Scenario, Sensitivity and Risk Analysis
A single forecast is a claim about the future that will be wrong. A range is a claim about what the business can absorb, which is useful. This guide is about building that range properly, and about presenting it so that it does not collapse back into a point estimate the moment it reaches a board paper.
- Sensitivity against scenario, and why they are not the same tool
- Choosing the two or three assumptions that actually matter
- Covenant and debt service headroom
- What Indian regulators do and do not require here
- Presenting a range without losing it
Sensitivity and scenario are different tools for different questions
Sensitivity
Move one assumption, hold everything else. Answers: how much does the result depend on this input? Its value is diagnostic. It tells you which two or three of your forty assumptions are worth arguing about.
Its limitation is that assumptions do not move one at a time. A demand shock does not leave your input costs untouched.
Scenario
Move a coherent set of assumptions together, in a way that describes a world that could actually happen. Answers: what does the business look like if things go this way?
Its value is that it can be discussed. A board can have a view on whether a scenario is plausible. Nobody has a view on a plus or minus ten per cent sensitivity table.
Finding the assumptions that carry the answer
Run a one at a time sensitivity across every material assumption and rank them by how much they move the output you care about. In most operating businesses two or three assumptions carry most of the variance, and they are frequently not the ones that took longest to build. Price and retention usually beat headcount phasing. In a leveraged structure, the interest rate and the refinancing assumption usually beat everything operational.
Two traps. The first is calibrating the sensitivity range to what feels comfortable rather than to what has actually happened: if your input cost moved thirty per cent in the last three years, a plus or minus five per cent test is decoration. The second is testing an assumption in isolation when it is mechanically linked to another, which produces a number that cannot occur.
Once the two or three are identified, build scenarios around them rather than around everything. Three well constructed scenarios beat twelve, because three can be discussed and twelve can only be filed.
How to build scenarios so they can be changed later
Scenarios belong in the model's structure, not in copies of the file. A separate workbook per scenario guarantees that within a month the scenarios will differ in ways nobody intended, because a correction made in one will not have reached the others.
The structure that works is a scenario block on the assumptions sheet: one column per scenario, a single switch cell, and every assumption in the model reading from the active column. Changing the switch changes the whole model. Adding a scenario is adding a column. Correcting an error corrects it everywhere at once.
Name scenarios for what they describe, not for how they feel. Base, upside and downside tell a reader nothing. Demand holds, competitor enters at a lower price and principal customer does not renew are scenarios a board can actually have an argument about, which is the point.
Covenant and debt service testing
For any business carrying debt, the question is not what the profit is in a bad year but whether the covenants hold and the debt is serviced. That is a different calculation and it belongs in the model explicitly.
Build the covenant tests as their own block: the ratio as defined in the facility agreement, calculated each test date, with headroom shown as a percentage rather than a pass or fail flag. A pass tells you nothing about how close you came. Then run the scenarios through that block and record the breach point, which is the value of the driving assumption at which the covenant fails. That single number is usually the most useful output of the entire exercise, and it is the one a credit committee will ask for.
Define the ratios as the agreement defines them, not as the textbook does. Earnings before interest, tax, depreciation and amortisation in a facility agreement is a defined term with adjustments and exclusions, and modelling the textbook version produces headroom you do not have.
What Indian regulators actually require, which is less than commonly claimed
It is often written that the Reserve Bank requires stress testing of a borrower's projections. It does not, at least not in those terms: a full text search of the 2025 Directions turns up no occurrence of sensitivity analysis, scenario analysis or net present value. The Reserve Bank of India (Project Finance) Directions, 2025, which took effect on 1 October 2025, require a techno-economic viability study where the aggregate exposure of all lenders is 100 crore rupees or more, certification of completion stages by an independent engineer or architect, and ongoing monitoring by the lender of project performance and any build up of stress. Debt service coverage appears in those Directions only as a parameter to be reported, in the project profile at Annex 3. No minimum is prescribed.
What follows is not that stress testing is optional. It follows that stress testing is a lender's credit requirement and a board's governance requirement, not a statutory formula you can satisfy by hitting a published number. Individual lenders set their own cover ratio floors, and those are the numbers that will actually be applied to your model.
For a listed company, risk management is a board level obligation under Regulation 21 of the Securities and Exchange Board of India (Listing Obligations and Disclosure Requirements) Regulations, 2015, which requires a risk management committee. Regulation 21(5) now applies the requirement to two populations: the top 1,000 listed entities, and a high value debt listed entity. The second limb is the one usually missed, and it was added by the Fifth Amendment Regulations with effect from 7 September 2021. The top-1,000 figure was itself raised from 500 by the Second Amendment Regulations with effect from 5 May 2021.
How the top 1,000 is determined changed in 2024, and this is the part that dates older guidance. The words requiring the ranking to be taken by market capitalisation as at the end of the immediately preceding financial year were deleted from Regulation 21(5)(i) by the amendment regulations notified on 17 May 2024, because the ranking mechanism moved into the generally applicable Regulation 3. Under that mechanism the stock exchanges prepare the list from the average market capitalisation between 1 July and 31 December of the relevant calendar year rather than from a single 31 March snapshot, the first such list having been prepared as at 31 December 2024. A newly covered entity complies within the period Regulation 3 prescribes, and an entity does not leave the population until it has stayed outside the threshold for three consecutive years.
So anything that still describes the test as a 31 March market capitalisation is describing the position before May 2024. Where the regulation applies, scenario analysis is one of the few ways a committee can discharge that duty in a way that leaves a record.
How a range survives contact with a board paper
The failure mode is well known. Three scenarios go into the pack, the discussion settles on the middle one, and by the next meeting the middle one is the plan. The range was produced and then discarded.
Three things prevent it. Do not label a scenario base, because base becomes the plan. Label them by their assumption. Second, lead with the breach point rather than the range: the sentence that survives a board meeting is that the covenant fails if the renewal rate falls below seventy-one per cent, not that earnings fall somewhere between two numbers. Third, state what each scenario would require you to do, because a scenario with no decision attached is information without consequence, and that is what gets forgotten.
Questions about ranges and stress testing
Do we need Monte Carlo simulation?
Rarely, and almost never as the primary tool. Simulation requires distributions and correlations for every input, and in most businesses those are guesses given a statistical appearance. A well chosen set of three scenarios usually communicates more and can be defended in a way a distribution assumption cannot.
How severe should the downside be?
Severe enough to break something. A downside case where every covenant still passes comfortably has not tested anything. The purpose is to find the breach point, and you cannot find it without crossing it.
Who should own the scenarios?
The people who own the assumptions. A scenario built by the finance team alone will be arithmetically clean and commercially unrecognisable to the people running the business, which is precisely why it will be ignored.
Where to go next
Financial Modeling Services
Back to the main page: what we build, how an engagement runs, and how to reach us.
Choosing the Right Financial Model for the Decision You Face
Which model answers which question, what makes a model decision grade rather than merely arithmetically correct, and the three questions a reviewer asks before reading a single number.
Three Statement Models, Forecasting and Budgeting
The operating core: how the three statements tie, how to choose drivers that a business actually manages, what the accounting basis does to a forecast, and the integrity checks that catch most errors in under ten minutes.
Fundraising Models: Investor Financials, Cap Table and Dilution
The model as an investor reads it: burn and runway, the cap table, convertible instruments under Indian exchange control, employee option pools and the tax that attaches to them.
Project Finance and Infrastructure Models
A different model shape: bankability rather than profitability, the Reserve Bank's project finance regime as it stands since October 2025, debt sizing and cover ratios, and where viability gap funding fits.
Send an enquiry
If you already have a model, a scenario and covenant layer can usually be added to it without a rebuild. Send it over and we will tell you which.
Position as at 18 September 2026. Reviewed every six months.
This page is general information, not professional advice. A financial model is only as good as the rules it assumes will still be there when the forecast period arrives, and most of those rules moved recently. The Income-tax Act 2025 replaced the 1961 Act on 1 April 2026 and renumbered every section. The Reserve Bank's project finance regime was rewritten with effect from 1 October 2025. The external commercial borrowing framework was rewritten in February 2026. The treatment of a share buyback changed twice inside eighteen months. Take professional advice before acting on anything on this page. We are happy to be that adviser, but we do not act on a web page, ours or anyone else's, without one.