Probability-Curve Forecasting
Replace a single forecast with weighted scenarios and explicit escape routes
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 99%
Rosenberg reframes forecasting as a probability curve rather than a contest to publish one correct number. Start with a base case, state how much conviction it deserves, and map scenarios B, C, and D that describe how the forecast could be wrong. Rank those alternatives and examine the full distribution, including whether its tails are thin or fat. Then compare the reward if the leading view is right with the cost if it fails. The output is not merely a prediction: it is decision support that helps a portfolio manager tighten the range of plausible outcomes, recognize when probabilities shift, and move to a prepared alternative. A forecast therefore remains useful even when its base case has not changed or ultimately proves wrong.
Origin
Rosenberg says the method emerged after a Gluskin Sheff portfolio manager asked him for his conviction level and scenarios B, C, and D, revealing that investors need probability bands rather than a lone base case.
Core principles
- 01A forecast is a distribution of outcomes, not one point estimate
- 02Conviction can change even when the base case does not
- 03The cost of being wrong matters alongside the reward of being right
- 04A usable forecast includes an insurance policy for failure
How to run it
- 1
State the base case
Describe the most likely outcome without presenting it as certain. Treat it as one point within a wider distribution.
Pro tip Use a concrete outcome that decision-makers can connect to portfolio choices.
Watch out Do not present the base case as ironclad.
- 2
Assign conviction
Attach a probability or conviction level to the base case. Reassess that level whenever new information arrives.
Pro tip Report conviction changes even when the base-case forecast itself is unchanged.
- 3
Build alternative scenarios
Specify scenarios B, C, and D and explain where the base case could be wrong. Include both hotter and colder outcomes rather than one generic miss.
Pro tip Identify which alternative has moved up or down in the ranking.
Watch out Do not hide uncertainty behind a single point estimate.
- 4
Map the distribution
Consider the whole range of outcomes and whether the probability curve has fat or thin tails. Tighten the probability bands as the evidence improves.
Watch out A low-probability outcome can still matter when its portfolio cost is large.
- 5
Compare payoff and loss
Evaluate the reward from being right against the cost of being wrong across the possible outcomes. Use that comparison to support the actual decision.
Pro tip Focus on expected values rather than forecast prestige.
- 6
Prepare the escape route
Define the insurance policy or scenario the decision-maker can move into if the base case fails. Admit the error and switch with discipline.
Pro tip Make the fallback actionable before the forecast is tested.
Watch out Do not marry the forecast.
In the wild
Rosenberg describes entering an investment meeting with the same base-case forecast but telling the portfolio managers that his conviction had fallen and that scenarios B, C, and D had changed rank. The room treated those probability changes as important information even though the headline forecast was unchanged.
→ Portfolio managers received a more useful view of changing risk than a repeated point forecast would have provided.
Rosenberg contrasts a rigid fourth-quarter GDP forecast with asking whether a miss would be lower or higher than expected. The alternative framing attaches confidence to the central estimate and specifies the direction of plausible errors.
→ The forecast becomes actionable for an investor instead of merely being right or wrong.
Common mistakes
Publishing one ironclad number
A point estimate conceals confidence and the direction of likely errors, making it less useful to someone managing money.
Ignoring changing conviction
The probabilities can change materially even when the base case remains the same.
Having no plan B
Forecasts will be wrong, so failing to define an escape route leaves the portfolio manager without an insurance policy.
Is it for you?
Best for
It is best for investors and other decision-makers who must act under uncertainty and absorb the consequences of error.
Not ideal for
It is not ideal for questions with a deterministic answer or no meaningful alternative outcomes.
From the transcript
“Your whole world is one job, probability curve.”
“The economist role is not to focus just on the base case. It's to focus on the whole range of outcomes.”
“You cannot marry your forecast, Mary, your partner, don't marry your forecast.”
From the episode
Interview With David Rosenberg: Masters in Business (Audio)
David Rosenberg