More on housing and startups

Image source

People are starting to notice the epic collapse of startup activity of recent years. Via Arnold Kling, see this great note by Glenn Reynolds and a reader suggesting that the decline in housing collateral could be a large factor.

I think so too, as I've discussed before. I'm working on a paper that I hope will shed some light on this question. A few things to keep in mind:

1. Part of the decline is secular; I noted this here. This has coincided with a more general secular decline in business dynamism, and we still don't know what's driving it. The startup problem seems to matter for the broader dynamism decline, though. It's difficult to disentangle the secular component from the Great Recession component.

2. I note here that (a) national house price indices and home equity peak about the same time as startup activity (2006, before the "recession" started) and (b) residential investment peaks about that time as well despite other investment series peaking at least a year later. That's far from a smoking gun, but it is suggestive.

3. More formal empirical evidence for this link is emerging. I discussed one paper here (this one exploits the famous Saiz housing supply elasticity instrument). Another paper, this one exploiting time series and regional variation, obtains similar results. Both of these papers cast doubt on the notion that the Mian and Sufi channel (the household balance sheet channel) is sufficient for understanding the full consequences of house prices (in part because the two papers I mentioned find effects in tradeables in addition to nontradeables).

4. The full details of a housing collateral/startup channel require some unpacking. For example: you could build a really simple model with frictionless housing markets and housing collateral that would not give you a clear house price/startup relationship. To see this, suppose housing and nondurable consumption enter into utility as Cobb-Douglas, so expenditure shares are fixed. Then a house price decline just causes people to buy more houses. You need something more; lots of housing market frictions (which is reasonable) or a simultaneous decline in loan-to-value ratios will probably do it.

5. It would be nice to be able to quantitatively compare the consequences of the main channels through which housing collapse smashed the economy. These include this housing startup channel; the Mian and Sufi consumption channel; the standard residential investment/construction industry channel; and the bank balance sheet channel. Someone should write a paper about this... (working on it).

6. Figuring out the cause of the startup collapse is important since startups account for almost all net job creation.

Crazy prices on Deep Space Nine

Image source

I broadly agree with Matthew Yglesias about Star Trek. But his post reminded me of a complicated aspect of the series: its economy. Particularly when watching Deep Space Nine, it can be difficult to reconcile the apparent plenty provided by replicator technology with the "profits" obsession of the Ferengi. The Star Trek universe has a currency--typically gold-pressed latinum--and it appears to have value and uses even in an environment with little scarcity.

Also, some of the prices don't make any sense.

Memory Alpha provides this discussion of the currency with examples of prices mentioned during the series. I'm going to convert some of these prices to dollar values by using a very lucky mention--wages. "Quark pays his Bajoran employees one slip of latinum a day during the Cardassian Occupation." We can probably assume that these are low-skilled wages, and we can use what we know about low-skill wages in dollars to build exchange rates.

We're also given a set of conversion rates between latinum denominations: 1 bar = 20 strips = 2000 slips.

I don't know what sort of labor market is supposed to have existed during the Cardassian Occupation; maybe all employers had monopsony power in labor markets, or maybe labor was scarce. I'll try three different specifications:

• Low wage conversion assumption: $1/hour or $8/day
• Minimum wage conversion assumption: $7.25/hour or $58/day
• High wage conversion assumption: $20/hour or $160/day

Of course, they may not be working 8-hour days, but I think these specifications cover reasonable scenarios. Now consider a few of the mentions of latinum and how they convert to dollars:

	Slips	Low wage conversion ($)	Minimum wage conversion ($)	High wage conversion ($)
Crate of root beer	10	80	580	1,600
Pajamas	300	2,400	17,400	48,000
Cadet's uniform	500	4,000	29,000	80,000
Dress	1,700	13,600	98,600	272,000
Wreckage of a ship	6,000	48,000	348,000	960,000
Nog's life savings	10,000	80,000	580,000	1,600,000
Quark's wager on Sisko vs. Q fight	10,000	80,000	580,000	1,600,000
A day's revenue at Quark's	10,000	80,000	580,000	1,600,000
Morica Bilby, shipping consultant, weekly wages	10,000	80,000	580,000	1,600,000
Someone's bar tab at Quark's	44,000	352,000	2,552,000	7,040,000
2,000 tons of Kohlanese barley	378,000	3,024,000	21,924,000	60,480,000
Quark's evacuation stash	1,200,000	9,600,000	69,600,000	192,000,000
Offer to buy Quark's bar	10,000,000	80,000,000	580,000,000	1,600,000,000

These are some pretty startling numbers. A cadet's uniform costs between $4,000 and $80,000. A dress is between $13,600 and $272,000. Quark is a very wealthy man (or else has a serious gambling problem); he wagers half a million dollars on a fight and keeps tens of millions in cash under his bed for emergencies.

I would say that these prices are pretty inconsistent. Note that this observation does not depend on my dollar conversion choices; just look at the "Slips" column and observe that pajamas cost 300 days of wages in the food service industry. Maybe the post-scarcity economy leads to strange preferences and relative prices. Or maybe the writers didn't think very hard about latinum mentions.

It almost makes Star Trek seem unrealistic!

Employment services and misclassification

Lately there has been some talk about temporary help services (see here and here). This industry, and the industry of employment services more generally, is interesting not only for its potential business cycle implications but also for its economic measurement implications.

In Census and BLS data, "employment services" is an industry category (4-digit NAICS 5613) that includes job placement services, temp agencies, and other services that allow businesses to outsource HR and other tasks. We have seen some interesting activity in employment services during the last 20 years (click for larger image):

Here I've plotted "employment services" employment as a share of "professional and business services" employment (red line). Observe that this ratio has risen by almost 5 percentage points since 1990. Since readers may know that services generally have made huge gains in employment during this time, I also provide "employment services" employment as a share of total private nonfarm employment (blue line).

What interests me is the fact that a lot of employees in this industry are misclassified by industry codes. People on the payrolls of temp agencies could actually be working in any industry. This may become a measurement problem if employment services resume their gains of recent decades; to the extent that these workers are misclassified, US data overstate the number of workers in these narrow services industries and understate the number of workers for the industries in which temporary employees are working.

Consider an example. If I have a manufacturing plant with a big HR department, but I decide to close the HR shop and pay an HR services firm to do that work, very little has actually changed in the industry composition of the US economy--but the data will record that the manufacturing sector shrank and the services sector grew.

Consider another example. Suppose a change occurs among retailers that makes them want to fill existing jobs with temporary, rather than permanent, workers, and they do this by contracting with temp agencies. Again, the actual industry composition of employment hasn't changed, but the data will indicate a smaller retail sector and a larger services sector.

Something to keep in mind for those watching the evolution of the US industry composition.

How to think like Eric Falkenstein

I've become less enamored with trying to change opinions, because ideas need a zeitgeist, and if that's not fertile nothing you say will matter. . . . I don't see lot of value to being an advocate, though I know someone needs to sit down and write thoughtful things to counteract all the instinctive first-order solutions people think are great ideas (poor? give them money!). The problem is that it's hard not to become a partisan hack if you write too much, to pick on the other side's worst arguments, which regardless of what side you are on, will be indefensible and so prove nothing.

The original post is here. This, along with the fact that the number of political opinions I have has been on a steady decline, is why I don't post about politics anymore.

Inside the GDP sausage factory

Image source

It's useful to be aware of the difficulty of measuring the US economy. Those interested in today's advance GDP report might do well to look through this, a handbook about the concepts and methods behind US income and product accounting. GDP estimates are constructed from a combination of many datasets, most of which are produced by the Census Bureau but some of which come from BLS, BEA, the Department of Agriculture, Treasury, IRS, OMB, and state governments. Putting everything together is quite a task. "The source data available to BEA are not always ideal for the preparation of the NIPAs" (3-2).

The most reliable GDP estimates are based on the Economic Census, which occurs every five years. Between censuses, statisticians must rely on surveys that have sampling properties chosen based on, well, the most recent Economic Census (and the Business Register, which forms the backbone of many business datasets). Advance estimates are basically built entirely on survey data.

Today we see the advance estimate for 2013Q1 GDP. Here's what the handbook says:

For most of the product-side components, the [advance] estimate is based on source data for either 2 or 3 months of the quarter. In most cases, however, the source data for the second and third months of the quarter are subject to revision by the issuing agencies. Where source data are not available, the estimate is based primarily on BEA projections. (3-7)

The components for which only 2 months of data are typically available include several categories of construction, inventories, exports, and imports. Missing data have to be filled in somehow, and the solution will probably be something based on trends--so it may sometimes be difficult for advance estimates to catch turning points. Also, a rough rule of thumb might be that data frequency and data quality are negatively related (not to mention that survey quality may decline as time since the last Economic Census increases). In short, advance estimates require a lot of guesswork.

And this is to say nothing of the microdata. Survey microdata can be pretty nasty.

The people at BEA have a pretty tough job. It is therefore not surprising that GDP sometimes receives pretty big revisions. Advance data should probably be taken with a grain of salt.

Reinhart/Rogoff and policy outcomes: Let's be careful about drawing causal inference

By now everyone knows about the Great Reinhart and Rogoff Implosion of 2013. If you don't, read this. What has most amused me is that many journalists seem to think that the R&R 90% threshold had a nontrivial, causal impact on US fiscal policy. Consider the following headlines:

• The Excel depression
• How an Excel Error fueled panic over the federal debt
• FAQ: Reinhart, Rogoff, and the Excel error that changed history

One journalist, Tim Fernholz, set out to answer the question, "How influential was the Rogoff-Reinhart study warning that high debt kills growth?" (here). But he does not answer that question, at least if "influence" is meant to refer to policy outcomes. The question he does answer is, "Are there examples of policymakers citing R&R in support of fiscal restraint?" The answer to that question, as Fernholz shows, is "yes." Fair enough.

If we operate from the assumption that policymakers are highly amenable to evidence; and if we further assume that a descriptive (not causal) result from a single empirical study is enough to drive a policymaker's decisions, then maybe a story about policymakers discussing R&R is enough to show that it influenced actual policies. But I think those are bad assumptions. Here's my Twitter conversation with Fernholz:

First off, he's a nice guy--some journalists don't respond to critical Twitter questions from nobodies. As I understand it, Fernholz believes that R&R90% affected the opinions of marginal, outcome-influencing lawmakers enough to change their vote. There is a counterfactual world in which R&R did the work more acceptably and fiscal policy outcomes in the United States are more accommodative.

Call me skeptical. Chalk it up to my cynicism about the degree to which politicians care about evidence. In any case, if they cared that much about evidence, they might have asked someone whether the R&R90% result could reasonably have been considered to be a causal point estimate. I think any responsible "wonk" would have told them they need more data and theory before changing their mind. In any case, the fact that a few policymakers, like Paul Ryan, mentioned R&R90% in support of their policy preferences is not sufficient evidence that the result drove policy. It's just as likely--some might say more likely--that policymakers have preexisting preferences about policy which they justify by grabbing whatever research results seem to support them.

This is slightly ironic since a reasonable criticism of R&R90% before we knew it was bogus was that there was no causal story and no attempt to tease one out of the data. We have to be careful about touting descriptive results as causal. Let's apply that standard to our attempts to blame Reinhart and Rogoff for whatever fiscal policies happened that we don't like.

Bottom line: If this matters, I think a really useful thing for a journalist to do would be to back up assertions of a causal chain from R&R90% to US fiscal policy outcomes with a story about the lawmakers who had the marginal vote and were swayed by the research. Fernholz seems to have a headstart since he already knows of some legislators who were swayed by the R&R claims.

The housing crisis was regional

I think a lot of people sometimes think of housing as a national market. It is, in some ways. But different regions experienced substantially different house price dynamics prior to and during the Great Recession.

Figure 1 plots the FHFA house price index for North Dakota (click for larger image). Observe that this state experienced no housing bust. Several other states (DC, Oklahoma, Texas, and West Virginia) experienced more of a plateau than a decline, though they still show a slight peak and trough pattern.

Figure 1

Among states that did experience declines in house prices, there is some heterogeneity in the timing of peak and trough as well as the depth of the decline. Figure 2 shows the distribution of states over peak timing; each bar indicates how many state housing markets peaked during the indicated quarter (click for larger image). I include all states but North Dakota.

Figure 2

Observe that most states peaked in the middle of 2007; interestingly, the national Case-Shiller index peaked in early 2006. The states peaking in 2006 include Nevada, Arizona, California, and Florida--the "sand states" that experienced the largest booms and busts led the national market by a year.

Figure 3 shows the distribution of states over trough timing--the timing of the state housing bottom (click for larger image). The data end in 2012q4, so some states may still fall further; but observe that no states were at a bottom at the end of the sample.

Figure 3

Many states bottomed in early 2011, and most of the rest bottomed a year later.

The biggest differences among states were the depth of the housing market decline. Figure 4 shows peak-to-trough house price declines as a percent of the index peak, where peak and trough are defined state by state (click for larger image).

Figure 4

By the FHFA index, Nevada saw a stunning 60 percent decline in house prices, with the other sand states not far behind at around 50 percent. On the other hand, thirteen states experienced declines of less than 10 percent.

The housing collapse was a few states--accounting for a nontrivial portion of the national housing market--beginning to dive in mid-2006, with most other states following a full year later. The early states saw the largest price declines. The price collapse ended for most states in 2011, with real growth just now starting to return. This variation in both the timing and the depth of the crisis is a reminder that housing markets are regional, but the fact that nearly every state experienced some degree of decline is telling.