Gains:
- Ability to establish basic/optimistic/pessimistic scenarios with clear assumptions and produce transparent, manually verifiable spreadsheets with artificial intelligence
- Ability to apply sensitivity and stress testing logic (how much the outcome changes when you change an assumption) and identify critical assumptions
- Be able to maintain that the scenario is not a prediction but a conditional 'if-then' exercise, and that ownership of the assumption and final judgment belongs to the economist.
The future is uncertain, and the economist's job is not to cover up this uncertainty with a precise prediction, but to make it visible in orderly options. Scenario modeling is exactly this: "If the exchange rate goes to this level, if the growth happens like this, if the interest rate goes like this, what will be the result?" A budget projection, a debt sustainability analysis, an investment feasibility — all are thought through scenarios. Artificial intelligence is a powerful accelerator here: it translates assumptions into spreadsheets, generating three to five scenarios in minutes. But the most critical sentence is this: The scenario is not a prediction; It is a conditional "if-then" exercise based on explicit assumptions, and ownership of the assumptions rests with the economist. Artificial intelligence does the calculation; It is up to the human being to decide which assumption is reasonable.
Anatomy of the script
Parts of a solid screenplay:
- Basic (base) scenario. The most likely set of assumptions are; the “expected” pathway.
- Optimistic and pessimistic scenarios. Reasonable upper and lower trails around base. These should be justified, not arbitrary.
- Clear assumptions. The numbers on which each scenario is based (growth, inflation, exchange rate, interest) are written one by one. The implicit assumption is the untenable scenario.
- Transparent account. Formulas from input to output must be visible and manually verifiable. A black box scenario won't work.
- Conclusion and comment. What does the difference between the scenarios show? Which risk is of what magnitude?
Tip: Do not present optimistic/pessimistic scenarios as a "forecast band". These are not probabilities, but conditional consequences: "If the exchange rate increases by 30%, the debt burden will be." The sentence should always begin with "if" so that the reader sees the condition.
Sensitivity analysis: which assumption is decisive?
The outcome of a scenario depends on dozens of assumptions, but not all of them are of equal importance. Sensitivity analysis measures how much the outcome changes when you change one assumption, ceteris paribus. If the outcome is very sensitive to an assumption (small change makes a big difference in outcome), that assumption is critical and deserves the most attention, verification, and communication.
Stress testing is an extension of this: testing how the system/budget/debt behaves under a moderate but harsh negative shock (e.g. sudden jump in interest rates, collapse in the export market). The goal is not to predict the worst, but to foresee breaking points.
Caution: If you tell the AI to "make a pessimistic scenario", it may give you an arbitrary number. The pessimistic scenario is not a "falling from the sky", but a justified assumption: it must have a historical or logical basis, such as "a movement similar to the 2018 exchange rate shock". You justify the assumption.
Step by step scenario study
- Define the question and the outcome. What do we project? (debt/GDP, budget balance, cash flow)
- Set driver variables. A small number of main inputs (growth, inflation, interest, exchange rate) determine the outcome.
- Make base assumptions and justify them. The source and logic of each number is written.
- Produce the scripts. Move assumptions for optimist/pessimist with justification.
- Set up a transparent account. Have the AI print the formulas as a visible table/code.
- Run sensitivity. Shake each key assumption separately and measure the response.
- Comment and filter. Which assumption is critical? Are the results reasonable? The final judgment is the economist's.
three mini cases
Case 1 — Debt sustainability. A treasury analyst projected the public debt/GDP ratio for 5 years. He had the AI create three scenarios: base (growth 4%, interest rate X%), optimistic, pessimistic. Sensitivity analysis showed that the result was most sensitive to the real interest-growth gap (r−g): if growth fell below the interest rate, the debt ratio climbed rapidly. The analyst highlighted this single critical assumption in the report; he focused his energy on validating it.
Case 2 — Correction of the arbitrarily pessimistic scenario. In one institution, AI gave “pessimistic scenario: growth -10%” but it had no basis. The economist objected: this was even harsher than the deepest crisis in the country's history and was unjustified. The pessimistic scenario was downgraded to -3% based on the average severity of the past two recessions. The scenario was now defensible.
Case 3 — Transparent account saved. On a budget projection, the results looked odd. Because the calculation was transparent, the analyst watched the table and found the error: the artificial intelligence had applied inflation to the income side but forgot to apply it to the expense side. In a black box model, this error would remain hidden; visible formulas revealed it.
Scenario design table
item
good practice
bad practice
Assumptions
Clear, reasoned, sourced
secret, arbitrary
Scenario range
Historical/logically based
Random "high/low"
account
Transparent, manually verifiable
black box
presentation
"if-then", conditional
sure as guess
Sensitivity
Critical assumption marked
All assumptions are considered equal
Four copyable templates
1) Scenario skeleton:
Get that output [e.g. debt/GDP] I will project 5 years. Driver variables: growth, inflation, interest, exchange rate. Suggest me a CLEAR assumption table for base/optimistic/pessimistic; Write a rationale next to each assumption. Don't do the math yet; Let me confirm the assumptions first.
2) Transparent spreadsheet/code:
Calculate the projection with the assumptions I approve. Keep the formulas VISIBLE (so it's obvious how each row was calculated). Take one year as an example and show the calculation step by step so I can verify it manually. Don't use a black box.
3) Sensitivity analysis:
In the base scenario, adjust each main assumption by ±X% one by one (others constant) and make a table of how much the result changes. Mark the 2 assumptions that affect the result the most as "critical". Suggest validation for these critical assumptions.
4) Stress test:
Give this reasonable but harsh shock: [e.g. +X point sudden increase in interest rate, Y% decrease in exports]. Show how output changes under this shock, whether it exceeds a critical threshold (e.g. debt/GDP unsustainable limit). Make it clear that this is a conditional stress scenario, not a prediction.
Weak prompt / Strong prompt
Weak prompt:
Write me an optimistic and pessimistic economic scenario.
Artificial intelligence produces unjustified, unsourced, black-box scenarios; You cannot see or verify assumptions.
Powerful prompt:
I will project the public debt/GDP ratio for 2025-2029. Base case: growth 4%, inflation 25%, average interest rate X%, primary balance Y% (I added my sources). Play assumptions with REASON for optimist/pessimist (for pessimist, reference the severity of the last two recessions). Set up the calculation as a transparent table, display a year to be verified manually, then run the r−g sensitivity. State in the text that the scenarios are conditional, not predictions.
Common mistakes
- Hiding assumptions. The unseen assumption is untenable.
- Arbitrarily optimistic/pessimistic. Unjustified extreme numbers destroy trust.
- Presenting the scenario as a prediction. The "if" thought condition disappears.
- Black box account. The error is hidden, verification becomes impossible.
- Sensitivity skipping. Mismanaging risk without knowing the critical assumption.
- The assumption is to cede ownership to artificial intelligence. Judgment belongs to man.
In summary
Scenario modeling is the art of translating uncertainty into conditional conclusions through clear and justified assumptions. Base/optimistic/pessimistic scenarios must be robust, calculations must be transparent and manually verifiable; Sensitivity analysis should reveal the critical assumption. Artificial intelligence speeds up the calculation and options, but the justification, reasonableness and final judgment of the assumptions rest with the economist. The scenario is not a prediction, but an "if-then" exercise.
Application task
Choose a projection for your business (budget, cash flow, debt ratio). Produce a justified assumption table with the 1st template, transparent calculation with the 2nd template, and sensitivity analysis with the 3rd template. Identify the critical assumption that most affects the outcome and verify it from the original source. Check an extreme scenario suggested by the AI for justification and correct it if necessary.
checklist
- [ ] I have written the assumptions of each scenario clearly and with justification.
- [ ] I put the optimist/pessimist on a historical or logical basis.
- [ ] I kept the account transparent and manually verified at least one period.
- [ ] I identified the critical assumption through sensitivity analysis.
- [ ] I confirmed the critical assumption from the original source.
- [ ] I presented the scenarios as conditional “if-then” rather than predictions.
- [ ] I have retained presumptive ownership and final judgment.