Jewellery Peer Comps

A trading-comps tool for Titan Company and its listed Indian jewellery peers. It reads financials from Screener.in Excel exports and prices from NSE's bhavcopy (or from Screener), computes EV/EBITDA, P/E and EV/Sales for each company, and turns the peer range into an implied value per share for Titan, drawn as a football-field chart next to my own two Titan valuations.

Why comps, when I already have a valuation

My Titan valuations (titan-valuation) value the stock at ₹2,131 a share with a free cash flow (FCFF) DCF and ₹4,058 with a discounted-earnings model, against a price of ₹4,515.7 on 1 Oct 2026. Those models say what the business is worth if my forecasts are right. Comps say what the market is paying today for similar businesses. They fail in different ways: a forecast-based model is only as good as its growth and margin assumptions, while comps inherit whatever mood the market is in. Putting both on one chart shows whether the gap between my values and ₹4,515.7 is my forecasts being cautious or Titan simply trading at a premium to its peers.

Results

Prices are Screener's "Current Price" on 1 Oct 2026 (the as-of date is set in comps/config.py). Fundamentals are the latest fiscal year (FY2026, March year-end), trailing, in ₹ crore. The core peers are Kalyan, Senco, Thangamayil and P N Gadgil; Trent is shown for reference and left out of the statistics.

Company	Price (₹)	Mkt cap	EV	EV/Sales	EV/EBITDA	P/E	EBITDA margin	ROCE	Inventory days
Titan (target)	4,515.7	400,898	429,602	4.9x	51.4x	79.0x	9.5%	16.3%	178
Kalyan Jewellers	528.5	54,581	59,837	1.7x	23.4x	40.4x	7.2%	17.2%	145
Senco Gold	317.4	5,202	7,385	0.9x	7.5x	9.1x	11.7%	17.3%	229
Thangamayil (standalone)	4,861.0	15,109	15,641	1.8x	27.7x	43.0x	6.6%	22.5%	127
P N Gadgil	582.0	8,566	9,799	0.9x	15.9x	20.9x	5.9%	15.1%	127
Trent (reference)	2,580.0	137,574	139,852	7.0x	37.3x	80.0x	18.7%	25.0%	42

Core peer statistics (Titan and Trent excluded):

Multiple	p25	Median	p75	Min	Max
EV/EBITDA	13.8x	19.6x	24.4x	7.5x	27.7x
P/E	17.9x	30.7x	41.1x	9.1x	43.0x
EV/Sales	0.9x	1.3x	1.7x	0.9x	1.8x

Titan's implied value per share, applying those peer multiples to Titan's own numbers:

Method	p25	Median	p75	Min to max
EV/EBITDA	₹974	₹1,523	₹1,977	₹383 to ₹2,284
P/E	₹1,025	₹1,752	₹2,346	₹518 to ₹2,455
EV/Sales	₹585	₹964	₹1,369	₹541 to ₹1,492
My FCFF DCF		₹2,131
My earnings model		₹4,058
Price (1 Oct 2026)		₹4,515.7

On an LTM basis (--basis ltm, to Jun-26) the peer median EV/EBITDA is 18.8x, P/E 28.6x and EV/Sales 1.2x, and Titan's implied medians are ₹1,667, ₹1,857 and ₹920.

Observations

• Titan trades at 51.4x EV/EBITDA and 79.0x P/E, about 2.6x the core-peer medians (19.6x and 30.7x). Among the jewellers only Thangamayil (27.7x, 43.0x) and Kalyan (23.4x, 40.4x) come anywhere near it, and Senco (7.5x, 9.1x) is far below.
• Titan's multiples sit closer to Trent's (37.3x EV/EBITDA, 80.0x P/E) than to the jewellers'. Trent has twice Titan's EBITDA margin (18.7% vs 9.5%) and about a quarter of its inventory days (42 vs 178).
• All three peer-multiple methods imply values well below the market price (medians ₹964 to ₹1,752 against ₹4,515.7), so the market is paying a premium that these peers' multiples do not explain. My FCFF DCF (₹2,131) lands just above the top of the peer ranges, which supports it; my earnings model (₹4,058) is about 10% under the price, and the gap between my two models is mostly the cash Titan reinvests in gold inventory and stores.
• Profitability and returns do not separate Titan from the group: ROCE of 16.3% is inside the 15.1% to 22.5% peer range, and its EBITDA margin is above Kalyan, Thangamayil and P N Gadgil but below Senco. What does stand out is size (market cap about 4.8x the four core peers combined) and the highest EV/Sales (4.9x vs 0.9x to 1.8x).

Caveats

• Trailing multiples only. There are no consensus forward estimates, and the peers are growing fast (FY26 sales growth of 33% to 73%), which flatters trailing multiples for the faster growers.
• Thangamayil is standalone (no consolidated page on Screener); the other five are consolidated.
• Net debt is unadjusted. It is Screener's Borrowings less Cash & Bank, so gold loans and lease liabilities likely sit inside it and investments are ignored. Titan's borrowings are ₹30,621 cr against cash of ₹1,917 cr, about 3.4x EBITDA, so EV-based multiples are sensitive to this.
• Small sample. Four core peers make the percentiles coarse, and one peer (Senco) pulls the lower quartile down.
• One day's prices. The as-of date (1 Oct 2026) is a config setting, not read from the exports.

Full write-up: outputs/RESULTS.md (generated locally, not committed). Analysis only, not investment advice.

What's inside

Module	What it does
comps/screener.py	Reads a Screener "Export to Excel" workbook by section and row label into a dataclass
comps/bhavcopy.py	Reads close prices (series EQ) from NSE's capital-market bhavcopy, .csv or .zip
comps/metrics.py	EBITDA, EBIT, net debt, market cap, EV, multiples, growth, margins, ROCE, ROE, inventory days; fiscal year and LTM
comps/valuation.py	Peer min / 25th / median / 75th / max and the implied value per share for the target
comps/charts.py	The football-field chart (PNG)
comps/excel.py	outputs/comps.xlsx: Comps table, Implied values, Inputs & as-of dates, Notes
comps/config.py	Peer groups, my valuation, reference price and the method notes
comps/__main__.py	The command line

Quick start

git clone https://github.com/Dakshcore/jewellery-peer-comps.git
cd jewellery-peer-comps
pip install -r requirements.txt
# put the downloaded files in data/raw/screener/ (see below), then:
python -m comps --target TITAN --price-source screener
python -m comps --target TITAN --price-source bhavcopy --bhavcopy data/raw/BhavCopy_NSE_CM_0_0_0_20260925_F_0000.csv.zip
python -m unittest -v

Other options: --valuation "My FCFF DCF=2131" (my own valuation per share; repeat for several, defaults in comps/config.py), --ref-price 4515.7 and --ref-date 1-Oct-2026 (the price line), --basis ltm (multiples on the last twelve months instead of the latest fiscal year), --peers A,B,C, --reference TRENT, --include-reference (put the reference peers into the medians), --data-dir, --out-dir. Outputs go to outputs/: comps.xlsx and football_field.png.

Written for Python 3.13 with NumPy, pandas, openpyxl and Matplotlib (see requirements.txt).

Getting the data (by hand)

Nothing is scraped. Both inputs are files you download yourself, and data/raw/ is git-ignored.

1. Screener.in: for each company, open its page, click "Export to Excel", and save the file as data/raw/screener/<NSE_SYMBOL>.xlsx (TITAN.xlsx, KALYANKJIL.xlsx, SENCO.xlsx, THANGAMAYL.xlsx, PNGJL.xlsx, TRENT.xlsx). Use the consolidated page where there is one.
2. NSE bhavcopy (optional): download the capital-market bhavcopy for the day you want (BhavCopy_NSE_CM_0_0_0_YYYYMMDD_F_0000.csv.zip) and pass it with --bhavcopy. Without it, prices come from the "Current Price" in each Screener export, which is the price on the day you downloaded it.

More detail: docs/DATA_GUIDE.md. Note template: docs/NOTE_TEMPLATE.md.

The Screener layout is undocumented. The loader assumes the "Data Sheet" has section header rows (META, PROFIT & LOSS, Quarters, BALANCE SHEET, CASH FLOW, PRICE) and a "Report Date" row of dates in each time-series section, and it finds every row by its label inside its section, ignoring case and spacing. If a label is missing it stops and names it, so a layout surprise shows up as a clear error rather than a wrong number. Share counts are accepted in absolute numbers or crore: anything at or above 100,000 is divided by 10,000,000.

Peer set

Symbol	Company	Role	Screener basis
TITAN	Titan Company	Target	consolidated
KALYANKJIL	Kalyan Jewellers India	Core peer	consolidated
SENCO	Senco Gold	Core peer	consolidated
THANGAMAYL	Thangamayil Jewellery	Core peer	standalone (Screener has no consolidated page)
PNGJL	P N Gadgil Jewellers	Core peer	consolidated
TRENT	Trent	Reference: lifestyle retail, shown separately, not in the medians	consolidated

All six are listed on NSE with these symbols and have a March fiscal year-end. Edit comps/config.py to change the set.

Method

• EBITDA = Profit before tax + Interest + Depreciation − Other income. EBIT = EBITDA − Depreciation.
• Net debt = Borrowings − Cash & Bank, from the latest fiscal-year balance sheet. Market cap = price × shares in crore. EV = market cap + net debt. Money is in ₹ crore.
• Multiples: EV/Sales, EV/EBITDA and P/E (market cap / net profit), on the latest fiscal year or, with --basis ltm, on the last twelve months. A multiple with a zero or negative denominator is shown as n/a rather than a misleading number.
• LTM = the sum of the last 4 consecutive quarters in Screener's Quarters block, for Sales, EBITDA (same formula) and Net profit. If fewer than 4 consecutive quarters are available, LTM is n/a. LTM multiples still use the latest fiscal-year net debt.
• Also computed: revenue and EBITDA growth (1 year and 3-year CAGR), EBITDA and net margin, ROCE = EBIT / (Equity share capital + Reserves + Borrowings), ROE = Net profit / (Equity share capital + Reserves), and inventory days = Inventory / Sales × 365, which matters for jewellers because gold stock is most of their balance sheet.
• Peer tiers: the core jewellers set the medians. Trent is shown separately and left out of the statistics unless --include-reference is passed.
• Peer statistics: minimum, 25th percentile, median, 75th percentile and maximum, using linear interpolation.
• Implied value per share: for an EV multiple, (multiple × Titan's metric − Titan's net debt) / shares. For P/E, multiple × Titan's net profit / shares. The chart shows the 25th to 75th percentile as a dark bar, the minimum to maximum as a light bar, each of my own valuations as a diamond and the reference price as a dashed line.

Limitations

• Trailing multiples only. There is no free source of consensus estimates, so nothing here is forward-looking. A fast-growing company looks expensive on trailing numbers.
• Leases and gold loans sit in debt. Screener's Borrowings may include lease liabilities, and jewellers can carry gold metal loans in borrowings. Net debt is not adjusted for either, so EV is not perfectly comparable across companies. Cash & Bank excludes investments.
• Fiscal-year timing. The balance sheet is at the last fiscal year-end but the price is today's. The LTM basis reduces the mismatch for earnings, not for net debt.
• A small peer set. Four core peers make the percentiles coarse, and the peers are smaller and differently run from Titan (Titan also has watches, eyecare and other businesses with different economics). One outlier moves the median.
• Screener data is taken as given. Standalone versus consolidated bases differ across companies (Thangamayil is standalone), and exceptional items are not stripped out of profit.
• The Screener layout is undocumented. The tests use synthetic workbooks built from my understanding of it; the loader has also been run on real exports.
• Not investment advice.

Tests

python -m unittest -v runs 55 tests, all on synthetic data built in a temporary folder (no real data is committed):

• Screener loader: annual and quarterly series and META values, section-aware lookup of repeated labels, case/spacing/colon-insensitive labels, clear errors that name a missing row, section, sheet or file, optional rows, and share-count units.
• Bhavcopy: series EQ only, whitespace and case, blank closes, missing columns, and reading .csv and .zip files.
• Metrics: EBITDA, EBIT, net debt, market cap and EV on hand-computed numbers, multiples, margins, ROCE, ROE, inventory days, growth and CAGR, LTM from four consecutive quarters, and None (not a crash) for negative EBITDA, negative profit, zero sales, zero capital and a missing price.
• Valuation: percentiles (odd, even, with missing values, empty), implied values per share for each multiple on both bases, and skipping a multiple when the target's metric is not positive.
• Chart and CLI: the PNG is written, and the command line runs end to end on synthetic workbooks with Screener and bhavcopy prices, writes comps.xlsx and the PNG, keeps Trent out of the medians unless asked, skips a missing peer with a warning and fails clearly on a missing target.

Licence

MIT. See LICENSE.