AI & Methodology
We believe investors deserve to understand not just what the AI concludes, but how it reached that conclusion. Every model, every score, every confidence level — explained in plain language.
Knowledge Graph Nodes
50K+
Events, companies, sectors
Historical Data
14 yrs
2010–2024 Indian markets
Relationship Types
6
Causal edge categories
Cascade Depth
4 levels
Upstream to downstream
Feature Deep-Dive
MarketRipple's AI search goes far beyond keyword matching. It understands the financial intent behind your question and reasons through the answer step by step.
NLP pipeline performs entity extraction (companies, sectors, events, time periods), intent classification (comparison, trend, causation, prediction), and query decomposition into sub-questions.
Query is embedded into vector space and matched against MarketRipple's knowledge graph containing 50,000+ event-company-sector relationships. Semantic similarity retrieves top-K relevant context passages.
For complex financial queries, the model chains reasoning steps — first identifying affected sectors, then companies within those sectors, then evaluating historical analogues before forming a conclusion.
Supporting evidence is assembled from event timelines, corporate filings, regulatory announcements, and quantitative data. Each evidence piece is ranked by recency, source credibility, and relevance.
Final response is generated with inline citations linking back to source events, data points, and companies — so you can verify every claim independently.
Example: Query Processing Walkthrough
Query
"Which sectors benefit from a weak rupee?"
Entity Extraction
Entity: INR (Indian Rupee) · Relationship: weakness/depreciation · Outcome: sector beneficiaries
Intent Classification
Intent: Causal analysis → sector screening → beneficiary identification
Graph Traversal
INR depreciation node → outgoing edges to: IT Services (positive, conf 88%), Pharmaceuticals (positive, conf 82%), OMCs (negative, conf 79%), Airlines (negative, conf 81%)
Evidence Retrieval
Pulled: 14-year INR/sector performance correlation, Infosys FY24 earnings call (INR hedge ratio), historical ATF-INR relationship data
Response Generated
Primary beneficiaries: IT (dollar revenue, INR costs), Pharma exporters (US market revenue). Primary losers: Airlines (USD fuel), OMCs (crude imports). With citations.
Core Algorithm
The Ripple Engine is MarketRipple's proprietary system for tracing how a single market event propagates through the Indian economy — sector by sector, company by company.
MarketRipple maintains a live knowledge graph with nodes representing events, companies, sectors, commodities, currencies, and macroeconomic indicators. Edges represent causal relationships, each weighted by a confidence score derived from historical co-movement analysis and domain expertise.
Trained on 14 years of Indian market data (NSE/BSE price movements, earnings, macro releases). The model identifies which event-outcome pairs have statistically significant and economically meaningful relationships, filtering spurious correlations.
Bayesian inference combining: Source Reliability (official vs. secondary) × Historical Accuracy (backtested prediction accuracy) × Analyst Consensus (cross-model agreement) × Recency Weight (recent data weighted higher). Updated continuously as new evidence arrives.
Breadth-first propagation through the dependency graph up to 4 levels deep. Each level attenuates confidence by the edge weight, preventing overconfident downstream claims. Parallel branches are tracked independently to surface both positive and negative effects on the same company.
New events trigger automatic graph re-evaluation within minutes. The confidence of existing ripple chains is updated as confirming or disconfirming evidence accumulates. Stale predictions are flagged with a decay indicator when underlying conditions change.
Scoring Model
Every opportunity on the Radar is assigned a composite score (0–100) built from five independently weighted factors. Here is exactly how the maths works.
Opportunity Score Formula
Score = Σ (factor_score × weight) — normalised to 0–100
Event Impact Score
Weight: 30%
Magnitude (scale of financial impact) × Breadth (number of sectors and companies affected) × Duration (transient vs. structural change). Scored 0–10, then normalised to 0–30 contribution.
AI Confidence Score
Weight: 25%
Aggregate confidence across the full ripple chain. Higher confidence = more actionable signal. Computed as geometric mean of confidence at each causal step, weighted by depth.
Sector Momentum
Weight: 20%
Current price momentum, earnings revision trend, and FII/DII flow direction for the affected sector. Trailing 30-day momentum with mean-reversion adjustment for over-extended moves.
Historical Precedent
Weight: 15%
Outcome of similar past events — how the sector and specific companies actually performed. Backtested on 14 years of Indian market data (2010–2024). Adjusted for regime changes.
Time Sensitivity
Weight: 10%
Urgency of the opportunity (event recency, information decay rate) and reversibility (is this a one-time event or a structural change?). Higher score for time-sensitive, irreversible shifts.
Narrative Engine
MarketRipple's story generation pipeline transforms clusters of related events into coherent investment theses — with evidence, timeline, and risk factors.
Related events are grouped using semantic similarity and temporal proximity. A cluster of events about government infrastructure spending, RBI rate decisions, and NBFC credit growth would be grouped into a "Credit-Driven Infrastructure" story candidate.
The core investment hypothesis is extracted from the cluster: the fundamental driver, expected timeframe, and primary beneficiary profile. This becomes the story's thesis statement.
Supporting data is linked: affected companies (with rationale), quantitative data points (earnings, revenue exposure), historical analogues from prior market cycles, and expert viewpoints from analyst reports.
A coherent investment narrative is generated with a clear timeline (what has happened, what is happening, what is likely to happen next), risk factors, and actionable conclusions for different investor time horizons.
Automated fact-checking cross-references all named companies against their actual sector classifications, verifies that stated financial metrics are within plausible ranges, and flags internal contradictions before publishing.
Scoring Detail
Confidence is not a gut feeling — it is a computed estimate that accounts for evidence quality, recency, corroboration, and domain-specific model accuracy.
Source Credibility
Contributes 30% of total confidence
Official regulatory sources (RBI, SEBI, NSE, BSE, Ministry filings) receive maximum credibility weight. Verified financial journalism is weighted at 70–80%. Social media and unverified sources receive minimal weight and are flagged as low-credibility inputs.
Recency Weighting
Contributes 20% of total confidence
Information decays in relevance. Events from the past 7 days carry full weight; 7–30 days carry 80% weight; 30–90 days carry 60%. For structural relationships (sector-commodity linkages), historical data retains higher relevance.
Source Corroboration
Contributes 25% of total confidence
When 3+ independent sources confirm the same fact or relationship, confidence receives a significant boost. A single-source claim carries baseline confidence; five or more independent confirmations pushes a signal to Very High confidence tier.
Historical Accuracy
Contributes 15% of total confidence
Each event-outcome relationship is backtested on 14 years of data. The model's own historical prediction accuracy for that specific relationship type adjusts the base confidence. A relationship the model has predicted correctly 90% of the time earns higher confidence.
Domain Expertise
Contributes 10% of total confidence
Sector-specific sub-models for Banking, IT, Pharma, Commodities, and Infrastructure have been calibrated on sector-expert knowledge. Cross-sector effects use the general model; within-sector effects benefit from the specialised sub-model's higher precision.
Honest Disclosure
Transparency requires honesty about what AI can and cannot do. These are the genuine limitations of MarketRipple's analytical systems.
MarketRipple analyses probabilities and historical patterns. It cannot predict market movements with certainty. All insights are probabilistic — treat high-confidence signals as strong hypotheses, not facts.
Analysis is based exclusively on publicly available information — regulatory filings, market data, news, and official announcements. MarketRipple does not have access to private or insider information.
Historical patterns are the foundation of confidence scoring. In unprecedented events (novel pandemics, first-of-kind policy changes), confidence scores will be lower and scenario analysis wider — as they should be.
Confidence percentages represent the AI model's calibrated uncertainty — not statistical guarantees. A 75% confidence signal will be wrong roughly 25% of the time. Size positions accordingly.
AI models have training cutoff dates. For very recent structural changes in market microstructure or new regulations, historical pattern recognition may be less reliable until the model incorporates new data.
Analysis quality depends on data source refresh rates. Breaking news may take 3–10 minutes to fully propagate through the ripple engine. For intraday trading decisions, always verify against primary sources.
Human Judgment
MarketRipple is designed to augment human investment judgment, not replace it. AI excels at processing large volumes of structured information, identifying historical patterns, and surfacing non-obvious connections across thousands of data points simultaneously.
Human investors bring irreplaceable judgment: qualitative assessment of management quality, reading between the lines of regulatory intent, contrarian thinking that departs from consensus, and the lived experience of navigating market cycles.
We recommend using MarketRipple to generate and stress-test hypotheses, then verifying your conclusions against primary sources (BSE/NSE filings, RBI releases, company annual reports) before making any investment decision. MarketRipple is a powerful second opinion — your first opinion should always be your own informed judgment.
MarketRipple does not provide personalised investment advice. All analysis is for informational purposes only. Past patterns do not guarantee future outcomes.
Continue Exploring
Explore how MarketRipple reasons through real market events — and discover where all the underlying data comes from.