a labor market spider chart incorporating a variety of indicators.) Are there particular indicators that Google produces that could be useful in this regard?
Varian: Absolutely. Queries related to job search seem to be indicative of labor market activity. Interestingly, queries having to do with killing time also seem to be correlated with unemployment measures!

Curtis: What are the downsides or potential pitfalls of using these types of new data sources?
Varian: First, the real measures—like credit card spending—are probably more indicative of actual outcomes than search data. Search is about intention, and spending is about transactions. Second, there can be feedback from news media and the like that may distort the intention measures. A headline story about a jump in unemployment can stimulate a lot of "unemployment rate" searches, so you have to be careful about how you interpret the data. Third, we've only had one recession since Google has been available, and it was pretty clearly a financially driven recession. But there are other kinds of recessions having to do with supply shocks, like energy prices, or monetary policy, as in the early 1980s. So we need to be careful about generalizing too broadly from this one example.

Curtis: Given the predominance of new data coming from Google, Twitter, and Facebook, do you think that this will limit, or even make obsolete, the role of traditional government statistical agencies such as Census Bureau and the Bureau of Labor Statistics in the future? If not, do you believe there is the potential for collaboration between these agencies and companies such as Google?
Varian: The government statistical agencies are the gold standard for data collection. It is likely that real-time data can be helpful in providing leading indicators for the standard metrics, and supplementing them in various ways, but I think it is highly unlikely that they will replace them. I hope that the private and public sector can work together in fruitful ways to exploit new sources of real-time data in ways that are mutually beneficial.

Curtis: A few years ago, former Fed Chairman Bernanke challenged researchers when he said, "Do we need new measures of expectations or new surveys? Information on the price expectations of businesses—who are, after all, the price setters in the first instance—as well as information on nominal wage expectations is particularly scarce." Do data from Google have the potential to fill this need?
Varian: We have a new product called Google Consumer Surveys that can be used to survey a broad audience of consumers. We don't have ways to go after specific audiences such as business managers or workers looking for jobs. But I wouldn't rule that out in the future.

Curtis: MIT recently introduced a big-data measure of inflation called the Billion Prices Project. Can you see a big future in big data as a measure of inflation?
Varian: Yes, I think so. I know there are also projects looking at supermarket scanner data and the like. One difficulty with online data is that it leaves out gasoline, electricity, housing, large consumer durables, and other categories of consumption. On the other hand, it is quite good for discretionary consumer spending. So I think that online price surveys will enable inexpensive ways to gather certain sorts of price data, but it certainly won't replace existing methods.

By Mark Curtis, a visiting scholar in the Atlanta Fed's research department


April 28, 2014 in Economics, Forecasts, Technology, Web/Tech | Permalink

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341c834f53ef01a3fcfb87b0970b

Listed below are links to blogs that reference New Data Sources: A Conversation with Google's Hal Varian:

Comments

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

April 17, 2014

Using State-Level Data to Estimate How Labor Market Slack Affects Wages

At a recent speech in Miami, Atlanta Fed President Dennis Lockhart had this to say:

Wage growth by most measures has been very low. I take this as a signal of labor market weakness, and in turn a signal of a lack of significant upward unit cost pressure on inflation.

This macroblog post examines whether the data support this assertion (answer: yes) and whether wage inflation is more sensitive to some measures of labor underutilization than other measures (answer: apparently, yes). San Francisco Fed President John Williams touched on the latter topic in a recent speech (emphasis mine):

We generally look at the overall unemployment rate as a good yardstick of labor market slack and inflation pressures. However, its usefulness may be compromised today by the extraordinary number of long-term unemployed—defined as those out of the workforce for six months or longer... Standard models of inflation typically do not distinguish between the short- and long-term unemployed, because they're assumed to affect wage and price inflation in the same way. However, recent research suggests that the level of long-term unemployment may not influence inflation pressures to the same degree as short-term unemployment.

And Fed Chair Janet Yellen said this at her March 19 press conference:

With respect to the issue of short-term unemployment and its being more relevant for inflation and a better measure of the labor market, I've seen research along those lines. I think it would be tremendously premature to adopt any notion that says that that is an accurate read on either how inflation is determined or what constitutes slack in the labor market.

The research to which President Williams refers are papers by economists Robert Gordon and Mark Watson, respectively. (For further evidence, see this draft by Princeton economists Alan Krueger, Judd Cramer and David Cho.)

The analysis here builds on this research by broadening the measures of labor underutilization beyond the short-term and long-term unemployment rates that add up to the standard unemployment rate called U-3. The U-5 underutilization measure includes both conventional unemployment and "marginally attached workers" who are not in the labor force but who want a job and have actively looked in the past year. The difference between U-5 and U-3 is a very close proxy for the number of marginally attached relative to the size of the labor force.

U-6 encompasses U-5 as well as those who work less than 35 hours for an economic reason. The difference between U-6 and U-5 is a very close proxy for the share of "part-time for economic reason" workers in the labor force. These nonoverlapping measures of labor underutilization rates are all shown in the chart below.


The series are highly correlated, making it difficult to isolate the impact of any particular labor underutilization rate on wage inflation (e.g., "How much will wage inflation change if the short-term unemployment rate rises 1.0 percentage point, holding all of the underutilization measures in the above figure constant?").

We follow the approach of Staiger, Stock, and Watson (2001) by using state-level data to relate wage inflation to unemployment in a so-called "wage-Phillips curve." Because the 2007–09 recession hit some states harder than others, we can use the cross-sectional variation in outcomes across states to arrive at more precise estimates of the separate impacts of the labor underutilization measures on wage inflation (see the chart).


Five-year state-level wage inflation rates for 2008–13, using monthly Current Population Survey(CPS) microdata, are shown on the vertical axis. The CPS microdata are also used to construct all of the labor underutilization measures. Each circle represents an individual state (red for long-term unemployment and blue for short-term unemployment), and each circle's area is proportional to the state's population share. Two noteworthy states are pointed out for illustration. North Dakota has had lower unemployment and (much) higher wage inflation than the other states (presumably because of its energy boom). And California has had higher unemployment and (somewhat) lower wage inflation than average. Even after excluding North Dakota, we see a clear negative relationship between wage inflation and underutilization measured with either short-term or long-term unemployment.

Because short-term and long-term unemployment are highly correlated (also apparent in the above plot), one can't tell visually if one underutilization measure is more important for wage inflation than the other. To make this assessment, we need to estimate a regression. The regression—which also includes both U-5 minus U-3 and U-6 minus U-5—adjusts wages for changes in the composition of the workforce. This composition adjustment, also made by Staiger, Stock and Watson (2001), controls for the fact that lower-skilled workers tend to be laid off at a disproportionately higher rate during recessions, thereby putting upward pressure on wages. The regression also weights observations by population shares.

The regression estimates imply that short-term unemployment is the most important determinant of wage inflation while U-6 minus U-5—the proxy for "part-time for economic reason" workers—also has a statistically significant impact. The other two labor underutilization measures do not affect wage inflation statistically different from zero. Rather than provide regression coefficients, we decompose observed U.S. wage inflation for 1995–2013 into contributions from the labor underutilization measures, workforce composition changes, and everything else (see the chart).


Both short-term unemployment and workers who are part-time for economic reasons have pushed down wage inflation. But the "part-time for economic reason" impact has become relatively more important recently because of the stubbornly slow decline in undesired part-time employment.

Photo of Patrick HigginsBy Pat Higgins, a senior economist in the Atlanta Fed's research department


April 17, 2014 in Employment, Labor Markets, Unemployment | Permalink

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341c834f53ef01a73dad2442970d

Listed below are links to blogs that reference Using State-Level Data to Estimate How Labor Market Slack Affects Wages:

Comments

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

April 10, 2014

Reasons for the Decline in Prime-Age Labor Force Participation

As a follow up to this post on recent trends in labor force participation, we look specifically at the prime-age group of 25- to 54-year-olds. The participation decisions of this age cohort are less affected by the aging population and the longer-term trend toward lower participation of youths because of rising school enrollment rates. In that sense, they give us a cleaner window on responses of participation to changing business cycle conditions.

The labor force participation rate of the prime-age group fell from 83 percent just before the Great Recession to 81 percent in 2013. The participation rate of prime-age males has been trending down since the 1960s. The participation rate of women, which had been rising for most of the post-World War II period, appears to have plateaued in the 1990s and has more recently shared the declining pattern of participation for prime-age men. But the decline in participation for both groups appears to have accelerated between 2007 and 2013 (see chart 1).

140410_1

We look at the various reasons people cite for not participating in the labor force from the monthly Current Population Survey. These reasons give us some insight into the impact of changes in employment conditions since 2007 on labor force participation. The data on those not in the official labor force can be broken into two broad categories: those who say they don't currently want a job and those who say they do want a job but don't satisfy the active search criteria for being in the official labor force. Of the prime-age population not in the labor force, most say they don't currently want a job. At the end of 2007, about 15 percent of 25- to 54-year-olds said they didn't want a job, and slightly fewer than 2 percent said they did want a job. By the end of 2013, the don't-want-a-job share had reached nearly 17 percent, and the want-a-job share had risen to slightly above 2 percent (see chart 2).

140410_2

Prime-Age Nonparticipation: Currently Want a Job
Most of the rise in the share of the prime-age population in the want-a-job category is due to so-called marginally attached individuals—they are available and want a job, have looked for a job in the past year, but haven't looked in the past four weeks—especially those who say they are not currently looking because they have become discouraged about job-finding prospects (see the blue and orange lines of chart 3). In 2013, there were about 1.1 million prime-age marginally attached individuals compared to 0.7 million in 2007, and the prime-age marginally attached accounted for about half of all marginally attached in the population.

140410_3

The marginally attached are aptly named in the sense that they have a reasonably high propensity to reenter the labor force—more than 40 percent are in the labor force in the next month and more than 50 percent are in the labor force 12 months later (see chart 4). This macroblog post discusses what the relative stability in the flow rate from marginally attached to the labor force means for thinking about the amount of slack labor resources in the economy.

140410_4

Prime-Age Nonparticipation: Currently Don't Want a Job
As chart 2 makes evident, the vast majority of the rise in prime-age nonparticipation since 2009 is due to the increase in those saying they do not currently want a job. The largest contributors to the increase are individuals who say they are too ill or disabled to work or who are in school or training (see the orange and blues lines in chart 5).

140410_5

Those who say they don't want a job because they are disabled have a relatively low propensity to subsequently (re)enter the labor force. So if the trend of rising disability persists, it will put further downward pressure on prime-age participation. Those who say they don't currently want a job because they are in school or training have a much greater likelihood of (re)entering the labor force, although this tendency has declined slightly since 2007 (see chart 6).

140410_6

Note that the number of people in the Current Population Survey citing disability as the reason for not currently wanting a job is not the same as either the number of people applying for or receiving social security disability insurance. However, a similar trend has been evident in overall disability insurance applications and enrollments (see here).

Some of the rise in the share of prime-age individuals who say they don't want a job could be linked to erosion of skills resulting from prolonged unemployment or permanent changes in the composition of demand (a different mix of skills and job descriptions). It is likely that the rise in share of prime-age individuals not currently wanting a job because they are in school or in training is partly a response to the perception of inadequate skills. The increase in recent years is evident across all ages until about age 50 but is especially strong among the youngest prime-age individuals (see chart 7).

140410_7

But lack of required skills is not the only plausible explanation for the rise in the share of prime-age individuals who say they don't currently want a job. For instance, the increased incidence of disability is partly due to changes in the age distribution within the prime-age category. The share of the prime-age population between 50 and 54 years old—the tail of the baby boomer cohort—has increased significantly (see chart 8).

140410_8

This increase is important because the incidence of reported disability within the prime-age population increases with age and has become more common in recent years, especially for those older than 45 (see chart 9).

140410_9

Conclusions
The health of the labor market clearly affects the decision of prime-age individuals to enroll in school or training, apply for disability insurance, or stay home and take care of family. Discouragement over job prospects rose during the Great Recession, causing many unemployed people to drop out of the labor force. The rise in the number of prime-age marginally attached workers reflects this trend and can account for some of the decline in participation between 2007 and 2009.

But most of the postrecession rise in prime-age nonparticipation is from the people who say they don't currently want a job. How much does that increase reflect trends established well before the recession, and how much can be attributed to the recession and slow recovery? It's hard to say with much certainty. For example, participation by prime-age men has been on a secular decline for decades, but the pace accelerated after 2007—see here for more discussion.

Undoubtedly, some people will reenter the labor market as it strengthens further, especially those who left to undertake additional training. But for others, the prospect of not finding a satisfactory job will cause them to continue to stay out of the labor market. The increased incidence of disability reported among prime-age individuals suggests permanent detachment from the labor market and will put continued downward pressure on participation if the trend continues. The Bureau of Labor Statistics projects that the prime-age participation rate will stabilize around its 2013 level. Given all the contradictory factors in play, we think this projection should have a pretty wide confidence interval around it.

Note: All data shown are 12-month moving averages to emphasize persistent shifts in trends.

Melinda PittsBy Melinda Pitts, director, Center for Human Capital Studies,

John RobertsonJohn Robertson, a vice president and senior economist in the Atlanta Fed's research department, and

Ellyn TerryEllyn Terry, a senior economic analyst in the Atlanta Fed's research department

April 10, 2014 in Business Cycles, Employment, Labor Markets | Permalink

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341c834f53ef01a5119b7704970c

Listed below are links to blogs that reference Reasons for the Decline in Prime-Age Labor Force Participation :

Comments

have you considered that a number of people will say they dont want a job because they have experienced repeated frustration in finding one? it's better for one's psyche to lie to yourself and to others about such than to accept the fact that one has repeatedly been rejected...

Posted by: rjs | April 10, 2014 at 04:39 PM

Astonishing decline in male labor force participation since 1970s.

I would be interested to see more detailed age bracket than category of the 25 - 54 age brackets.

This is so we can see if the decline over time is consistent for all ages or the particular works from certain age that flows through remainder of their working life.

Jason

Posted by: Jason | April 12, 2014 at 10:21 PM

Clearly there is nobody who is unemployed who does not want a job. This article is simply a deceptive representation of the facts. The problem is largely that employers will not hire qualified people unless they have done the exact same job before. They will not for example hire an Architect to work as a project manager at a company that manufactures windows, because the HR people use IT to scan the resumes in place of interviews and will only choose from the set of people who have been employed by manufacturers of windows in the past.

Do the survey and the research over again and ask the right questions. The problem more than likely is that the most qualified people are being overlooked, are frustrated because they can,t crossover to a different industry or are victims of age discrimination. You can be certain that most people want to have a job. Sop, dig deeper.

Posted by: Terry L. Walker, ARCHITECT | April 14, 2014 at 11:01 AM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

April 08, 2014

A Closer Look at Post-2007 Labor Force Participation Trends

Introduction
The rate of labor force participation (the share of the civilian noninstitutionalized population aged 16 and older in the labor force) has declined significantly since 2007. To what extent were the Great Recession and tepid recovery responsible?

In this post and one that will follow, we offer a series of charts using data from the Current Population Survey to explore some of the possible reasons behind the 2007–13 drop in participation. This first post describes the impact of the changing-age composition of the population and changes in labor force participation within specific age cohorts—see Calculated Risk posts here and here for a related treatment, and also this recent BLS study. The next post will look at the issue of potential cyclical impacts on participation by examining the behavior of the prime-age population.

Putting the decline in context
After rising from the mid-1960s through 1990, the overall labor force participation rate was relatively stable between 1990 and 2007. But participation has declined sharply since 2007. By 2013, participation was at the lowest level since 1978 (see chart 1).

140407_1

For men, the longer-term declining trend of participation accelerated after 2007. For women, after having been relatively stable since the late 1990s, participation began to decline after 2009. The decline for both males and females since 2009 was similar (see chart 2).

140407_2

The impact of retirement
One of the most important features of labor force participation is that it varies considerably over the life cycle: the rate of participation is low among young individuals, peaks during the prime-age years of 25 to 54, and then declines (see chart 3). So a change in the age distribution of the population can result in a significant change in overall labor force participation.

140407_3

The age distribution of the population has been shifting outward for some time. This is a result of the so-called baby boomer generation—that is, people born between 1946 and 1964 (see chart 4). The oldest baby boomers turned 62 in 2008 and became eligible for Social Security retirement benefits.

140407_4

At the same time the age distribution of the population has shifted out, the rate of retirement of older Americans has been declining. Retirement rates have generally been drifting down since the early 2000s (see chart 5). The decline in age-specific retirement rates has resulted in rising age-specific labor force participation rates. For example, from 1999 to 2013, the share of 62-year-old retirees declined from 38 percent to 28 percent. The BLS projects that this trend will continue at a similar pace in coming years (see table 3 of the BLS report).

140407_5

Although the decline in the propensity to retire has put some upward pressure on overall labor force participation, that effect is dominated by the sheer increase in the number of people reaching retirement age. The net result has been a steep rise in the share of the population saying they are not in the labor force because they are retired (see chart 6).

140407_6

Participation by age group
Individuals aged 16–24
The labor force participation rate for young individuals (between 16 and 24 years old) has been generally declining since the late 1990s. After slowing in the mid-2000s, the decline accelerated again during the Great Recession. However, participation has been relatively stable since 2009 (see chart 7). Nonetheless, the BLS projects that the participation rate for 16- to 24-year-olds will decline further, albeit at a slower pace than it declined between 2000 and 2009, and will fall a little below 50 percent by 2022.

140407_7

The change in participation among young people can be attributed almost entirely to enrollment rates in education programs (see here) and lower labor force participation among enrollees (see chart 8). The change in the share of 16- to 24-year-olds who say they don't currently want a job because they are in school closely matches the change in labor force participation for the entire cohort.

140407_8

Individuals aged 25–54 (prime age)
Generally, people aged 25 to 54 are the group most likely to be participating in the labor market (see chart 3). These so-called prime-age individuals are less likely to be making retirement decisions than older individuals, and less likely to be enrolled in schooling or training than younger individuals.

However, the prime-age labor force participation rate declined considerably between 2007 and 2013, and at a much faster pace than had been seen in the years prior to the recession (see chart 9). Reflective of the overall gender-specific participation differences seen in chart 2, the decline in prime-age female participation did not take hold until after 2009, and since 2009 the decline in both prime-age male and female participation has been quite similar. Nevertheless, the BLS projects that prime-age participation will stabilize in coming years and prime-age participation in 2022 will be close to its 2013 level.

140407_9

Implications
The BLS projects that participation by age group will look like this in 2022 relative to 2013 (see chart 10).

140407_10

Participation by youths is projected to continue to fall. The participation of older workers is projected to increase, but it will remain significantly lower than that of the prime-age group. Combined with an age distribution that has also continued to shift outward (see chart 11), the overall participation rate is expected to decline over the next several years from its 2013 level of around 63.3 percent. From the BLS study:

A combination of demographic, structural, and cyclical factors has affected the overall labor force participation rate, as well as the participation rates of specific groups, in the past. BLS projects that, as has been the case for the last 10 years or so, these factors will exert downward pressure on the overall labor force participation rate over the 2012–2022 period and the rate will gradually decline further, to 61.6 percent in 2022.

140407_11

However, an important assumption in the BLS projection is that the post-2007 decline in prime-age participation will not persist. Indeed, the data for the first quarter of 2014 does suggest that some stabilization has occurred.

But separating what is trend from what is cyclical is challenging. The rapid pace of the decline in participation among the prime-age population between 2007 and 2013 is somewhat puzzling. Could this decline reflect a temporary cyclical effect or something more permanent? A follow-up blog will explore this question in more detail using the micro data from the Current Population Survey.

Note: All data shown are 12-month moving averages to emphasize persistent shifts in trends.

Update: The authors acknowledge a debt to Tomaz Cajner and Bruce Fallick for their influence on some of this material. We regret inadvertently omitting this acknowledgement in the original post.

Melinda PittsBy Melinda Pitts, director, Center for Human Capital Studies,

John RobertsonJohn Robertson, a vice president and senior economist in the Atlanta Fed's research department, and

Ellyn TerryEllyn Terry, a senior economic analyst in the Atlanta Fed's research department

April 8, 2014 in Business Cycles, Employment, Unemployment | Permalink

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341c834f53ef01a51199e182970c

Listed below are links to blogs that reference A Closer Look at Post-2007 Labor Force Participation Trends:

Comments

I am having difficulty reconciling Charts 1 and 2 as chart 2 seems to suggest much lower total participation rates than are suggested in Chart 1. Any help much appreciated.

Posted by: James Thomas | May 28, 2014 at 01:30 PM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in