Author: Jeremy Horpedahl

Big Win for Prediction Markets

Jeremy Horpedahl1 Comment

Last night was a big win for Trump, but it was also a big win for prediction markets. In January 2024, I suggested that one of the best ways to follow the election was by following prediction markets. That prediction turned out to be correct!

Before any polls had closed, prediction markets had Trump with about 60% odds of winning. That’s far from a sure thing, but it’s much better than many prediction models, which all had the race as basically a 50-50 toss-up with a very slight edge to Harris (though one simple model that I wrote about two weeks ago had Harris slightly losing the popular vote, a good call in hindsight). So going into the election results, you would have been more confident that a Trump win was a real possibility if you watched predictions markets

Last night after the results started coming in, the average over five different prediction markets from Election Betting Odds put Trump at over 90% odds by 11:00pm Eastern Time. By about 12:45am, he was already over 95%. These aren’t absolutely certain odds, but if you were watching the election night news coverage, they were still treating this as essentially a toss-up in the battleground states.

The Associated Press hadn’t even called Georgia, the second of the battleground states, by the time prediction markets were over 95% for the overall race! Decision Desk HQ, which is a very good source for calling races in real time, didn’t declare Trump the winner until 1:21am, when they called Pennsylvania (they also have a nice explanation of how they made the call). The AP didn’t declare Trump the winner until 5:34am, when they called Wisconsin.

Polymarket is the largest of the five markets in the Election Betting Odds average, and they are also a good source because they have markets for all of the battleground states (here’s the market for Michigan, which still hasn’t been called as of 11:30am on Wednesday by most news sources!). This table shows when the 90% and 95% thresholds were permanently crossed on Polymarket odds for each of the 5 early battleground states, in comparison with the DDHQ and the AP.

Notice that the 90% threshold consistently beats DDHQ by at least an hour (the one exception is North Carolina, where DDHQ called it very early — they are very good at what they do!). And the 90% threshold is consistently beating the AP by at least 3 hours.

None of this should be read as a criticism of the Associated Press. They should be cautious about predictions! But if you want to know things fast (or, before your bedtime in this case), prediction markets are clearly worth following.

How can prediction markets be so far ahead of media sources? Because there is a strong incentive to be right early: that’s how you make money in these markets! How exactly this is done is unclear, since the traders are all anonymous and we generally can’t ask them. But likely they are doing a similar analysis of counties results compared to the 2020 election, as DDHQ told us they did after the fact, just quicker (indeed, if you were watching news coverage, they were doing the same thing, just in an ad hoc way, and much more slowly).

Federal Spending in 2024 was $2.3 Trillion More Than 2019

Jeremy HorpedahlLeave a comment

In Fiscal Year 2019, the US federal government spent $4.45 trillion dollars. In Fiscal Year 2024, spending was $6.75 trillion, or an increase of $2.3 trillion dollars. If you adjusted the 2019 number for inflation with the CPI, it would only be about $1 trillion more. Where did that additional $2.3 trillion go?

It will probably not surprise you that most of the increase in spending went to the largest categories of spending. Historically these have been health, Social Security, and defense, but now we must also include interest spending (roughly equal in size to defense and Medicaid in 2024). Indeed, with these areas of spending, 72 percent of the increase is accounted for. Add in the next three functions, and we’ve already accounted for over 90 percent of the increase.

Importantly, most of these categories are outside of the annual federal budget process, meaning that Congress does not need to approve new spending each year (Congress could change them, just as it could change any law, but it’s not part of the annual budgeting process). The “mandatory” categories, as they are called in federal law, are shaded red. I’ve striped with red and blue the health and income security functions, because some of this is subject to the annual budget process, but most of it is not. For example, Medicaid is not subject to the budget process (biggest part of the “health” function) and SNAP is not subject to the budget process (a big part of income security — it is set by the Farm Bill, usually on a five-year cycle).

So, when we talk about the $2 trillion increase since 2019, or the roughly $2 trillion cuts that would be needed to balance the budget, keep in mind that most of this is not subject to the annual budget process. It would require Congress to consider them specifically to enact cuts — though some big categories, such as Social Security, would be automatically cut under current law once their trust funds are exhausted (coming up on about a decade for the Social Security Old-Age Trust Fund).

A Simple Presidential Election Model, Using Three Economic Variables

Jeremy Horpedahl2 Comments

As the presidential race finishes out the last two weeks, it’s clearly a close race. In the past I have recommended prediction markets, and right now these are giving Trump about 60% odds. There have been lately a few big bettors coming into the markets and primarily betting on Trump, so there has been speculation of manipulation, but even at 60-40 the race is pretty close to a toss-up.

Another tool many use to follow the election are prediction models, which usually incorporate polling data plus other information (such as economic conditions or even prediction markets themselves). One of the more well-known prediction models is from Nate Silver, who right now has the race pretty close to 50-50 (Trump is slightly ahead and has been rising recently).

But Silver’s model, and many like it, is likely very complicated and we don’t know what’s actually going into it (mostly polls, and he does tell us the relative importance of each, but the exact model is his trade secret). I think those models are useful and interesting to watch, but I actually prefer a much simpler model: Ray Fair’s President and House Vote-Share Models.

The model is simple and totally transparent. It uses just three variables, all of which come from the BEA GDP report, and focuses on economic growth and inflation (there are some dummy variables for things like incumbency advantage). Ray Fair even gives you a version of the model online, which you can play with yourself. Because the model uses data from the GDP report, we still have one more quarter of data (releasing next week), and there may be revisions to the data. So you can play with it (and one of the variables uses the 3 most recent quarters of growth), but mostly these numbers won’t change very much.

What does the model tell us?

Continue reading →

Where Are The 7 Million Missing Men?

Jeremy HorpedahlLeave a comment

You may have heard that there are roughly 7 million men of working age that are not currently in the labor force — that is, not currently working or looking for work. The statistic has been produced in various ways using slightly different definitions by different researchers, but the most well-known is from Nicholas Eberstadt who uses the age cohort of 25-55 years old and gets about 7 million (in 2015). More recently and perhaps more prominently is from Senator JD Vance, and as with almost all issues he has tied this to illegal immigration.

The 7-million-men statistic is true enough, and if we limit it to native-born American men with native-born parents (I assume this is the group Vance is concerned about), we can get right at 7 million non-working men in 2024 by expanding the age cohort slightly to 20-55 year olds.

Why are these men not working? According to what they report in the CPS ASEC, here are the reasons broken down by 5-year age cohort (I drop 55-year-olds here to keep the group sizes equal, which shrinks the total to 6.7 million men):

By far the largest reason given for not working is illness or disability, which is 42% of the total for all of these men, the largest reason for every age group except 20-24 (who are mostly in school if they aren’t working), and it’s the majority for workers ages 30-54 (about 56% of them report illness or disability). Slightly less than 10% report “could not find work” as the reason they weren’t working, which is about 650,000 men in this age group (and are native-born with native-born parents). And over half of those reporting that they couldn’t find work are under age 30 — for those ages 30-54, it’s only about 7% of the total.

More men report that they are taking care of the home/family (800,000) than report not being able to find work (650,000). And a lot more report that they are currently in school — almost 1.5 million, and even though they are mostly concentrated among 20–24-year-olds, about one-third of them are 25 or older.

It’s certainly true that the number of working age men in the labor force has fallen over time. In 1968, 97% of men ages 20-54 had worked at some point in the past 12 months (that’s for all men regardless of nativity, which isn’t available back that far in the CPS ASEC). In 2024, that was down to about 87%. But even if we could wave a magical wand and cure all of the men that are ill or disabled, this would add less than 3 million people to the labor force, not nearly enough to make up for all of the immigrants that Vance and others are suggesting have taken the jobs of native-born Americans.

What Are the Effects of TCJA? It’s A Little Hard to Say

Jeremy HorpedahlLeave a comment

The Tax Cuts and Jobs Act was passed in late 2017 and went into effect in 2018. For academic research to analyze the effects, that’s still a very recent change, which can make analyzing the effects challenging. In this case the challenge is especially important because major portions of the Act will expire at the end of next year, and there will be a major political debate about renewing portions of it in 2025.

Despite these challenges, a recent Journal of Economic Perspectives article does an excellent job of summarizing what we know about the effects so far. In “Sweeping Changes and an Uncertain Legacy: The Tax Cuts and Jobs Act of 2017,” the authors Gale, Hoopes, and Pomerleau first point out some of the obvious effects:

  1. TCJA increased budget deficits (i.e., it did not “pay for itself”)
  2. Most Americans got a tax cut (around 80%), which explains #1 — and only about 5% of Americans saw a tax increase (~15% weren’t affected either way)
  3. Following from #2, every quintile of income saw their after-tax income increase, though the benefits were heavily skewed towards the top of the distribution ($1,600 average increase, but $7,600 for the top quintile, and almost $200,000 for the top 0.1%)

Beyond these headline effects, it seems that most of the other effects were modest or difficult to estimate — especially given the economic disruptions of 2020 related to the pandemic.

For example, what about business investment? Through both lowering tax rates for corporations and changing some rules about deductions of expenses, we might have expected a boom in business investment (it was also stated goal of some proponents of the law). Many studies have tried to examine the potential impact, and the authors group these studies into three buckets: macro-simulations, comparisons of aggregate data, and using micro-data across industries (to better get at causation).

In general, the authors of this paper don’t find much convincing evidence that there was a boom in business investment. The investment share of GDP didn’t grow much compared to before the law, and other countries saw more growth in investment as a share of GDP. Could that be because GDP is larger, even though the share of investment hasn’t grown? Probably not, as GDP in the US is perhaps 1 percent larger than without the law — that’s not nothing, but it’s not a huge boom (and that’s not 1 percent per year higher growth, it’s just 1 percent).

Ultimately though, it is hard to say what the correct counterfactual would be for business investment, even with synthetic control analyses (the authors discuss a few synthetic control studies on pages 21-22, but they aren’t convinced).

What’s important about some of the main effects is that these were largely predictable, at least by economists. The authors point to a 2017 Clark Center poll of leading economists. Almost no economists thought GDP would be “substantially higher” from the tax changes, and economists were extremely certain that it would increase the level of federal debt (no one disagreed and only a few were uncertain).

Consumer Expenditures in 2023

Jeremy Horpedahl2 Comments

Today BLS released the annual update to the Consumer Expenditure Survey, which is exactly what it sounds like: a survey of US consumers about what their spending. The sample size is “20,000 independent interview surveys and 11,000 independent diary surveys” so it’s a pretty big sample. And this is a really great data source, because versions of it go back over 100 years (though the current, annual survey with a lot of detail starts in 1984).

What does this new data tell us? One area that has received a lot of attention lately is food spending (including a lot of attention on this blog), especially the cost of groceries. According to the CPI food at home index, grocery prices are up almost 26 percent since the beginning over 2020. That’s a lot! But incomes are up too, so how does this affect spending patterns?

Here’s what food and grocery spending for middle-quintile households looks like:

Compared to the pre-pandemic 2019 levels, consumers are spending slightly less of their income on food (12.7% vs. 13.2%), though a slightly larger share of their income is being spent on groceries (8.1% vs. 7.8%). Those changes are noticeable, though this isn’t the radical realignment of spending patterns you might expect from such a big change in food prices. The reason is clear: while grocery prices are up about 26%, middle-quintile incomes are up a similar 25% since 2019. That’s falling behind a little bit, but incomes have roughly kept pace with rising food prices. And from 2022 to 2023, both of these percentages decreased slightly, by about 0.3 percentage points.

Continue reading →

The Top 1 Percent Paid a Lot of Taxes in 2021

Jeremy HorpedahlLeave a comment

In 2021 the top 1 percent of taxpayers in the United States paid 36 percent of all federal taxes (they have 21.1 percent of income). This figure had been below 20 percent until the mid-1990s, and as recently as 2019 it was just 24.7 percent (they had 15.9 percent of the income that year).

The data comes from the latest CBO report on “The Distribution of Household Income in 2021.”

The increase is primarily due to a large number of high-income households realizing capital gains in 2021. With all the talk lately of potentially taxing unrealized capital gains, it’s important to note that we do tax realized gains, and these change a lot from year to year. Another contributing factor is that the share of the bottom 60 percent of households only paid 1 percent of federal taxes in 2021, a big drop from 2019 due to a big increase in temporary refundable tax credits.

Better Off Than 4 Years Ago? Median Family Income Edition

Jeremy Horpedahl1 Comment

Are you better off than you were four years ago? That question was asked at the Presidential debate last night. But more importantly, we also got a massive amount of new data on income and poverty from Census yesterday. That data allows us to make that just that comparison, although somewhat imperfectly.

The Census data is excellent and detailed, but it’s annual data, meaning that the release yesterday only goes through 2023. We won’t have 2024 data for another year. Such is the nature of good data. (Note: I’ve tried to address this same question with more real-time data, such as average wages). Still, it’s a useful comparison to make. It’s especially useful right now because the new 2023 data on income are (for most categories) the highest ever with one exception: 4 years ago, in 2019.

A reasonable read of the data on income (whether we use households, families, or persons) is that in 2023 the median American was no better off than in 2019, after adjusting for inflation. In fact, they were probably slightly worse off. I fully expect this will no longer be true when we have 2024 data: it will certainly be above 4 years prior (2020) and likely above 2019 too (more on this below). But we can’t say that for sure right now.

So let’s do a comparison of “are you better off than 4 years ago” for recent Presidents that were up for reelection (treating 2024 as a reelection year for Biden-Harris too), using the 4-year comparison that would have been available at the time using real median family income. Notice that this data would be off by one year, but it’s what would have been known at the time of the election.

Continue reading →

The Cumulative Effect of Small Changes in Economic Growth

Jeremy HorpedahlLeave a comment

A recent post from the blogger (Substacker?) Cremieux called Rich Country, Poor Country showed how small differences in economic growth add up over time. Because he used nominal GDP growth rates, I don’t think that post is exactly the right way to analyze the question, but I still think it’s a very important one. So in this post I will offer, not necessarily a critique of that post, but perhaps a better way of looking at the data.

For the data, I will use the Maddison Project Database, which attempts to create comparable GDP per capita estimates for countries going back as far as possible… for some, back thousands of years, but for most countries at least the last 100 years. And the estimates are stated in modern, purchasing power adjusted dollars, so they should be roughly comparable over time (if you think these estimates are a bit ambitious, please note that they are scaled back significantly from Angus Maddison’s original data, which had an estimate for every country going back to the year 1 AD). The most recent year in the data is currently 2022, so if I slip up in this post and say “today,” I mean 2022, or roughly today in the long sweep of history.

Like Cremieux’s post, I am interested in how much slightly lower economic growth rates can add up over time. Or even not so slightly lower growth rates, like 1 percentage point less per year — this is a huge number, because the compound annual average growth rate for the US from 1800 to 2022 is 1.42%. So let’s look at the data way back to 1800 (the first year the MPD gives us continuous annual estimates for the US) to see how changes in growth rates affect long-term growth.

It probably won’t surprise you that if our 1.42% growth rate had been 1 percentage point lower, the US would be much poorer today, but to put a precise number on it, we would be about where Bolivia is today (that is, ranked 116th out of the 169 countries in the MP Database). Note: I’m using a logarithmic scale, both so it’s easier to see the differences and because this is standard for showing long-run growth rates.

What is very interesting, I think, is that if our growth rate had been just 0.25 percentage points lower per year since 1800, we would be about where Spain is. Now, Spain is certainly a fine, modern developed country (they rank 34th of the 169 MPD countries). But Spain’s growth has not been spectacular lately. Average income in Spain is almost half of the US today (purchasing power adjusted!), which is another way to say that just 0.25 percentage points lower over 222 years reduces your growth rate by half.

That’s the power of economic growth.

And if our growth rate had been 0.5 percentage points lower, we’d be about where the big former Communist countries are today (both China and the former countries of the USSR are about equal today — about 1/3 of the income of the US).

What if we perform the same analysis for a shorter time horizon? If we go back 50 years to 1972, the effects are not quite as dramatic, but still visible.

Our cumulative annual growth rate since 1972 has been a bit higher than the long-run average, around 1.68%. Under these four alternative growth scenarios since 1972, the comparable countries don’t sound so bad. It probably wouldn’t be a huge deal if we were only at Australia’s level, losing just about a decade of economic growth. But it would be a huge failure if we were only at Italy’s current level of development. Under that 1 percentage point lower growth scenario, we would have had no net growth since about year 2000, which has roughly been the case for Italy.

All of these alternative scenarios show the power of economic growth to add up over time, but they do so in pessimistic way: what if growth had been slower. What if we look at the opposite: what if growth had been faster over some time horizon. Sticking with the 1972 medium-run example, if real growth rates had been 1 percentage point higher, our income today would be almost double what it actually is, about $95,000 compared with the current $58,000 (the MPD data is stated in 2011 dollars, so that sounds lower than it actually is now: over $80,000).

What if we went back even further? If our economic growth rate since 1800 had been 1 percentage point higher every year, our average income in 2022 would be an astonishing $517,000 — almost 10 times what it actually was in 2022. That’s a dizzying number to think about, and maybe that’s not a realistic alternative scenario.

But what if it had only been 0.25 percentage points higher since 1800 — that probably is a world that was possible. In that case, GDP per capita would be about double what it actually was in 2022, at over $100,000 (again, stated in 2011 dollars).