# Identifying Risky Multifamily Markets: A Forward Looking Approach

As real estate fundamentals have slowly improved with the sluggishly expanding economy, lenders have remained cautious. Commercial underwriting standards remain relatively guarded as lenders continue to focus on the implication of leases rolling to lower market rents and the implication of relatively slow improvements in market occupancies. As a means to mitigate such market risks, many lenders have focused on obtaining lending assignments on property in the Gateway markets. Higher quality properties with favorable leasing structures, combined with perceptions that liquidity is higher in the Gateway markets than other areas, have been attractive elements to lenders that wish to appropriately manage market risks. However, these same markets have demonstrated a degree of rent volatility over past history. As certain sectors of the commercial real estate market have begun to gain pace – the multifamily sector in particular – focus has shifted to risks related to future supply.

When identifying which markets may be more “risky” or subject to drops on rents and occupancies, lenders often tend to focus on the historic pattern of volatility. While historic measures of rent volatility may be useful in characterizing markets, they have several shortcomings when used to evaluate the overall volatility or riskiness of

The basic structure of the methodology is straightforward. The CBRE-EA team generates forecasts of rents and occupancy using fundamental supply and demand indicators, macroeconomic trends, and other data on local market conditions. For the multifamily sector, these forecasts include a quantification new projects that are expected to be completed over the short-term, along with employment, population, and macroeconomic indicators that drive the demand for multifamily space. From these supply and demand factors, a baseline model for rents and net operating income may be developed that shows the “most likely” path of future outcomes.

Of course, any forecast has a degree of error, or likelihood that the actual value may somewhat different that the forecast. To attempt to quantify this risk, CBRE-EA calculates the forecast errors from their model equations to determine a measure of uncertainty around the baseline forecast. This concept is best represented in Exhibit 1. The dark line shows that baseline forecast for a representative property’s net operating income or value over the forecast. Around this line are bands that measure one and two standard deviation forecast errors. A standard statistical “rule of thumb” indicates that the probability that the actual forecast value lies within the one standard deviation band, or “cone” is roughly 66%, while it is 95% within the two standard deviation band. Note that these bands widen as we move further out into the forecast horizon – we have less confidence that the actual value will fall within a narrow range of the baseline forecast. CBRE-EA develops a set of unique forecasts and their respective errors for each property type across a large number of metropolitan areas.

As you might expect, the trajectory, or baseline forecasts and their forecast error cones can vary significantly by market. In addition, the width of the cones is depending on a number of factors. In particular:

When identifying which markets may be more “risky” or subject to drops on rents and occupancies, lenders often tend to focus on the historic pattern of volatility. While historic measures of rent volatility may be useful in characterizing markets, they have several shortcomings when used to evaluate the overall volatility or riskiness of

*future*market performance. Historic measures fail to provide any information on the current point in the market’s real estate cycle, potential future supply conditions, nor an assessment of the trajectory of rents. To address these issues, several years ago, CBRE-EA developed a forward-looking methodology for quantifying market risks that incorporates a baseline econometric outlook for market performance in rents and occupancy, and a measure of risk around the achievement of forecast performance^{1}.The basic structure of the methodology is straightforward. The CBRE-EA team generates forecasts of rents and occupancy using fundamental supply and demand indicators, macroeconomic trends, and other data on local market conditions. For the multifamily sector, these forecasts include a quantification new projects that are expected to be completed over the short-term, along with employment, population, and macroeconomic indicators that drive the demand for multifamily space. From these supply and demand factors, a baseline model for rents and net operating income may be developed that shows the “most likely” path of future outcomes.

Of course, any forecast has a degree of error, or likelihood that the actual value may somewhat different that the forecast. To attempt to quantify this risk, CBRE-EA calculates the forecast errors from their model equations to determine a measure of uncertainty around the baseline forecast. This concept is best represented in Exhibit 1. The dark line shows that baseline forecast for a representative property’s net operating income or value over the forecast. Around this line are bands that measure one and two standard deviation forecast errors. A standard statistical “rule of thumb” indicates that the probability that the actual forecast value lies within the one standard deviation band, or “cone” is roughly 66%, while it is 95% within the two standard deviation band. Note that these bands widen as we move further out into the forecast horizon – we have less confidence that the actual value will fall within a narrow range of the baseline forecast. CBRE-EA develops a set of unique forecasts and their respective errors for each property type across a large number of metropolitan areas.

As you might expect, the trajectory, or baseline forecasts and their forecast error cones can vary significantly by market. In addition, the width of the cones is depending on a number of factors. In particular:

- A market with wide historic swings will tend to generate wider cones, or bands in the future, unless the model can explain the swings accurately. Therefore, a market that tends to have a high amount of historic volatility will also tend to have a high level of forecast risk.
- In general, areas where it is difficult to explain supply and demand trends, the model will have a poor fit and will tend to generate wider forecast cones. In addition, if there are low quality data, and there are limited variables that accurately capture what drives the market, there will be a tendency to have wider forecast cones.

A comparison of the performance of two different markets may illustrate these concepts better. Exhibit 2 shows the history and baseline net operating income forecast (indexed to 2005) for a representative property in the Hartford multifamily market. Note how Hartford’s net operating income performance is quite stable over time, and shows a gradual upward trend. As a result, this market can be modeled fairly well, providing a relatively high degree of confidence regarding the future path of net operating income. As a result, Exhibit 3 shows relatively narrow forecast cones for the market. Contrast this performance with San Jose’s which is featured in the following exhibits 4 and 5. San Jose’s multifamily market shown some rather dramatic swings in performance over the past two decades, reflecting the boom-and-bust nature of the its dominant industry, the technology sector. During the dot.com boom of the late 1990s and early 2000s, surging employment levels increased occupancy and rents to unprecedented levels. As the economy cooled during the early 2000s recession, occupancies fell just as new supply was added to the market, depressing net operating income. A similar, but much more muted cycle was observed over the past few years. As a result, our confidence in the future path of San Jose’s net operating income is significantly less than it is in Hartford; as a result, San Jose’s confidence bands, or cones are much wider.

So, how may we use this information to develop a measure of future riskiness of lending by market? One way would be to use the confidence bands to calculate the probability that net operating income falls below its level at origination over a specified loan term. For any given forecast year there is a distribution that allow us to calculate the probability that net operating income falls below a specified level. We may calculate the average of these probabilities over time to come up with a single probability measure of the “riskiness” of lending in a particular multifamily market. Exhibits 3 and 5 graphically illustrate this calculation for Hartford and San Jose. To calculate this “riskiness” measure, we are effectively calculating the cone area that lies below dashed line – which represents the NOI level at origination. Note that the size of this area, and riskiness of the market, is dependent on both the expected trajectory of market NOI as well as cone width. Therefore it explicitly takes into account future baseline scenario of market performance, as well as its uncertainty. As the exhibits show, we would assign a higher probability of an NOI shortfall to the San Jose market, and therefore consider it to be more “risky”.

We may then rank markets according to the probability that NOI falls below its level at origination. While this analysis does not provide an indicator of the actual default probability of a loan, since it does not explicitly take into account loan terms such as debt service coverage and LTV, it does provide an indicator of “market” risk that may be associated with overall default levels.

So, how may we use this information to develop a measure of future riskiness of lending by market? One way would be to use the confidence bands to calculate the probability that net operating income falls below its level at origination over a specified loan term. For any given forecast year there is a distribution that allow us to calculate the probability that net operating income falls below a specified level. We may calculate the average of these probabilities over time to come up with a single probability measure of the “riskiness” of lending in a particular multifamily market. Exhibits 3 and 5 graphically illustrate this calculation for Hartford and San Jose. To calculate this “riskiness” measure, we are effectively calculating the cone area that lies below dashed line – which represents the NOI level at origination. Note that the size of this area, and riskiness of the market, is dependent on both the expected trajectory of market NOI as well as cone width. Therefore it explicitly takes into account future baseline scenario of market performance, as well as its uncertainty. As the exhibits show, we would assign a higher probability of an NOI shortfall to the San Jose market, and therefore consider it to be more “risky”.

We may then rank markets according to the probability that NOI falls below its level at origination. While this analysis does not provide an indicator of the actual default probability of a loan, since it does not explicitly take into account loan terms such as debt service coverage and LTV, it does provide an indicator of “market” risk that may be associated with overall default levels.

____________________________

^{1}See Torto Wheaton Research, “Real Estate Risk: A Forward Looking Approach”, May 2001, at www.cbre-ea.com.