The FHA released its FY 2012 HECM Actuarial Study and MMI Fund Report on November 16. The report reflects many changes to their economic and modeling assumptions, resulting in significantly higher projected losses for the HECM program. This is a dramatic turnaround from just a year ago. We criticized last year’s FHA actuarial study for being “too rosy.” This year, we might echo William Blake in saying “Rose thou art sick.” Last year, our criticism centered principally on their assumptions on prepayment rates, home prices, and Tax and Insurance (T&I) defaults. At the same time, we saw reason for optimism because of the many changes (lower PLFs, higher MIP, HECM Saver, etc.) that the FHA has adopted. As a result, we predicted that the overall HECM picture, particularly for the Mutual Mortgage Insurance Fund (MMI), would gradually improve each year as old loans paid off and were replaced by new HECMs with lower Loan-to-Value (LTV) ratios, higher MIPs, and less underlying home price erosion.
The new assumptions in FHA’s FY 2012 model include slower prepayments, smaller home price increases, and higher default frequencies and severities, all of which result in a significantly higher estimate of expected losses. The Actuarial Study predicts that the FHA will lose $7.6 billion, a figure that represents the present value of the net inflows and outflows from currently outstanding HECMs in the MMI. HECMs originated prior to FY 2009 are insured by the General Insurance Fund (GI) and therefore are not included in this analysis. After giving $4.8 billion credit to previously built-up capital and income (mostly from initial and ongoing MIP), the MMI fund has an economic value of negative $2.8 billion. As a result, says the FHA, they will change the program again, including a third round of PLF reductions. As much as we criticized their assumptions last year, we have to ask: have they gone too far?
To understand the answer to that question, we should first understand the dimensions of the MMI fund. About 300,000 HECM loans remain of the approximately 320,000 originated since the beginning of FY 2009. These are the HECM loans that comprise the reverse mortgage portion of the MMI fund. The FY 2009 vintage is the largest of the four vintages in the MMI and is also worst performing: home prices have declined about 11% since October 2008, when that Fiscal Year began. Moreover, the FY 2009 HECMs were originated under a higher Principal Limit Factor (PLF) table, so these loans have significantly higher LTVs than later vintages. FHA receives only 0.50% annual mortgage insurance premium (MIP) for these loans. Over 100,000 FY 2009 loans remain outstanding. The FY 2010 vintage has significantly lower PLFs, but also has the lower 0.50% MIP. This vintage has suffered only a 5% average home price decline since October 30, 2009. The last two vintages, FYs 2011 and 2012, have lower PLFs, a much higher annual MIP (1.25%), and have suffered comparatively little, if any, home price deterioration.
From this and other FHA reports we estimate that these 300,000 loans have approximately $48.5 billion in outstanding principal balance and $77 billion in Maximum Claim Amount. That implies that FHA will collect about $375 million in MIP next year from the loans and less with each subsequent year as loans pay off. We estimate the present value of the ongoing MIP for these loans at about $1.6 billion. Squaring this with the Actuarial Report’s estimate of the net present value of the MMI Fund HECM cash flow at -$7.6 billion implies that the Report estimates the present value of future losses at over $9 billion. This is a staggering number, representing nearly 20% of the current loan balance. The FHA expresses this number in its financial statements as the Liabilities for Loan Loss Guarantees (LLG), defined as “the net present value of anticipated cash outflows and cash inflows.” Our opinion: this estimate seems way too high. Even in a market with flat home prices (no increase or decrease for the life of the loan portfolio), we would not expect a newer vintage of HECMs to cause FHA to suffer losses exceeding the present value of 12% of the current balance. A flat market is hardly the base case, which usually allows for home price appreciation that should lower expected losses.
The actuarial report is not the final word; the FY 2012 Financial Statement Audit shows a $5.5 Billion MMI LLG, which probably reflects the favorable home price increases of the third quarter of 2012 and other adjustments. Still, this implies an average 15% loss level, which we think overstates the risk of the HECMs in the MMI fund.
The cash flow model needed to analyze the MMI Fund is necessarily complex. It must have good loan level data input, sound statistical methods, and reflect a comprehensive knowledge of reverse mortgage cash flow with all its idiosyncrasies. It is impossible to evaluate any model conclusively without examining it first hand, but the actuarial report provides a lengthy description of its methodology. Although this description is long on statistical methodology and short on cash flow methodology, it does provide some insight into how it reached its conclusions.
The really big differences came from changes in the model’s methodology for measuring loss from defaults. These include tax and insurance defaults and also the losses resulting from loans that end up in foreclosure. These loans have very high loss severities because of the legal, maintenance, and disposition costs that arise during the lengthy process of foreclosure. According to the report, changes from the updated valuation model account for $2.4 billion in additional losses, nearly equal to the overall negative value of the fund. This includes slower payoff assumptions.
We warned last year that the Actuarial Analysis loan payoff speed estimates were too high. The 2011 Analysis suggested an annual prepayment rate of 6%, even for new vintages. The new report shows 4-5% prepayment rates in the near term (page A-12), still a bit higher than what we are observing, but more realistic than last year’s forecast. We are told 54,591 HECMs were originated in FY 2012. In its latest statistical release, HUD states that 595,342 HECM loans remained outstanding at the end of the year; the corresponding number in last year’s report was 560,843. That implies that about 20,000 loans paid off, a prepayment rate of approximately 3.5%. One can easily do this math going back through previous years’ FHA Annual Management Reports to find that HECM prepayment rates have been falling steadily for almost seven years. Still, the report’s payoff speed adjustment is a step in the right direction.
The report gives a detailed description of the regression model used to estimate T&I loss frequency, but not so much detail on its impact on loss severity. The report does not break out how much of the $2.4 billion loss is due to T&I defaults vs. slower payoffs or any other factor. The report describes its HECM Cash Flow Analysis methodology in Appendix C, where it states (on page C-3) that “the HECM forecasting model assumes that the assignment occurs when the projected UPB reaches 98 percent of the MCA threshold.” But this is not necessarily correct: assignment cannot take place if the loan is in default. In that case, the investor (which could be an HMBS issuer) must hold on to the loan. This is an important feature that can have a major impact on loss severity. If the cash flow model assumes automatic assignment regardless of default status it is overestimating the T&I advances and losses suffered by FHA.
In any case, the lifetime T&I default rates predicted by the model, as shown on page D-8, would give us some comfort if they made any sense, but they don’t. Page D-7 describes Exhibit D-6 (which is on page D-8, confused already?) as “… T&I default probabilities were forecasted for all active loans at the end of June 30, 2012. The resultant cumulative lifetime T&I default rates by historical fiscal years of endorsement for the active loans appear in the Exhibit D-6 below.” (By “Active” loans it appears that they mean any loan still outstanding, not to be confused with the Loan Status Category “Active” used by servicers to describe a loan that is still outstanding but not in default.) But the T&I default numbers look lower than the T&I default rates already achieved. Is this an estimate of additional defaults only? If so, what is the impact of loans already in T&I default? Most importantly, if 3 of the 4 vintages of the MMI will not exceed 1% in Lifetime T&I default rates, then how can this add significantly to the loss estimate?
The second largest factor in the model was the Updated Loan Conveyance Projection. The report points out an alarming swing in conveyance rates at termination, that is, the rate at which properties are conveyed to FHA upon loan termination, as opposed to owners or estates engaging in direct sales. Whereas last year 30% of all terminations resulted in property conveyance (e.g. foreclosure, deed-in-lieu) and 70% were paid off by the estate, in FY ’12 the proportions switched to 70%/30% conveyance and normal payoff, respectively. This adds nearly $2 billion to the loss estimate. This is a huge change; the report cites declining home prices, which are clearly the main driver, but such a large and abrupt change might have other immediate causes, such as a change in tactics by a large HECM investor.
In FHA’s Annual Report to Congress for the FY 2012 Financial Status of the MMI Fund (p 29), FHA notes that:
“ … Research indicates that this was directly tied to falling home prices. Owners and estate executors faced with mortgage balances greater than property value at the time of borrower exit from the home are less willing to engage in marketing and sale of the property than are those with positive equity in the home. In such cases, there is no financial benefit from managing the property sale and so those responsible for the home are more likely to convey the property to HUD for sale. … Property management and marketing costs associated with the disposition of homes conveyed to HUD typically cost approximately 12 percent of property value and thus increase the severity of loss for FHA.”
So far so good; this is an accurate description of this risk factor and consistent with our own research. When a loan passes the crossover point, at which time the loan balance exceeds the net property value, default frequency and severity rise sharply. The FHA continues:
“ .. The actuarial projections now include consideration of how conveyance rates vary with changes in house prices when estimating the economic value of the HECM portfolio. The actuarial projections now include consideration of
how conveyance rates vary with changes in house prices when estimating the economic value of the HECM portfolio.”
This is conceptually sound, but the key question is: how do you predict loss frequency and severity accurately? Most of these conveyances are from loans in the GI Fund that were originated in 2005 to 2008. Many of these loans are deeply underwater, which is why they have such sharply high frequency and severity of loss. They cannot be relied upon as a historical benchmark for the MMI loans; these distressed GI loans will have much higher frequency and severity of losses compared to the MMI loans. Conveyance rates vary indirectly from changes in home prices; what really matters is do these changes in home prices result in crossover and negative equity, and if so, by how much?
The experience of the GI Fund loans is applicable to only the most severe scenarios, which should be considered given recent history, but should by no means serve as average benchmarks for frequency and severity of loss. The report admits that “The lack of long-run performance data potentially limits the robustness of the models’ predictive capacity for later policy years.” That is precisely why the numbers should be used carefully; we have very little data on foreclosure frequency and severity for the MMI loans, as comparatively few have paid off and fewer still of these payoffs were underwater. The best option is to provide alternatives to conveyance: we will discuss this topic further in our next blog.
Also, the question we posed above for T&I default is relevant here too: does the model properly measure the cases in which FHA does NOT bear the brunt of the loss due to conveyance? For payoffs that take place before FHA assignment, the investor can bear the brunt of the loss, especially if it results in an Appraisal-Based Claim, where the investor ends up with an REO property for more than six months. Referring again to page C-3 of the report, the “historical severity rate” is used to calculate Type I (pre-assignment) claims and seems to imply that all loans that hit 98% MCA are put to FHA. As we explained above, historical severity is probably too high and defaulted loans cannot be assigned to FHA.
The third largest factor was the Updated Economic Forecast for home prices. This portion of the model is also appropriately more conservative; this resulted in a downward adjustment to the MMI fund value of $462 million.
The fourth largest factor was the introduction of Monte Carlo Stochastic Home Price Appreciation (HPA) and interest rate simulation. This is a very technical topic, but conceptually it is a valid approach and one intended to avoid an overreliance on a base case. Instead, it calculates several future scenarios, based on assumed probabilities and volatilities of underlying variables, and takes the weighted average of the results. This can result in a more robust model, but bear in mind that it discards one set of base assumptions in favor of another, namely the assumed volatilities and probabilities. The model’s accuracy is therefore limited by the availability of robust historical data. For example, according to the report, home price data before 1980 is not considered reliable (page F-4). Is the volatile boom and bust experience since 1980 typical?
In any case, the Monte Carlo methodology raises expected losses by $412 million. This is not surprising, due to the phenomena known as Jensen’s inequality, explained by the report on page F-1:
“ … when HPA goes negative, default losses increase at an increasing rate as HPA falls, but when HPA goes positive and keeps increasing, default losses can only go as low as zero and premium income does not increase. This is what we observed for HECMs.”
In fact, it is also true for proprietary reverse mortgages and forward mortgages. Our objection is more philosophical than quantitative. Among other advantages, the Monte Carlo method is intended to capture the “Black Swan” risk, the doomsday scenarios that could cause losses as high as $28.3 billion (page iv). According to the report, this worst-case scenario is one of 100 equally likely outcomes. This implies that this one doomsday scenario accounts for $283 million of expected loss. A repeat of the 2006-2009 crisis (or worse) would surely cause very high losses of the magnitude that doomed the private mortgage insurers. But if FHA is pricing its risk like a private insurer does (or should), it risks straying from its mission. This is especially true if the coming product changes are too severe and price out some of the senior homeowners who need help. Something to think about: what is government mortgage insurance for?
A few other conclusions that appeared in the report seem puzzling:
The Principal Limit discussion (page 2) says the expected mortgage interest rate floor is 3.0%. In fact, the PLF tables might show 3%, but the effective floor is 5%. The PLFs do not increase for expected rates lower than 5%. We will explain the significance of this in our next installment.
The report seems to say that the value of the FY 2012 book of business is negative $385 million (page 14). How can that be when the nearly identical FY 2013 vintage is positive?
In the actuarial model, all loans are assumed to pay off after 35 years (page 42). For a pool of nearly 300,000 loans, FHA will certainly end up with several 97 year-old borrowers who took out a HECM at age 62. Our decades-long observations from the proprietary reverse mortgage market confirm this. The loss severities for these loans can be quite high, but the present value impact today is much lower.
The model appears to project annual, not monthly, cash flows. If this is the case, it may overstate losses as the crossover amounts are determined in arrears (at period end). All else equal, this will increase estimated crossover amounts and therefore lower economic value overall.
Implications of the Model Results
FHA’s own numbers show a steady improvement over the next several fiscal years. As we explained last year, with each passing year more of the bad old loans pay off, and the remaining portfolio contains proportionately more of the profitable (to FHA) loans.
The FHA concludes that the MMI deficit means they must further change the product. The report does state fairly clearly that:
1. The fixed rate HECM Standard will be “merged” into the fixed rate HECM Saver;
2. FHA intends to reduce the amount of the initial draw; it’s not clear if this language refers to capping fixed rate proceeds which the report stated is a “longer term” objective, or is another change not well articulated in this report;
3. PLFs will be lowered “across the board” for all products;
4. Some form of economic incentive will be implemented to motivate estates to dispose of properties on their own. No hint was provided as to what or how that might work, but it is intended to provide some flexibility that will reduce loss severity;
5. Proceeds on fixed rate loans will be limited to the amount necessary to pay off “mandatory obligations,” i.e. mortgage liens, closing costs, delinquent tax, and insurance premiums. The report references immediately limiting the amount borrowers will be allowed to draw at origination but gives no specific program detail; and
6. Financial assessment and some form of T&I set-aside continue to be on the docket for introduction in the future.
The FHA must abide by the rules set by Congress, so they must bring the MMI back into balance. But remember, the MMI consists of 4 vintages of HECM that are really 3 different standard products: high PLF/Low MIP (FY ’09), Medium PLF/Low MIP (FY ’10), and Lower PLF/High MIP (FY ’11 and ’12). The last two vintages also include the HECM Saver. If the MMI consisted of only the last two vintages, we would not be having this discussion. Instead, the FHA is forced into a situation where the product must become even more conservative to subsidize the other two vintages (especially ’09).
This leads to another problem: further PLF cuts may be counterproductive as they will drive down origination volumes and actually hinder the MMI’s recovery. The report acknowledges this with respect to the forward program, for example, higher MIPs charged for longer periods will result in faster prepayments by credit worthy borrowers, leaving a remaining pool of lower quality.
The report seems to agree with us that the fund will improve over time. At the same time, it predicts a level of loss that seems too severe. FHA must operate under the budget rules that force it to act now and close the gap sooner, which paradoxically means changing a product that, in its current form, is probably not contributing to the problem.
We do not suggest that a perfectly modeled MMI fund would give us cause for complacency: the program should be reformed further. For example, the T&I Chickens have come home to roost; we advocated a T&I set aside or escrow over 3 years ago. The inaction on this issue has been costly.
Our next blog will discuss what FHA should do next, with an emphasis on using tools in the current program’s features and less on the root-canal of across-the-board PLF cuts.
One last word: far from being the disproportionate cause of FHA’s problems, it is the forward mortgage program that has disproportionately higher delinquencies, defaults, realized losses, and leverage. HECM has, if any, borne a disproportionate share of the reforms. We searched the FHA reports in vain for any discussion of lowering forward mortgage LTVs. No amount of adjustment to MIP and FICO requirements will fix the problems caused by overleverage. Unlike the current forward mortgage FHA program, the existing HECM loans being originated today are not making the situation worse.