Here’s Autumn again, carrying upon its still breeze a very special flavor of stagnation. With an ineffectiveness that has become emblematic, the US government has shut down. The world stands perplexed. Although large political events can often go unnoticed in daily life, the government shutdown does not seem to fall in that category. It has caused quite a few very visible ripple effects.
Aside from the fact that government-employed friends of mine are furloughed, my brother Andrew was kicked out of Yosemite this month. Rangers announced the government shutdown and gave everyone 2 days to vacate the park, which led to a scramble of many rock climbers to get final routes in before the park closed (or, for the more anarchistic ones, to make sure they were up on a wall and thus unable to be kicked out when the shutdown commenced). The always-booked-to-the-brim famous climber campsite, Camp 4, became eerily empty. Andrew told me a few bummer stories, like about a New Zealander he met who was just in the beginning of a month long US national parks tour. Not a great way for our country to treat guests.
On a more peripheral but personally upsetting note, the shutdown has prevented my all-time favorite rockclimber, Chris Sharma, from joining an all-star team of climbers on a ridiculously hard new route in Yosemite called the Dawn Wall. Since the Yosemite climbing season is short, this will likely throw off attempts at the Dawn Wall for an entire year.
Come on, US government!
To be fair, these issues are mere inconveniences compared to cancer patients put on hold for clinical trials, scientists waiting for grants to come through, and federal employees unable to pay bills or legally seek other work while their paychecks are withheld. The shutdown has been a major inconvenience and burden for a great number of people.
With all this in the news, I became curious about what kind of measurable effect a government shutdown might have on the economy. There seems to be a lot of supposition and back-of-the-envelope calculations about this in blogs and the news, but little data. My question was simple: did past government shutdowns affect any indicators of US economic health in a length-of-shutdown dependent way? While I’m not an economist, I did a naïve set of calculations and found quite a surprising signal -- in short, I found that yes, the economy does seem to suffer after government shutdowns, in the form of inflation. The longer the shutdown, the more inflation is seen.
The data I analyzed is the consumer price index (CPI), which gives the cost of a battery of goods across the US at a given moment in time. The CPI is available on a monthly basis since at least 1914, which easily covers the period from 1976 to 1995, during which there have been 17 government shutdowns. (That's right, I said Seventeen! I could barely believe it when I first read it.) I lumped together two shutdowns that occurred back-to-back in September-October 1984. The dates of the shutdowns can be found here.
I first looked at whether there is a correlation between the number of days of a government shutdown, and the % change in the CPI afterwards. I found that in fact yes, there is such a correlation. Below I've plotted % change in CPI from the day before each shutdown to a day 4-months after each shutdown (y-axis), versus the number of days of each shutdown (x-axis). For the statistically inclined, the Spearman's correlation is: rho=0.77, p=4e-4. For the non-statistically inclined, this is considered a strong and non-random correlation.
Also, curiously, shutdowns during democratic presidencies have generally been longer than those during Republican presidencies. Hmm...
It is possible that the correlation between length of shutdown and % increase in CPI is simply there because the CPI nearly always rises over time, and thus will rise more during the period of a long shutdown than the period of a short one. To eliminate this confounding factor (or other, potentially more hidden ones), I scrambled the dates of the shutdowns, keeping them the exact same lengths, to see if scrambled dates would show the same correlations as the real ones. In short, the answer is no -- the actual shutdown dates showed a much more significant correlation than scrambled ones. The results of this test are plotted here:
The important thing is that for the first few months, the blue bars (correlations for actual shutdown dates) are higher than any of the errorbars for randomized shutdown dates. I circled the datapoints of most interest in red. This eliminates a lot of the chance that the shutdown-length dependency of CPI increase is merely an artifact.
You can also see on this plot that the correlation between shutdown length and % change in CPI is the highest when checked 4 months after the end of the shutdowns. This implies that it takes about 4 months for the inflationary effects of a shutdown to become most apparent. If anybody reading this blog is an economist, I would be curious if it is common to see a 4-month lag in changes to CPI (or other economic indicators) after a causative economic event.
Aside from looking at the % change in CPI, I was also curious if the CPI increases following long shutdowns were more likely than after short shutdowns to exceed extrapolations based on pre-shutdown data. Suffice to say, this is another test that the CPI tends to rise faster after long shutdowns than after short ones.
To check this, I looked at the % difference after each shutdown between the actual CPI and the CPI expected based on a linear fit of data from before the shutdown. I checked this various numbers of months after each shutdown (using an equal amount of pre-shutdown time for the projection), and found that indeed, the CPI increase exceeds projections by more after long shutdowns than after short ones. You can see it in the below plot, as, again, the blue bars (correlations for actual shutdown dates) are higher than any of the errorbars for random shuffles (with the effect being most dominant 4 months out). Again, I circled the most interesting datapoints:
By the way, I did not see similar trends in the S&P500, the NASDAQ composite, or the GDP. I’m guessing that the stock indicators equilibrate to the expected slump after a shutdown, and the GDP is simply too rough of an economic estimate to be useful since it’s posted only once per year. It's also possible I just didn't look hard enough.
One note -- even with my paltry knowledge of economics, I'm aware that an increase in inflation is often a sign of economic growth. Despite this, I would wager that the post-shutdown inflation is not a sign of economic health, but rather a sign of economic illness. I'm sure there's a way to tell between helpful and harmful inflation. Maybe a nice follow-up study for an enterprising economist?
It's a shame to me that our government can't do such an elementary task as passing a budget. This hurts Americans on many levels, including, I would argue, broad economic ones. It also makes us look like damn fools. I only hope that in future shutdowns, congress will be so kind as to consider my family, my friends, and the entire US economy. Not to mention, I would like them to also consider the Dawn Wall.
