Causal architecture
Directional graphs, not correlation heatmaps. Every answer surfaces the chain behind a market move — causality and correlation as distinct outputs.
The market moves,you should understand why
Ask any question about any market. Plutonal tests the thesis across causal, statistical, and quantitative models — and shows the evidence behind the answer.
Plutonal accepts market research questions such as: Which publicly traded companies are best positioned to benefit from the global AI data-center buildout?
Stress-test the thesis before you trust it.
Compare bull, base, and bear cases, the conditions each requires, the companies most exposed, and the signals that would confirm or invalidate the result.
Trace the chain behind the market move.
Follow an event through mechanisms and companies, with confidence at each link and weak points made explicit.
See where the models agree, and where they do not.
Read the verdict, confidence, and agreement across the model families behind the answer.
Put every exposure on the map.
Place companies, infrastructure, and market exposure in geographic context, then follow the relationships between them.
Scan every relevant company at a glance.
Compare conviction, price, regime, risk, and quantitative signals across the research universe.
Compare the whole universe, side by side.
Sort the research universe by verdict, price action, momentum, and liquidity signals.
Turn the comparison into a verdict.
Read the cross-company conclusion, key findings, conviction levels, and highest-risk names in one analytical summary.
See how the leading names differ by factor exposure.
Compare RSI momentum, trend strength, volatility, and volume signals across the dynamically selected companies.
See which signals confirm the thesis—and which contradict it.
Review company-level confirmations and contradictions, then trace the supporting causal chain beneath them.
Statistically derived reference levels. Validated against historical analogues. Research only. Not a buy or sell recommendation.
Institutional-grade data.Institutional-grade methods.Retail access.
3,000+ equities across 40 global markets. 120+ data sources. Every number sourced, every series traceable.
NYSE
Nasdaq
LSE
Cboe
Bloomberg
Reuters
FactSet
S&P Capital IQ
Morningstar
SEC EDGAR
FCA
ESMA
FRED
BLS
BEA
OECD
IMF
World Bank
BIS
Three stepsfrom questionto evidence.
Built for analysts working outside the institutional perimeter.
01 Ask
A question, in plain English.
Scope can be a single ticker, a sector, a macro linkage, or a structural thesis. If you can write it as a question, Plutonal can answer it.
02 Compute
80+ models run in parallel.
Granger causality, vector autoregression, GARCH, factor decomposition, regime-switching — selected by the question and validated against historical analogues.
03 Receive
A brief with evidence and confidence.
Verdict, confidence interval, causal chain with dated lags, full evidence trail, and sourced data rows — exportable and traceable end-to-end.
What the terminal does.
The proprietary engine orchestrates six product surfaces. Each does one thing, rigorously.
Quantitative engine
80+ models running simultaneously. Granger causality, VAR, GARCH, factor decomposition, regime-switching. Bias factors surfaced alongside every verdict.
Visual research layer
Lens renders geographic and structural questions as annotated maps and ranked panels. Charts carry confidence cones, pattern marks, and regime overlays.
Context & discourse
Sentiment rendered as a time series, not a single score. News, filings, and social signal from X (Twitter) — all event-annotated and dated.
Watchlist
Alerts fire on regime shifts and statistical thresholds — not on every price movement. Every trigger traces to one of 120+ verified data sources.
Choose how you research.
Every paid plan includes the complete research engine. Select access based on how extensively you expect to use it.
The shape of a Plutonal brief.
Statistically derived reference levels. Validated against historical analogues. Research only. Not a buy or sell recommendation.
Asian Paints · operating margin, forecast vs. realised
RealisedForecast90% CI
NVDA · last 5-day move, decomposed
MarketSector / styleIdiosyncratic
S&P 500 · current regime vs. 2018 analogue
Realised vol2018 analogueCurrent state
Brent crude → Indian paint manufacturers, causal transmission
Brent · M1Paint mfrs · cost indexGranger lag ≈ 8w
The statistical workis named, not hidden.
The orchestration layer is proprietary. The techniques themselves are standard. Naming them, with specification, is the point.
These are the techniques institutional quants use. Plutonal runs them on your questions. Each method answers a different question about market behaviour — whether one thing causes another, how several things move together, when the market’s rules have shifted, and why a specific thing moved.
Granger causality
Tests whether the past of one series improves the forecast of another beyond its own past. The difference between “these move together” and “this one moves first.”
Significance threshold p < 0.01 · BIC-selected lag order · bootstrap-validated at 5,000 draws
Vector autoregression
A small system of equations in which every variable depends on every other variable’s recent history. Captures feedback, not just one-way effects.
Minimum sector R² benchmark 0.55 · Ljung-Box residual test at 20 lags · up to 8-variable systems
GARCH volatility
Volatility as state-dependent and clustered in time rather than constant. The same price move is a different event in a calm regime than in a turbulent one.
GARCH(1,1) baseline · EGARCH for asymmetric shocks · Student-t innovations where normality is rejected
Factor decomposition
A move is broken into exposures — sector, style, macro, idiosyncratic — so a name-specific signal can be separated from what the whole market is doing to every name.
Fama-French 5-factor + momentum + liquidity · rolling 252-day window · shrinkage-regularised
Bayesian forecasting
Projections of future values rendered with credible intervals, not point estimates. When the model is uncertain, the uncertainty is visible. When the model is confident, the interval narrows.
Bayesian posterior sampling · 10,000-draw Monte Carlo · 80% and 95% credible intervals reported on every projection
Sentiment Analysis
Structured analysis of financial text — filings, transcripts, news, commentary — using the same techniques used inside institutional research. What’s being said, by whom, in what tone, and how that’s changed.
Named-entity recognition · FinBERT-calibrated sentiment · topic coherence tested on holdouts · source-weighted aggregation
Regime-switching is overlaid on every method: parameters are re-estimated per detected state, so a relationship that holds in calm markets but fails in crises is reported with both regimes dated. All methods are walk-forward validated on rolling out-of-sample windows — the model is tested on data it has not seen before the result is reported.
There is no reason, in 2026, thatquantitative research should stillbe an institutional privilege.
Get started
Ask your first question in under a minute.
Sign in and put a real research question to the engine. Every answer comes back with the evidence and the confidence behind it.