Open-Source Trading Ecosystem: Landscape Analysis
Commodity ETF Algorithmic Trading Platform -- Evaluated April 2026
Verdict legend: FORK (use as foundation) | VENDOR (use as dependency) | INSPIRE (learn from, build own) | IGNORE (skip)
1. Backtest / Research Frameworks
QuantConnect Lean
| Attribute |
Detail |
| License |
Apache 2.0 |
| Stars |
~9.4k |
| Activity |
Active -- commits Dec 2025, regular releases |
| Language |
C# core, Python/C# strategies |
| Quality |
Enterprise-grade; 200+ contributors; heavy .NET dependency tree |
| Commodity Fit |
Supports futures, options, equities. Cloud-first model, local runner available. No native commodity curve/roll logic. |
| Verdict |
INSPIRE -- Architecture patterns for multi-asset OMS are excellent. C# core is wrong language for a Python/Rust stack. Cloud lock-in risk. Study the brokerage abstraction layer and the universe selection framework, then build natively. |
Zipline-Reloaded
| Attribute |
Detail |
| License |
Apache 2.0 |
| Stars |
~1.6k |
| Activity |
Maintained -- Stefan Jansen keeps it updated for his ML for Trading book; issues active through 2025 |
| Language |
Python |
| Quality |
Clean event-driven architecture but showing age; Quantopian-era design assumptions |
| Commodity Fit |
Equity-centric pipeline. No futures, no roll logic, no margin. Adding commodity support would require heavy modification. |
| Verdict |
IGNORE -- Equity-only DNA. The event-driven concepts are better realized in nautilus_trader. |
Backtrader
| Attribute |
Detail |
| License |
GPL-3.0 |
| Stars |
~16-21k (varies by source) |
| Activity |
Stale -- Core maintainer (mementum) largely inactive since 2021. Community PRs trickle in but are rarely merged. |
| Language |
Python |
| Quality |
Excellent teaching codebase. OOP design is intuitive but single-threaded and slow on large datasets. |
| Commodity Fit |
Has futures data feed support and basic roll logic. Community contributed commodity examples. Performance ceiling is low. |
| Verdict |
IGNORE -- GPL license is problematic. No active maintainer. Performance does not scale to production workloads. Legacy project. |
vectorbt (open-source) / vectorbt-pro
| Attribute |
Detail |
| License |
Apache 2.0 + Commons Clause (OSS); proprietary (Pro) |
| Stars |
~6.8k (OSS) |
| Activity |
Active -- Pro is actively developed by Oleg Polakow; OSS version is feature-frozen |
| Language |
Python (NumPy/Numba) |
| Quality |
Exceptional vectorized design. Pro adds portfolio simulation, cross-validation, parameter optimization at scale. The fastest pure-Python backtester. |
| Commodity Fit |
Asset-agnostic -- works on any OHLCV time series. No built-in futures roll, margin, or curve awareness. You supply the data, it crunches combinations. |
| Verdict |
VENDOR (Pro) -- The parameter sweep / signal research engine is unmatched for alpha discovery. Use as a dependency for research notebooks. Do NOT use as the execution backbone -- it has no OMS. The Commons Clause on OSS prevents selling a product built on it, so the paid Pro license is the correct path. Budget ~$50/mo per seat. |
bt (pmorissette)
| Attribute |
Detail |
| License |
MIT |
| Stars |
~2.8k |
| Activity |
Active -- Multiple releases in 2025-2026 (v1.1.1 through v1.1.5) |
| Language |
Python |
| Quality |
Clean, tree-based strategy composition. Built on ffn for analytics. Lightweight and composable. |
| Commodity Fit |
Portfolio-level backtesting with rebalancing logic. Good for testing ETF allocation strategies across commodity baskets. No execution layer. |
| Verdict |
VENDOR -- Excellent for portfolio-level strategy research (ETF weighting, rebalancing rules). MIT license is clean. Use alongside vectorbt-pro: vbt for signal research, bt for portfolio construction testing. |
nautilus_trader
| Attribute |
Detail |
| License |
LGPL-3.0 |
| Stars |
~17k |
| Activity |
Very Active -- Bi-weekly releases, Apr 2026 latest. Rust core with Python bindings. |
| Language |
Rust + Python |
| Quality |
Production-grade. Deterministic event-driven architecture. Nanosecond resolution. Handles 5M+ rows/sec. Margin accounts, futures contract management, custom fee models. |
| Commodity Fit |
First-class futures support (CME, ICE). FuturesContract instrument type with expiry/multiplier. Margin accounts simulate commodity futures accurately. IB integration for live trading. |
| Verdict |
FORK -- This is the execution engine. LGPL-3.0 allows linking as a library without GPL contamination of your proprietary code. The Rust core gives the performance ceiling needed. Fork to add: (1) commodity-specific roll logic, (2) curve-aware position management, (3) ETF creation/redemption simulation. The architecture already supports multi-venue, multi-strategy, multi-asset natively. |
hummingbot
| Attribute |
Detail |
| License |
Apache 2.0 |
| Stars |
~14k |
| Activity |
Active -- Regular releases, strong community |
| Language |
Python |
| Quality |
Well-structured for market-making on crypto exchanges. Extensive connector ecosystem. |
| Commodity Fit |
Crypto-native. No futures exchange support (CME/ICE). No margin account modeling for traditional commodities. |
| Verdict |
IGNORE -- Wrong asset class. Crypto market-making patterns do not transfer to commodity ETF execution. |
freqtrade
| Attribute |
Detail |
| License |
GPL-3.0 |
| Stars |
~48k |
| Activity |
Very Active -- March 2026 release; massive community |
| Language |
Python |
| Quality |
Excellent for retail crypto trading. Good plugin architecture (FreqAI for ML). |
| Commodity Fit |
Crypto-only. No traditional exchange support. No futures, no margin modeling. |
| Verdict |
IGNORE -- GPL license is toxic for commercial use. Crypto-only. The FreqAI integration pattern for ML strategy plugins is worth studying, but the codebase itself is not applicable. |
FinRL
| Attribute |
Detail |
| License |
MIT |
| Stars |
~14.7k |
| Activity |
Active -- FinRL-X (next-gen) in development; contests run through 2025 |
| Language |
Python |
| Quality |
Research-grade. Good for RL experimentation. Not production-ready -- training loops and environments need significant hardening. |
| Commodity Fit |
Asset-agnostic RL environments. Can model commodity portfolio allocation as an RL problem. No execution layer. |
| Verdict |
INSPIRE -- Study the RL environment design for portfolio allocation. Do not use as a dependency -- the code is research-quality. Build your own RL training environment that plugs into nautilus_trader's backtest engine. |
qlib (Microsoft)
| Attribute |
Detail |
| License |
MIT |
| Stars |
~36k |
| Activity |
Very Active -- Rapid growth (6k+ stars in a single month in 2025); RD-Agent integration for automated ML research |
| Language |
Python |
| Quality |
Institutional-grade ML pipeline. Server-client architecture. Comprehensive data handler, model zoo, workflow orchestration. The RD-Agent is a genuine innovation -- automated feature engineering and model selection. |
| Commodity Fit |
Equity/factor-model centric by default, but the data handler and model pipeline are asset-agnostic. Custom data loaders can ingest commodity data. The alpha factor framework translates well to commodity signals (basis, roll yield, positioning). |
| Verdict |
VENDOR -- Use as the ML research platform. The data pipeline, model training, and experiment tracking are best-in-class for quant research. Feed commodity-specific features into qlib's model zoo. The MIT license is clean. Integrate qlib's signal output into nautilus_trader for execution. |
mlfinlab (Hudson & Thames)
| Attribute |
Detail |
| License |
Proprietary (All Rights Reserved; commercial license required) |
| Stars |
~4.7k |
| Activity |
Maintained behind paywall |
| Language |
Python |
| Quality |
Implements Marcos Lopez de Prado's "Advances in Financial Machine Learning" algorithms. Well-tested. |
| Commodity Fit |
Meta-labeling, triple-barrier method, fractional differentiation -- all applicable to commodity signals. |
| Verdict |
IGNORE -- Proprietary license makes it unsuitable as a dependency. The algorithms from the book are well-documented; implement the specific ones you need (triple-barrier, CUSUM filter, entropy features) directly. Several open-source reimplementations exist. |
2. Execution / OMS
nautilus_trader
See above. FORK -- primary execution engine.
Lean (QuantConnect)
See above. INSPIRE -- wrong language stack (C#).
barter-rs
| Attribute |
Detail |
| License |
MIT |
| Stars |
~2k (estimated) |
| Activity |
Active -- Last updated Nov 2025; PRs through late 2025 |
| Language |
Rust |
| Quality |
Clean Rust crate ecosystem (engine, data, execution, integration). Modular design. Explicitly marked as educational/research, NOT production. |
| Commodity Fit |
Asset-agnostic event-driven architecture in Rust. Could theoretically handle futures. |
| Verdict |
INSPIRE -- The crate separation pattern (instrument, data, execution, integration) is a clean Rust architecture reference. nautilus_trader is more mature and production-ready. Study barter-rs for Rust idioms if extending nautilus_trader's Rust core. |
StockSharp
| Attribute |
Detail |
| License |
Apache 2.0 |
| Stars |
~9.4k |
| Activity |
Active |
| Language |
C# |
| Quality |
Comprehensive connector library (100+ exchanges/brokers). HFT-capable. |
| Commodity Fit |
Supports futures exchanges. Good brokerage abstraction. |
| Verdict |
IGNORE -- C# ecosystem. nautilus_trader covers the same ground in Rust/Python with better performance. |
Roq
| Attribute |
Detail |
| License |
Permissive (API open, gateways commercial) |
| Stars |
Low (niche) |
| Activity |
Active -- Specialized HFT toolkit |
| Language |
C++23 |
| Quality |
Ultra-low latency (single-digit microsecond). Shared memory IPC. Production-grade for market making. |
| Commodity Fit |
CME/ICE gateway support. Purpose-built for futures market making. |
| Verdict |
INSPIRE -- Overkill unless you need sub-10us latency. Study the shared memory architecture and FIX bridge patterns. If you eventually need HFT-grade commodity futures execution, revisit. For ETF-level rebalancing (seconds, not microseconds), nautilus_trader is sufficient. |
OpenAlgo
| Attribute |
Detail |
| License |
AGPL-3.0 (platform); MIT (SDKs) |
| Stars |
~1.5k (estimated) |
| Activity |
Active -- Indian market focus |
| Language |
Python |
| Quality |
Good for Indian retail brokers (Zerodha, Angel One). Clean REST API. |
| Commodity Fit |
Indian commodity exchanges (MCX). Not applicable to CME/ICE commodity ETFs. |
| Verdict |
IGNORE -- Wrong market geography. AGPL license is restrictive. |
3. Data / Feature Stores
ArcticDB (Man Group)
| Attribute |
Detail |
| License |
BSL 1.1 (production use requires commercial license); v1.0 converts to Apache 2.0 after Mar 2025 |
| Stars |
~2.2k |
| Activity |
Active -- Bloomberg co-development; regular releases |
| Language |
C++ core, Python bindings |
| Quality |
Purpose-built for financial time series at Man Group scale. Serverless (S3/LMDB backend). Versioned DataFrames. Handles billions of rows. |
| Commodity Fit |
Ideal for storing tick data, OHLCV bars, futures curves, and feature matrices. The versioning model is excellent for reproducible research. |
| Verdict |
VENDOR (with caveat) -- The BSL license means you need a commercial agreement for production use. For research and development, it is free. Evaluate the cost of a commercial license vs. building on Parquet+DuckDB. If Man Group's pricing is reasonable, ArcticDB is the best-in-class option for financial time series storage. If not, fall back to the Parquet+DuckDB+Polars stack. |
Parquet + DuckDB
| Attribute |
Detail |
| License |
Apache 2.0 (Parquet format) / MIT (DuckDB) |
| Stars |
DuckDB: ~30k+ |
| Activity |
Very Active -- DuckDB v1.5.0 released March 2026 |
| Language |
C++ (DuckDB); Parquet is a format |
| Quality |
DuckDB is exceptional. OLAP queries on Parquet files with zero infrastructure. Embedded, no server. Native Polars integration. |
| Commodity Fit |
Store futures curves, COT data, EIA/USDA timeseries as partitioned Parquet files. Query with SQL or Polars. Zero ops burden. |
| Verdict |
VENDOR -- The default storage layer. Parquet for persistence, DuckDB for ad-hoc analytics, Polars for DataFrame operations. This stack is free, performant, and has no license risk. Use ArcticDB only if you need its versioning/branching features and can afford the license. |
Polars
| Attribute |
Detail |
| License |
MIT |
| Stars |
~35k |
| Activity |
Very Active -- Series A funding (EUR 18M, Sep 2025); Polars Cloud and Distributed Engine in beta |
| Language |
Rust core, Python/Node/R bindings |
| Quality |
10-100x faster than pandas. Lazy evaluation, streaming, out-of-core processing. Apache Arrow memory model. |
| Commodity Fit |
Replace pandas everywhere in the pipeline. Commodity data wrangling (curve construction, roll calendars, feature engineering) runs dramatically faster. |
| Verdict |
VENDOR -- Replace pandas with Polars across the entire codebase. MIT license, massive momentum, commercial backing. This is the DataFrame standard going forward. |
qlib (data layer)
See Backtest section. VENDOR -- qlib's data handler is excellent for ML feature pipelines. Use its data infrastructure for model training; use Parquet+Polars for everything else.
tsfresh
| Attribute |
Detail |
| License |
MIT |
| Stars |
~8.5k (estimated) |
| Activity |
Maintained -- Academic publications ongoing through 2025 |
| Language |
Python |
| Quality |
Systematic feature extraction (hundreds of features). Automated relevance testing. Well-documented. |
| Commodity Fit |
Extract statistical features from commodity price series, volume profiles, spread dynamics. The automated feature selection reduces overfitting risk. |
| Verdict |
VENDOR -- Use for systematic feature engineering on commodity time series. Feed extracted features into qlib's model pipeline. MIT license, lightweight dependency. |
| Attribute |
Detail |
| License |
BSD-3-Clause |
| Stars |
~7k (estimated) |
| Activity |
Maintained -- Alteryx-backed |
| Language |
Python |
| Quality |
Deep feature synthesis for relational data. tsfresh integration available. |
| Commodity Fit |
Less relevant than tsfresh for pure time series. Useful if you have relational data (e.g., COT report entities, EIA facility-level data). |
| Verdict |
IGNORE -- tsfresh is more directly applicable. featuretools solves a relational problem we don't have. |
4. Portfolio / Risk
riskfolio-lib
| Attribute |
Detail |
| License |
BSD-3-Clause |
| Stars |
~3.8k |
| Activity |
Active -- v7.2.1 released Feb 2026 |
| Language |
Python |
| Quality |
Most comprehensive portfolio optimization library available. 30+ risk measures, Black-Litterman, HRP, nested clustering, factor models. Built on CVXPY. |
| Commodity Fit |
Directly applicable to commodity ETF portfolio construction. Supports the risk models and constraints needed for commodity allocation (CVaR, drawdown limits, sector constraints). |
| Verdict |
VENDOR -- Primary portfolio optimization library. BSD license is clean. The breadth of optimization methods exceeds all alternatives. |
PyPortfolioOpt
| Attribute |
Detail |
| License |
MIT |
| Stars |
~4.8k |
| Activity |
Active -- Latest release Feb 2026 |
| Language |
Python |
| Quality |
Clean API. Efficient Frontier, Black-Litterman, HRP. Good documentation. Less comprehensive than riskfolio-lib but simpler API. |
| Commodity Fit |
Good for basic mean-variance and Black-Litterman allocation across commodity ETFs. |
| Verdict |
IGNORE -- riskfolio-lib is strictly superior in breadth. PyPortfolioOpt is simpler but we don't need simpler, we need comprehensive. |
skfolio
| Attribute |
Detail |
| License |
BSD-3-Clause |
| Stars |
~1.9k |
| Activity |
Active -- v0.15.3 latest; academic paper published 2025; enterprise support via Skfolio Labs |
| Language |
Python |
| Quality |
Scikit-learn compatible API. Cross-validation, walk-forward optimization built in. The sklearn integration is genuinely powerful -- GridSearchCV for portfolio parameters. |
| Commodity Fit |
Walk-forward portfolio optimization with cross-validation is critical for commodity strategies that exhibit regime changes. |
| Verdict |
VENDOR -- Use alongside riskfolio-lib. riskfolio-lib for optimization breadth, skfolio for sklearn-compatible backtesting/cross-validation of portfolio models. The two complement each other. |
empyrical-reloaded
| Attribute |
Detail |
| License |
Apache 2.0 |
| Stars |
~300 (estimated) |
| Activity |
Maintained -- Stefan Jansen maintains the reloaded fork |
| Language |
Python |
| Quality |
Canonical financial metrics (Sharpe, Sortino, max drawdown, etc.). Lightweight, well-tested. |
| Commodity Fit |
Asset-agnostic performance metrics. |
| Verdict |
VENDOR -- Lightweight dependency for performance calculations. Use as utility library. |
quantstats
| Attribute |
Detail |
| License |
Apache 2.0 |
| Stars |
~6.9k |
| Activity |
Maintained -- Ran Aroussi's original; quantstats-reloaded fork also exists |
| Language |
Python |
| Quality |
Beautiful HTML tear sheets. Comprehensive metrics. Some dependency drift with yfinance. |
| Commodity Fit |
Asset-agnostic reporting. Generate investor-grade performance reports. |
| Verdict |
VENDOR -- Use for reporting and tear sheet generation. The HTML output is presentation-ready. |
pyfolio-reloaded
| Attribute |
Detail |
| License |
Apache 2.0 |
| Stars |
~300 (estimated) |
| Activity |
Maintained -- Updated through 2025 |
| Language |
Python |
| Quality |
Bayesian tear sheets, sector exposure analysis. More analytical depth than quantstats but less polished output. |
| Commodity Fit |
Sector/factor exposure analysis relevant for commodity baskets. |
| Verdict |
IGNORE -- quantstats covers reporting better. The unique Bayesian features are niche. |
5. UI Inspiration
| Attribute |
Detail |
| License |
AGPL-3.0 |
| Stars |
~35k |
| Activity |
Very Active -- 350+ data endpoints, AI agent integration |
| Quality |
The gold standard for open-source financial terminal UX. Plugin architecture, data standardization layer, CLI + web UI. |
| Commodity Fit |
Has commodity data endpoints. The data standardization pattern (normalize across providers) is directly applicable. |
| Verdict |
INSPIRE -- Do NOT fork (AGPL). Study: (1) the data provider abstraction layer, (2) the charting/visualization patterns, (3) the CLI/TUI design. Build your own commodity-focused terminal UI taking these patterns. |
Reactive Trader Cloud (Adaptive)
| Attribute |
Detail |
| License |
Apache 2.0 |
| Stars |
~1.8k |
| Activity |
Archived -- Repo marked "NO LONGER MAINTAINED" as of 2025 |
| Quality |
Beautiful real-time trading UI in React/RxJS. Excellent reactive programming patterns. |
| Commodity Fit |
FX-focused but the real-time blotter, P&L, and position management UI patterns are universal. |
| Verdict |
INSPIRE -- The codebase is frozen but the UI patterns are timeless. Study the real-time position blotter, the streaming P&L component, and the reactive data flow architecture. Reference implementation for building your trading dashboard. |
TradingView
| Attribute |
Detail |
| License |
Proprietary (Lightweight Charts library is Apache 2.0) |
| Activity |
Commercial product |
| Quality |
Industry-standard charting. The open-source Lightweight Charts library is excellent. |
| Commodity Fit |
Futures and commodity charting is core to TradingView. |
| Verdict |
VENDOR (Lightweight Charts) -- Use TradingView's open-source Lightweight Charts library for interactive price charts in your UI. Apache 2.0 license. Do not attempt to rebuild charting from scratch. |
pysystemtrade (Rob Carver)
| Attribute |
Detail |
| License |
GPL-3.0 |
| Stars |
~2.7k |
| Activity |
Active -- Moved to pst-group org Jan 2026; Andy Geach is primary maintainer |
| Quality |
The most complete open-source futures trading system. Roll calendars, carry strategies, portfolio optimization for diversified futures. Implements Rob Carver's books end-to-end. |
| Commodity Fit |
Purpose-built for diversified futures. Roll logic, carry/basis signals, dynamic optimization, IB execution. The commodity futures expertise embedded in this codebase is unmatched. |
| Verdict |
INSPIRE (GPL prevents forking) -- This is the single most important reference for commodity futures trading logic. Study exhaustively: (1) roll calendar construction, (2) carry signal calculation, (3) forecast combination, (4) dynamic position sizing. Reimplement the key algorithms in your own codebase. The GPL license means you cannot fork it, but the algorithms are published in Rob Carver's books and can be independently implemented. |
Futures Curve / Roll Libraries
| Tool |
Notes |
curves (bananapy) |
Forward curve construction from traded prices. Small but useful. MIT. |
| RTL (risktoollib) |
R package with rolladjust() and chart_fwd_curves(). R-only. |
| Commodity Derivatives Pricing Engine |
Academic project. Cost-of-carry models, convenience yield. |
| Verdict |
INSPIRE -- No single library solves commodity curves well. Build your own curve engine. Study pysystemtrade for roll logic, bananapy/curves for interpolation, and RTL for analytics patterns. |
COT Data Libraries
| Tool |
License |
Notes |
cot_reports (NDelventhal) |
MIT |
All CFTC report types. Clean API. |
cftc-cot (Mcamin) |
MIT |
ETL-pipeline oriented. Good for automation. |
pycot (philsv) |
MIT |
Lightweight wrapper. |
| Verdict |
VENDOR cot_reports -- Most comprehensive COT coverage. MIT license. Wrap in your own data pipeline. |
|
Government Data Pipelines (EIA / USDA / NOAA)
| Source |
Python Wrappers |
Notes |
| EIA API v2 |
myeia, eia-client-python, pyEIA |
Energy supply/demand/inventory data |
| USDA ERS |
USDA-REE-ERS/data-apis |
Crop production, WASDE reports |
| NOAA |
Various weather APIs |
Weather impact on agriculture/energy |
| Verdict |
BUILD -- No integrated pipeline exists. The individual wrappers are thin. Build a unified commodity data ingestion service that normalizes EIA, USDA, NOAA, and CFTC data into a consistent schema in Parquet format. This is a competitive moat. |
|
Satellite / AIS Data
| Tool |
Notes |
| AISdb (AISViz) |
AIS vessel tracking database. Python. |
| AISStream |
Free websocket API for global vessel AIS data. |
| GOST_AIS (World Bank) |
AIS processing for trade flow analysis. |
| Verdict |
INSPIRE -- Alternative data for commodity flow analysis (tanker tracking, grain shipments). AISStream provides free real-time data. Build a custom pipeline that correlates vessel movements with commodity prices. This is a genuine alpha source for commodity ETFs. Long-term roadmap item. |
7. Architecture Recommendation
Validating Your Prior
Your prior stack: nautilus_trader + vectorbt-pro (core engines), qlib (ML research), ArcticDB (storage), OpenBB (UI inspiration).
Verdict: 80% confirmed, 20% adjusted.
| Component |
Your Prior |
My Recommendation |
Change Reason |
| Execution Engine |
nautilus_trader |
nautilus_trader (FORK) |
Confirmed. Best-in-class. LGPL is workable. |
| Signal Research |
vectorbt-pro |
vectorbt-pro (VENDOR) |
Confirmed. Unmatched for parameter sweeps. |
| ML Platform |
qlib |
qlib (VENDOR) |
Confirmed. MIT license, Microsoft-backed, RD-Agent is transformative. |
| Storage |
ArcticDB |
Parquet + DuckDB + Polars (primary); ArcticDB (if licensed) |
Override. BSL license requires commercial agreement for production. Parquet+DuckDB is free, performant, zero-ops. Use ArcticDB only if Man Group offers reasonable terms. |
| UI Inspiration |
OpenBB |
OpenBB + Reactive Trader Cloud + TradingView Lightweight Charts |
Expand. OpenBB for data patterns, Reactive Trader for real-time blotter UX, Lightweight Charts for charting. |
Recommended Core Stack
RESEARCH LAYER
vectorbt-pro .......... Signal discovery, parameter optimization
qlib .................. ML model training, experiment tracking, RD-Agent
bt (pmorissette) ...... Portfolio-level strategy backtesting
tsfresh ............... Systematic feature extraction
EXECUTION LAYER
nautilus_trader ....... Backtest + live execution engine (FORK)
Custom roll engine .... Futures curve, roll logic (built in-house, informed by pysystemtrade)
DATA LAYER
Polars ................ DataFrame engine (replaces pandas everywhere)
Parquet ............... Persistence format
DuckDB ................ Ad-hoc analytics / SQL queries
ArcticDB .............. Optional: versioned research datasets (if licensed)
cot_reports ........... CFTC positioning data
Custom pipelines ...... EIA, USDA, NOAA ingestion (build)
PORTFOLIO / RISK
riskfolio-lib ......... Optimization (30+ risk measures, CVXPY)
skfolio ............... Walk-forward cross-validation of portfolio models
quantstats ............ Performance reporting / tear sheets
empyrical-reloaded .... Financial metrics utility
UI LAYER
TradingView
Lightweight Charts .. Interactive price charting (Apache 2.0)
Custom dashboard ...... Inspired by OpenBB (data layer) + Reactive Trader (real-time blotter)
COMMODITY-SPECIFIC (build in-house)
Roll calendar engine
Futures curve constructor
Carry / basis signal calculator
COT positioning analytics
EIA/USDA/NOAA data normalizer
Vessel AIS flow tracker (future)
Why This Stack Wins
-
No GPL contamination. nautilus_trader (LGPL) links cleanly. Everything else is MIT/BSD/Apache. The proprietary strategy code stays proprietary.
-
Rust + Python hybrid. nautilus_trader's Rust core handles the hot path (order matching, event dispatch). Python handles strategy logic, ML, and analytics. Best of both worlds.
-
Commodity moat is in-house. The commodity-specific logic (curves, rolls, carry, government data pipelines) is where alpha lives. No open-source library solves this well. Building it yourself is not a liability -- it is the product.
-
Storage is zero-ops. Parquet files on disk (or S3), queried by DuckDB or Polars. No database server to manage. Version control via git-lfs or ArcticDB if licensed.
-
ML pipeline is institutional-grade. qlib provides the experiment infrastructure that would take months to build. The RD-Agent for automated feature/model research is a genuine multiplier.
-
Portfolio construction is best-in-class. riskfolio-lib + skfolio together cover every optimization method in the literature, with proper cross-validation to avoid overfitting -- critical for commodity strategies that exhibit regime changes.