Forecast Lab guide
How to build bear / base / bull forecasts with AI (Forecast Lab method)
A single price target hides the assumption that drives it. Three scenarios make the bet honest by showing what must happen in a bear, base, and bull path—and what evidence would prove each path wrong. This method puts assumptions first and targets second.
Free includes 3 AI research analyses per week, with no card required. Forecast Lab is included with Elite Investor.
Why scenarios beat one number
A precise target can make uncertainty disappear on the screen without reducing it in the business. Change next year’s margin by a few points, use a different diluted share count, or move the valuation multiple, and the answer may change substantially. One number hides that sensitivity and invites you to debate the output instead of the assumptions.
Bear, base, and bull cases force the important questions into view. What must be true about sales growth? Which costs scale, and which remain fixed? Does the company need a richer multiple, or can operating results support the case? The range is not there to guarantee that reality lands between two prices. It is there to expose the business paths behind the math.
Readiness checks add a necessary refusal step. A model should not display an impossible target when its required inputs are missing or when the target does not follow from the stated growth and margin path. A blank or unavailable scenario is more honest than a polished number built from incompatible periods.
The Forecast Lab method
Work in this order. Each step makes the next one auditable, so the final scenarios can be challenged without rebuilding the whole model from memory.
- 1
Lock the as-of fundamentals
Start with sales, earnings, and diluted share count for a specific period. Record the source and as-of date beside every input. If the latest filing and a data provider differ, stop and reconcile the period, accounting basis, currency, split adjustment, and whether the figure is trailing or forward-looking. A scenario built on mixed dates is not conservative or aggressive; it is simply incoherent.
- Ask the AI…
- “List the latest sales, normalized earnings, and diluted share count. For each figure, show the period, source, as-of date, and whether it is reported, adjusted, trailing, or estimated.”
- You still own…
- Open the filing or named provider and verify the three starting figures. Keep unresolved differences visible instead of selecting the number that produces the target you prefer.
- 2
Write three assumption sets
Define bear, base, and bull through explicit growth, margin, and valuation assumptions. The bear case is not ‘bad vibes,’ and the bull case is not optimism with a larger multiple. Each case needs a business path: what happens to demand, pricing, costs, competition, and capital intensity? Change only assumptions you can explain, then state why each differs from the others.
- Ask the AI…
- “Create three internally consistent assumption sets for 12 months. Separate sales growth, operating or net margin, and the valuation multiple or exit logic. Explain the evidence behind every difference.”
- You still own…
- Decide whether the ranges are plausible for this business. Compare them with its own history and industry economics; do not let the model widen a range merely to make the output look balanced.
- 3
Normalize earnings and sales
Scenarios must use comparable foundations. Separate recurring operations from restructuring charges, unusual tax effects, acquisitions, disposals, and other one-time items. Keep reported and adjusted figures labeled rather than blending them. Normalization is not permission to erase every cost management dislikes: a charge that returns each year may be part of the business.
- Ask the AI…
- “Reconcile reported sales and earnings to normalized figures. List every adjustment, its source, whether it is genuinely non-recurring, and the effect it has on each scenario.”
- You still own…
- Review the reconciliation line by line. Restore recurring costs, keep share-based compensation visible, and make sure every case starts from the same normalized base.
- 4
Run readiness checks
Do not trust a target just because the formula produced one. Check that price, shares, financial periods, and valuation inputs are present and sensible. Then test whether projected earnings or sales actually follow from the stated growth and margin path. If the math requires negative shares, an unexplained margin jump, or a multiple applied to the wrong period, the scenario is not ready.
- Ask the AI…
- “Audit each case for missing inputs, period mismatches, impossible values, inconsistent per-share math, and a target that does not follow from the stated growth, margin, and valuation assumptions.”
- You still own…
- Treat a failed readiness check as a stop sign. Correct the inputs or leave the scenario unavailable; never patch the target manually to make it look reasonable.
- 5
Write what would falsify each case
A scenario becomes useful when future evidence can prove it wrong. Give every case one observable falsifier tied to the business: a revenue-growth threshold, margin range, customer-retention measure, cash-conversion result, or management guidance change. Avoid vague statements such as ‘the story weakens’ because they can be reinterpreted after the fact.
- Ask the AI…
- “For each scenario, name one measurable result within the next reported period that would make its assumptions no longer credible. State the source where that result can be checked.”
- You still own…
- Choose thresholds before the next result arrives. Record them in your research notes so a price move or persuasive management explanation cannot rewrite the test later.
- 6
Decide only after you can restate the cases
Finish by explaining the three cases without looking at the target prices. If you can describe what drives sales, margins, and valuation—and what would invalidate each path—you understand the model. If you remember only the highest and lowest target, the model is driving you. Scenarios organize uncertainty; they do not issue a recommendation or predict which case will occur.
- Ask the AI…
- “Summarize each case in two plain-language sentences without using its target price. Then list the assumption with the greatest effect on the range.”
- You still own…
- Keep the decision separate from the output. Consider your own objectives, time horizon, concentration, and risk tolerance, and seek qualified advice where appropriate.
Printable checklist
Assumptions before targets
Print or save this list with each model. If one item is missing, the scenario set is unfinished no matter how convincing its targets look.
- Source and as-of date appear beside every input figure.
- Reported and adjusted figures are labeled separately.
- Share count matches the diluted per-share math.
- Bear, base, and bull differ because named assumptions differ.
- Every case has one specific, observable falsifier.
- No scenario implies a trade; the output remains research only.
Mini example (hypothetical — not advice)
Imagine Northstar Systems, a fictional software company with 100 million hypothetical currency units of normalized annual sales. Its diluted share count is fixed at 50 million for this exercise. These figures do not describe a real company or security.
| Case | Sales growth | Net margin | Exit logic | Falsifier |
|---|---|---|---|---|
| Bear | 2% | 8% | Lower multiple as growth slows | Sales growth holds above 8% |
| Base | 8% | 10% | Multiple remains near its normal range | Margin falls below 8% |
| Bull | 14% | 13% | Higher earnings support a modest re-rating | Growth remains below 10% |
Now suppose an AI returns a bull target that requires a 20% margin while the written bull case says 13%. Readiness should reject it: the target and assumptions describe different companies. The fix is not to keep the attractive target and hide the margin. Recalculate from the stated case or mark the scenario unavailable until the inconsistency is resolved.
How Ticker-Terminal fits
Forecast Lab builds bear, base, and bull 12-month research scenarios from normalized earnings and sales models. Readiness checks prevent users from seeing an impossible target when required inputs or the underlying math do not support it. The result is a set of research scenarios—not a prediction, recommendation, or promise of performance.
Start on the free plan with 3 AI research analyses per week and no card. Elite Investor includes Forecast Lab plus portfolio and earnings AI; see pricing for the current plan details. Pair the workflow with an Academy daily lesson or Chart of the Day to practice turning fresh evidence into better assumptions rather than chasing a new target.
Related reading
How to research a stock with AI
Place scenarios inside a complete seven-step research workflow.
AI investing without the hype
Learn where AI helps, where it misleads, and how to verify its answers.
Read an earnings report with AI
Use six layers and five sanity checks before updating assumptions.
Ticker-Terminal Learn
Browse free lessons on psychology, market cycles, and AI research.
Which AI is best for stock research?
Compare research tools by data, transparency, and repeatability.
Start with research. Add scenario depth when ready.
Use 3 AI research analyses per week on the free plan with no card. When you want the full bear, base, and bull workflow, Elite Investor includes Forecast Lab. Keep every target tied to a source, an assumption, and a falsifier.
Free guide series
Get the “how to research a stock” checklist
Two emails. The 7-step checklist now, then the next guide in the series — starting with AI stock analysis: earnings breakdown.
- The 7-step “how to research a stock” checklist, sent immediately
- The next recommended guide in the series, two days later
- Occasional new guides only — no daily market noise, unsubscribe in one click