Research on markets, economics, and questions worth testing properly.
CMB Research publishes empirical work by Chinmay Belgaonkar. Current research covers Indian equities, quantitative finance, prediction markets, labour economics, and empirical research methods.
Each project is shown at the stage it has actually reached.
Design-stage work is labeled as design-stage work. Historical holdout validation is not presented as future-time proof. A failed test stays visible instead of disappearing from the site.
Persistent Relative Weakness and Rebound Dynamics in Indian Equities
The project studies whether stocks that remain weak versus both the market and their sector contain any prospectively observable information about what happens next.
Price Provenance and Belief Updating in Prediction Markets
A completed 36-call pilot tests whether an LLM discounts the same quote when its source is manipulated. A harder 60-call reliability gradient is frozen for the next run.
Young-Graduate Employment Across Indian States
The PLFS project separates two questions that are often mixed together: whether young graduates find work, and what kind of employment they get once they do.
Three rules guide the work on this site.
The purpose is not to maximize the number of positive findings. The purpose is to make weak findings harder to keep.
Keep discovery and validation separate
Patterns found through exploration are labeled as exploratory. When possible, the candidate is frozen before a new holdout is inspected.
Keep failed ideas in the record
Null results, data problems, failed holdouts, and methodological errors remain visible because they are part of the research process.
Say what the evidence does not show
A sector-relative relationship is not the same thing as a stock-price decline. A historical holdout is not the same thing as a future prospective test.
A promising shorting result turned out to depend on hindsight.
An early rebound test inside prolonged weakness looked strong enough to resemble a potentially useful shorting strategy. The result was later rejected because the rebound peak had been selected using future information.
Short the rebound after extreme weakness
The original result was removed from the candidate set. A fixed Day-20 redesign produced a much smaller sector-relative relationship.
- Initial result: retrospective rebound peaks appeared to be followed by sharp declines.
- Problem: the rebound peak was selected from the next 20 sessions, so the historical anchor already contained future information.
- Audit result: the apparent decline disappeared when the selected maximum occurred at the end of the search window.
- Redesign: fixed Day 5, Day 10, and Day 20 landmarks used only information available at the time.
The polished paper is only part of the record.
The research log captures shorter notes on hypotheses, surprises, failed tests, and changes in interpretation while the work is still developing.
Chinmay Belgaonkar
Chinmay studied Computer Information Systems and data analytics at Georgia State University and later worked in derivatives-clearing operations at Morgan Stanley. Current interests span quantitative finance, economics, prediction markets, and applied data analysis.
A recurring theme across the projects is turning a vague question into something measurable, then finding out whether the result survives a harder test.