Every analysis here ships its dataset and the code that produced every figure. Nothing on this site is investment advice.
Every payroll and GDP vintage since 2010 from ALFRED without a key, the alfredgraph.csv vintage_date parameters tested, and how far first prints moved.
Sixty month-of-year tests on the Nasdaq, Treasuries, oil, gasoline and natural gas: 10 pass at p < 0.05, and a 5% false discovery rate keeps 4, all gasoline.
Nelson-Siegel fitted to 11 Treasury yields on 5,171 days misses by 9.4 bp on a median day; since May 2020 the 20-year sits above a curve fitted without it daily.
From January 2021 to August 2026, real hourly pay fell 0.8% by the CPI and rose 1.7% by the PCE price index. How to deflate, checked against the BLS, and six traps.
Rates, VIX and the Nasdaq in a five-day window around every FOMC statement since 2003, with permutation tests, a multiple-testing correction, and what survives it.
Rolling 126-day correlations between the 10-year yield, equities, and the dollar, with a simulation that separates a real regime change from sampling noise.
Every inversion of the 10-year minus 2-year Treasury spread since the 2-year series began in 1976, with dates, depth, duration, and what followed, from FRED.
Shelter is 35.6% of the CPI-U basket for 2026 and energy 6.4%, yet energy supplied 38% of the index's rise from December to August. The weights, and how to use them.
Price, yield to maturity, Macaulay and modified duration, and convexity derived and coded from scratch, then applied to the Treasury par curve of 2026-09-09.
modified duration of a 10-year par bond at the 4.83% yield of 9 September 2026
Data and code: three CSV files and one Python script, attached to the note.
Price of the 10-year par bond after a parallel yield shift: exact (solid) and duration-only (dashed)
days the 99% normal VaR was exceeded in 2,013 backtest days, against 20.1 expected
Historical, parametric normal, and Cornish-Fisher VaR computed on ten years of FRED's S&P 500 series, then backtested day by day with the Kupiec coverage test.
Every sentence mentioning AI in Target's 10-K filed 2026-03-11 (accession 0000027419-26-000016), counted by section and quoted, with a script for any filing.
The three free EDGAR JSON endpoints in one standard-library script: a company's filings by accession number, full-text search hits, and deduplicated XBRL facts.
Simple and log returns on ten years of FRED's S&P 500 series: the sum identity, variance drag, the size-of-move gap, and the mistakes each choice invites.
Four 2026 benchmarks measured LLMs computing ratios, verifying statements, answering from full 10-Ks, and summarising MD&A. Accuracies and failure modes, cited.
What 2026 papers report on language and foundation-model forecasts in finance, by arXiv id, and four simple baselines run on archived FRED vintages since 2010.
| Desk | Notes | What it covers |
|---|---|---|
| Data Analysis | 9 | A question, a public dataset, the code, and the answer, with every figure reproducible. |
| Methods | 7 | Time series, risk measures, backtesting discipline, and the statistics that keep an analysis honest. |
| Data Sources | 6 | FRED, SEC EDGAR, BLS, exchange feeds: how to pull them, parse them, and not misread them. |
| LLMs in Finance | 5 | Where language models are used in financial work, what the papers and filings show, and what they get wrong. |
| Briefs | 2 | Short readings of a release, a filing, or a paper, dated and sourced. |