Evidence-Based Worldview
Replace intuition-led decisions with data, models, and disciplined execution
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 96%
The Evidence-Based Worldview begins by translating a decision into measurable outcomes, risks, and hypotheses. Relevant data is then analyzed through quantitative methods that can process more information consistently than unaided judgment. In investing, Montgomery applies this by purchasing groups of stocks with desired characteristics rather than predicting individual winners. The method extends beyond security selection to portfolio construction, transaction costs, business strategy, and operational efficiency. Models do not eliminate human judgment because people choose the assumptions, data, and algorithms. A sound process therefore tests results across conditions, examines implementation costs, and explicitly checks for behavioral bias before acting. The output is not certainty but a more disciplined, repeatable decision with transparent evidence behind it.
Origin
John Montgomery developed this worldview from engineering, statistics, quantitative modeling, and investment research before founding Bridgeway Capital Management. Extracted from Masters in Business.
Core principles
- 01Start with evidence rather than labels or stories
- 02Use models to process more information consistently
- 03Buy diversified groups rather than relying on single predictions
- 04Evaluate implementation costs alongside theoretical returns
- 05Treat models as human-built tools that can contain bias
How to run it
- 1
Define the decision
State the choice, desired outcome, constraints, and relevant risks in measurable terms. Avoid beginning with a fashionable label or preferred answer.
Pro tip Write down what evidence would disprove your initial view.
Watch out A vague objective allows almost any result to look successful.
- 2
Collect representative evidence
Gather data spanning different market environments or operating conditions. Include adverse periods rather than studying only recent successes.
Pro tip Use historical extremes to understand the nature of risk.
Watch out A narrow sample can hide fragility.
- 3
Form a testable mechanism
Explain why the inputs should produce the expected output. Distinguish compensation for risk from returns caused by persistent human behavior.
Pro tip Investigate anomalies that contradict your preferred explanation.
Watch out Correlation without a plausible mechanism may not persist.
- 4
Model groups and systems
Use quantitative methods to evaluate diversified groups and system-level effects rather than relying on isolated predictions. Design the portfolio or process around the target characteristics.
Pro tip Diversification makes the tested effect less dependent on one case.
Watch out A precise model can still encode poor assumptions.
- 5
Price the execution
Include trading, taxes, implementation effort, and other costs when evaluating the model. A theoretical advantage is useful only if it survives execution.
Pro tip Compare the model's gross and net expected results.
Watch out Ignoring costs can turn an apparent edge into a loss.
- 6
Audit human influence
Identify who built the model, what they expected to find, and which subjective choices shaped it. Invite challenges from people with different views.
Pro tip Begin reviews by confessing any bias toward the result.
Watch out Quantitative output is not automatically objective.
- 7
Implement and learn
Apply the decision consistently, monitor the intended measures, and update the process when evidence changes. Preserve discipline during emotionally difficult periods.
Pro tip Separate a broken process from a sound process temporarily out of favor.
Watch out Abandoning a model solely after underperformance can lock in the behavior gap.
In the wild
Bridgeway studies characteristics such as value, size, momentum, and volatility across groups of stocks. The firm constructs portfolios around tested characteristics rather than purchasing a stock because someone believes it will outperform.
→ Security selection becomes diversified, testable, and less dependent on individual forecasts.
The team examined how market-cap weighting adds more money as a stock rises and can behave like momentum without a defined exit. It researched alternatives and adopted a roughly equal-weighted structure with favorable downside characteristics.
→ The resulting strategy provided a cushion in nine of the S&P 500's ten worst quarters since its inception.
Common mistakes
Trusting a smart label
Marketing terms such as “smart beta” can imply intelligence without explaining the underlying evidence or mechanism.
Treating models as objective
Algorithms are built by people and can preserve their assumptions, incentives, and biases.
Ignoring implementation
A model that looks attractive before transaction costs, taxes, or execution constraints may fail in practice.
Is it for you?
Best for
It is best for investors and decision-makers who face repeated choices supported by measurable evidence.
Not ideal for
It is not ideal when outcomes cannot be measured, relevant evidence is unavailable, or values must outweigh optimization.
From the transcript
“I always go back to the statistics, the logic, the data, the evidence. Evidence based investing is what we do.”
“We don't purchase individual stocks. We buy groups of stocks.”
“And the models are only as good as the modeling process that you've got”
From the episode
Bridgeway's John Montgomery Gives Away Half of Its Profits
John Montgomery