IROHA documentation

IROHA is OpenKuber's causal engine for Indian equities. For any large NSE price move it reports what the model expected, the market and sector parts, and a company-specific part matched to dated events, keeping only causes that pass evidence tests. These docs cover the concepts, coverage, data and limits. The API reference is provided to pilot customers.

How it works

1. Detect large moves: windows where a stock moves well beyond what its market and sector explain.
2. Decompose each move into model call, market, sector and company-specific parts.
3. Match the company-specific part to filings, results, corporate actions and news dated inside the window.
4. Test each candidate cause; drop those that fail and report the rest as unexplained.
5. Run the same model forward for scenarios and stress tests.

Read the full walk-through on the IROHA page and a worked example on RAYMOND.

Glossary

Attribution

Splitting a price move into the parts explained by different causes.

Model call (expected move)

What the causal model expected the stock to do over a window, before the window began.

Market component

The part of a move that matches what the index did over the same window.

Sector component

The part of a move that matches what the stock’s sector did, beyond the market.

Company-specific component

What is left after the expected, market and sector parts. IROHA matches it to dated company events.

Evidence test

A statistical check a cause must pass before IROHA shows it: a pooled test across many stocks, placebo dates and intraday timing.

Placebo date

A date with no event, used to check that an apparent effect is not just noise.

Unexplained

The part of a move no tested cause accounts for. IROHA reports it rather than guessing.

Scenario (what-if)

A forward range for each stock under a chosen market, sector or event shock.

Coverage and data

• API: the 14 Bank Nifty constituents and the Bank Nifty index.
• Pilots: your own universe of NSE-listed stocks.
• Evidence: NSE exchange filings since 2012, news, macro releases and flows. The pooled test spans more than 1,400 NSE symbols.
• Delivery: reports during pilots and the API (see Developers).

Limits

• IROHA explains past moves; it does not forecast prices or rate stocks.
• Attribution figures are model estimates with uncertainty.
• Moves with no cause that passes the tests are reported as unexplained.

How IROHA compares

IROHA works alongside terminals, document search and fundamentals databases. Each comparison says what each product is for.

• OpenKuber vs AlphaSense (AI search over documents)
• OpenKuber vs Trendlyne (Indian equity research platform)
• OpenKuber vs ACE Equity Nxt (Indian fundamentals database)
• OpenKuber vs Capitaline (Indian fundamentals database)
• OpenKuber vs causaLens (Causal AI platform)
• OpenKuber vs Bloomberg Terminal (Global data terminal)
• OpenKuber vs LSEG Workspace (Global data terminal)
• OpenKuber vs Tijori Finance (Indian company data)
• OpenKuber vs CMIE Prowess (Indian company database)