Skip to content

Module 10 · Financial Modelling

What this module does. Turns the driver logic of Modules 04–06 into a working three-statement model: the architecture and conventions, building and tying out the historicals, forecasting each schedule (revenue, costs, working capital, capex/D&A, debt/interest, tax, dividends, cash as the plug), scenarios/sensitivities/QA, and a hands-on build of the Kaveri model in Python with tools/fi/model.py that reproduces ₹320 and then stresses it.

Why it matters. A model is how you find out whether your assumptions are mutually consistent — and the only way to see the balance sheet your forecast implies (Kaveri's model shows cash building to ₹220 Cr by FY29, which raises a capital-allocation question no DCF alone would).

Lessons

# Lesson One line Time
10.1 Model architecture Inputs → schedules → statements → valuation → outputs; conventions; Excel vs Python 60 min
10.2 Building the historicals Sources, mapping, reclassification, normalisation, tie-outs, vendor pitfalls 75 min
10.3 Forecasting drivers Revenue and cost builds, days-based NWC, capex/D&A, debt and circularity, tax, dividends, the plug; Kaveri FY27–29 100 min
10.4 Scenarios, sensitivities & model QA Switches, data tables, tornado, stress tests, the 17-point review 75 min
10.5 Build-along: the Kaveri model in Python The whole pipeline; the receivables stress; pointing it at a real company 120 min hands-on

Connections

  • Back: 02.6, 06.3, 06.4.
  • Forward: the model is the engine of the memo (11.3) and the capstone (15.3).
  • Tools: tools/fi/model.py, tools/examples/build_kaveri_model.py, tools/tests/test_model.py.

You're ready to move on when you can…

  • Lay out a model with inputs in one place, checks on the summary page and cash as the only plug.
  • Build and tie out five years of historicals for a real company from primary sources.
  • Forecast every schedule from drivers and explain the circularity choice.
  • Reproduce Kaveri's FY27–FY36 statements and the ₹320 value with project(); run the receivables stress.
  • Produce scenario switches, two-way grids, a tornado and a stress test, and pass the QA checklist.
  • Map a real company's Yahoo/Screener data into the model layout and run it.

Tested by: Mock 3 — Valuation (model-based problems) and the final exam case.

Exercises · Solutions