Measuring Shortages since 1900

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1407 May 2025 Measuring Shortages since 1900 Dario Caldara, Matteo Iacoviello, and David Yu Please cite this paper as: Caldara,Dario,MatteoIacoviello,andDavidYu(2025). “MeasuringShortagessince1900,” International Finance Discussion Papers 1407. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2025.1407. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

∗ Measuring Shortages since 1900 Dario Caldara Matteo Iacoviello David Yu May 8, 2025 Abstract This paper introduces a monthly shortage index spanning 1900 to the present, constructed from 25 millionnewspaperarticles. Theindex capturesshortages across industry, labor, food, and energy, and spikes during economic crises and wars. We validate the index and show that it provides information beyond traditional macroeconomic indicators. Using predictive regressions, we find that shortages are associated with persistently high inflation and lower economic activity. A structural VAR model reveals that, compared to a traditional supply shock, surprise movements in shortages produce less inflation relative to their GDP impact, suggesting that shortages are associated with constraints on price adjustment that limit inflation but magnify the decline in real activity. We also show that post-pandemic shortages and inflation were primarily driven by supply forces, with demand factors playing a less important role. KEYWORDS: Shortages; Inflation; Textual Analysis; Predictive Regressions; Structural VAR Model. JEL CLASSIFICATION: C32, C55, E31, N10. ∗Corresponding author: Matteo Iacoviello (matteo.iacoviello@frb.gov). The data on shortages presented in thispaperandregularupdatesoftheindexareavailableathttps://www.matteoiacoviello.com/shortages. html. Caldara and Iacoviello are at the Federal Reserve Board. Yu is at the University of California Los Angeles. We thank Sam Nate for outstanding research assistance. The views expressed in this paper are solely theresponsibilityoftheauthorsandshouldnotbeinterpretedasreflectingtheviewsoftheBoardofGovernors of the Federal Reserve System or of anyone else associated with the Federal Reserve System.

1 Introduction Shortages—–defined as the inability of supply to meet demand at prevailing prices–—have been a recurring feature of economic life throughout the 20th and early 21st centuries. From widespread disruptions during major wars to the acute shortages of goods and labor during the COVID-19 pandemic, these episodes have strained supply chains, hindered production, and placed significant burdens on households and businesses. Despite their importance, there has been limited research on the long-run trends and patterns of shortages across different sectors, as well as on their economic effects. This paper addresses this gap by providing a comprehensive analysis of shortages over a 125-year period and demonstrating their persistent effects on inflation and economic activity. In the first part of this paper, we construct a monthly shortage index based on newspaper articles spanning from 1900 through the present (Section 2). This index—–covering industry, labor, food and energy shortages—is derived from approximately 25 million articles from six major U.S. newspapers, providing a novel, long-term perspective on global supply disruptions. The index captures historical events effectively, showing significant spikes during periods of economic turmoil such as the two world wars and the 1970s oil crises, and reaching its highest level in 40 years during the COVID-19 pandemic. Although developed using U.S. news sources, our index captures shortages on a global scale. We demonstrate this property by constructing sub-indexes for six major advanced economies—Canada, France, Germany, the United Kingdom, Japan, and the United States. These indexes spike around common global events, but also reflect country-specific variation, capturing both differential exposure to global shocks and idiosyncratic episodes, such as the Suez Crisis in 1956 or the 2011 earthquake and tsunami in Japan. In Section 3, we validate our index through narrative investigation, text analysis based on artificial intelligence (AI), and comparison to alternative measures of supply-side disruptions. The second part of this paper shows that shortages are associated to higher inflation and lower economic activity in the United States. Using predictive regressions (Section 4), we estimate a negative correlation between shortages and future GDP and a positive relationship between shortages and future prices, particularly for durable consumption goods and private fixed investment. These results remain robust even when controlling for traditional indicators of inflation pressures such as commodity prices and wages, suggesting that our shortage 2

index provides additional information not captured by conventional measures. An out-ofsample forecasting exercise reveals that a model incorporating the shortage index outperforms competing models at a three-month horizon, highlighting the practical value of tracking shortages for inflation prediction. To investigate how the relationship between shortages and activity varies over time, we estimate rolling regressions of future prices and future GDP on shortages. Throughout most historical periods, shortages have been associated with higher future inflation. The predictive power for future GDP growth is typically negative, with notable exceptions during World War II and the COVID-19 pandemic—periods when shortages coincided with strong demand pressures—suggesting that both supply and demand forces are important for understanding shortage dynamics. The varying impact of shortages and their persistent impact on inflation motivate our structural vector autoregressive (VAR) model in Section 5. Unlike predictive regressions, a fully identified VAR model provides a comprehensive and economically interpretable framework for understanding joint movements in GDP, inflation, and shortages. The model also includes commodity prices and short-term interest rates, allowing us to identify “traditional” demand and supply shocks, commodity price shocks, monetary policy shocks, and shortage shocks.1 Our identification strategy is as follows. Demand and supply shocks are identified based on the slope of economy-wide demand and supply curves and can affect prices, activity, and shortages contemporaneously, thus allowing shortages to react on impact in response to standard demand and supply shocks, for instance because price adjustment is not instantaneous. Shortage shocks, by contrast, are assumed to affect prices and activity with one-quarter lag—reflecting that while newspapers may immediately report on shortages, their broader economic impact typically may materialize more gradually. This approach conservatively attributes immediate movements in activity and prices to traditional shocks, while preserving a distinct role for episodes of acute shortages. We interpret shortage shocks as unexpected disruptions that require two concurrent conditions to materialize. First, there must be supply constraints stemming from events such as extreme weather, natural disasters, geopolitical conflicts, labor strikes, or production failures. Second, unusual market frictions must impede normal price adjustment processes. These 1The inclusion of commodity prices is also motivated by a large and growing literature that attributes a role to energy shocks in driving post-pandemic inflation dynamics (Baumeister, 2023; Dao et al., 2024). 3

frictions include hoarding behavior, regulatory interventions, social norms against price increases, or coordination failures. Together, these two conditions result in quantity rationing that persists beyond what traditional market clearing would predict, and differentiate shortage shocks from traditional adverse supply shocks.2 The VAR results reveal several key insights. First, while traditional economic forces contribute to the identified movements in the shortage index, exogenous shortage shocks remain the dominant driving force behind these fluctuations. Second, shortage shocks have significant and persistent effects on GDP and inflation, showcasing the shortage index’s value as a leading indicator of activity and prices. Third, compared to traditional contractionary supply shocks, shortage shocks produce less inflation relative to their impact on economic activity—suggesting that shortages often involve constraints on price adjustment that limit inflation but magnify the decline in real activity. Historical decompositions from our model provide new perspectives on major economic episodes. Shortages in the 1950s were largely driven by the Korean War, post-World War II pent-up demand, and the rapid economic transition from wartime to peacetime. The 1970s oil embargo triggered shortages beyond what could be explained by commodity price shocks alone. During the COVID-19 pandemic, shortages emerged from a confluence of demand, supply, and shortage-specific shocks. Counterfactual simulations reveal that, according to the model, the rise in shortages and inflation that started in 2020 was predominantly driven by supply forces—a combination of traditional supply shocks, shocks to commodity prices, and shocks to shortages—with demand forces playing a somewhat secondary role. The reversal of adverse supply shocks and negative demand shocks in 2022 and 2023—capturing the tightening of monetary policy and more favorable supply conditions—led to a faster-than-expected decline in inflation.3 Our approach builds on previous work combining newspaper sources and textual analysis to measure economic phenomena, such as the Economic Policy Uncertainty Index (Baker, Bloom, and Davis, 2016) and the Geopolitical Risk Index (Caldara and Iacoviello, 2022). 2The combination of supply disruptions with market frictions that impede price adjustment can potentially result in larger quantity reductions relative to price increases, distinguishing shortage shocks from conventional supply shocks where prices bear relatively more of the adjustment burden. 3Our analysis of the post-pandemic drivers of inflation relates to a large and growing literature. Giannone and Primiceri (2024) and Bergholt et al. (2024) find that the pandemic-era rise in inflation was mostly driven by demand factors. By contrast, Ascari et al. (2024), Comin et al. (2024), Shapiro (2024) and Di Giovanni et al. (2023) find a relatively larger role for supply factors. 4

While other researchers have developed news-based shortage indicators for shorter periods (Lamont, 1997; Chen and Houle, 2023; Burriel et al., 2023) or constructed alternative supply disruption measures using transportation and shipping data (Benigno et al., 2022; Bai et al., 2024; Liu et al., 2024), our contribution is distinct in three ways.4 First, we develop the first comprehensive shortage index covering more than 125 years, offering a unique long-term perspective unavailable in existing indicators. Second, our disaggregated indexes capture events in different markets at both global and country-specific levels, providing valuable insights for historical and cross-country analysis. Third, our empirical analysis sheds light on causes and consequences of shortages, and demonstrates that shortages have more persistent effects on inflation than previously recognized. Unlike Bernanke and Blanchard (2023), who use Google searches for the term “shortage” as a proxy for supply disruptions and find only short-lived inflationary effects from shortages, our longer-term analysis reveals significantly more prolonged inflationary responses—underscoring the importance of examining supply disruptions across extended time horizons. The remainder of the paper is structured as follows. Section 2 discusses the construction of the index and its evolution over time. Section 3 validates the index. Sections 4 and 5 examine the relationship between the shortage index and U.S. economic outcomes using predictive regressions, forecasting analysis, and a structural VAR model. Section 6 concludes. 2 The Shortage Index In this section, we begin by discussing the construction of the shortage index. We then show how the index captures key episodes of shortages in U.S. history. 2.1 Construction of the Index The shortage index is a monthly news-based indicator that measures the intensity of shortages in materials, goods, labor, and energy in the United States. It is constructed from a sample of 4Goldfarb (2013) provides a taxonomy of shortages that aligns well with our framework. Goldfarb identifies six shortage types: (1) demand deadline shortages (e.g. Christmas toys), (2) dynamic shortages from supply/demand lags (labor shortages), (3) below-market pricing by suppliers (concert tickets), (4) governmentregulated prices (parking), (5) capacity choice under demand variance (airline seats), and (6) sudden supply shocks (natural disasters). These types stem from key underlying causes—planning errors, non-market-clearing pricing, supply constraints, and reaction lags—that correspond well with our idea that news about shortages are likely to occur when the combination of supply constraints and market frictions impedes price adjustment. 5

approximately 20,000 news articles per month, spanning from 1900 through present—about 25 million articles in total. These articles are published in major U.S. newspapers: the Boston Globe, the Chicago Tribune, the Los Angeles Times, The New York Times, The Wall Street Journal, and The Washington Post. Each month, the index counts articles that discuss shortages in energy, food, industry or labor markets—the set S visualized in the top panel of Figure 1—normalized by the total number of articles, denoted as the set A. A higher index value indicates more intense shortages. Below, we outline the steps used to create the search query, reported in Table 1, for isolating set S. First, we construct a broader set of articles, B, the pink area in Figure 1. Articles in this set must mention at least one shortage term—namely ‘shortage,’ ‘scarcity,’ ‘bottleneck,’ or ‘rationing’—in conjunction with an economics-related term, such as ‘economy,’ ‘market,’ or ‘commerce.’ The shortage terms used are those most frequently linked to economically relevant constraints on production or on the availability of goods to consumers.5 Including an economics-related term reduces the occurrence of false positives—articles mentioning shortages unrelated to economic phenomena, such as shortages of empathy or goodwill. Second, we draw a random sample of about 3,300 articles from the set B and extract the 1,000 most frequent collocates, words appearing within five words of the shortage terms. From these collocates, we select those indicative of shortages in specific sectors. The most common collocates, excluding stopwords, include ‘oil,’ ‘water,’ ‘war,’ ‘time,’ ‘coal,’ ‘days,’ ‘food,’ ‘cars,’ ‘people,’ ‘government,’ ‘million,’ ‘labor,’ ‘state,’ ‘home,’ ‘steel,’ and ‘fuel.’ Subsequently, generic words like ’time,’ ’days,’ ’people,’ and ’government,’ which convey little information about shortages, are removed. The remaining collocates are grouped into four topics: food, industry, labor, and energy.6 Using the lists of shortage terms, economic terms and topic-specific terms, we construct the search query shown in Table 1.7 An article is included in the shortage index (set S in Figure 1) if it meets two conditions: (1) A shortage term must appear within five words of a topic-specific term, and (2) the article must contain at least one economics term. If an article meets the first condition for multiple topic-specific categories, it is counted once for each—so the total 5Potential synonyms of shortage, such as ‘lack,’ ‘paucity,’ or ‘insufficiency,’ were excluded because they have a broader range of meanings and are less likely to be specifically associated with economic shortages. 6Exploratory work identified additional categories (housing, health, and education shortages), but these were excluded because they either reflect long-standing social issues or contribute minimally to the evolution of the overall index. 7See the Appendix for the data sources. 6

number of shortage articles is the sum across categories. This method gives greater weight to articles discussing multiple types of shortages, enhancing the index’s informational content. Section 3 documents that the requirement of proximity between shortage and topic-specific terms is crucial to reducing false positives and improving search accuracy. Our article classification into four topics is further supported by an ex-post Latent Dirichlet Allocation (LDA) analysis on a sample of articles meeting our inclusion criteria. LDA, a widely used unsupervised machine learning technique for topic modeling, identifies hidden topics by analyzing word co-occurrences. The analysis uses two inputs: a text corpus consisting of a random sample of 13,623 newspaper abstracts that mention shortages (about 4 percent of the approximately 330,000 articles from 1900 to 2023) and the number of topics (set to four).8 The output of the LDA analysis is a set of topic distributions—one for each abstract. Specifically, for each article, LDA provides a probability vector indicating the degree to which each of the four topics is present. These probabilities allow us to validate the topic classification used in our searches and allow us to track trends in topic prevalence over time. The LDA results are presented in the bottom panel of Figure 1. The word clouds display the most frequent terms for each topic, which we interpret as follows. Topic 1 corresponds to energy; Topic 2 to water, food, and agricultural products; Topic 3 to goods and industrial products (e.g., coal, steel, railroads, and cars); and Topic 4 to the labor market. These findings validate our selection of topics for the index, confirming that the topics capture the primary dimensions of economic shortages reflected in the news. In the next section, we discuss the evolution of topics over time. 2.2 Shortages in History We now present the shortage index, highlight its spikes, and discuss the historical context of these events. Figure 2 shows the monthly shortage index from 1900 through 2024. The index is calculated as the monthly share of articles discussing economic shortages (number of articles in S divided by the number of articles in A), scaled so that its average is 100 between 1900 and 2019. Figure 3 presents a breakdown of the index by topic. Table 2 lists the 30 largest 8Standard pre-processing was applied, including the removal of stopwords, numbers, and word stemming. 7

spikes in the index, along with descriptions of the key events associated with each episode.9 The index varies considerably over time, with the most significant spikes linked to events related to the four classification topics—energy, food, industry, and labor. Adverse geopolitical events—especially wars—are strongly associated with severe shortages. For instance, the index rises sharply during World War I and during World War II, peaking at over 1,000 (10 times the sample mean) in January 1943, with spikes recorded across all shortage categories. Other events, such as the Suez Crisis and the Iraqi invasion of Kuwait in 1990, also coincide with substantial spikes.10 Many of these spikes are linked to energy shortages, often resulting from wars and instability in the Middle East, such as during the oil shocks of the 1970s. Labor shortages, particularly those caused by strikes, were historically significant, especially in the early part of the sample. For example, coal-related strikes in 1903, 1919, and 1922 produced notable spikes. However, strike-related shortages have become less frequent in recent decades as labor markets evolved and other factors gained prominence. More recently, the index spiked on several occasions the COVID-19 pandemic. The first spike corresponds to shortages in medical equipment and healthcare workers at the pandemic’s onset. A second, larger spike occurred at the beginning of 2022, driven by global supply bottlenecks as economies reopened after prolonged mobility restrictions in 2020 and 2021. As shown in Figure 3, this second spike primarily reflected shortages of labor and industry components. The stacked bar chart at the bottom of Figure 3 visualizes the probabilistic topic mixture calculated with the LDA algorithm for each abstract, sorted by year. Each year’s news coverage reflects varying degrees of focus on different shortage topics, confirming the taxonomy ofshortagesfromtheanalysisofthesub-indexes. Intheearlypartofthesample, discussionsare primarily centered around industry-related shortages, with food and water shortages receiving some attention. Energy shortages become prominent during the 1970s, largely due to events like the oil crisis, which significantly affected the global economy. In the more recent period, following the COVID-19 pandemic, labor shortages have increasingly dominated the news, reflecting the widespread disruptions in workforce availability and supply chains that persisted post-pandemic. 9We calculate spikes by first extracting the residuals from a regression of the shortage index, h , on its t values two months before and two months after. The largest residuals are our candidate spikes. We then discard any spike if it is not a local maximum within the 13-month window [t−6,t+6], ensuring that we capture true local peaks. The table reports the 30 largest residuals. 10Federle et al. (2024) show that wars are associated with supply disruptions both domestically and abroad. 8

2.3 Key Drivers of Shortages and Breakdown by Country To determine whether shortages tend to occur alongside particular phenomena, we examine their co-occurrence in the news with other economic events—specifically, labor strikes, natural disasters, adverse geopolitical events, pandemics, and price controls. Figure 4 compares the actual joint probability of these events being mentioned with shortages to the probability expected under independence. The lift ratio, defined as the ratio of the observed joint probability to the product of the individual probabilities, serves as a measure of association. A lift ratio greater than one indicates that these events co-occur more frequently than would be expected by chance. The results indicate that price controls have the strongest association with shortages, co-occurring 7.6 times more frequently than expected under independence. Other events show weaker associations, but still co-occur between twice and three times more often as expected. These patterns suggest that shortages are not randomly distributed across different economic disruptions but instead cluster with certain events, with price controls exhibiting the strongest association. Figure 5 breaks down the shortage index by country. Specifically, we “slice” the main index by counting the number of articles that satisfy the shortage search query and mention either the name of a country or one of its major cities. The country-specific shortage indexes reveal both similarities and differences. As expected, they all tend to spike around global events; however, there are notable differences as well. Countries are differentially exposed to global shortages in terms of timing—for instance European countries experienced earlier and more prolonged shortages during the two world wars than North American countries—and intensity, as evidenced by France and Germany during World War I and by Japan in the late 1970s. The figure also reveals some country-specific episodes of shortages. For instance, it shows oil and energy shortages in France and the United Kingdom following the Suez Crisis, concerns about shortages in Germany around the time of the construction of the Berlin Wall in 1961, or shortages in the aftermath of the 2011 tsunami in Japan.11 In sum, the country-specific indexes offer valuable insights for comparative and cross-country research. Their ability to capture country variation in shortage events also provides additional evidence that our methodology accurately measures shortages, reinforcing the robustness of our search-query-based approach. 11In line with our measurement, the Suez Crisis did not create major energy shortages in Germany because its energy supply was less reliant on Middle Eastern sources compared to France and the United Kingdom 9

3 Assessing the Accuracy of the Shortage Index We perform two exercises to assess the index’s accuracy. First, we confirm that the newspaper articles in the index mention actual shortage-related concerns. Second, we evaluate whether the index aligns with alternative proxies for shortages during the overlapping periods. A key advantage of our index is its long historical coverage—from 1900 through present—compared to existing measures that span only shorter, more recent time frames. 3.1 Validation of the Shortage Index Audit of the Overall Index We verify that our index accurately measures shortages—minimizing both Type-I and Type- II errors—by sampling the abstracts of 872 articles from the shortage set S.12 By design, each article contains at least one economics-related term along with a reference to shortages, scarcities, or bottlenecks near a topic word (energy, food, industry, or labor). For each article, we extract the first snippet of text that references shortages. These snippets are centered on the shortage word and are limited to 110 characters—a length chosen to balance brevity as well as computational and cognitive load. For example, one sample snippet is: “...Ford motor company said today that still lingering steel shortages would force a closing next week of eighteen...” (New York Times, 1952) We then used Claude AI Assistant (Anthropic, 2024) to determine whether each snippet mentions current or prospective shortages, rationing, scarcity, or bottlenecks related to industry, labor, materials, food, or water. Claude returned a table of results, coding snippets as 1 (shortage mentioned), 0 (shortage not mentioned), or 99 (unsure). Additionally, Claude provided a brief explanation for each coding decision. Beforeinitiatingclassification,weprovidedClaudewithexamples—–listedintheAppendix—– of how to code the snippets, ensuring that the training sample included false positives (e.g., mentions of the lack or end of shortages). Although AI-based validation is not foolproof, we 12The abstracts must also satisfy the conditions to belong in the set S. Our search query is run on the six U.S. newspapers listed in Section 2. Abstracts for the six newspapers are only available until 2014. For the period 2015–2023, we sampled abstracts from a broader set of newspapers, including British, Canadian, and Australian newspapers. Abstracts in the database are typically short portions of text, often containing the opening sentences or the initial paragraphs of the article. A list of the 872 abstracts is available upon request. 10

found that Claude performed comparably to a human evaluator, even extrapolating contextual cues (such as linking a sentence to a particular country or individual). For example, when processing the snippet “Economy may be slowing but Lowe is banking on labour shortages gradually leading to an increase...,” Claude classified it as 1, and noted that the Reserve Bank of Australia was expecting labor shortages to lead to wage growth.13 The audit results are summarized in Table 3. Out of 872 articles in the set S, only 6.3 percent were classified by Claude as false positives. For example, Claude coded the snippet “...says that while there is not necessarily a shortage of people wanting to work in management” as 0, explaining that there was “No shortage of people wanting to work in management.” Additional examples are provided in the Appendix. We then repeated the audit for a sample of 298 articles outside the set S. Among these, only one article mentioned shortage of manufacturing plants that were not captured by our search query (“recycling of newsprint was held back by a shortage of de-inking plants”). Notably, our search query deliberately excluded the word ”plants” because preliminary tests revealed that its inclusion produced numerous false positives.14 As a last check, we confirmed that restricting the search to include shortage words near key terms related to industry, labor, food, or energy significantly improves accuracy. In a broader set of articles in which the presence of shortage words near these key terms is allowed but not required, 84.2 percent of articles mention economically relevant shortages. This approach yields a Type I error rate of 15.8 percent, considerably higher than that achieved with our preferred search query. False positives in the broader set include articles referencing non-economic shortages, such as shortages of political campaign funds, a lack of quality baseball photos, legislative bottlenecks, and shortages of sunshine. Validation of Categorical Shortage Indexes The categorical shortage sub-indexes are correlated, yet each contains unique information about different aspects of shortages. We show this result in the Appendix, where we reports estimates from regressions in which selected price and wage measures are each regressed on the four categorical shortage indexes for industrial products, labor, energy and food. We 13The excerpt quoted is from March 2019, when Philip Lowe was the Governor of the Reserve Bank of Australia. 14For example, see the article “Brighten Up Indoors with Colorful Plants” (Los Angeles Times, February 4, 1996) that states “There’s no shortage of plants with brightly colored foliage to liven up your kitchen.” 11

find that the industry shortage index predicts producer price index (PPI) inflation, the labor shortage index predicts earnings growth, and the energy and food shortage indexes predict consumer price index (CPI) energy inflation and CPI food inflation, respectively. In other words, each subindex forecasts the prices (and earnings) of products most likely to be affected by that specific shortage, thus providing further support to the idea that our index and its disaggregated components contains useful informational content. 3.2 Comparison with Other Indicators of Supply Constraints Figure 6 plots our shortage index alongside the New York Fed Global Supply Chain Pressure Index (GSCPI) (Benigno et al., 2022), the ISM Supplier Deliveries Index (SDI), and the Supply Bottlenecks Index (SBI) for the United States developed by Burriel et al. (2023). The GSCPI measures global supply chain conditions using data on both manufacturing and transportation costs. The SDI is based on a monthly survey that asks firms whether they are experiencing longer or shorter wait times compared with the previous month; the SDI value represents the share of respondents reporting longer wait times, plus half the share reporting no change. The SBI uses a text-based newspaper search to quantify supply chain issues. To facilitate the comparison, we standardize each variable to have a mean of 0 and a variance of 1 over the overlapping period. Figure 6 shows that our index shares similar features with these three indicators. Over the full period of the GSCPI, the correlation between our index and the GSCPI is 0.73. Both measures increased sharply at the onset of COVID-19 in early 2020, and again at the beginning of 2022 as supply chain bottlenecks intensified. The correlation between our index and the SDI is 0.25; nonetheless, both measures spiked around the 1979 oil crisis and during COVID-19. One possible explanation is that both events caused significant transportation delays—whether due to rising fuel costs or supply bottlenecks—that contributed to manufacturing shortages. Finally, our index has a high correlation of 0.90 with the U.S. SBI, indicating alignment between these measures over the common sample, although most of the correlation is driven by the sharp rise in shortages after 2020 (the correlation drops to 0.61 when the sample is cut in December 2019). Thus, while all indicators provide useful insights into shortages and broader supply constraints, a key advantage of our index is its long historical coverage. Its availability from 1900 through present (other indexes start only in 1980s or later) makes it particularly valuable for 12

historical research on long-term trends and cyclical patterns in shortages. 4 Shortages as Predictors of Inflation and Activity In this section, we investigate the predictive power of our shortage index for key macroeconomic outcomes in the United States. We first analyze how shortages relate to inflation and economic activity using a predictive regression framework. Our results indicate that higher levels of shortages are associated with prolonged periods of elevated inflation and a deceleration in economic activity. We then extend our analysis to a real-time forecasting exercise, demonstrating that incorporating the shortage index improves out-of-sample forecasts of headline inflation. Together, these findings underscore the importance of supply-side disruptions in shaping macroeconomic dynamics and validate our search-query-based approach. 4.1 Predictive Regressions: Shortages, Inflation, and Activity We explore the relationship among shortages, inflation, and economic activity using the following predictive regression: p (cid:88) ∆Y = α+βSHORTAGE + γ′X +ε (1) t+h t i t−i t+h i=0 where (cid:18) (cid:19) 400 Y t+h ∆Y = ln t+h h Y t represents the annualized log change of a variable of interest Y between period t and forecast t horizon h, and SHORTAGE denotes the shortage index. The vector X contains control t variables. We use quarterly data from 1950 through 2023 for real per-capita GDP, personal consumptionexpenditures, andprivatefixedinvestment. Inflationforeachcategoryismeasured using the associated price deflator. For each price and economic activity indicator, we estimate equation (1) by OLS, with standard errors calculated following Newey and West (1987). As control variables, we include quarterly changes of the dependent variable and the corresponding economic indicator or price deflator, both contemporaneously and with three lags. For example, the regression for real GDP growth includes contemporaneous and lagged values of both real GDP growth and inflation (measured by the log change in the GDP deflator). 13

Figure 7 presents the results for four-quarter-ahead regressions (the Appendix reports results for the one- and eight-quarter horizons). For ease of comparison, we report standardized estimates of the coefficient β. In the full sample, higher shortages are associated with a rise in inflation and a decline in economic activity—the typical effects of supply-side disruptions. The inflationary effects of shortages are fairly evenly distributed across GDP components, with prices in the services consumption category being the least affected. While both durable goods consumption and private investment decline, there is no statistically significant effect on the consumption of nondurable goods or services. Quantitatively, the standardized coefficients imply that a one-standard-deviation increase in the shortage index is associated with a rise in inflation (as measured by changes in the GDP deflator) of 0.41 percentage point, and a decline in GDP growth of 0.32 percentage point. Figure 7 also documents time variation in the relationship among shortages, inflation, and economic growth. When we split the sample into two periods, pre- and post-2000s, shortages exhibit a positive correlation with future sectoral inflation throughout, but their link with future economic activity shifts from negative in the earlier period to positive (and statistically insignificant) in the latter. The positive correlation between shortages and activity during the more recent period, which includes the COVID-19 pandemic, suggests demand factors may have been more prominent drivers of shortages relative to the pre-2000 period, an issue that we return to in Section 5. A further concern is whether the shortage index contains information beyond that captured by other macroeconomic variables. We address this issue by augmenting our baseline regression with additional controls, including oil prices, commodity prices, wage growth, and inflation expectations. The inclusion of these controls only slightly attenuates the effect of shortages on inflation and activity, but the coefficients remain in large part statistically significant (see the Appendix for details). These findings suggest that our shortage index contains additional information beyond that captured by traditional macroeconomic indicators. To assess the robustness of our findings, we estimate equation (1) on a rolling sample using a 30-year window, focusing on real GDP growth and its deflator and extending the sample back to 1900:Q1.15 Figure 8 shows that shortages have consistently exerted inflationary effects since World War II, except during the 10-year period preceding the pandemic, when the effect 15We extend the sample for GDP, the GDP deflator, and population using data from Ramey and Zubairy (2018). 14

becomes negligible and imprecisely estimated. Moreover, shortages have had adverse effects on economic activity except during periods—such as the pandemic and World War II—when elevated demand alongside binding supply constraints may have temporarily sustained output levels while intensifying shortages. 4.2 Forecasting Inflation Building on the work of Stock and Watson (1999), we also examine the efficacy of the shortage index in forecasting inflation at shorter horizons. Specifically, we assess whether our index can improve three-month ahead forecasts of annual inflation. To this end, we use monthly data to estimate the following specification: π12 = c+βπ12 +γ′x +δh , (2) t+3 t t t where π12 is the 12-month percent change in the headline CPI index three months ahead; π12 t+3 t is current 12-month CPI inflation; x is a vector of economic variables, including unemployment t and the 12-month change in oil prices; and h is the shortage index. To keep the specification t as simple as possible, we only include the current value of each variable. However, to reduce the influence of extreme swings in the variables during the pandemic, we smooth shortages and each series in x using 12-month lagged moving averages. The design is a full-blown t real-time forecasting exercise, as three-month-ahead inflation forecasts are made using only data available up to each forecast date. We use a 30-year rolling window, and implement a series of real-time rolling forecasts from 1990:M1 to 2024:M9 (the last observation for inflation is 2024:M12). The choice of the forecasting framework is motivated by the findings of Ang et al. (2007), who compare alternative inflation forecasting methods and show that while survey-based and atheoretical models often yield the best predictions for inflation, economic models that build on the Phillips curve can still perform well while maintaining a structural grounding. In particular, we select a specification that delivers good out-of-sample performance while maintaining interpretability and coherence with macroeconomic fundamentals. Our primary finding is that a model incorporating inflation expectations, unemployment, oil prices, and the shortage index outperforms competing models that exclude the shortage measure. For the whole forecasting window covering the 1990–2024 period, the root mean squared error (RMSE) of the forecast with shortages is 0.86—7 percent lower than the RMSE 15

of 0.92 of the model that excludes shortages, and 8 percent lower than a simple autoregressive model that only uses current inflation. In the 2020–2024 period, the RMSE of the model including shortages is 1.04, an improvement of about 20 percent relative to the model without shortages, which has an RMSE of 1.31. As illustrated in Figure 9, the forecasting model that includes shortages predicts a sharper rise in inflation from 2021 through 2022, thereby reducing forecast errors.16 In summary, both the predictive regressions and the out-of-sample forecasting exercise indicate that shortages have significant effects on inflation and economic activity. 5 Shortages: A VAR Analysis In this section, we present a structural VAR model of the U.S. economy that incorporates shortages. Compared to the predictive regressions discussed earlier, a full structural model offers a more comprehensive explanation of how movements in shortages interact with broader economic activity. The VAR approach has the additional benefit of capturing how the relationshipbetweenshortagesandeconomicactivitymayevolveovertimeasdifferenteconomic shocks become more or less dominant. This time-varying relationship is the expected outcome when the composition of shocks underlying movements in shortages changes and each shock is associated with distinct patterns of economic responses.17 5.1 The Model The VAR model includes five quarterly variables: activity, inflation, commodity prices, shortages, and the short-term interest rate. These are summarized by the vector X = (y ,π ,h ,c ,r )′ (3) t t t t t t where activity (y ) is the four-quarter percent change in U.S. real GDP; inflation (π ) is the t t 16The shortage index also improves the forecast for core inflation (CPI All Items Less Food and Energy). In the full 1990–2024 period, the RMSE of the forecast with shortages is 0.40—lower than the RMSE of 0.44 for the model that excludes shortages. In the 2020–2024 period, the RMSE is 0.83 for the model with shortages compared to 0.77 for the model without shortages. 17A more comprehensive approach to quantifying time-varying relationships among shortages, economic activity, and inflation would involve estimating a VAR model with time-varying parameters or other forms of nonlinearities. For instance, periods of intense shortages might coincide with—or even induce—changes in the structural parameters of the model, giving rise to nonlinear dynamics. We leave these issues for future research. 16

four-quarter percent change in the U.S. headline CPI index; shortages (h ) are expressed in t levels (standardized); commodity prices (c ) are measured by the four-quarter percent change t in the Reuters–CRB Spot Commodity Price Index for raw industrials and foodstuffs; and the short-term interest rate (r ) is the annualized three-month Treasury bill rate. All series are t demeaned. The sample runs from 1950:Q1 through 2023:Q4. The VAR representation is given by: p (cid:88) AX = B X +Cu , (4) t j t−j t j=1 where u = (uS,uD,uH,uC,uR)′ is the vector of structural shocks with zero mean and a t diagonal covariance matrix E[u u′] = Σ . Without loss of generality, we normalize one entry t t u per row of A to 1. The structural relationships among variables, with z = (X′ ,X′ ,...,X′ )′ denoting t−1 t−1 t−2 t−p lagged variables, are described by the following equations: π = uS +κ y +[bS]′z , (5) t t t t−1 y = uD −δ π +[bD]′z , (6) t t t t−1 h = uH +θ uD +θ uS +[bH]′z , (7) t t D t S t t−1 c = uC +χ uD +χ uS +χ uH +[bC]′z , (8) t t D t S t H t t−1 r = uR +α π +α y +α h +α c +[bR]′z , (9) t t π t Y t H t C t t−1 Equation (5) represents the economy-wide aggregate supply equation, where inflation is positively related to activity (κ > 0) and subject to adverse supply shocks uS. Equation t (6) describes aggregate demand, which is negatively related to inflation (δ > 0) and driven by demand shocks uD. Equation (7) models shortages, which are assumed to respond to t demand and supply shocks as well as to an exogenous component uH capturing newsworthy t unexpected disruptions in the flow of goods, services, and factors of production that cause demand to temporarily exceed supply. Examples of exogenous shortage shocks include instances where supply disruptions, such as extreme weather events, production failures, or geopolitical disruptions, co-occur alongside market frictions, such as sudden hoarding behavior, emergency regulatory interventions, rapid adoption of social norms against price increases, or coordination failures, resulting in persistent quantity rationing beyond what traditional market clearing 17

would predict. Equation (8) describes the behavior of commodity prices, that respond to demand, supply and shortage shocks along with commodity-market specific disturbances (uC). t Finally, equation (9) describes the monetary policy rule, where uR denotes monetary policy t shocks. Our model incorporates specific timing assumptions regarding shortages and their economic impacts. In particular, exogenous shortage shocks affect GDP and inflation with at least a one-quarter delay, while they can influence commodity prices and interest rates within the same quarter. This timing structure reflects how news about shortages tends to immediately impact fast-moving variables like commodity prices and interest rates, before affecting broader economic measures such as inflation and economic activity. Our approach distinctly treats shortages as contemporaneously responsive to economic conditions caused by traditional demand and supply shocks. This strategy differentiates our analysis from both Burriel et al. (2023), who order shortages first in a monthly VAR model using data from 1990 through 2020, and Bernanke and Blanchard (2023), who employ a quarterly VAR model from 1990 through 2023 using the assumption that shortages affect inflation within the same quarter. 5.2 Priors and Estimation Following Baumeister and Hamilton (2019), we treat identification and estimation of this structural VAR model as a Bayesian exercise, imposing prior restrictions directly on the parameters in the matrices A, B , and C. These priors help identify the structural shocks j and their dynamic effects. For example, if one were to assume a flat within-quarter Phillips curve—through a prior distribution centered at κ = 0 with zero variance—, the impact matrix D = A−1C would become lower triangular, providing a recursive VAR interpretation with the ordering (y ,π ,h ,c ,r ). t t t t t We estimate the model with p = 2 lags, setting most cross-partial terms beyond the first lag to zero to avoid overfitting. The Appendix describes the prior distributions for all model parameters, along with their posterior mean, standard deviation, and 80 percent credible intervals. Key prior assumptions include: • Supply and Demand: The priors for κ and δ in equations (5) and (6) follow inverse gamma distributions with a mean of 0.2 and a standard deviation of 1. At these values, a 1% increase in activity driven by demand shocks raises inflation by 0.2 percentage points, 18

and a 1% supply-driven increase in inflation reduces GDP growth by 0.2 percentage points. These assumptions are broadly consistent both with estimates from DSGE models—see for instance Smets and Wouters (2007)—and with structural VAR models of the U.S. economy—see for instance Baumeister and Hamilton (2018). • Commodity Prices: The priors for χ , χ and χ are drawn from an inverse gamma S D H distribution with a mean equal to 2 and a standard deviation equal to 5. Thus, the prior imposes that commodity prices rise in response to expansionary demand shocks, contractionary supply shocks, and shortage shocks—but with a generous range of possible outcomes. • Shortages: The parameters θ and θ have inverse gamma priors with a mean of 0.2 S D and a standard deviation of 1. These priors assume that positive demand shocks and contractionary supply shocks lead to higher shortages. • Monetary Policy: The priors assume that there is a systematic positive response of interest rates to inflation and GDP, with short-run response coefficients of 0.1; the long-run response is assumed to be larger due to the autoregressive parameters in the rule itself, as captured by the coefficients ρ and ρ . 1,RR 2,RR • Lagged Coefficients: For the lag coefficients in B , we set the prior means so that the j variables exhibit first- and second-order autocorrelations of 0.65 and 0.10, respectively, with most lagged indirect effects set to zero (except for specific responses of inflation and activity to interest rate shocks). • Standard Deviation of Shocks: We calibrate the prior means for the standard deviation of the five structural shocks using a simple moment-matching exercise. Following Justiniano and Primiceri (2008), we try to generate plausible volatilities of the five variables in the VAR model. The calibration of the prior also targets three additional moments: (i) Shortage shocks should account for approximately 50 percent of the variance in the shortage index; (ii) commodity price shocks should account for approximately 50 percent of the variance in commodity prices; and (iii) demand-related shocks (traditional demand and monetary policy shocks) and supply-related shocks (traditional supply, shortage, and commodity price shocks) should each account for about 50 percent of the variance of inflation. The objective is to balance the contribution of demand and supply 19

factors to inflation as well as the contribution of shortages and non-shortage shocks to fluctuations in the shortage index.18 All prior standard deviations are relatively diffuse. Priors for key parameters are shown in Figure 10. All prior densities are intentionally wide to allow the data to update both the location and spread of these distributions. The Appendix shows that the prior-induced impulse responses have much wider confidence intervals than the corresponding posterior intervals, indicating that the data are informative about the model parameters. 5.3 Results We estimate the model using the algorithms in Dynare (Adjemian et al., 2011). The first 20 observations serve as a training sample, and we draw 20,000 posterior samples using a random walk Metropolis-Hastings algorithm. Parameter Estimates Figure 10 shows the prior distributions and the estimated posterior distributions for the key parameters of the model. The location and spread of all posterior distributions are substantially updated compared to the priors, revealing that the data and the structure of the model are informative about the structural parameters. The posterior mean for δ in the demand equation is 0.12, which implies that a supply shock that increases inflation by 1 percentage point leads to a decrease in GDP growth of about 0.1 percentage point. The posterior mean for κ is around 0.12—a demand shock that raises GDP growth by 1 percentage point leads on impact to an rise in inflation of around 0.1 percentage point. All told, the posterior distributions reveal a “flatter” supply curve—lower sensitivity of inflation to shifts in demand—and a “steeper” demand curve—higher sensitivity of inflation to shifts in supply—than assumed by the prior distributions. The top right panel of Figure 10 illustrates one of the underlying forces driving the inflationaryeffectsofshortagesinourmodel. Aftercontrollingforcurrentandlaggedmovements in all other variables, the posterior density measuring the effect of shortages in period t on inflation in period t+1—as captured by the parameter ρ —shows that most of its probability 1,πH 18We use a standard minimum distance procedure, by minimizing the Euclidean distance between VARimplied moments on the one hand, and empirical and additional target moments on the other. 20

mass lies in the positive region. This finding indicates a high likelihood that lagged shortages have a positive effect on inflation. Estimates for θ and θ are consistent with the notion that shortages respond to the D S business cycle, albeit to a smaller degree than assumed by the priors. The posterior estimate for both parameters is positive but relatively small. This finding echoes results from the variance decomposition: The share of the forecast error variance of shortages accounted for by shortage shocks is 66.8 at the posterior mean, higher than its prior value that was calibrated at 51 percent. Finally, the last row of Figure 10 describes the estimated response of interest rates to inflation and output, which is positive. Additionally, the parameter estimates suggest a relatively modest, negative response of interest rates to increases in shortages. Impulse Responses Figure 11 reports the impulse responses to the estimated shocks in the model. An exogenous increase in shortages leads to persistent rise in inflation alongside a slow, persistent decline in GDP. Quantitatively, a one-standard deviation shock to shortages leads to a material rise in inflation after one year of about 0.2 percentage points.19 As shortages return towards the baseline, inflation keeps rising and peaks about two years after the shock, remaining elevated thereafter. Commodity prices rise significantly. There is an initial, modest decline in interest rates that is quickly followed by an overall tightening in the monetary policy stance. Shortages also respond endogenously to other shocks: An expansion in aggregate demand or a contraction in aggregate supply tend to increase shortages. Similarly, commodity price shocks, such as those observed in the 1970s and 1980s, also drive up shortages.20 Finally, an interest rate shock leads to a reduction in shortages, alongside lower inflation, commodity prices, and GDP. All told, we find bidirectional causality: Exogenous shortage shocks trigger business cycle fluctuations, and traditional macroeconomic shocks generate movements in shortages. We highlight two important insights from our impulse response analysis. First, a shock 19Atthe posteriormean, theshare of the unconditional variance ofinflation accounted forby shortage shocks is 6.3 percent, compared to 1.4 percent at the prior mean. As shown later, this share is substantially larger in the post-pandemic period. 20Our identification strategy assumes that shortages respond within the quarter to shocks to commodity prices, while commodity prices respond to shortage shocks only with a one-quarter lag. This timing restriction helpsexplaintwokeypatternsinourimpulseresponses: Commoditypricesrespondlittletoexogenousshortage shocks, while shortages jump on impact following commodity price shocks. 21

that leads to an immediate jump in shortages has significant and persistent effects on GDP and inflation, with the maximum effects occurring between one and two years after the shock. The size and timing of these effects highlight the value of the shortage index as a leading indicator for macroeconomic variables. Second, compared to the traditional contractionary supply shock identified in the model, a shortage shock produces a smaller effect on inflation relative to its impact on economic activity. For a 1-percentage point rise in inflation, GDP growth bottoms out at 0.41 percent after a shortage shock, compared to a 0.26 percent decline after a supply shock. This observation suggests that, all else equal, a supply-driven rise in shortages is associated with more muted price pressures—for instance due to price controls, sluggish adjustment of prices, or other constraints—at the cost of a larger decline in economic activity. The Effects of Shortages throughout History Figure 12 presents the historical decomposition of the model. In the early 1950s, shortages were largely driven by the Korean War and by demand shocks associated with the rapid transition from wartime to peacetime economies after World War II. Shortages were muted throughout the 1960s. In the 1970s, the oil embargo by OPEC producers caused a significant rise in shortages beyond what commodity price shocks alone would predict. This shortage shock contributed to higher inflation and, to a lesser higher, slower economic activity. Shortages moved little in the 1990s and in the early part of the 21st century. To better highlight the macroeconomic impact of shortages in recent years, Figure 13 zooms in on the pandemic and post-pandemic period that began in 2020, revealing three distinct periods. The first period, running from 2020:Q1 through 2021:Q1, is characterized by shortage shocks driven by both tight labor markets and supply chain disruptions, along with depressed aggregate demand. Shortages are low because adverse shortage shocks are offset by depressed demand, and these two countervailing effects leave inflation close to baseline. In the second period, from 2021:Q2 through 2022:Q3, adverse shortage shocks continue to exert pressure on economic activity; these shocks occur in a context of stronger aggregate demand and rising commodity prices, leading to upward pressure on inflation. The third period, from 2022:Q4 through the end of 2023, is marked by a reduction in shortages, although shortages remain above their pre-pandemic levels. Importantly, as the estimated effect of shortage shocks on inflation is delayed and long-lived, past shortages continue to be a prominent contributor to 22

inflation well into 2023. All told, the exercise illustrates that shortage shocks account for a non-negligible share of the post-pandemic rise in inflation in the United States. Most of the discussion in policy and research about the drivers of inflation and economic activity since 2020 has been framed in terms of demand and supply forces. We connect to this debate by presenting in Figure 14 counterfactual paths for shortages, inflation, and GDP growth under alternative historical realizations of demand- and supply-side forces. For ease of exposition, we define supply-side forces as the sum of supply, commodity, and shortage shocks—the model shocks that move inflation and GDP in opposite directions—while demandside forces are the sum of demand and monetary policy shocks—the model shocks that move inflation and GDP in the same direction.21 Figure 14 illustrates that, according to the model, the post-pandemic increase in inflation is predominantly driven by supply forces, with demand forces playing a secondary but quantitatively relevant role. Supply shocks in 2020 and 2021 had the cumulative effect of increasing inflation by 6 percentage points and reducing GDP growth by nearly 2 percentage points. Negative demand shocks exerted downward pressure on both inflation and GDP growth in 2020; the reversal of these shocks contributed to a rebound in both indicators in 2021. Additional shocks in 2022 and 2023—capturing the tightening of monetary policy and more favorable supply conditions—led to a faster-than-predicted decline in inflation, while leaving, on net, a smaller imprint on economic activity.22 Our results differ from those of Giannone and Primiceri (2024), who use a VAR models identified through sign restrictions and find that demand factors predominantly drove the post-pandemic rise in inflation in the euro area and the United States. These authors emphasize that a key element of their findings is a flat aggregate demand curve; When the aggregate demand curve is flat, supply shocks exert only modest inflationary pressures, and large inflation movements must be driven primarily by demand shocks. Our estimates for the United States point to a steeper aggregate demand curve, implying a larger role for supply shocks in driving 21Each line in the figure represents a counterfactual path that assumes that: (i) no shocks are active through 2019, and (ii) shocks become active from 2020:Q1 through the end of the specified date. 22Theresultsforthepost-pandemicepisodelineupwiththebroaderresultofourestimatedVARthatsupply forces are relatively more important than demand forces in explaining the volatility of inflation throughout the sample. At the posterior mean, the share of the inflation forecast error variance that is accounted for by supply forces is 68.3 percent, higher than its value at the prior, which was calibrated to be near 50 percent. 23

inflation.23 The Appendix contains robustness checks, showing results for alternative VAR models estimated under different specifications, including a flat slope of the demand curve, a steep slope of the supply curve, and a specification that replaces GDP growth with the CBO measure of the output gap. Both the flat demand curve and the steep supply curve model exhibit a worse fit (in terms of marginal likelihood) than the baseline model, leading us to prefer our baseline specification. The contribution of shortages to inflation remains substantial across specifications, except when the demand curve is assumed to be flat. 6 Conclusions This paper introduces a novel monthly newspaper-based shortage index that covers the period from 1900 through the present. The index captures the intensity of shortages across sectors— industry, labor, food and energy—and exhibits pronounced spikes during periods of economic turmoil, such as the world wars, the oil crises of the 1970s, and the COVID-19 pandemic. Validation exercises confirm that the index accurately measures shortages and correlates strongly with other indicators of supply constraints. The extensive historical coverage of the index makes it a valuable tool for understanding long-term economic dynamics. Predictive regressions demonstrate that, throughout history, increases in the shortage index are associated with persistently higher inflation and lower economic activity. Furthermore, the relationship between inflation and shortages appears to be stronger during the COVID-19 pandemic. Structural VAR analysis further decomposes shortage movements into components arising from traditional business cycle shocks and exogenous innovations. Our results indicate that shortages cannot be explained solely by conventional demand, supply, commodity, and monetary shocks; surprise shortage innovations contribute importantly to inflation. Historically, the 1950s saw shortages driven mainly by demand factors, while the 1970s witnessed unusual 23Otherpapershaveanalyzedthepost-pandemicinflationusingtime-seriesandquantitativemodels. Bergholt et al. (2024) find that demand factors primarily drove the pandemic-era inflation, Ban´bura et al. (2023) and Ascari et al. (2024) attribute the bulk of pandemic inflation in the euro area to supply shocks. Comin et al. (2024) find that capacity constraints explain half of the increase in post-pandemic inflation. Our results underscore the dominant role of supply shocks, with demand shocks playing a smaller role. Our findings align with Shapiro (2024) and Di Giovanni et al. (2023), who find that an important role for supply factors and supply constraints in driving inflation in 2021 and 2022. Bernanke and Blanchard (2023) find that most of the post-pandemic inflation surge was the result of shocks to prices—reflecting a mix of supply-side shocks, strong aggregate demand, and sectoral shocks—given wages. 24

shortage shocks from the OPEC oil embargo. The COVID-19 period reflects a mix of supply and demand factors that led to a sustained inflationary impact, with supply shocks—including shocks to shortages—playing a predominant role. Overall, ourshortageindexprovidesacomprehensivelong-runperspectiveontheprevalence, drivers, and consequences of shortages in the United States over the past century. These findings underscore the complex interplay among various forces behind shortages and highlight their significant role in boosting inflation, offering valuable insights to policy makers and researchers. Although our analysis relies on linear models to estimate the effects of shocks to the shortage index, future research could extend this framework to account for nonlinear dynamics using both data and theoretical models. Recent work has highlighted how supply chain disruptions and input bottlenecks can amplify shocks through production complementarities and rigidities (Bonadio et al., 2021; Baqaee and Farhi, 2022). Similarly, shortages in the form of binding capacity constraints and pricing frictions can lead to non-linear inflation responses (Guerrieri et al., 2022; Di Giovanni et al., 2023). 25

References Adjemian, S., H. Bastani, M. Juillard, F. Karame´, J. Maih, F. Mihoubi, G. Perendia, J. Pfeifer, M. Ratto, and S. Villemot (2011): “Dynare: Reference Manual, Version 4,” Dynare Working Papers 1, CEPREMAP. Ang, A., G. Bekaert, and M. Wei (2007): “Do macro variables, asset markets, or surveys forecast inflation better?” Journal of Monetary Economics, 54, 1163–1212. Anthropic (2024): “Claude,” Conversational AI model developed by Anthropic, version from March 30, 2024. Available at https://www.anthropic.com. Ascari, G., D. Bonam, and A. Smadu (2024): “Global Supply Chain Pressures, Inflation, and Implications for Monetary Policy,” Journal of International Money and Finance, 142, 103029. Bai, X., J. Ferna`ndez-Villaverde, Y. Li, and F. Zanetti (2024): “The Causal Effects of Global Supply Chain Disruptions on Macroeconomic Outcomes: Evidence and Theory,” NBER Working Paper 32098, National Bureau of Economic Research. Baker, S. R., N. Bloom, and S. J. Davis(2016): “MeasuringEconomicPolicyUncertainty,” The Quarterly Journal of Economics, 131, 1593. Ban´bura, M., E. Bobeica, and C. Mart´ınez Herna´ndez (2023): “What Drives Core Inflation? The Role of Supply Shocks,” Working Papers 2875, European Central Bank. Baqaee, D. and E. Farhi (2022): “Supply and Demand in Disaggregated Keynesian Economies with an Application to the COVID-19 Crisis,” American Economic Review, 112, 1397–1436. Baumeister, C. (2023): “Pandemic, War, Inflation: Oil Markets at a Crossroads?” NBER Working Papers 31496, National Bureau of Economic Research, Inc. Baumeister, C. and J. D. Hamilton (2018): “Inference in structural vector autoregressions when the identifying assumptions are not fully believed: Re-evaluating the role of monetary policy in economic fluctuations,” Journal of Monetary Economics, 100, 48–65. 26

——— (2019): “Structural interpretation of vector autoregressions with incomplete identification: Revisiting the role of oil supply and demand shocks,” American Economic Review, 109, 1873–1910. Benigno, G., J. Di Giovanni, J. J. Groen, and A. I. Noble (2022): “The GSCPI: A new barometer of global supply chain pressures,” FRB of New York Staff Report. Bergholt, D., F. Canova, F. Furlanetto, N. Maffei-Faccioli, and P. Ulvedal (2024): “What Drives the Recent Surge in Inflation? The Historical Decomposition Roller Coaster,” Mimeo, BI Norwegian Business School. Bernanke, B. and O. Blanchard (2023): “What caused the US pandemic-era inflation?” American Economic Journal: Macroeconomics, forthcoming. Bonadio, B., Z. Huo, A. A. Levchenko, and N. Pandalai-Nayar (2021): “Global Supply Chains in the Pandemic,” Journal of International Economics, 133, 103527. Burriel, P., I. Kataryniuk, C. Moreno Pe´rez, and F. Viani (2023): “A new supply bottlenecks index based on newspaper data,” Banco de Espana Working Paper. Caldara, D. and M. Iacoviello(2022): “Measuringgeopoliticalrisk,” American Economic Review, 112, 1194–1225. Chen, L. and S. Houle (2023): “Turning Words into Numbers: Measuring News Media Coverage of Shortages,” Tech. rep., Bank of Canada. Comin, D., R. Johnson, and C. Jones (2024): “Supply Chain Constraints and Inflation,” NBER Working Papers 31179, National Bureau of Economic Research, Inc. Dao, M. C., P.-O. Gourinchas, D. Leigh, and P. Mishra (2024): “Understanding the international rise and fall of inflation since 2020,” Journal of Monetary Economics, 148, 103658, inflation in the COVID Era and Beyond. Di Giovanni, J., S. Kalemli-Ozcan, A. Silva, and M. Yildirim (2023): “Pandemic-Era Inflation Drivers and Global Spillovers,” NBER Working Papers 31887, National Bureau of Economic Research, Inc. 27

Federle, J., A. Meier, G. J. Mu¨ller, W. Mutschler, and M. Schularick (2024): “The price of war,” Tech. rep., Kiel Working Paper. Giannone, D. and G. Primiceri (2024): “The Drivers of Post-Pandemic Inflation,” NBER Working Papers 32859, National Bureau of Economic Research, Inc. Goldfarb, R. S. (2013): “Shortage, Shortage, Who’s Got the Shortage?” The Journal of Economic Education, 44, 277–297. Guerrieri, V., G. Lorenzoni, L. Straub, and I. Werning (2022): “Macroeconomic Implications of COVID-19: Can Negative Supply Shocks Cause Demand Shortages?” American Economic Review, 112, 1437–1474. Justiniano, A. and G. E. Primiceri (2008): “The Time-Varying Volatility of Macroeconomic Fluctuations,” American Economic Review, 98, 604–41. Lamont, O. (1997): “Do shortages cause inflation?” in Reducing Inflation: Motivation and Strategy, University of Chicago Press, 281–306. Liu, E., V. Smirnyagin, and A. Tsyvinski(2024): “SupplyChainDisruptionsandSupplier Capital in US Firms,” Cowles Foundation Discussion Paper 2402, Cowles Foundation for Research in Economics, Yale University. Newey, W. K. and K. D. West (1987): “Hypothesis testing with efficient method of moments estimation,” International Economic Review, 777–787. Ramey, V. A. and S. Zubairy (2018): “Government spending multipliers in good times and in bad: evidence from US historical data,” Journal of Political Economy, 126, 850–901. Shapiro, A. (2024): “Decomposing Supply and Demand Driven Inflation,” Journal of Money, Credit & Banking, forthcoming. Smets, F. and R. Wouters (2007): “Shocks and frictions in US business cycles: A Bayesian DSGE approach,” American economic review, 97, 586–606. Stock, J. H. and M. W. Watson (1999): “Forecasting inflation,” Journal of Monetary Economics, 44, 293–335. 28

Table 1: Search Query for the Construction of the Shortage Index Search Name Search Query Peak Month Energy Shortages (shortages N/5 energy) AND economics Dec-1973 Food Shortages (shortages N/5 food) AND economics Mar-1943 Industry Shortages (shortages N/5 industry) AND economics Aug-1942 Labor Shortages (shortages N/5 labor) AND economics Jan-2022 Articles articles – Topic Sets Components shortages shortage* OR bottleneck* OR scarcit* OR rationing* energy oil OR gas OR coal OR fuel OR fuels OR gasoline OR energy OR heating OR petroleum OR electricity OR refinery OR pipeline OR petrol food food OR wheat OR meat OR milk OR crop OR crops OR grain OR farm OR agriculture OR famine OR feed OR farmer OR farmers OR water OR fertilizer OR drought industry steel OR copper OR iron OR metal* OR automotive OR textile OR machinery OR equipment OR transportation OR railway OR airline OR freight OR shipping OR transit OR deliveries OR shipment* OR ships OR chip* OR semiconductor* OR infrastructure OR materials OR distribution OR car OR cars OR parts OR goods OR material OR auto OR computer OR ‘supply chain’ OR components labor labor OR workers OR job* OR work OR employment OR manpower OR worker OR staff OR professional* OR technician* OR staffing OR skills OR workforce OR personnel OR strike* OR union* economics economic OR industr* OR production OR manufactur* OR economy OR trade OR commerce OR business OR budget OR tax OR fiscal OR corporation OR market OR price OR capacity OR company OR demand OR sales OR factory OR wages OR suppl* articles the AND be AND to AND of AND and AND at AND in Note: The truncation character (*) denotes a search encompassing all possible endings of a word, e.g. “scarcit*” includes “scarcity” and “scarcities.” “AND” and “OR” are logical operators, N/5 denotes a proximity operator requiring words appearing within five words from each other. 29

Table 2: Largest Shortage Spikes, 1900-2023 Month Index Surprise Event (st.dev.) Jan-1903 174 2.84 Nationwidecoalshortages Dec-1906 191 2.58 ShortageofcoalandfreightcarsinMidwest Nov-1916 235 2.54 Nationwidecoalshortages Jan-1918 578 7.63 Fuelandcoalshortages Dec-1919 300 2.75 Fuelandcoalshortagesduetowarandstrikes Jul-1920 317 2.32 Freightcarshortageaffectscoalandsteeltransportation Aug-1922 281 3.24 Coalshortageduetostrikes Aug-1930 93 1.28 Droughtleadstofoodandwatershortages Jul-1934 134 1.70 StrikebyTeamstersunionsintheWestCoast Sep-1939 152 1.81 SteelshortageduetothebeginningofWorldWarII Aug-1941 516 3.30 War-relatedenergy,materialsandlaborshortages Jan-1943 1085 4.87 War-relatedoil,laborandfoodshortages Jan-1945 563 4.17 War-relatedwidespreadshortages Aug-1945 556 3.46 Laborshortagesattheendofwar May-1946 589 4.31 Strikesbycoalworkersandfuelshortages Jan-1948 460 4.01 Metal,fuelandfoodshortages Feb-1950 226 2.30 Coalshortagesamidstrikes Jan-1951 286 2.44 Laborshortagesduetodemandfromdefenseindustries Jan-1952 271 2.87 Nationwideandworldwideshortages Dec-1956 172 1.80 OilshortagesduetoSuezCrisis Dec-1973 1085 11.28 Gasolineshortagesdueto1973oilcrisis Jan-1975 267 2.02 Concernsaboutgasolinerationing Feb-1977 422 6.81 Carter’sappealonenergyconservation Feb-1978 158 1.59 Concernsaboutenergyshortages May-1979 579 4.81 Concernsaboutenergyshortages Aug-1981 153 1.18 Gasolineshortagesdueto1979oilcrisis Aug-1990 160 1.47 Concernsaboutenergyshortages Apr-2020 297 4.58 MedicalshortagesduetoCOVID-19pandemic Jan-2022 554 2.69 Laborshortages Oct-2023 213 1.20 UnitedAutoWorkersstrike Note: The table lists the 30 largest spikes in the shortage index. For this table, the spikes are identified using the residuals of an autoregression and a condition on local maxima described in footnote 9. 30

Table 3: Validation of the Shortage Index Articles Share of Total Validation Mentioning Type I Type II Set Newspaper Sample Actual Error Error Articles in Set Shortages Shortages NEAR Topic Words (S) 1.58% 872 817 6.30% – Not Shortages (A\S) 98.42% 298 1 – 0.33% All Shortages (B) 2.93% 334 284 14.97% – Note: Validation of the shortage index using a sample of newspaper articles used to construct the index, denoted by the set Shortages NEAR Topic Words (S), a sample of articles excluded from the construction of the index (denoted by the set A\S), and a sample of articles mentioning All Shortages (B). 31

Figure 1: Grouping of Newspaper Articles for the Construction of the Shortages Index (a) Grouping of Newspaper Articles (b) Word Clouds and Topic Classification of Articles Note: Panel (a) shows a diagram representing the set of articles used in the construction of the index. Panel (b)showswordcloudsassociatedwiththe4mostinfluentialtopicschosenbyLatentDirichletAllocation(LDA) analysis on a large set of articles. 32

Figure 2: The Shortage Index Index: 1900-2019=100 Oil, labor, Gasoline Food Shortages Shortages 1000 800 Coal Energy 600 Shortages Strikes Shortages COVID Shortages 400 Freight Car Shortages Labor Shortages 200 Sho C r o ta a g l es C S r u is e i z s Sh L o a rt b a o g r es Sh E o n r e ta rg g y es 0 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2024 Note: The figure plots the shortage index from January 1900 through December 2024. Updated charts and data can be found at https://www.matteoiacoviello.com/shortages.html. 33

Figure 3: Decomposition of the Shortage Index by Category Energy Shortages Food Shortages 400 250 200 300 150 200 100 100 50 0 0 1900 1920 1940 1960 1980 2000 2024 1900 1920 1940 1960 1980 2000 2024 Industry Shortages Labor Shortages 200 200 150 150 100 100 50 50 0 0 1900 1920 1940 1960 1980 2000 2024 1900 1920 1940 1960 1980 2000 2024 12-month moving average. Scale Equal to Contrib(utiaon) to TDotael Incdeox.mposition of the Shortage Index by Category (b) Topic Classification for the Index Note: Panel (a) shows shortage indexes by category, scaled so that the sum of the four indexes adds up to the total index shown in Figure 2. Panel (b) shows the evolution over time of the topic classification by a Latent Dirichlet Allocation algorithm for a large sample of articles used in the construction of the index. Data are from January 1900 through June 2024. 34

Figure 4: Phenomena Associated with Shortages 7 6 5 4 3 2 1 0 Geopolitical Natural Disasters Pandemics Labor strikes Price Controls Category P(Shortage) P(Category) P(Joint) P(Indep.) Geopolitical 1.27% 2.63% 0.084% 0.033% Natural Disasters 1.27% 2.85% 0.103% 0.036% Pandemics 1.27% 1.23% 0.047% 0.016% Labor strikes 1.27% 2.69% 0.104% 0.034% Price Controls 1.27% 0.80% 0.077% 0.010% ).borP tnioJ tnednepednI / .borP tnioJ lautcA( tfiL Comparison of Joint Probability Lifts Relative to Independence Assumption 7.61 Independent case (Lift = 1.00) 3.03 3.06 2.85 2.52 Note: The figure plots lift ratios, calculated as the ratio between the realized joint probability of events in 5 categories being mentioned in newspaper articles alongside shortages, and the probability that would be expected if these events and shortages were independent. The higher the lift ratio, the stronger the association between events and shortages. In the table, P(Shortage) is the frequency of news articles mentioning shortages; P(Category) is the frequency of news articles mentioning the specific category in each row; P(Joint) is the frequency of news articles mentioning both shortages and the specific category in each row; and P(Indep.) is the hypothetical joint probability under the independence assumption. 35

Figure 5: Shortage Index across Countries Standardized index (12 month average) 6 5 USA CAN 4 FRA GBR 3 GER JPN 2 1 0 −1 1900 1920 1940 1960 1980 2000 2024 Note: The figure shows the shortage index for the United States, Canada, France, Germany, the United Kingdom, and Japan. The country indexes are proportional to the share of newspaper articles mentioning shortages and the name of a foreign country and/or one of its major cities. Indexes are standardized to have mean zero and unit standard deviation over the period 1900 through 2023. 36

Figure 6: Comparison to Other Supply Constraints Measures 6 Shortage Index NY Fed GSCPI ISM Supplier Deliveries Index 4 Supply Bottlenecks Index USA 2 0 -2 -4 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 2024 All Indexes Standardized to have zero mean and unit standard deviation over 2000-2023 Period Note: The figure compares the shortage index to alternative measures of supply constraints from January1975throughJune2024. TheGSCPIistheNewYorkFedGlobalSupplyChainPressureindex (Benigno et al., 2022), measuring global supply chain conditions using data on both manufacturing and transportation costs. The ISM Supplier Deliveries Index is computed as the share of respondents reporting longer delivery times plus half the share of respondents reporting no change in delivery times. U.S. The Supply Bottlenecks Index is from Burriel et al. (2023). 37

Figure 7: Relationship between Shortages and Inflation and Activity 4-Quarters Ahead GDP Durable Consumption Nondurable Consumption Consumption Services Private Investment −0.4 0.0 0.4 0.8 1.2 Standardized Coefficient on Shortages selbairaV tnednepeD Effects on Inflation Effects on Activity −0.5 0.0 0.5 1.0 1.5 Standardized Coefficient on Shortages Sample: 2000:Q1 − 2023:Q4 1950:Q1 − 1999:Q4 1950:Q1 − 2023:Q4 Note: The figure reports standardized coefficients of predictive regressions of inflation and economic activity on the shortage index estimated on quarterly data. The dependent variable for each regression is the log difference between t+4 and t of the variable listed in each row (the price deflator for inflation and real quantities for growth). Each regression includes as controls the quarterly changes of the dependent variable and the associated economic indicator or price deflator, contemporaneously and with 3 lags. Data are quarterly. The full sample runs from 1950:Q1 to 2023:Q4. We also partition the sample into 2 periods: 1950:Q1 to 1999:Q4, and 2000:Q1 to 2023:Q4. Heteroskedasticity and autocorrelation consistent 90 percent confidence intervals are reported using shaded horizontal bars and computed according to Newey and West (1987). 38

Figure 8: Effect of Shortages on Inflation and Real GDP Growth (30-Year Window) (a) One-Year Ahead Inflation 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 −0.2 −0.2 −0.4 −0.4 −0.6 −0.6 −0.8 −0.8 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020 )dtS( noitalfnI (b) One-Year Ahead Real GDP Growth 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 −0.2 −0.2 −0.4 −0.4 −0.6 −0.6 −0.8 −0.8 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020 )dtS( htworG PDG laeR Significant at: 68% level 90% level 95% level Note: The black line shows the time-varying estimated effect of shortages on inflation (panel (a)) and real GDP growth (panel (b)). The estimates are based on regressions using rolling 30-year windows. In each regression, the dependent variable is the 4-quarter ahead difference in log real GDP per capita or the 4-quarter ahead difference in log GDP deflator. On the right-hand side, the main explanatory variable is our shortage index. As controls, we include the 1-quarter change in both log real GDP per capita and log GDP deflator, in quarter t plus 3 lags. Heteroskedasticity and autocorrelation consistent significance levels are computed according to Newey and West (1987). 39

Figure 9: Inflation and Its Forecasts 10 Headline Inflation AR AR+UR+Oil 8 AR+UR+Oil+SH 6 4 2 0 2019m1 2020m1 2021m1 2022m1 2023m1 2024m1 2024m12 Each month plots actual inflation against the expectation calculated 3 months before for the same period Note: The figure plots realizations of year-on-year, total consumer price index inflation with corresponding forecasts constructed 3 months earlier. The series are aligned so that the vertical distance between the plot of inflation and the forecast represents the forecast error. The model ‘AR’ predicts 3-month-ahead inflation using current inflation only. The model ‘AR+UR+Oil’ adds to the prediction model the unemployment rate and the 12-month change in oil prices. The model ‘AR+UR+Oil+SH’ adds shortages. 40

Figure 10: Prior and Posterior Densities of Structural VAR Parameters 20 10 10 Prior Posterior 10 Posterior Mean 5 5 Mean: 0.12 Mean: 0.12 Mean: 0.15 0 0 0 0.2 0.4 0.6 0.8 1 0 0.5 1 -0.2 0 0.2 0.4 0.6 1, h 10 40 0.6 20 5 0.4 Mean: 0.05 Mean: 0.13 0.2 Mean: 1.76 0 0 0 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 1 2 3 4 D S H 6 6 10 4 4 5 2 Mean: 0.09 Mean: 0.09 2 Mean: -0.08 0 0 0 -0.2 0 0.2 0.4 -0.2 0 0.2 0.4 -0.4 -0.2 0 0.2 0.4 Y H Note: Prior and posterior densities of selected parameters of the structural VAR model. See Section 5 for additional details and the Appendix for the full list of prior and posterior moments of the model parameters. 41

Figure 11: Effects of Shortages on US Activity and Inflation in the Structural VAR Model (a) Shock to Shortages s n0.6 Shortages tp p0.4 Inflation tp p 0 GDP Growth tp p Commod.Inflation 0.2 Interest Rate o ita ,e g ,e g ,e g 1 iv e D d ra d n a tS 0 0 . . 2 4 0 0 10 20 n a h c re tra u q -4 0.2 0 0 10 20 n a h c re tra u q -4 - - 0 0 . . 2 1 0 10 20 n a h c re tra u q -4 0.5 0 0 10 20 tp P -0 0 . . 1 1 0 0 10 20 Quarters Quarters Quarters Quarters Quarters (b) Shock to Demand s n Shortages tp p Inflation tp p2 GDP Growth tp p Commod.Inflation Interest Rate o ita 0.2 ,e g 0.6 ,e g ,e g 2 0.6 iv e D d ra d n a tS 0.1 0 0 10 20 n a h c re tra u q -4 0 0 . . 2 4 0 0 10 20 n a h c re tra u q -4 0 1 0 10 20 n a h c re tra u q -4 0 1 0 10 20 tp P 0 0 . . 2 4 0 0 10 20 Quarters Quarters Quarters Quarters Quarters (c) Shock to Supply s n 0.2 Shortages tp p Inflation tp p 0 GDP Growth tp p4 Commod.Inflation Interest Rate o ita ,e g 0.6 ,e g ,e g 0.4 iv e D d ra d n a tS-0 0 . . 1 1 0 0 10 20 n a h c re tra u q -4 0 0 . . 2 4 0 0 10 20 n a h c re tra u q -4 - - - 0 0 0 . . . 3 2 1 0 10 20 n a h c re tra u q -4 0 2 0 10 20 tp P0.2 0 0 10 20 Quarters Quarters Quarters Quarters Quarters (d) Shock to Commodity Prices s n0.2 Shortages tp p0.6 Inflation tp p GDP Growth tp p Commod.Inflation 0.4 Interest Rate o ita iv e D d ra d n a tS 0.1 0 0 10 20 ,e g n a h c re tra u q -4 0 0 . . 2 4 0 0 10 20 ,e g n a h c re tra u q -4 - - 0 0 . . 2 1 0 0 10 20 ,e g n a h c re tra u q -4 0 2 4 0 10 20 tp P 0.2 0 0 10 20 Quarters Quarters Quarters Quarters Quarters (e) Shock to Interest Rate s n 0 Shortages tp p 0 Inflation tp p GDP Growth tp p 0.5 Commod.Inflation 1 Interest Rate o ita ,e g ,e g 0 ,e g iv e D d ra d n a tS -0 - . 0 0 . 5 1 0 10 20 n a h c re tra u q -4 - - - 0 0 0 . . . 3 2 1 0 10 20 n a h c re tra u q -4 - - 0 0 . . 2 1 0 10 20 n a h c re tra u q -4 -0.5 0 0 10 20 tp P 0.5 0 0 10 20 Quarters Quarters Quarters Quarters Quarters Note: Impulse responses to structural shocks of size one standard deviation estimated using the structural VAR model described in Section 5. Solid lines denote the responses at the posterior mean. Shaded areas denote 80 percent confidence intervals. 42

Figure 12: Historical Decomposition from the Structural VAR Model: Full Sample Shortages s n6 Int.Rate Demand o ita Commodity Shortages iv4 Supply e D d ra 2 d n a tS0 1950 1960 1970 1980 1990 2000 2010 2020 Inflation tp 10 p ,e g 5 n a h c 0 re tra u -5 q -4 -10 1950 1960 1970 1980 1990 2000 2010 2020 GDP Growth 10 tp p ,e 5 g n a h 0 c re tra -5 u q -4-10 1950 1960 1970 1980 1990 2000 2010 2020 Commodity Inflation tp 50 p ,e g n a h c re 0 tra u q -4 -50 1950 1960 1970 1980 1990 2000 2010 2020 Interest Rate 10 5 tp p 0 -5 -10 1950 1960 1970 1980 1990 2000 2010 2020 Note: This figure displays the historical contributions of structural shocks to the evolution of the model’s endogenous variables. All variables are expressed as deviations from their sample means. 43

Figure 13: Historical Decomposition from the Structural VAR Model: 2020-2023 Shortages 8 s n Int.Rate Demand o ita 6 Commodity Shortages iv Supply e4 D d ra2 d n a tS0 2020Q1 2021Q1 2022Q1 2023Q1 2023Q4 Inflation tp p ,e 4 g n a 2 h c re 0 tra-2 u q -4-4 2020Q1 2021Q1 2022Q1 2023Q1 2023Q4 GDP Growth 10 tp p ,e 5 g n a h 0 c re tra -5 u q -4-10 2020Q1 2021Q1 2022Q1 2023Q1 2023Q4 Commodity Inflation tp p ,e 40 g n a h 20 c re tra 0 u q -4 -20 2020Q1 2021Q1 2022Q1 2023Q1 2023Q4 Interest Rate 4 2 tp 0 p -2 -4 2020Q1 2021Q1 2022Q1 2023Q1 2023Q4 Note: This figure displays the contributions of structural shocks to the evolution of the model’s endogenous variables since 2020. All variables are expressed in deviation from their 2019:Q4 value, and any shocks occurred before 2019:Q4 are set equal to zero. 44

Figure 14: Historical Contribution of Demand-side and Supply-side Forces: 2020-2024 (a) Demand-side shocks Shortages Inflation GDP Growth 1 2 10 1 0.5 5 0 0 0 -1 -0.5 -5 23Q4 -2 22Q4 21Q4 -1 -10 -3 20Q4 19Q4 -1.5 -4 -15 2020Q1 2023Q4 2020Q1 2023Q4 2020Q1 2023Q4 (b) Supply-side shocks Shortages Inflation GDP Growth 5 8 5 23Q4 4 22Q4 6 21Q4 3 20Q4 4 19Q4 2 0 2 1 0 0 -1 -2 -5 2020Q1 2023Q4 2020Q1 2023Q4 2020Q1 2023Q4 Note: The figure shows the contribution between 2019:Q4 and 2024:Q4 of demand-side forces (panel (a)) and supply-side forces (panel (b)) to shortages, inflation and GDP growth. Demand forces are the sum of demand and monetary shocks. Supply forces are the sum of shortage, commodity, and supply shocks. Variables are plotted in deviation from their 2019:Q4 value. Each line in the figure represents a counterfactual path that assumes (i) no shocks are active through 2019 and (ii) shocks become active from 2020:Q1 through the end of the specified date. For instance, the green lines labeled 19Q4 correspond to a scenario with no shocks in history; the purple lines labeled 20Q4 assume that shocks are active from 2020:Q1 through 2020:Q4; the yellow lines labeled 21Q4 assume that shocks are active from 2020:Q1 through 2021:Q4. 45

Appendix for: Measuring Shortages since 1900 A Data Sources This Section describes in detail the data used in the paper in addition to the shortage indexes presented throughout the paper and available at https://www.matteoiacoviello. com/shortages.html. The shortage indexes presented in the paper are constructed and analyzed running search queries using ProQuest Global Newsstream Collection, ProQuest Historical Newspapers, and the textual analytics tools of ProQuest TDM Studio (see https://tdmstudio.proquest.com/ home). Section 3 and the Appendix describe the results of regressions of selected price and wage indicators on the categorical shortage indexes. In addition to the shortages indexes, the series we use in these regressions and the corresponding data sources are: • Wages, measured by BLS Average Hourly Earnings in Total Private Industries (Haver mnemonics: LETPRIVA@USECON); • Producer Prices, measured by PPI: Intermediate Demand Processed Goods, seasonally adjusted (Haver mnemonics: PC1@USECON); • CPI: Energy (Haver mnemonics: PCUSEN@USECON); • CPI: Food (Haver mnemonics: PCUFON@USECON). Section 3 also compares our shortage index with alternative measures of supply constraints. Data sources are as follows: • The Global Supply Chain Pressure Index (GSCPI) published by the Federal Reserve Bank ofNewYorkanddescribedinBenignoetal.(2022)(Havermnemonics: GSCPI@SURVEYS); • The ISM Supplier Deliveries Index (SDI) produced by the Institute for Supply Management (Haver mnemonics: NAPMVDI@SURVEYS); • U.S. Supply Bottlenecks Index (SBI) from Burriel et al. (2023). The predictive regressions in Section 4 use data on real GDP, personal consumption expenditures (durables, non-durables, and services), and private fixed investment and their associated deflators. Data for all variables cover the period 1950:Q1-2023:Q4 and are taken from the Bureau of Economic Analysis National Income and Product Account Historical Tables. Variables are expressed in per capita terms dividing by total population from the U.S. Census Bureau, taken from Haver (Total Population: All Ages including Armed Forces Overseas; A.1

Haver mnemonics: POPT@USECON). The population series starts in 1952 and is backcast to 1950 using population data from Ramey and Zubairy (2018). In Section 4, we run the predictive regressions for GDP and its deflator on a rolling sample for the period 1900:Q1-2023:Q4. To this end, we backcast the GDP, GDP deflator, and populationdatadescribedinthepreviousparagraphusingthegrowthratesofthecorresponding series in Ramey and Zubairy (2018), which are available for the earlier period. The regressions in Section 4 also use as controls the following series: • Oil Prices, measured by Spot Oil Price: West Texas Intermediate. (Haver mnemonics: PZTEXP@USECON); • Commodity Prices, measured by the S&P Goldman Sachs Commodity Nearby Index (Haver mnemonics: GSCI@USECON); • OilPricesandCommodityPricesareexpressedinrealtermsbydividingfortheSeasonally Adjusted PCE Chain Price Index (Haver mnemonics: JCBM@USECON); • Wages, measured by BLS Average Hourly Earnings in Total Private Industries from the BLS (Haver mnemonics: LETPRIVA@USECON); • 10-year Inflation Expectations from the Federal Reserve Bank of Cleveland, retrieved from FRED (Fred mnemonics: EXPINF10YR). In the forecasting exercise of Section 4, we use the following variables: • CPI for All Urban Consumers (Haver mnemonics: PCUN@USECON); • Civilian Unemployment Rate: 16 years and over (Haver mnemonics: LR@USECON); • Oil Prices (Haver mnemonics: PZTEXP@USECON); • CPI for All Items less Food and Energy (Haver mnemonics: PCUSLFEN@USECON). The VAR analysis of Section 5 uses the following variables: • U.S. real GDP, from the Bureau of Economic Analysis; • CPI for All Items less Food and Energy (Haver mnemonics: PCUSLFEN@USECON); • Commodity Prices, measured by the CRB spot index for commodity prices (Haver mnemonics: PZALL@USECON); • The3-MonthTreasuryBillrate,SecondaryMarket(Havermnemonics: FTBS3@USECON). The robustness analysis constructs replaces real GDP with the output gap constructed using the output gap measure constructed by the Congressional Budget Office (Haver mnemonics: OGAPQ@USECON). A.2

B Additional Details of Audit Conducted with Claude AI We used the Claude AI Assistant to help with auditing the index. Claude was instructed to return a table of results, coding articles as 1 (shortage mentioned), 0 (shortage not mentioned), or 99 (unsure whether the existence of shortages was mentioned). In addition to the classification, Claude was asked to provide a brief explanation for each snippet’s coding. Before initiating the classification, we provided Claude with examples of how we would code the snippets and ensured that the training sample included false positives, such as mentions of the lack or end of shortages. The specific prompt was as follows: “I give you 872 snippets of text, each about 110 characters long. For each of them, can you tell me whether they mention current or prospective shortages, rationing, scarcity, or bottlenecks related to goods, labor, materials, food, or water? Just return a table with yes=1, no=0, unsure=99, and a brief explanation. For instance: • Article 1 mentions prospective shortages since it states that steel shortages will prevail in the near future, so it is coded as 1. • Article 2 says steel shortages caused a plant closure, so it is coded as 1. • Article 3 says the shortage of cars is crimping coal production, so it is coded as 1. • Article 4 mentions the shortage of cars, so it is coded as 1. • Article 10 mentions a shortage of workspace, so it is not related to a work shortage and is coded as 0. • Article 329 says no shortage of cars has been experienced, so it is coded as 0.” Although using AI for validation is not foolproof, we found that Claude performed as well as, if not better than, a human classifier. For example, Claude demonstrated an ability to extrapolate the context of a particular sentence to a country or person. In one instance, the snippet “Economy may be slowing, but Lowe is banking on labor shortages gradually leading to an increase...” was classified as 1 by Claude, with the explanation: “Reserve Bank [of Australia] expecting labor shortages to lead to wage growth.” The results of the audit are in Table 3. Out of 872 articles belonging to the set S, only 6.3 percent were deemed by Claude as false positives. Claude classified the snippet “...says that while there is not necessarily a shortage of people wanting to work in management” as 0 with the explanation “No shortage of people wanting to work in management.” Similarly, the snippet “a motive for mr. newt gingrich’s knife job, had no shortage of conspiracy theories, most leading to the...” was classified as 0 with the explanation “Speaker’s ouster sparked many conspiracy theories but not actual shortages.” In some cases, Claude classified as 0 articles that we would have probably classified as 1. For instance, the snippet “canada’s action today in temporarily suspending meat rationing” was classified as 0 since Claude gave more weight to the temporary suspension of the rationing rather than its existence. Of note, Claude classified only one snippet as “unsure,” and we found it had good reason to do so: the snippet was sampled from a short article in the Chicago Tribune in 1929 that reported short snippets of information on miscellaneous news items, a common practice in early 20th-century journalism. The snippet mentioned “Food shortage in Batanes islands. (Neither do we.)...” and featured incomplete comparison, lack of context, grammatical mismatch, and ambiguous meaning. A.3

C Additional Empirical Results Validation of the Categorical Shortage Indexes Table A.1 reports the coefficients from regressions where selected price and wage measures are regressed on the four categorical shortage indexes for industrial products, labor, energy and food. The dependent variables in these regressions are prices indexes that are more likely to be impact by shortages in these four categories: the 3-month ahead log difference of the Producer Price Index (PPI) for processed goods for intermediate demand, the Consumer Price Index (CPI) for energy, CPI for food, and the 12-month ahead log change in average hourly earnings (total private). Each regression includes three lags of the monthly changes in the dependent variable and one lag of the log change in industrial production.1 When the four categorical indexes are included one at a time, we find that the industry shortage sub-index predicts PPI materials inflation, the labor shortage index predicts earnings growth, and the energy and food shortage indexes predict energy CPI inflation and food CPI inflation, respectively. In other words, each index forecasts the relevant prices in the sectors most likely to be affected by that specific shortage. When all four indexes are included simultaneously, these results largely hold, with one exception: the food shortage index no longer predicts CPI food inflation. In addition, energy shortages predict PPI materials inflation and (weakly) wage growth. Additional Results from Predictive Regressions We explore the relationship between shortages, inflation, and economic activity using the following predictive regression: p (cid:88) ∆Y = α+βSHORTAGE + γ′X +ε (10) t+h t i t−i t+h i=0 where (cid:18) (cid:19) 400 Y t+h ∆Y = ln t+h h Y t represents the annualized log change of a variable of interest Y between period t and forecast t horizon h, and SHORTAGE denotes the level of the shortage index. The vector X contains t control variables. We use quarterly data from 1950 through 2023 from the National Income and Product Accounts (NIPA) for real per-capita GDP, personal consumption expenditures, and private fixed investment. Inflation for each category is measured using the associated price deflator. For GDP and its price deflator, we extend the sample back to 1900 using data from Ramey and Zubairy (2018). Data on total population also come from Ramey and Zubairy (2018), extended through 2023 using the POP series from FRED. For each price and economic activity indicator, we estimate equation (10) by ordinary least squares, with standard errors calculated following Newey and West (1987). As control variables, we include quarterly changes of the dependent variable and the corresponding economic indicator or price deflator, both contemporaneously and with three lags. For example, 1We use the 12-month change for earnings because earnings adjust more slowly than commodity prices or materials. Our results are similar when using the 6-month change in earnings. A.4

the regression for real GDP growth includes contemporaneous and lagged values of both real GDP growth and inflation (measured by the log change in the GDP deflator). Figures A.1 and A.2 report results for the one- and eight-quarter horizons, respectively. For ease of comparison, we report standardized estimates of the coefficient β. A standardized coefficient represents the change in standard deviation units in the dependent variable following a one-standard-deviation change in the explanatory variable. Does the Shortage Index Contain Information Beyond Traditional Indicators? One concern discussed in the paper is whether the shortage index contains information beyond that captured by other macroeconomic variables. We address this issue by augmenting our baseline regression with additional controls, including oil prices, commodity prices, wage growth, and inflation expectations. The inclusion of these controls only slightly attenuates the effect of shortages on inflation and activity, but the coefficients remain in large part statistically significant. See Figure A.3 for details. These findings suggest that our shortage index contains additional information beyond that captured by traditional macroeconomic indicators. VAR Results: Additional Detail Table A.2 provides details of the prior distributions for all model parameters, along with their posteriormean, standarddeviation, and80percentcredibleintervals. FigureA.4showsthatthe prior-induced impulse responses have much wider confidence intervals than the corresponding posterior intervals, indicating that the data are informative about the model parameters. Alternative VAR Specifications Figure A.5 illustrates the response of inflation to a shortage shock and the role of shortages and other shocks in explaining the post-2020 fluctuations in inflation under the baseline model and three alternative specifications. The contribution of shortages remains substantial across specifications, except when we impose the strong assumption of a flat demand curve (panel (b)), fixing the slope of the short-run inverse demand curve at δ = 3. Notably, a flatter demand curve assigns a larger role to demand shocks in accounting for the higher inflation observed in 2022. However, the fit of this model is worse than for the baseline model, as measured by the marginal likelihood of the models (the log marginal data density is −2574.1 for the flatter demand curve model compared to −2249.9 for the benchmark model). Assuming a steeper supply curve (panel (c)) only marginally reduces the role of supply shocks in explaining the post-pandemic inflation surge, but also lowers the marginal likelihood of the model relative the baseline (the log marginal data density is is −2,393.93). Finally, the results are unchanged relative to the benchmark model when activity is measured using the CBO measure of the output gap instead of 4-quarter GDP growth (panel (d)). A.5

Table A.1: Predictive Regressions with Categorical Shortage Indexes Wages PPI Materials CPI Energy CPI Food (1) (2) (3) (4) (5) (6) (7) (8) Labor 0.17∗∗∗ 0.11∗∗∗ -0.06 0.03 0.03 (4.55) (2.61) (-1.03) (0.61) (0.47) Industry 0.05 0.34∗∗∗ 0.23∗∗ -0.00 0.15 (0.71) (5.63) (2.00) (-0.01) (0.95) Energy 0.12∗ 0.21∗∗∗ 0.30∗∗∗ 0.29∗∗∗ 0.07 (1.90) (2.63) (5.24) (3.68) (.58) Food -0.06 -0.05 -0.02 0.15∗∗∗ 0.06 (-1.26) (-1.26) (-0.32) (3.01) (1.13) Controls Yes Yes Yes Yes Yes Yes Yes Yes Obs. 668 668 672 672 672 672 672 672 Note: The table reports standardized coefficients of predictive monthly regressions of selected prices and wages indicators on the categorical shortage indexes. The dependent variable for each regression is the 3-month log difference of the producer price index (PPI) for processed goods for intermediate demand, consumer price index (CPI) energy, CPI Food, and the 12-month log change in average hourly earnings (total private). Each regression includes as controls three lags of the monthly changes of the dependent variable and one lag of the log change in industrial production. The sample runs from 1964M1 through 2019m12. Heteroskedasticity and autocorrelation consistent t-statistics are reported in parentheses and computed according to Newey and West (1987). ∗ p<0.10, ∗∗ p<0.05, ∗∗∗ p<0.01 A.6

Table A.2: VAR Model: Estimated Parameters Prior Posterior Parameter Type Mean St. Dev Mean St. Dev 10% 90% σ Inv. Gamma 2.10 0.50 1.58 0.07 1.49 1.67 εd σ Inv. Gamma 1.20 0.50 0.67 0.03 0.63 0.70 εs σ Inv. Gamma 6.10 0.50 5.56 0.20 5.30 5.80 εc σ Inv. Gamma 0.50 0.50 0.57 0.03 0.54 0.60 εh σ Inv. Gamma 2.00 0.50 0.72 0.03 0.68 0.76 εr ρ Beta 0.65 0.10 0.68 0.04 0.63 0.73 1,HH ρ Beta 0.65 0.10 0.75 0.04 0.70 0.80 1,YY ρ Beta 0.65 0.10 0.89 0.03 0.85 0.92 1,ππ ρ Beta 0.65 0.10 0.83 0.03 0.80 0.87 1,CC ρ Beta 0.65 0.10 0.85 0.03 0.81 0.88 1,RR ρ Beta 0.10 0.05 0.02 0.01 0.01 0.03 1,πR ρ Beta 0.50 0.10 0.11 0.02 0.08 0.13 1,YR δ Inv. Gamma 0.20 1.00 0.12 0.05 0.07 0.18 κ Inv. Gamma 0.20 1.00 0.12 0.02 0.09 0.15 θ Inv. Gamma 0.20 1.00 0.05 0.01 0.04 0.06 D θ Inv. Gamma 0.20 1.00 0.13 0.04 0.08 0.19 S α Normal 0.00 0.10 0.02 0.01 0.01 0.03 C α Normal 0.10 0.10 0.09 0.03 0.06 0.13 Y α Normal 0.00 0.10 -0.08 0.05 -0.15 -0.02 H α Normal 0.10 0.10 0.09 0.05 0.02 0.16 π χ Inv. Gamma 2.00 5.00 1.49 0.21 1.23 1.75 D χ Inv. Gamma 2.00 5.00 3.93 0.53 3.19 4.60 S χ Inv. Gamma 2.00 5.00 1.76 0.56 1.10 2.46 H ρ Normal 0.00 0.10 0.04 0.02 0.02 0.06 1,HY ρ Normal 0.00 0.10 0.04 0.02 0.01 0.07 1,Hπ ρ Normal 0.00 0.10 0.01 0.00 0.01 0.01 1,HC ρ Normal 0.00 0.10 -0.09 0.08 -0.19 0.02 1,YH ρ Normal 0.00 0.10 0.12 0.05 0.06 0.20 1,Yπ ρ Normal 0.00 0.10 -0.00 0.01 -0.01 0.01 1,YC ρ Normal 0.10 0.10 0.15 0.04 0.10 0.21 1,πH ρ Normal 0.00 0.10 -0.04 0.03 -0.07 -0.01 1,πY ρ Normal 0.10 0.10 0.02 0.00 0.02 0.02 1,πC ρ Normal 0.10 0.10 0.07 0.10 -0.06 0.20 1,CH ρ Normal 0.00 0.10 0.05 0.08 -0.06 0.15 1,CY ρ Normal 0.00 0.10 -0.18 0.09 -0.30 -0.07 1,Cπ ρ Normal 0.00 0.10 0.03 0.05 -0.04 0.10 1,Rπ ρ Normal 0.00 0.10 0.01 0.03 -0.03 0.05 1,RY ρ Normal 0.00 0.10 -0.02 0.01 -0.03 -0.01 1,RC ρ Normal 0.00 0.10 0.07 0.06 -0.00 0.14 1,RH ρ Beta 0.10 0.05 0.13 0.06 0.06 0.21 1,CR ρ Normal 0.00 0.10 0.02 0.02 0.00 0.05 1,HR ρ Beta 0.10 0.05 0.04 0.02 0.02 0.07 2,HH ρ Beta 0.10 0.05 0.05 0.02 0.02 0.07 2,YY ρ Beta 0.10 0.05 0.06 0.02 0.03 0.10 2,ππ ρ Beta 0.10 0.05 0.03 0.02 0.01 0.05 2,CC ρ Beta 0.10 0.05 0.06 0.02 0.03 0.09 2,RR Note: The table shows estimated parameters of the model described in Section 5. The coefficients describing lagged response of π, y, c, and h to interest rates—ρ —are written with a negative sign 1,·R in the model specification of the VAR model to impose a negative lagged impact of interest rates on each of these variables. A.7

Figure A.1: Relationship between Shortages and Inflation and Activity One Quarter Ahead GDP Durable Consumption Nondurable Consumption Consumption Services Private Investment −0.5 0.0 0.5 Standardized Coefficient on Shortages selbairaV tnednepeD Effects on Inflation Effects on Activity 0.0 0.5 1.0 1.5 Standardized Coefficient on Shortages Sample: 2000:Q1 − 2023:Q4 1950:Q1 − 1999:Q4 1950:Q1 − 2023:Q4 Note: The figure reports standardized coefficients of predictive regressions of inflation and economic activity on the shortage index estimated on quarterly data. The dependent variable for each regression isthelogdifferencebetweent+1andtofthevariablelistedineachrow(thepricedeflatorforinflation and real quantities for growth). Each regression includes as controls the quarterly changes of the dependent variable and the associated economic indicator or price deflator, contemporaneously and with three lags. Data are quarterly. The full sample runs from 1950:Q1 to 2023:Q4. We also partition the sample into two periods: 1950:Q1 to 1999:Q4, and 2000:Q1 to 2023:Q4. Heteroskedasticity and autocorrelation consistent 90% confidence intervals are reported using shaded horizontal bars and computed according to Newey and West (1987). A.8

Figure A.2: Relationship between Shortages and Inflation and Activity 2-Years Ahead GDP Durable Consumption Nondurable Consumption Consumption Services Private Investment −0.5 0.0 0.5 1.0 1.5 Standardized Coefficient on Shortages selbairaV tnednepeD Effects on Inflation Effects on Activity −0.5 0.0 0.5 1.0 Standardized Coefficient on Shortages Sample: 2000:Q1 − 2023:Q4 1950:Q1 − 1999:Q4 1950:Q1 − 2023:Q4 Note: The figure reports standardized coefficients of predictive regressions of inflation and economic activity on the shortage index estimated on quarterly data. The dependent variable for each regression isthelogdifferencebetweent+8andtofthevariablelistedineachrow(thepricedeflatorforinflation and real quantities for growth). Each regression includes as controls the quarterly changes of the dependent variable and the associated economic indicator or price deflator, contemporaneously and with three lags. Data are quarterly. The full sample runs from 1950:Q1 to 2023:Q4. We also partition the sample into two periods: 1950:Q1 to 1999:Q4, and 2000:Q1 to 2023:Q4. Heteroskedasticity and autocorrelation consistent 90% confidence intervals are reported using shaded horizontal bars and computed according to Newey and West (1987). A.9

Figure A.3: Effects of Shortages on Inflation and Economic Activity 1-Year Ahead, with and without controls Baseline w/ Oil w/ Commodities w/ Wages w/ Inf. Exp. (10 Yr.) −0.5 0.0 0.5 1.0 1.5 Standardized Coefficient on Shortages selbairaV tnednepeD Effects on Inflation Effects on Activity −1.0 −0.5 0.0 0.5 Standardized Coefficient on Shortages Sample: 2000:Q1 − 2023:Q4 1950:Q1 − 1999:Q4 1950:Q1 − 2023:Q4 Note: The figure reports standardized coefficients of predictive regressions of inflation and economic activity on the shortage index estimated on quarterly data. The dependent variable for each regression is the log difference between t+4 and t of the variable listed in each row (the price deflator for inflation and real quantities for growth). Each row in the Figure shows how the standardized coefficients change with the inclusion of the controls listed. The full sample runs from 1950:Q1 to 2023:Q4. We also partition the sample into two periods: 1950:Q1 to 1999:Q4, and 2000:Q1 to 2023:Q4. Heteroskedasticity and autocorrelation consistent 90% confidence intervals are reported using shaded horizontal bars and computed according to Newey and West (1987). The regressions controlling for inflation expectations for the earlier sample start in 1982:Q1. Inflation expectations are from the Federal Reserve Bank of Cleveland via FRED. Heteroskedasticity and autocorrelation consistent 90% confidence intervals are reported using shaded horizontal bars and computed according to Newey and West (1987). A.10

Figure A.4: Effects of Shortages in the Estimated VAR Model: Prior vs Posterior (a) Shock to Shortages s n0.6 Shortages tp p Inflation tp p 0.2 GDP Growth tp p2 Commod.Inflation Interest Rate o ita iv e D d ra d n a tS 0 0 . . 2 4 0 0 10 20 ,e g n a h c re tra u q -4 0 0 . . 2 4 0 0 10 20 ,e g n a h c re tra u q -4 - - 0 0 . . 4 2 0 0 10 20 ,e g n a h c re tra u q -4 0 1 0 10 20 tp P -0 0 . . 2 2 0 0 10 20 Quarters Quarters Quarters Quarters Quarters (b) Shock to Demand s n 1 Shortages tp p 2 Inflation tp p 2 GDP Growth tp p Commod.Inflation 1 Interest Rate o ita ,e g ,e g ,e g4 iv e D d ra d n a tS-1 0 0 10 20 n a h c re tra u q -4 -1 0 1 0 10 20 n a h c re tra u q -4 -2 0 0 10 20 n a h c re tra u q -4 0 2 0 10 20 tp P -0 0 . . 5 5 0 0 10 20 Quarters Quarters Quarters Quarters Quarters (c) Shock to Supply s n o ita 0.2 Shortages tp p ,e g 0.6 Inflation tp p ,e g 0 GDP Growth tp p ,e g 4 Commod.Inflation 0.4 Interest Rate iv e D d ra d n a tS -0.2 0 0 10 20 n a h c re tra u q -4 -0 0 0 . . . 2 2 4 0 0 10 20 n a h c re tra u q -4 -0 - .5 1 0 10 20 n a h c re tra u q -4 0 2 0 10 20 tp P -0 0 . . 2 2 0 0 10 20 Quarters Quarters Quarters Quarters Quarters (d) Shock to Commodity Prices s n 1 Shortages tp p 2 Inflation tp p 2 GDP Growth tp p Commod.Inflation 2 Interest Rate o ita iv e D d ra d n a tS -1 0 0 10 20 ,e g n a h c re tra u q -4 -1 0 1 0 10 20 ,e g n a h c re tra u q -4 -2 0 0 10 20 ,e g n a h c re tra u q -4 0 2 4 0 10 20 tp P -1 0 1 0 10 20 Quarters Quarters Quarters Quarters Quarters (e) Shock to Interest Rate s n 0.2 Shortages tp p Inflation tp p 0.2 GDP Growth tp p 0.5 Commod.Inflation 1 Interest Rate o ita iv e D d ra d n a tS-0.2 0 0 10 20 ,e g n a h c re tra u q -4 - - 0 0 . . 4 2 0 0 10 20 ,e g n a h c re tra u q -4 - - 0 0 . . 4 2 0 0 10 20 ,e g n a h c re tra u q -4 -0.5 0 0 10 20 tp P 0.5 0 0 10 20 Quarters Quarters Quarters Quarters Quarters Note: Gray lines and areas, prior; Blue lines and areas, posterior. Impulse Response from the VAR: comparison of prior vs. posterior. Posterior shocks are one-standard deviation in size. Solid lines denote the response at the mean. Shaded areas denote 80 percent confidence intervals. The impact of the shocks at the prior mean has been normalized to match the impact effect of the estimated shocks at the posterior mean on (a) shortages, (b) GDP, (c) inflation, (d) commodities, (e) interest rates. A.11

Figure A.5: Comparison of baseline VAR model with alternative versions (a) Baseline Model s Shortages tp Inflation tp Inflation n o ita 0.6 p ,e g 0.3 p ,e g 1 8 0 S C R u o p mm iv0.4 n n 6 Dem e D a h 0.2 a h 4 Shrt d r a d n a tS 0. 0 2 0 5 10 15 20 c r e tr a u q - 4 - 0 0 . . 0 1 1 0 5 10 15 20 c r e tr a u q - 4 - - - 0 2 6 4 2 2020Q1 2022Q1 2023Q4 Quarters Quarters (b) Flatter Demand Curve (δ = 3) s Shortages tp Inflation tp Inflation n o ita 0.6 p ,e g 0.3 p ,e g 1 8 0 iv0.4 n n 6 e a 0.2 a 4 D h h d r a d n a tS 0. 0 2 0 5 10 15 20 c r e tr a u q - 4 - 0 0 . . 0 1 1 0 5 10 15 20 c r e tr a u q - 4 - - - 0 2 6 4 2 2020Q1 2022Q1 2023Q4 Quarters Quarters (c) Steeper Supply Curve (κ = 1) s Shortages tp Inflation tp Inflation n o ita 0.6 p ,e g 0.3 p ,e g 1 8 0 iv0.4 n n 6 e a 0.2 a 4 D h h d r a d n a tS 0. 0 2 0 5 10 15 20 c r e tr a u q - 4 - 0 0 . . 0 1 1 0 5 10 15 20 c r e tr a u q - 4 - - - 0 2 6 4 2 2020Q1 2022Q1 2023Q4 Quarters Quarters (d) Using Output Gap instead of 4-quarter GDP growth s Shortages tp Inflation tp Inflation n o ita 0.6 p ,e g 0.3 p ,e g 1 8 0 iv0.4 n n 6 e a 0.2 a 4 D h h d r a d n a tS 0. 0 2 0 5 10 15 20 c r e tr a u q - 4 - 0 0 . . 0 1 1 0 5 10 15 20 c r e tr a u q - 4 - - - 0 2 6 4 2 2020Q1 2022Q1 2023Q4 Quarters Quarters Note: Model (a) is the baseline model. Models (b) and (c) are estimated after fixing δ = 4 and κ = 1, respectively. Model (d) replaces 4-quarter GDP growth with the output gap constructed using the CBO measure of the output gap. A.12