NSE Relative-Strength — Sectors & Industries

Ranks all four NSE classification levels — Macro, Sector, Industry, Basic Industry — equally weighted, by relative strength against the NIFTY 500, with a McClellan-style Market Breadth oscillator.

100% free data: Yahoo Finance for prices, NSE for the classification. No API keys, no paid subscriptions.

What it produces

A website (FastAPI backend + vanilla JS frontend, served on localhost:8000) with:

• Four tabs (Macro / Sector / Industry / Basic Industry), each an equal-weight RS ranking of its groups; expand any group to see its constituent stocks.
• Per-stock: RS sparkline, RS_STS%, LTP, EMA20/50/100/150/200 flags, the RS line vs its 21-day EMA (above/below), % off 52-week high, ADR%.
• Filters: search, price-above-EMA, %-off-high, ADR, RS ≥ N%, RS above EMA21 — combined with a TradingView watchlist export.
• Market Breadth oscillator (advancers − decliners, RANA-based).

Method

• Daily closes per stock from Yahoo (SYMBOL.NS); benchmark = NIFTY 500 (^CRSLDX).
• Equal-weight index per group: each stock base-100 on day 0 of the window, averaged across constituents each day.
• RS[t] = group_index[t] / benchmark_norm[t] (both base-100).
• RS_STS% = PERCENTRANK.INC(RS_series, latest RS) over the 26-day window.
• Stocks without a close every window day (suspensions, halts, new listings) are excluded from the view — the window count is reported in the UI.

Data sources

• Universe + classification: nse_stock_master.csv — all ~2,400 NSE stocks with their 4-level industry classification, built directly from NSE.
• Prices: Yahoo Finance chart API (free, no key). The cache (.yh_price_cache.json) grows over time: the first run pulls ~2y per symbol, later runs only top up the last month and append, so daily runs are fast.
• Yahoo occasionally publishes null OHLC for index symbols on days stocks traded fine — such benchmark gaps are filled from Investing.com (real index close), with an equal-weight synthetic from stock returns as last resort.

Scripts

Script	Purpose
build_rs.py	Fetch Yahoo prices, append history, rebuild rs_data.json + rs_view.html.
refresh_classification.py	When NSE lists new stocks: fetch their 4-level classification (headless browser) and append to the master.
run_daily.sh	Wrapper that runs build_rs.py and logs to run.log.
run_server.sh	Serve the website (FastAPI backend + frontend) on localhost:8000.

Daily use

bash run_daily.sh     # 1. fetch today's prices, rebuild data (~few min)
bash run_server.sh    # 2. start the website
# 3. open http://localhost:8000

Or, with the server running, click "Update Prices" / "Refresh Stocks" in the website header.

Build options

python3 build_rs.py               # full fetch + rebuild (default)
python3 build_rs.py --refresh     # force full ~2y re-fetch of every stock
python3 build_rs.py --html-only   # rebuild from cache only (no fetching)
python3 build_rs.py --fast        # incremental top-up (what "Update Prices" uses)
python3 refresh_classification.py --dry-run   # list new symbols, no browser
python3 refresh_classification.py --headful  # show the browser window

Tests

.venv/bin/python -m pytest tests/ -q

Covers the benchmark gap-fill fallback chain (Yahoo index gaps → Investing.com scrape → synthetic equal-weight), Yahoo null-OHLC handling, RS maths, the RS-vs-EMA21 flag, and integrity of the generated artifacts (RS window alignment, market-breadth continuity, window-coverage consistency).

Setup (first time)

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
playwright install chromium       # only needed for refresh_classification.py
python3 build_rs.py               # initial build

Notes

• refresh_classification.py needs a real browser because NSE's classification API sits behind Akamai. Use the full Chromium build (channel="chromium") — the headless-shell is fingerprint-detected and connection-reset by Akamai.
• NSE geo-blocks non-Indian IPs (403 on every endpoint). Run refresh_classification.py from an Indian IP / VPN; the equity list falls back to a browser-session fetch if plain HTTP is blocked. Yahoo-based steps (prices, RS, breadth) work from anywhere.
• Price caches, rs_data.json and rs_view.html are generated and git-ignored.