Government payments and farm debt utilization during the pandemic

The COVID-19 pandemic posed significant challenges for farmers, intensifying pre-existing issues (Kauffman, 2013; Marchant and Wang, 2018; Zhang, 2021; Thilmany et al., 2022) while introducing new uncertainties in the supply chain, input costs, marketing strategies, and market demand (Johansson et al., 2021; McLaughlin et al., 2023). In response, the government support was augmented under the 2018 Farm Bill with several provisions aimed at increasing access to credit. Much of this support was gradually withdrawn as the pandemic receded, assuming it was no longer needed. However, access to credit has remained crucial for both the survival and growth of farm operations (Key, 2022), both during an economic challenge and afterward. Therefore, understanding the factors related with access to and use of debt can reveal ongoing vulnerabilities and risks that farmers face, which can limit access to capital and hinder growth. In this study, we contribute to this line of literature and the 2024 Farm Bill discussions by examining the differences in short-term debt use during the pandemic across farms of various sizes, as well as among farm operators of different races, genders, and levels of experience.

Farmers and agricultural operations exhibit varying levels of debt and financial vulnerability, influenced by operation size, operator diversity, market access, and management practices. Previous studies have highlighted disparities among farm operators and operation types in farm survival, financial performance and financial position (Fairlie et al., 2022; Jablonski et al., 2022; Giri et al., 2022; Giri and Subedi, 2024). Additionally, inequalities in credit access have been shown to persist in programs like the Paycheck Protection Programs (Sant’Anna et al., 2023), USDA loan programs (Escalante et al., 2018) and non-traditional lending (McDonald et al., 2021).

During the peak pandemic time, farmers received only $5.9 billion in Paycheck Protection Program (PPP) loans, far less than the total amount that the farm sector could have received (Giri et al., 2021b), indicating selective utilization possibly due to a lack of awareness about government programs. Subedi and Giri (2024) also highlight that producers use several sources of credit that their choice of lender varies based on the type of loan (real estate or non-real estate) and farm size. Furthermore, racial inequities in access to relief programs were also highlighted during the initial phase of the Coronavirus Food Assistance Program (CFAP) (Reiley, 2021), the primary relief program from USDA for producers impacted by the COVID pandemic. This highlights some selection in awareness about and access to the government programs and support made available during the pandemic. Without an investigation into the heterogeneity in previous utilization, ensuring balanced access to information and resources may continue to be difficult in future programs like the upcoming 2024 Farm Bill.

This study explores the heterogeneity in short-term debt use and the degree of indebtedness across farm sizes, and operator gender, race, and levels of experience. Specifically, by analyzing the USDA’s Agricultural Resource Management Survey (ARMS) data for 2020 and 2021, we examine how short-term debt use and amounts changed due to pandemic-specific provisions. We only consider loans with a repayment period of less than a year. In 2020 and 2021, ARMS also collected comprehensive information on COVID-19 government payments, making this a suitable data source to conduct a comprehensive comparison of access to short-term debt during the pandemic. We use a combination of machine learning methods and economic intuition in the form of post-double selection LASSO estimation. Therefore, we contribute to the relatively new literature that uses machine learning methods to predict several outcomes in the agricultural finance domain (Ifft et al., 2018; Chen et al., 2021; Storm et al., 2020; Lu et al., 2022).

Record-high government payments and near-zero interest rates, along with more guaranteed loans, provided higher liquidity and favorable credit conditions for the farm sector, expectedly alleviating financial challenges and increasing access to credit across all types of farmers. Analyzing whether such an effect was experienced unilaterally or differentially among farms by sizes and operator demographics is highly relevant for evaluating and considering the updates to debt-related provisions in the next Farm Bill. This enables policymakers to tailor future Farm Bill provisions to meet the specific financial needs of various farm operations, contributing to more effective agricultural policy design.

We find positive associations between government payments and short-term debt use for all farm types and farmer demographics, except for residence farms and non-white farm operators, suggesting their limited access to credit. A priori, one might expect similar limitations for beginning farmers, but their debt use is at par with, if not higher than, their more experienced counterparts. This may stem from previous Farm Bill provisions, demonstrating that special provisions improved access to credit for this group. Moreover, estimations indicate substantial variation in how government payments may have impacted the amounts of short-term debt across farms and farm operators. We observe insignificant changes in short-term debt when pandemic-related government payments increased during the pandemic. Conversely, lower interest rates during the pandemic are significantly associated with higher short-term debt use for all farm and farmer categories other than female operators. Lower interest rates are also associated with higher amounts of short-term debt for all farm and farmer categories except female and non-white operators. Our results also indicate that farms that could already access credit, like commercial farms, increased their short-term debt during the pandemic in association with lower interest rates. One explanation of these results is that while government payments provided stability, they did not necessarily encourage additional borrowing. Instead, the borrowing behavior appears more responsive to the cost of credit, as evidenced by the strong association between lower interest rates and increased short-term debt.

The Agriculture Improvement Act of 2018, also called the 2018 Farm Act, provides critical support to farmers who face limited access to credit through traditional lending markets. Title V of the 2018 Farm Act details these provisions related to farm debt and grants USDA the authority to establish credit-related programs. These programs specifically target individuals who encounter difficulties accessing credit from conventional sources. Furthermore, the 2018 Farm Act focuses on supporting credit access for beginning farmers, farmers with limited financial means, and military veterans (USDA ERS, 2019a). Although credit provisions by the government, including this Farm Act, are set to support all farmers facing financial challenges, recent literature highlights several disparities in access to financial support (Sant’Anna et al., 2023; McDonald et al., 2021; Ghimire et al., 2020), and farm debt is one such support.

The overall farm debt had been steeply rising before the pandemic; however, real farm debt fell for the first time in a decade during the pandemic. After the pandemic, farm sector debt continued to increase and was forecast to exceed half a trillion dollars in 2023 (Figure 1). This comes with a risk of higher farm leverages if farm incomes or asset values fall (Ifft et al., 2015), placing farmers in a critical position. As farm incomes are heterogeneous, the changes in aggregate farm debt before, during, and after the pandemic convey a distributional impact on farmers during the pandemic when financial stress was high. Farm sector debt includes both real estate as well as non-real estate debt, or alternatively is comprised of short-term and long-term loans, where debt use varies significantly by farm type and farm income. As the pandemic would primarily impact the short-term financial health of farm operations, this study focuses only on short-term debt.

Several farm and farmer characteristics are related to overall as well as short-term debt use. Ifft et al. (2014) highlighted the role of farm size, commodity specialization, and farmer characteristics in determining debt use and magnitude. They found that large-scale family farms (with annual gross cash income of $1 million or more) held the largest share of farm business debt; dairy farm operations, and those specializing in poultry had the highest average debt-to-asset ratios that decreased as the operator’s age increased. Additional factors that are related to debt use and composition are government payments (Kropp and Katchova, 2011; Katchova, 2005), interest rates (Martinez et al., 2023), financial vulnerability, and cash flow (Prager et al., 2018), status as a beginning farmer (Dodson and Ahrendsen, 2016; Tulman et al., 2016) and current leverage (Brewer et al., 2014). We include all of these factors and measure their relative importance by grouping farms by their size, financial status, and operator characteristics. This allows us to look at the role of government payments in addressing financial needs of each group separately, identifying the most effective utilization of government support, along with assessing debt use in light of the low interest rates during the pandemic.

Recent USDA data shows that producers had record-high net cash incomes of $149.3 billion and $202.2 billion in 2021 and 2022, respectively, which at $148.6 billion in 2023 is expected to remain higher than the pre-pandemic average (USDA ERS, 2023). Although record-high commodity prices boosted cash receipts, a major component was the rise in government payments. Direct government payments of more than $45 billion in 2020 were the highest on record in both nominal and real terms (Giri et al., 2022). Additionally, the Federal Reserve Bank maintained historically low interest rates during the pandemic in 2020 and 2021, as depicted in Figure 2. This suggests that the farm sector was financially stronger during the pandemic compared to the period before, placing them in a favorable position to access credit. Higher government payments improve liquidity while lower interest rates make it easier to access credit, refinance debt, and revise farm investments, making it a favorable time for the farm sector.

Figure 3, based on data from the August 31, 2023, release of the Farm Income and Wealth Statistics data product by the USDA’s Economic Research Service, illustrates the nominal government payments from 2000 through 2023. The 2020 direct payments reached record highs due to COVID-19 assistance, particularly through the Coronavirus Food Assistance Program (CFAP). In fact, CFAP alone distributed more payments than the total average payments to the farm sector over the preceding 20 pre-pandemic years. Notably, a study by Giri et al. (2021a) revealed that nearly all producers, approximately 97% based on cash receipts, were eligible to receive CFAP payments, making it one of the most comprehensive USDA programs in history. According to the USDA-ERS ARMS web tool (2023), 40% of all farm operations received some form of government payments in 2020. This marked a substantial increase from the 31% observed in the preceding year (2019) and the 34% in the subsequent year (2021). As eligibility was high across the board, one might expect that short-term debt needs would be substituted by the government payments, decreasing both debt use, and the amounts of short-term debt.

Moreover, beginning in March 2021, the Federal Open Market Committee (FOMC) of the Federal Reserve System initiated a series of eight consecutive rate hikes through March 2023 to combat elevated and persistent inflation. Figure 2 illustrates the short-term federal funds rate, which increased to 4.65% in March 2023, contrasting sharply with the less than one percent rate observed in March 2020. An increase in loan interest rates directly translates to higher interest expenses for farm operations, potentially leading to reduced loan demand. The USDA-ERS forecasts that interest expenses in 2023 will be the fastest-growing category among production expenses, with sector-level interest expenses estimated at $33.85 billion, reflecting a $6.21 billion increase, or 22%, compared to the $27.64 billion recorded in 2022 (Giri and Subedi, 2023). This further supports the notion of debt being substituted to some extent with government payments and presents a unique opportunity to examine the effects of interest rate fluctuations on farm debt. This underscores the importance of understanding the lasting implications of interest rate changes observed during the pandemic and the study period for farm operations, particularly for those considering new loans or lacking fixed interest rate arrangements.

This study uses 2020 and 2021 Phase 3 of the USDA’s Agricultural Resource Management Survey (ARMS), which annually collects data from a nationally representative cross-section of about 30,000 farms across 48 contiguous states in the US. ARMS is also the only annual source of information about the financial well-being of the farm sector and detailed farm operation characteristics. Since the 2020 and 2021 ARMS collected data on government payments, including those from COVID-19 related programs, it allows an empirical examination of the tradeoffs among various sources of credit and cash flow constraint relaxation faced by farmers. Using the pooled data for 2020 and 2021 also allows us to capture the plausibly unexpected changes in interest rates as they sharply fell in 2020 and started rising again in 2021. We use this data to explore descriptive statistics related to debt use, farm debt amount, and pandemic support as they vary by farm size, financial conditions, and operator demographics. Subsequently, we leverage a shrinkage model to identify the most important characteristics that determine debt use for various farm and farmer types.

Our main outcome variables, short-term debt use and short-term debt amount, both pertain to production or other loans with a repayment period of one year or less. Short-term debt use is a binary variable for identifying the farm operations that held any short-term debt owed on December 31st of the respective year, while short-term debt amount is a sum of the balance owed on all short-term loans. Both outcomes are denoted by $yi$ in estimation equations. We group all lender types together and only classify debt as either short-term or not to capture the changes in farm liquidity and short-term debt brought about by the pandemic and pandemic-related support.

We begin with two possible determinants of short-term debt use and the degree of indebtedness. First, the average interest rates on production or other loans with a repayment period of one year or less, denoted $inti$⁠, and second, the pandemic related government payments, denoted $govi$ for farm i. The pandemic related government payments include all the COVID-19 related grants, funds, and loans available for 2020 and 2021. These are only the financial payments received by operations (or operators). Both variables are included as our primary control variables, and referred to as model variables in the LASSO. In addition to the two model variables, several farm and farmer characteristics are considered; they are described in Table 1, and summarized by farm and operator categories in Tables 1 and 2. Table 1 separately presents the summary statistics for residence, intermediate, and commercial farms, while Table 2 summarizes the data by farmer gender, race and experience.

Additionally, the key factors we consider in our study can further be classified into supply and demand side factors affecting short term debt use. Supply factors refer to the conditions and constraints set by lenders that affect the availability of credit, such as interest rates, lending policies, and government support programs. Lower interest rates during the pandemic made credit more accessible, and government payment programs provided direct payments and guarantees that influenced lenders' willingness to extend credit. Therefore, our two model variables are both supply side factors. Their relationship with debt use during the pandemic is also related to the demand side factors. Demand side factors pertain to the needs and preferences of farmers seeking credit, including financial needs, risk tolerance, and awareness of available programs. The remaining farm and farmer specific control variables, described in Table 1, therefore, depict the demand side determinants of debt use. Additionally, government payments may also be considered a demand side factor since it can affect the need for loans.

Following USDA’s definitions, residence farms are those with gross cash farm income less than $350,000 and where the principal operator has either retired from farming or has a primary occupation other than farming; intermediate farms are those with gross cash farm income less than $350,000 and a principal operator whose primary occupation is farming; and commercial farms as those with more than $350,000 in gross cash farm income and nonfamily farms. Since race and farmer experience are key variables for our analysis, we drop the observations where these are missing, which were about 10% of the pooled sample. We further define beginning farm operators as those with 10 years or less of farming experience (Key and Lyons, 2019; Jablonski et al., 2022).

Table 1 presents the averages of debt use and debt amounts, our primary outcome variables, and several farm and farmer characteristics used as control variables in our analysis for all farms, along with the subsamples of residence, intermediate, and commercial farms. The summary statistics show that debt use increases with farm size and is much higher for commercial farms compared to residence or intermediate farms, perhaps due to higher reliance on leverage and easier access to credit for a larger-size farm operation, also indicated by their debt-to-asset ratios of 5%, 7%, and 22%, respectively. The average interest rate across all debt is the highest for commercial farms, suggesting a relation to their higher debt utilization. Several other variables highlight differences in farms, a notable one being a lower ratio of male operators for residence and intermediate farms compared to commercial farms in Table 1. This motivates an investigation into the differences in farms based on operator characteristics, which are presented in Table 2.

Table 2 displays the averages of the same farm and farmer variables for six categories of primary farm operators: male, female, non-white, white, experienced, and beginning operators. Among these, beginning farmers have the highest debt use of 31% compared to the lowest use by female and non-white operators, 19% and 23%, respectively. All of these individual farm operator categories have a lower debt use and debt-to-asset ratio than commercial farms. That is because select farms from each operator category comprise the financially vulnerable sample of farm operations.

During the pandemic, the average COVID-19 related government payments were highest for commercial farms at $38,846, with intermediate farms falling second at $2,460 (Table 1). Moreover, on average, male farm operators received almost twice as much COVID-19 government payments at about $6,070 compared to female operators who received $3,151, though their gross farm income and acreage was also much larger (Table 2). Beginning farmers rank fourth in the category of farm operators in the ranking of payments, only above female and non-white operators, at $3,930. When gross cash income is accounted for, commercial farms, male operators, white operators and experienced operators all received about $30 of government payment per $1,000 of gross farm income. Female farm operators received the highest amounts of support per $1,000 in their gross farm income at $50, followed by non-white and beginning operators and intermediate farms at about $35 with the lowest support being utilized by residence farms at $27 per $1,000 in gross farm income. Higher amounts of government payment utilization, therefore, does not seem to have a clear correlation with lower debt needs or use during the pandemic. Instead, the categories of farms and farm operators with the highest COVID-19 government payments were often the ones with higher debt use both overall and short-term.

Our study follows Belloni et al. (2014) and utilizes the double selection LASSO (Least Absolute Shrinkage and Selection Operator) for model selection followed by an OLS regression (StataCorp, 2021) in analyzing the role of government payments and interest rates on farm debt use during the COVID-19 pandemic period. The approach assumes only that farm debt use is related to interest rates and government payments, and the rest of the control variables that influence farm debt use are selected during the estimation process. The empirical approach combines the use of machine learning methods with economic theory to select an appropriate set of controls for short-term debt use and amount of debt. We begin with the following model signifying the data generating process

where the outcome variable y_i is either short-term debt use or the amount of short-term debt for farm i, and $inti$ and $govi$ are model variables referring to the interest rate and government payments, respectively. The variables $zi$ are other relevant control variables, and an unknown function $g(zi)$ represents the relationship between other control variables and debt use. The LASSO estimation does not assume which other control variables will be included in the model and in what functional form (for example, as interaction terms with other control variables). For our study, we include farm and operator characteristics that have typically been used in previous studies (crop insurance (Ifft et al., 2015), commodity specialization and farmer characteristics (Ifft et al., 2014), farm financials (Prager et al., 2018), race and gender (Escalante et al., 2018; Sant’Anna et al., 2023)) on farm debt use in the set $zi$⁠. Then, we use linear approximation of the function $g(zi)$ function with all the control variables in Tables 1 and 2 and all interactions between the categorical and continuous variables to construct an exhaustive list of controls $xi$⁠. This transforms the model to:

where $xi′αz$ is the linear approximation of $g(zi)$⁠, and $rzi$ is the resulting approximation error. $xi′αz$ contains all the control variables in Tables 1 and 2 and all interactions between the categorical and continuous variables, resulting in a fully saturated model with over 200 variables, from which LASSO would select the list of variables that belong in the final regression. Since there are a large number of variables in $xi$ (i.e. $xi$ is high dimensional), estimation and inference may be challenging. The assumption of sparsity, stating that only a small number of variables need to be selected in the linear approximation of $g(zi)$ to make the approximation error $rzi$ small relative to the estimation error $ϵi$ (Belloni et al., 2014). Therefore, we apply the double selection LASSO method to select the variables that approximate $g(zi)$ reasonably enough for $rzi$ to be relatively small.

The variables selection, or alternatively the approximation of $g(zi)$⁠, which belongs in the true unobserved model, is done in two steps. First, we estimate a linear LASSO of $yi$ on $xi$ (not including the main variables of interest, $inti$ and $govi$⁠) and denote the vector of estimated coefficients $θ$⁠. Specifically, we use the Rigorous LASSO estimator from Belloni et al. (2012) to solve the following optimization problem to select a subset of controls:

where $λ$ is the penalty level, $p$ is the number of variables in $x$,$n$ is the sample size, and $ljˆ$ are variable-specific penalty loadings for each $θj$ which are selected according to Belloni et al. (2012) to accommodate the heteroscedastic and non-Gaussian error. The LASSO penalty term in equation (3), $λn∑j=1p|ljˆ θj|$⁠, forces the coefficients of variables explaining the least variation towards zero, allowing us to select a subset from control variables $xi$ that have non-zero $θ$’s from the LASSO optimization.

In the second step, we estimate two additional LASSO models, one for each of the main model variables, $inti$ and $govi$⁠, respectively on all the control variables $xi$⁠. The optimization problem is solved again to choose the variables with the non-zero coefficients from the estimated coefficient vectors $δ$ and $γ$ resulting from LASSO estimation for $inti$ and $govi$⁠, respectively, on the control variables $xi$⁠. That is, we select the relevant control variables using LASSO from the following optimization problems:

and

Identifying the main variables associated with either interest rate, $inti$⁠, or government payments, $govi$⁠, explicitly and ensuring their inclusion in the following regression minimizes the omitted variable bias. In our study, these steps determine the main farm and operator characteristics that affect farm debt use in addition to interest rates and government payments, while also allowing for heterogeneity among different farm types.

Finally, we use $xi∼$ to denote the set of selected controls after the two selection estimations depicted by equations (3) and (4a and 4b), i.e. LASSO double selection, and estimate the following reduced form model using least squares:

Utilizing the LASSO two-step selection process combined with estimating equation (5) using OLS on a subset of controls represented by $xi ∼$ offers a significant advantage over either method alone. This approach allows us to observe and verify the relevance of the selected variables $xi ∼$ based on contextual knowledge and literature, ensuring that approximation error $rzi$ is small relative to $ϵi$⁠. Estimating an OLS regression on contextually motivated interactions involving over 200 variables without the LASSO selection would result in high variance in the OLS estimates, especially given the sample size and model complexity. Conversely, relying solely on LASSO without monitoring the double selection process might lead to the exclusion of essential variables, causing omitted variable bias. Therefore, our chosen post double selection LASSO OLS method enhances overall accuracy by combining the strengths of both LASSO and OLS approaches.

An additional contribution of our study is that the LASSO can determine which other farm and farmer characteristics in addition to interest rates and government payments, are selected as the main variables explaining the variation in farm debt use. This allows us to study the heterogeneity in the two outcome variables, short-term debt use and the amount of short-term debt, by grouping farmers into different samples based on farm size, farm financial status, and operator race, gender, and experience. We use the ARMS survey main weights during both LASSO steps. For the final post-double selection estimation in equation (5), we use the main weights to estimate the coefficients and the 30 replicate weights and the delete-a-group jackknife method for the re-sampling process to calculate the standard errors (Dubman, 2000; USDA-ERS, 2024).

The empirical results are discussed in the following sequence. We begin with a comparative analysis of the key factors associated with short-term debt use during the pandemic to understand any credit access constraints. We divide our sample into groups based on farm operation size, beginning versus established farmer status, and race and gender of the primary operator. This allows us to map patterns in debt use, highlighting the implications for Farm Bill provisions. Following that, we consider the variations in the short-term degree of indebtedness conditional on debt use to highlight whether government payments had comparable effects on credit access and financial constraints across diverse groups of farms and farmers.

Tables 3 and 4 present estimations of short-term debt usage, highlighting the differences in factors associated with debt use across different farm types and farm operator demographics. The results depict that higher government payments during the COVID-19 pandemic were associated with increased short-term debt use for all farm types and farmer demographics other than residence farms and non-white operators. This suggests that non-white farm operators may have limited access to credit, or, despite the available programs, they may have faced other barriers that present in a lower debt use for them. Interestingly, despite being a group typically expected to face credit limitations, beginning farmers displayed higher debt use, likely due to specific provisions in previous Farm Bills, demonstrating that special provisions could provide easier access to credit for this group ^[1]. This overall positive correlation suggests that farms that received higher government support payments during the pandemic may have been more likely to experience cash constraints, necessitating debt use to address their financial needs.

Results also show a positive association between interest rates and short-term debt use across all farm sizes and most operator types, except for female operators, where the association is insignificant. This aligns with the idea that farms using debt typically face higher loan interest rates. Notably, non-white operators had the highest significant increase in debt use with higher interest rates (6.34%), while beginning farmers (3.85%) and commercial farms (2.16%) showed the lowest significant increases. Higher debt use due to higher interest rates and accessible government support may stem from persistently lower net cash incomes (Giri and Subedi, 2023). It could indicate that farm operations could not avoid higher interest rates to relieve short-term liquidity constraints. Conversely, increases in debt use (for beginning and commercial farms, along with female operators) as interest rates increase but remain quite low during the pandemic, might result from either stable farm financials, as often observed among large commercial farm operations, or limited access to credit. If farm operators faced challenges in accessing credit, changes in interest rates would not be significantly related to their debt use, as observed among female operators.

A notable phenomenon presented in Tables 3 and 4 is the selection of variables explaining debt use and the relationship between debt use, COVID-19 government payments, and interest rates. LASSO selected different controls based on farm type (residence, intermediate, and commercial farms) and farmer demographics (male, female, white, non-white, experienced, and beginning), as shown in Tables 3 and 4. The variables selected by LASSO, significant or insignificant, highlight the most important characteristics that explain debt use in each farm type and farmer group. While gross farm income remains important for residence, intermediate and commercial farms along with male, white and experienced farm operators, its absence in explaining debt use for female, non-white and beginning farm operators highlights that once gender, experience, or race are accounted for, farm income no longer explains unique variation in debt use. Table 4 further verifies this by highlighting the differences in both the number and types of variables selected as most important in explaining debt use for white and non-white operators. Notably, fewer variables were selected for the farm and farmer categories traditionally expected to face higher challenges. Residence and intermediate farms had only 10 and 8 critical determinants for debt use, respectively, while commercial farms had 15 influential factors. Similarly, 2, 8, and 12 variables were selected for the subsamples of non-white, beginning, and female farmers, respectively, compared to 29, 27, and 23 variables for the categories of white, experienced and male farm operators, respectively. This underscores the significance of size, experience, and farmer demographics as crucial factors influencing debt use, diminishing the explanatory power of other determinants sometimes including farm financials.

The double selection by LASSO includes the variables that explain the most variation in the outcome. Even if some estimates are insignificant, the inclusion of each variable in underscores its importance in relation to the outcome (short-term debt use) and model variables (interest rates and government payments). This difference in variable selection by LASSO emphasizes the relative importance of factors such as farm financials, race, and gender in explaining variations in binary debt use and continuous debt utilization.

Overall, we also observe the interaction of supply and demand factors in explaining debt use across different farm sizes and operator characteristics. Larger commercial farms typically have better access to credit due to stronger financial positions and collateral availability, benefiting more from favorable supply factors like low interest rates. In contrast, residence and intermediate farms may face greater challenges due to stricter lending criteria and less collateral. Traditionally financially vulnerable groups might have higher demand for credit but face significant supply-side barriers due to perceived higher risk by lenders (Thilmany et al., 2022), even with government support programs in place. Non-white and female operators may often face additional supply-side constraints, such as discriminatory lending practices or lack of targeted support (Atkins et al., 2022; Sant’Anna et al., 2023) despite possibly having a similar or greater demand for credit compared to their white and male counterparts, which may explain their relatively lower debt use and less farm financial factors being selected by LASSO. Beginning farmers, despite higher demand for credit to establish their operations, might benefit from specific provisions in the Farm Bill that ease supply-side constraints.

Next, we examine the association between government payments and interest rates and farms’ level of indebtedness using the amount of short-term debt. The amount of short-term debt represents the dollar value (in millions of dollars) of loans with a duration of less than one year. The results, specifically estimated using equation (5) for the subset of farms that have obtained short-term debt, are presented in Tables 5 and 6.

Results in Table 5 present the estimations for all farms, residence, intermediate, and commercial farms while Table 6 includes estimations for male, female, non-white, white, experienced, and beginning farm operators. The findings indicate that increased pandemic-related government payments were not significantly associated with lower short-term debt for any farm type or farmer group. However, we observe a weak negative coefficient for residence and intermediate farms, and non-white farm operators. An association between lower debt and higher government payments indicates a substitution effect between the two sources of financial support, debt and government payments, in addressing a farm operation’s financial needs. The farms where government payments may have substituted short-term debt were residence and intermediate farms, and non-white operators.

Overall, once we restrict the estimation to a sample of those with short-term debt, the average change in short-term debt is not significantly associated with pandemic-related government payments during 2020 and 2021 for these farmers likely due to a high variation in the access to credit as well as the decision to use short-term debt. Some farms may have increased their debt use due to a combination of very favorable credit terms (low interest rates) and stronger financial conditions from pandemic-related assistance. The farms that did not increase their debt amounts could either have chosen not to do so, or they may have restricted access to credit.

On the other hand, the favorable credit conditions during the pandemic allowed farms that could arguably access credit more easily, like commercial farms and white or experienced farm operators to increase their short-term debt during the pandemic in response to lower interest rates. Our findings show that a one percentage point decrease in the short-term interest rate is significantly associated with an increase of $17,500 in short-term debt for the sample of all farms. Despite the relatively low average short-term interest rate for debt holders during the pandemic (as shown in Figure 3), our estimate suggests that a decrease in the already low interest rate was associated with a significant increase in the amount of short-term debt. Notably, this relationship was primarily driven by commercial farms, as evidenced by the fourth column in Table 5. For these farms, a one percentage point decrease in the short-term interest rate was associated with an average increase of $47,600 in short-term debt. Experienced and white operators followed this with an average increase in short-term debt of $19,000, and $18,900 for a percentage point decrease in the short-term interest rate, respectively. As the prevailing interest rate was very low, an increase in the interest rate may not necessarily increase interest expense significantly for commercial farms and white or experienced operators, perhaps allowing them to gain advantage of COVID-19 easing policies.

Consistent with previous estimations, LASSO selected a smaller number of controls for the subsamples of smaller sized farms, and female, non-white, beginning farm operators compared to commercial farms, and farms with male, white, and experienced operators highlighting the explaining power of size, gender, race, and experience in farm debt.

To ensure that the LASSO selection of control variables does not introduce regularization bias or omitted variable bias, we also estimated equation (5) using the entire sample of all farms (column 1 in Tables 1, 3, and 5). In this approach, the variables selected for the overall sample are used consistently across all subsamples, providing a uniform set of variables for all regressions. This consistency allows for direct comparison of coefficients across all estimations. The results for the two variables of interest in these models are shown in Table 7, which presents estimates using both the LASSO selection process and the same set of variables.

The estimated coefficients and their precision closely match the results presented in Tables 3–6. Given this similarity, one might question the necessity of using LASSO selection. The key advantage of LASSO is that it identifies the most parsimonious set of variables that explain sufficient variation in the outcome variable. In other words, when a variable is selected by LASSO, it highlights its relative importance in explaining variations in debt use.

Our analysis shows significant differences in the factors influencing debt use among different groups. As seen in Tables 3 and 4, fewer variables were selected for categories traditionally facing higher challenges. For example, residence and intermediate farms had only 10 and 8 critical determinants for debt use, respectively, compared to 15 for commercial farms. More notably, 2, 8, and 12 variables were selected for the subsamples of non-white, beginning, and female farmers, respectively, compared to 29, 27, and 23 variables for the categories of white, experienced and male farm operators, respectively. Therefore, this selection of variables highlights the significance of operator’s gender, race, experience and the size of a farm operation in their respective debt use. It also indicates the diminishing role of other major determinants like farm financials relative to farmer demographics.

Table 6 further illustrates these differences. For non-white operators, debt utilization is influenced by 13 variables, with only a few being financial factors. In contrast, for white farmers, 19 variables are relevant, including several financial characteristics. This suggests that debt utilization for white farmers is driven by numerous farm-specific financial factors, whereas for non-white farmers, race plays a more dominant role. Note that our results and LASSO selection does not serve as evidence of discrimination, rather, it underscores the homogeneity of specific farmer groups' financial characteristics, resulting in lower debt use for the entire demographic. This finding is essential for designing support programs such as the Farm Bill credit provisions. It suggests that systematic trends in debt utilization among groups like female and non-white operators may stem from barriers not reflected in farm financial statements.

Our study yields two critical findings that bear significant implications for the forthcoming 2024 Farm Bill discussions. Firstly, it sheds light on the debt use by farmers during the COVID-19 pandemic, providing valuable insights into potential liquidity constraints that led farmers to seek loans. This observation is underscored by the substantial associations between changes in interest rates and debt use among farmers, regardless of whether interest rates increased or decreased. The presence of these significant associations indicates that farmers, particularly those expected to face financial challenges, recognized the need for farm debt. Notably, our study reveals that non-white farm operators exhibited the most substantial increase in debt use when facing higher interest rates. We observed beginning farm operators and commercial farms recorded the smallest increases in debt use. This pattern suggests that their existing access to credit may have been associated with higher debt use even when interest rates were higher, potentially as a response to cash constraints during the pandemic.

Furthermore, the 2018 Farm Bill introduced crucial debt-related provisions that aimed to assist farmers who lacked access to traditional debt sources. This legislative support explains, in part, the positive association observed between interest rates and debt use across various farm typologies and farm operator categories. Notably, the Farm Bill significantly increased the total loan authorization levels for direct and guaranteed farm loan programs from $4.226 billion to $10 billion per fiscal year between 2019 and 2023, maintaining a consistent 30–70% allocation between direct and guaranteed loans (USDA ERS, 2019a, b). This expansion in available credit may have reduced competition between farm operations when seeking debt, fostering more equitable access. The contribution of our study lies in providing updated insights, specifically during the pandemic, characterized by historically low interest rates and record-high government payments. Overall, our study enhances the understanding of adapting policies related to interest rates, government payments, loan programs, and debt restructuring in the post-pandemic era to support farmers, manage financial risks, and optimize credit markets. These insights are particularly pertinent to the upcoming 2024 Farm Bill.