Self Attempt 02

Asteria Investment Office — Reusable Task-Solving Procedure

This skill covers institutional portfolio risk tasks in the Asteria Investment Office domain: credit trade construction, equity correlation reviews, active allocation view refreshes, fixed-income risk rebalancing, and multi-asset committee decision files.

1. Environment & Data Precedence

API base URL as provided by the task environment. All data comes from these HTTP GET endpoints:

Hard data-precedence rule: The current environment API is authoritative. Local payload data (request JSONs, memos, stale snapshots) may contain outdated marks. When current API data conflicts with a local payload, prefer the API. Capture this in any data_precedence output field as "current_environment_over_stale_payload".

Global as-of date: All API responses share the same as_of_date (verify by checking /api/portfolios, /api/policies, /api/market/energy). Use this date in every output that requires as_of_date. The stale-data cutoff printed in /api/market/energy tells you which local dates to distrust.

2. Policy Lookup Chain

Every portfolio links to a constraint policy via constraint_policy_id in its summary. Look up that policy in the /api/policies response — it is a flat object with keys matching each policy_id. The policy provides:

• max_hy_allocation_pct — hard cap on high-yield exposure
• duration_band_years — [min, max] allowed weighted modified duration
• issuer_concentration_limit_pct — single-issuer cap (typically 12%)
• subsector_min_count_for_diversified — minimum distinct subsectors across selected holdings
• target_hy_reduction_pct — for risk-reduction policies, the minimum percentage-point reduction required
• correlation_high_threshold / correlation_low_threshold — for correlation policies
• conviction_thresholds / view_score_thresholds — for allocation-mapping policies

The allocation mapping policy (POL_ALLOCATION_MAPPING) defines:

• View thresholds: score ≥ 0.35 → OW, score ≤ -0.35 → UW, otherwise N
• Conviction thresholds: abs(score) ≥ 0.70 → HIGH, 0.35 ≤ abs(score) < 0.70 → MEDIUM, abs(score) < 0.35 → LOW
• View rank: OW = 1, N = 0, UW = -1

Multi-asset policies compose from the single-asset policies (e.g., POL_MULTI_ASSET_DEFAULT uses POL_ALLOCATION_MAPPING + POL_CORRELATION_DEFAULT + POL_CREDIT_DEFAULT).

3. Task-Type Procedures

3A. Credit Trade Construction (e.g., PF-EN-ALTA)

Goal: Select eligible bonds for BUY/SELL orders that improve carry while respecting constraints.

Procedure:

1. Fetch current portfolio from /api/portfolios/<id>. Record each holding: instrument_id, quantity_usd_m, sleeve.
2. Enrich each holding by matching instrument_id to the bond universe (/api/instruments/bonds). Pull: rating_bucket, modified_duration_years, yield_to_maturity_pct, subsector, issuer_id, energy_linked.
3. Enrich issuers from /api/issuers — pull watchlist flag per issuer_id.
4. Compute current portfolio metrics:
   • total_market_value_usd_m = sum of all quantity_usd_m
   • hy_allocation_pct = 100 × (sum of HY-rated quantity_usd_m) / total_market_value_usd_m
   • weighted_modified_duration_years = Σ(qty_i × dur_i) / Σ(qty_i)
   • weighted_yield_to_maturity_pct = Σ(qty_i × ytm_i) / Σ(qty_i)
5. Filter candidate bonds: from the bond universe, select bonds where:
   • candidate is true (or is an existing holding)
   • The issuer is not on the watchlist (if watchlist avoidance is required)
   • Matches any sector/theme preferences from the request
6. Select trades: choose the required number of BUY/SELL tickets at the required sizes.
7. Recompute post-trade metrics by adding new holdings to the portfolio and re-running the weighted-average formulas.
8. Check each constraint:
   • hy_cap_pass: post-trade HY% ≤ max_hy_allocation_pct
   • duration_band_pass: post-trade duration ∈ [band_min, band_max]
   • issuer_diversification_pass: no single issuer > issuer_concentration_limit_pct of post-trade MV
   • subsector_diversification_pass: ≥ subsector_min_count distinct subsectors among the selected (new) positions
   • watchlist_avoidance_pass: none of the BUY tickets are watchlisted issuers

Rounding: notional_usd_m to 1 decimal; post_trade_metrics to 2 decimals.

Ordering: trade_package sorted ascending by instrument_id.

3B. Equity Correlation Review (e.g., PF-INT-NEXVEN)

Goal: Compute Pearson correlation matrix from monthly index levels and identify concentration/diversification pairs.

Procedure:

1. Determine the review window from the request payload (level_start_date, level_end_date). Also available from the policy's correlation section.
2. Fetch index levels from /api/index-levels for each index in the universe. Each index returns an array of {date, level} objects.
3. Filter levels to the review window dates (inclusive of both start and end).
4. Compute monthly simple returns for each index:
   • r_t = (level_t / level_{t-1}) - 1
   • Result: N levels → N-1 returns. Count = return_observations.
   • Do NOT use log returns; use simple (discrete) returns.
5. Compute the Pearson correlation for every unordered pair of indices:
   • ρ(X,Y) = Σ((x_i - x̄)(y_i - ȳ)) / √(Σ(x_i - x̄)² × Σ(y_i - ȳ)²)
   • Round to 3 decimal places.
6. Identify extreme pairs:
   • highest_positive: the pair with the largest correlation value
   • lowest: the pair with the smallest (most negative) correlation value
7. Concentration analysis: check whether IDX_CHINA and the Asia-Pacific index (IDX_AC_ASIA_PAC_EX_JP) correlation exceeds the high threshold (0.8). If so, china_asia_dependence_flag = true and primary_code = "CHINA_ASIA_DEPENDENCE". Also set high_threshold_breached = true.
8. Diversification candidates: from {IDX_EM_EX_CHINA, IDX_INDIA, IDX_LATAM}, select indices whose pairwise correlations are lowest (best diversifiers). List in alphabetical order.
9. Sleeve actions: suggest trim/add/rotate based on concentration and diversification findings.

Pair formatting: each pair_id is a 2-element list of index IDs sorted alphabetically.

Return observations: equals the number of monthly return data points (levels - 1).

3C. Active Allocation View Refresh (e.g., Q2 2026 CIO desk)

Goal: Produce active views (UW/N/OW), change vs prior quarter, conviction, and rationale for each requested opportunity set.

Procedure:

1. Read the allocation policy from /api/policies (section allocation_mapping). Extract score thresholds and conviction thresholds.
2. Read prior views from /api/allocation/prior-views. Filter to quarter == target_quarter and previous_quarter == prior_quarter (the prior-views endpoint stores the current assigned view carried forward from the previous quarter). Important: the prior-views endpoint gives you the Q2_2026 assigned views that were determined in Q1_2026 — these ARE the prior views for the current refresh.
3. Read macro signals from /api/macro-signals. Filter to quarter == target_quarter. Each entry gives: opportunity_set, score, rationale_code, drivers.
4. For each requested opportunity set:
   • Look up the macro signal score
   • Determine view: score ≥ OW_min → OW; score ≤ UW_max → UW; else N
   • Determine conviction: |score| ≥ HIGH_abs_min → HIGH; |score| ≥ MEDIUM_abs_min → MEDIUM; else LOW
   • Determine change: compare the current view to the prior view from step 2:
     • Prior N → Current OW: UP
     • Prior UW → Current N: UP
     • Prior OW → Current N: DOWN
     • Prior N → Current UW: DOWN
     • Same view: UNCHANGED
   • Rationale_code: use the rationale_code directly from the macro signal
   • Asset_class: from /api/allocation/opportunity-sets
5. Select risk overlay based on the aggregate direction of views:
   • If HY is UW and duration is OW → DURATION_QUALITY_TILT / tilt_to_duration_quality
   • If HY is severely UW (|score| large) → CREDIT_RISK_REDUCTION / trim_credit_beta
   • If equities are broadly OW → EQUITY_BETA_EXTENSION / add_cyclical_equity_beta
   • If currency signals diverge defensively → CURRENCY_DEFENSIVE_HEDGE / add_currency_hedge
   • If no clear tilt → NO_OVERLAY / hold_policy_weights
   • Rationale_codes: list the most relevant codes in business priority order.

Ordering: allocation_views rows in the order listed by the request payload's focus_opportunity_sets. Risk overlay rationale_codes in business priority order.

Policy ID: use the allocation-mapping policy_id from the policies response (POL_ALLOCATION_MAPPING).

3D. Fixed-Income Risk Rebalance (e.g., PF-FI-LUMEN)

Goal: Propose SELL/BUY rotation that reduces HY and watchlist exposure while keeping duration in band.

Procedure:

1. Fetch current portfolio and enrich with bond/issuer data (same as 3A steps 1-3).
2. Identify watchlist bonds in the portfolio — cross-reference each holding's issuer_id against /api/issuers watchlist flag.
3. Identify HY bonds in the portfolio — bonds with rating_bucket == "HY".
4. Compute pre-trade metrics (same formulas as 3A step 4).
5. Construct rotation:
   • SELL side: prioritize watchlisted HY bonds first, then non-watchlist HY as needed to meet the HY reduction target. Use quantities from the current portfolio, or scale to meet the reduction requirement.
   • BUY side: select candidate IG bonds (not watchlisted, not already held in large concentration) to replace the sold bonds. Match total proceeds approximately (dollar-neutral rotation).
6. Recompute post-trade metrics:
   • post_trade_hy_allocation_pct = 100 × (post-trade HY qty) / total MV
   • post_trade_duration_years = weighted average of post-trade holdings
   • hy_reduction_pct_points = pre_trade HY% − post_trade HY%
   • post_trade_watchlist_exposure_usd_m = sum of remaining watchlist-issuer quantities
7. Check exception flags:
   • hy_cap_pass: post_trade HY% ≤ max_hy_allocation_pct
   • duration_band_pass: post_trade duration ∈ [band_min, band_max]
   • target_hy_reduction_met: hy_reduction_pct_points ≥ target_hy_reduction_pct (from policy)
   • watchlist_exposure_cleared: post_trade_watchlist_exposure_usd_m == 0

Rounding: quantity_usd_m to 1 decimal; risk_metrics percentages and duration to 2 decimals; watchlist exposure to 1 decimal.

Ordering: trades sorted by action (SELL before BUY), then by instrument_id ascending within each action. watchlist_sell_ids sorted ascending.

Risk note code: choose based on the dominant concern addressed:

• hy_cap_pressure if HY was near/above cap
• watchlist_concentration if watchlist positions were material
• duration_preservation if duration was maintained through the rotation
• carry_tradeoff if carry was sacrificed for risk reduction
• no_action if no material risk

3E. Multi-Asset Committee File (e.g., PF-MA-HELIO)

Goal: Combine correlation findings with active allocation views into a single committee decision file.

Procedure:

1. Correlation sub-task: Run the correlation procedure (3B) on the subset of indices specified in the request. Use the "current 12-month monthly-level window" from the policy (review_window_start to review_window_end from POL_CORRELATION_DEFAULT).
2. Identify concentration and diversifier pairs from the correlation matrix:
   • highest_concentration: the pair with the highest correlation
   • best_diversifier: the pair with the lowest correlation
3. Allocation sub-task: Run the allocation view procedure (3C) on the requested opportunity sets. Include prior_view (from prior quarter), signal_score (from macro signals), view, change, conviction, and rationale_code.
4. Sleeve actions: For each sleeve/opportunity set, combine the correlation insight with the allocation view:
   • High correlation + UW view → trim or hedge
   • High correlation + OW view → monitor (concentration concern despite positive view)
   • Low correlation + OW view → add
   • Low correlation + N view → hold
5. Determine rebalance trigger: correlation_cap_breach if any pair exceeds the high threshold; committee_review for scheduled review; hy_cap_pressure or watchlist_concentration if credit risk is the primary concern.
6. Set portfolio_risk_concentration_flag: true if any material concentration exists (e.g., a correlation pair above the high threshold, or a severely UW allocation signal).
7. Set next_step: approve_rotation if clear actions emerge; approve_with_monitoring if views are mixed; defer_pending_risk_review if concerns unresolved; reject_constraint_breach if a hard constraint fails.

Ordering: correlation_summary items by pair_role (highest_concentration first, then best_diversifier). target_sleeve_actions and allocation_views rows in request payload order. Index IDs within each pair sorted alphabetically.

4. Common Formulas

Weighted Average

weighted_avg = Σ(quantity_i × metric_i) / Σ(quantity_i)

Used for: portfolio duration, portfolio YTM. Quantities in USD millions.

High-Yield Allocation Percentage

hy_pct = 100 × Σ(quantity of HY-rated holdings) / Σ(all quantities)

HY-rated means rating_bucket == "HY" from the bond universe.

Issuer Concentration

concentration_pct = 100 × max_issuer_exposure / total_market_value

Where max_issuer_exposure is the largest sum of quantities for a single issuer_id. Check that this ≤ issuer_concentration_limit_pct.

Monthly Simple Return

r_t = (level_t / level_{t-1}) - 1

For t from 1 to N-1 where N is the number of monthly level observations in the window.

Pearson Correlation

ρ = Σ((x_i - x̄)(y_i - ȳ)) / sqrt(Σ(x_i - x̄)² × Σ(y_i - ȳ)²)

Where x_i and y_i are the monthly simple return series for two indices. Round to 3 decimal places. Number of observations = number of return pairs.

Signal Score to View

if score >= 0.35 → "OW"
elif score <= -0.35 → "UW"
else → "N"

Signal Score to Conviction

if abs(score) >= 0.70 → "HIGH"
elif abs(score) >= 0.35 → "MEDIUM"
else → "LOW"

View Change Detection

Compare current view to prior-quarter view:

• N → OW or UW → N or UW → OW (improving direction) → "UP"
• OW → N or N → UW or OW → UW (deteriorating direction) → "DOWN"
• Same view → "UNCHANGED"

5. Output Conventions

Precision (decimal places)

Ordering Rules

Boolean Flags

Use JSON true / false (not strings). All constraint check flags are boolean.

Dates

Format as YYYY-MM-DD. Use the API's as_of_date (not the local request date) when the output asks for the current environment date.

6. Common Pitfalls

1. Stale vs current data: Local payloads (desk requests, memos, "stale" snapshots) may have dates or marks that conflict with the live API. Always prefer the API. The /api/market/energy endpoint includes a stale_data_warning field with the cutoff date.

2. Rating bucket, not rating letter: Use rating_bucket (IG/HY) for HY allocation calculations, not the individual rating letter grade. IG = investment grade, HY = high yield.

3. Watchlist is on the issuer, not the bond: Check watchlist via /api/issuers by matching issuer_id. A bond's recommended_theme_tags may hint at watchlist risk (e.g., "WATCHLIST_RISK") but the authoritative field is the issuer record.

4. Duration band is inclusive: [min, max] from policy means post-trade duration must be ≥ min AND ≤ max.

5. Return observations count: For N monthly index levels, there are N-1 monthly returns. The review window's return_observations is N-1, not N.

6. Simple returns, not log returns: Always use (P_t / P_{t-1}) - 1, never ln(P_t / P_{t-1}).

7. Prior views look-up: The /api/allocation/prior-views endpoint returns entries keyed by quarter (the target quarter) and previous_quarter. For a Q2_2026 refresh, filter to quarter == "Q2_2026" and previous_quarter == "Q1_2026". These entries represent the views assigned in Q1_2026 that are the "prior" for the Q2 refresh.

8. View change direction: UP means the view became more favorable (UW→N→OW direction), DOWN means less favorable. Check the view_rank from the policy: OW=1, N=0, UW=-1. Compare ranks numerically when in doubt.

9. Energy-linked filter: The energy_linked boolean on bonds determines eligibility for energy-credit portfolios. A bond can be in the Utilities or Materials sector but still be energy_linked: true.

10. Subsector vs sector: For diversification checks, use subsector (more granular), not sector. The constraint subsector_min_count_for_diversified counts distinct subsectors.

11. Multi-asset policy composition: POL_MULTI_ASSET_DEFAULT and POL_MULTI_ASSET_RISK compose from sub-policies. Their fields uses_allocation_mapping, uses_correlation_default, uses_credit_default, uses_credit_risk_reduction tell you which sub-policy thresholds to apply. Always check these flags.

12. Risk overlay selection is holistic: Don't pick the overlay from just one signal. Consider the overall pattern: if duration signals are OW while credit/HY signals are UW, the overlay should address credit risk (CREDIT_RISK_REDUCTION) rather than duration (DURATION_QUALITY_TILT). If equity signals dominate, consider EQUITY_BETA_EXTENSION. If no clear tilt, NO_OVERLAY.

13. Candidate flag: Bonds with candidate: false are existing holdings that should not be considered for new purchases (they may be legacy positions). Bonds with candidate: true are available for selection.

14. Correlation pair_id ordering: Always sort the two index IDs alphabetically within a pair. The answer template requires ascending alphabetical order.

15. Even split for ticket packages: When the desk specifies "exactly N tickets totaling $X million, split evenly," each ticket is $X/N million.