Vampires and Recessions: A Correlational Exercise

Mon, 07 Sep 2026 00:00:00 +0000

This is a note, not a paper. No causal claims are made. A PDF version is here.

Capital is dead labour, that, vampire-like, only lives by sucking living labour, and lives the more, the more labour it sucks. — Karl Marx, Capital, Volume I, Chapter 10 (1867)

Vampires are back. That means a recession is coming. — A TikTok, September 2026

1. The question

Does the popularity of vampires in film track the business cycle? The claim circulating this month is that it does, and that the current wave of vampire films is therefore a recession indicator. The evidence offered is a list of four dates. Dracula was released in 1931, in the Great Depression. Frank Langella’s Dracula, Werner Herzog’s Nosferatu the Vampyre, and Love at First Bite were released in 1979, during stagflation. Twilight and True Blood were released in 2008, the year of the Lehman Brothers collapse. Robert Eggers’ Nosferatu (December 2024), Ryan Coogler’s Sinners (April 2025), and AMC’s Interview with the Vampire are the current wave.

I wanted to know whether this survives contact with data. Two things made the exercise more interesting than I expected. First, the idea has a long intellectual history that predates cinema, and the history makes a specific prediction about which economic variable should matter, which is not the one the TikTok uses. Second, the naive test confirms the claim, and it takes several rounds of doing the analysis properly for the result to change.

This is a correlational exercise. There is no exogenous variation in vampire output and no natural experiment. I say this up front so that nothing below is read as a causal claim.

2. Where the idea comes from

The link between vampires and money is not internet folklore. Marx wrote the sentence in the epigraph in 1867, thirty years before Bram Stoker published Dracula. The literary criticism built on it is substantial. Franco Moretti, in “The Dialectic of Fear” (New Left Review, 1982), reads Dracula as monopoly capital: a foreign aristocrat who accumulates bodies the way a firm accumulates assets, opposed by a coalition of professionals defending free competition. Mark Neocleous argues in History of Political Thought (2003) that the vampire is not a rhetorical ornament in Marx but the central image of his critique, dead labour ruling the living. David McNally’s Monsters of the Market (2011) extends the reading to zombies and global capitalism. Nina Auerbach’s Our Vampires, Ourselves (1995) supplies the summary most often quoted: “every age embraces the vampire it needs, and gets the vampire it deserves.”

Economists have used the vampire too. Dennis Snower’s “Macroeconomic Policy and the Optimal Destruction of Vampires” appeared in the Journal of Political Economy in 1982. It is a satire, but a formally correct one. Vampires reduce the labour force by consuming it, households divert resources to protection, and the paper shows that full eradication is not welfare-optimal.

The specific “monster cycle” claim has a traceable origin. In October 2008, Annalee Newitz published a chart on io9 titled “War and Social Upheaval Cause Spikes in Zombie Movie Production.” It is the ancestor of every subsequent version. Two details matter. The chart is about wars, not recessions, and it explicitly excludes vampires from the count. Newitz followed it two weeks later with “Zombies Are Red, Vampires Are Blue,” arguing that zombies were the Bush-era monster and vampires the Obama-era monster. Peter Rowe made a similar argument in the San Diego Union-Tribune that month. I could not access Rowe’s original article, so I report its numbers only as cited by others: 183 zombie films across the Bush years, and a 2008 in which vampire films outnumbered zombie films two to one.

The party version of the claim has been checked and does not hold. Dan Sanchez noted in 2019 that zombies dominated under both Obama and Trump. Salon revived the recession version in 2022 with the same four-date list. By May 2025, CNN was covering “recession indicator memes” as a genre, and vampires were one entry among dozens. The vampire index belongs to the same family as the lipstick index and the hemline index: pattern-matching on a small set of memorable years, never tested out of sample. To the best of my knowledge, no one has run the test.

3. Data

There is no official count of vampire films, so I built one from web content, following the same logic I have used elsewhere to measure things for which no administrative data exist. I queried Wikidata for every film tagged with the genre “vampire film” or with vampires as its main subject, and took the earliest release year. This gives 566 films between 1920 and 2025. As a check, I scraped Wikipedia’s list of vampire films; the two yearly series correlate at 0.85. I did the same for zombie films (548), for all horror films (10,697), for television series, and for twenty other genres that I use as placebos below.

For the economy I use FRED: the unemployment rate and the CBO’s natural rate, real and potential GDP, consumer sentiment, the NBER recession dates, Moody’s corporate bond yields, household debt, and the Fed funds rate. I add Robert Shiller’s stock market data for real returns and drawdowns, the World Inequality Database for the top 1 percent income share, and Google Trends for monthly search interest since 2004. The unemployment series begins in 1948, which leaves 77 usable years, 21 of them with at least one recession month.

Two limitations of the film data matter for everything that follows. (i) Coverage before 1960 is thin: the entire 1930s contain eight vampire films, so the 1931 Dracula observation is a single movie and cannot be tested. (ii) Coverage improves over time because volunteers tag recent films more thoroughly, so the raw series trends upward for reasons unrelated to the economy.

4. Results

This exercise produced four sets of results.

4.1 The naive test confirms the claim

Figure 1 shows vampire films per year with NBER recessions shaded.

The largest year in the data is 2009, with 28 films, and 2009 is a recession year. The correlation between annual vampire film counts and the unemployment rate over 1948–2025 is 0.31, and it is significant at conventional levels. If this were the end of the analysis, the TikTok would be right.

4.2 Filtering both sides removes the result

The problem is that both series trend. Vampire output rose from 1948 to 2012, partly because film production rose and partly because of the coverage issue above. Unemployment was higher on average in the second half of the period than in the first. Two trending series will correlate regardless of whether they are related.

The fair comparison is between cycles. I use the unemployment gap (the rate minus the CBO natural rate), the output gap, and Hodrick-Prescott cycles on both the macro series and the log film counts. I also change the denominator to vampire films as a share of all horror films, which removes the general rise and fall of cheap horror production, and I add television series, since the current wave is mostly television. Figure 2 shows the filtered series.

The correlation of 0.31 falls to 0.05. Against the unemployment gap it is 0.03. When the gap leads films by one year, which allows for production time, it is −0.08. Vampire share of horror gives 0.03, and adding television gives 0.02. None is close to significance with standard errors that allow for autocorrelation. Across 120 combinations of outcome, transform, and lag against the unemployment gap, six reach the 5 percent level, which is what chance would produce, and none of the six involves vampires.

The simplest version says the same. Recession years averaged 6.2 vampire films and expansion years averaged 7.0.

I also ran an event study around the 11 recession starts since 1949, comparing the average path of the vampire share of horror against 2,000 draws of 11 random years. Figure 3 shows it.

The vampire series rises above the placebo band in the two years after a recession begins, and a naive reading gives a p-value of 0.07. But the series is also outside the band before the recession. The year before a recession starts is a local low for vampires, and the post-recession rise is measured from that low. Re-basing on the average of the three preceding years, the effect is zero. With 11 events, this is as far as the data go.

4.3 Zombies are not a valid control, and the valid controls show nothing

My first comparison group was zombies. This was a mistake worth explaining. The same literature that calls the vampire a capitalist monster says the same of the zombie: the mass, the unemployed, the consumer who cannot stop consuming. If the hypothesis is that economic distress produces monster films, zombies are treated, not control. A control has to be a genre the theory says should not respond.

I therefore ran the identical test on twenty Wikidata genres: westerns, musicals, war films, romantic comedies, spy films, superhero films, sports films, film noir, Christmas films, biographical films, disaster films, heist films, martial arts films, science fiction, and the horror sub-genres (werewolves, ghosts, slashers, monsters), plus zombies and vampires. Figure 4 reports the contemporaneous correlation of each genre’s cycle with the unemployment gap.

Vampires rank eighth of twenty by absolute correlation. Forty percent of genres are at least as correlated with unemployment as vampires, and the median genre has the same correlation vampires do. One genre in twenty clears the 10 percent level, again what chance delivers. The genre with the largest correlation is the werewolf film, for which I am not aware of any theory.

4.4 The theory’s own variable does better, and its second variable has the wrong sign

Unemployment is the variable the TikTok uses, but it is not the variable the theory predicts. In Marx and in Moretti, the vampire is capital. The zombie is the unemployed mass. If the metaphor has empirical content, vampire output should respond to financial conditions: crashes, credit, the fortunes of asset owners. I tested the vampire cycle against real stock returns, market drawdowns, the corporate credit spread, consumer sentiment, changes in household debt, changes in the Fed funds rate, and the top 1 percent income share.

Most of these are flat. Credit spreads, drawdowns, household debt, and the Fed funds rate show nothing. Two variables do not. The prior year’s real stock return correlates with the vampire cycle at −0.40: a bad year in the market is followed by a year with more vampire films. Same-year consumer sentiment correlates at −0.33. On both, vampires are the top-ranked genre of twenty, with the next genre at 0.17 and zombies at zero.

I tried to eliminate these results before reporting them. The first suspect was the filter. Both correlations came with a mirror image at the opposite lag: the stock return one year after the films correlated at +0.36, and sentiment two years after at +0.43. A two-sided smoother can manufacture this pattern, because it uses next year’s films to define this year’s trend. To check, I redid the exercise with one-sided filtering in the Box-Jenkins manner: fit a time-series model to the returns, apply the same filter to the film series, and cross-correlate the residuals, with a moving-block bootstrap for inference. Figure 5 shows the comparison.

The stock return result survives. Annual real returns are uncorrelated at one year, so the one-sided filter reduces to a linear detrend, and there the prior-year correlation is −0.40 while the next-year mirror is 0.05. Where a mirror does reappear under other filters, it comes from the fact that real returns mean-revert at a two-year horizon (autocorrelation −0.21 at lag 2), so any series that follows crashes will also precede rebounds. That is a property of the market, not of the filter. The result is not driven by any single year (leave-one-out changes it by at most 0.04), the rank correlation is −0.36, and vampires remain first of twenty genres with the second at 0.18. After the fourteen years since 1949 in which real returns fell more than 15 percent, the vampire share of horror was at or above trend in eleven.

The consumer sentiment result does not survive. Pre-whitened, it falls to −0.16 and is not significant.

Three caveats apply to the stock return result. (i) It is concentrated before 1985, where the correlation is −0.54; after 1985 it is −0.17 and not significant, so the modern period carries little of it. (ii) The magnitude is small. A year with a −30 percent real return, compared with a year at +10 percent, is associated with about 3.5 percent more vampire share of horror relative to trend, against a mean share of 6 percent. (iii) It is one of 24 finance cells I examined, selected after seeing all 24. The genre ranking protects against this, but the hypothesis was not pre-registered.

The metaphor’s most direct prediction concerns inequality, and here the sign is wrong. If vampires are the monster of extraction, vampire films should be most common when the returns to owning things are highest. Figure 6 shows the ten-year moving average of the vampire share of horror against the top 1 percent income share.

The vampire share peaked in the early 1970s, which is the trough of the top 1 percent share in the century of data, and has fallen through the entire post-1980 rise in inequality. Across eleven decades the correlation is −0.62. Zombies show no pattern and westerns, a neutral genre, are slightly positive. These are two slow-moving series with about ten effective observations, and I present the figure as descriptive. But the descriptive fact is that Hollywood made the most vampire films in the decade when the United States had the fewest rich people, which is not what the theory predicts. Auerbach’s reading would be that the vampire of the 1970s, Anne Rice’s guilty and reluctant Louis, was the vampire an egalitarian decade needed. That may be right about meaning. It is not a prediction about volume.

5. Search interest

Film counts measure what studios made. Google Trends measures what people looked for, monthly, since 2004. That is 264 months and roughly two business cycles. Figure 7 shows searches for “vampire” and “zombie” against unemployment.

Here the unemployment correlation is present: 0.43 on filtered series and 0.51 in twelve-month changes, against 0.08 for zombies. The obvious concern is that the peak in vampire searches and the peak in unemployment are the same month, October 2009, and that month is the release of New Moon. Dropping June 2008 through June 2013 leaves the correlation at 0.40. Dropping 2020 and 2021 as well leaves it at 0.48 in filtered series, though only 0.29 and insignificant in changes. Against this, the 2020 recession, with the largest unemployment spike in the sample, produced no increase in vampire searches at all. With two cycles, one of which contains the most successful vampire franchise in history and the other of which is a counterexample, I would call this suggestive at most. Search interest is also not production.

6. What I take from this

The timing claim does not hold. Once both series are filtered, vampire films do not cluster in recessions, do not follow them, and do not precede them, and their relationship with unemployment is indistinguishable from that of westerns or romantic comedies. The raw correlation that supports the claim is two trends laid on top of each other. Zombies, the supposed contrast case, behave the same way.

The financial version of the claim does better, in a direction consistent with the theory. Vampire output follows bad years in the stock market, only vampire output does, and the pattern survives the check designed to eliminate it. It is small, and it is mostly a pre-1985 phenomenon. I would not have predicted it.

The inequality version, which is the theory’s central prediction, has the wrong sign. Vampires were most common when the top income share was lowest.

None of this touches the interpretive claim that Marx, Moretti, and Auerbach were making, which never depended on release timing. Vampires are the monster of extraction: an aristocrat who does not work, lives off the living, and compounds. Whether each age gets the vampire it deserves is a claim about decades, not quarters, and it is not the kind of claim a time series can settle. Whether the current wave means anything for the economy is a different question, and the answer from this data is no.

Method notes

Films. Wikidata SPARQL for items with genre “vampire film” (Q2137852) or main subject “vampire” (Q46721), earliest publication year, 1920–2025, N = 566. Zombies (Q3072049, Q9406), N = 548. Horror (Q200092 and subclasses), N = 10,697; Wikidata does not class vampire film as a subclass of horror, so the vampire share of horror is a ratio of overlapping sets. Twenty placebo genres by direct genre tag. Television via program genre and subject tags. Wikipedia list cross-check: r = 0.85 between yearly counts.

Macro. FRED: UNRATE, NROU, GDPC1, GDPPOT, UMCSENT, USREC, BAA, AAA, CMDEBT, FEDFUNDS. Shiller’s data for real returns and drawdowns. World Inequality Database for the top 1 percent pre-tax income share. Google Trends via pytrends, US, monthly, “vampire,” “zombie,” and “recession,” scaled jointly.

Filtering and inference. Hodrick-Prescott cycles with λ = 6.25 on annual series (Ravn and Uhlig), λ = 100 as a check, and log differences; outcome cycles on log(1 + x); monthly series with λ = 129,600. Newey-West standard errors with 2 annual or 12 monthly lags. Event study placebo: 2,000 draws of the same number of random start years. Negative binomial count models with log horror films as an offset and a quadratic trend agree with the correlations: −3 percent vampire films per point of unemployment gap, not significant. Pre-whitening: AR(p) by AIC on the linearly detrended regressor (p = 0 for real returns), the same polynomial applied to detrended log film shares, residual cross-correlations with moving-block bootstrap p-values (block length 3, 2,000 draws), leave-one-out influence, Spearman, and subsample splits at 1960 and 1985.

Caveats. 1,440 correlation cells with no multiple-testing correction; the finance results were selected after examining the grid. Wikidata coverage is single digits before 1960 and incomplete after 2022. Two-sided HP filters are least reliable at the endpoints, so 2023–2025 cycle values will revise. Code, data, and full results tables are available on request until the repository is public.

Sources

Marx, Capital I, ch. 10. Moretti, “The Dialectic of Fear,” NLR I/136, 1982. Neocleous, “The Political Economy of the Dead,” History of Political Thought 24(4), 2003. McNally, Monsters of the Market, Brill 2011. Auerbach, Our Vampires, Ourselves, Chicago 1995. Snower, “Macroeconomic Policy and the Optimal Destruction of Vampires,” JPE 90(3), 1982. Newitz, io9, 29 Oct and 10 Nov 2008. Rowe, San Diego Union-Tribune, Nov 2008 (via secondary sources). Sanchez, FEE.org, 29 Oct 2019. McFarland, Salon, 23 Oct 2022. CNN Business, 4 May 2025. Full annotated list with URLs in the project repository.

Curiosities | Fernanda Sobrino