Consensus-Edge Forecasting
Find important outcomes that credible evidence supports before consensus sees them
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 97%
Consensus-Edge Forecasting starts by treating forecasting as disciplined judgment rather than a mechanical model. The analyst maps what the consensus expects, studies how people behaved under comparable historical conditions, and reviews leading indicators such as policy, consumer conditions, debt, and market structure. Agreement with consensus is passed over quickly because it offers little differentiated value. Attention shifts to outcomes that satisfy three tests: they matter, evidence suggests they are likely, and they are not yet within the consensus view. The method does not reward automatic contrarianism; disagreement must arise from a coherent mechanism and relevant evidence. The output is a differentiated forecast accompanied by assumptions and invalidation conditions, allowing the analyst to update it as circumstances change.
Origin
Extracted from Masters in Business, where Gary Shilling described the principles behind his economic forecasting practice.
Core principles
- 01Human nature changes slowly, making relevant history useful
- 02Consensus agreement does not create differentiated value
- 03Contrarianism alone is not an edge
- 04The best opportunities are important, probable, and not yet widely recognized
- 05Judgment must combine indicators, history, and contextual clues
How to run it
- 1
Map the Consensus
Document the prevailing forecast and the assumptions analysts use to support it. Separate genuine agreement from views repeated without independent examination.
Pro tip Use published forecasts, positioning, and common narratives to triangulate what is already priced or expected.
Watch out Do not assume the consensus is wrong merely because it is popular.
- 2
Select Relevant History
Look for periods driven by similar human incentives, constraints, and reactions. Focus on mechanisms rather than superficial resemblance.
Pro tip Ask which historical episode rhymes with the current situation rather than demanding an exact repeat.
Watch out The wrong historical analogue can create false confidence.
- 3
Combine the Inputs
Review leading indicators, policy choices, consumer conditions, financial pressures, and qualitative evidence. Use judgment to determine which signals matter most.
Pro tip Explicitly distinguish robust signals from noisy or fashionable data releases.
Watch out Large models can conceal repeated manual adjustments and nonsensical outputs.
- 4
Search Outside Consensus
Identify outcomes that are important and credibly likely but not yet included in conventional expectations. Explain the causal chain behind each candidate.
Pro tip Prioritize overlooked mechanisms rather than provocative conclusions.
Watch out Novelty without probability is speculation, not forecasting.
- 5
Commit and Monitor
State the forecast, its time horizon, and the evidence that would disprove it. Update the view when the mechanism changes, not merely when prices fluctuate.
Pro tip Keep a written record so forecast quality can be judged honestly over time.
Watch out Markets can remain disconnected from fundamentals long enough to punish an undisciplined position.
In the wild
Shilling concluded that resistance to government excess would unwind inflation while investors remained positioned for continued inflation. He argued that stocks and bonds, then neglected by consensus, would benefit as inflation and yields declined.
→ The forecast became his defining long-term call as Treasury yields fell and bond prices rose.
An analyst sees widespread forecasts of permanent stagnation. Instead of opposing them automatically, she studies the adoption lags of earlier general-purpose technologies, examines current investment and productivity channels, and defines measurable conditions under which emerging technology could lift growth.
→ She develops a differentiated forecast with explicit evidence and invalidation criteria.
Common mistakes
Contrarianism for Its Own Sake
Rejecting consensus automatically replaces analysis with identity. A differentiated view must still be important, probable, and mechanistically supported.
Choosing Superficial Analogies
Periods that look similar can have different causal structures. Historical comparisons should match incentives and constraints, not just visible outcomes.
Treating Models as Automatic Answers
A model repeatedly repaired with discretionary inputs is not independent of human judgment. Its output should be tested for economic coherence.
Is it for you?
Best for
It is best for economists, investors, strategists, and researchers making decisions under uncertainty.
Not ideal for
It is not ideal for situations where reliable base data is unavailable or being different matters more than being correct.
From the transcript
“One is that is that human nature changes very slowly over time. So people react to summer circumstances in Simmler ways and others. History is…”
“And the second one is that you don't add value by rehashing the cansensus.”
“But where we find something that's important is h is likely to happen because you're judged by your forecasting record, Elemanly. And third is not…”
From the episode
An Interview With Gary Shilling: Masters in Business (Audio)
Gary Shilling