Financial data literacy
Reproduce timing, returns, risk, and baselines in guided notebooks and explain them in two minutes.
한국어 안내
16-week high-school research studio
AI + FinanceFrom code to evidence.
Test data timing, baselines, costs, errors, and risk before celebrating a financial AI performance curve.
MIT · Harvard · Stanford · Wharton · ChicagoAn independent CIT course that clearly separates official teaching, faculty research code, and university-linked open-source projects.
Research and simulation onlyNo real money, brokerage accounts, personal financial data, or security recommendations. This course does not provide investment, tax, legal, lending, or credit advice.
Core questionCould we have produced this result at that time, after costs, and reproduce it now?
16 weeksweekly research lab
72 hoursguided + independent
4 pathsreadiness-based depth
Financial AI is not
a prediction contest.
A strong result exposes the conditions under which it can fail.
Students read public or synthetic financial data, build simple baselines, then compare ML, optimization, NLP, and agent simulations. Every conclusion carries a timestamp, cost assumption, repeat range, failure case, and human review boundary.
35%data and finance
35%AI, code, simulation
30%evaluation and responsibility
- QuestionSeparate prediction, explanation, optimization, simulation
- TimeRecord when a value was observed and available
- BaselineCompute a simple method on the same period and costs
- AI testMeasure repeats, errors, subconditions, market effects
- DecisionDefend claims, limits, stop rules, and next evidence
Evidence map
We label the relationship,
not just the university.
Official courses, faculty research code, and university-linked open source are not equivalent. Students verify where each resource came from and the conditions under which it can be used.
Official university teaching
- MIT Sloan 15.S52Financial ML, domain knowledge, regulation, fairness
- MIT Deep Learning GitHubOfficial course code and reproducible notebook structure
- Harvard Applied Quant FinanceData, strategies, portfolios, risk
- Wharton Financial AnalyticsFrom data acquisition to ML and inference
Faculty research and companion code
- Stanford CVXPortfolioReturn, risk, trading cost, and holding cost
- CVXPortfolio GitHubCompanion code linked by the Stanford faculty page
- Chicago ML_CodesAuthor code for empirical asset-pricing ML
- Wharton PyWRDSOfficial WRDS Python code with public fallback
University-linked open source
- Columbia-linked FinRLFinancial reinforcement-learning research
- Georgia Tech-linked ABIDESAgent-based synthetic markets
- Johns Hopkins-linked HARKHeterogeneous household simulations
CIT is not an official partner, accreditor, or credit-granting unit of these universities or projects. The links make the curriculum's evidence and code provenance transparent.
Four evidence levels
Start by readiness,
not grade alone.
All students share the same 16-week questions and safety rules. Code depth, primary-source reading, and independence change.
Grades 9-12 are the recommended age range. Students may change paths after a diagnostic or during the term.
Evidence data card, baseline notebook, explanation
Financial ML experiments
Compare time splits, costs, small models, and optimization on one evaluation table.
University code replication
Run a scaled replication and analyze sensitivity to seeds, costs, and subconditions.
Independent AI finance research
Operate a reproducible repository, red team, and external review, then defend the work.
16-week curriculum
Read, reproduce,
and challenge.
Each week includes a 180-minute guided lab and 90 minutes of independent work. Fixed snapshots and synthetic fallbacks preserve the research decision when a tool or restricted service is unavailable.
Questions, data, baselines, leakage
- W01Research questions and financial boundaries
- W02FRED, SEC timing and provenance
- W03Returns, risk, and baselines
- W04Backtest audit and leakage
Research code, optimization, households, fairness
- W05Chicago asset-pricing ML replication
- W06Stanford cost-aware portfolios
- W07HARK household-risk simulation
- W08MIT financial ML fairness
Market agents and financial language models
- W09ABIDES market microstructure
- W10FinRL rewards and interaction
- W11FinGPT NLP and claim tracing
- W12FinRobot evidence-grounded research
Reproduction package and defense
- W13Wharton data access and fallback
- W14Capstone first reproducible run
- W15NIST red team and stop decision
- W16Poster and six-minute defense
Keep the research trail,
not just the return curve.
The final evidence package lets another reviewer find the basis and boundary of every major claim.
- Data evidenceSource, unit, observation date, availability date, snapshot, transformation
- Evaluation evidenceTime split, non-AI baseline, costs, errors, repeat range, sensitivity
- Code evidenceLocked environment and seed, run order, expected output, offline fallback
- Responsibility evidenceModel card, unsupported-claim refusal, red-team results, stop rules
- Communication evidenceResearch poster, six-minute defense, AI and external-code disclosure
Five capstone tracks
Different questions,
one evidence standard.
- Cost-aware virtual portfolio research
- Scaled asset-pricing ML replication
- Multi-agent synthetic-market simulation
- Evidence-grounded public filing and news researcher
- Synthetic household economic-shock policy simulation
Every track uses public or synthetic data. Security recommendations, live trading, and personal financial decisions are outside the capstone boundary.
Non-negotiable boundaries
Study finance without
testing on student money.
A no-trading boundary makes room to investigate failures in models, markets, and claims honestly.
- Zero real money or accountsNo brokerage account, trading API, leverage, live order, or account sharing.
- Zero personal financial dataNo student or family income, credit, account, transaction, or debt data.
- Zero investment instructionsNo security pick, buy or sell timing, guaranteed return, or urgency prompt.
- Zero inflated university claimsUniversity names identify sources, not affiliation, accreditation, credit, or admission benefit.
- Zero hidden AI contributionStudents disclose AI help, external code, and licenses and explain the core work themselves.
Primary sources
If a claim cannot link back,
it is not course evidence.
CIT does not copy university courses. We use official pages to verify questions, tool boundaries, and usage conditions, then write original high-school explanations, synthetic data, labs, and rubrics.
Reviewed 2026-08-27 / repositories, licenses, APIs, and links rechecked before teaching
MIT and Harvard
- MIT Sloan AI in FinanceFinancial ML traps, economic loss, generative AI, fairnessOfficial university syllabus
- MIT OCW AI in FinanceExplainability, privacy, robustness, resilienceOfficial open course
- MIT Introduction to Deep Learning GitHubOfficial course code and notebook reproducibilityOfficial course repository
- Harvard Applied Quantitative Finance and MLData, strategies, portfolios, risk managementOfficial course listing
- Harvard AC215 GitHubVersioning, tests, automation, monitoringOfficial course repository
Stanford, Wharton, Chicago
- Stanford CVXPortfolioMulti-period portfolios with cost constraintsFaculty research and companion software
- CVXPortfolio GitHubPortfolio optimization and back-testing codeCompanion code linked by the Stanford faculty page
- Wharton Financial AnalyticsData acquisition through ML and inferenceOpen faculty course material
- Wharton PyWRDSResearch-data access and public fallbackOfficial university GitHub
- Chicago ML_CodesAuthor code for empirical asset-pricing MLFaculty research code
- Wharton AI and Analytics researchAI trading, financial stability, regulationOfficial research program
Columbia, Georgia Tech, Johns Hopkins
- Columbia CDFT software directoryRelationship evidence for FinRL, FinGPT, and FinRobotOfficial university research-center page
- Georgia Tech ML4T course projectRelationship evidence for student research using ABIDESOfficial university course site
- Johns Hopkins computational economics syllabusRelationship evidence naming HARK as a course toolFaculty syllabus
- FinRLFinancial reinforcement-learning environmentsLinked by Columbia's official software page
- FinGPTFinancial language models and data pipelineLinked by Columbia's official software page
- FinRobotEvidence-grounded financial research agentLinked by Columbia's official software page
- ABIDESClosed agent-based market simulationLinked to Georgia Tech research and teaching
- HARKHeterogeneous household economic modelsLinked to Johns Hopkins faculty research
Public evidence and risk
- FRED APIEconomic series, timing, notes, snapshotsOfficial Federal Reserve data
- SEC EDGAR APIsPublic filings and Company FactsOfficial SEC API documentation
- NIST AI RMFGovern, Map, Measure, ManageOfficial risk-management framework
- Investor.gov scam warningsGuaranteed returns, pressure, FOMO, AI hypeSEC investor-education material
Relationship notice: University and project names identify the sources of public teaching and research materials. This CIT course is not official, affiliated, endorsed, accredited, credit-bearing, or guaranteed by any listed institution.
Questions before
the first lab
Python and statistics experience, the student's finance research interest, and independent-work habits help determine the starting path.
01Do students trade stocks or crypto with real money?
No. There is no real money, brokerage account, trading API, leverage, or live order. Work stays inside public snapshots and synthetic simulations.
02Is this an official university course?
No. It is an independent CIT course that labels official teaching, faculty research code, and university-linked open source separately. It does not imply affiliation, endorsement, accreditation, credit, or admission benefit.
03Can a beginner start?
Yes. Foundation uses guided notebooks and synthetic data for returns, risk, timing, and baselines. Experienced students deepen implementation through Applied, Research, or Studio.
04Is a paid WRDS account required?
No. Students learn the access pattern, then complete the core analysis with a public snapshot or synthetic fallback using the same schema. Account sharing is not allowed.
05Do you use family financial data?
No. CIT does not collect real account, income, credit, transaction, debt, or family financial information. Public, synthetic, and aggregated data are used.
06What becomes portfolio evidence?
A reproducible repository, data card, non-AI baseline, evaluation and errors, model or system card, red-team results, research poster, and six-minute defense.
Before building financial AI,
choose a question worth testing.
We assess current experience and research interests, then place the student from Foundation through Studio.