Capstone Project

BT4103 Business Analytics Capstone Project Proposal

Industry Partner: Dymon Asia

Latest Sponsor Direction

• Start with FX, then expand to fixed income (FI) and commodities later. The cross-asset proposal below describes the broader ambition; FX is the initial scope.
• Initial workstreams, as relayed from the discussion with Dymon:
  1. “pull CFTC pipeline query” — begin with the CFTC data pipeline for FX; clarify whether Dymon has an existing query/pipeline to use.
  2. “Survey Positioning data provided by Dymon” — work with the survey positioning data supplied by Dymon; its format, coverage, frequency and intended role still need clarification.
• Open questions: Which FX markets should be included first? How should the survey data be used alongside CFTC positioning (comparison, model input, or validation)?

Part 1: Project Information

Project Name

CFTC Positioning Nowcast and Price-Action Crowding Dashboard

Project Description

• Develop a cross-asset analytics platform that combines weekly CFTC Commitments of Traders data with daily price action to estimate how speculative positioning evolves between official releases.
• Create a standardized data pipeline across FX, rates, equity indices, energy, metals and agricultural futures, mapping relevant participant categories such as Managed Money and Leveraged Funds.
• Engineer features from returns, momentum, volume, open interest, volatility, futures-curve structure and cross-asset macro variables; compare regularized regression and tree-based models using time-series cross-validation.
• Produce interpretable positioning and crowding indicators, including historical percentile/z-score, new-long versus short-covering classification, long-liquidation versus new-short classification, price-positioning divergence and squeeze/reversal-risk scores.
• The platform will be a research and decision-support tool, not an autonomous trading or execution system.

Total Workload

Approximately 16 hours per week per student for 12 weeks (team of 3-5 students).

Project Data Sources

Public CFTC historical Commitments of Traders data; Price/Volume per instrument yfinance api.

Platform and Tools

Python (pandas, NumPy, scikit-learn, statsmodels, XGBoost/LightGBM, SHAP), SQL, Git, Jupyter, and Streamlit or Power BI for dashboard development.

Project Deliverables

• Automated and reusable data ingestion, cleaning and contract-mapping pipeline.
• Validated positioning-nowcast models with walk-forward testing, confidence ranges and feature interpretation.
• Interactive cross-asset dashboard showing reported positions, estimated current positions, crowding scores, divergences and regime classifications.
• Reusable codebase, technical documentation, model evaluation report and final presentation.

Reference Repositories

• CFTC COT Viewer — Dashboard reference for net positioning, open-interest normalization, z-scores and historical percentiles.
• cot_reports — Python data-ingestion reference for downloading historical CFTC reports into pandas DataFrames, including Traders in Financial Futures (TFF) reports for the FX workstream.

Definitions

• CFTC (Commodity Futures Trading Commission): The US regulator that publishes the market-position data used by this project.
• Commitments of Traders (COT) data: A weekly CFTC snapshot of the futures positions held by different categories of traders.
• Futures: Exchange-traded contracts tied to the future price of an asset, such as a currency, stock index, oil, or gold.
• Contract mapping: Matching instrument identifiers and expiring futures contracts to the correct market and CFTC series.
• Cross-asset: Covering multiple types of markets instead of only one.
• FX: Foreign-exchange or currency markets.
• Rates: Interest-rate markets, including government-bond and short-term interest-rate futures.
• Equity indices: Measurements of groups of stocks, such as the S&P 500.
• Price action: How a market’s price moves over time.
• Position / positioning: A trader’s market exposure and whether it benefits from prices rising or falling.
• Speculative positioning: Positions intended mainly to profit from price movements rather than to reduce an existing business risk.
• Nowcast: An estimate of the current state when the official data arrives late. Here, it means estimating positioning between weekly COT releases.
• Managed Money: A CFTC trader category generally covering professional money managers in commodity futures.
• Leveraged Funds: A CFTC category generally covering hedge funds and similar leveraged managers in financial futures.
• Return: The percentage change in an asset’s price over a period.
• Momentum: The strength and direction of recent price movement.
• Volume: The number of contracts traded during a period.
• Open interest: The number of futures contracts that remain open and unsettled.
• Volatility: The size and variability of price movements.
• Futures-curve structure: The relationship between prices of futures on the same asset with different expiry dates.
• Cross-asset macro variables: Wider economic or market indicators—such as interest rates or the US dollar—used to help explain another asset’s positioning.
• Regularized regression: A regression model penalized for unnecessary complexity, helping it generalize beyond its training data.
• Tree-based model: A model that learns decision rules from the data and can capture nonlinear relationships. XGBoost and LightGBM are examples.
• Time-series cross-validation: Testing a model on later periods after training it only on earlier periods, so future information cannot leak into the past.
• Historical percentile: The percentage of past observations below the current value; the 95th percentile is higher than about 95% of past values.
• Z-score: How many standard deviations a value is above or below its historical average.
• New long: Traders opening positions that benefit from rising prices.
• Short covering: Traders buying to close positions that benefited from falling prices.
• Long liquidation: Traders selling to close positions that benefited from rising prices.
• New short: Traders opening positions that benefit from falling prices.
• Price-positioning divergence: Price and positioning moving in conflicting directions.
• Crowding: An unusually large concentration of traders positioned in the same direction.
• Squeeze risk: The risk of a rapid move caused by many losing traders exiting similar positions at once.
• Reversal risk: The risk that the current price trend changes direction.
• Autonomous trading or execution system: Software that independently decides and places trades; this project explicitly is not one.