← Back to list
weekly2026-04-174 min read

Week 16

factor-investingquantitative-researchfund-admindashboardrklbwatch-factor

This week was three parallel tracks running full-tilt: finishing off the Watch Factor study with a clean null, closing Mar 2026 books on the fund (with two methodology fixes I had been putting off), and kicking off a new event-study research project on whether RKLB's CEO media appearances actually move the stock.

1. Watch Factor Research — the null finding

Finished the full quantitative test of the "luxury watch prices predict luxury stocks" hypothesis. Total: 3,030 daily Granger tests + 2,072 monthly Granger tests + event study + walk-forward backtests, all with Holm-Šidák correction at 5% FWER. The cleanest single test — Omega ChronoPulse → Swatch Group (UHR.SW) — comes out at static correlation -0.010 and minimum raw Granger p = 0.137. The widely-cited -0.68 correlation with Bitcoin is a spurious regression artifact; the actual return correlation is -0.04. Best Deflated Sharpe Ratio across all backtested strategies: 0.008. Comprehensive null.

→ Full research and methodology

2. Fund Operation Agent — March NAV + methodology fixes

Closed the Mar 2026 books. Headline numbers: Net NAV $4,673,981.72, Class B NAV/share 810.22 (MTD -5.24%), Class C NAV/share 814.27 (MTD -5.16%). Rough month driven by -$298.9k unrealized equities.

Two methodology fixes that had been sitting in the backlog finally got resolved:

  • Option assignment RGL — switched from IBKR's method ($0 RGL, premium absorbed into stock cost) to the Amicorp/TJZ method (recognize gross premium as RGL, stock cost = strike only). This surfaced an extra $10,900 of RGL on the TSLA P390 assignment and brought my validator in line with how Amicorp books it. Verified against the Nov 2025 BABA P165/P170 assignments — the same $13,181 diff that had nagged me for months is exactly the premium.
  • Cost basis — TJZ vs IBKR — every stock's cost basis differs between the two systems (IBKR uses FIFO, TJZ uses something else I haven't fully nailed down yet). For the assigned stocks, TJZ uses strike only, IBKR uses strike minus premium. TSLA cost moved from $807,358 → $824,000 after switching, avg price $412.00/share. Prior UGL restated from -$25,745 → -$31,490.
  • Setup Fee formula — corrected the monthly amortization with gov subsidy. Base = 240,000 - N×(240,000/36) - 150,000 HKD, denominator = 36 - N, re-computed every month. Mar 2026 (N=3): $405.01.

3. TGIF-I Dashboard upgrades

Big week for the daily-report dashboard too. Headline: IB canonical TWR is now the source of truth for all five banner numbers and the NAV history chart. nav_source.py parses 13 IB CSV reports and matches IB's TWR across 115 trading days with zero diff — 96 previously-wrong days (including that 11/06 $1M bug) are now fixed.

On top of that:

  • New Stock Analysis section — dual-axis chart (cumulative return % left, stock price $ right) per ticker, with trade markers (buy / sell / option writes & closes) layered in. Hover shows shares + avg cost on normal days, full trade breakdown on trade days.
  • FIFO vs Weighted Average toggle — implemented lot-based FIFO alongside weighted avg. Only TSLA actually differs ($412 FIFO vs $414.11 weighted, driven by the 11/28 partial sell). Toggle defaults to FIFO.
  • Historical cost lines — two toggles to switch avg-cost / amortized-cost between a flat "current value" horizontal line and a full stepped-line history that jumps with each trade.
  • New benchmarks — XYLD (S&P 500 covered-call ETF) and NDX, both auto-pulled from Yahoo. Verified IB's SPX matches Yahoo to 0.0000%, and IB's NDX drifts ~0.01%/day (~0.35% over 6 months).
  • Dark/Light theme — CSS variables + View Transitions API cross-fade, localStorage remembers preference.
  • Mobile fixes — touch scrub tooltips theme-aware on both charts, hovermode: false on mobile to stop Plotly's default hover colliding with my custom touch tooltip, Plotly.purge() + tooltip reset on ticker switch.

4. RKLB Management Exposure — new research project

Started a brand-new research line: do RKLB's exec media appearances actually move the stock, or is it pure selection bias? "Peter Beck went on CNBC → RKLB up 3%" sounds like causation, but CEOs tend to go on CNBC because there's good news to talk about — so the appearance and the price move could both be downstream of the earnings beat.

Designed a full event study around this, with two-factor (SPY + ARKX) market model, independent catalysts.csv for selection-bias control, and "clean events" filter (≥3 days from any catalyst). First-pass results are directionally very suggestive — raw Beck appearances show +6.1% CAR[-1,+1], but clean Beck appearances flip to -3.9%, which is exactly the selection-bias pattern the methodology was designed to detect. N=4 though, so p=0.34 — not significant, needs a bigger sample.

Also learned RKLB has β_SPY = -0.96 but β_ARKX = +2.37 in the two-factor model. It's essentially a 2.4x levered bet on the space sector, not the broader market.

→ Full research and methodology

→ Watch Factor research · RKLB Management Exposure research