Questions
What is synthetic data for investment research?
TL;DR
- Synthetic data means data generated by a model rather than collected directly. In investment research it covers three different things: simulated market data, generated text and simulated survey respondents.
- Neon Blue builds simulated audiences: AI models that answer the way a defined group of people would, grounded in Neon Blue's own paid studies and AI-moderated interviews with real people.
- Simulated respondents answer consumer questions the way a defined group of people would. They are useful only if they're built from real people's data and labeled as simulated.
- Treat a simulated read as a hypothesis to weigh, not evidence on its own, and keep it separate from real responses.
What kinds of synthetic data do investors use?
| Type | What it is | Typical use |
|---|---|---|
| Simulated market data | Generated price or return series | Stress tests and strategy research |
| Generated text | Model-written summaries, transcripts or reports | Research productivity |
| Simulated respondents | Models that answer survey and interview questions as a defined group would | Consumer and buyer questions behind model lines |
This page is about the third type. It is the one that answers questions like "will these customers pay the new price?"
Why are investors skeptical of simulated respondents?
With reason. The common objections are fair:
- "It's made up." A model with no grounding in real people produces plausible answers with no anchor.
- "An LLM gives consensus." A generic model reflects what is commonly written, which is often the Street's view.
- "Bias compounds." A flaw in how the model reasons doesn't average out across respondents.
- "I can't trace it." If an answer can't be traced to real people, it is a hypothesis, not evidence.
What separates a usable simulated read from a made-up one?
- Grounding. Simulated respondents are built from real people's interview and survey data, not generic LLM personas.
- Labeling. Every simulated result is marked as simulated and kept separate from real responses.
- A check you can run. On your own names, compare a simulated read with real responses for the questions you would actually use.
- Stated limits. The vendor says which questions it isn't suited to.
How do funds use simulated respondents?
As one labeled input to a consumer variable in a fundamental model: who will pay, switch, churn or adopt, and at what price. Not as a stock call. Neon Blue doesn't predict stocks or give price targets.
Hypothetical: a consumer analyst covers a meal kit company that is adding a cheaper plan. The model needs a guess at how many current subscribers move down.
- Question: would you keep your plan, move to the cheaper plan or cancel?
- Audience: current subscribers by tenure and order frequency.
- Read: lower-frequency subscribers lean toward the cheaper plan, while frequent ones mostly stay.
- Model: the analyst adds a trade-down line for lower-frequency subscribers, labeled as simulated, and scores it when the company reports.
Want this run on your own question? Neon Blue's first study is free: start a free first study.
Where doesn't it fit?
- Fast-moving, news-driven variables. A read reflects how people say they'll act around a known change.
- Lines driven by supply, costs or management decisions.
- Systematic strategies that need years of point-in-time history.
- Anywhere you'd use it as the only input to a large decision.
Questions about synthetic data in investment research
Is synthetic data the same as alternative data?
No. Alternative data is any non-traditional source, mostly observed, like card spend or web traffic. Synthetic data is generated by a model. Simulated respondents are one kind.
Is synthetic data a substitute for surveys?
No. It is built from real survey and interview data and works alongside it.
How should a fund evaluate a simulated source?
Ask what it is built from, whether results are labeled, how it is tested against real responses and what it can't answer. Then track its reads against reported results on your own names.
How is Neon Blue's grounding built?
From Neon Blue's own paid studies and AI-moderated interviews with real people.
How do I start a free first study?
Bring one consumer question behind a line in your model. Neon Blue's first study is free. Start a free first study