10.2 · Building the historicals¶

Why this matters: every forecast is an extrapolation from history, so a model is only as good as the historical block it starts from. Getting five to ten years of clean, consistently classified, tied-out statements is unglamorous and takes half the modelling time — and it is where you learn what the company actually does. This lesson covers where the data comes from, how to standardise it, what to reclassify, and how to prove the historicals tie before you forecast anything.

Learning objectives — after this lesson you can:

Get financial data for an Indian company from the primary source (annual reports), from Screener.in and from yfinance/OpenBB via tools/fi/data.py, and know each source's limits.
Map reported lines to a standard model layout and make consistent reclassification decisions (other income, exceptional items, leases, interest classification).
Normalise history for one-offs and policy changes.
Prove that the historicals tie: balance sheet balances, cash flow reconciles, retained earnings roll forward.
Recognise data-vendor pitfalls: restatements, standalone/consolidated mix-ups, unit errors, missing rows.

Prerequisites: 02.6 Linking the three statements, 03.2 Anatomy of an annual report · Time: ~75 min (plus a day the first time you do it on a real company)

1. Sources, in order of authority¶

Source	What you get	Cost	Limits
Annual reports (company website; BSE/NSE)	Audited consolidated and standalone statements with every note; 2 years per report, so 5 reports for 10 years	Free	Manual extraction; restatements mean last year's comparatives may differ from last year's report — use the latest figure for each year
Quarterly results (exchange filings, Reg 33)	Quarterly P&L, half-yearly balance sheet, segment results	Free	Less detail; no cash-flow statement quarterly
Screener.in (free tier; Excel export)	10+ years of standardised P&L, BS, CFS, ratios, quarterly; consolidated by default	Free	Its standardisation merges lines (e.g., "other income" includes some items you may want separate); "operating profit" is EBITDA ex other income; check units (₹ Cr) and consolidated vs standalone toggle
Yahoo Finance via `yfinance` / OpenBB	4–5 years annual, limited quarterly, in INR; standardised US-style labels	Free	Coverage of Indian companies is patchy (quarterly cash flow often missing; some lines mislabelled); some VPNs block Yahoo; row names vary — always print them
Paid terminals / databases (Capitaline, Ace Equity, Prowess, Bloomberg, Refinitiv, Tikr)	Long histories, standardised, restated	₹ thousands to lakhs a year	Standardisation choices are theirs

The course convention: build from annual reports for the company you are seriously studying; use Screener or fi.data for screening and first looks; never mix sources within one model without reconciling.

2. Fetching with the course tools¶

import io, sys
sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding="utf-8")   # ₹ on Windows
from fi.data import fetch_statements, fetch_price_info, revenue_row

stm = fetch_statements("ASIANPAINT.NS", period="annual", in_crore=True)   # disable VPN first
inc, bs, cf = stm["income"], stm["balance"], stm["cash"]
print(inc.index.tolist())          # ALWAYS inspect the row names before using them
print(revenue_row(inc))            # handles 'total_revenue' vs 'operating_revenue'
info = fetch_price_info("ASIANPAINT.NS")
print(info["price"], info["market_cap"], info["shares_outstanding"], info.get("beta"))

fetch_statements(ticker, period="annual"|"quarter", provider="yfinance", limit=5, use_openbb=None, in_crore=False) uses OpenBB's obb.equity.fundamental.income/balance/cash when OpenBB is installed and falls back to raw yfinance; in_crore=True divides rupee values by 10⁷. Quarterly uses period="quarter" (OpenBB's spelling). The tools appendix lists the known quirks (missing quarterly cash flows for some companies; label differences). For the running example, load_kaveri() and load_nirmal() return the course's clean frames.

3. The standard layout¶

Map every reported line to a fixed set of model rows. The Kaveri layout (tools/data/kaveri_pumps_annual.csv) is the course standard for a non-financial company:

Block	Model rows
P&L	`rev, mat, emp, sbc, oth, ebitda, dep_ppe, rou_amort, int_amort, da, ebit, other_income, debt_int, lease_int, fin_cost, exceptional, pbt, current_tax, deferred_tax, tax, pat`
Balance sheet	`gross_block, acc_dep, net_block, cwip, rou, intangibles, inventory, receivables, oca, cash, cur_inv, total_assets, share_capital, other_equity, equity, lt_debt, st_debt, lease_liab, payables, ocl, dtl, total_le`
Cash flow	`op_before_wc, d_inv, d_rec, d_oca, d_pay, d_ocl, wc_change, cfo, capex_ppe, capex_int, d_cur_inv, asset_sale, cfi, d_lt, d_st, lease_payment, dividends, cff, net_cash`

Mapping decisions you must make — and record on a Notes sheet:

Reported item	Model treatment	Reason
Other income	Separate row below EBIT; exclude from EBITDA/EBIT	Treasury income is valued via cash in the bridge, not capitalised at an operating multiple
Exceptional items	Separate row; excluded from "adjusted" metrics with tax effect	One-offs distort trends (02.3)
Purchases of stock-in-trade and change in inventories	Fold into `mat` (cost of goods)	Gross margin needs all three
Share-based payment	Inside `emp`, shown as memo `sbc`	Real cost; non-cash add-back in CFS
Leases (Ind AS 116)	`rou` asset, `lease_liab`, `rou_amort` in D&A, `lease_int` in finance costs, payments in CFF	Consistent with treating leases as debt; restate pre-FY20 years if comparing
Interest paid / received	Paid in CFF, received in CFI (Kaveri's policy) — or restate all to CFO	Pick one and apply to every company you compare
Current maturities of long-term debt	Inside `st_debt` (Schedule III) — or reclassify to `lt_debt` for maturity analysis	Note which
Provisions (current)	Inside `ocl`
Deferred tax	`dtl` net (or DTA separately if material)	Quasi-equity for capital employed
Non-controlling interests	Separate equity row; subtract in the bridge
Investments in associates	Non-operating; add in the bridge at value
Discontinued operations	Exclude from continuing lines; note separately

4. Normalising history¶

Before you compute ratios or trends, adjust for:

One-offs: exceptional items (with tax); large provision reversals; asset sales in other income; COVID quarters.
Policy changes: Ind AS 116 (FY20) — restate earlier EBITDA or later; revenue recognition changes (Ind AS 115, FY19, especially real estate); tax-regime switch (Sec. 115BAA from FY20 — effective rates jump).
Corporate actions: mergers/demergers (pro-forma history if available in the scheme documents); bonus and splits (per-share history restated); changes in year-end.
Consolidation scope: a subsidiary acquired mid-year contributes a partial year; disclosed in the business-combination note.

Keep both the reported and the normalised series; forecast from the normalised, reconcile to the reported.

5. Tie-outs — prove it before you forecast¶

Three identities must hold in every historical year, to the rounding of the source:

Check	Formula	Kaveri FY26
Balance sheet	total_assets − total_le = 0	1,143.6 − 1,143.6 = 0
Cash reconciliation	opening cash + net_cash − closing cash = 0	28.0 + 4.5 − 32.5 = 0
Retained-earnings roll	other_equity₍ₜ₋₁₎ + PAT − dividends + SBC (± OCI, issuance) − other_equity₍ₜ₎ = 0	607.2 + 90.5 − 24.0 + 2.4 − 676.1 = 0
Fixed-asset roll	gross_block₍ₜ₋₁₎ + capex + transfers from CWIP − disposals − gross_block₍ₜ₎ = 0	780.0 + 36.0 + 14.0 − 0 − 830.0 = 0
Working-capital tie	Δ(inventory + receivables + oca − payables − ocl) on BS = −wc_change on CFS	✓ by construction

from fi.data import load_kaveri
k = load_kaveri()
for c in k.columns:
    assert abs(k.loc["total_assets", c] - k.loc["total_le", c]) < 0.05
prev_cash = 34.0    # 31-Mar-2020 opening
for c in k.columns:
    assert abs(prev_cash + k.loc["net_cash", c] - k.loc["cash", c]) < 0.05
    prev_cash = k.loc["cash", c]
print("historicals tie")

If a tie fails on real data, the usual causes: OCI items or share issuance missing from the equity roll; a reclassification between current and non-current between years; disposals or impairments not captured in the fixed-asset roll; FX translation in a group with overseas subsidiaries; or a vendor's standardised line that double-counts. Find it before forecasting — the forecast inherits every error.

6. Vendor and extraction pitfalls¶

Pitfall	Symptom	Fix
Standalone vs consolidated mix	Revenue jumps or falls 30% in one year with no event	Check the toggle/source; use consolidated throughout
Restatements	Last year's number differs between two annual reports	Take each year from the latest report that shows it; note the restatement reason
Units	Numbers 10x or 100x off; ₹ lakh vs crore; ₹ million	Read the header; `in_crore=True` for yfinance
Missing rows	"ebitda" absent in Yahoo; D&A inside operating expenses	Build from components; print `index.tolist()`
Sign conventions	Vendor shows expenses negative	Standardise on entry
Fiscal-year labels	Yahoo labels FY by calendar date (2026-03-31)	Map to FY26
Merged line items	Screener's "other income" mixing treasury and operating items	Go to the report for the split when it matters
Quarterly patchiness (Indian companies on Yahoo)	Missing quarterly cash flow; four vs five quarters	Use exchange filings for quarterly detail

Trader's lens

Historicals are your market data feed. A pricing model on a corrupted feed produces confident nonsense; so does a DCF on unreconciled statements. The tie-outs are the equivalent of checking that bid ≤ ask and that the close matches the exchange print — cheap, boring, and the only thing standing between you and a model that is precisely wrong.

India notes

Consolidated statements are the model base; standalone is read for governance and dividends (02.8).
The Schedule III format is stable enough that a mapping template built for one manufacturer works for most; lenders, insurers and holdcos need their own layouts (Module 07).
Pre-FY17 numbers are under old Indian GAAP for most companies; Ind AS transition-year reports give restated comparatives for one year only — treat older data as a different regime.

Common mistakes

Forecasting from Screener's standardised lines without checking what they contain.
Mixing sources across years.
Skipping the tie-outs because "the vendor's numbers must balance".
Forgetting Ind AS 116 when comparing FY19 with FY20+ margins.
Treating restated comparatives as errors in your model rather than in the source.

Key terms¶

Term	Meaning
Historicals	The block of actual financial statements a model is built on
Standard layout / mapping	The fixed set of model rows and the rules for placing reported lines in them
Reclassification	Moving an item between categories (e.g., other income out of EBITDA) for analytical consistency
Normalisation	Adjusting history for one-offs and policy changes
Tie-out	A check that related figures agree (BS balances; CFS reconciles; rolls close)
Restatement	Revision of previously reported figures
Roll-forward	Opening + additions − reductions = closing, for equity, fixed assets, provisions
Consolidated / standalone	Group vs parent-only statements

Check your understanding¶

Screener shows "operating profit" of ₹186 Cr for Kaveri FY26 and your model shows EBITDA of ₹181.9 Cr. Why?

Answer

Screener's operating profit typically includes other income (3.9) in some layouts, or treats an item (e.g., a component of other expenses) differently; 181.9 + 3.9 = 185.8 ≈ 186. Confirm the definition; use your own consistently.

A company adopted Ind AS 116 in FY20; its EBITDA margin rose from 9% to 13% with no operational change. How do you make FY19 and FY21 comparable?

Answer

Either add FY21's lease payments (from the CFS) back into opex to get a pre-116 EBITDA, or subtract FY19's rent from opex and treat it as D&A + interest. State the convention; compare EBIT if in doubt.

Your retained-earnings roll shows a ₹12 Cr unexplained gap. List three candidate causes.

Answer

OCI items (actuarial remeasurements, FVOCI gains) taken directly to equity; share issuance/ESOP exercise credited to securities premium; a transfer to/from a reserve, dividend distribution tax (pre-2020), or a prior-period adjustment/restatement.

Why does the course exclude other income from EBITDA and EBIT?

Answer

Because treasury income is generated by cash and investments that are added separately in the equity bridge; including it in EBIT would capitalise it at an operating multiple and count the cash twice. It also makes margins comparable across companies with different cash piles.

fetch_statements("XYZ.NS") returns an income frame with no ebitda row. What do you do?

Answer

Print the index, find `operating_income`/`ebit` and `depreciation_and_amortization` (or reconstruct from revenue and expense rows), build EBITDA yourself, and check it against the company's own presentation for one year before using it.

Go deeper¶

tools/fi/data.py and tools/examples/fetch_indian_company.py — read the code; run it on two companies and compare the row names.
Screener.in's "Export to Excel" (free with an account) — the fastest way to a 10-year standardised history; then reconcile one year to the annual report to learn its mapping.
Any company's last five annual reports — build the 10-year P&L by hand once; you will never again trust a vendor line without checking.

← Previous: 10.1 Model architecture · Module index · Next: 10.3 Forecasting drivers →