Changelog
All notable changes to this project are documented here.
Newest entries first. Dates in ISO format (YYYY-MM-DD).
This project began as a single trend-following ETF strategy (see README.md)
and is evolving into a multi-strategy systematic book. This file tracks that evolution.
[2026-06-26] — Research documentation + VRP gate validation
Added
RESEARCH.md — full methodology, backtest process, results, and rationale
for each strategy addition, with honest markers where figures are not yet
measured.OPERATIONS.md, CHANGELOG.md, and a reframed README.md introduction
linking the four-document set (narrative / operations / research / history).- VRP gate-validation backtest (2015-01-01 → 2026-06-26) run and saved to
vrp_backtest_results.txt. Both required gate-fires confirmed (Feb 2018,
Mar 2020 = True). Sleeve standalone: CAGR 7.5%, Sharpe 0.39, max drawdown
−65.4% (≈ −6.5% at a 10% portfolio cap).
Findings
- VRP gate is a regime detector, not a gap-loss preventer: on 2018-02-05 the
exit fired on the close after SVXY had already gapped ~80% at the open. The
sleeve’s survivability rests on the 10–15% position cap, not the gate.
Documented in
RESEARCH.md.
- Confirmed VRP futures-curve proxy is the CBOE
^VIX/^VIX3M indices (not
VIXY/VIXM); the reverse-split issue does not apply.
- Mean-reversion diversification case confirmed: anchor + MR blended under ERC
weighting (≈67% MR / 33% anchor) gives max drawdown −17.73%, 9.23pp
shallower than the anchor’s −26.96%, with blended Sharpe 0.95 vs anchor 1.07.
Worst residual drawdown narrows to the COVID window (2020-02-19 → 2020-03-23).
Saved to
blended_backtest_results.txt.
- PEAD lookahead-bias check came back clean after fixing a quarter-matching flaw
in the verification script: the earlier DIVERGED (17% on MKC) was a
different-quarter artifact, not a data revision. Matched-quarter comparison
shows 0.00% divergence — yfinance serves point-in-time estimates. Verdict rests
on one stock so far; to be re-confirmed as the earnings DB fills through Q2.
Saved to
pit_verification_results.txt.
- PEAD aggregate stats measured (2025-06-02 → 2026-06-01, cohort-ranking version):
24 trades, 62.5% hit rate, +1.23% avg/trade, Sharpe 0.59, max drawdown −16.3%.
Sample is small (~4 quarters) — indicative, not validated; re-run after Q2 2026.
The file’s 28.8% CAGR is a sequential-compounding artifact and is not quoted.
Earlier seasonal-SUE version not reproducible (replaced in place, no history).
Saved to
pead_aggregate_stats.txt.
Deployment / sample-growth items (not open research)
- PEAD: re-verify PIT and re-run stats after Q2 2026 earnings adds ~5–6 events.
- VRP: portfolio-level position-cap enforcement required before any allocation.
[2026-06-26] — IB Gateway migration
Changed
- Migrated execution from TWS (port 7497) to IB Gateway (port 4002) for
headless, unattended operation. IBC 3.24.0 (IbcAlpha) handles auto-login
to paper account DUP447680.
- Updated the connection port in
data_feed.py, order_engine.py, and
monitor.py (old TWS line kept commented as a fallback).
Added
- Headless launch via
C:\IBC\StartGateway.bat. IBC config kept outside the
repo, with credentials gitignored.
- Full pipeline dry-run passed end-to-end through Gateway: data feed → signals
(trend / mean-reversion / VRP / PEAD) → governance → ERC risk
(gross exposure 134.97%) → 15 orders staged → monitor → dashboard push.
Known gaps
StartGateway.bat not yet wired into Task Scheduler ahead of the 7:00 AM
submit job. Until scheduled, Gateway must be launched manually before market
open.
[2026-06-24] — Multi-strategy split architecture
Added
- Mean-reversion sleeve went live: 50-day z-score, ±1.5 dead-band,
asymmetric momentum filter (excludes top-quartile momentum names from the
short book), 2-position-per-sector cap.
- VRP sleeve built (SVXY-based volatility-premium carry via ^VIX/^VIX3M) —
currently gated out by governance.
- PEAD sleeve built (post-earnings-announcement drift, cross-sectional
cohort SUE ranking) — orders paused pending a qualifying 5+ stock cohort.
- Shared governance gate (three-condition VIX risk governor with two-day
deactivation confirmation) and ERC portfolio risk layer (equal risk
contribution weighting, 11% vol target).
Changed
- Split the pipeline into two stages to satisfy order-type timing constraints:
4:15 PM signal generation (
daily_routine_v2.bat) and 7:00 AM submission
(premarket_submit.bat).
- Order type changed from MKT/DAY to MKT+OPG (open auction). MOO was rejected
by the paper gateway (Error 321: requires submission before 3:58 PM,
incompatible with the 4:15 PM pipeline).
- Migrated all scripts from
OneDrive\Desktop\TrendFollowing to
C:\QuantTrading\TrendFollowing\ to eliminate file-lock race conditions.
[2026-06-18] — Paper trading live
Added
- Trend-following anchor strategy entered its first positions on IBKR paper
account DUP447680 (IWM, EEM, SMH, XLK, QQQ, SPY, EWJ).
- Daily monitoring and monthly rebalancing automated via Windows Task Scheduler.
- Live dashboard deployed to GitHub Pages
(
isaacnicas.github.io/quant-portfolio/live-dashboard.html), fed by a daily
JSON push.
[earlier] — Backtest and research
Added
- Eighteen-year backtest (2008–2026) of the trend-following ETF strategy across
US/international equities, government bonds, gold, and currencies, with
realistic trading costs.
- TSMOM + CS-Mom signal blend, regime filter, and fast-exit logic.
Fixed
- Over-layered safeguards bug: three independent risk overlays multiplied
together, leaving the strategy ~30% invested on ordinary days and returning
~2%/yr. Simplified to one clear rule → over 13%/yr without materially more
risk.
- Fast-exit calendar lock: the circuit breaker stayed defensive until the
next monthly check-in, missing short-selloff recoveries. Reworked to listen
for recovery daily so it can re-engage as soon as the data supports it.