The impact of the COVID‐19 pandemic on retail dairy prices – Liu – – Agribusiness

1 INTRODUCTION

The coronavirus was first identified in humans in the 1960s and multiple strains have existed since then that have affected humans worldwide (CDC, 2020). However, the coronavirus disease that surfaced in December 2019 (coronavirus disease 2019 [COVID‐19]) quickly spread throughout the world leading to the World Health Organization (WHO) declaring it a pandemic on March 11, 2020 (WHO, 2020). As the virus spread throughout the United States, President Trump declared a national emergency on March 13, 2020, and states throughout the nation quickly began declaring their own state of emergencies. As attempts to control the spread of the virus were undertaken, many local and state government restrictions were put in place on business activities and the movement of residents. Stay at home orders were issued, schools and restaurants were closed, and in many areas only essential services, such as grocery stores, were permitted to stay open.

Goolsbee and Syverson (2020) found that individual choice, possibly tied to fear of infection, was more important than government restrictions. However, of relevance to our study, they also found that government restrictions significantly reallocated consumer activity away from restaurants and toward grocery and other food retailers. Thus, with the closure of schools and restaurants, consumers shifted their purchasing behavior. Consumers began limiting their purchases of food away from home and significantly increasing food at home purchases. Furthermore, Lusk and McCluskey (2020) found that sales in grocery outlets increased 90% relative to the year prior during the mid‐to‐late March period. Consumers were stocking up, potentially hoarding, as great uncertainty existed around the pandemic. This change in demand also created stress in the supply chain, further increasing worries about food availability and price increases (Iyer, 2020). Grocery stores also placed quantity restrictions on high demand items to support more customers and limit panic buying (Repko, 2020).

With a significant increase in demand and decreases in supply, one can consider the impact on food prices from an economic theory perspective. When demand increases, economic theory would predict an increase in price because of the rightward shift of the demand curve. The price of a good would further increase as the supply decreases and the supply curve shifts to the left. Thus, one would naturally expect higher food prices in response to the current pandemic. However, the current pandemic is on a scope and scale that has not been seen in our modern‐day economy. The closest one might come to an empirical understanding of the impact on food prices from such an event is to look at other sudden shocks in demand and/or supply. One such event type is a natural disaster. While common natural disasters are snowstorms, hurricanes, and earthquakes, these are isolated and geographically constrained events. This may, however, provide guidance on what might be expected.

Historically there has been limited literature on the impact of disasters on food prices with most studies focusing on measuring the impact on output (Cavallo & Noy, 2011). The potential impact from a market shock can vary in length, depending on the time period of interest. Parker (2018) illustrates how disasters may affect prices from a short‐run and medium‐run perspective. In the short‐run, there are concerns from natural disasters over death and injury to people (labor) as well as damage to infrastructure and destruction of harvests. This may ultimately affect infrastructure and harvesting that cannot operate or occur without workers and thus create supply issues. From a medium‐run perspective, there are concerns about longer lasting impacts on prices from decreased investment and lost infrastructure. While the current pandemic is not expected to impact physical infrastructure, there are impacts on labor, distribution, and the potential for decreased capital investment as a result of changing markets and changing consumer behaviors. Parker (2018) ultimately finds that different economies (advanced vs. developing) are affected differently, while different types of disasters also have differing impacts. In fact, middle to low income countries were found to have increases in food prices while other studies of high‐income countries experiencing similar disasters did not have significant price impacts (Abe et al., 2014; Cavallo et al. 2014; Doyle & Noy 2015).

Meanwhile, from a longer‐term perspective, Leibtag (2006) documented food prices in the region affected by Hurricane Katrina. This study found that prices increased a year later at similar rates to other regions that were not impacted, despite production cost increases and supply disruptions ranging from transportation to processing of food. Only time can determine what the long‐term impacts of the COVID‐19 pandemic might be, but short‐term impacts can be measured in conjunction by examining a change in price at the point in time when the disaster occurred.

Our research focuses on the impacts of COVID‐19 on retail prices of dairy products in the United States. We focus on dairy as a fundamental set of staple goods, including milk, cheese, yogurt, and butter. Furthermore, while the dairy industry experienced a demand shock from increased food at home purchasing, there was also a significant supply shock from disruptions in export markets and an inability to process milk for institutional buyers as a result of school and restaurant closures. The supply shock resulted in a large volume of milk being dumped and calls for extra culling of herds because of an inability of farmers to get their product to market (Huffstutter, 2020).

Using weekly data from the US Department of Agriculture (USDA) Agricultural Marketing Service (AMS), we analyze a change in dairy prices coinciding with COVID‐19 restrictions. Using a regression discontinuity (RD) design, we find that prices of all dairy products decreased by eight percent at the time that stay at home orders and changes in retail activities took place. When focusing on specific dairy categories, we find a varying degree of price declines, while many prices did not change. While this is contrary to initial expectations of increasing prices, it can be explained by consumers experiencing greater price sensitivity as a result of uncertainty and record setting unemployment, thus more elastic demand during this period. It can further be rationalized by sticky prices, retailers focusing on the health and safety of employees and customers, a focus on developing alternative marketing strategies (online ordering plus pickup and delivery services), as well as a retailer’s desire to maintain good relations with consumers by keeping relatively stable prices. While not primary to our research question, we also analyze store promotional advertisements for dairy products and find the expected decline in the number of store ads. With the disruptions in the supply chain and the sharp increase in demand, there was less need for stores to promote these products.

We present a model specification in the next section with justification of our estimation methods. Next, we outline the data and descriptive statistics used in the RD design. A discussion of results is then provided, where we also look at reasons for our findings. We finally conclude with ideas for future research once sufficient time has passed since the beginning of COVID‐19 impacts and data are available.

2 MODEL SPECIFICATION

In this analysis, we have a real‐world problem where we observe retailer’s pricing strategies on dairy products during a period of a public health emergency. To analyze the impact of this market shock, a natural experiment and difference‐in‐difference method seems ideal. These analyses typically require the construction of an appropriate control group that does not receive the treatment and can then be used as a benchmark to measure how other treated subjects respond. However, COVID‐19 was so widespread that all retailers and dairy product prices were subject to the potential impact, thus no product can be considered as untreated and a control group is not possible to construct. An alternative method to estimate the treatment effect when a control is not present is the RD design.

The RD design is a quasi‐experimental research design that allows for identification of causal effects of a treatment variable with good internal validity (Thistlethwaite & Campbell, 1960; Hahn et al., 2001). It is therefore like an experimental design, except that levels of the treatment variable are not randomly assigned by the researcher. Instead, there is a jump in the conditional mean value of the treatment variable at a known cut‐off in another variable. This cut‐off variable is called the assignment variable, which is perfectly observed and allows us to estimate the effect of the treatment as if it were randomly assigned in the neighborhood of the known cut‐off. By comparing observations lying closely on either side of the cut‐off, it is possible to estimate the average treatment effects in environments when randomization is not feasible.

The RD design has been used to estimate program effects in a wide variety of economic contexts, including labor markets, health, crime, environment, and other areas. Davis (2008) measures the effects of driving restrictions on air quality using high‐frequency measures from monitoring stations using the RD design. He finds that there is no evidence that the restrictions have improved air quality. Bento et al. (2014) examines the introduction of the Clean Air Vehicle Stickers policy in California with a RD design and finds that the welfare effects of the policy are negative. Using a RD design, Rabinowitz and Liu (2014) examine the causal effects of a change in state regulation on retail milk prices and finds that the regulation change has led to lower prices and increased consumer welfare. Nataraj and Hanemann (2008) use a RD design to investigate consumers’ water use when there is an introduction of a new price block in an increasing block pricing schedule for water. Exploiting longitudinal insurance claims data, and a cost‐sharing subsidy that has exempted copayment and coinsurance of healthcare services for children under the age of 3 in Taiwan, Han et al. (2020) use a regression discontinuity design to estimate its effect on children’s healthcare utilization.

We specify the RD design using the retail prices of dairy products as the outcome variable. In this quasi‐experimental design, the treatment is the time period when COVID‐19 restrictions were rapidly occurring throughout the United States, which is denoted by a dummy variable . An assignment variable, which is time t, is also assigned in this RD design. Although there is no clear single date that COVID‐19 restrictions were put in place throughout the entire United States, we choose to use the week of March 20–27, 2020, which is denoted by time c, as the cut‐off week. This is when most states were actively issuing restrictions on retail operations, schools and other institutions were closing, and state emergency declarations were being issued. Thus, we hypothesize that this week represented a significant shock to the US economy and more specifically, the retail sector and dairy industry. Therefore, for any dairy product in the market:

In other words, takes the value zero in all periods before the start of major restrictions and takes the value one in all subsequent periods.

While the distribution of the virus and specific state level restrictions were not uniformly distributed, all retailers and dairy products were subject to the impact of the pandemic and “participation” was mandatory. It is thus appropriate to implement a sharp regression discontinuity design (SRD). A nonparametric local linear regression is then performed to estimate the treatment effect of the regulatory change (Hahn et al., 2001).1 Specifically, we fit linear regression functions to the observations within a distance h on either side of the discontinuity point (the cut‐off point) and h as the bandwidth. In the SRD estimation, the optimal bandwidth is calculated following Imbens and Kalyanaraman (2009) to minimize MSE, or squared bias plus variance. A triangle Kernel is used and robust standard errors are calculated.

Of all the coefficients, β is the one of special interest because it captures the effect of COVID‐19 restrictions on retail dairy prices. The regression discontinuity method yields consistent estimates of β and can be interpreted as the causal effect of these restrictions.

3 DATA AND DESCRIPTIVE STATISTICS

To estimate the model, we use advertised prices for dairy products at major retail supermarkets outlets. The data are collected by the USDA AMS and reported in the Dairy Market News. This report is a compilation of surveys from nearly 150 retailers, comprising over 23,000 individual stores, with online weekly advertised features. Available data include prices and the number of individual stores’ online advertisement in a week for several dairy products, including cheese, cream cheese, milk, ice cream, and yogurt. It is important to note that because prices are collected directly from a retailer’s website that we do not have sales information, thus we cannot calculate sales‐weighted average prices, but only simple average prices. We also do not include organic dairy products due to limited data.

In this estimation we use 1 year of weekly data from June 2019 to June 2020. During this time period, we use the week of March 20–27, 2020 as a cut‐off date for the COVID‐19 pandemic shock, which identifies the start of the most significant COVID‐19 restrictions including stay at home orders and changes in retail activities. We also obtained a Class I raw milk price from AMS to account for the single greatest input in production of dairy products.

Summary statistics for dairy product prices are reported in Table 1. The top panel shows the mean overall price and standard deviation for all dairy products pooled together. The average price of a dairy product over the entire data period is $2.80 per unit, with an average raw milk price of $1.73 per gallon. We also break the sample into two periods, before and after the start of major COVID‐19 restrictions. The overall average price for all dairy products in our sample drops slightly from $2.80 to $2.77. Again, it is important to note that these prices represent the average shelf price of the products available to consumers and are not representative of the overall purchasing behavior of consumers. Except for the raw milk price, all prices are reported as a per unit price with the unit of measure noted when available. The raw milk price, typically reported in a per hundredweight (cwt), is converted to a per gallon price and is shown to have dropped significantly from $1.78 per gallon to $1.53 per gallon.

In the lower panel of Table 1, we break down the average prices by dairy product category and find that the decreasing trend on average prices nationwide is observed for most categories, including cream cheese, milk, flavored milk, ice cream, and yogurt. Most cheese products, however, present a slight increase in prices after the shock.

In addition to price information, the AMS also collects the number of stores that that provide online ads for each dairy product, which provides an indicator of retailer activities and promotional strategies before and after the shock. As shown in the top panel of Table 2, on average, there are 3423 stores that advertise dairy products each week. However, the volume of online ads dropped from an average of 3553 a week to an average of 2921 a week, showing a possible slowdown in retail promotion during the latter period. We also breakdown the volume of advertisement by product category. Among all dairy categories, ice cream, cheese, and yogurt are the most advertised products, but overall all dairy products show a drop in promotional activities after the shock.

4 RESULTS

Table 3 presents the RD estimates of the impact of the COVID‐19 restrictions. The top panel provides the estimated overall impact of all dairy products in the data and we find a −0.08 statistically significant coefficient. Because the dependent variable is in the log of price, the effect can be interpreted as an approximate percentage change in price. Therefore, the overall effect of the COVID‐19 restrictions in the dairy market is about an eight percent decrease in price.

To have a clearer understanding of the impact on prices, we further perform a series of RD estimations for each of the dairy categories separately,3 with the results presented in the lower panel of Table 3.4 We find the pandemic had a significant negative impact on the price of around a third of the products, such as butter, shredded cheese 8 oz, cream cheese, half‐gallon milk, sour cream, and Greek yogurt. Specifically, we find a 15.8% decrease in the price of butter. With an average unit price of $3.41 before the pandemic shock, a 15.8% decrease corresponds to an average unit price that is 53.8 cents lower after COVID‐19 shock. We also find a statistically significant negative effect of −8.5% of 8 oz shredded cheese, −13.5% of cream cheese, −28.8% of half gallon milk, −8% of sour cream, and −10.7% of Greek yogurt. For other products, such as block cheeses and ice cream, the pandemic had no significant impact on prices.

It is also useful to look at a graphical representation to visually inspect for a discontinuity at the point of interest. Figure 1 plots the results of local RD regressions estimating the effects of the COVID‐19 shock on retail prices of various dairy products. In particular, Figure 1 focuses on products categories that are estimated to be significantly affected by the COVID‐19 shock. For example, in Figure 1a, the red vertical line indicates the start of the COVID‐19 restrictions, and it shows a sharp drop in prices for butter at the point of our identified pandemic shock. Similarly, Figure 1b–f shows drops in retail prices of different magnitude after the start of the COVID‐19 restrictions for 8 oz shred cheese, cream cheese, half‐gallon milk, and Greek yogurt, and sour cream. All these graphs show a significant drop in prices after the start of COVID‐19 restrictions, which is consistent with our previous RD estimates presented in Table 3.

For some product categories, such as butter, half‐gallon milk, and Greek yogurt, we notice that there is a slight increase following the initial drop in prices due to the COVID‐19 shock. While for other product categories, such as cream cheese and sour cream, we observe a further drop in price after the initial price decrease. Therefore, even though all products were negatively affected by the COVID‐19 shock, the impact is heterogenous across different dairy products. These results are consistent with the feature of RD estimation, which captures the price impact around the neighborhood of the COVID‐19 shock, rather than a long‐term analysis.

The COVID‐19 impact estimates for products such as ice cream and many cheese products are not statistically different from zero. We also see this from the relatively continuous local regression functions shown in Figure 2 that are not significantly affected by the COVID‐19 shock. This is an important distinction when we see an overall average price that is lower after the restrictions were put in place. Thus, we can conclude that it is not the result of the COVID‐19 market shock event that is responsible for the price change shown during the longer‐term period.

One feature of the COVID‐19 pandemic is that it is difficult to pinpoint an exact start date, compared with other policy analyses. In this analysis, we use the week of March 20–27, 2020 as a cut‐off date for the COVID‐19 pandemic shock, which identifies the start of the most significant COVID‐19 restrictions including stay at home orders and changes in retail activities. The week of March 13–20, when the pandemic was declared by the WHO and the President, or the week after our choice when almost all states had announced the stay‐at‐home orders or restrictions are also possible candidate for the RD design. To test this variation, we run some robustness checks on the choice of the COVID‐19 shock impact dates. Specifically, we run the RD regression to estimate the overall impact on all dairy products in the data, using two other different impact dates: 1 week before and 1 week after our identified shock period. The results are presented in Table 4. In the results using the week of March 13–20 as the impact date, the COVID‐19 coefficient is negative but insignificant, suggesting that there is no significant price drop in the neighborhood of that period. When using the week of March 27–April 3, we still estimate a significant drop in prices, suggesting a continuing impact of the COVID‐19 shock. These robustness check implies it is the state level stay‐at‐home orders and restrictions that affect the retail prices and not the declaration of the pandemic.

In addition to retail price, we further examine the impact of COVID‐19 shocks on retail stores promotional activities by focusing on the volume of advertisements. Table 5 presents the RD regression estimates of the overall impact of all dairy products, as well as the impact by category. Figure 3 provides a visual presentation of the impact on the number of ads for categories that are estimated to be significantly affected by the COVID‐19 shock. Overall, the number of store ad promotions decreased by 27.1% in volume after the COVID‐19 restrictions and the impacts also vary in product categories. It is possible that with the disruptions in the supply chain and the sharp increase in demand, there was less need for stores to promote these products.

5 CONCLUSION

The COVID‐19 pandemic has changed modern life throughout the world. In the United States, many state and local governments have resorted to stay‐at‐home orders and retail restrictions in an attempt to curtail the spread of the virus. Even where government restrictions have not continued, private companies and individuals have continued to change their activities. While grocery stores are considered essential services, many changes have occurred in their daily operations in an effort to protect consumers and employees and to provide new fulfillment methods. Health guidelines continue to be issued and changes occur as more is learned about this virus. Even food and agricultural markets along the supply chain have experienced different market impacts and for different lengths of time. This paper explored the impact of the initial COVID‐19 restrictions on dairy prices and promotional advertising in the United States.

We used an RD design to estimate the impact on dairy prices in the neighborhood of the COVID‐19 shock. Overall, there was a negative impact on dairy prices as a result of this shock, however, when examining specific dairy products there were many that showed no significant impact. Even for those products where we see average price decreases in the latter part of our data period, we are able to conclude that it is not the result of the COVID‐19 market shock. Also using an RD design, we examined retail promotional advertising of dairy products during the same period. Here we find the expected negative impact. Given the uncertainty in supply and increases in demand there was less need for traditional store promotions to occur.

Our price results were not expected, as economic theory would predict a price increase would occur. However, one must recognize the complexity of the situation and economic uncertainty that existed. There are multiple reasons why prices of dairy products may have stayed the same or decreased at that time. Prices can often be sticky, that is, grocery stores may prefer to respond more slowly to the situation. This can also be an effort by retailers to maintain good customer relations by not being seen as taking advantage of the consumer by raising price. Alternatively, retailers may be concerned about being seen as violating anti‐price gouging laws. It is also possible that the impacts are implemented over a longer‐term and thus not immediate at the time of the initial shock, which would indicate that we have not captured the impact in our analysis. This may depend on the length of time that restrictions are in place, how consumers respond, and the severity of the supply chain disruptions. There is also the potential that economic conditions were such that an extremely elastic section of the demand curve was reached at that time.

There are some notable limitations to our research. Given the short timeframe from the COVID‐19 shock to performing this study, there was limited data available for analysis. A longer time series of data would help show longer‐term impacts, even though our question is focused on a point‐in‐time. Furthermore, we only use simple average prices, rather than a sales weighted average. Thus, the best data would be scanner data that would include both price and quantity information. Using scanner data would also eliminate the possibility of including prices of items that are not actually available to purchase due to supply limitations. A vast opportunity exists to extend this study in the future as data become available and longer‐term impacts can be measured.

The implications of our research are significant to policymakers. There was a need to quickly respond to the pandemic, not just from a health perspective but also from an economic perspective. With many businesses closed, unemployment reached record levels in the United States and large stimulus packages were put in place to alleviate issues of lost revenue and income. Furthermore, concerns existed over the price of food as the pandemic worsened (Kang & Bunge, 2020). While these issues go much further than just dairy consumption, understanding the initial impacts of the shock to the economy on the retail dairy sector helps add perspective on these impacts so that future disaster related responses may benefit from this knowledge.