meanydeany / Independent work
I trade futures and study the decisions behind them. My work connects financial econometrics, trading-behavior analysis, and market microstructure.
Personal trading
Actual account observations, separate from backtests and model evaluations.
Performance, calendar, and data coverageRead-only account results. Flow-adjusted reporting begins 01 Aug 2026.
Authenticated tracking / since 01 Aug 2026
152 wins / 70 losses / 222 resolved UTC days / funding excluded / snapshot fallback
Authenticated window / account-wide
Published realized component
All open positions / not realized
24 completed / net sent ≈ 5,933.11 USDT eq. + fees ≈ 9.77 USDT eq. / CSV through 25 Sep 2026 / live update connecting
| Instrument | Direction | Position side | Open orders |
|---|---|---|---|
Connecting to the public feed No position state is assumed while data is unavailable. | |||
| Period | Return | Net PnL |
|---|---|---|
| 7 days | — | — |
| 30 days | — | — |
Legacy rolling figures use the published Modified Dietz mark-to-market calculation.
These are sanitized account observations, not ASRA strategy results or an execution interface. All non-zero symbols and hedge sides remain visible in the public feed. Exact sizes, prices, per-position PnL, balances and credentials are not published. Open-order presence is not proof of a protective stop.
Account tracking begins at 2026-08-01T00:00:00Z. The source supplies flow-adjusted performance; the interface does not recompute it. Freshness follows each feed’s published TTL and original observation time. Updates are validated before publication. A failed update never means that the account is flat.
Daily PnL and PnL % are published read-only account telemetry derived from the authenticated ledger. The primary daily metric from V2 follows the Binance Futures PNL Analysis wallet-balance basis: ending wallet balance minus beginning wallet balance minus net capital inflow; the percentage denominator uses beginning wallet balance plus positive inflow. Open-position unrealized PnL is therefore excluded from the primary daily figure until realized. A separate flow-adjusted mark-to-market PnL/return is retained for research context. The public calendar currently extends back to 2024-11-13 because that is the earliest date present in the supplied Binance Futures export. It is not asserted to be the first-ever trade or account inception date. Historical CSV cells show stablecoin realized cash PnL only: supported BTCUSDT/BTCUSDC REALIZED_PNL, FUNDING_FEE and USDT/USDC COMMISSION. Capital flows are excluded; BNB-denominated commission and ambiguous income types are not converted into USD and make that date visibly partial. From 2026-08-01 onward authenticated ledger rows take precedence. The trade-day win rate is a separate realized-trading statistic rather than an account-return measure. It counts UTC dates with actual futures fills and a resolved realized outcome, classifies each day after trading commissions, excludes funding entirely, and excludes fill-only or flat days from the win-rate denominator. Historical BNB commissions are translated with contemporaneous BNBUSDT 1-minute close prices, so the published percentage is explicitly approximate. Historical base runs through 2026-09-28 UTC; the live read-only feed extends the same funding-excluded definition from 2026-09-29 onward. External-withdrawal totals combine the supplied historical export through 25 Sep 2026 with completed external withdrawals from the authenticated Binance withdrawal-history API after that boundary. Only completed external withdrawals are added; pending, rejected and internal transfers are excluded. The source syncs every five minutes and the browser checks every 30 seconds. Withdrawal status and observation time are shown separately; failed updates retain the last verified value. Stablecoins are valued at par. Historical XRP retains the export’s transaction-implied estimate; subsequent non-stablecoin withdrawals use the closed one-minute USDT price at the withdrawal application time and remain estimates. Withdrawal totals are not trading PnL. No address, TXID or raw withdrawal row is published. Personal notes stay browser-local and are never included in the public feed.
Selected research
What I asked, what I built, and what the evidence actually supports.
Personal trading / Machine learning
Can a model reproduce my LONG/SHORT decisions using only information available before entry?
I reconstructed trading episodes, aligned pre-entry features, and separated direction imitation from entry quality and tail-loss detection.
The final chronological holdout supports partial imitation. It does not establish profitable entries or a reliable risk veto.
Read the case studyFinancial econometrics / Risk
Does market context improve one-hour BTC risk forecasts beyond historical volatility persistence?
I challenged the model against a HAR-basis baseline, then tested whether forecast improvements translated into a useful risk policy.
Independent forecast confirmation survived. The separate preregistered policy-utility claim did not.
Read the case studyMSc thesis / Futures
Can daily volatility regimes provide useful context for intraday futures strategies?
I compared an EGARCH-based participation gate and exposure-scaling bundle in a long-only intraday breakout framework, using historical ablation and walk-forward analysis.
The revised thesis treats the backtests as retrospective development comparisons. Timing, provenance and execution-accounting limits prevent an independent performance claim.
Read the case studyCurrent work / Market microstructure
A signal can look useful and still fail at the point of execution. I am studying order flow, fill uncertainty, and what prices do after a hypothetical fill.
Trades and order-book updates from multiple venues.
Sequence discipline, state checks, and reproducible event streams.
Fill proxies, adverse selection, markouts, and cost-aware tests.
Research workflow, not a live order route. Local compute benchmarks are not exchange latency, and hypothetical fills are not realized PnL.
How I work
My academic training spans undergraduate Quantitative Economics & Econometrics at UC San Diego and an MSc in Economics at the University of Copenhagen. Trading gives me questions; research helps me test them; engineering makes those tests repeatable.
AI Systematic Research Architecture
I use AI-assisted development and review to implement experiments and challenge assumptions. I remain responsible for the question, the validation design, and the interpretation.
A better forecast, a successful replay, and a profitable trade are different claims. I keep their evidence separate rather than presenting every project as one working trading algorithm.
Get in touch