gauss314 Financial Market Skills

The repository holds 32 skill directories, grouped in the README as 21 global data skills, six Argentina-specific ones (BCRA Macro, Data912, MAE, BYMA, CAFCI, INDEC), two broker execution skills and three tools. Eighteen of the global data skills are marked free and two freemium, with five requiring an API key, among them FRED Macro, Alpha Vantage, Alpaca Data and Finnhub. Installation is one command into any SKILL.md-standard agent — npx skills add gauss314/skills -g for selected skills, npx skills add gauss314/skills --all for all 32.

The three tools are the quant core, and they run on numpy, pandas and scipy alone. The backtesting skill follows a five-stage methodology taken from the course material (Data → Research → Metrics → Parameterisation → Validation) with about 30 vectorized risk and performance ratios, ten taxonomic classes of indicators, an event-driven engine, Johnson SU and t/Gaussian-copula forward simulation, walk-forward cross-validation with an IS/OOS split and gap, parametric stress testing and fundamental screens (Altman Z, Piotroski F, DuPont); it ships SPY benchmark returns since 1980, momentum and contrarian strategy return series, sample portfolios and validation cases. The option-pricing skill spans 15 CLI modes (Black-Scholes, binomial CRR, trinomial, Monte Carlo with antithetics, Longstaff-Schwartz, Bjerksund-Stensland/BAW, Heston, Bates, greeks, implied volatility) and the portfolio skill 12 (Markowitz, Black-Litterman with Idzorek omega, HRP, HERC, NCO), neither depending on Riskfolio-Lib or PyPortfolioOpt. Internal counts disagree — the README credits the event-driven engine with eight built-in strategies where the skill’s own frontmatter says six or more — and the author’s throughput and validation claims (419k Black-Scholes options per second, a 33-check validation suite, portfolio output matching library results exactly) are self-reported and were not reproduced here.

Two skills place real orders: Alpaca Trading (US stocks and options, paper and live base URLs both documented) and Primary (Argentine futures). Nine or more of the data skills are scrapers against commercial sites — Finviz, Macrotrends, MarketScreener, MarketWatch, CompaniesMarketCap, SimplyWallSt, Barchart, Investing.com — and MIT covers the skill code, not permission to scrape those sources; each site’s terms of service remain the user’s problem. Documentation is mixed English and Spanish: the top-level README is English, while the alpaca-trading skill body and the backtesting skill’s six reference documents are Spanish. This is a single-author project written for UCEMA’s AI courses, created 2026-06-02 with roughly twelve days of commits behind it, and dormant since 2026-06-14.