Sensitivity and Stress Testing: Break-Even Points, Shock Sizes and What a Lender Should See

A Base Case Is One Path: What Sensitivity Work Adds

A projection in a feasibility study is a single path through an uncertain future. It fixes a rent, an occupancy, a cost budget and an interest rate, runs them through the operating statement and arrives at one coverage ratio. That ratio answers whether the loan pays if everything happens as assumed. It says nothing about how far any assumption can drift before the loan stops paying, which is the question a credit officer exists to ask. Sensitivity and stress testing answer it, and the answer is only as good as two choices a study makes and must disclose: which variables it moves, and by how much.

The federal texts that touch the subject name the variables and leave the magnitudes open. Bank policy comes first. Under the interagency guidelines that sit behind every bank's real estate lending policy, a policy that covers construction and development lending and completed commercial property should set out the feasibility study and the sensitivity and risk testing it expects, with the depth matched to the project's size and type, and the guidelines' own examples of what to test are how projected income responds when interest rates, vacancy or operating expenses change ^[1]. A 2006 interagency statement addressed to lenders with heavy commercial real estate exposure lists the same item among its underwriting standards, calling it sensitivity analysis or stress testing ^[2]. The OCC's commercial real estate booklet asks that the factors behind a property's income be weighed under both normal and stressed conditions, with capitalization rates, interest rates and coverage ratios stress-tested ^[3]. None of these passages sets a shock size, and the OCC states in its community bank stress testing bulletin that it does not endorse a particular method ^[4]. The sizes are the study's to choose, and therefore the study's to justify.

The USDA OneRD rule approaches the same point from the side of the projection itself. Its credit evaluation section requires projections that depart from historical performance to be substantiated and documented, and says the borrower's projections should be consistent with past performance ^[5]. Its application section asks for projected statements, or a financial model, that start from the current statements and continue until the project has shown a minimum of two years at full capacity or in stable operation ^[6], and every projection has to rest on a list of assumptions showing its basis ^[7]. For Community Facilities guarantees the rule goes a step further: the financial feasibility report must take into account any rate adjustment the promissory note allows ^[8]. On the SBA side, 13 CFR 120 allows the agency to make a feasibility study one of the conditions it attaches to a loan ^[9], and the thresholds a 7(a) study measures against are the coverage floors the site's SBA program page states: 1.15x for a standard 7(a) loan, which a startup must reach within two years of funding ^[10].

Conventions. This article follows the conventions of the Loan Analytics worked examples. Coverage is tested against the program floor the site's program page states, where the program has one, and against a stated 1.25x lender assumption, and the higher of the two binds. Equity is measured at the loan the cash flow supports, with fees recomputed on that loan. An acquisition is tested on the seller's historical earnings, not on projections.

This article sets out the method: which variables to move, how to size each move from the published record, how to solve the break-even points exactly, how to rank the drivers, how to build a combined case that hangs together, and what the lender should be able to read off the result. The site's own methodology page describes the sensitivity stage as the point where the inputs with the most influence on the result are stressed and coverage is recomputed for each of them ^[11]; what follows is the arithmetic behind that description. Two archived posts cover neighbouring ground. The Base Case Is a Hypothesis. The Stress Test Is the Evidence. calibrates shocks for a new hotel, and Multifamily Underwriting Amid Sustained Cost Shocks: A DSCR Stress Test follows operating cost pressure through apartment coverage. How the coverage ratio itself is assembled, and which floor binds, is the subject of the companion article on debt service coverage, and the three sensitivities the interagency guidelines name are run on a bank loan in the conventional loan method article.

Three Tools for Three Questions

A study can put three different questions to its own projection, and each needs a different tool.

The first question is which input matters most. One-variable sensitivity answers it: hold every input at its base value, move one by a stated amount, recompute coverage, put that input back, and move the next. The output is a ranking. It cannot say what a bad year looks like when several inputs move at once, but it tells the analyst and the reader where the projection is fragile, and therefore where the evidence has to be strongest.

The second question is what a plausible bad episode would do. Scenario analysis answers it. The OCC's bulletin describes scenario analysis as applying historical or hypothetical scenarios, extreme ones included, to judge the impact of events and circumstances ^[12], and it notes that the variables are usually set by a scenario with a coherent, logical narrative of why the events happen and in which combination and order ^[13]. The same bulletin calls the loan-level version transaction stress testing, which estimates loss on a single loan by judging how changing economic conditions affect the borrower's capacity to service its debt ^[14]. That is the nearest supervisory description of what the downside section of a feasibility study does.

The third question runs the other way: how far would each input have to move before the loan fails? The OCC defines reverse stress testing as assuming a specific adverse outcome and working back to the kinds of events that could produce it ^[15]. Its example concerns a bank's capital ratios, but the logic carries over to one loan without change. Fix the outcome (coverage of exactly 1.00x, or coverage at the lender's minimum) and solve for the occupancy, the rent, the rate or the cost budget that produces it. That value is the break-even point, and its distance from the base case is the cushion.

The three tools go wrong in different ways. One-variable tests understate risk when inputs move together. A stacked scenario overstates it when the moves it combines came from separate episodes in the record. A break-even point on its own is a number without a yardstick: a cushion of fifteen occupancy points is generous in a market whose occupancy has never moved more than three, and thin in one that has moved twenty. A sound downside section therefore uses all three and ties each to the same evidence. For portfolio stress testing the OCC asks for a projected base case together with one or more adverse scenarios drawn from macro and local data ^[16], projected over at least two years, because losses tend to surface over the two years that follow a downturn ^[17]. A study can borrow both disciplines at the scale of one project: at least one adverse scenario, and a horizon that reaches through the first two years of stable operation the OneRD rule already asks to see.

Which Variables to Move

The candidates come straight out of the cash flow. Revenue is the number of units, times the price per unit, times occupancy. Costs divide into two groups: those that rise and fall with collected revenue (a management fee charged as a share of collections, credit loss, the turnover cost of each move-out) and those that hold regardless of collections (property taxes, insurance, base payroll, the replacement reserve). Debt service follows from the loan amount, the rate and the amortization. Each of these is a possible driver, and a study should say which ones it tests and why it leaves the others alone.

The supervisory texts point to the same list. The OCC's booklet names historical, current and projected rental rates, operating expenses, capital expenditures, and vacancy and absorption rates among the factors an income analysis typically weighs ^[18]. For construction lending, its stress testing bulletin adds selling rates, leasing activity and oversupply as variables a loan may be sensitive to ^[19]. The interagency guidelines give interest rates, vacancy rates and operating expenses as examples ^[1].

The rate needs the most care, because how it moves depends on the note and not on the market alone. The OCC's booklet warns that when a project carries floating rate debt against fixed rents, rising rates can damage the borrower's repayment capacity ^[20]. On a variable-rate 7(a) loan the regulation settles how much of a market move reaches the borrower. On such a note the base is either prime or the Optional Peg Rate ^[21], or an alternative that SBA has approved and announced in the Federal Register ^[22]. Whatever the base, the note's rate fluctuates by the same amount as the base rate moves ^[23], and after the first change it may change no more often than monthly ^[24]. For loans above $350,000 the regulation allows no more than 3.0 percentage points of spread above the base ^[25], and because that ceiling governs the spread while leaving the base free, the full movement of the base rate passes into the note. The exception is a note that sets its own ceiling rate, which the regulation allows only if the ceiling sits no further above the initial rate than the floor sits below it ^[26]. The regulation also allows the initial amortization to be recomputed as rates move, as SBA directs ^[27], which changes the payment, and a study should say which payment rule it models. On a B&I guaranteed loan the lender may reprice a variable rate at different intervals during the term, but not more often than quarterly ^[28]. A fixed-rate loan with a reset date needs the shock applied at the reset and not in the first month; a fixed-rate loan with no reset inside the horizon needs no rate shock at all, and a study that runs one anyway is padding its table.

Timing is a variable too. A project that opens late or fills slowly carries debt service before it carries rent, and the companion article on ramp-up and absorption explains how a study proves a lease-up schedule; a delayed ramp belongs among the scenarios of any new project. Seasonality is another. The OneRD credit evaluation includes a projected quarterly cash flow analysis for borrowers whose cash flow is seasonal or cyclical ^[29]. An annual coverage figure can hide a quarter in which the property cannot pay.

What a study should not move is anything fixed by contract for the horizon tested: a signed lease at a stated rent, a fixed-rate note before its reset, a guaranteed maximum price on construction. Shocking a contracted number tests nothing except the contract. The study should instead state what is fixed, for how long, and what follows on the date it stops being fixed.

Sizing Each Shock From the Published Record

A shock size is an assumption, and it deserves the same sourcing as any other assumption in the model. Round numbers such as a ten percent revenue cut or a two point rate rise are easy to compute and hard to defend, because nothing ties them to the property's market or to its note. The method is to take each driver's adverse move from a published series that measures it, to state the series, the geography, the window and the frequency, and then to judge whether the project's own market is likely to be more or less volatile than the series. The rate and housing series below show the approach, and Table 1 lists the moves they record.

Rates. The Federal Reserve's H.15 release publishes weekly averages of the bank prime loan rate, which the second column of its weekly data package describes as the average majority prime rate charged by banks on short-term loans to business ^[30]. The weekly series shows several large moves in two decades. Prime stood at 4.00 percent in the week ending June 23, 2004 ^[31] and at 8.25 percent by the week ending July 5, 2006 ^[32]. It was still 8.25 percent in the week ending September 12, 2007 ^[33] and had fallen to 3.25 percent by the week ending December 24, 2008 ^[34]. It was again 3.25 percent in the week ending March 16, 2022 ^[35] and reached 8.50 percent in the week ending August 2, 2023 ^[36], a rise of 5.25 points inside eighteen months. It was still 8.50 percent in the week ending September 18, 2024 and 8.00 percent a week later ^[37], and the November 9, 2026 release shows 7.00 percent ^[38]. For a floating-rate loan, the move from 2022 to 2023 is the larger of the two rises in the period shown, and a base case that cannot absorb it is a wager that nothing like it recurs within the life of the loan.

Occupancy. The Census Bureau's Housing Vacancy Survey publishes quarterly rental vacancy rates from 1956 onward ^[39], each the share of the rental inventory that is vacant for rent ^[40]. That is a physical measure; a project's economic occupancy also reflects concessions and credit loss, so a study that converts one into the other should say how. Nationally the rate rose from 7.9 percent in Q1 2000 ^[41] to 10.4 percent in Q1 2004 ^[42], a rise of 2.5 points over four years, and it peaked at 11.1 percent in Q3 2009 ^[43]. It fell from 9.7 percent in Q1 2011 ^[44] to 7.1 percent in Q1 2015 ^[45]. More recently it climbed from 5.6 percent in Q2 2022 ^[46] to 7.6 percent in Q3 2026 ^[47]. Some regional rates moved further than the national one. Over the same four years from 2000, the Midwest rate went from 8.5 percent ^[41] to 12.3 percent ^[42], and in Q3 2009 the South stood at 14.2 percent ^[43]. For sizing, the lesson is that a national move averages regions that moved by different amounts, and a submarket can sit at either end of that spread. A study for a particular submarket should use the most local series available and print the national series beside it.

Rent. The same survey publishes a median asking rent, and for a vacant unit the survey's rent is the amount asked at the time of interview rather than a contract rent agreed on an occupied unit ^[48]. Nationally it was $723 in the first quarter of 2009 and $680 in the fourth quarter ^[49], a fall of 5.9 percent within the year, while the annual median moved only from $708 for 2009 ^[49] to $694 for 2011 ^[50]. The quarterly fall shows how far asking rents on new leases dropped at the worst point of that year; the annual figures show how little the level moved across the cycle once a full year is taken together. Both describe asking rents on vacant units rather than rents on leases in place, so applying either to a whole rent roll at once overstates the first year and approximates the effect only once leases have turned over. The survey also publishes margins of error for its rental vacancy rates ^[51], and a study should not mistake sampling noise for a market shock: a swing from one quarter to the next that sits inside the margin is not evidence of volatility.

Lodging. Lodging revenue has moved far more than any housing figure above. The Census Bureau's Quarterly Services Survey put traveler accommodation revenue for the second quarter of 2020 at 70.8 percent below the same quarter of 2019, on figures not adjusted for seasonality ^[52]. That quarter is a tail event, not a recession case, and a hotel study should say which of the two it is testing. The housing series do not transfer to lodging. A hotel study should draw its occupancy and rate shocks from the subject's own operating history where one exists and from the competitive set it surveyed, and should state which.

Costs. No public series is used for the expense shock in the illustration that follows; the size of that shock is the illustration's own assumption, marked as such among the inputs. The companion article on operating expense evidence covers how a study builds an expense budget a reviewer can trace, from the property's own records, tax bills and insurance quotations.

Four rules follow from the record. Match the series to the line it shocks: an asking rent series shocks new leases, a vacancy series shocks occupancy, a base rate series shocks a floating note. Prefer the most local series available and print the national one as a reference. Keep the window comparable to the horizon of the loan. And state whether the move tested is the worst in the window examined, so that the reader knows whether the test is a recession case or a tail.

The Illustration: One Apartment Property, Every Input Printed

The worked illustration is a 120-unit apartment building financed with a conventional floating-rate loan priced at prime plus a spread. Every operating input is an assumption chosen to show the method, not a market finding, and the shock sizes come from the series above. A conventional loan carries no program floor of its own, so the illustration holds the property to the stated 1.25x lender assumption and also marks 1.00x, the point where it stops paying its debt from operations. Every input, with its source or its label as an assumption, appears in Table 2.

Rounded as shown; lines may not foot by one unit in the last place.

A rent of $1,450 on 120 units gives gross potential rent of $2,088,000. At 93.0 percent economic occupancy the building collects $1,941,840, of which 8.0 percent goes to costs that move with collections and $640,000 to fixed costs, leaving net operating income of $1,146,493. At a note rate of 9.00 percent (prime of 7.00 percent plus a spread of 2.00 points) on a 25-year amortization, debt service on the $8,400,000 loan is $845,910, and coverage is 1.36x with $300,583 left after debt service, as Table 3 shows.

Coverage of 1.36x clears the assumed 1.25x minimum; the question is how much of that margin survives the record.

Break-Even Points, Solved Exactly

A break-even point is the value of one input at which coverage falls to a stated level while every other input holds at its base. Two levels matter. At 1.00x the property's income exactly pays its debt service and nothing is left over; below it, the debt can be paid only from money outside the project. At the lender's minimum, here the assumed 1.25x, the property just meets the coverage the lender requires. Income-producing real estate loans may carry a coverage covenant ^[53], and where one is written at that level, falling below it is a breach even while the loan still pays; the covenant's own definition of coverage may differ from the one used to underwrite the loan ^[54], so the solve should use the definition the lender names. A study should solve both points, because the second sits much closer to the base case than the first.

Break-even occupancy has a closed form. Take the rent the building would collect if every unit were let, deduct the share of it that goes to costs moving with collections, and the result is what each point of occupancy contributes toward fixed costs and debt. Divide fixed costs plus debt service by one hundredth of that full-occupancy contribution and the answer is the occupancy at 1.00x; replace debt service with debt service times the lender minimum and the answer is the occupancy at that minimum. Break-even rent follows the same logic with occupancy held at its base and the monthly rent solved instead.

A familiar shortcut treats every operating cost as fixed: it adds them all to debt service and divides by gross potential rent. It is quick, and whenever the base case clears break-even it overstates the answer, by an amount that grows with the variable share. When occupancy falls, the costs that move with collections fall with it, so treating them as fixed overstates the break-even. In the illustration the gap is small, 78.6 percent against the exact 77.4 percent, because only 8.0 percent of collections is variable. In a hotel, where housekeeping, commissions and franchise fees scale with every room sold, the gap is far wider, which is why the hotel worked example builds its break-even from what each revenue dollar contributes after variable costs.

The rate has no closed form. Debt service depends on the rate through the amortization formula, and that formula cannot be rearranged to give the rate directly. The model therefore solves for it with Newton's method: start at the note rate, compute the payment and its slope, step to the rate at which a straight line through that point would meet net operating income, and repeat until the step vanishes. Eight steps are carried, and the answer stops changing after the third. Because an iterated answer invites the question of whether it has converged, Table 4 prints the proof beside each solution: coverage recomputed at the solved break-even rate, which reads exactly 1.00x, and coverage at the rate solved for the assumed lender minimum, which reads exactly 1.25x.

Read the table from the top. Occupancy can fall from 93.0 percent to 77.4 percent before the building stops covering its debt, a cushion of 15.6 points; the national move tested is 2.5 points, so the cushion covers it 6.3 times. Rent can fall from $1,450 to $1,206, a cushion of 16.8 percent against a recorded move of 5.9 percent, covered 2.8 times. The note rate can rise from 9.00 percent to 13.13 percent, a cushion of 4.13 points against a recorded move of 5.25 points, so the cushion covers only 0.79 of it. Expressed as income, net operating income can fall 26.2 percent before coverage reaches 1.00x. The points at the lender minimum are far closer: an occupancy of 88.4 percent, a rent of $1,378 or a note rate of 10.02 percent each takes coverage down to the assumed 1.25x on its own.

Those two sets of numbers carry the analysis. On occupancy and rent this loan has room against the national moves tested here. On the rate it does not: a repeat of the move in prime from 2022 to 2023 would push the property below 1.00x, and a rise of little more than one point would take coverage below the assumed minimum. The occupancy result also depends on geography. The Midwest move of the early 2000s was larger than the national one, so a study in a submarket with that kind of history would test the local move rather than the national.

Break-even is a contract term as well as an analytical one. The OCC's booklet notes that standby and forward take-out commitments require completed construction and, in most cases, carry performance conditions such as lease-up to break even or better at minimum rental rates ^[55], so on a construction loan the occupancy break-even can decide whether the permanent financing funds at all.

Ranking the Drivers: The Tornado Table

A tornado chart lays the one-variable results side by side with the widest bar at the top, so that the driver which moves coverage most is read first. The table form used here does the same work: each column applies one recorded move on its own, and the columns run from the move that takes the most coverage to the move that takes the least.

The order in Table 5 is rate, rent, fixed costs, occupancy. The recorded rate move costs 0.43x of coverage and uses 119.7 percent of the cushion above 1.00x, which is another way of saying that it breaks the loan: coverage falls to 0.93x. The recorded rent move costs 0.13x and uses 35.3 percent of the cushion. The assumed rise of ten percent in fixed costs costs 0.08x. The recorded occupancy move costs 0.06x and uses 16.0 percent.

Two features of the method matter more than this particular order. First, each driver moves by its own evidence, not by a common percentage. A uniform ten percent shock would move occupancy by more than nine points, nearly four times the national move tested, while moving the note rate by less than one point, a fraction of what prime did in 2022 and 2023. A uniform shock ranks drivers by arithmetic sensitivity, which is a property of the formula; an evidence-sized shock ranks them by exposure, which is a property of the loan and its market. The lender's question is about exposure.

Second, the table shows only the adverse side. A full tornado also plots the favourable move for each driver, and a study may print both, but a credit decision turns on the downside, and the ranking a reviewer acts on is the ranking of the adverse bars. Where a driver's favourable and adverse moves differ in size, as the prime record does (a rise of 5.25 points from March 2022 against a fall of 1.50 points since September 2024), the asymmetry is a finding in its own right and belongs in the text.

The order tells the study where to spend its evidence, and it tells the borrower what to fix before the request goes in. Here the loan is exposed to the base rate far more than to the rental market, so the remedies worth testing sit on the debt side: a fixed rate, a rate cap, a smaller loan or a longer amortization, each of which the model can test by changing a single input.

Combined Stress: Coherent Episodes Against a Stacked Case

One-variable results cannot show what happens when several inputs move at the same time, and some do. In 2009 the national rental vacancy rate rose from 10.1 percent in the first quarter to 11.1 percent in the third ^[43] while the median asking rent fell from $723 to $680 between the first and fourth quarters ^[49], so occupancy and price turned against owners together. A study needs a combined case for that reason, and the OCC's bulletin supplies the discipline for building one: in its description, scenario variables are usually set by a narrative that explains which events occur together and in what order ^[13].

The record also shows moves that came in different episodes. The vacancy rise from 2000 to 2004 came while prime ended far below where it began: 8.50 percent in the week ending January 5, 2000 ^[56] against 4.00 percent in the week ending June 23, 2004 ^[31], the last full week at that level before the climb to 8.25 percent ^[32]. The rate surge from 2022 to 2023 began with national vacancy at 5.8 percent in Q1 2022 and 5.6 percent in Q2 ^[46], well below the 7.9 percent of early 2000 ^[41]. A scenario that stacks the largest vacancy move of one episode onto the largest rate move of another combines moves the record shows in different periods. It can still be worth running, but it should carry a label saying so, and it should not be the only adverse case the lender sees.

Table 6 runs two coherent episodes and the stacked case. The demand downturn moves occupancy and rent by their recorded amounts and holds the rate at its base, taking no credit for a fall in prime like the one to 3.25 percent by December 2008 ^[34]. The rate and cost rise moves the rate by the 2022 to 2023 record and fixed costs by the assumed ten percent, with occupancy and rent held. The stacked case applies all four moves together.

The demand downturn takes coverage to 1.18x: the property still pays its debt with $149,164 to spare, but it sits below the assumed 1.25x minimum. The rate and cost rise takes coverage to 0.88x, a shortfall of $150,218 a year, and the stacked case to 0.76x, a shortfall of $301,637. In both rate cases the break-even occupancy is above full occupancy (100.8 percent and 107.2 percent), which means that no occupancy the building could reach would pay the debt at that rate: the shortfall cannot be leased away.

The last two rows turn each scenario into a loan amount, often the most useful form for a lender. The income in the rate and cost case carries $7,376,376 at 1.00x and $5,901,101 at the assumed 1.25x minimum, against the $8,400,000 requested. That gap measures how far the structure would have to change (a smaller loan, more equity, a fixed rate or a cap) for the loan to survive the scenario, and it is a finding a study can state without recommending any one remedy.

What a Lender Should See in the Downside Section

A downside section a credit officer can rely on lets the officer check every step without rebuilding the model. At a minimum it should contain the following.

• The inputs, each with its basis. Every operating input labelled as an assumption or tied to a source, so the reader can separate evidence from judgement; the OneRD requirement that projections be supported by a list of assumptions ^[7] writes the same idea into a regulation.
• The shock list, each with its record. For every driver tested, the series, the geography, the window, the frequency and the move, as in Table 1, with a sentence on whether the project's market is likely to be more or less volatile than the series.
• The one-variable results in ranked order. The coverage at each move and the coverage given up, with the order explained in words.
• At least one coherent adverse scenario. A scenario whose moves the record shows together, with any stacked hypothetical marked as one.
• The break-even points, solved and checked. Occupancy, price and rate at 1.00x and at the binding minimum, with coverage recomputed at each solved value so the solution can be verified at a glance.
• The cushion set against the record. Each cushion expressed in the units of its series and divided by the recorded move, so a reader can see at once whether a cushion covers the record several times over or only part of it.
• The threshold, stated. The program floor where one applies (for a 504 loan the site's SBA program page states 1.15x, measured either on the borrower's history or on projections for two years ^[57]), the assumed 1.25x lender minimum, and which of the two binds.
• The horizon. The years tested, reaching at least through the first two years of stable operation that OneRD projections must cover ^[6], with the ramp-up years before them where the project is new.

A section built this way lets the lender answer the question that matters at this stage: how far is the base case from failure, and is that distance large compared with what the market has actually done? Coverage of 1.36x is a statement about one path. A rate cushion that covers 0.79 of the last tightening is a statement about the loan.

The Loan Record as a Frame, Not a Probability

A stress test concerns one project. The SBA loan record concerns many thousands of loans, and it is tempting to borrow its loss rates as the chance that this particular project fails. The record does not support that use. The Loan Analytics dataset measures charge-offs by approval cohort, and every rate it publishes is a floor: the 7(a) cohort approved in FY2017 to FY2021 shows charge-offs to date of 6.33 percent ^[58] of disbursed loans, while for 504 loans approved in those years the figure is 0.35 percent ^[59]. With the horizon held at five years from approval, the 7(a) cohort shows 3.72 percent ^[60] and the 504 cohort 0.17 percent ^[61].

The five-year measure also differs from one approval year to the next. For 7(a) loans it was 4.89 percent ^[62] for the FY2010 cohort and 2.58 percent ^[63] for the FY2020 cohort, the highest and the lowest of the twelve cohorts from FY2010 through FY2021. The dataset does not say why, and this article does not guess. The spread alone shows that a program-wide rate is not a constant that can be applied to any loan made in any year.

For hotels and motels, the asset class of the hotel worked example, the 7(a) floor for the FY2017 to FY2021 cohort is 0.99 percent ^[64], and 0.27 percent ^[64] of those loans were written off inside five years; under 504, the floor for the class is 0.55 percent ^[65]. These are approval-cohort floors on pools of loans the file describes only in outline. They say how often loans of a kind have been written off so far. They do not say which shock caused a loss or what coverage the loans were underwritten to.

Their proper place in a study is as a frame. A class floor that sits high argues for a demanding stress case and for a clear account of the evidence behind the base case; a floor that sits low is not an argument for a gentler one. Why a published SBA loss rate can only understate is the subject of The Approval-Cohort Charge-Off Floor: Why a Published SBA Loss Rate Can Only Understate, and what the record leaves out is set out in What the SBA Loan File Cannot Tell You: The Limits of FOIA Data in Underwriting. The figures this section relies on are listed below with their keys.

Loan Data Behind This Article

Every figure in this table comes from the Loan Analytics loan dataset, analytics-loan-data freeze-3 (commit 360c856), and carries its key. SBA rates are approval-cohort floors and are withheld when fewer than 30 loans stand behind them.

Faults That Empty a Sensitivity Table of Meaning, and What to Check

A sensitivity table can be long, tidy and still leave the lender's question unanswered. Each fault below can be checked from the page.

Shocks with no source. A round-number shock with no series behind it cannot be compared with anything. Ask for the series, the window and the geography of every move, and check that the move tested is the move the series shows.

A rate shock on the wrong date. Applying a rate shock to a fixed-rate note from the first month, or leaving it out at a reset that falls inside the horizon, misstates the exposure. On a variable 7(a) note the base rate passes straight through ^[23]; on a variable B&I note repricing comes no more often than quarterly ^[28]; on a fixed note the shock waits for the reset.

Revenue moved, costs frozen. Cutting occupancy while holding the costs that move with collections overstates the damage, and the error grows with the variable share. The model should carry the split between variable and fixed costs so that each shock passes through correctly, and the break-even should be solved from contribution rather than from the shortcut.

Fees left out of debt service. Recurring fees that scale with the loan belong in stressed debt service. The hotel worked example carries the USDA retention fee inside its coverage test for that reason.

Break-even only at 1.00x, or found by trial and error. A cushion to 1.00x can look generous while the cushion to the lender minimum is a fraction of it, as Table 4 shows for every driver, and a rate found by trying values in a spreadsheet and rounding cannot be verified by the reader. Solve both points and print the coverage recomputed at each solved value.

A national series applied locally without comment. A national rate can understate a local move, as the Midwest vacancy figures in Table 1 show. Where a local series exists, use it; where none does, say so and explain how, if at all, the national move was scaled.

A noisy series read as volatility. Quarterly survey estimates carry sampling error, and the Housing Vacancy Survey publishes margins of error for its rates ^[51]. A study should check that the move it tests is larger than the noise around it.

No ramp-up case. A stabilised-year test says nothing about the years before stabilisation; a delayed opening and a slower lease-up belong among the scenarios of any project without an operating history.

A base case detached from history. For a USDA guarantee, projections that depart from the borrower's historical performance must be substantiated and documented ^[5]. A stress test that starts from a base case well above history, without explaining the gap, measures its cushion from the wrong starting point.

The three worked examples this article draws on apply the same method to three different revenue units. The hotel example states its break-even in occupancy, daily rate and revenue per available room; the short-term rental example states it per booked cabin-night; the apartment example states it in economic occupancy and average rent. In each, the useful number is not the break-even itself but its distance from the base case, read against what the market has shown it can do.

Related Pages on Loan Analytics

• The Loan Analytics methodology series (all articles)
• How a feasibility study is done: the Loan Analytics methodology
• SBA feasibility studies (7(a) and 504)
• USDA feasibility studies
• Bankable feasibility studies for conventional loans
• Loan data by state
• Loan data by asset class
• Worked examples (case studies)
• Worked Example: USDA Business and Industry Hotel Feasibility Study in Connecticut
• Worked Example: Conventional Community Bank Construction-to-Permanent Short-Term Rental Feasibility Study in New York
• Worked Example: Conventional Bank Construction-to-Permanent Multifamily Feasibility Study in Nebraska
• The Base Case Is a Hypothesis. The Stress Test Is the Evidence.
• Multifamily Underwriting Amid Sustained Cost Shocks: A DSCR Stress Test
• Debt Service Coverage in a Feasibility Study: How the Ratio Is Built, and Which Floor Binds
• The Approval-Cohort Charge-Off Floor: Why a Published SBA Loss Rate Can Only Understate

Commission a Feasibility Study

Loan Analytics prepares lender-grade feasibility studies for SBA 7(a), SBA 504, USDA Business and Industry, conventional and EB-5 financing, built on the methods this series sets out. A study is not an appraisal: it tests whether a project's market, costs and cash flow can carry the debt a lender is asked to approve. To discuss a study for your project, contact Loan Analytics.

Daniel Smith, MAI, Loan Analytics