8c7b9a4706
Logs the open question (noise/denominator artifact vs real depeg micro-edge deliberately self-limited) + how to settle it. VIOLET keeps BLUE's exclusion regardless (follow BLUE). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2.5 KiB
VIOLET Research TODO — Why does the IRP pick stablecoins when unguarded?
Status: Research TODO (logged 2026-06-14). Low priority, high curiosity. Not blocking.
Observation
The VIOLET shadow soak, running BLUE's muted-IRP gold config (min_irp_alignment=0.0)
WITHOUT BLUE's stablecoin exclusion gate, repeatedly ranked USDCUSDT top and would
have shorted it at full conviction (9.0). BLUE hardcodes these out
(_STABLECOIN_SYMBOLS, nautilus_event_trader.py:24/3906 — "must never be traded"), and
VIOLET now replicates that (commit V3.1). So in production it's correctly excluded.
The question
Why does the IRP ranker select stablecoins when unguarded — and is the exclusion leaving a real (small, desirable-to-forgo) edge on the table?
Two hypotheses:
- Noise/artifact. Stablecoin price series are near-flat; the IRP efficiency/ARS metric may divide by a tiny denominator (low realized vol), inflating the score on essentially flat micro-movements → a spurious "clean signal." If so, exclusion is pure correctness (avoids dividing-by-near-zero noise).
- Real micro-edge, deliberately self-limited. Stablecoin depeg oscillations (USDC/USDT wobbling around $1) are mean-reverting and genuinely tradeable at tiny size — the IRP may be detecting a real, low-volatility, low-risk signal. BLUE's exclusion would then be forgoing a small but real edge BY DESIGN (correctly — pegged assets carry depeg tail risk, poor borrow, regulatory/operational hazards that dwarf the micro-edge; not worth it).
How to settle it (when researched)
- Backtest the EXCLUDED stablecoin picks as if traded (small size): realized PnL, Sharpe, max depeg drawdown, borrow/funding cost. Does the micro-edge survive costs + the depeg tail?
- Inspect the IRP ARS/efficiency computation on flat series (rank_assets_irp_nb): is the high score a denominator artifact (hypothesis 1) or a real efficiency signal (2)?
- Cross-check against the dual-leverage/conviction the sizer assigns (it gave conviction 9 on USDC vel_div −0.0585 — is that vel_div itself an artifact of a flat series?).
Why it matters (mildly)
If hypothesis 2 holds, it informs whether a separate, tiny-size, depeg-aware stablecoin strategy could be a future color — orthogonal to BLUE's directional short edge. If hypothesis 1, it's just confirmation the exclusion is correctness. Either way, VIOLET keeps the exclusion (follow BLUE). See blue_margin_envelope_study, violet_v3_alpha_doctrine (#12 follow-BLUE).