The Paycheck Protection Program and small business performance: Evidence from craft breweries

It should be noted that the smallest loan size explored in Hubbard and Strain (2020) was US $150,000. In contrast, the average loan size in our sample was US $128,197 and the median loan size was US $56,711. The analysis presented here is a necessary extension of Hubbard and Strain, as it teases out the relationship between PPP funding and smaller business performance.

According to the SBA (2021a), the following businesses were eligible to apply for a first-round PPP loan: “(i) sole proprietors, independent contractors, and self-employed persons; (ii) Any small business concern that meets SBA’s size standards (either the industry size standard or the alternative size standard); (iii) Any business, 501(c)(3) non-profit organization, 501(c)(19) veterans organization, or tribal business concern (sec. 31(b)(2)(C) of the Small Business Act) with the greater of: 500 employees, or that meets the SBA industry size standard if more than 500; (iv) Any business with a NAICS code that begins with 72 (Accommodations and Food Services) that has more than one physical location and employs less than 500 per location”.

Fairlie and Fossen (2021b) demonstrate that sales in California plummeted 17% YoY during Q2 of 2020, but the analysis does not concern itself with PPP. Their analysis divides sales growth by different business types, and their results suggest that accommodation businesses and (alcoholic) drinking places were the two sectors that experienced the steepest decline in Q2 YoY sales.

The Brewers Association (2022c) defines “small” as producing less than six million barrels of beer per year, and they define “independent” as having less than 25% ownership from a business that is not itself a craft brewer.

Goolsbee and Syverson (2021) find that from March 1, 2020 to April 12, 2020, total foot traffic fell by 60 percentage points. Their methodology, which allows them to identify the causal effect of county-level governmental on foot traffic, suggests that shelter-in-place policies explained just 7 percentage points of the decline. Instead, much of the decline in consumer foot traffic was attributable to voluntary changes in behavior due to the perception about the risk of contracting COVID-19. In other words, businesses in counties with and without COVID-19 health policies both experienced, on average, substantial declines in consumer foot traffic during the early months of the pandemic. Those businesses operating in counties with shelter-in-place policies, on average, saw a decline in consumer foot traffic that was approximately only one-tenth larger than those in counties without the governmental mandates, holding all else constant.

When asked what made the craft beer industry more vulnerable to COVID-19 health policies and changes in consumer behavior relative to other sectors, Chief Economist of the Brewers Association Bart Watson wrote:

“The craft beer industry provides an interesting lens through which to study the economic effect of the COVID-19 pandemic, particularly due to the geographic and business model variations that occurred in performance. Craft brewers had high exposure to onsite hospitality shutdowns and shifts in consumer mobility, both due to the primary onsite business model of taprooms and brewpubs, as well as the much higher proportion of draught beer sales for most craft brewers relative to the overall beer industry” (Watson, personal communication, May 26, 2022).

Breweries were identified as permanently or temporarily closed in one of two ways. First, breweries could be identified as permanently closed by the Brewers Association, which was captured in the initial dataset the association provided. Then, Google searches were used to identify permanently closed breweries that did not report their operational status to the Brewers Association as well as temporarily closed breweries that were not identified in the initial dataset. Specifically, for each observation included in the dataset, we searched the brewery name and identified businesses that Google listed as temporarily or permanently closed. The internet searches and data collection were conducted in July 2021. Breweries that were identified as temporarily or permanently closed based on Google searches were then sent back to the Brewers Association for confirmation. The Brewers Association then analyzed the list and sent us an updated copy of the closures. Their revised set of closures was used in the analysis.

Note, any brewery that closed before April 3, 2020 was excluded from the analysis. Therefore, in removing the first 3 months of closures in 2020, this study underreports the closure rate. As many closures occurred after COVID-19 was declared a national emergency but before the SBA began distributing PPP loans on April 3, 2020. Using historical closure data from Brewers Association (2022d), the brewery closure rate in 2020 was approximately 4.8%, up from 4.2% in 2019 and higher than the long-run average of 2.7% over the past decade. The reader is directed to the Appendix accompanying this manuscript for further information on the differences between the closure rates reported in this manuscript and the ones reported on the Brewers Association website.

Data are available for 6304 (70%) of 8946 breweries for 2019, and 6892 (77%) for 2020. For observations without production volume, yearly production is treated as missing data. Year-over-year (YoY) changes in production volume from 2019 to 2020 are then calculated, allowing for an assessment of how production changed throughout the pandemic. In examining the change in YoY production, there are several outliers, mainly driven by breweries that were in the process of expanding production in the time of interest or opened later in 2019 (and their estimate does not reflect a full year of production). Therefore, in the following analysis, attention is limited to breweries that experienced a negative 100% to positive 100% change in YoY production from 2019 to 2020. For example, the median YoY change in production from 2019 to 2020 is a 12.5% decline, while the mean YoY change is a positive 26.3% change in production. A 0% change in production from 2019 to 2020 is at the 75th percentile, suggesting the distribution is skewed to the left with a long tail to the right. By construction, there is a necessary lower bound of − 100% change in YoY production (i.e., shutdown with zero production in 2020). An upper bound of + 100% change in YoY production is imposed to remove significant outliers. For example, a brewery could have opened in November of 2019, had 2 months of production, and this figure reflects their 2019 annual production. Suppose that the brewery remains open for all 12 months of 2020 and reports their 2020 annual production. Then it is reasonable to expect a 500% increase in YoY production from 2019 to 2020. For this reason, the upper bound limit of + 100% is placed on YoY production volume changes. After removing breweries that fail to meet the specified criteria, the sample contains 5877 breweries with production data, or 93% of the original 6304 observations with production data.

The Brewers Association provided data on the breweries that received funding from the Restaurant Revitalization Fund (RRF). Part of the American Rescue Plan Act of 2021 (Public Law 117–2), passed into law on March 11, 2021, the RRF was an additional government aid program run through the US SBA to support restaurants, bars, and other businesses that provide food or drink services (SBA, 2021e). The program, which ran from May 3, 2021 to July 2, 2021, supported more than 100,000 approved applicants and totaled US $28.6 billion (SBA, 2021f). The Brewers Association identified 1539 breweries that received RRF funds. By segment, the data suggests that 633 brewpubs, 257 microbreweries, 15 regional breweries, and 634 taprooms received RRFs from the SBA.

Data on PPP loan recipients can be accessed, here: https://​data.​sba.​gov/​dataset/​ppp-foia

One shortcoming with fuzzy matching is that breweries located in large plazas, malls, etc. may share a street address with another full-service restaurant but have different suite numbers. Loan recipients oftentimes failed to list their suite number on their application. Google Maps was used to make manual corrections to improperly matched locations.

Most unmatched observations included cideries, wineries, distilleries, pubs, and restaurants that were incorrectly coded into NAICS 312,120. Other businesses coded in NAICS 312,120 did not engage in alcohol production or distribution. Additionally, several observations that were coded into NAICS code 312,120 do not fit the Brewers Association’s definition of a brewery (e.g., kombucha brewers not registered with the Brewers Association). The most notable groups excluded from the analysis are breweries in planning or proprietor brewers. These groups were not included in the universe of breweries provided by the Brewers Association, so they are excluded from the analysis. While this may be seen as a limitation, these observations accounted for less than 20% of the unmatched observations, i.e., less than 3% of total PPP observations.

The statistics on the number of jobs supported by the PPP come directly from the PPP application, where applicants had to list the number of workers employed at the business. Unfortunately, the Brewers Association data did not contain statistics on brewery employment over time, meaning the study cannot observe changes in employment as an outcome variable. It is also worth noting that some PPP loan recipients may have more than one business specified under a parent company (e.g., a brewery is one of the trademarks of a larger company), overstating the number of jobs reported.

Specifically, observations are removed if (i) they are missing data in 2018, 2019, and/or 2020; (ii) the brewery experienced greater than a 100% increase in YoY production from 2018 to 2019 and/or 2019–2020; and (iii) they are statistical outliers that significantly skew the average. Breweries listed as producing 1 bbl of beer per year are also removed, as this may be evidence of an error in the industry production dataset where “1” signals an indicator of having produced in the corresponding year. Additionally, breweries producing above the 99th percentile of annual production in 2018 are excluded from the analysis. Specifically, 99% of the sample produces at or below 66,669 barrels of beer per year, while the remaining 1% of observations range from 66,784 to 2,175,784 barrels per year. Similar statistics hold for the 2019 and 2020 data, with 99th percentiles of 55,660 and 50,084 bbls per year, respectively. As such, the 2018 data is used as the production cutoff point.

Including state-level control variables in the analysis did not improve explanatory power, produced point estimates that were identical in magnitude to the preferred specification, and may be inappropriate given the inclusion of county-level fixed effects. We also consider the inclusion of local bar and restaurant policy over time instead of traditional FIPS codes. However, their inclusion is insignificant and reduces the degrees of freedom. As such, we elect to use the FIPS codes to control for unobserved, county-level heterogeneity. The relationship between local bar and restaurant policy and performance is discussed further in Sect. 5 and in the Appendix accompanying this manuscript.

It is reasonable to expect breweries that produce a higher volume of beer per year to benefit from economies of scale and have access to more technologically advanced equipment requiring less labor. Additionally, companies may have multiple locations, with one serving as their headquarters (i.e., primary location) and the other(s) as (a) secondary location(s). Finally, a control variable is included for whether a brewery received an RRF loan. Note that the distribution of RRF loans comes immediately before the data collection on open/closed status. This is important because it is possible that some breweries were temporarily closed in, for example, early-May 2021, and then opened when they received RRF funding in late-May 2021. With the data collection in July 2021, brewery operational status only observes whether a business was open in July 2021, not seeing that they were temporarily closed weeks before. The indicator variable controlling for RRF loan funding is included in the regression analysis to overcome this shortcoming.

The described specification was also run with the breweries that were closed as of July 2021. Unsurprisingly, results were more pronounced when these breweries were included as they saw the most substantial declines in YoY production. As a result, the approach described in the article is believed to be the conservative empirical decision that dampens estimated results.

The data used in the analysis is available at the Alcohol and Tobacco Tax and Trade Bureau (TTB) website: https://​www.​ttb.​gov/​beer/​statistics [last accessed October 25, 2022].

Brewpubs operate much like full-service restaurants, making local restaurant policy restrictiveness the ideal policy variable to analyze.

These results are exploratory, as Hale et al. (2021) note the difficulty of teasing out the causal effect of COVID-19 policies due to potential confounders and endogeneity concerns. Identifying the causal effect of local restaurant policies on performance lies outside the scope of this manuscript and is left to future research. However, we can offer initial insights suggesting that policies do not appear to be the leading driver behind declines in production. For a more detailed overview of the data analysis and procedures, please see the Appendix accompanying this manuscript.

Evaluating the causal effect of loan timing on performance in the quasi-experimental setting rest on two identifying assumptions: (i) loan timing did not affect loan demand and (ii) the firms before and after the gap in tranches are similar to one another (Doniger & Kay, 2022). To further assert the validity of the second assumption, we also run the analysis with a constrained framework that analyzes breweries that receive funding within 3 days of the gap between the first and second tranche. That is, the restricted framework explores the YoY production of breweries that receive funding in the last 3 days of first-tranche funding (April 14–16, 2020) and the first 3 days of second-tranche funding (April 27–29, 2020). The results are largely consistent with the findings reported in the main text (Tables 6 and 7). Specifically, tightening the window of analysis leads to (i) similar growth rates from 2018 to 2019 for both the treatment (8.0%) and control (6.7%); (ii) positive and statistically significant point estimates that are of similar magnitude; and (iii) a similar average treatment effect on the treated (ATT) when using propensity score matching. The 7-day window is our preferred specification, as the 3-day window cuts the experimental group from 1349 to 748 (55%) and has less variation in propensity scores across observations. Nonetheless, the results of the tightened analysis with the 3-day window are available in the Appendix accompanying this manuscript.

Attention is also restricted to breweries with production data from 2018 to 2020 to explore pre-trends. The sample begins with the 5555 breweries that were in operation as of July 2021 and had 2019–2020 YoY production volume changes within the bounds of − 100 to + 100% (analysis shown in Table 5). An additional restriction imposes an upper-bound limit on YoY production from 2018 to 2019 to mirror the restriction imposed on 2019–2020 YoY production changes. Lastly, outliers that significantly skew the sample mean are excluded. This includes breweries producing below the 5th percentile (≤ 100 bbls) and above the 95th percentile (≥ 7757 bbls) in 2018. Then, given the quasi-experimental setting, this portion of the study only analyzes observations with loan approval dates between April 10 and 16, 2020, or April 27–May 3, 2020.

County-level fixed effects are excluded given the sample size and diminished explanatory power. The models presented in Table 6 is also run with county-level fixed effects. The magnitude of the point estimate is similar (0.025), though it loses statistical significance. By including county-level FIPS codes, the analysis loses its explanatory power and increases its standard errors, leading to lower t-statistics. The inclusion of county-level fixed effects here may also not be appropriate given that of the 1346 observations across 577 counties, 321 counties (56%) are represented by a single observation.

A more detailed overview of the propensity score methodology and results are provided in the Appendix accompanying this manuscript.

The reader is directed to Caliendo and Kopeinig (2008) for an overview on the various matching methods as well as to Huntington-Klein (2022) for a discussion on the benefits and drawbacks of different propensity score matching procedures.

The PPP represents just one of the many policy levers pulled by the US government to minimize the economic damages from COVID-19. Each country had its own unique response to COVID-19, and Hale et al. (2021) tracked government responses to COVID-19 across more than 180 countries. Categorizing policy responses into three overarching themes (containment and closure, economic response, and health systems), the data tracks 19 types of policy responses throughout 2020. The Oxford COVID-19 Government Response Tracker data discussed in Hale (2021) is available at: https://​github.​com/​OxCGRT/​covid-policy-tracker [last accessed October 28, 2022]. While comparing government responses to COVID-19 is not the main objective of this manuscript, other studies have considered the effects of COVID-19 economic policy responses on small business performance and entrepreneurial behavior in countries such as the UK (Belghitar et al., 2022; Yue and Cowling, 2021), Germany (Block, Fisch, and Hirschmann, 2022a, b; Block, Kritikos et al., 2022a, b; Dörr, Licht, and Murmann, 2022), and China (Liu et al., 2022). These studies have shown that (i) firms adjusted their liquidity decisions in response to COVID-19; (ii) government policies often helped reduce the financial strain on small businesses; and (iii) government policies were sometimes not well targeted, leading to more substantial adverse effects on smaller businesses, entrepreneurs, and self-employed individuals. These findings generally align with those presented in this article.

Each market segment has its unique business model, indicative of different production levels, packaging decisions, revenue streams, etc. For brewpubs, food sales constitute a large portion of their revenue relative to the other market segments, and the food sales are largely driven by on-premise dining. When public health policies limited or shut down indoor dining, and consumer foot traffic fell due to the perceived risk of contracting COVID-19 (Goolsbee and Syverson, 2021), brewpubs saw a large decline in a primary revenue channel. Furthermore, a reliance on sales from indoor dining meant that brewpubs were also primarily selling their beer on-premise. While true that microbreweries and taprooms also rely heavily on on-premise consumption, brewpubs oftentimes have a less diverse portfolio of revenue streams. In other words, it is more common for the other market segments to have canning equipment, relationships with aluminum suppliers (upstream of the supply chain), and relationships with beer distributors and retailers (downstream), making the response to a shift in consumer behavior more likely. Without the necessary equipment and the relationships across the supply chain, brewpubs were particularly vulnerable. Thus, while state governments implemented policies alleviating some of the revenue declines—for example, allowing for to-go beer and brewery delivery—other barriers hindered a brewpub’s ability to pivot away from its original business model. Industry reports and anecdotal accounts have also highlighted the disproportionate effect the pandemic has had on brewpubs (Brewers Association, 2022d; Watson, 2022).

Given the number of craft breweries in the USA, the manual matching procedures described in this study were practical. With access to a verified dataset of producers from the largest craft beer industry group, the PPP observations were mapped to known breweries. However, even with less than 9000 total producers, the matching procedures proved extremely time-intensive. If future research wishes to examine the effect of PPP on a larger industry (e.g., full-service restaurants) or full sector (e.g., accommodation and food services), the matching methods are feasible but necessarily will be even more time- and labor-intensive. Additionally, researchers would need to obtain a verified dataset of existing full-service restaurants before COVID-19 and consider potentially miscoded observations to gain a more accurate representation of the impact of the PPP on the industry. While proprietary datasets (like the National Establishment Time-Series) exist, they have significant limitations which we were able to alleviate through partnering with the Brewers Association (i.e., being able to cross-validate closures, enhancing confidence that we accurately identified all firms that closed from the full universe of businesses in this specific industry and having non-imputed measures of production data). Thus, while analyzing a single, smaller industry has potential drawbacks for identifying the overall causal effect of the loan program, the practicality and internal validity of the methodology described in this paper offer substantial improvements over alternatives.

The stated objective of the PPP was to keep workers on payroll. But this study evaluates performance based on changes in annual production: a secondary policy outcome but a primary concern for small businesses. The current analysis is limited by data availability, and future studies that causally link receiving PPP funding to employment and production outcomes would significantly improve our understanding of the program’s effectiveness.