Model
Combine indicators only when the combination improves on a clear baseline and remains understandable enough to review.
Research
Quantitative research at Cornell, from question to contribution.
CATC is a student-run quantitative research and algorithmic trading club at Cornell University. Members start with a market question, identify the data that can answer it, build a repeatable measurement, and test whether it adds useful information. Strong work becomes something the club can review and keep building on.
The CATC loop
Choose a stepLearn · 01Professional insight
How would a practitioner frame the problem?
The research map
Follow the question, not the jargon.
The later stages matter, but the quality of the data and indicator shapes everything that follows.
01
Question
What market behavior are we trying to understand?
02
Data
Which information can answer it, and can we trust that information?
03
Indicator
Can the data become a repeatable measurement?
04
Model
Does combining measurements improve on a simple baseline?
05
Risk
Where might the result be fragile, concentrated, or misleading?
06
Downstream use
Is the result useful enough to document, share, and build on?
Data foundation
Build the data foundation.
Data work starts with a research question, not a download button. The team compares sources, builds a repeatable path from source to table, and keeps checking the dataset after it launches.
SEC Form 4 data pipeline
No CATC result or performance claimResearch question
Can clusters of insider purchases after a sharp earnings drop reveal a market overreaction?
To test the idea, separate deliberate open-market purchases from transactions that happen for other reasons.
Already availablePrices · earnings dates · returns
Still neededClean, classified Form 4 purchase histories
1 / 4
RelevanceReputationAvailable historyCoverageFreshnessConsistencyFeasibilityCost and permitted useRepeatable ingestionClear schemasStandard names and unitsMissing-value rules
Indicators
Turn data into an indicator, then test whether it adds useful information.
Teams use statistics and machine learning where they help, compare different approaches against simple baselines, test the result on data it hasn’t seen before, and ask whether it actually adds useful information.
“More complicated” and “more useful” are not the same thing.
Illustrative example
Insider Conviction Score
APublic Form 4 filingsStart with official records of insider transactions.
+BOpen-market buysKeep purchases made with an insider’s own money.
+COther transactionsSeparate awards, option exercises, automatic sales, and similar activity.
Not every insider transaction carries the same information.
1 / 4
After an indicator
Go deeper only when the evidence earns it.
Models, risk, and downstream use remain part of the research conversation, but they do not rescue weak data or an unstable indicator.
Risk
Ask where a result is concentrated, fragile, or sensitive to changing conditions before treating it as useful.
Downstream use
Document an accepted result so another team can understand it, challenge it, and build the next piece of research or tooling from it.
For technical visitors: what can become reusable?
Datasets · ingestion code · schema definitions · validation checks · health monitoring · transformations · indicators · diagnostics · model experiments · documentation · review notes