IROHA answers one question for any large NSE price move: what caused it. It first estimates what the stock would have done given the market and its sector, then attributes the rest to macro releases, flows and company events, and keeps only causes that pass statistical evidence tests. The output is a dated, sourced explanation a research desk can check and reuse.
The three layers
Expected move. What the causal model expected the stock to do on that day, from its own history and the state of the market.
Market and sector. How much of the move matches what the index and the stock's sector did over the same window.
Company-specific shock. What is left, matched to dated events: exchange filings, results, corporate actions and news.
Evidence tests: how a cause is accepted or rejected
A cause is shown only if it survives tests, not because it is plausible. IROHA uses a pooled test across the 500 most liquid NSE stocks each day, 1,423 different stocks and their filings since 2012, to check whether a type of event moves prices at all; a mechanism test to check that news pushes the price the way it says; placebo dates to check that the effect does not appear on days with no event; and, for the graph as a whole, a local Markov check and a residual check. Causes that fail are dropped, and the remainder is reported as unexplained.
Worked example: RAYMOND
From 21 Aug – 7 Oct 2026 RAYMOND rose +52.8%. The market contributed −0.5 points and textile peers −1.9; the remaining +53.1 points trace to orders won by its aerospace subsidiary, where policy news and company news pass every causal test. The full RAYMOND study covers all 52 big moves since 2015.
RAYMOND · 52 big moves since 2015
call leaned right leaned wrong41/52
151617181920212223242526
#52 · 21 Aug – 7 Oct 2026 · Company news
+52.8%move
Raymond Limited’s Aerospace Subsidiary Wins Multi-Programme Aerospace Orders from a Leading Indian Aerospace and Defence Major
Passed every causal test: policy news, news.
Where the move came from · pts
Model call leaned the right way+2.0
Market −0.5
Textiles peers −1.9
Company event +53.1
Call + market + sector + company = +52.8 pts. Data as of 27 October 2026.
From explanation to action
The same causal model runs forward. Set a market move, a sector move or an event, and IROHA returns a range for each holding. Research desks use these ranges for stress tests; portfolio managers use them to size positions after a shock.
Coverage and data sources
Universe: NSE-listed equities. The API covers the 14 Bank Nifty constituents and the index today; pilots run on your own universe.
Evidence: NSE exchange filings since 2012, news, macro releases and flows.
Delivery: web reports for pilot customers and an API.
Limits: what IROHA does not claim
It explains past moves. It is not a price forecast and gives no buy, sell or hold view.
Some moves have no cause that passes the tests. IROHA reports them as unexplained rather than inventing one.
Attribution percentages are model estimates with uncertainty, not accounting identities.
Explore the causal map
RAYMOND’s causal graph as a map: each territory is a variable IROHA uses, the small cells are the data behind it, and the arrows are causal links. Hover a territory for its evidence. The live report adds forward shocks on the same map.
World modelMarketIndustry peersMacroCompany newsPolicy newsResultsNSE filingsconfirmedcandidate
Large cells are the variables IROHA uses for RAYMOND; the small cells around each are the data behind it. Hover a territory for its evidence and links; arrows are causal links, solid when confirmed. 1,267 sessions, 2021-09-06 to 2026-10-14.
Questions
How is causal attribution different from correlation?
Correlation shows two series moved together. Causal attribution estimates how much of a move a specific driver produced after controlling for the market and sector, and tests whether that effect holds.
Which data does IROHA use?
Prices, NSE exchange filings, news, macro series and market-structure data such as flows.
Does IROHA predict prices?
No. IROHA explains past moves and builds scenario ranges. It does not issue price targets or buy and sell calls.
