ARBOR REALTY TRUST INC (ABR) stock intrinsic value fair value estimate & owner earnings analysis

Arbor Realty Trust is really two businesses inside one company, and they need to be judged separately. The first is the agency business. Arbor makes apartment loans under its Fannie Mae and Freddie Mac licenses, sells them to the agencies, and keeps the right to service them for a fee. The servicing book was $36.7 billion at mid-2026, with an average remaining life of about six years. Arbor has been a top-10 Fannie Mae lender for 19 straight years. This business earned $125.7 million in 2023, $97.3 million in 2024 and $77.1 million in 2025 under standard accounting. Adding back the amortization of servicing rights and removing the non-cash gains booked when new servicing is created, it produced about $94 million of cash in 2025. Interest earned on borrower escrow balances is shrinking (about $66 million a year in the spring of 2026, about $55 million by mid-year), and reserves for Arbor's share of losses on agency loans are rising. So we use about $85 million a year (about $0.46 per share) as its earning power. This is the durable part of the company. The second is the bridge-lending business, which holds most of the capital. Arbor makes short-term loans, mostly against apartment buildings, funded largely with borrowed money. The loan book was $12.1 billion at mid-2026 against about $11.1 billion of borrowings. Roughly $1.25 billion of common shareholders' equity sits in this business, along with all $634 million of the preferred stock. Before loan losses and preferred dividends, it now produces roughly $110 million a year of cash. The preferred dividend takes $41.4 million of that. A full-cycle allowance for loan losses takes about $48 million. That leaves the lending business earning only about $20-25 million a year for common shareholders, roughly a 2% return on the equity committed to it. This is the central fact about the company today. The lending business is large, but after a normal cycle's losses and the preferred dividend it earns very little for the common shares. The value rests on the agency franchise plus whatever lending capital survives the current cleanup. Why we use a full-cycle loss rate. Loan losses never appear in cash flow, because a bad loan shows up as a smaller repayment, so we charge them separately. Losses were negligible from 2016 to 2022. They were then $5.7 million in 2023 and $12.2 million in 2024, and about $158 million in 2025 including write-downs and losses on foreclosed property. Averaging only the good years understates the cost of this business. Spreading the whole 2016-2029 experience, including the losses we expect from the current problem loans, gives about $0.26 per share a year (about 0.4% of the book), and we use that permanently in the base case. This lender has also been through a far worse cycle. In 2008 and 2009 it took $373 million of loan-loss provisions on a book of about $2.2 billion. Its GAAP book value fell to $3.81 per share by the end of 2009, and it stopped paying a common dividend. The current problem. At mid-2026, of $8.7 billion of apartment bridge loans, $360 million were rated doubtful (Arbor's own term for a probable loss), up from $216 million six months earlier. Another $587 million were substandard (Arbor expects to modify or foreclose) and $3.4 billion special mention (an observed weakness or a modification). Foreclosed property was $546 million. Most of the weak loans were made in 2021-2022 at peak prices, and Texas and Florida make up about 40% of the book. Several facts point to more losses than Arbor has reserved for. Its loss reserve fell from $239 million to $146 million during 2025 while the problem loans did not shrink. It stood at $163 million at mid-2026. At the end of 2025, specific reserves on its impaired loans equaled only about 11% of their value. In May 2026 Arbor said it planned to cut foreclosed property from $521 million to $250-300 million by year-end, yet by June it had risen to $546 million. Its unencumbered assets fell from 1.57 times unsecured debt at the end of 2025 to 1.40 times at mid-2026, against a required minimum of 1.20 times. With the 10-year Treasury near 5.2%, borrowers find it harder to refinance out of bridge loans into permanent agency loans. Other facts point the other way. Arbor's loans mostly sit first in line, at an average 77% of Arbor's estimate of property value. It sold $265 million of foreclosed property in 2025 for a loss of only $9.2 million against carrying value. Its $3.2 billion single-family rental book was at 64% loan-to-value with few problems at the end of 2025. The shareholder class action was dismissed in March 2026. July 2026 financing. Arbor sold $375 million of 6.25% convertible notes due 2029 (convertible at about $6.10 per share), repaid $270 million of 4.50% notes, and bought back about 21 million shares at $5.42. Our share count is about 184 million, including 16.2 million partnership units exchangeable into common shares. Base and bull values include the roughly 61 million shares the notes would become if the stock is above $6.10 when they come due. Base case (about $8.70 per share). We build losses from Arbor's own ratings. Doubtful loans lose about 40% and substandard loans about 20%. One in ten special-mention loans defaults at a 25% loss, and foreclosed property is written down another 10%. That totals about $400 million ($2.17 per share) from late 2026 through 2029, roughly two and a half times Arbor's reserve. Total cash earnings before losses and preferred dividends dip to $1.04 per share in 2027 and stay there through 2028. That is about 3% below the current run rate, against a 14% fall in 2025. The difference is that Arbor's move to more expensive unsecured debt ended with its April and September 2026 note redemptions. We credit only a small recovery, to $1.12 by 2029, from foreclosed property already being sold (Arbor reported $195 million of problem assets resolved or scheduled for resolution in July and early August 2026). That is far less than the $75-95 million a year Arbor says its problem assets cost it, because new problem loans have kept replacing resolved ones. After the full-cycle loss rate and preferred dividends, common shareholders keep about $0.64 per share a year. About $0.46 of that comes from the agency business. The agency licenses and servicing book carry the earnings beyond ten years; the lending business adds little. We credit no growth from agency volume, spreads or interest rates. Bear case (about $0.40). Losses reach about $950 million ($5.15 per share). In this case doubtful and substandard loans lose about half their value, a third of special-mention loans default at a 30% loss, and foreclosed property is written down another 22%. The agency business earns about $70 million a year, and the lending business shrinks as its capital is used up, so cash earnings fall to about $0.76 per share. For three years losses far exceed earnings, consuming about $600 million of cash. That is more than Arbor's cushion above its unsecured-debt asset test, which was about $370 million at mid-2026. In this case the company would have to issue shares, sell assets or restructure. After its lenders and preferred holders are covered, almost nothing is left for common shareholders: the business is worth only slightly more than it owes. This is not the worst outcome imaginable, but a plausible one built from Arbor's own loan ratings and from what happened to this lender in 2008-2009. Bull case (about $13.00). Problem assets are largely resolved by 2027, with total losses of about $165 million ($0.90 per share). Capital returns to earning loans and cash earnings recover to $1.35 per share by 2029. That is still below the $1.48 of 2025 and the $1.72 of 2024. Long-run losses run at $0.20 a year. How this differs from our earlier estimate today (base about $10.60, bear about $3.50, bull about $13.00). We now judge the two businesses separately, and that shows how little the lending book earns after a normal cycle. We raised the long-run loss rate from the 2016-2025 average ($0.17) to a full-cycle figure ($0.26). We built base-case losses from Arbor's own risk ratings, which gives about $400 million instead of $275 million. We stopped crediting most of the earnings recovery Arbor's management describes, because foreclosed property rose in the second quarter despite management's plan to reduce it. The bull case is unchanged. Margin-of-safety price. The bear case is close to zero, so no price protects against it. The margin-of-safety price therefore depends on evidence that the credit problem has stopped growing. If problem assets fall for two consecutive quarters while Arbor's unsecured-debt cushion holds, the losses in our bear case become much less likely, and the margin-of-safety price is $4.50. From that price the base case implies roughly 25-35% a year over three to four years, including dividends. Until that happens, no price qualifies, because the loss ratings were still worsening at the latest report.