Evan Soltas

Fiscal Policy and the ZLB

noreply@blogger.com (Evan Soltas) — Sat, 19 Mar 2016 00:38:00 +0000

I have been doing some reading for my undergraduate thesis, which looks at the role of credit-supply shocks in the Spain during its housing boom and bust, and I came across some interesting thoughts from Bob Hall. Commenting on research by Alan Auerbach and Yuriy Gorodnichenko, Hall makes some useful points that contradict a lot of the received wisdom about the efficacy of fiscal policy:

I conclude that the chapter uncovers a proposition of great importance in macroeconomics—that the response to government purchases is substantially greater in weak economies than in strong ones. The finding is a true challenge to current thinking. The first thing to clear away is that the finding has little to do with the current thought that the multiplier is much higher when the interest rate is at its lower bound of zero...Standard macro models have labor and product supply functions that are close to linear over the range of activity in the OECD post-1960 sample. The simple idea that output and employment are constrained at full employment is not reflected in any modern model that I know of. [Bolding is by me, not Hall.]

On the economics blogosphere, the "current thought" is also that, because monetary policy is in certain respects (that is, if only by social convention) constrained when the policy rate hits zero, fiscal policy becomes discontinuously powerful at the zero lower bound. Once the policy rate is a quarter of a percentage point, time to turn off fiscal policy, one might infer. Scott Sumner is one of the clearest and most persuasive exponents of this view—see here, for instance.

It turns out that the best evidence on fiscal policy does not support it. That conclusion is new to me, which is to say that I think I have written things that rely on that view, and I would now consider them to be wrong. Hall suggests there either have to be (1) much more powerful nonlinearities in supply functions around the natural rate of unemployment than economists currently think there are or (2) nonlinear monetary-policy response functions, so that the strength of the central bank's response to the marginal bit of inflation is increasing in the level of inflation.

If you combine the Brainard conservatism principle and some intertemporal thinking about the zero lower bound, then you get that optimal monetary policy should do little to offset positive demand shocks even when interest rates are low and positive.

To be clear on what I mean here: First, it is a classic result that, when the efficacy of monetary policy is uncertain, it should not fully stabilize demand. Second, if the zero lower bound poses any restrictions on monetary policy, and it obviously does, if only in many indirect ways, then the appropriate amount of conservatism actually increases the risk of a future zero lower bound event rises, which is basically a function of the current policy rate. Fiscal multipliers are, as a result, well above zero when the policy rate is positive and decrease slowly and smoothly in the policy rate.

Trump and Nonwhite Names

noreply@blogger.com (Evan Soltas) — Thu, 17 Mar 2016 04:26:00 +0000

This piece from David Wasserman is, in my view, one of the most interesting perspectives on the role that racism and xenophobia are playing in the 2016 Republican primary. Wasserman:

If Donald Trump somehow falls three delegates short of reaching the magic 1,237 delegates needed for the Republican nomination, he may be haunted by an obscure outcome from the primary voting in Illinois on Tuesday. There’s clear evidence that Trump supporters in Illinois gave fewer votes to Trump-pledged delegate candidates who have minority or foreign-sounding names like “Sadiq,” “Fakroddin” and “Uribe,” potentially costing him three of the state’s 69 delegates. This pattern appears to be a phenomenon unique to Trump’s supporters.

His test of the hypothesis is informal but compelling. He compares the names of the Trump delegates who came in last and those who came in first within each district. The idea was interesting enough to me, though, that it deserved a more rigorous test.

Why do I find this set-up so compelling? First, due to an unusual primary voting system in Illinois, we have multiple delegates running for the same candidate, and delegates' names appear on the ballot, which allows me to isolate the effect of bias from the effect of political disagreement. Donald Trump isn't any different whether you let one Raja Sadiq represent him, or whether you give that job to a Doug Hartmann—one just happens to be white, and the other is not. Second, we have multiple candidates running for the same nomination. Third, we have multiple districts in which these votes are occurring simultaneously. These three factors practically create a laboratory to test bias among Trump voters.

I took the same data as Wasserman but for all 458 individuals who ran to Republican delegates for Illinois—multiple individuals run for each presidential candidate—recording their first and last names, the candidate they represented, and the district in which they ran.

Then I did something simple. I mapped data from the 2000 Census on the racial breakdown of the population by last name to each of the candidates. The non-Hispanic white percentage of each last name, in particular, gave me an objective measure to test the phenomenon Wasserman discovered.

Importantly, the purpose isn't to determine whether any delegate is white or not—but rather what voters who are sensitive to race might think. About 95 percent of Americans last name "Schumann," for instance, are white; only about 5 percent with the last name "Fuentes" are.

Wasserman's discovery holds up in my test: Trump delegates won significantly more votes when they had "whiter" last names relative to other delegates in their district. This effect does not appear for any of the other Republican candidates, and it is strongest in districts with high Trump vote shares.

Trump delegates who were likely to be perceived as nonwhite, in particular, won about 2 percentage points less of the vote (95% confidence interval: 0.8 p.p. to 3.1 p.p. less) than those who were likely to be perceived as white. In the districts where Trump performed most strongly, the vote cost of a "nonwhite" last name was closer to 3 percentage points, as compared to no effect at all where Trump performed worst.

My conclusion from all this is that Trump voters are significantly more racially motivated than Illinois Republicans who did not vote for Trump. Since Donald Trump won Illinois by 8 percentage points, the effect was politically important but not decisive.

What I am going to do next with this is see if these results hold in the 2012 Republican primary and wait for Pennsylvania, which also has a primary system in which delegates' names appear on the ballot.

* * *

My data are available here: CSV format, DTA format.

Update (3/18/16): By popular demand, and because I agree that open science should mean "push-button replication," my .DO file is available here. It will allow you to replicate my results in Stata.

Update (3/21/16): I now have done this analysis for the 2012 Illinois primary. Non-Romney delegates (i.e. mostly those for Newt Gingrich and Rick Santorum) received a 0.9 percentage point boost (95% CI: 0.2 p.p. to 1.7 p.p.) for names that were likely to be perceived as white. No significant patterns by perceived race appeared for Romney delegates.

Did Michigan Vote Against Trade?

noreply@blogger.com (Evan Soltas) — Thu, 10 Mar 2016 06:17:00 +0000

Paul Krugman writes:

The Sanders win defied all the polls, and nobody really knows why. But a widespread guess is that his attacks on trade agreements resonated with a broader audience than his attacks on Wall Street; and this message was especially powerful in Michigan, the former auto superpower. And while I hate attempts to claim symmetry between the parties — Trump is trying to become America’s Mussolini, Sanders at worst America’s Michael Foot — Trump has been tilling some of the same ground. So here’s the question: is the backlash against globalization finally getting real political traction?

It's certainly plausible that Donald Trump's and Bernie Sanders' strong performances in the Michigan primary are manifestations of a protectionist backlash. I was curious, and so using David Dorn's data on local declines in manufacturing and vote data from the Michigan Bureau of Elections, I checked.

The idea: To see if Trump and Sanders benefited from protectionist sentiment, I will compare their respective primary vote shares in a given county against how badly affected that county was by the decline of manufacturing employment.

The protectionist-backlash explanation of Trump and Sanders does not seem to hold up:

Sanders' county-level vote share in the Democratic primary is not significantly correlated with that county's change in manufacturing employment, measured as a share of its working-age population. No matter how intense the local decline in manufacturing, Sanders won about half the Democratic vote.

Trump's vote share in the Republican primary is in fact significantly positively correlated with that measure, meaning that he performed most strongly in areas where manufacturing's decline has been least important. Where manufacturing's decline was most intense, Trump received about 30 percent of the Republican vote, and where it was lightest, he received about 50 percent of the Republican vote.

Here are vote-weighted scatterplots to visualize these conclusions:

By the way, I completely expected Krugman to be right on this one—and I remain quite surprised that the protectionist-backlash explanation isn't apparent in the data. But it's not.

* * *

Note: Dorn's data is actually at the "commuting zone" level, so I assumed that manufacturing declines were constant across counties within each commuting zone. I have data on 21 commuting zones and 83 counties in Michigan.

If you want to try to prove me wrong, I've made it easy for you: My pre-processed data are available here for download.

Boom, Bust, and Inference

noreply@blogger.com (Evan Soltas) — Thu, 10 Mar 2016 03:32:00 +0000

Matt Klein circulated this nice graph from Renaissance Macro on Twitter, and he suggested that we might infer from it that productivity growth responds to cyclical conditions.

Since this argument has been getting a lot of airtime recently—from Jared Bernstein and Gerald Friedman, to name two examples—I wanted to explain why I think we should be skeptical about this piece of evidence and others like it, and to explain what kind of evidence we should find persuasive.

The deep underlying economic question is: Under what conditions can we make inferences about structural relationships from cyclical patterns?

We see that productivity growth is correlated to the output gap. In fact, it is not only labor productivity but also total factor productivity that behaves in this way. Does that mean that running small output gaps would raise the long-run rate of productivity growth? Not necessarily.

Here is a counterexample. Suppose that firms must split their workers between two activities, production and maintenance. When firms allocate more workers to production, they make more money. When they allocate more to maintenance, they make less.

It makes sense for firms to defer maintenance to times when more production is less beneficial to them on the margin, as in recessions. But no firm can defer maintenance forever. Despite a cyclical relationship, it is not true that a policymaker could permanently raise productivity by keeping the output gap low. One can make similar arguments about firms' use of overtime or firms' willingness to shift capital investments through time.

The problem with these endogenous-supply explanations is that, if given too much power, become hard to square with what we know about economic growth in the long run. If it were true that countries could raise productivity growth by maintaining high aggregate demand, output should be only weakly related to fundamentals. In fact, the cross-country evidence on economic growth points to the opposite conclusion. One can predict output quite well from simple structural facts about an economy, like the capital stock, education, technology, and institutions.

Growth economics should limit what kind of long-run gains we can rightly expect. Yet it shouldn't eliminate them. For compelling evidence that short-run economic activity has long-run consequences, look to work by Larry Summers and Antonio Fatás as well as by Lawrence Ball.

Eurozone countries that applied the harshest austerity from 2009 to 2011, Summers and Fatás show, saw the largest declines in economic growth expected in the long run as well as the short run. Fiscal consolidations that create output gaps, they conclude, also have disturbingly large effects on long-run output. Ball shows that, across developed countries, the size of a country's output gap is extraordinarily predictive of the loss in potential output.

Both of these findings seem to require a universe in which long-run output depends on keeping the economy near full employment. The growth literature, on the other hand, should warn us about extending this conclusion too far, and convincing ourselves we can engineer sustainable supply-side booms by boosting aggregate demand.

Who Is Voting For Trump?

noreply@blogger.com (Evan Soltas) — Mon, 29 Feb 2016 07:16:00 +0000

"I live in a rather special world. I only know one person who voted for Nixon. Where they are I don't know. They're outside my ken. But sometimes when I'm in a theater I can feel them." — Pauline Kael, 1972

"I began to ask everyone I met: Do you know anyone who supports Donald Trump? In more cases than not—actually, in nearly all the cases—the answer was no." —Byron York, 2016

* * *

Two factors explain 60 percent of all the variance in Trump's support across 227 New Hampshire voting precincts:

Cook Partisan Voting Index: For every 1 percentage point more liberal the precinct, Donald Trump's share of votes rises by 0.48 percentage points.

College attendance: For every 1 percentage point more college graduates over the age of 25, Donald Trump's share of votes falls by 0.65 percentage points.

Household income and population are insignificant statistically or substantively. Here is a visualization of Trump voters as both moderate and low-education, where I split the 227 precincts into PVI and education quintiles:

* * *

My data set is here for your amusement. Compiled from New Hampshire Public Radio, Will Tucker, and the American Community Survey.

Alone, Would Rubio Beat Trump?

noreply@blogger.com (Evan Soltas) — Mon, 29 Feb 2016 04:26:00 +0000

Donald Trump could be beaten, Republicans say, if only the opposition would unite behind an anti-Trump candidate.

Not so fast. Using precinct-level data from the New Hampshire primary, I tested this "united we stand, divided we fall" hypothesis, which has so quickly gained credence among anti-Trump Republicans. It came up short.

Here's the basic idea. Let's say that Republicans come in two flavors—moderate and conservative—and are split half-and-half between those categories. Suppose that Trump wins every moderate vote and none of the conservative votes, and that the non-Trump candidates divide up the conservatives. Then the GOP looks like this:

Let's imagine that voters have a specific behavior: When a candidate drops out, the candidate's voters go to the remaining candidates in proportion to their support within their flavor. Consider, for example, what would happen under this assumption when Jeb Bush dropped out.

Since, in my simplified example, the five candidates split conservative votes equally, when Bush dropped out, every candidate but Trump divides Bush's votes equally, and Trump gets nothing.

That's what the unite-against-Trump crowd wants to believe, because what happens if every candidate but one gets out should be obvious: The remaining anti-Trump becomes competitive with Trump.

But consider another possible state of the universe, again simplified: Moderates and conservatives split their votes in the same way among candidates, like below.

Now forcing candidates out, so as to unite around an anti-Trump, does not work. Donald Trump's voting share gains in proportion to the other's, as you can see:

These childish sketches make an important point: The viability of the "unite-against-Trump" strategy depends on how polarized the Republican electorate is. If Trump's voters form a discrete bloc, it will work. If they don't, it won't. Trump will gain in proportion to the remaining candidates, and so no ground will be gained against him.

So, what does the Republican electorate look like? Obviously, they don't split neatly into two categories. They do, however, widely vary in the intensity of their conservatism. That is, there's a spectrum. Maybe uniting Republicans who are on the conservative end of the ideological spectrum against Trump will work.

That's a testable hypothesis, unlike the vague hope that uniting around a Trump candidate will benefit the remaining candidate more than it benefits Trump. Let's test it with data from the New Hampshire primary.

Will Tucker, a New Hampshire politics blogger, has computed a Cook PVI measure for every New Hampshire voting precinct. This measures how conservative each precinct is relative to the rest of the U.S. using their voting behavior from prior elections. If Republicans won on average 55 percent of the vote in the precinct of Hooksett, N.H. and 50 percent of the national vote, for example, then Hooksett would have a Cook PVI of +5.

I also obtained precinct-level voting data from the CartoDB page of New Hampshire Public Radio.

We can easily compare Donald Trump's share of votes in the 2016 N.H. Republican primary to that N.H. precinct's Cook PVI. What we see confirms, in fact, that Donald Trump's voters were more moderate, in terms of their precinct's Cook PVI:

What this graph tells us is that precincts that usually lean 10 percentage points to the Democrats, Donald Trump won about 47 percent of the vote—but in precincts that usually lean 10 percent to the Republicans, Trump won 35 percent of the vote. That's a highly significant moderate slant.

Let's recall our earlier belief about how voters would move when a candidate drops out: They reallocate in proportion to other voters like them. This is a more sophisticated assumption than splitting the votes proportionally between candidates—e.g. if there were three candidates with a third each, and one dropped out, leaving the two remaining with half—and takes seriously the idea that voters are ideological.

To model the effects of a candidate dropping out, I first have to model the ideological distribution of their voters. Imagine that Donald Trump won 40 percent of the vote in Hooksett, N.H., which, we suppose, had 1000 Republican primary ballots cast and had a Cook PVI of +5. Then what I do is assume that there are 400 Trump supporters with a Cook PVI of +5.

Repeat this process for every candidate and every precinct, and what you get is an approximate ideological distribution of voters for each Republican presidential candidate.

(Caveat: I say "approximate" because we don't have data on individuals. If most of the ideological diversity is within-precinct rather than between-precinct, this won't work. Fortunately, N.H. precincts are tiny! The average one had 1,000 ballots cast, and the standard deviation of their Cook PVI was 8.2 percentage points—the difference between a blow-out victory for the Republicans and one for the Democrats. There is a lot of between-precinct ideological diversity in N.H.)

For example, here is the ideological distribution of New Hampshire's Trump voters:

Where the line is high, the Trump supporters are densest. Notice that the modal Trump voter, denoted with the dashed line, is actually to the left of zero in the Cook PVI measure. As you go far to either ideological extreme, the distribution goes to zero, because there aren't that many extreme precincts.

I've used something called kernel density estimation to produce "smooth" distributions of voters for each candidate—don't worry, you don't need to know what it is. This allows us to see how the candidates split New Hampshire Republicans, sorting them from moderate to conservative:

You should notice a problem for the "unite-against-Trump" strategy if you think back to our simplified model. Yes, Trump's voters lean to the moderate side. But not that much! If Republicans respond to candidates dropping out in the way we think they do, Trump gains almost exactly as much from dropouts as do the remaining non-Trump candidates.

We can show this a bit more precisely. Let's say that everyone but Rubio and Trump dropped out before New Hampshire. By our assumptions, we can figure out the new vote shares with the following formula:

where p is the Cook PVI. That is just the math-y way of saying: "Take everyone who voted for someone else than Trump or Rubio, and split them up between Trump and Rubio according to the votes of people ideologically like them."

We can compare that estimate to a baseline in which we just divide voters evenly between the remaining candidates:

If ideology is a promising dividing line to unite Republicans, then the non-Trump percentage will be higher in the method that accounts for ideology than in the method that doesn't. If ideology doesn't work, the figures will be about the same.

Here they are: Without accounting for ideology, my estimate is that Trump would have won 77.0 percent of the New Hampshire vote in a head-on primary against Rubio. When I account for ideology, Trump would have won....76.9 percent of the N.H. primary vote.

What my analysis suggests is that ideology is not strong enough a differentiator between Trump voters and other Republican voters for a "conservatives against Trump" strategy to work.

Yet there are surely some other lines that divide Trump voters from most Republicans. Exit polls, for example, tell us that income and education are two such lines: Trump voters earn less and are less educated. Other polls tell us that Trump is not really anyone's second choice—that is, Republicans either love him or despise him.

So maybe the "unite against Trump" strategy can work. But it won't work by uniting an ideological coalition. It will need to run along lines of class, or personality, or something else.

Update (2/29/16): In my follow-up post, I downloaded precinct-level education data. I just tried doing this counterfactual analysis using both PVI and education. It puts Donald Trump at 76.7 percent of the N.H. primary vote. So the Trump coalition is sufficiently broad that (I think) the "united against Trump" strategy is probably mistaken, in that Rubio or whatever alternative would not gain against Trump from having candidates drop out.

In fact, the candidate whose voters looked most like Trump's in N.H. was Ted Cruz. Having Cruz drop out would have actually been (mildly) counterproductive. Better to have Kasich or Bush drop out.

Bad Energy Debt and the Banks

noreply@blogger.com (Evan Soltas) — Tue, 23 Feb 2016 16:29:00 +0000

Usually, a falling oil price is good news for just about everybody.

Consumers benefit from cheaper gasoline. Investors expect higher profits from companies that use oil to make their products. Only the energy industry gets squeezed — and that’s too small, in the grand scheme of the economy, to count for much.

Yet it feels different this time, as oil hovers near $30 a barrel. Nobody seems to be celebrating the decline in energy prices.

Consumers are saving almost every dollar they would have spent on oil, finds Matt Busigin, a portfolio manager. Investors are even grimmer. Stocks have fallen sharply since the start of 2016, led not only by energy but also by financials. Deutsche Bank, most notably, has been under severe pressure from investors worried about the chance it might default.

That link to the financial system has people on edge. Andrew Levin, a former adviser to Ben Bernanke and Janet Yellen, has been ringing the “recession” alarm bell as loud as he can. Larry Summers has warned policymakers to “heed the fears of financial markets.”

The banks are tied up in this for a simple reason: America’s fracking boom was brought to you by very aggressive financing. Buying land, drilling a well, renting equipment, hiring a team, and securing pipeline or rail space to ship out the oil — all that takes capital, and the banks provided it at low interest rates with little equity from the borrower.

Banks lent so much to frackers that the cost of debt service consumed 60 percent of cash flow before oil prices fell, according to the Energy Information Administration. The collapse in oil prices makes that kind of debt unpayable. Frackers will default and force banks to eat the loss.

If we are to heed the financial markets, we need to know what they are saying. Is the decline really about energy?

To answer that question, I took the weekly-average stock prices of 37 banks since 2003 from Yahoo Finance and data on the energy exposure of those banks from a Raymond James report to investors earlier this month. The Raymond James data measures banks' exposures as the energy share of their total loan books.

If finance’s current troubles are an extension of the collapse in energy prices, then banks with the largest energy exposures will see their stock prices fall relatively more than banks with lesser energy exposures.

I took the data and estimated the following regression:

where s is the logarithm of the stock price of bank b in week t and n is its energy exposure. Then the alphas are fixed effects for bank and time, and the epsilon is the error term. The coefficients we’re interested in are the beta terms, which show how banks’ relative stock prices have moved in line with their energy exposures.

Here’s a graph of it from 2013 to present:

What we see is that, for every 1 percentage point increase in a bank’s exposure to energy, its stock has declined 3.86 percentage points since January 2013 relative to other bank stocks, with most of the drop occurring in two waves: late 2014 and right now. (That’s tracing the two big drops in the price of oil.) Energy exposure, in fact, explains about 40 percent of the decline in bank equity prices since the beginning of the year.

You might worry that banks that have high exposures to the energy sector also tend to respond to be risky in general — that is, their sensitivity to the performance of the financial sector is relatively high. That could explain the pattern above, given the sharp decline in financial stocks. It turns out, however, that’s not what’s going on. We can distinguish between energy beta and financial-sector beta — and what we are measuring is mostly an energy effect. Controlling for financial-sector beta, for every 1 percentage point increase in a bank’s exposure, its stock has declined 2.94 percentage points since January 2013 relative to other bank stocks.

Furthermore, there’s a useful fact about banks. Say their debt-to-equity leverage ratio is L. Then a 1 percent decline in the value of assets implies an L+1 percent decline in the value of equity. As a result, given banks’ energy exposures and their leverage, we can estimate the discount that markets are applying to energy assets. For a stock decline D, the implied discount on assets is D/(L+1).

The average leverage ratio across the 37 banks is 8.51. (My leverage data come from here.) Therefore, since a 1-percent increase in energy assets currently predicts a 3.86-percent lower stock price, the market thinks the discount on energy assets should be huge: 40.6 percent of their value, or 30.9 percent if you correct for financial-sector beta. Even if you assume that the bank’s loss given default is near 100 percent on these energy loans, the market thinks the coming wave of energy defaults is going to be historically massive.

At the moment, the stress in the financial sector has a clear cause: the energy debt held by banks.

* * *

My dataset (157KB) is available here for download as a .dta file for a limited time. If it is no longer online when you read this post, send me an email.

Image: Well pads along Little Missouri River, with Elkhorn Unit of Theodore Roosevelt National Park in background. Chris Boyer, Kestrel Aerial Services, Inc., on May 20, 2014, in Billings, ND. Link: https://www.flickr.com/photos/npca/15700621196/in/album-72157646828990714/.

Does iTunes Compete with Spotify?

noreply@blogger.com (Evan Soltas) — Sun, 17 Jan 2016 06:04:00 +0000

iTunes and Spotify are both leaders in digital music. For that reason they are in competition. Right?

Well, to an economist, there's a bit of a problem. Both iTunes and Spotify, at least in their basic versions, are free -- and when goods are free, the notion of competition gets cloudy.

Economists, for example, assess whether two products are substitutes or complements for each other with a concept called the cross-elasticity of demand. When the price of one product rises, that is, how does demand for the other product respond? When it rises, the products are substitutes. When it falls, they are complements. Simple.

The idea can't get off the ground without a price. Most of the time, that's not a problem. Ford competes with Toyota, apples with oranges. No price, no scarcity -- and, it might seem, no economics. But the growing amount of economic activity on the Internet, where competition is so obvious and yet so many products are free, requires a more flexible notion of competition.

Let's stick with iTunes and Spotify as a case study of two high-profile Internet competitors. In the data analysis that follows, I find that iTunes loses a user for every five new users of Spotify and that the introduction of Spotify has so far reduced iTunes use by 15 percent globally. There is, in other words, competition on the Internet -- and competition even when goods are free.

[See update below: More data, bigger results.]

Using Google Trends, I downloaded weekly Google search data for iTunes and Spotify for 57 countries between 2004 and 2015. These, I think, are good proxies for signups for iTunes and Spotify, since people will search on Google to download either. (In my conversation with Josh Gans, I found that searches with "Spotify" or "iTunes" are usually about downloading them.) And the search data seem to match what we see in the limited public data on iTunes and Spotify use. I picked the countries based on where iTunes and Spotify had launched, looking for their target markets.

iTunes was initially released in 2001, and Spotify was in October 2008, both in select countries only. What we see in the Google data is that iTunes has been trending downwards since 2012. Perhaps not coincidentally, that was also when Spotify began expanding beyond its initial markets in northern Europe. That, of course, hardly proves that Spotify was responsible.

For iTunes or Spotify to enter a country, they must strike deals with big music labels in the context of local intellectual property law. Spotify negotiated for two years, for instance, before entering the US. As a result of this legal thicket, Spotify rolled out gradually across my sample of countries -- and I can exploit that gradual expansion to measure the effect of Spotify on iTunes. Using official Spotify press releases, like this one, I added Spotify's entry dates (or lack of entry, for some of the 57 countries) to my dataset. You can download my data here.

To measure the effect of Spotify on iTunes, I'll use an instrumental variables regression, one simple enough that even people who aren't stats nerds or economists can follow.

This method requires an important, but believable, assumption: The only way the timing of Spotify's market entry affects iTunes use is through Spotify use. Said differently, nobody decided to stop listening to music, or switch to Pandora, because Spotify became available. Which wouldn't make much sense. Formally, for the economists, my claim to instrument exogeneity follows from the independence of irrelevant alternatives.

The first step is to build a simple regression model that predicts Spotify use in a given country in a given week from whether Spotify was available in that country, whether it had launched in that country in that week, and how many weeks it had been available in that country.

I found, not surprisingly, that Spotify use was higher in countries where Spotify had entered, higher in the launch week, and growing in the number of weeks since launch. (I also included a term to capture the fact that this growth slowed eventually.) What is perhaps surprising is that my simple model explains 75 percent of the variance in Spotify use among the 57 countries from 2004 to 2015.

Here, for instance, is what the model's predictions look like versus actual data for the US:

Next, I take my model's predicted values for Spotify use for the 57 countries and use those to predict iTunes use. That regression, under our assumptions, actually measures the causal effect of Spotify on iTunes.

I find that, for every one new user of Spotify, iTunes loses -0.23 users -- although the 95-percent confidence interval is quite wide, at -0.40 users to -0.05 users. Roughly speaking, for every five that Spotify gains, iTunes loses one.

I can also use that conversion rate to figure out the total effect of Spotify on iTunes use over time. iTunes use is 15 percentage points lower because of Spotify use than it would have been in a world without Spotify.

Spotify thus explains around a third of the decline in iTunes use since its peak in 2012. There seems to be more to iTunes' decline than competition alone. Yet competition does exist, quite clearly, online and among goods without prices. It would be worth doing a similar analysis for MySpace and Facebook, or Netflix and Hulu and HBO GO.

Note about graphs: It doesn't affect any of the results, but the year labels are slightly off.

Update (1/18/16): I have expanded my dataset to 72 countries, and the marginal effect of Spotify on iTunes is -0.33, with a 95% confidence interval of (-0.51,-0.16). This implies that Spotify has caused a 20-percent drop in iTunes use, and I am thus able to explain about half of the decline of iTunes since 2012. The rest, presumably, must be related to the fact that iTunes is just an awful piece of software.

The First Count

noreply@blogger.com (Evan Soltas) — Fri, 18 Dec 2015 20:23:00 +0000

An unofficial count of the number of anti-Muslim hate crimes since the Paris and San Bernardino attacks is now available, and it looks like my and Seth's model got it right.

Updating the search data to the most recent week, the model estimates there were about 37 hate crimes against Muslims since the Paris attacks. There were 38. And the model estimates there were about 16 hate crimes against Muslims since San Bernardino. There were 18.

Had anti-Muslim sentiment been at normal levels, there would have been about 8 and 4 respectively, so the model did a remarkable job getting right the magnitude of the deviation from normal. These were all "out-of-sample" forecasts -- that is, none of our estimates were made including the last month of data.

A few caveats. First, the model's performance is unusually accurate this time. It's better to expect us to miss by about five to ten hate crimes or so in future surges like this one, rather than being off by one or two. The model leaves more unexplained than this single performance would suggest. Second, this is an unofficial count; as we said in the article, we will be waiting until year-end 2016 for the official numbers, which might be different.

Rise of Hate Search: Follow-Up

noreply@blogger.com (Evan Soltas) — Wed, 16 Dec 2015 21:18:00 +0000

by Evan Soltas and Seth Stephens-Davidowitz

We are adding more detail than we could fit in our op-ed in The New York Times. For all data files, refer to Seth's website.

Simplest Prejudice Measure

The simplest prejudice measure we used was the search “Muslims are ___” that was completed with a negative adjective.

We estimate the top 5 negative searches of this form for Muslims are “Muslims are evil,” “Muslims are terrorists,” “Muslims are bad,” “Muslims are violent,” and “Muslims are dangerous.”

The reason we started with this measure is it is possible to get a similar measure for other groups, which we also did and will more fully explicate in a piece in January.

Racial Threat

To test racial threat versus the contact hypothesis, we looked at anti-Muslim searches in the 10 counties with the highest proportion of Muslims.

These were found here.

We used the negative-adjective searches discussed in the previous section. These included “are Muslims evil?” and “Muslims are evil.” The volumes were found on Google AdWords. Unfortunately, Google AdWords, the only source that gives county-level data, does not include search rates. So, instead, we estimated total searches based on searches for the 10 most common Google searches in the United States, as found on Google Trends.

The search volumes and calculations are in the file MuslimsUSRates.csv.

Note that, even if this suggests that proximity does not lower discrimination, there is strong evidence that organized and facilitated intergroup contact may reduce biases, as a meta-analysis by Pettigrew and Tropp finds.

Islamophobia and Anti-Muslim Hate Crimes

We compared anti-Islam hate crimes to a bunch of searches that both may suggest Muslims are in the news and that Islamophobic is high.

The simplest thing to do is just use the measure of prejudice we developed previously. This is search volume for “ “Muslims are evil,” “Muslims are terrorists,” “Muslims are bad,” “Muslims are violent,” and “Muslims are dangerous.” It was the measure we used to compare prejudice against other groups and in different locations.

At the weekly level, they were highly correlated (r=.16; t=3.7). They were even higher-correlated restricting the data since 2008, when the Google data has become less noisy (r=.26; t=4.78.

They were also highly correlated at the monthly level (r = .37; t=4.57).

Anti-Muslim hate crimes were not similarly correlated to any other prejudice, using this simple, blunt measure.

In addition, the relationship was not explained by trends or monthly factors. (All this data is available at WeeklyPrejudicePlusHateCrimes.csv and MonthlyPrejudicePlusHateCrimes.csv.)

However, to best predict what searches matter and what don’t, we downloaded a large set of weekly searches and compared it to weekly hate crimes. Since the goal was prediction, we just want to let the data speak to what searches best predict hate crimes. We used about 35 common search phrases related to Muslims or Islam, which we found by using Google Correlate, the “top searches” feature within Google Trends, and Google auto-complete. In pre-processing, we normalized all series to means of zero and standard deviations of one. The LASSO selected 12 terms, yielding an L1-norm of 0.82.

Some were obviously Islamophobic, such as “I hate Muslims.” Others had some clearly non-Islamophobic uses. One of the striking things in the data, however, is that even seemingly innocent searches, such as Koran, include many potentially Islamophobic searches, such as those related to burning the Koran.

We put all the data in an OLS Lasso model. The Lasso model generally chose shorter searches -- one or two word searches, rather than many-word searches. A probable reason is that these data were much less noisy at the weekly level. We chose the constraint on the L1- norm by 10-fold cross-validation, minimizing mean squared error. We also tested a Poisson LASSO regression model, which is more appropriate for count data; this yielded virtually identical results and predictive power.

We could explain about 10 percent of the weekly variation with Google searches and about 25 percent of the monthly variation.

Our initial data was through 2013, which was the only data available online. However, we recently obtained new monthly data from 2014. The model was just as strong predicting this new, out-of-sample data, which is strong evidence for its reliability.

The data and R code can be found at LassoData.csv and LassoCode.csv and hatecrimepredictors.csv.

We are writing a full paper on these results. We are also examining to what extent prejudiced searches towards other groups can predict hate crimes against those groups.

Response to Obama’s Speech

Searches During Obama’s Speech.csv includes data for the minute-by-minute search response to Obama’s speech.

AthletesTerroristsSoldiers.csv includes the hourly data on searches for “Muslim terrorists,” “Muslim athletes,” and “Muslim soldiers.” It was the data used to make the accompanying graphic.

Response to San Bernardino

MuslimsBelieveKill.csv shows the paths of a likely-hateful search (“Muslims kill”) and a likely-informative search (“What do Muslims believe?”) after the San Bernardino attacks. Both rise, but Muslims kill rises far more.

We also have data on a large set of such searches, that are available upon request.

Political Responses to Terror Attacks

SyrianRefugees.xlsx includes data on daily search volumes for several common positive and negative searches about Syrian refugees from September 7, 2015 to December 2, 2015.

CloseMosques.xlsx includes hourly data on all searches including the word “mosques” and several common searches that suggest support for closing mosques.

Going to Oxford

noreply@blogger.com (Evan Soltas) — Mon, 23 Nov 2015 05:39:00 +0000

I'm overjoyed to announce that I have been selected as a 2016 Rhodes Scholar. I'm deeply grateful to the selection committee and am excited for this wonderful opportunity to study at Oxford. More details to come.

The Swiss Shock: A Case Study

noreply@blogger.com (Evan Soltas) — Mon, 26 Oct 2015 15:58:00 +0000

In January 2015, the Swiss central bank removed its floor on the exchange rate between the Swiss franc and the euro, allowing its currency to appreciate without limit. The immediate effect was a 20-percent increase in the value of the Swiss franc relative to the euro -- one of the largest revaluations of a developed-world currency in recent history.

The move sent tremors through the financial markets, which had been using the Swiss franc as a funding currency for carry trades and Swiss banks as a haven from the chaos of the Eurozone. Swiss exporters and the tourism industry screamed that the central bank's move would render the country uncompetitive.

I've been fascinated by this move -- whose proximate cause, I have suggested, was the resumption of capital inflows after a two-year pause, and the Swiss central bank's latent unwillingness to sterilize further inflows -- and so I've been waiting to do some post-mortem work.

What are the effects of changes in exchange rates on the macroeconomy? Switzerland provides a beautiful, clean case study. The revaluation was unanticipated before it happened and huge when it did.

To do my analysis, I'll use the synthetic control method that has been pioneered by Alberto Abadie at Harvard. You can read about that method here, but the basic intuition is that you can construct a comparison for the treated unit (in our case, Switzerland) by taking the weighted average of untreated units, where the weights are optimized so that the "synthetic Switzerland" matches actual Switzerland before the treatment as closely as possible.

The two macro variables I want to look at are stock prices and consumer prices. What I find are that the revaluation has reduced consumer prices by 1.5 percentage points but had no significant real effect on Swiss stocks.

I construct synthetic Switzerlands for consumer prices and stock returns separately. For consumer prices, the algorithm says that synthetic Switzerland is a mix of nine European countries, but is mostly a mix of Slovakia, Sweden, Netherlands, and Denmark. I use monthly data from 2004 to 2014 to do the matching. I found it interesting that it picked smaller countries, many of which have their own currencies.

What we find is that, in both actual and synthetic Switzerland, prices were flat prior to the exchange-rate shock. Such is Europe in 2014. Then, starting in January 2015, we see actual Swiss prices begin to diverge from consumer prices in synthetic Switzerland. As of September 2015, Swiss prices are now 1.5 percentage points lower than they would have been absent the revaluation.

I use data from the iShares MSCI indexes for stock returns, and I find that synthetic Switzerland is mostly a mix of Netherlands, Belgium, Sweden, and the United Kingdom. (Worth noting: On a totally different dataset, the algorithm picks roughly the same states.) Turns out we can predict daily stock returns in Switzerland quite well, as this scatterplot of actual versus synthetic Switzerland shows.

But I'm not finding any significant effect on Swiss stocks. Here are the cumulative returns for actual versus synthetic Switzerland from January 2014 to present, and any effect should appear starting in January 2015.

Perhaps this makes American investors less concerned about the effect of the appreciating U.S. dollar on their portfolios. The implication of these findings is that any nominal decline in stock prices is offset by currency appreciation.

I'll try to look next Swiss unemployment, their trade balance, and other real macroeconomic variables.

Boom and Bust and Biotech

noreply@blogger.com (Evan Soltas) — Tue, 29 Sep 2015 00:41:00 +0000

Biotechnology and pharmaceuticals stock prices have declined about 20 percent in the last week, wiping out hundreds of billions of dollars in market capitalization. That drop is in the wake of popular outrage at the headline-grabbing Martin Shkreli, whose firm acquired the antimalarial drug Daraprim and had planned to raise its price 50-fold, as well as rumblings of a substantive public-policy response to pharmaceutical prices from the Hillary Clinton campaign.

It seems to me this market reaction raises two possibilities.

First, is this decline just the "inevitable" correction (in the sense of Blanchard and Watson's rational bubbles) for five years of strong performance from biotech stocks?

Blanchard and Watson proposed an idea of bubbles in which an asset price rises faster than other asset prices most of the time but has a small chance of falling catastrophically. As crazy as that sounds, it works fine in expected-value terms, relative to other investments, because the two cancel each other out.

There is some evidence for this proposition in the stylized fact that it's been the riskiest, best-performing biotech/pharma stocks which have corrected most sharply. For example, just compare the new-school Valeant, Celgene, Regeneron, and Biogen with the old-school Pfizer, Merck, Novartis, and Eli Lilly.

Alternatively, let's suppose that some of the decline reflects actual changes in the expected future profits of biotech. Shouldn't it be disturbing that a rough draft of a policy proposal to restrict drug prices caused biotech to implode? What does that say about the social value of these biotech innovations?

Not good things, I think. To the extent that price regulations hit firms differentially, they will hit firms most dependent on the high-price business model that regulators find objectionable. The market has just told us that new-school biotech is built around this model.

Intuitively, a good product does not depend on which way the regulatory winds blow. A lot of the new-school biotech firms just proved they absolutely do. That should concern anyone who is hopeful about the future of health.

Up. But Not Up, Up, and Away.

noreply@blogger.com (Evan Soltas) — Sat, 26 Sep 2015 15:19:00 +0000

For the last year, I've been putting together simple short-term forecasts for inflation using only oil prices. Since oil has been driving so much of the movement lately, the forecasts have been quite accurate. This January, before most of the decline in inflation, I said there was a 50-50 chance of deflation -- and, sure enough, PCE inflation hit just 0.2 percent in the summer.

What oil giveth, however, the year-over-year calculation taketh away.

Since inflation is usually examined on a year-over-year basis, there are often "base effects" -- if prices fall or rise sharply in one month, that shock is carried over for the next 11 months. And then, the next month, the effect vanishes as the shocked month drops off the base, sending inflation just as sharply in the opposite direction.

Here we go, then. Since much of the decline in oil prices happened in the summer and fall of 2014, inflation is likely to rise sharply in winter 2015. That's what you can see in the forecast above. My baseline forecast is that core and headline will be at or around 2 percent by January 2016.

The model is not really a forecasting model, because almost all its power comes from simple propagation of the oil price into inflation and then the capture of the base effect, so I would not take seriously any longer-term forecasts it produces. It's not designed, for example, to consider the effects of unemployment or wage growth on inflation, which are arguably more important to inflation in the medium run. It's just meant to give you some visibility six months or so ahead. There it succeeds.

So inflation will go up -- but not up, up, and away. The Fed's inflation target is a two-percent annual increase in prices, as measured by the PCE price index. It looks as though, once the oil shock dissipates, the inflation will be on target.

My only concern is what happens when everything snaps into place so quickly. In six months, with inflation at 2 percent and unemployment in the high 4 percent range, that might produce a substantial amount of pressure for the Fed to tighten quickly, even when the Fed should be responding to the signal in the inflation process, not oil-driven noise.

Update (10/13/15): I added +/- 1 standard error bars to the forecast generated by Monte Carlo simulation. I think a reasonable confidence that inflation will return to target is, at this point, warranted.

What's the Case for Big Banks?

noreply@blogger.com (Evan Soltas) — Fri, 25 Sep 2015 20:18:00 +0000

Ever since the global liberalization and deregulation of financial services began in the 1980s, banks have argued that it would be better if they were bigger.

Allow consolidation, bank executives have said, and customers can have what they want from their bank: a full suite of financial services from a bank with global reach and a deep knowledge of financial markets.

To be that kind of bank, however, takes scale. A smaller bank could take deposits and lend so people could buy cars and homes and so businesses could make payroll. But it would never be able to promise an individual a competitive interest rate or to help a business pay a supplier in Japan.

All that seems fair enough. There's a problem with the argument, though: It doesn't seem to be true in the data. If large banks had a competitive advantage, we would be able to see it in their return on assets. If return on assets doesn't increase with bank size, then it's hard to see why free markets would favor scale in banking.

To the banks' point, it does seem to be true that community banks -- those with assets under $1 billion -- operate at a severe disadvantage. The data suggest that, holding leverage constant, they could increase their value by roughly 20 to 30 percent from scale efficiencies. (This probably explains that the consolidation has been concentrated in these banks.)

Yet there's no evidence that returns to scale are increasing beyond that point. And community banks only control 14 percent of bank assets. So that's a free-market case against community banking, rather than one for the rise of the very largest banks.

This was something that researchers noticed in the early 1990s, back when bank consolidation was just beginning. It remains true. So, tell me, what's the free-market argument for big banks? In a world of "too-big-to-fail," the benefits that are supposed to offset the cost of implicit government subsidies are elusive.

More Thoughts on Productivity

noreply@blogger.com (Evan Soltas) — Thu, 17 Sep 2015 16:54:00 +0000

Well, that could have gone better.

Josh Bivens and Larry Mishel have written a response to my blog post on labor productivity and compensation. Since we've had a short discussion over email today, I figured it would be worthwhile to outline a short reply.

Let's stipulate that the analysis could have been better. For example, it would have been useful for me to have better-defined, non-overlapping industry categories or data on the value of intermediate inputs. Above all, the ability of industry data at all to give insight into individual labor productivity is limited -- and they allude to the compositional issue in their blog post -- but, without some unit of analysis above the individual, measuring labor productivity is almost impossible. To the extent that we want to make any comparison at all between productivity and compensation, we need to accept certain trade-offs. This is one of them.

And the mistake I made in preparing the data, of course, is on me.

Yet I think Bivens and Mishel don't recognize that there is a good reason to look at nominal definitions of productivity and compensation. Notably, they misrepresent the analysis with an analogy to Zimbabwe's hyperinflation, saying that inflation invalidates my results. This is wrong.

My results are driven by relative changes in compensation and in productivity. This means that, had I deflated all my data by any measure of prices -- CPI, PCE, etc. -- it would not change my results.

What Bivens and Mishel do in their blog post, I would say, examines an different relationship than than I do, because they adjust productivity for industry-specific price indexes. So we reach two distinct conclusions that are both correct, which is easily missed in their write-up:

I show that there is a robust relationship between changes in the economic value of output produced per hour and changes in the hourly compensation of employees.
They show that there is no relationship between changes in the volume of output produced per hour and changes in the hourly compensation of employees.

The key difference, of course, is that I examined the value of output, and they examined the volume. Both results are meaningful. Together, they imply that, to a considerable extent, the economy adjusts to industries' different rates of productivity growth by changing the relative prices of output. You might think of how consumer technology is both much cheaper and produced more efficiently, than say, haircuts.

There is an underlying normative issue here as to whether workers should be compensated for the gains in the economic value of their output or in increases in the volume of it. To say that one analysis is "right" or "wrong" implicitly takes a position on the normative issue. I don't have any special insight on it.

Addendum: Brad DeLong and Mike Konczal have asked me to say how my interpretation has changed, and rightly so. Originally, I found a relationship between growth in labor productivity and in labor compensation strong enough to explain most (80%) of the variance across industries from 1987 to 2013. In my revised results, this drops to a third. If before I would have said that compensation growth is very well explained by productivity growth, now I think a reasonable view of my results is to say that it is a substantial contributor, but not by any means the full story.

Raising the Retirement Age

noreply@blogger.com (Evan Soltas) — Sun, 13 Sep 2015 17:03:00 +0000

New advancements in linking have made it possible, and indeed easy, to determine the year of birth, and almost the month of birth, for individuals surveyed in the Current Population Survey.

In an important effort, three researchers at the Minnesota Population Center have made it substantially easier to track individuals over time through the Current Population Survey. While economists have known this was possible for the better part of twenty years, it's been both imperfect and time-consuming. This new linking tool makes linking much better and dramatically easier.

It could spark a wave of new research in the social sciences. In fact, I've realized it's easy to determine almost exactly the month and year of birth for about half of individuals who participate in the Current Population Survey.

Survey participants show up in two sets of four consecutive months, so there are six opportunities for their ages to change going from one month to another. If you're, say, 21 in August but 22 in September, then you must have had your birthday between August and September. (This is where the technicality of "almost" the month of birth comes in, since the U.S. government does surveys in the middle of the month.)

So, what can we use this for? Here's one application: What is the effect of changes in the full retirement age for Social Security on the labor-force participation of the elderly?

Starting in 2000, that age rose from 65 to 66 in increments of two months. It will start rising again 2017. Given this two-month pattern, it has been challenging to get sharp estimates of the effect of the age increases. (The best work I know of on this topic comes from Giovanni Mastrobuoni.)

Yet this effort gets much easier when you know birth year and month more precisely. That's because you can look at the change in labor-force participation in the months right around the retirement age. What we should see, as the retirement age has increased, is a "moving bump" in the labor force participation rate -- the jump should be at 65 years precisely for some, 65 years and two months for the next year, and so on.

And it turns out we do see that: There's a 7-percent drop in labor force participation right at the month of the full retirement age that has shifted out with the statutory increases. Since the majority of Americans have already retired by their mid-60s, this means that the full retirement age knocks out a substantial share (about a quarter) of the remaining elderly workforce.

Raising the retirement age, then, would keep some older workers working. Whether we should do that is, of course, an entirely different question -- one for voters, not economists, to answer.

* * *

My dataset (15 MB) is available here.

China's Fish Story

noreply@blogger.com (Evan Soltas) — Wed, 09 Sep 2015 04:27:00 +0000

Even Li Keqiang, the Chinese prime minister, has said he doesn't trust his country's official GDP figures. To keep tabs on China's growth, Keqiang famously watches three numbers that are harder to fake: electricity consumption, railroad cargo traffic, and bank lending.

His skepticism echoed many similar comments from small industry of experts who try to estimate Chinese GDP growth from other data points. But why does China have a data-quality problem at all?

It's not a Chinese conspiracy to keep its growth secret. If that were the case, Keqiang would surely be among those in the know. Yet he is as in the dark about China's growth as you or I am. Instead, the story is one of weak political institutions.

To explain the fish story of China's GDP data, in fact, it helps to turn to an actual fish story -- the story about how, for most of the 1990s, China likely reported fake data on fishing in its territorial waters to the Food and Agriculture Organization of the United Nations.

Reg Watson and Daniel Pauly, who have spent their lives studying global fisheries, found back in 2001 that something strange was going on in Chinese waters: They had become vastly more productive over the last decade and were now far more productive than a statistical model, one that fit the rest of the world's fisheries well, said they should have been.

Where were all those extra fish coming from? China, Watson and Pauly concluded, was just making them up. They were overstating their annual catch by some 5 million tons, which was half of the official figures.

Usually fishing data is underreported, as fishermen conceal some of their catch and governments lowball with the numbers to meet quotas. Why, then, the overstatement? Watson and Pauly:

We believe that explanation lies in China's socialist economy, in which the state entities that monitor the economy are also given the task of increasing its output. Until recently, Chinese officials, at all levels, have tended to be promoted on the basis of production increases from their areas or production units. This practice, which originated with the founding of the People's Republic of China in 1949, became more widespread since the onset of agricultural reforms that freed the agricultural sector from state directives in the late 1970s.

This, to me, sounds much like what could be going on in Chinese GDP data today. Like Chinese officials promoted on fishy data about, well, fish, others are being judged on the basis of economic statistics: GDP growth, unemployment, and industrial output.

And we see the same pattern in Chinese GDP data as we did in the old fish data -- growth that looks both too strong and too smooth. It should be an important precedent for Christopher Balding and other skeptics who argue China might be overstating its GDP by more than 10 percent.

If China wants reliable economic data, it must stop judging its officials on their regions' economic track records and establish real independence for its statistical agencies. Letting officials control their own information and then judging them on it is a recipe for fraud and ignorance. And China's current crisis should teach its leaders that the costs of ignorance are far too high.

Wake Me Up When September Ends

noreply@blogger.com (Evan Soltas) — Sat, 05 Sep 2015 02:40:00 +0000

Does it matter exactly when the Federal Reserve raises its policy rate for the first time? Should it be September, December, or maybe even 2016?

Financial markets seem to think that call matters. And it's not just the overwhelming volume of analysis coming from the financial media. It's actually the markets themselves. Most notably, the ten-year yield has moved closely with expectations of the first rate hike as measured by futures contracts in the federal-funds market, where the Fed's target should matter most.

It's a bit surprising, then, to see economists argue that it shouldn't matter. Alan Blinder, for instance, wrote that the importance of the first hike had been "oversold." His argument is pretty reasonable. The choice is over 25 basis points higher or lower for several months at most. There's not a macroeconomic model out there which would tell you such a policy decision would have a significant economic impact. It's better to focus on the exit path over the longer term, Blinder and others have said, which really should matter.

Nevertheless, like Tim Duy, I think this perspective misses the mark. To quote Duy:

The lack of consensus regarding the timing of the first hike tells me that we don't fully understand the Fed's reaction function and, importantly, their confidence in their estimates of the natural rate of unemployment. The timing of the first hike will thus define that reaction function and thus send an important signal about the Fed's overall policy intentions.

Right. When the central bank has private information about its long-run target for the policy rate and gives a care about how its actions effect or disturb the bond market, as former governor Jeremy Stein showed in a recent paper with Adi Sunderam, it gets stuck in a bind.

Here's how the problem works, as per Stein and Sunderam. Say the central bank decides internally that its long-term target for the policy rate is too low. Because the central does not want to shock the bond market with a big change, it moves gradually. But markets aren't stupid. Understanding policy inertia, they infer from small moves in the short run what will happen in the longer run. As a result, the effort to avoid shocking the bond market doesn't work, essentially because a small hike today has more informational content about future hikes. The central bank becomes trapped by its own inertia rather than doing what it thinks would be best for the economy.

It's a brilliant insight, one that should be as influential as similar time-inconsistency arguments about inflation have been from Finn Kydland, Ed Prescott, and Ken Rogoff. And it poses an intellectual problem for the "wake me up when September ends" crowd, because a 25-basis-point hike in September isn't just a one-off decision, but one that is informative about years of potential actions down the road.

What, then, is the Fed to do about Stein and Sunderam's trap? They might recommend that the Fed should clarify that it really doesn't care if the bond market gets surprised every now and then. Another suggestion that comes out of their paper is that the Fed should try to water down its own forward guidance so that policy is less anchored by them and markets are less confident in future policy and, as a result, less susceptible to surprise. It could, for example, just kill the so-called "dot plot," in which FOMC members state their expectations for future interest rates. Or it could axe the primary dealer survey, in which the Fed's market-makers opine on what they expect the Fed will do.

But wait, you ask, doesn't the Fed publish its estimate of the long-run policy rate? How can that be private information? Yes, they do, and it's right here. However, the broader point of Stein and Sunderam's analysis is what matters. The Fed will always have some private information about its own reaction function. If it didn't, and markets knew how the Fed would react to every possible contingency, then there would be no market reaction to FOMC announcements -- and this is manifestly false. The only way the Fed could have a fully-public reaction function, I'd argue, is to commit to a Taylor rule or something of that spirit. It won't.

Stein and Sunderam's advice, of course, has to be weighed against the standard case for forward guidance from Michael Woodford, which contends that, when interest rates fall to zero, recovery today requires aggressive commitments about tomorrow. Some reconciliation of these two positions is necessary.

If the Fed chooses to wait, just about the worst thing it could do, Stein and Sunderam would say, would be to indicate in any way that this choice was informed by recent volatility in financial markets. And if it chooses to go ahead, the Fed will need to find some way to convince markets it's not kidding around when it says "data-dependent" -- a sales job, by the way, on which it totally gave up amid the taper.

Killing the dot plot or the primary dealer survey would send that message. It might talk a little about its reaction function instead.

Inequality and Productivity

noreply@blogger.com (Evan Soltas) — Thu, 03 Sep 2015 21:05:00 +0000

If you regularly read online about economics, you probably have seen this graph from Jared Bernstein and Larry Mishel so many times that you have lost count:

What it seems to say is that compensation had once kept pace with productivity but does not any longer -- in essence, that workers are getting a raw deal. And Mishel and Josh Bivens are back with a big, new update to the analysis. But there are some issues with the graph, despite the latest defenses.

Debating the divergence: some background

Let's review the debate. (My own contribution is towards the end of this post.) To an economist, the first red flag should be that this divergence looks nothing like the change in the labor share of income over the same period. The labor share was stable until about fifteen years ago, and since then has declined by about 10 percent, or about 6 percentage points. That translates to productivity having outgrown compensation by about 25 percent, not 140 percent as in the above graph.

So something is clearly strange here. That something, as James Sherk of the Heritage Institute has explained in a research memo, is a distinction in how productivity and compensation are adjusted for inflation. And that's a an important question -- why is the average price of consumption rising faster than the average price of output? -- but it's not at all an issue of workers not being compensated for rising productivity. It's a change in the terms of trade. Since workers buy a different basket of goods than businesses produce, the prices of those baskets can diverge.

Another key issue here is who counts as a worker: The above graph focuses only on compensation of production and non-supervisory workers. But supervisory workers are obviously part of the production process. While the graph certainly demonstrates inequality, in some vague sense, to compare the productivity of all workers against the compensation of a subset destroys the graph's ability to test the actual claim, which is that workers are not being compensated for their productivity.

What Mishel and others at the Economic Policy Institute have most recently argued, then, is that we should be focused on inequality within labor income. Even if, in an apples-to-apples comparison, mean labor compensation of all workers has kept up with productivity, median labor compensation hasn't. So what, they ask, explains the difference between mean and median labor compensation?

Mishel and Bivens's answer, in the latest study, is a "portfolio of intentional policy decisions" that have hamstrung labor. Capital deepening and broad gains in labor quality should lead us to believe that productivity has risen broadly, they say, and yet less than 10 percent of workers have seen their compensation keep up with productivity gains.

Inequality within labor compensation is a interesting place for the debate to have ended up. For one thing, it's where the academic conversation is: see, for example, Matt Rognlie's recent Brookings paper. Yet, on the other hand, it seems to make the key question -- are workers being compensated for rising productivity? -- intractable.

Why? Because it's hard to assess the productivity of individual workers. The whole debate over CEO pay, for example, has foundered on this issue. You can see that in Bivens and Mishel's recent paper on CEO pay, in which they concede that the evidence that CEO pay is untethered to productivity must be suggestive and circumstantial. In their latest, they again say it is hard to draw a link, and they're right.

It's easy to be pessimistic, then, about economists' ability to answer this productivity-compensation question. Mishel and Scott Winship at the Manhattan Institute have gone blue in the face, without any apparent resolution, arguing about how the productivity of the median worker has changed since the 1970s.

Another approach, then

It is important to concede up-front that there is, at the moment, no way to measure productivity at the individual level. But we can aggregate upwards, at least to the firm level, where there is a meaningful measure of output to be had. In this post, I'll use detailed industry-level data from the Bureau of Labor Statistics because firm-level data requires special clearances with the US government that I don't have.

The implication of this aggregation, however, is that I can't say anything about divergences between productivity and compensation within sectors. Which could be important. Critically, my results have no bearing on the debate about the very top of the income distribution; using sectors, I can only look at the body of the distribution. My industry breakdown, though, is reasonably granular: 246 industry categories.

With those caveats in mind, here's the big takeaway: Between 1987 and 2013, changes in sector-level labor productivity explain about a third of the changes in sector-level hourly labor compensation. And those productivity increases were paid as compensation to labor somewhat below the prevailing labor shares of income.

If you really want to know, 35 percent of the variance in the change between 1987 and 2013 in sector-level log hourly labor compensation is explained by changes in log labor productivity over the same period. A one-percentage point increase in productivity generated a 0.41-percentage-point increase in compensation. I've used 1987 and 2013 because this data was collected starting in 1987 and much of the data is still missing for 2014. As always, you can find my cleaned dataset here for your own analysis. [These results have been revised since the post went live. See below.]

We should ask if this result makes sense from a theoretical perspective. The key substance of labor economics has for decades circled around a key question: How true is the basic intuition that workers are paid their marginal product?

You might think it should be true. If workers aren't compensated for their productivity, it seems, they'll switch firms or industries. Yet there's a countervailing argument, associated with the economist William Baumol, which leads to the opposite result: If productivity rises more slowly in one industry than in others, its workers will demand wages in line with their opportunities in other industries -- and so, at the industry level, we shouldn't expect a link between productivity and compensation.

Whether the classical viewpoint or Baumol's is correct turns upon the strength of workers' "outside option" to exit industries where productivity growth is lagging and enter industries with faster productivity growth. If this outside option is weak, then workers' wages are determined by industry-level productivity; if this outside option is strong, then workers' wages are determined by the productivity of their best-alternative industry.

Mishel and Bivens, in their latest study, have argued that the industry-level approach doesn't make sense. Yet I don't think their critique really goes anywhere. (Read it yourself and be the judge.) Yes, measures of labor productivity reflect the average, not marginal, product of labor. Yes, workers in low-productivity industries could move to higher-productivity industries, so we can't say that low-paid workers are inherently and always unproductive.

Neither of these points, however, seems to have any bearing on the industry-level comparison. If labor productivity predicts labor compensation, then it seems fair to say that, at the industry level, workers have been compensated for their productivity gains.

My reading of the evidence, then, is distinctly different than that of Mishel and Bivens. We'd agree that, since 2000, the decline in the labor share of income is concerning. And we'd agree that some of the apparent divergence between compensation and productivity is attributable to changes in relative prices of consumption versus output, a phenomenon which isn't readily linked to inequality.

Where we differ is the extent to which changes in productivity explain changes in compensation. At least for the body of the income distribution, this evidence should lead us to explanations centered on productivity rather than on labor-market institutions.

Note: There was a data error in this post that has since been fixed. It weakens the conclusions relative to what was first posted, though there is still a robust relationship between productivity growth and compensation growth. Specifically, we go from an R2 of about 0.81 to 0.34 and a slope of nearly 1 to a slope of 0.41. See this post for more.

A Fact about China's Crash

noreply@blogger.com (Evan Soltas) — Tue, 01 Sep 2015 02:35:00 +0000

Imagine I took all the stocks in the Shanghai index on June 11, 2015, the height of the bubble in Chinese equities, and created 50 different investment portfolios.

Portfolio #1 would invest only in those stocks that had cumulatively performed worst since March, roughly when the boom began. Portfolio #2 would invest in stocks that had performed slightly better, and so on until I had broken the whole index into 50 portfolios, with the very last portfolio invested in the stocks that ran hottest in the boom.

The performance of those portfolio since June shows a clear pattern, and a very familiar one to students of America's housing bubble and bust. The bubbliest portfolios, and the bubbliest stocks have performed the worst amid the crash -- just like Las Vegas in 2005 versus in 2009.

What went up, in essence, is now coming down. This graph shows that fact.

Indeed, this "what goes up must come down" result is so strong that it explains a third of all cumulative returns of individual stocks, and virtually all of the cumulative returns of the portfolios, over this period. (This statistical performance compares favorably to most tests of the CAPM or the Fama-French 3-factor model on US equities.) About half of all the bubble-related gains, from this perspective, have been unwound.

One shouldn't infer directly from this that the crash is "good." But, if you thought that the boom was nuts, you might be relieved to know that the crash is quite focused on undoing the boom and isn't just dragging everything lower.

This post would not have been possible without help with Python from my friend Evan Chow. Evan helped me scrape Yahoo Finance for individual .csv files from the Shanghai index. I wrote a Stata program to merge the files together into a single panel dataset and completed the analysis.

How Are Economists Connected?

noreply@blogger.com (Evan Soltas) — Mon, 31 Aug 2015 04:27:00 +0000

The National Bureau of Economic Research, an organization of top economists that serves as a sort of clearinghouse for new research papers, counts nearly 1,400 members. Their interests vary widely, but upon joining the NBER, they sign up for research programs that represent their favored topics.

The NBER has 20 such programs, and economists usually sign up for one or two, although some sign up for more. (Twelve economists are signed up for five programs. Andrei Shleifer is the only one signed up for six.) As a result, we have over 700 connections between topics.

With so many members signing up for different combinations of programs, the NBER's member interest list gives a picture into the field. Not only can it tell us what fields are popular and unpopular, but also, it shows us what combinations are comparatively more or less common -- a window, perhaps, into the connections economists draw within their own field.

So I scraped the NBER's member list and got started. (As always, my data set is available here.)

The first metric I looked at was the correlation of registrations, as you can see in the matrix below. (Click it to enlarge.) You should interpret a significantly positive cell as "economists often put these topics together," a cell with a value near zero as "there are no strong connections between these two topics," and a significantly negative cell as "economists tend not to put these two topics together."

Some things immediately popped out at me. NBER members that were interested in monetary economics, for instance, also tend to be interested in economic fluctuations and growth. Those that are interested in the economics of education also tend to do work on labor economics. Both of those connections make a great deal of sense!

The areas where economists seem to pick and choose are also fascinating. Labor economists seem to dislike asset pricing. Those interested by economic fluctuations and growth stay away from education. And so on.

What Ails the American Startup?

noreply@blogger.com (Evan Soltas) — Fri, 28 Aug 2015 18:04:00 +0000

For all the hoopla about Silicon Valley, the data are clear: These are rough times to be a young business in America. In the early 1980s, about 12 percent of all firms were less than a year old. In 2012, however, only 8 percent were.

This raises a good question: What's going on? Why are new firms struggling to gain a foothold? Data from the Business Dynamics Statistics of the US Census offer an interesting answer: The problem isn't with the startups. It's with the economy in which they are starting up.

To reach that conclusion, though, we first need to learn a little bit about entrepreneurship in America. You've probably heard the factoid that 9 out 10 restaurants fail in their first year -- it's false, but never mind -- and actually, only about a quarter of all new firms go bust in their first year. Five years later, 45 percent of firms have survived. It's a pattern, technically called a "survival function," that has repeated itself since at least 1977, when the Census began collecting this data, as the next graph shows.

Let's take that survival function for granted, then, and focus on two specific phenomena. The first is a year-level effect: something that hits all firms in a given year the same amount, no matter when they were founded. The second is a cohort-level effect: something that hits firms founded in a given year the same amount and sticks permanently with that cohort of firms. (Economists: Scroll to the end of the post for the modeling details.)

You might think of the first as a cyclical or structural shock to the economy and the second as whether it was just a big or small "class" of new firms that year. Using the Census data, we can track the number of firms in each cohort for their first five years of existence, allowing us to disentangle the cohort and year effects. We can answer the question: Are the startups getting worse? Or is survival getting harder?

I find that about half of the decline in new firms from 1977 to 2012 can be ascribed to the year-level effect, and that there has been no average change in the cohort-level effect over the same period. The startups aren't that much worse, essentially, but the economy is much harsher towards them. With the same cohort strength but the prior economy, we would have about 200,000 more startups per year -- and about 700,000 more firms less than five years old. Since the US has about 5 million firms, that's a substantial change.

We can compare the actual decline to a counterfactual without the year-level effects:

Here are few more graphs to make sense of this. The first shows the cohort-level effect, and you should notice the lack of a down trend, but also the strong cyclicality, which shows the "smothered in the cradle" effect of recessions on new firm formation. High cohort effects can be thought of as years in which lots of startups launched successfully, whereas low cohort effects are bad years, with few successful launches.

The second shows the year-level effect, and you should notice the persistent down trend, indicating that, for any given firm, survival is becoming harder.

I've also taken the change in the year-level effect, so that we can see more clearly when survival has become harder. What we see, clearly, are two bloodbaths -- the 1980 and 2008 recessions -- and then a slow decline between them, without any obvious cyclicality.

There's a big takeaway here: The decline in new firms seems to be driven by changes that are making new firm survival more difficult in general, not just a decline in the cohort size itself.

* * *

Technical explanation

Let n_ft be the log number of firms founded in year f and alive in year t. I specify the model:

n_ft = b_f + b_t + b_t-f + e_ft,

where all the b terms are OLS coefficients and e is an error term. Then b_f can be thought of as a cohort-level effect, b_t as a year-level effect, and b_t-f as a survival function. Note that this isn't actually a survival model but rather more of a quick-and-dirty test with panel-data techniques, and if b_t increases year-over-year, the model doesn't make any sense. (Fortunately, this isn't a problem for our data set.)

My cleaned dataset is available here.

Who Is On the RUC?

noreply@blogger.com (Evan Soltas) — Tue, 02 Jun 2015 21:43:00 +0000

For the last year, I have been working to reconstruct the membership of the RUC, which is probably the most important policy entity in healthcare you've never heard of. The short of it is that RUC is a private organization with a critical public function: it advises the Centers for Medicare and Medicaid Services on how to set the relative prices for physician reimbursement within Medicare.

For example, it's the RUC's job to decide that, say, one treatment of a heart attack is equivalent in value to two treatments of pneumonia. It has come under extensive criticism -- see here, here, and here -- for basically being an unaccountable shadow government that acts in the interest of the American Medical Association and specialist doctors, rather than the medical community as a whole, patients, or the taxpayer. To be clear, I am repeating, not endorsing, that phrasing of the critique of RUC.

Initially, it was my intention, working with Judd Cramer, a friend and grad student at Princeton interested in labor economics, to try to link changes in the composition of the RUC to changes in Medicare's relative prices, known in health-policy circles as RVUs. But we never finished the project, mostly because I was overwhelmed with work this year -- I took a more-than-full load of classes and also wrote this research paper as independent work on the side.

Then the plan was to publish the list in an article with extensive commentary and discussion. In particular, I was very interested in potential conflicts of interest among RUC members, as prior work by Roy Poses has shown this to be a real problem. Yet, to do that, I really needed a complete and fully accurate membership list. That, as I have learned over the last few months, is basically impossible. RUC has been overseen by the AMA since 1991. It now has 32 seats, though it has expanded over the years. This means there are 736 person-years to account for. I could get all but 23 of them.

Over the last year, however, various health-policy researchers have found out that I have been working on this project -- and so I have an increasingly long list of people whom I've been telling to wait.

Yet I've decided that it's in the public interest for me just to publish the list already. (It's the document at the top of this post.) I do so with two honest caveats. First, it's incomplete. I'm missing a handful of years for certain seats, as my efforts to track down some person-years failed. Second, there are probably some inaccuracies. I do not think it is ridden with errors, but I would frankly be surprised if I got everything right. That's just the nature of trying to research a body that has made an extraordinary effort to remain cloaked in secrecy. (The type of error that I think is most likely is that I got some of the years wrong. I think all the names are correct; I am pretty sure anyone I claim was on RUC was in fact on RUC, for approximately the period I say they were. My guess is that I will be off by a year, say, for 10 percent of the people.)

Here is how I put this list together: dozens of hours of archival research. First, I managed to track down old AMA Board of Trustees reports. Those sometimes contained RUC appointments. Second, the medical-specialty newspapers and journals often mention who is currently serving on the RUC on the specialty's behalf. Third, the résumés and websites of ex-RUC doctors often list their full years of service; sometimes you can also find these in articles for the medical-specialty publications when they retire. Fourth, the AMA recently began publishing the current membership as part of an (admirable, but highly incomplete) effort towards transparency. Fifth, I relied on other efforts that Roy Poses and Brian Klepper, among others, have made, to identify RUC members.

I will also try to release some of the related research that I have done on RUC in the coming days. It was past time for me, however, to share this document. Thank you to the many who helped or cheered along this project.

SNAP and Food Security

noreply@blogger.com (Evan Soltas) — Tue, 02 Jun 2015 15:55:00 +0000

"SNAP and Food Security: Evidence from Terminations" is the title of my first-ever working paper, which I wrote for my junior-year independent work at Princeton. What I do in the paper is try to measure very carefully the effect of participating in SNAP on households' food security, and the basic idea of how I do that is pretty simple:

[C]onsider two similar groups of households. The first group receives SNAP benefits in both November and December of a given year. The second group receives SNAP benefits in November but not in December. The difference in December food security between the two groups provides an intuitive estimate of the effect of SNAP benefits on food security in December.

With that kind of comparison in mind, here's what I find:

SNAP participation increases the probability of food security by 10 percentage points (22 percent), with gains concentrated in reducing the probability of extreme food insecurity by 8 percentage points (36 percent), an effect that is broadly comparable to that of a change in household income from $10,000 to $20,000.

Naturally, there's a whole lot more in the paper itself.