| name | financial-statement-analysis |
| scope | co-consult |
| description | Comprehensive Korean financial statement analysis pipeline. Collects DART data, validates accounting integrity, normalizes to canonical financial model, extracts KPIs (profitability, growth, leverage, cash flow), builds a 5+ level ROIC value driver tree, and generates a structured Markdown report.
Use when: "financial statement analysis", "ROIC analysis", "value driver tree", "financial KPI", "profitability analysis", "cash flow analysis", "fundamental analysis", "DART analysis", "financial metrics extraction", "corporate financial analysis".
|
| status | active |
| owner | data-analyst |
| version | 1.0.0 |
| last_reviewed | "2026-07-19T00:00:00.000Z" |
| prerequisites | ["k-dart"] |
| metadata | {"type":"analysis","tier":"medium","triggers":["financial statement analysis","ROIC analysis","value driver tree","financial KPI","profitability analysis","cash flow analysis","DART analysis","fundamental analysis","corporate financial analysis","financial metrics extraction"]} |
Financial Statement Analysis Workflow
Context
End-to-end financial statement analysis pipeline for Korean listed companies.
Uses DART (Financial Supervisory Service electronic disclosure system) data
to produce validated, normalized financial analysis with ROIC-based value driver decomposition.
Architecture: Bun (TypeScript) orchestration + Python pandas data processing.
This hybrid approach complies with the project's Computational Integrity policy
(Class A financial calculations must use validated external tools).
Owner Agent: data-analyst (dispatched by PM)
When to Use
- A consulting engagement requires ROIC, profitability, growth, leverage, or cash-flow analysis of a Korean listed company
- The client or engagement lead asks for a "financial statement analysis", "fundamental analysis", or "value driver tree"
data-analyst needs validated, normalized financial data before building a business case in financial-modeling- Do NOT use for non-Korean companies (no DART coverage) or for engagements needing only qualitative company research — use
company-intelligence instead
Prerequisites
k-dart skill — for DART data collection (Phase ①)- Python 3 with
pandas installed — for data processing (Phases ②–⑤)
- Bun runtime — for orchestration scripts
Pipeline Stages
① DART Collection (k-dart)
↓
② Validation Engine (python/validate.py)
↓
③ Normalization (python/normalize.py)
↓
④ KPI Extraction (python/kpi.py)
↓
⑤ ROIC Value Driver Tree (python/driver_tree.py)
↓
⑥ Report Generation (scripts/financial-report.ts)
① DART Data Collection
Use the k-dart skill to collect raw financial data via DART OpenAPI fnlttSinglAcntAll.
Output: deliverables/<company>/dart/<company-name>-<corp_code>-<start>-<end>.json
Required parameters:
- Company name or ticker
- Corp code (8-digit DART code)
- Year range (e.g., 2020–2025)
fs_div: CFS (consolidated) preferred, OFS (separate) as fallbackreprt_code: 11011 (annual business report)
② Validation Engine
Validates raw DART data for accounting integrity.
Rules:
| Rule | Description | Threshold |
|---|
| Accounting Equation | Total Assets = Total Liabilities + Total Equity | Residual < 5B KRW |
| Cash Flow Reconciliation | Beginning Cash + OCF + ICF + FCF = Ending Cash | Residual < 1B KRW |
| NI → Retained Earnings | Retained earnings increase ≈ net income | 20% tolerance |
| Anomaly Detection | Flag YoY changes > 200% or sign flips | Base > 1B KRW |
Runner: bun scripts/co-consult/financial-validate.ts <dart.json> --output <path>
③ Normalization
Converts Korean DART account names to a Canonical Financial Model using
industry-specific mapping tables.
Mapping table: python/mappings/ifrs_general.json
Canonical fields (36 per year):
- Income: revenue, cogs, gross_profit, sg_and_a, rd_expense, depreciation, operating_income, ebit, ebt, nopat, tax_expense, net_income, fin_expense, fin_income
- Balance Sheet: total_assets, current_assets, cash, receivables, inventory, non_current_assets, ppe, intangible_assets, total_liabilities, current_liabilities, st_debt, lt_debt, lease_liabilities, right_of_use_assets, total_equity, retained_earnings
- Derived: invested_capital, working_capital
- Cash Flow: operating_cf, investing_cf, financing_cf, free_cash_flow, dividends_paid
- Cross-period: prev_revenue
Runner: bun scripts/co-consult/financial-normalize.ts <dart.json> --mapping <path> --output <path>
④ KPI Extraction
Computes financial KPIs from the canonical model.
KPI Groups:
| Group | KPIs |
|---|
| Profitability | Gross Margin, Operating Margin, EBITDA Margin, Net Margin, ROE, ROA, ROIC |
| Growth | Revenue YoY, Operating Income YoY, Net Income YoY, Revenue 3yr CAGR |
| Leverage & Liquidity | D/E, D/A, Equity Ratio, Current Ratio, Quick Ratio, Interest Coverage, Net Debt |
| Cash Flow | OCF Margin, FCF, OCF/Debt, Dividend Payout, Cash Ratio, Asset Turnover, Receivables Turnover, Fixed Asset Turnover |
| Efficiency | COGS/Revenue, SG&A/Revenue, R&D/Revenue, Depreciation/PPE |
Runner: bun scripts/co-consult/financial-kpi.ts <canonical.json> --output <path>
⑤ ROIC Value Driver Tree
Builds a 5+ level decomposition of ROIC into its operational drivers.
Tree structure:
ROIC (L0)
├── NOPAT (L1)
│ ├── Revenue (L2) → Growth Drivers (L3) → YoY, CAGR (L4)
│ ├── Operating Margin (L2) → Gross/SGA/R&D/Depreciation (L3) → Detail (L4)
│ └── Tax Efficiency (L2) → Effective Rate, NOPAT Margin (L3)
└── Invested Capital (L1)
├── Working Capital (L2) → Receivables/Inventory/Payables Days (L3) → Ratios (L4)
├── Net Fixed Assets (L2) → PPE/Revenue, Capex (L3) → Turnover (L4)
├── Intangible Assets (L2) → Intangibles/Revenue, YoY (L3)
└── Cash & Debt (L2) → Net Cash, D/E (L3)
Each node contains: value, YoY change, weight (contribution), comment (AI-filled).
Runner: bun scripts/co-consult/financial-driver-tree.ts <canonical.json> --output <path>
⑥ Report Generation
Generates a structured Markdown report from all pipeline outputs.
Sections:
- Executive Summary
- Financial Highlights
- Profitability Analysis (with Returns)
- Growth Analysis
- Leverage & Liquidity
- Cash Flow Analysis
- ROIC & Value Driver Tree
- Investment View (AI agent synthesizes)
- Appendix A: Validation Summary
- Appendix B: Data Coverage
Runner: bun scripts/co-consult/financial-report.ts <canonical.json> <validation.json> <kpi.json> <tree.json> --output <path>
End-to-End Pipeline
Run all stages in sequence:
bun scripts/co-consult/financial-pipeline.ts <dart.json> --company <name> --output-dir <dir>
This creates the full output structure:
<output-dir>/
├── dart/dart-YYYY-MM-DD.json
├── validation/validation-report-YYYY-MM-DD.json
├── canonical/canonical-model-YYYY-MM-DD.json
├── kpi/kpi-report-YYYY-MM-DD.json
├── driver-tree/driver-tree-YYYY-MM-DD.json
└── reports/financial-analysis-<company>-YYYY-MM-DD.md
Execution Steps
(also referred to as the Execution Protocol for the data-analyst agent)
- Receive dispatch from PM with company name and year range
- Collect data using
k-dart skill (or use existing data if available)
- Run pipeline:
bun scripts/co-consult/financial-pipeline.ts <dart-file> --company <name>
- Review outputs — check validation pass rate (>90% acceptable), coverage (>80% acceptable)
- Synthesize Investment View — fill section 8 of the report with analytical insights
- Hand off to
communications-lead or strategy-analyst as appropriate
Output Format
The pipeline emits one Markdown report (see §⑥ Report Generation for the section list) plus the intermediate JSON artifacts it was built from:
| Output | Destination | Format |
|---|
| Raw DART data | deliverables/<company>/dart/ | JSON |
| Validation report | deliverables/<company>/validation/ | JSON |
| Canonical model | deliverables/<company>/canonical/ | JSON |
| KPI report | deliverables/<company>/kpi/ | JSON |
| Driver tree | deliverables/<company>/driver-tree/ | JSON |
| Final report | deliverables/reports/ | Markdown |
Error Handling
| Stage | Failure Mode | Recovery |
|---|
| ② Validation | Pass rate < 80% | Flag data quality issues, proceed with warnings |
| ③ Normalization | Coverage < 60% | Check industry mapping, may need custom mapping |
| ④ KPI | Missing key fields | Report which KPIs are unavailable due to missing inputs |
| ⑤ Driver Tree | NOPAT/IC unavailable | Report tree with null nodes and notes |
Related Skills
k-dart — DART data collection (prerequisite)financial-modeling — Consulting ROI/NPV business cases (complementary)company-intelligence — Broader company research (upstream)
Notes
- All financial computations are performed by Python pandas (Class A Computational Integrity compliance)
- A NodeJS-Polars validation engine was considered as an alternative to
python/validate.py and explicitly rejected — the pipeline keeps the Bun/TypeScript orchestration + Python/pandas computation split as-is
- The AI agent must NOT calculate financial figures directly
- Korean DART account names vary by company; the normalization mapping absorbs this variation
- The pipeline is loosely coupled — each stage can run independently with JSON I/O
- Future expansion: SEC EDGAR (US), EDINET (Japan) support via new mapping tables only
