In 2012, economists from the FHFA published a research paper describing a countercyclical approach for estimating the capital level required for mortgage portfolios to withstand future shocks to the housing sector. This approach uses state-level countercyclical stressed housing price paths (CSPs) based on where a state’s housing price levels are relative to its long-run trends and on the historical downside volatility of the state’s housing prices. FHFA has published a set of 51 of these 30-year CSPs (one for each state plus DC) starting from 13 different launch dates (each of the past 7 quarters—Q4 2013 through Q2 2015—as well as 6 quarters from 2003 to 2010). While the approach is not new, we believe it provides an interesting alternative to the Federal Reserve’s annual Dodd Frank Act Stress Test (DFAST) stressed housing price scenarios because it is more transparent and more granular. Thpaper compares FHFA’s CSPs to the DFAST stressed HPI scenarios and outlines an approach for applying the state-level granularity of the CSPs to national-level DFAST Severely Adverse scenario.
This paper is divided into four parts. The first two parts review the Fed’s and FHFA’s approach for stress scenarios from both a practical and theoretical perspective. The third part compares the Fed’s DFAST scenarios to FHFA’s state-level CSPs. And finally, the fourth part outlines our approach for using FHFA’s state-level CSPs to create state-level scenarios that are consistent with the Fed’s national DFAST “Severely Adverse” scenario.
Review of Fed’s Methodology for Creating Stress Scenarios
The idea of a countercyclical capital cushion predates the financial crisis but has gained traction with regulators and central bankers in recent frameworks and guidance. The objectives of the Basel III proposal on countercyclical capital include “conserv[ing] capital to build buffers at individual banks and the banking sector that can be used in stress” and “achiev[ing] the broader macroprudential goal of protecting the banking sector from periods of excess credit growth.”1 In practice, this means setting up rules that force banks to hold a capital “buffer [that] would grow during economic expansions and decrease during contractions.”2
This countercyclical approach contrasts with the procyclicality of the previous capital regime. Under Basel II, a bank with an internal capital model based on historical credit loss experience would have a higher capital requirement during downturns, when credit losses are higher, thus discouraging lending. Conversely, in expansionary times, when credit losses are low, banks would have lower capital requirements. Similarly, a value‐at‐risk (VAR) approach for estimating fluctuations in a portfolio’s market value would suggest a lower need for capital during period of low market volatility.
Based on our reading of the relevant sections of the Dodd Frank Act, the Fed should be using a countercyclical approach to develop its DFAST stressed scenarios. The Dodd Frank Act amended the Bank Holding Company Act (passed in 1956) to clarify some vague guidance on the Federal Reserve Board’s authority to regulate banks. Here is the original wording from the 1956 Act: “The Board is authorized to issue such regulations and orders including regulations and orders relating to the capital requirements for bank lending companies, as may be necessary to enable it to administer and carry out the purposes of this Act and prevent evasions thereof.”3
The Dodd Frank Act added the following lines, which explicitly require that the FRB take a countercyclical approach to setting capital requirements: “In establishing capital regulations pursuant to this subsection, the appropriate Federal banking agency shall seek to make such requirements countercyclical so that the amount of capital required to be maintained by a company increases in times of economic expansion and decreases in times of economic contraction, consistent with the safety and soundness of the company.”
Exhibit 1. DFAST Severely Adverse HPI Scenarios: 2013, 2014, and 2015
But while this guidance to use a countercyclical approach is explicit, the Fed’s method for incorporating countercyclicality into the housing price scenarios is not transparent. For instance, the 2014 and 2015 Severely Adverse Scenarios are almost identical: in both scenarios, housing prices decline by 25.7% over 10 quarters (from launch points of Q3 2013 and Q3 2014, respectively), even though housing prices increased by 5.8% between the two launch points (4.7% on a real basis when adjusting for 1% inflation over the Q3 2014‐Q3 2014 time period). If the Fed views that 5‐6% housing price growth rate as normal, it would imply that the housing market was at the same place in the cycle in Q3 2014 as in Q3 2013 and thus there would be no need to adjust the severely adverse scenario from 2014 to 2015. This is certainly a plausible explanation for the unchanged scenario. And it would be consistent with the increasing severity of the Severely Adverse scenario in 2014 relative to 2013. The 2013 Severely Adverse scenario sent housing prices down by 21.1% from the Q3 2012 launching point. But housing prices rose by 10.4% from Q3 2012 to Q3 2013 (9.6% on a real basis when adjusting for 0.7% inflation from Q3 2012 to Q3 2013), so it could be argued that in Q3 2013, the launch point for the 2014 Scenarios, the housing market was at a higher point in the macroeconomic cycle and thus warranted a “Severely Adverse” scenario with a bigger housing price decline than in the prior year’s scenario. But the Fed does not disclose its methods for formulating the scenarios, so this is just speculation.
Review of FHFA’s Methodology for Creating Countercyclical Stress Paths
By contrast, FHFA is relatively transparent in describing its methodology for generating its state‐ level countercyclical stressed housing price paths (CSPs). Each state’s CSP is derived from three components:
Exhibit 2. FHFA Countercyclical Stress Path Throughs by Launch Date
- The state’s long‐term trend line is calculated using the state’s real (inflation‐adjusted) HPI growth rate between 1975 and 2001, which FHFA deemed a reasonable pre‐bubble end‐ point.
- The state’s trough of the “worst HPI downturn” is measured as the lowest historical housing price as a percentage of the long‐term trend line during the same 1975‐ 2001 time span.
- A general 10‐year “HPI Shock Time Path” is used for all states. Over the first 3 years, this path takes HPI from the state’s current level to the state’s trough, where it remains for 4 years. Then it returns to the state’s trend line over the final 3 years.
Moreover, FHFA has retroactively published CSPs for 13 different launch dates between 2003 and 2015, so we can look at the severity of projected house price declines in different stages of the cycle. As shown in Figure 2, FHFA’s approach for developing stress paths projects bigger declines when launching from house price levels in the bubble years prior to 2008 than in subsequent years after the onset of the Great Recession.
Exhibit 3. Actual HPI vs. Countercyclical Stress Paths (California) (Source: FHFA)
Figure 3 is an extract from FHFA’s presentation comparing two CSPs applied to California housing prices at two different stages of the macroeconomic cycle. Applying the CSP In 2003, in the early‐to‐mid stage of the bubble, would have sent housing prices down by 37.2% over 3 years. But applying the CSP in 2005, at close to the peak of the bubble, would have sent housing prices down by 53.5%. Note that both paths revert to the same long‐term trend. Figures 4‐7 compare the 2003 and 2005 CSPs for 3 states and the US to the actual experienced housing prices. In general, the actual peak‐to‐trough house price decline was less severe than the troughs produced in either of the two stressed scenarios, but there are several exceptions, including Nevada (see figure 6).
Exhibits 4 and 5. Actual HPI vs. Countercyclical Stress Paths (California and Florida)
The case of Nevada illustrates one of the drawbacks we see in this backward‐looking approach to determining a state’s long‐term growth trend and volatility. In Figure 8, we extend Nevada’s historical HPI trend back to 1975 and use real instead of nominal HPI levels to show that Nevada’s housing prices were relatively stable through 2001. But Nevada’s low historical volatility over this short 27 year time period limited the extent to which housing prices could overshoot on the way down in the post‐ bubble correction and thus the CSPs under‐predicted the severity of the plunge. The SEC‐mandated disclaimer in investment literature that “past performance does not necessarily predict future results” seems appropriate in this context as well.
Exhibits 6 and 7. Actual HPI vs. Countercyclical Stress Paths (Nevada and U.S.)
The other drawback to this backward‐looking approach is that it implicitly assumes that the demographic, economic, regulatory, and geographic drivers of a state’s historical housing price trends remain in place today and that any deviations from that historical trend reflect cyclical changes in the housing market. But it is also possible that recent deviations from the 1975‐2001 trend for some states may reflect structural shifts in the housing market. For instance, Illinois grew at a healthy pace from 1975 to 2001, which generates a long‐term average annual nominal growth rate of about 5% (1% real growth after adjusting for inflation). But because Illinois home prices remain depressed after the financial crisis, as of the October 2014 launch date, they did not have far to fall to hit the historical trend and volatility‐based trough.
In fact, Illinois is one of three states (the other two are Massachusetts and North Carolina), where FHFA applies a minimum 5% real decline (after controlling for mild appreciation over the 3‐year glide path to the trough, this translates to a 2.6% nominal decline) because the current housing price levels are at a low point in the cycle (i.e., well below the historical trend line). But the current house price levels may seem reasonable to people familiar with the Illinois’ poor fiscal health,5 and it might not be difficult to imagine a more severe drop in housing prices than the projected 2.6% decline.
Exhibit 8. Real HPI vs. Countercyclical Stress Paths (Nevada: 1975‐2015)
Exhibit 9. Nominal Actual HPI & Oct. 2014 Countercyclical Stress Path vs. Historical Trend Line: Illinois
Exhibit 10. Nominal Actual HPI & Oct. 2014 Countercyclical Stress Path vs. Historical Trend Line: Washington, DC
Conversely, house prices in Washington, DC rose significantly during the bubble years but did not see much of a decline in the aftermath before resuming their ascent. Thus DC prices remain well above the historical trend line derived from the 1975‐2001 experience. As a result, DC’s October 2014 CSP sends housing prices down by 59.2% in real terms (58.1% in nominal terms). It is certainly plausible to argue that DC is more vulnerable to a big decline in housing prices than other regions that have already experienced big post‐bubble corrections and that DC remains in bubble territory due to a temporary post‐ crisis increase in demand for real estate close to the federal government. But it is equally plausible to argue that the increased demand is the result of secular forcesthat will continue to attract high paying jobs to DC and boost housing prices. For instance, one could argue that the post‐crisis landscape has ensured a regulatory environment that will keep Washington residential real estate in high demand by variouslobbyists and consultants. (Full disclosure: the author is a DC homeowner who would be underwater if the stress path for DC were realized.)
Comparison of FHFA’s CSPs to the Fed’s DFAST Scenarios
The most obvious difference between the FHFA’s CSPs and the Fed’s DFAST scenarios is that the Fed provides only a national level forecast, whereas FHFA provides state‐level stressed paths. We will address the different level of granularity provided by the two sources in the next section. This section compares the forecasts on two other dimensions: stress severity and timing. Figure 11 compares the two adverse HPI scenarios (the “Adverse” and the “Severely Adverse”) from the 2015 DFAST to two FHFA CSPs for the U.S. from different launch dates. The FHFA’s October 2014 CSP shares a launch date with the Fed’s two DFAST scenarios, so these three forecasts are all based on actual HPI trends through Q3 2014 and the first quarter forecast corresponds to December 2014. The FHFA CSP produces a national decline of 16.1% over 12 quarters, which is slightly more severe than the DFAST’s “Adverse” scenario, which produces a national decline of only 13.7%. But it is significantly less severe than the 25.7% decline in the DFAST “Severely Adverse” scenario. The decline is also more rapid in the DFAST scenarios, with house prices hitting a trough in only 10 quarters compared to the 12 quarter path to trough in the CSP scenario.
Figure 11 also includes the FHFA’s CSP from October 2005, which shows a 32.8% decline in housing prices over 12 quarters. This is one of the CSPs shown in Figure 7, but it is useful to see it again because it demonstrates the sensitivity of FHFA’s CSPs to the stage in the housing cycle: housing prices in October 2005 had further to fall because they were in bubble territory whereas house price levels in October 2014 were more in line with historical trends. The October 2005 CSP sends housing prices down by more than twice as much as the October 2014 CSP and tracks very closely with the DFAST “Severely Adverse” scenario, though as noted above, housing prices start to recover afterthe 10th quarterin the DFAST scenario but do not bottom out in the CSP scenario until the 12th quarter. More generally, FHFA’s CSPs for launch dates during the bubble years (from 2003 to 2007) all project more severe housing price declines than the 2015 DFAST “Severely Adverse” scenario, and all CSPs for launch dates after the onset of the Great Recession project less severe housing price declines. This is shown in Figure 12, which superimposes the house price decline projected by the 2015 DFAST “Severely Adverse” scenario onto the declines projected by the FHFA’s 13 different national‐level CSPs (as originally shown in Figure 2).
Exhibit 11. 2014 Dodd-Frank Stress Scenarios vs. US Countercyclical FHFA Stress Scenarios by Quarter
Exhibit 12. Severity of FHFA's Countercyclical Stress Paths by Launch Date
The severity of the stress paths is even more variable across states than over time. As we saw in Figures 9 and 10, FHFA’s CSPs from October 2014 project declines ranging from a negligible 2.6% for Illinois to 58.1% for Washington, DC. Figure 13 shows the full range of these house price declines across all 50 states as well as the difference between the national average decline and the DFAST “Severely Adverse” decline (this difference is captured as the DFAST adjustment). The CSPsfor twelve of the states project declines greater than the 25.7% decline projected in the DFAST “Severely Adverse” scenario.
Outline of Methodology to Develop State-Level DFAST HPI Scenarios
The final section of this paper outlines an approach for using FHFA’s state‐level CSPs to develop state‐ level scenarios that are consistent with the Fed’s national DFAST “Severely Adverse” HPI scenario. The goal of this approach isto establish a set ofstate‐level HPI projections that: 1) retain the same level of differentiation in house price declines acrossstates as found in the FHFA CSPs; 2) retain the basic shape and timing of the DFAST “Severely Adverse” HPI path; and 3) roll up at the portfolio level to the DFAST’s national “Severely Adverse” HPIscenario. This approach might be useful for a bank that is required to apply the DFAST scenarios to their portfolio but wants to (or is encouraged by regulators to) use more geographic granularity than is provided by the DFAST national HPI forecast. While we illustrate this approach using the state‐level HPI declines projected in the FHFA CSPs, the approach would work equally well with granular HPI projections from other economic forecasters either at the state level or more granular unit of analysis.4
Exhibit 13. Severity of FHFA's Countercyclical Stress Paths by State (Launch Date: Oct. 2014)
The first step is to look at each the housing price forecasts for each state and calculate the troughs. Figure 13 shows these troughs for the FHFA CSPs.
The second step is to use the portfolio’s footprint to calculate the portfolio‐level average decline. Each state‐level projected decline is weighted by the state’s share of the portfolio. In this example we assume the portfolio has the same geographic footprint as the universe of agency loans (using June 2015 pool‐level data pulled from our RS Edge Platform) and calculate a weighted average decline of 16.0%, which is not materially different than the 16.1% decline projected by FHFA’s October 2014 CSP for the U.S.
The third step is to compare this weighted average portfolio decline from the FHFA CSPs (16.0%) to the target portfolio decline from the DFAST “Severely Adverse Scenario” (25.7%) to determine an adjustment to apply to the FHFA state‐level declines. We considered two different methods for applying the adjustment – multiplicative and additive. Using the ratio of the DFAST target to FHFA house price decline would yield a multiplier of 1.61 (25.7% / 16.0% = 1.61) that could be applied to FHFA’s projected decline for each state to achieve the portfolio level house price decline of 25.7%. The problem with this “multiplier” approach is that it exacerbates the already wide variance across states and makes the scenarios for the states with big projected declines even more implausible. For instance, the projected decline in housing prices for DC would increase from 58% to 93%. As discussed before, it is difficult to imagine DC housing prices declining to 42% of their current levels. A decline to 7% of current levels is not plausible. Our preferred approach is simply to calculate the difference between the DFAST target and FHFA house price decline to arrive at a ‐9.7% additive adjustment.
The fourth step is to apply this additive adjustment to each state‐level projected decline to arrive at a DFAST‐adjusted projected decline or trough for each state. So for instance, FHFA’s CSP for Florida projects a 23.4% decline. Adding the ‐9.7% DFAST adjustment produces a DFAST‐adjusted trough of ‐33.1%.
The fifth and final step is to shift the DFAST “Severely Adverse” Scenario so that the trough in the 10th quarter is equal to the state’s DFAST‐adjusted trough. To make this modification, we first compare the state’s DFAST‐adjusted trough to the DFAST national trough to determine the additive adjustment to apply to the DFAST national path. In the example of Florida, where the DFAST‐adjusted trough is‐33.1%, the additive adjustment would be ‐7.5% (‐33.1% ‐ ‐25.7% = ‐7.5%). We then phase this adjustment in over the 10 quarters, so that in the case of Florida, the first quarter projected HPI level would be 0.75% below the national “Severely Adverse” HPI projection (1/10 * ‐ 7.5% = ‐0.75%) and the 9th quarter would be 6.75% below the national projection (9/10 * ‐7.5% = ‐6.75%). This results in a set of 51 state‐level DFAST‐adjusted HPI paths, each with a similar shape to the original specified national HPI path. Figure 14 shows steps 4 and 5 of this process for Florida. Figure 15 shows the weighted average HPI path for the portfolio (which, mathematically, is exactly equal to the DFAST “Severely Adverse” scenario path) as well as paths for selected states, including two with less severe DFAST‐ adjusted paths than the portfolio average (Illinois and New York) and three with more severe paths (Florida, California, and DC).
Exhibit 14. Construction of Florida's DFAST-Adjusted HPI Path
Exhibit 15. DFAST‐Adjusted HPI Paths for Selected States
In this paper, we compared the FHFA’s state‐level countercyclical stressed housing price paths (CSPs) to the Federal Reserve’s national HPI scenarios. While the Fed’s scenarios are more widely known and used because regulators require them for CCAR and DFAST scenario runs, we believe the FHFA CSPs can be referenced as a complement to the Fed’s scenarios to build more granular HPI paths. We find that the FHFA’s formulaic approach for constructing its CSPs, which is based on state‐level historical trends, can create nonsensical stress paths and we highlight two of these in the paper: the October 2014 CSPs consider housing price declines of 57.8% in Washington DC and only 3.5% in Illinois. But we recognize thatsuch a formulaic approach does not capture emerging risks and valuation dynamics in a housing market. Overall we believe the approach provides a reasonable starting point for granular countercyclical HPI forecasts, and we credit FHFA for its transparent approach. We recommend that portfolio or risk managers considering using these CSPs review the state‐level projected declines to make sure they are reasonable, especially for regions where their portfolio is concentrated, and substitute their own views when appropriate. This paper outlines a flexible approach that applies state‐level stress declines—either those projected by FHFA’s CSPs or those provided by a risk manager—to the national‐level DFAST “Severely Adverse” HPI path to create a set of state‐level HPI projections that roll up at the portfolio level to the DFAST’s national HPI scenario.
CLICK HERE to download a PDF version.
 https://www.fas.org/sgp/crs/misc/R42744.pdf : “Congressional Research Service U.S. Implementation Of The Basil Capital Regulatory Framework.” April 9, 2014.
 Two think tanks, Mercatus and the Pew Trusts, ranked Illinois 50th on fiscal health as of July 2015 and 47th on its level of reserves as of September 2015. http://mercatus.org/statefiscalrankings
 We have not discussed the effect of the level of granularity on performance projections. This is largely beyond the scope of this paper but deserves brief mention. Typically, housing prices affect mortgage performance either directly through a current LTV calculation that estimates a borrower’s level of equity or indirectly through a variable that captures general macro‐economic health. If the relationship between these variables and default rates is linear, adding granularity to a forecast should not change overall default rates as long as the granular forecasts produce average declines equal to the higher level forecast. But if the relationship is convex (that is, if default rates are increasing at an increasing rate as LTV ratios increase), adding granularity to a forecast should increase the overall default rate.