Quant research infrastructure
BTC Futures Research Assistant
Research infrastructure for immutable volatility forecasts, factual forward outcome validation, and operational integrity monitoring.
Interactive research exhibit
Synthetic Volatility Shock Lab
Compare how four volatility specifications transform the same deterministic synthetic event into normalized next-hour variance evidence.
No fitted private coefficients, production forecasts, model ranking, or trading authority.
Synthetic scenario
Scenario design
A single negative standardized return shock arrives at hour zero after a stable pre-event variance state.
- Event marker
- Negative synthetic shock at relative hour 0
- Comparison goal
- Highlights the stronger conceptual response of signed-shock and threshold-asymmetric specifications.
Deterministic conceptual illustration. Values are normalized for visual comparison and are not fitted forecasts, empirical results, or performance rankings.
Model output is retained as research evidence only. Evidence does not authorize execution.
| Relative hour | Normalized next-hour variance forecast |
|---|---|
| -6 | 0.27 |
| -5 | 0.28 |
| -4 | 0.28 |
| -3 | 0.29 |
| -2 | 0.29 |
| -1 | 0.30 |
| 0 | 0.96 |
| +1 | 0.84 |
| +2 | 0.75 |
| +3 | 0.67 |
| +4 | 0.60 |
| +5 | 0.54 |
| +6 | 0.49 |
| +7 | 0.45 |
| +8 | 0.41 |
| +9 | 0.38 |
| +10 | 0.36 |
| +11 | 0.34 |
| +12 | 0.33 |
Research-only
No live or paper trading approval
No exchange, broker, or order-routing connection
No entry, short-permission, or strategy-approval role
Model outputs remain descriptive research evidence. They are never presented as strategy approval.
Research architecture
Asset context to immutable evidence
A compact system map. Provenance stays attached to each asset context.
Asset layer
BTCUSDT Futures
E-mini Nasdaq-100 Futures (NQ)
E-mini S&P 500 Futures (ES)
Gold Futures (GC)
Model layer
Evidence layer
Purpose and contribution
Evidence infrastructure, not a strategy layer
Project Overview
The system preserves factual research history so forecasts, realized outcomes, and operational checks can be reproduced and audited without rewriting prior evidence.
Its role ends at evidence generation. Strategy interpretation and execution sit outside the system boundary.
- Not a market-direction predictor or entry-signal generator
- Records model forecasts, realized outcomes, and operational integrity separately
- Makes research evidence reproducible, auditable, and resistant to retrospective rewriting
- Separates factual evidence generation from later strategy interpretation
Research contribution
- 01Designed the research-only evidence architecture
- 02Implemented and integrated GARCH, EGARCH, GJR-GARCH, and HAR-RV workflows
- 03Built deterministic fit, state, and forward-outcome ledgers
- 04Implemented anti-lookahead outcome validation
- 05Implemented source provenance and content hashing
- 06Engineered cron scheduling
- 07Engineered process locks and append locks
- 08Implemented health checks and integrity blockers
- 09Separated research evidence from trading execution
- 10Designed the architecture for future multi-asset reuse
Research lineage
Multi-asset research progression
Volatility Research Foundation
Academic and methodological foundation
BTCUSDT Research Infrastructure
Initial public snapshotInitial public forecast-evidence baseline
Unified Multi-Asset Framework
Planned researchLong-term architecture objective
Gold Futures Generalization
Planned researchPlanned commodity research; not operational
NQ and ES established the academic foundation for volatility-regime research. The public BTCUSDT showcase documents an initial forecast-evidence infrastructure baseline. Gold futures are a planned commodity generalization asset. The long-term objective is a unified multi-asset research and automation framework.
Public snapshot forecast category
Baseline forecast models
The initial public snapshot records four statistical specifications producing comparable raw next-hour decimal variance evidence.
GARCH(1,1)-t
01A symmetric conditional-variance model with Student-t innovations for heavy-tailed returns.
EGARCH(1,1)-t
02A log-variance specification that represents asymmetric volatility responses without positivity constraints.
GJR-GARCH(1,1)-t
03A threshold model that allows negative and positive return shocks to affect variance differently.
HAR-RV
04A heterogeneous autoregressive model built from realized-volatility components across multiple horizons.
Their outputs are not market regimes, direction forecasts, entry signals, vetoes, sizing instructions, leverage decisions, or trading permissions.
Feature ecosystem
Realized-volatility research map
Research input
5-minute decimal log returns
Research input
Realized Variance
Research input
Realized Volatility
Diagnostic feature
RV48
Historical research component
Diagnostic feature, not an independent shadow model in the public baseline.
Forecast construction
HAR components
Daily
Research input
Weekly
Research input
Monthly
Research input
HAR-RV
Forecast model
Vol-of-vol
Volatility-state research feature
Research branches
Jump component
Diagnostic feature
HAR-Jump
Regime research
Defensive-policy candidate
Historical research component
Realized-volatility components support forecasting, diagnostics, and defensive research. They do not independently authorize trading.
HAR-Jump is not an entry model and does not provide short permission. Vol-of-vol does not provide trading permission. Defensive research candidates are not automatic veto rules.
Historical research context
Prior benchmarks and challengers
Earlier benchmark specifications provide academic context for the current public research architecture without implying comparative performance.
Academic contextEWMA volatility
Prior benchmark research
ATR volatility
Prior benchmark research
Parkinson volatility
Prior benchmark research
Garman-Klass volatility
Prior benchmark research
These are historical benchmark contexts, not shadow models recorded in the initial public baseline. No comparative performance claim is presented.
Uncertainty
Monte Carlo variance uncertainty
Illustrative parameterization
A fixed seed generates 250 normalized mean-reverting log-variance paths across 48 research steps. Parameters are not fitted.
5th-95th percentile band
25th-75th percentile band
Median variance path
Illustrative scenario dispersion under a normalized mean-reverting log-variance process. This is not a return forecast, trading simulation, or investment projection.
Evidence discipline
Evidence maturity
Initial forward outcome rows
8
Shadow models at baseline
4
Baseline maturity
BOOTSTRAP
At the initial public baseline, outcome comparisons and descriptive model ordering were intentionally suppressed below the common reviewable threshold.
Evidence pipeline
Evidence Pipeline Explorer
Follow one deterministic synthetic event from completed observations through forward validation and integrity review.
Interaction changes only this conceptual visualization. It does not alter a model, forecast, policy, or system state.
Stage 1 of 7
Completed Market Data
- Purpose
- Use completed observations with explicit source identity.
- Required
- Completed 5-minute candles
- Produced
- A completed candle set with explicit source identity.
- Timing rule
- No incomplete candle enters the research window.
- Prohibited
- No interpolation, forward fill, or silent repair.
- Integrity
- The source identity remains attached to the completed observations.
Integrity refusals
Failure Mode Atlas
A reliable research pipeline should fail visibly when evidence is premature, duplicated, orphaned, stale, or changed during publication.
Refusal case 1
Premature Outcome
- Trigger
- The forecast horizon has not closed.
- Why dangerous
- Future information would be treated as factual too early.
- System response
- Reject outcome publication.
- Evidence retained
- Forecast state remains unchanged.
- Boundary protected
- Anti-lookahead integrity
Conceptual illustration. This atlas describes research-integrity behavior, not a live incident, production alert, or trading control.
Research operations
Operational scheduler
Seven scheduled jobs
- 01Hourly GARCH-family state generation
- 02Hourly HAR-RV state generation
- 03Hourly forward outcome processing
- 04Daily GARCH-family fitting
- 05Daily HAR-RV fitting
- 06Hourly health validation
- 07Daily maturity review
Operating controls
- Staggered UTC scheduling
- Serialized scheduler execution
- Process locks and separate append locks
- Scheduler status and schedule-hash validation
- Automatic cron observation
- Production fit generation is scheduler-controlled
- Production state generation is scheduler-controlled
- Forward-outcome appends are scheduler-controlled
- Manual production appends are outside normal operating procedure
Operational health is research infrastructure evidence, not a trading signal.
Public showcase baseline · 2026-07-13 UTC
Initial public operational snapshot
Shadow models at baseline
4
Scheduled jobs at baseline
7
Ledger blockers at initial validation
0
Semantic duplicates at initial validation
0
Initial forward outcome rows
8
Baseline outcome maturity
BOOTSTRAP
Execution integration
NOT INCLUDED
Fixed portfolio baseline for the initial public showcase. It does not refresh, does not represent a production server, and is not a live monitor, model ranking, strategy approval, or trading-readiness indicator.
Engineering scope
What this project demonstrates
BTC research system stack
Python · pandas · NumPy · arch · SQLite · CSV event ledgers · Linux · cron · flock · GitHub · AWS Lightsail
Non-negotiable constraints
Research boundaries
- Research-only system
- No live trading
- No paper trading approval
- No Binance execution
- No broker integration
- No order routing
- No entry permission
- No short permission
- No leverage
- No position sizing
- No automatic veto rule
- Policy state is not entry permission
- Model maturity is not strategy approval
- Lower forecast loss is not trading permission
- No investment advice