archived model combinations
GARCH-family ticker-model combinations.
← All work
Risk analyticsHistorical evidence live
Ruang Risiko IDX
Indonesian equity risk research that makes returns, volatility, drawdown, downside estimates, direction probabilities, provenance, and model evidence inspectable.
Risk analyticsVerified 2026-09-07
Ruang Risiko IDX
36archived model combinations
9,072walk-forward rows
0failed archived rows
Quick read
Project summary
Problem
A price forecast alone does not describe risk. Recruiters and users need to see uncertainty, data quality, benchmark context, and how a model behaved outside its fitting window.
My role
I designed and built the reviewed research workflow across market-data validation, quarantine rules, volatility and direction modeling, out-of-sample evaluation, and the public research interface.
Result
9,072 successful archived walk-forward forecast rows with 0 failed archived rows.
Stack
Next.js · Vercel · yfinance · GARCH family models · scikit-learn
Status
Historical evidence live
Evidence reviewed
7 Sep 2026
walk-forward rows
Successful archived forecast evidence.
failed archived rows
Recorded in the reviewed forecast evidence.
reviewed market scope
Five IDX equities plus the market benchmark in the reviewed project scope.
Overview
Ruang Risiko IDX is a production risk-research product for Indonesian equities. It is designed to show not only an output, but also the evidence, provenance, and limitations behind the output.
Problem
A price forecast alone does not describe risk. Recruiters and users need to see uncertainty, data quality, benchmark context, and how a model behaved outside its fitting window.
Why it matters
Risk research is more useful when uncertainty, data quality, benchmark context, and out-of-sample behavior are visible instead of collapsing the decision into one forecast number.
My role
I designed and built the reviewed research workflow across market-data validation, quarantine rules, volatility and direction modeling, out-of-sample evaluation, and the public research interface.
Product
The public product combines market context with risk and model evidence while keeping the interface educational rather than transactional.
- Cakrawala Lite Market Board and IHSG Benchmark Lens for market context.
- Risk Map views for drawdown and volatility.
- Returns, Value at Risk, direction probability, model evidence, provenance, and explicit limitations.
- Responsive controls, keyboard semantics, and reduced-motion behavior in the verified production interface.
Data
Yahoo Finance market data enters through a replaceable yfinance provider, then passes validation and quarantine before a reviewed read-only artifact is published.
- Invalid OHLC relationships are quarantined rather than silently repaired.
- Unsupported tickers, malformed provenance, missing sections, and unsafe advisory output fail closed in the public artifact contract.
- The public universe is fixed to five IDX equities plus IHSG in the verified scope.
Method
Volatility and direction are treated as separate modeling questions, and model choice remains ticker-specific.
- GARCH, EGARCH, and GJR-GARCH variants are compared with out-of-sample forecast evidence.
- Logistic Regression, Random Forest, and a constant-probability baseline are compared for next-observation direction probability.
- Kronos-small and IBM Granite TTM remain experimental challengers until traceable evidence supports promotion.
Results
The strongest reviewed evidence is the archived walk-forward record rather than a feature count.
- 36 archived GARCH-family ticker-model combinations.
- 9,072 successful walk-forward forecast rows.
- 0 failed rows in the archived walk-forward evidence reviewed for this portfolio.
- Verified Vercel production for five IDX stocks plus IHSG.
Validation
The project favors temporal validity and visible failure states over headline fit metrics.
- Out-of-sample evidence is used to compare volatility models.
- Classical direction models are compared with an explicit constant-probability baseline.
- Public output is gated to avoid personalized transactional instructions or price targets.
Architecture
The reviewed production path separates acquisition, validation, research artifacts, and the public interface.
- Provider layer for replaceable market-data acquisition.
- Validation and quarantine before reviewed artifacts are exposed.
- Read-only public artifact contract for the Vercel interface.
- Research and scope gates before an artifact can reach the public data-refresh path.
- Market data providerReplaceable acquisition layer
- Validation and quarantineReject invalid OHLC and malformed inputs
- Reviewed artifactsRead-only evidence contract
- Risk and model analyticsVolatility, direction, drawdown, and VaR
- Public interfaceEducational research experience
Tradeoffs
The public market universe is intentionally narrow. The product prioritizes inspectable evidence and stable behavior over adding more tickers or model families quickly.
Limitations
The portfolio does not present experimental foundation models as promoted production models. The public product is educational risk research, not personalized financial advice. The latest verified artifact is stale historical research evidence with a 2026-07-31 data cutoff, while the current provider, current prices, and current forecast remain unavailable.
Live Product
The Vercel production route returned HTTP 200 on 7 September 2026 and served a recovered verified real-data research evidence artifact. The artifact retains 9,072 successful historical walk-forward forecasts with 0 failed rows, but it is marked STALE with a 2026-07-31 data cutoff and keeps current prices and the current forecast unavailable.
Source
A repository URL is not published in this prototype because the current remote source state was not reverified during this milestone. This prevents a stale or dead source link from being presented as evidence.
Technical status: Production route live with stale historical evidence; current refresh unavailable
Repository status: Repository link intentionally withheld until remote source access is reverified.
Next milestone
Strengthen probability calibration and forecast reliability diagnostics using leakage-safe out-of-sample evidence before widening any model claim.