Multifamily property credit risk

The multifamily mortgage market plays a critical role in the U.S. housing finance system, supporting millions of rental units. At the heart of this network are government-sponsored enterprises (GSEs) — Fannie Mae and Freddie Mac — which play a pivotal role by purchasing, securitizing and guaranteeing multifamily loans. Informed by the lessons of the 2008 Great Recession — which shed a light on the vulnerabilities of GSEs’ mortgage-related products — and given today’s changing environment of economic policies, understanding the drivers of credit risk has become increasingly important. Lenders, investors and policymakers want to know the likelihood that a loan may default, but there is no crystal ball that can reveal exactly which loans will go bad. Alternatively, we can explore historical loan data and ask: What are the most important risk factors that might predict whether an owner of a multifamily property will default on a loan? Identifying these critical risk factors is difficult, as loan defaults are relatively rare in the multifamily housing market and the relationships between risks and their determinants are highly nonlinear by interactions among loan characteristics, property attributes and economic conditions.

Despite the challenges, we sought to uncover the most important risk factors by exploring Fannie Mae loan-level performance for multifamily property loan data from 2000 to 2024. We employed machine learning techniques and the University of Florida’s supercomputer, HiPerGator, to analyze this big dataset — which totaled nearly 69,000 loans and 4.9 million monthly records. Below we explain the work and highlight our findings. The bottom line: we found that underwriting variables at the loan level — such as loan-to-value ratio (LTV), debt service coverage ratio (DSCR), property’s total unit count and year built — are fundamental predictors of default risk, which is consistent with both theory and prior empirical studies. But our machine-learning models also highlight the significant contribution of broader local and national economic factors, which importantly provide meaningful predictive power even within Fannie Mae’s relatively high-quality loan portfolio.

Determinates of multifamily loan defaults

To explore what determines multifamily loan defaults, we conducted a “knowledge discovery in databases,” so-called data mining, which refers to algorithmic discovery of patterns in data, using a panel dataset of Fannie Mae’s multifamily loan performance combined with state and national economic data. Fannie Mae’s multifamily performance data includes loan features at origination and loan performance data month-by-month. Each mortgage has a number of detailed features at origination, such as borrower’s original LTV ratio, original balance, original interest rate, loan type, property type, ZIP code, property location and many more. This performance data includes: how many days behind payment the mortgage currently is, current balance, whether the mortgage is real estate-owned (REO) by a lender, in foreclosure, or has been paid off. In addition to loan data, we used state-level and national-level economic data to serve for post-origination changes in default risk.

We first examined the historical delinquency rates for mortgages acquired by Fannie Mae from 2000-2024. Figure 1 depicts the mortgage delinquency rates of multifamily performance in different delinquency levels. Loans 90 days or more delinquent are considered in serious delinquency and classified as default in our analysis. While the overall incidence of default remains low — peaking at approximately 0.6% — delinquency rates rose sharply during two major stress periods: the Great Recession (2007–2010) and the COVID-19 pandemic (2020–2021). More recently, delinquency rates have begun to rise again, starting in the latter months of 2024. Although rare, each default can trigger significant financial loss and even small increases in default rates are highly consequential for whole economy.

Source: Fannie Mae

Following theoretical studies (e.g. options-based and double-trigger models) and empirical studies of mortgage delinquency, we explored the strong factors among complex interplay of many elements driving the credit risk. We broke down three key categories of risk—loan characteristics, property fundamentals and macroeconomic conditions — to explain how each contributes to the probability of default.

1. Loan characteristics: LTV ratio, DSCR, loan size, loan age and mortgage spread
2. Property characteristics: Year built, property type and number of properties
3. Economic factors: Unemployment rate, rent growth and rental vacancy, and housing price index

Loan characteristics

LTV: Loan-to-value is the ratio of the actual unpaid principal balance of the mortgage loan to the combined value of all underlying properties. High LTVs — those above 100% — indicate that a mortgage is worth more than the property’s value, often referred to as negative equity, which increases the risk of default. Figure 2 illustrates the number of loans and the default rate across different loan acquisition LTV at ranges. With the high standards in mortgage acquisition, the LTV of loans in the Fannie Mae portfolio is capped at 80%. Even with positive equity, defaults still exist among loans, especially those with higher LTV. For instance, 2.6% of loans with LTVs between 60% and 80% were in default, according to the latest data available.

Source: Fannie Mae

DSCR: Debt service coverage ratio is the net cash flow to an actual or calculated principal and interest payment, measuring a property’s net operating income relative to its debt obligations. A healthy DSCR is one of the most crucial variables of mortgage performance. Fannie Mae data reveals, from the Great Recession, average yearly DSCRs of multifamily loans is on the rise and has increased 34% from 1.6 in 2009 to 2.15 in 2024. Figure 3 illustrates the relationship between underwritten DSCR and default rates. As expected, loans with lower DSCRs exhibit significantly higher default rates. The leftmost bar — representing loans with lowest DSCRs, including below 1.0 — are associated with sharply elevated default rate (approximately 3.3%), as borrowers have insufficient cash flow to comfortably cover debt obligations. As DSCR increases, default rates decline rapidly, reflecting improved borrower capacity to service debt. This clear pattern underscores DSCR’s strong predictive power as an early indicator of credit risk.

Source: Fannie Mae

Loan size: The size of a loan often serves as a proxy for fixed costs or the sophistication of the borrower to evaluate the credit risk. We expect that the higher the loan size, the larger the potential loss severity in the event of default, however the patterns of the original amount borrowed in Fannie Mae data shown in Figure 4 are not as straight forward as we expect. A large portion of the pool is loans with a loan size below $26 million, and these loans have highest rate of default (roughly 1.9%). One explanation may be that smaller loans are often associated with smaller multifamily assets. Market stresses can impair the borrower’s ability to service debt.

Source: Fannie Mae

Mortgage spread: Mortgage spread is the difference between a loan’s interest rate at origination and the contemporaneous 10-year Treasury yield, reflecting credit and market risk priced in at the time of underwriting. A higher spread typically reflects greater perceived risk at loan origination. Figure 5 illustrates the relationship between mortgage spread and propensity of default. As expected, loans originated with higher spread exhibit higher default rates. Loans with negative spread where original interest rates were below the 10-year Treasury yield benchmark show lowest default rate at 0.84%. As spreads increase, so does the default propensity: loans with widest spread (3%–9.1%), although fewer in number, have the highest rate of default at 3.73%.

Source: Fannie Mae, Federal Reserve

Property factors

Type: The multifamily business describes financing for residential buildings with five or more units, including apartment communities, cooperative properties, senior, dedicated student and manufactured housing communities. The data shows a variation in default rates across different multifamily property types and reveals significant variety in credit performance depending on property subtype. Student, senior and manufactured housing communities are higher credit risk property types, as these segments often face more concentrated tenant bases, heightened operational risks and greater exposure to economic or demographic changes (such as the COVID pandemic).

Year built: Figure 6 shows a nonlinear relationship between property’s year built and loan default rates. The oldest properties (built before 1950) have highest rate of default (3.4%-4.3%). The properties built between 1925-2000 exhibit gradual decline of default rates (3.4%, 2.7% and 2.1% in next three bars), probably reflecting stabilized operations, experienced ownership and lower leverage. Properties built in the years 2000 to 2008 show an increased default risk (2.5%), lending during the credit expansion and housing boom prior to the 2008 financial crisis. Despite newer construction, riskier underwriting practices likely contributed to higher defaults. The properties built since 2008 have a significantly lower default rate at 1.33%, the lowest among all cohorts. This likely reflects property’s modern design and lenders’ stricter post-crisis underwriting.

Source: Fannie Mae

Total units: The default rates by the total number of units acquired reveals an inverse relationship between property size and credit risk (Figure 7). Loans secured by smaller multifamily properties exhibit substantially higher default rates relative to those backed by higher-unit-count assets. Similar to loan size, this pattern can be explained by the fact that smaller properties are often managed by less experienced owners, are limited in financial flexibility and more vulnerable to tenant turnover. In contrast, larger properties often have professional management, get benefit from economics of scales and greater resilience to localized economic shocks.

Source: Fannie Mae

Economic factors

Economic growth is also a key factor in the performance of mortgage-related assets. In a growing economy, rising employment levels and household incomes support tenants’ ability to meet rental obligations, thereby sustaining property cash flows and debt service capacity. Homebuilding typically increases to meet the rise in demand. Mortgage delinquencies typically fall in an expanding economy and vice versa.

Unemployment rate: Job loss is a primary driver of tenant distress. In multifamily housing, high unemployment can lead to increased vacancies and rent delinquencies, putting stress on property income and, ultimately, the mortgage. Figure 8 shows the average state unemployment rate during a period when the number of loans was increasing. The mortgage default rates were highest (above 4%) in states with unemployment over 7.4%.

Source: Bureau of Labor Statistics

Income: This metric reflects the financial health of the local tenant base. Higher personal incomes support rent growth and payment stability. When personal income stagnates or declines, rental collections often suffer, putting pressure on landlords and increasing the likelihood of default. Loans in states with quarterly personal income lower than $1.2 trillion suffer a high rate of default, 2.8%-3.2% (Figure 9).

Source: Bureau of Economics Analysis

House price growth: Changes in home prices (a proxy for apartment value) affect the amount of equity that borrowers have in their properties. When property prices rise, owners build equity, making it easier to refinance or cover financial stress. Higher values can also lead to more cash-out refinancing, increasing loan balances. But when property prices fall, equity shrinks, refinancing becomes harder and default risk grows — especially if rents decline at the same time. As expected, Fannie Mae data shows the pattern of higher default rates (3.6%) for loans in the states with a low House Price Index (HPI) and default rates decrease when the index is higher (Figure 10). The HPI, maintained by the Federal Housing Finance Agency (FHFA), has a benchmark set to 1991’s index of 100.

Source: FHFA

Rent growth: Rising rents improve cash flow and bolster DSCR, making defaults less likely. Conversely, stagnating or declining rents can create financial strain, especially for highly leveraged properties. Rent dynamics often reflect deeper shifts in supply and demand. We used CPI rent from Bureau of Labor Statistics as a proxy for the rent growth rate. Fannie Mae data confirms that during the period of low rate of growth, the loans have a high rate of defaults, 3.8%-5.8% (Figure 11).

Source: Fannie Mae, Federal Reserve

Vacancy rate: High vacancy rates signal oversupply or weak demand. For landlords, vacancies mean lost income, while fixed expenses remain. Persistently high vacancies erode the ability to service debt, and over time, increase default risk. Fannie Mae data shows that the loans during the time of high national rental vacancy (9.6%-10.6%) have high default rates, above 4.9% (Figure 12).

Source: Fannie Mae, Federal Reserve

Findings: Default prediction with machine learning

Overcoming these limits, we employed machine-learning models to predict multifamily mortgage defaults and highlight variables with the most predictive power. The models^[1] use a technique known as explainable artificial intelligence to select the variables that have the most predictive power for the mortgage defaults. In Figure 13 we list the 20 most important factors. As we have shown, well-known important factors that predict multifamily mortgage default include various loan characteristics — i.e., acquisition LTV, mortgage spread, acquisition loan size and underwritten DSCR — and property factors, i.e., total units acquired and year built. This finding is consistent with research literature. Our novel contribution is identifying broader economic indicators, including state population, state housing price indices, rent growth (CPI rent) and other national economic indices, as important drivers of default risk. These additional factors enhance predictive performance even in the Fannie Mae dataset, where the loan pool is generally considered premium with low LTV and high DSCR profiles.

Based on SHAP value in our prediction models. SHAP = SHapley Additive exPlanations value from Lundberg and Lee (2017)

Sources: Federal Reserve Economic Data, Fannie Mae, Bergstrom Real Estate Center

1) Machine Learning models in use: Logistic Regression with L1 Regularization, XGBoost, and Deep Neural Network

Takeaways

Our analysis of Fannie Mae’s multifamily mortgage performance data offers several important insights. First, consistent with theory and prior empirical research, loan-level underwriting factors — such as LTV, DSCR, total units and year built — remain core predictors of default risk. Second, our machine-learning models reveal that broader macroeconomic variables also play a meaningful role. Notably, these local and national economic factors continue to provide predictive power for a relatively high-quality loan pool such as Fannie Mae’s. Looking ahead, the outlook for multifamily credit risk remains clouded by uncertainty. In 2025, rent growth is expected to remain positive but below long-term historical averages. At the same time, vacancy rates are likely to continue rising, according to Fannie Mae’s 2025 Outlook, and interest rates are projected to stay elevated. These factors are creating a challenging environment for borrowers with loans approaching maturity. For lenders, investors and policymakers alike, staying abreast of changing market dynamics and being flexible will be key in navigating the next market cycle of multifamily finance.

---

Anh Tran, Ph.D., is a real estate researcher at the UF Bergstrom Real Estate Center.

---

Data sources:

Data used in this article is from Fannie Mae’s multifamily performance database, U.S. Bureau of Economic Analysis, Bureau of Labor Statistics and the Federal Reserve Bank of St. Louis. The information was subject to data-cleaning process due to missing values and outliers.

References:

Archer, W.R., Elmer, P.J., Harrison, D.M. and Ling, D.C., 2002. Determinants of multifamily mortgage default. Real estate economics, 30(3), pp.445-473.

Mamonov, S., & Benbunan-Fich, R. (2017). What Can We Learn from Past Mistakes? Lessons from Data Mining the Fannie Mae Mortgage Portfolio. The Journal of Real Estate Research, 39(2), -262.

Pennington-Cross, A., & Smith, B. C. (2020). Early termination of small loans in the multifamily mortgage market. Real Estate Economics, 48(4), 1198-1233.