CA Fixed Income Allocation Model
Three-state fixed-income sleeve for CAD-mandate portfolios. 70% Canadian aggregate bond core + 30% rotating overlay across high yield, long-duration Canada govt, and short-duration cash, gated by Canadian rate-favorability signals and a US HY OAS credit-cycle discriminator.
Architecture
The CA Fixed Income Allocation Model holds a permanent 70% core in iShares Canadian Universe Bond Index (XBB.TO) with a 30% tactical overlay that rotates monthly across three states:
- CREDIT_ON → overlay in iShares US HY Bond CAD-Hedged (XHY.TO)
- DURATION_ON → overlay in BMO Long Federal Bond Index (ZFL.TO)
- DEFENSIVE → overlay in BMO Short Federal Bond Index (ZSB.TO)
Signal construction
The state is determined by a two-stage gate. First, aCanadian rate-favorability vote tests whether duration is currently penalized. Second, conditional on a favorable rate environment, a US HY OAS 60-month z-score discriminator routes between credit-on and duration-on.
Rate-favorability vote (Canadian)
Three binary components, threshold 2 of 3:
- Canada 10Y − 3M T-bill spread ≥ 0 (FRED, BoC inputs)
- ZFL.TO trailing 6-month total return ≥ 0
- XRB.TO trailing 6-month total return ≥ 0
If < 2 votes pass, the overlay rotates to ZSB regardless of credit conditions. This is the DEFENSIVE state.
Credit/duration discriminator (US HY OAS z-score)
A 60-month rolling z-score of the US ICE BofA HY OAS series (FRED BAMLH0A0HYM2) is computed each month-end. Window: 60 months, minimum 36 months.
- z ≤ 0 → CREDIT_ON (overlay = XHY)
- z > 0 → DURATION_ON (overlay = ZFL)
Why US HY OAS, not Canadian: Canadian credit spread series are sparse and discontinuous. US HY OAS proxies the global credit cycle, which CA risk assets co-move with on a 0–1 day lag. Documented honestly in the limitations tab.
Locked parameters
| Parameter | Value | Notes |
|---|---|---|
| Core weight | 70% | XBB.TO permanent |
| Overlay weight | 30% | quarterly routine · monthly gate |
| Rate vote threshold | 2 of 3 | fixed in spec |
| ZFL lookback | 6 months | rate leg |
| XRB lookback | 6 months | rate leg |
| HY OAS z window | 60 months | min 36 months |
| HY OAS z threshold | 0.0 | credit/duration split |
| Hysteresis | 1mo exit + 1mo cooling | dampens whipsaw |
| Routine rebalance | Quarterly | Mar/Jun/Sep/Dec month-end |
| Defensive gate | Monthly | evaluated every month-end |
Research lineage
This is the third architecture attempt. The first two failed pre-registered acceptance gates and were honestly recorded as null results.
- Phase 1 (May 2026): Direct fork of the US Fixed Income model. Failed H1 IR gate vs B1 benchmark on the Canadian universe. Diagnostic showed all 3 rate-favorability votes were correlated bond-trend signals — no orthogonal signal.
- Phase 2 (May 2026): Parameter grid sweep and signal substitution variants. Also failed. Pre-reg discipline rejected retroactive parameter tuning.
- Phase 2.5 diagnostic: Alpha attribution isolated the issue — relmom discriminator (HYG 3M vs ZFL 3M) was structurally inverted, picking the wrong overlay 67% of the time vs 50% random.
- Phase 3 architectural fix: Replace the bond-trend relmom with US HY OAS z-score as the discriminator. Pre-registered, passes H1 IS Bonferroni-corrected gate (Sharpe +0.081 over B1, p = 0.0152). Locked as production 2026-05-26.
Walk-forward and forward-OOS
The 2026-05-26 IS evaluation is on the full available window (2014-02 onward, ~12.4 yr) and is by construction not OOS. The portal honestly flags this contamination on the Limitations tab.
The forward-OOS firewall begins on the first production publish (2026-05-31). The first formal H1 re-evaluation is scheduled for 2029-05-26 (+36 months). Until then, all performance numbers are IS and should be treated as a sanity check, not validation.