Edge-to-Model Research Loop
Turn expert observations into replicable, tested decision models
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 97%
The Edge-to-Model Research Loop begins with conversations rather than datasets. Researchers speak with experienced investors, advisers, pensions, and other practitioners to uncover the distinctive observation or decision process that constitutes each person's edge. They then translate that insight into variables, rules, and a replicable model. Historical tests determine whether the proposed mechanism has predictive value, under which market regimes it works, and whether it improves an existing collection of signals. Ideas that survive testing can enter the research mix; ideas that fail are revised or discarded. The loop combines qualitative discovery with quantitative falsification. It also guards against blindly copying experts: the observation is valuable as a hypothesis, while repeatability and evidence determine whether it becomes a usable framework.
Origin
Extracted from Masters in Business from Savita Subramanian's description of how conversations with professional investors generate ideas for Bank of America's quantitative research.
Core principles
- 01Experienced practitioners possess different forms of edge
- 02A useful observation must be translated into measurable variables
- 03Replication separates transferable insight from personal intuition
- 04Backtesting should decide whether an idea enters the model portfolio
- 05Signals must be rechecked as regimes and incentives change
How to run it
- 1
Sample diverse practitioners
Talk with experienced investors across different roles, mandates, and styles to expose the research team to distinct forms of edge.
Pro tip Ask for concrete decisions and signals rather than broad market opinions.
Watch out A homogeneous interview group will reproduce the same blind spots.
- 2
Isolate the edge
Identify the specific observation, relationship, or decision rule that appears to improve the practitioner's judgment.
Pro tip Separate the underlying mechanism from the person's storytelling or reputation.
Watch out Professional survival does not prove that every stated belief has predictive value.
- 3
Operationalize the insight
Translate the proposed edge into observable inputs, explicit transformations, and a defined output or decision rule.
Pro tip Write the rule so another researcher could implement it independently.
Watch out Do not quietly change the hypothesis after seeing preliminary results.
- 4
Build the replicable model
Implement the rule using consistent data and document how each input is constructed.
Pro tip Preserve point-in-time data where possible to prevent look-ahead bias.
Watch out A model that depends on unavailable historical information cannot be honestly backtested.
- 5
Test across regimes
Evaluate predictive power, robustness, transaction costs, and behavior in different market environments.
Pro tip Compare the signal with simple baselines and existing models.
Watch out A strong aggregate result can conceal failure in the regime where the model will actually be used.
- 6
Decide its role
Retain the model, combine it with complementary signals, revise it, or reject it based on the evidence.
Pro tip Specify monitoring criteria for signal decay or gaming.
Watch out Do not promote a model merely because its original source is influential.
In the wild
A pension investor suggests that low price-to-earnings stocks work only when financing conditions and earnings revisions are supportive. A research team converts that observation into valuation, rate-regime, and revision variables, then tests the combinations across several cycles against a valuation-only baseline.
→ The team learns whether regime conditioning improves the signal and either adds the tested version to its model set or rejects the hypothesis.
Common mistakes
Treating anecdotes as evidence
Practitioner observations generate hypotheses; they do not replace replicable testing.
Building an untestable story
If an edge cannot be converted into defined inputs and rules, its apparent success cannot be independently evaluated.
Ignoring regime dependence
An indicator may work in one environment and fail in another, so aggregate backtests can create false confidence.
Is it for you?
Best for
It is best for research teams combining domain expertise, qualitative interviews, behavioral hypotheses, and quantitative testing.
Not ideal for
It is not ideal when observations cannot be measured, historical data are inadequate, or a backtest cannot account for changing incentives and regimes.
From the transcript
“everyone out there who's been a professional investor for a while has some edge”
“what I like to do is try to recreate. That framework in a model, a replicable model, and then test it to see whether it's…”
“my best ideas to this day have come from talking to our really smart clients out there on the field.”
From the episode
Savita Subramanian on Equity and Quantitative Strategy
Savita Subramanian