Finance tools by Jason Zou
This site documents six tools I built for equity research and portfolio work. They run as Claude skills and plugins. Each one has its own page explaining how it works, what checks it runs on itself and which decisions it doesn't make for you. The site also runs a cost-aware classifier, a web app that measures which Claude model is worth paying for when labelling news headlines.
- 01Stock diligence pack · builds a linked three-statement DCF on a listed company and writes the report around it
- 02Reverse DCF · works out the growth and margins a share price already assumes
- 03Direct indexing basket · tracks an index with a smaller basket of its stocks
- 04Stock screener · puts any ticker on the same fixed dashboard
- 05Cost-aware classifier · finds the cheapest Claude model good enough to label news headlines for a sentiment signal
- 06Two smaller tools · a correlated substitutes finder and a MACD signal test
All six are my own work. Spotify is the worked example used across the site, starting with the chart on the right.
- Source
- SPOT_DCF.xlsx, rebuilt from the plugin's input file and recalculated for all four cases; 14 tabs, 1,984 formulas
- Checks
- 17, all passed in the Street case (the case built on consensus forecasts)
- Output
- Test-run output, not a valuation of Spotify. The Street case comes out 5.2% below the share price. The management case, built from management's own targets, comes out 10.8% above it
The tools
Each card opens a page on one of the seven tools, three of which have public repositories. Skills are folders of instructions and scripts that Claude runs. A plugin bundles several skills, and a subagent is a separate model pass with its own brief.
Stock diligence pack
Builds a linked three-statement DCF, with a debt schedule and the bridge from enterprise value to equity, then writes the report and a memo on each forecast lever. Spotify is the worked example on this site.
Reverse DCF
Starts from the share price and works out the perpetual growth and exit multiple it implies. Then it shows which growth and margin pairs clear that price.
Direct indexing basket
Builds a smaller basket of stocks that tracks an equity index, by default 75 names for the S&P 500. It backtests the basket and reports its tracking error (how far its returns stray from the index's) at each basket size.
Stock screener
Pulls prices, five years of fundamentals, drawdowns, what analysts expect (the Street) and news for one ticker, and lays them out on a fixed six-tab dashboard. The full dashboard runs on its page with synthetic data.
Correlated substitutes finder
Takes one stock and finds the three liquid US names whose returns behave most like it, ranked on a composite score rather than raw correlation.
MACD signal test
Tests whether the monthly MACD crossover has worked on a given stock. It scores every crossover from five years of data on its forward return, and the rules for the result are fixed in advance.
An example run of the diligence pack
Spotify is the worked example throughout this site. The workbook shown here is rebuilt from the plugin's own input file and recalculated. Each figure below links to where it sits on the diligence page.
| CAGR ↓ · margin → | 16% | 20% | 24% | 28% | 32% |
|---|---|---|---|---|---|
| 8% | 315 | 368 | 422 | 475 | 529 |
| 10% | 357 | 421 | 484 | 547 | 610 |
| 12% | 407 | 482 | 556 | 631 | 706 |
| 14% | 465 | 553 | 640 | 728 | 816 |
| 16% | 531 | 635 | 738 | 841 | 945 |
The reverse DCF
Before asking what the stock is worth, the model works out what would have to be true for today's price to be right. It sets the answer out as a grid of growth and margin pairs.
- 01Start with the market enterprise value, taken from the share price and the equity bridge.
- 02Work out the terminal value that enterprise value requires, given the explicit cash flows.
- 03Solve in closed form for the perpetual growth rate and exit multiple that terminal value implies.
- 04Find the growth and margin pairs that clear the price.
How the tools work
All six tools follow some part of the same five stages, and the diligence pack runs all five. The screener and MACD test stop at Check, and the basket builder and substitutes finder stop at Build. Scroll through the stages and the panel follows.
- 01
Source
The coordinating agent saves the filings, decks and transcripts, and a script,
data_pull.py, pulls prices, statements, estimates, earnings dates, insider transactions, holders and FX. Gaps are recorded in the manifest (the file that lists every source and every gap) rather than filled in. - 02
Build
Every numeric exhibit is computed. The workbook is built from one input file, so a figure copied from a filing lives in one place and every formula reads from it.
- 03
Check
The model is recalculated and run through its checks across four cases and three switches, such as the choice of terminal method. It is not delivered until it recalculates with zero formula errors and passes every check in the Street case.
- 04
Review
Five reviewers, each prompted as a different reader, return ranked findings tagged FAIL or IMPROVE. They do not edit anything themselves; the fixes are made afterwards and the file re-rendered.
- 05
Decide
The view on the company, the scenario weights, any price target and the sign-off on adapted assumptions are yours. The tools do not make recommendations.
| Input | Source | Rule |
|---|---|---|
| Statements, five years | Filings first, then the data provider's standardised statements | Reconcile when they differ by more than 2% |
| Prices, ten years | Yahoo Finance | Fixed reference dates for every return |
| Estimates and targets | Consensus as reported | Blank rather than guessed when a broker did not disclose |
| Management targets | Investor Day, transcripts | Recorded with the date they were given |
| Gaps | MANIFEST.md | Listed with what was tried |
| Check | Rule | Street |
|---|---|---|
| Balance sheet balances every year | < 0.5 | OK |
| Cash never negative | ≥ 0 | OK |
| WACC exceeds terminal growth | > 2 pts | OK |
| Terminal value share of EV | 40–85% | OK |
| CF closing cash equals BS cash | 0 | OK |
| Debt schedule ties to balance sheet | 0 | OK |
| Sensitivity grid centre equals DCF price | 0 | OK |
| Overall status · 17 checks | ALL OK |
- R1Investment partner · reads the model for what the price already implies, the probability-weighted value, the return skew and a pre-mortem
- R2Banking managing director · reviews the capital structure, debt capacity, change-of-control constraints and strategic options
- R3Research partner · goes through the model against consensus line by line, then the guidance ledger and the setup for the next print
- R4Red team · hunts for claims with no mechanism behind them, figures quoted from memory and benchmarks that contradict the narrative
- R5QA · checks the numbers in the prose against the tables, and looks for leftover placeholders and breaches of the style rules
The tool
- Data pulls, exhibits and charts
- Building, recalculating and checking the model
- Drafting sections and rendering the report
- A five-reader review with ranked findings
You
- The view on the company
- Scenario weights and any price target
- Sign-off on adapted assumptions
- Which open items matter
About
I study Economics and Philosophy at the University of Chicago. The tools I've built so far help you get up to speed on a company, read what a share price assumes, track an index with fewer names and give any ticker the same first look.
All of them run as Claude skills or plugins that I designed and built. Spotify is the worked example for the diligence pack, and the screener has been run on eight tickers.
Notes
These notes say which numbers on this site come from a real run, which are illustrative, and what the technical terms mean.
Repositories and contact
Three of the seven tools have public repositories. The diligence plugin, which includes the reverse DCF, and the MACD test aren't published yet. You can still read the Spotify workbook, the plugin source and the XMPL dashboard on the pages above without downloading anything.