DCF Valuation Pipeline

A local DCF (Discounted Cash Flow) valuation system for Brazilian equities using the Damodaran methodology. An AI agent extracts financial data from company earnings releases (PDFs), runs a structured 5-step pipeline, and presents the results in a web dashboard with inline editing.

Features

• 5-step pipeline driven by an AI coding agent (pi) — from PDF extraction to sensitivity analysis
• Web dashboard — lists all analyzed tickers, detail view per company with all pipeline steps
• Inline editing — edit any extracted or assumed value directly in the browser; saves back to JSON instantly
• One-click recalculation — after editing step 1 or 2, hit "Recalcular DCF" in step 4 to rerun the model
• Zero frontend dependencies — vanilla HTML/CSS/JS, served by Python stdlib http.server

Requirements

• Python 3.8+
• pi CLI (npm install -g @earendil-works/pi-coding-agent)
• pypdf (pip install pypdf)

Project Structure

Market/
├── Acoes/
│   └── <TICKER>/
│       ├── release.pdf          # earnings release (pipeline input)
│       ├── step1.json           # extracted financials
│       ├── step2.json           # market parameters
│       ├── step3.json           # analyst assumptions
│       ├── step4.json           # DCF calculation output
│       └── step5.json           # sensitivity / scenarios
│
├── Valuation/
│   ├── script/
│   │   ├── valuation_pipeline.py   # pipeline orchestrator
│   │   └── step4_dcf.py            # DCF calculation engine
│   ├── web/
│   │   ├── server.py               # local HTTP server
│   │   ├── index.html              # SPA shell
│   │   ├── style.css               # styles
│   │   └── app.js                  # frontend logic
│   ├── data/
│   │   ├── beta-by-sector.md       # Damodaran beta table
│   │   ├── country-default-spreads-and-risk-premiums.md
│   │   └── ratings.md              # synthetic rating spreads
│   └── instructions/
│       ├── 00_pipeline_index.md        # pipeline index and transition rules
│       ├── step{1-5}_*.md              # step instructions for the AI agent
│       └── system_prompt_agente_dcf.md # AI agent system prompt

Running the Pipeline

Place the earnings release PDF at Acoes/<TICKER>/release.pdf, then:

# Run all 5 steps
python Valuation/script/valuation_pipeline.py FIQE3

# Resume from a specific step
python Valuation/script/valuation_pipeline.py FIQE3 --start-from 3

# Run only one step
python Valuation/script/valuation_pipeline.py FIQE3 --only 1

# Dry run (print prompts, don't execute)
python Valuation/script/valuation_pipeline.py FIQE3 --dry-run

Steps 1–3 and 5 are handled by the pi AI agent. Step 4 runs step4_dcf.py directly as a Python script.

Running the Dashboard

# From the Market/ root
python Valuation/web/server.py

# Custom port
python Valuation/web/server.py 3000

Open http://localhost:8000/Valuation/web/ in your browser.

The 5-Step Pipeline

Step	Name	What happens
1	Release Extraction	AI agent reads the earnings PDF and extracts income statement, balance sheet, and market data into JSON
2	Market Parameters	AI agent looks up Damodaran tables (ERP, beta, ratings) and calculates WACC inputs
3	Analyst Assumptions	AI agent consolidates all data and sets growth, margin, and perpetuity assumptions
4	DCF Calculation	Python script runs the full Damodaran DCF model (WACC → FCFF projections → terminal value → equity bridge)
5	Sensitivity / Scenarios	AI agent runs bear/base/bull scenarios and breakeven analysis

Dashboard Overview

List view — one card per ticker showing market price vs. intrinsic value, upside/downside badge, and pipeline completion bar.

Detail view — accordion with one section per step:

• Step 1: DRE, balance sheet, and market data tables with confidence indicators
• Step 2: Cost of capital parameters with source references and alerts
• Step 3: Assumption cards (growth rates, margins, perpetuity) + analyst narrative
• Step 4: KPI summary, 10-year FCF projection table with visual bars, equity value bridge, validation checklist, and Recalcular DCF button
• Step 5: Bear/Base/Bull scenario cards and breakeven sensitivity table

All numeric fields are editable inline — click any value, type the new number, press Enter to save. The JSON file is updated immediately on disk.

Recalculating After Edits

The ▶ Recalcular DCF button in Step 4:

1. Reads current step1.json and propagates financials → step3.dados_historicos / dados_mercado
2. Reads current step2.json and propagates cost of capital → step3.parametros_custo_capital
3. Saves the updated step3.json
4. Runs step4_dcf.py and writes the new step4.json
5. Re-renders the dashboard in place — no page reload needed

Server API

Method	Path	Description
GET	/api/tickers	List of ticker folders in Acoes/
PUT	/api/passo/<ticker>/<n>	Overwrite Acoes/<ticker>/step<n>.json
POST	/api/run-passo4/<ticker>	Propagate edits and rerun the DCF script
GET	/*	Static file serving from Market/ root