meanydeany / Research
Risk forecasting, strategy research, measurement, and academic work. Each result keeps its own evidence class, assessment window, and limitations.
Results are grouped by research question. Evidence class, work state, and claim outcome remain separate.
Interactive / Synthetic research
Compare model call/put premiums across strike and expiry, then change spot and volatility to explore pricing scenarios. Study fixed-tenor implied variance, 25-delta skew and term structure in the separate synthetic-quote demonstration. No live calibration or trading benefit has been established.
Explore the 3D BTC option-price model →Risk forecasting
Independent forecast assessment, development comparison and the completed policy test answer different questions. Their evidence labels and assessment windows do not transfer between records.
Context adds short-horizon BTC risk information beyond historical volatility persistence: H+C improves both primary losses against HAR-basis on the frozen independent assessment.
Risk information, not directional alpha, a trading signal, sizing approval or execution authority.
Inspect independent risk forecastPolicy translation test
Forecast information survived; the preregistered policy-utility claim did not.
Better risk forecasts did not automatically become a better risk policy.
Inspect policy translation testThe risk candidate passed the strong-baseline challenge described in the reviewed research brief.
Passing a development comparison does not itself establish independent confirmation.
Inspect risk baseline challengeReturn / methodology
Positive controls exposed blind spots. Target and selection repairs recovered methodological power, and synthetic scheduler robustness supported simpler fixed-native selection. The completed V3 test did not demonstrate incremental historical return transfer within its frozen validation domain.
Positive controls exposed a measurement blind spot; a repaired nonlinear sensor recovered detection within the tested synthetic settings.
Recovery in synthetic controls does not establish historical BTC nonlinear alpha or market predictability.
Inspect nonlinear sensor recoveryMeasurement repair
Native-horizon target alignment and selection repair recovered methodological power in the supplied research progression.
Do not attribute repaired power specifically to adaptive scheduling. The completed repair's evidence class and test domain remain unresolved.
Inspect native-horizon selectionThe registered synthetic test found no meaningful adaptive incremental value beyond fixed native-horizon selection, supporting a simpler research architecture within the tested scope.
Supports simplification within the tested synthetic scope, not exact equivalence, historical BTC alpha or execution authority.
Inspect native scheduler robustnessHistorical transfer test
The repaired fixed-native Activity and nonlinear challengers did not improve historical one-hour BTC return forecasts over the frozen OHLCV baseline. Zero finalists; the locked return holdout remained unopened.
The repaired Activity / nonlinear challengers did not add historical return information beyond the frozen OHLCV baseline in the tested domain. This does not establish that all BTC return information is absent.
Inspect candidate generator v3Strategy research
R15 connects short-horizon signal design, volatility-aware modeling and ensemble evaluation. Historical gross results and legacy strategy studies retain their own assumptions; they are separate from independent forecast confirmation and actual account performance.
I built and compared a raw-return benchmark, a volatility-normalized model and a dual-head ensemble. Both extensions produced positive gross results; RAW remains the fixed reference while incremental value is assessed against predefined statistical criteria.
Hypothetical, zero-friction historical development results. Paired comparisons did not establish a sufficiently consistent upgrade; neither extension has been promoted. No net executable-return or live-account claim.
Inspect r15 btc researchThe sparse C4 regime router exceeded the retained baseline on historical risk-adjusted performance under the original comparison, with selection and cost-stress limitations.
C4 was selected after inspecting historical tournament results; this is not independent validation.
Inspect c4 historical challengerAcademic research
MSc thesis
The thesis compares an EGARCH participation gate and exposure-scaling bundle in a long-only intraday breakout framework. The revised interpretation is retrospective development evidence, without separately identifying the gate, scaling or asymmetry effect.
Full-sample volatility ranks use future observations; development exposure, source provenance and execution accounting limit the reported backtests. The revision is not a fresh empirical replication.
Inspect futures volatility thesisThe reconstructed seminar revision explains exploratory GSADF testing and BSADF episode dating. It corrects the interpretation while retaining the original study as a historical research record.
The original figure conflicts with its reported scalar statistic; exact episodes and return/volatility comparisons remain unverified. The revision was reconstructed from the PDF, without editable original source.
Inspect bitcoin bubble diagnosticsCurrent / next
Current work focuses on event-level BTC market microstructure, passive execution, and low-latency market-data systems. Earlier ASRA, forecasting, and historical strategy studies remain frozen under their original evidence boundaries.
Current direction / Market microstructure
Do observable order-flow and liquidity states change fill probability and post-fill markouts enough to support robust, cost-aware passive-execution payoff?
5-minute and 1-minute layers remain as context and benchmarks. The new information layer is event-level market mechanics, not faster bar indicators.
Current build / Systems engineering
Build an event-driven core that captures, validates, reconstructs and replays trades and order-book updates with explicit timestamp and sequence discipline.
Raw WebSocket capture → sequence validation → local order book → deterministic replay → microstructure features → fill-probability, adverse-selection and markout analysis in Python.
Working principles
No single diagnostic establishes a system’s validity. The standard is point-in-time data, robustness, provenance, operational integrity, and claims that match the evidence.
A model output earns interpretation through validation. It does not become an entry, veto, sizing, or execution instruction by default.
Features, states, and outcomes use information available at the evaluated timestamp. Alignment rules are explicit and missing data is not silently repaired.
Ablation, alternative specifications, walk-forward evaluation, and subperiod checks matter more than a single favorable summary statistic.
Historical research records should remain inspectable as hypotheses evolve. Source identity and transformation history belong beside the result.
Duplicate protection, stale-input detection, locking, scheduling, and failure states are part of research validity when systems generate evidence repeatedly.
Descriptive evidence can inform later policy research, but policy state is not entry permission and neither layer authorizes execution.