Analysis of Competing Hypotheses in Investment Research

The Satisficing Trap

The natural approach to analysis follows a predictable pattern: the analyst identifies what appears to be the most likely explanation, evidence is gathered and organised according to whether it supports this initial view, the hypothesis is accepted if it provides "reasonable fit," and a brief review of alternatives confirms nothing obvious was missed.

This approach—selecting the first hypothesis that seems adequate rather than systematically evaluating all possibilities—is called satisficing. It dominates analytical practice because it is cognitively efficient. The mind seeks closure. Maintaining uncertainty is uncomfortable. The satisficing approach provides an answer quickly and moves on.

The problem is that satisficing produces systematically inferior judgments.

Three Structural Weaknesses

Weakness 1: Selective Perception. The initial hypothesis functions as a perceptual filter. Once an analyst commits to an explanation, they see evidence through that lens. Information consistent with the hypothesis registers clearly. Information inconsistent with it is rationalised, dismissed, or simply not noticed. This filtering is not conscious. The analyst experiences themselves as objectively evaluating evidence. But the hypothesis has already shaped what evidence seems relevant and how it is interpreted.

Weakness 2: Incomplete Hypothesis Generation. Research consistently shows that people are poor at generating complete sets of possibilities. When confronted with complex situations, analysts typically consider only a fraction of the reasonable alternatives. If the correct explanation is not among the hypotheses being considered, it cannot be found. The quality of hypothesis generation sets the ceiling on analytical quality.

Weakness 3: Failure to Assess Diagnosticity. Most evidence is consistent with multiple hypotheses. A strong management team is consistent with future outperformance—and with hubris preceding value destruction. Revenue growth is consistent with market share gains—and with unsustainable price competition. Evidence has diagnostic value only when it helps discriminate between alternatives. Without a complete set of alternative hypotheses, the analyst cannot assess whether evidence actually discriminates.

The Confirmation Trap

The deepest problem is psychological: analysts naturally seek evidence that confirms their hypotheses rather than evidence that would disprove them.

Consider: how often do investors actively seek the bear case on their holdings? How often do they read critical analysis with the same attention they give supportive research?

The natural tendency is to notice confirming evidence, weight it heavily, and explain away contradictions: "That's temporary." "The critic doesn't understand the business." "It's already priced in." "This time is different."

When information is processed this way, almost any hypothesis can be "confirmed." The analyst accumulates supporting evidence while rationalising contradictions—and becomes increasingly confident in a view that may be fundamentally mistaken.

The Logic of Disconfirmation

The correct approach inverts the natural tendency.

A hypothesis cannot be proved by accumulating consistent evidence—because the same evidence may be consistent with other hypotheses. But a hypothesis can be disproved by evidence incompatible with it.

This is counterintuitive. It requires conscious discipline. It imposes cognitive strain. But it is the only approach that reliably distinguishes between hypotheses that happen to fit available evidence and hypotheses that are actually correct.

The Core Insight

Analysis of Competing Hypotheses addresses the structural weaknesses of intuitive analysis through a single fundamental shift: instead of evaluating hypotheses one at a time, all reasonable hypotheses compete simultaneously against each other.

This shift transforms analytical practice. Instead of asking "does the evidence support my hypothesis?", the analyst asks "does this evidence help me distinguish between hypotheses?" Instead of accumulating confirming evidence, the analyst identifies which evidence actually discriminates. Instead of seeking to confirm the preferred view, the analyst seeks to disprove all views and accepts whichever survives.

The methodology forces what intuition resists: genuine consideration of alternatives, honest assessment of whether evidence discriminates, and systematic search for disconfirmation.

The Eight-Step Process

Step 1: Generate All Reasonable Hypotheses. Begin by identifying all possibilities that merit consideration. Use diverse perspectives. Elicit every reasonable alternative before evaluating any. Critical distinction: differentiate between unproven hypotheses (no evidence they're correct) and disproved hypotheses (positive evidence they're wrong). Keep unproven hypotheses alive until they can actually be disproved.

Step 2: List All Significant Evidence. Assemble everything that bears on the question—not just concrete data but assumptions, logical deductions, and notably, absence of evidence. For each hypothesis, ask: if this were true, what should I expect to see? What should I expect not to see? If expected evidence is absent, why?

Step 3: Construct and Analyse the Matrix. Create a matrix with hypotheses across the top and evidence down the side. For each item of evidence, assess: is it consistent, inconsistent, or irrelevant to each hypothesis? The critical procedure: work across rows, not down columns. Take one piece of evidence at a time and assess its relationship to all hypotheses. This forces the diagnosticity question.

Step 4: Refine the Matrix. Reconsider hypothesis wording. Combine hypotheses if no evidence distinguishes between them. Remove non-diagnostic evidence from the working analysis. Add factors that are influencing judgment but weren't initially listed.

Step 5: Draw Tentative Conclusions. Now work down columns—evaluate each hypothesis against all evidence. Proceed by seeking to disprove rather than confirm. The hypothesis with the least evidence against it is probably most likely.

Key insight: pluses (consistent evidence) are far less significant than minuses (inconsistent evidence). Any reasonable hypothesis can accumulate consistent evidence. What matters is evidence that rules hypotheses out.

Step 6: Sensitivity Analysis. Identify the "linchpin" evidence—the few items that actually drive the conclusion. Stress-test these. Are there questionable assumptions? Alternative interpretations? Could the evidence be incomplete or misleading?

Step 7: Report Conclusions with Alternatives. Present the relative likelihood of all hypotheses, not just the most likely one. Any analytical conclusion is incomplete without discussion of alternatives that were considered and why they were rejected.

Step 8: Specify Revision Triggers. Define in advance what would indicate events are taking a different course. Pre-committing to revision triggers makes it harder to rationalise away contradictory developments later.

Systematic Hypothesis Generation (Step 1)

The platform generates hypothesis sets through multiple analytical strategies applied simultaneously:

Situational Analysis: Based on the specific evidence—filings, transcripts, competitive position, management history—what explanations does the current situation suggest?

Theoretical Application: Based on patterns across many similar situations—base rates, historical outcomes, research findings—what hypotheses do comparable cases suggest?

Historical Comparison: Based on analogous situations—this company's history, peer experiences, adjacent sector parallels—what precedents might illuminate the current situation?

Adversarial Generation: What is the strongest possible counter-thesis? Not a strawman but a genuine attempt to construct an alternative explanation that fits the evidence.

The analyst receives a complete set of possibilities, not generated ad hoc but through systematic application of multiple analytical frameworks.

Evidence Assembly and Absence Detection (Step 2)

The platform assembles evidence systematically across domains:

Absence detection: For each hypothesis, the system identifies what evidence should exist if the hypothesis is true. Is that evidence present? If not, why not? The dog that didn't bark—absent evidence that should be present—is surfaced explicitly.

Matrix Construction and Diagnosticity (Step 3)

The platform constructs and maintains the hypothesis-evidence matrix automatically. Each evidence item is assessed against each hypothesis—consistent, inconsistent, or irrelevant. The relationships are stored explicitly and updated as new evidence arrives.

Diagnosticity calculation: For each evidence item, the system assesses: does this help discriminate between hypotheses? Evidence consistent with all hypotheses is flagged as non-diagnostic. Evidence that supports some hypotheses while undermining others is flagged as high-diagnostic.

The analyst sees not just evidence but its analytical significance—something impossible without explicit representation of competing hypotheses.

Systematic Disconfirmation (Step 5)

The platform implements disconfirmation as systematic workflow:

Adversarial search: For each hypothesis, the system searches specifically for evidence that would undermine it. Not evidence that might be relevant—evidence that would disprove the thesis if true.

Counter-argument construction: The system constructs the strongest possible case against each hypothesis. What would a skilled bear say about this bull thesis? What would a skilled bull say about this bear view?

Survival scoring: Rather than ranking hypotheses by supporting evidence (pluses), the system ranks by absence of disconfirming evidence (fewest minuses). The survivor wins—the hypothesis that has withstood the most attempts at disproof.

Linchpin Analysis (Step 6)

The platform identifies and stress-tests linchpin evidence:

Linchpin identification: The system identifies which evidence items have the highest impact on hypothesis rankings. If removing or reinterpreting a single item would change the conclusion, that item is a linchpin.

Assumption surfacing: For each linchpin, the system makes explicit the assumptions required for the evidence to mean what it is taken to mean. Are those assumptions warranted?

Alternative interpretation: The system generates alternative interpretations of linchpin evidence. What else could this evidence mean? Under what circumstances would the standard interpretation be wrong?

Tripwire Monitoring (Step 8)

The platform implements pre-commitment to revision triggers:

Tripwire definition: For each conclusion, the system identifies what evidence would indicate the analysis is wrong. What would have to happen for the rejected hypotheses to become more likely?

Continuous monitoring: The system monitors for tripwire conditions across all evidence sources. When a trigger condition is met, it alerts immediately with context.

Rationalisation resistance: Because tripwires are defined in advance—before contradictory evidence emerges—the analyst cannot easily rationalise them away. The pre-commitment creates accountability.

Workflow Consistency

The critical difference: these processes execute identically every time.

Human analysts apply rigorous method inconsistently—when fresh, when time permits, when stakes seem high enough. The platform applies the same hypothesis generation, the same diagnosticity assessment, the same disconfirmation search, the same linchpin identification to every analysis.

No degradation under time pressure. No skipped steps when deadlines are tight. No variation with fatigue or workload.

What the System Does

• Generate comprehensive hypothesis sets using multiple analytical strategies
• Assemble evidence systematically across domains
• Assess diagnosticity explicitly for every evidence item
• Seek disconfirming evidence for every hypothesis
• Identify linchpin evidence and stress-test assumptions
• Structure output to show alternatives and reasoning
• Monitor for tripwire conditions continuously
• Execute the same process consistently every time

What the System Does Not Do

• Render investment judgment
• Determine which hypothesis is correct
• Decide whether evidence is credible
• Assess whether an analogy applies
• Evaluate management quality
• Determine appropriate action

These remain human judgments. The system implements method. The analyst applies judgment informed by that method.