RAYMOND · shock story · NSE
What moved RAYMOND, and why.
Loading…
RAYMOND: price and shocks
Headlines per week
The shocks, one by one
The evidence
every cause is tested before it is shownCausal check · RAYMOND
Loading…
Pooled test · ~500 stocks
Loading…
RAYMOND · next 21 sessions vs median–
What could move RAYMOND next?
Next 21 sessions vs the median NSE stock · range from RAYMOND's own past forecast errors ·
Choose a scenario
Expected · next 21 sessions
–
Chance it beats the median stock
Likely range · 10th–90th pct
What changes the forecast
pts, largest firstHow it works
One causal model for every shock.
A self-supervised market model learns how ~1,500 NSE stocks behave together, every day since 2005. A news and filings engine reads what the market reads. A causal layer tests which of those events actually moved the price, so every shock gets a cause you can check.
How to use this page
about 5 minutes- 01Know the stock
This page covers RAYMOND (NSE). Today it makes aerospace & defence parts and precision auto components; its clothing and real-estate businesses now list as separate companies.
- 02See what moved it
The Story tab shows every big price move since 2015, what caused it, and the evidence behind each cause.
- 03Try “what if”
The What if tab shows how the stock might do next month if the market, oil, the rupee, its peers or its calendar change.
Reading the Story tab
- The four numbers at the top: how many big moves we found, how many have a named cause, how many pass the full causal test, and the biggest one. Click a number to filter the cards.
- The price chart: the share price since 2015; zoom with All / 5Y / 3Y / 1Y. Each numbered square marks a big move, a month in which the stock rose or fell far more than it normally does. The colour shows the cause: company news company filing whole industry moved whole market moved corporate event. Click a colour chip to show or hide that kind.
- Headlines per week: the small bars under the chart. A burst of news next to a big move points to the news behind it.
- The cards: one per big move, split into four measured parts: the model's call the day before, the market move and the sector move during the window, and the company event behind the rest, named by its cause. A green ✓ means the model had already leaned the right way. Click a card, then use ← / → to step through the moves.
- The evidence: the causal check tests RAYMOND's own causes; the pooled test checks every type of event across ~500 stocks, so a cause is only shown as confirmed when it holds across the market.
Using the What-if tab
- Scenarios are colour-coded by what they change: market economy industry & peers company. Each shows its estimated effect before you pick it; click one to apply it (click again to clear), then fine-tune with the sliders.
- Market & macro (left): how much NIFTY, small caps, the rupee, oil, foreign flows or market fear change over the next month. Each slider shows its estimated effect on RAYMOND, and the typical buttons set a normal-sized move.
- Company & peers (right): what if RAYMOND, its industry, a group of similar stocks or any single stock closes higher or lower today, or its results date moves.
- Reading the answer: the big number is how much better or worse RAYMOND might do than a typical NSE stock over the next month. It is not a price target. The ring is the chance it beats a typical stock; the bar is the realistic range, based on how far off past forecasts for this stock have been.
Live model vs saved data
- When the page is connected to the Shock Engine server, everything is computed live and the status at the top says ready.
- When it isn't (for example a shared copy), it shows RAYMOND's saved data and the status says snapshot · date; What-if answers are then a quick estimate from the saved sensitivities.
Words used on this page
- Big move
- A month in which the stock rose or fell far more than it usually does, compared with a typical NSE stock.
- Model's call
- What the market model expected for the next month, the day before the move, from the stock's own trend and similar stocks.
- Market move / sector move
- What the whole market and the stock's industry peers actually did during the move, measured, not guessed.
- Company event
- The rest of the move: the order win, results, demerger or other event specific to the company, named from its headlines and NSE filings.
- Typical NSE stock
- The middle stock of ~500 liquid NSE stocks. Comparing against it removes moves that hit every stock at once.
- Next month
- The next 21 trading days.
- Confirmed / candidate
- Confirmed: the cause passes every test, including placebo dates and the pooled test across ~500 stocks. Candidate: significant, still being tested.
- Market model
- A self-supervised model that learns how ~1,500 NSE stocks move together.
- Causal model
- OpenKuber's layer that tests which events actually moved a price: causal graphs, counterfactuals and pooled tests.
What goes in, and what comes out
Data
Prices & volumes2,044 NSE equities, daily since 2005; 5-minute bars for 2,100 instruments
Exchange filings1.48M company announcements since 2012, every listed stock, timestamped to the second
Newsheadlines naming listed companies
MacroBrent, USD/INR, foreign (FPI) flows, daily
Market structuredelivery %, results calendar, corporate actions
→
Models
Market modela cross-sectional transformer over ~1,500 stocks a day, trained self-supervised; ranks the next 21 sessions
Scenario modelthe same model told a NIFTY / VIX / small-cap path, so market scenarios flow through every stock
News & filings enginecompany matching, event types, financial tone, relevance; filing types from the exchange's own subjects
Causal graphthree discovery engines vote on what drives the stock; counterfactuals measure each driver's share of a shock
Pooled causal modeldouble machine learning over ~500 stocks: which event types move prices, and by how much
→
Outputs
Shock storyevery shock found, split into what the model expected and the surprise, and the surprise named from news and filings
What-if enginemarket, macro, news, peer and calendar knobs → a forecast with a realistic range
Evidenceeach cause tagged confirmed or candidate, with the tests behind it
How every claim is tested
- Pooled across ~500 stocks. An event type counts as a cause only if it moves prices across the market, not in one lucky stock: 1.8M stock-days since 2012.
- Placebo dates. The same events moved to random dates must show nothing, and they don't.
- Negative control. Routine filings (trading-window notices, newspaper ads) should have no effect, and they don't: −0.006%.
- Intraday timing. For filings made during market hours, the 30 minutes after the filing are compared with the same time on other days; order wins add +0.60% within half an hour.
- Stable over time. Each effect keeps its sign in 2012–16, 2017–20 and 2021–26.
- Clean news. Headlines that only report a price move are never used as a cause.
- Forecast skill. Across 2,853 days the model's call for RAYMOND lines up with the next month's return (rank correlation 0.21), and it leaned the right way before most of the big moves.
- Realistic ranges. The forecast range comes from RAYMOND's own past forecast errors.
What past headlines did next
Return from the day before the headline to 21 sessions after it, vs the average single-company headline
| Event | Headlines | Avg | Beat median |
|---|
What makes it stronger
the roadmapMore data
New signals the model can learn from on day one
- Derivatives positioning · open interest, implied volatility, put/call and rollover for every F&O stock since 2005
- Insider & promoter trades and quarterly shareholding, since 2016
- Real-time news with second-level timestamps for all 2,000+ listed stocks
- Earnings calls & filing text read by an LLM extractor for guidance, orders and risks
- Global context · commodity, currency and peer markets, supply-chain links
More compute
GPU scale turns into accuracy and tighter ranges
- Larger market models trained on multi-GPU clusters (A100 / H100)
- Seed & architecture ensembles for calibrated, tighter forecast ranges
- Intraday model on 5-minute bars for 2,100 instruments
- Causal discovery at full scale · thousands of drivers per stock, run in parallel
- Daily retraining so the model tracks every new regime
Infrastructure
From research engine to live product
- Streaming pipeline · filings and news scored seconds after they land
- Live shock alerts with the cause attached, minutes after the move
- Nightly causal refresh for every stock, fully automated
- Audit trail · every causal claim versioned with its data and tests
- API for brokers, wealth platforms and risk desks
Coverage & product
The same engine, many more surfaces
- Every NSE & BSE stock, then other emerging markets
- Portfolio view · attribute a portfolio's shocks to their causes
- Stress tests · run any scenario across a whole book
- Sector & theme maps · who moves with whom, and why
- Explainable alerts for PMs, analysts and compliance