CRE Financial Modeling Inputs Lenders Scrutinize Most

Commercial real estate lending hit an estimated $706 billion in 2025, up 40% from $505 billion the year before, according to the Mortgage Bankers Association's April 2026 report. Add the roughly $957 billion in loans that matured last year, plus another $1.5 trillion coming due in 2026, and credit committees are moving more paper than they have since before the pandemic. That volume presses down on a handful of inputs. When standards tighten at the same moment volume rises, the assumptions that used to slide through review stop sliding.

CBRE's Lending Momentum Index climbed 67% year-over-year in the fourth quarter of 2025, landing at 1.2 and closing in on pre-pandemic levels. Even so, lenders have continued tightening standards on commercial real estate credit, approving fewer deals, capping leverage lower, and asking for more documentation. Sponsors are refinancing into higher rates and into an underwriting environment that scrutinizes fewer inputs but hits each one harder. Knowing which inputs those are, and why, is the difference between a package that closes and one that sits on someone's desk for another month.

How lenders approach a model before they read a single number

A credit officer opening a sponsor's model asks where the optimism is hiding before asking what the returns look like. There's always some hiding somewhere. Underwriting, at its core, is the job of finding it before it becomes the bank's problem instead of the sponsor's.

Lenders sort a model into two buckets in their heads: inputs they can check against outside data, and inputs that rest on a judgment call. Nearly all the scrutiny lands on the second bucket. In-place rent, checked against a signed lease, is close to settled fact. A cap rate assumption for a disposition three years out is judgment stacked on judgment.

The operating story of a property reads the same way every time: where the income comes from, how durable it is, what it costs to keep the asset in shape, what's left after debt service. Lenders read those in order and check for consistency across them. Rent growth outpacing submarket supply, an exit cap that doesn't match recent transaction evidence, expense ratios frozen in 2021. These usually trace back to one root cause: the sponsor's numbers never got tested against anything outside the spreadsheet.

Version control matters too, before anyone even checks a number against outside data. Move inputs by hand across tabs enough times and assumptions drift without leaving a trace. That drift shows up again and again in CMBS files as a model passes from underwriter to credit officer to capital markets desk. Lenders also run scenarios where several assumptions go bad at once, because that's closer to how real distress shows up.

NOI: the input everything else is built on, and the first place lenders push back

Net operating income sits at the center of the model, the numerator in DSCR, the denominator in cap rate valuation. Get it wrong and coverage and value distort at the same time, in opposite directions.

Underwriting desks routinely run a stressed NOI, cutting it 5 to 10 percent to simulate turnover or softening rents. Before that stress test even runs, lenders are already picking apart the baseline. Do in-place rents reflect signed leases, or asking rents nobody's actually tested in the market? Did the sponsor normalize for one-time income, like free-rent burn-off, that inflates the trailing twelve months without reflecting anything repeatable? Is rent growth tied to submarket data, or lifted from a metro-wide average that has nothing to do with the actual building? Do expense lines reflect current insurance, tax, and management cost inflation, or last year's actuals rolled forward on autopilot?

That last question has gotten sharper lately. Construction costs, operating expenses, and tenant improvement allowances all moved up hard under sticky inflation, so any expense assumption underwritten before 2023 carries real risk of understatement baked in. A model can hand over a clean NOI figure and still misrepresent what the asset earns, if either side of that ledger anchors to stale data. Lenders know where those anchor points hide, and they go looking there first, every time.

DSCR: where interest rate assumptions meet income assumptions, and both get tested at once

Debt service coverage ratio compares NOI to annual debt payments. Above 1.0 means the property covers its debt. The real question is how much cushion sits above that line, and whether the cushion survives a stress scenario.

Rate movement has already changed the math for sponsors refinancing this year. Average DSCR loan rates fell from 8.73% in early 2024 to around 7.76% by the first quarter of 2025. Real relief, but still well above the environment many existing loans were originally underwritten in. Lenders stress the modeled rate directly, typically adding 100 to 200 basis points, to see if coverage still clears the floor. On floating-rate loans, this is standard practice, not a special case.

Two situations fail this test more often than sponsors expect. First, a deal underwritten at a favorable rate sitting just above the lender's DSCR floor, where 100 basis points of stress sinks it. Second, and more common this cycle, a straight refinance where the original loan carried a much lower rate. Even with NOI flat, debt service is higher now, and that alone produces a coverage problem the sponsor's own model may never have caught. Commercial mortgage loan spreads averaged 197 basis points in the fourth quarter of 2025, tighter at 142 for multifamily. With the 10-year Treasury at 4.25% as of April 2026, implied coupon rates worked out to roughly 5.79% for multifamily, 5.87% for industrial, 6.01% for retail, 6.45% for office. A sponsor modeling a rate that doesn't line up gets flagged fast. The lender is pricing off the same market the sponsor is supposed to be watching.

Cap rate assumptions: the most contentious number in any CRE model

No input generates more friction between sponsor and lender than the cap rate, and it isn't close. The going-in cap justifies purchase price. The exit cap drives IRR across the entire hold. An aggressive number on either one distorts the whole return story without ever making the income section look wrong.

The divergence problem shows up across every deal type. An exit cap underwritten at 4.5% in 2021, in a market that went on to trade between 5.25% and 5.75% through 2024, leaves a disposition value gap running through every projected return in the model. Current benchmarks give lenders a clean way to check this. Core multifamily going-in cap rates averaged 4.75% in the second quarter of 2025 per CBRE Research, with rent growth assumptions running 2.8% to 3.3% across core and value-add product. Retail net lease cap rates averaged 6.96% in the first quarter of 2025, up 58 basis points year-over-year. CBRE's 2026 forecast calls for cap rate compression of 5 to 15 basis points across most property types, a modest signal that leaves most of a real underwriting gap unclosed on its own.

Lenders apply something close to a 90-day rule: a cap rate assumption older than that gets checked again before it makes it into a final model, whether the sponsor thought to do it first or not. Tier divergence makes this worse. The spread between Class A and commodity assets has widened materially, and an exit cap averaged across tiers misrepresents whatever specific asset is actually on the table. A sponsor who can point to a comp closed last quarter, right tier, right submarket, gets that number accepted far faster than one citing a blended market average. That gap in evidence is often the whole reason a file needs another lap before it reaches committee.

Rent roll quality and vacancy: where market-level data misleads and submarket data is what actually matters

Rent roll and vacancy diligence has gotten sharper because sticky inflation changed how lenders read tenant credit and lease structure, not just the headline rent figure. Underwriters check whether rents come from executed leases or a market survey, how the expiration schedule lines up against rollover risk during the hold, and whether tenant credit can actually support the roll. That last question matters most in retail and office, where one sick anchor drags the rest of the property down with it. On active acquisitions, lenders expect asking rents in the model to match comps executed in the last 30 to 60 days, not comps pulled from earlier in the shopping period.

Vacancy is where sponsors trip up most, because the mistake looks defensible on paper. In gateway office markets, Class A CBD vacancy runs 12% to 15%, while commodity suburban product sits above 20%. A model built on a blended metro figure misstates risk in one direction or the other, depending on which product the asset actually is. Take a Sun Belt multifamily deal underwritten at 4% rent growth off metro-level data, where submarket vacancy actually runs 12% and roughly 3,000 new units are set to deliver within 18 months. That rent growth number just isn't achievable. The model pulled from published data, but the data wasn't the right unit of analysis for that asset, so the projection collapses under scrutiny despite looking clean on its face. Retail vacancy nationally hit 4.2% in 2025, the lowest reading since 2007, a genuinely good number that still hides wide variance by format, trade area, and anchor mix.

Lenders want vacancy sourced at the submarket level, from a named and current data provider. Not a year-old market report pasted into the back of the offering memo.

Expense assumptions: the inputs sponsors underestimate and lenders quietly recast

Lenders build their own NOI rather than trust the sponsor's. When the two numbers diverge, and they often do, the gap almost always lives on the expense side, not the revenue side.

Insurance has moved the most visibly, especially in coastal and Sun Belt markets, where catastrophic risk repricing has pushed premiums well past whatever the prior year's actuals show. Property taxes carry a related trap: a reassessment triggered by the sale price itself can reset the tax basis in year one, a cost any model built off the prior owner's actuals will miss entirely. Management fees get understated when a sponsor plans to self-manage and skips the market-rate management cost a lender will apply anyway. Capital reserves get recast often too. Lenders apply their own replacement figures for roofing, HVAC, and building systems based on the asset's age, and a model showing minimal or zero reserves gets adjusted upward with little discussion. Tenant improvement and leasing commission budgets round things out, particularly contested in office and retail, where longer free-rent periods and heavier build-out expectations have pushed real leasing costs up.

Sticky inflation is the thread running through all of it. A model that treats its expense ratio as flat while individual cost lines inflate underneath it has an internal contradiction an underwriter can spot in about ten minutes. Lenders want expense assumptions sourced to current market operating data for comparable assets, not trailing twelve-month actuals lifted straight off the current owner's P&L.

How lenders use sensitivity analysis to find the assumption doing the most work

Standard underwriting tests income declines, expense increases, and rate changes together, in combinations that approximate real stress. Occupancy softness and rate increases tend to show up in the same environment, not one at a time, so testing them separately misses the point.

A sensitivity table tells an underwriter a lot about how a model is actually built. If DSCR holds up fine under rate stress but collapses under a 5% NOI decline, the whole model is leaning on an income assumption harder than it should. If returns survive an income shock but fall apart when the exit cap moves just 25 basis points, the IRR is being carried by a terminal value guess rather than operating performance, and lenders will call that equity-story math, not credit-story math. If the sensitivity table only ever tests one variable at a time, that alone tells the lender the sponsor never modeled realistic correlated stress in the first place.

Lenders then rerun the sponsor's deal entirely under their own assumptions, a parallel model that's gotten more rigorous as delinquency rates have risen and contested refinancings have forced older underwriting back under a microscope. A sponsor whose own sensitivity table already reflects lender-grade stress has, in effect, already answered the credit officer's questions before they're asked. Review moves faster, and the credibility built on one deal carries into the next.

The data infrastructure behind inputs that hold up under scrutiny

The volume of data available to underwriters has expanded dramatically over the past several years, and that scale increase solves nothing on its own. The harder question was never how much data exists. It's which dataset applies to which specific assumption on which specific asset, and most of the failures below come from getting that match wrong.

The failure modes lenders catch most often are mundane, which is exactly why they keep happening. Rent and cap rate assumptions that were accurate when the deal was first underwritten but never got refreshed before the final package went out, missing the 30-to-60-day window active acquisitions require. Manual transfers across Excel tabs and files, introducing rounding errors or broken links as a model moves from underwriter to credit officer to capital markets desk in a CMBS process. Submarket data swapped for a metro average, not out of laziness but because submarket figures weren't sitting there ready when the sponsor was building the file under deadline. Version control failures, where the file actually sent to the lender isn't the one that got stress-tested internally days before.

These are the ordinary friction points of building financial models under time pressure, and they separate a package that survives due diligence from one that doesn't. Every number needs a source, a date, and something outside the spreadsheet to check it against. Simple to write down. Much harder to actually clear when you're the one building the file under deadline, at midnight, with the committee meeting in the morning.