Produce committee-grade JSON answers for the Asteria Investment Office across five portfolio archetypes — (a) energy-credit trade packages, (b) international equity correlation reviews, (c) active allocation-view refreshes, (d) fixed-income risk rebalances, and (e) multi-asset committee decisions. Covers the exact endpoints to pull, the deterministic computation recipes (Pearson correlation on monthly simple returns, notional-weighted modified duration and YTM, HY allocation %, carry, concentration), the policy thresholds that drive views/conviction/constraint checks, and the output-schema conventions (field names, enum sets, rounding, id ordering) that the desk's answer templates require. Use whenever a task references an Asteria portfolio (PF-*), index (IDX_*), bond (BND_*), issuer (ISS_*), policy (POL_*), or an answer_template.json.
Produce committee-grade JSON answers for the Asteria Investment Office across five portfolio archetypes — (a) energy-credit trade packages, (b) international equity correlation reviews, (c) active allocation-view refreshes, (d) fixed-income risk rebalances, and (e) multi-asset committee decisions. Covers the exact endpoints to pull, the deterministic computation recipes (Pearson correlation on monthly simple returns, notional-weighted modified duration and YTM, HY allocation %, carry, concentration), the policy thresholds that drive views/conviction/constraint checks, and the output-schema conventions (field names, enum sets, rounding, id ordering) that the desk's answer templates require. Use whenever a task references an Asteria portfolio (PF-*), index (IDX_*), bond (BND_*), issuer (ISS_*), policy (POL_*), or an answer_template.json.
Asteria Investment Office — committee JSON solver
0. Golden rules (apply to every archetype)
The HTTP environment is the book of record. Local payloads
(desk_request, review_request, memo, committee_request, "stale" snapshots,
exception boards, prior-week shortlists) are intake context only. Whenever a
local number disagrees with the environment (holding quantities, marks, HY %,
duration, currency views, index levels), USE THE ENVIRONMENT VALUE. A common
trap: a memo lists a stale holding quantity (e.g. 10.0) while the live
portfolio shows 12.0 — compute with 12.0. When the schema has a
data_precedence field, the value is current_environment_over_stale_payload
in essentially every refresh/rebalance case.
as_of_date = the environment's current as-of date (the as_of_date
returned by /api/policies, /api/portfolios/<id>, /api/market/energy,
etc.), NOT any date written in the local payload.
Follow the answer_template exactly: required keys, enum value sets
(verbatim casing), rounding precision, list lengths, and ordering rules.
Echo fields literally (portfolio_id, task_id, quarters).
required_value
Rounding: round only at the end, to the precision the template declares
per field. Trailing-zero forms are fine (5.8 == 5.80). Compute with full
precision; never round intermediate weighted sums.
Ordering: respect each field's stated order — "ascending by instrument_id",
"alphabetical by index id", "SELL before BUY then id ascending", or "in the
request payload's focus order / item_order". Within a correlation pair, sort
the two ids alphabetically.
Use python (not python3) and curl. Save fetched JSON locally and
compute deterministically; do not eyeball.
View from signal score: OW if score ≥ OW_min (0.35); UW if score ≤ UW_max (-0.35); else N. Boundaries are inclusive of OW/UW.
Conviction from |score|: HIGH if |score| ≥ 0.70; LOW if |score| < 0.35; else MEDIUM.
view_rank: UW=-1, N=0, OW=+1 (used to derive change).
Correlation (POL_CORRELATION_DEFAULT): correlation_high_threshold = 0.8, correlation_low_threshold = 0.2. Review window dates are given by the policy/request (e.g. 2025-05-30 → 2026-04-30).
Credit risk reduction (POL_CREDIT_RISK_REDUCTION): same band/cap but target_hy_reduction_pct = 4.0.
Multi-asset sets compose the above; POL_MULTI_ASSET_RISK escalates to committee on "two_or_more_material_exceptions".
Always take the policy_id from the portfolio's own constraints.policy_id and use that policy's numbers.
3. Core computation recipes
3a. Pearson correlation (archetypes b, e)
For each index pull its monthly level series for the window; sort by date.
Compute simple returns between consecutive levels: r_t = level_t/level_{t-1} − 1.
A 12-month window of levels yields 11 returns → return_observations = 11.
Pearson r over the paired return vectors. Round to 3 decimals.
highest_positive / highest_concentration = pair with the max correlation.
lowest / best_diversifier = pair with the minimum (most negative) correlation.
Each pair_id/pair is the two index ids sorted alphabetically.
3b. Notional-weighted portfolio metrics (archetypes a, d)
Build the post-trade position list (current holdings ± trades), then weight by
quantity_usd_m:
total_market_value_usd_m = Σ quantity (BUY-only packages add to MV; sell-and-fund rebalances keep MV constant when sells == buys).
watchlist exposure = Σ qty of held instruments whose issuer.watchlist is true.
Note HY% and HY-reduction depend only on which HY notional is added/removed —
they are invariant to which IG bond you buy; only duration/YTM depend on buy identity.
3c. Signal → view mapping (archetypes c, e)
For each requested opportunity_set, take the current macro score for the
target quarter, map to view/conviction via §2, set rationale_code to the macro
signal's own rationale_code, set asset_class from the opportunity-set
taxonomy, and set change = sign of (new view_rank − prior view_rank) →
UP/DOWN/UNCHANGED. The prior view comes from /api/allocation/prior-views
(the row whose quarter == target quarter; its recorded view is the standing
prior). Always recompute the view from the live score — do not copy the
recorded prior view as the new view.
4. Archetype playbooks
(a) Energy-credit trade package (e.g. PF-EN-ALTA)
Goal: N BUY tickets of equal notional that raise carry while staying inside
credit constraints and suitable for an income pitch.
Eligibility filter for a candidate bond (ALL must hold):
candidate == true AND energy_linked == true
issuer watchlist == false
not already a current holding (new sleeve money)
single-bond modified_duration_years within the duration band [3.0, 5.0]
— this per-bond duration gate is decisive. A long-dated bond (e.g. dur 5.9)
is ineligible even though the portfolio average might still land in band.
Selection among eligible bonds:
Anchor on the strongest energy signal (typically LNG/gas — check
/api/market/energy; LNG export pull is usually the top score). Pick the
IG LNG/gas name (theme tags like LNG_EXPORTS / GAS_DEMAND).
Pair it with the highest-YTM diversifier that has a different issuer AND
different subsector, to maximize carry while passing diversification.
Verify post-trade: HY% ≤ cap, weighted duration in band, both
selected_issuer_diversification and selected_subsector_diversification
true, watchlist_avoidance true. Set those constraint booleans honestly.
Schema conventions: trade_package length 2, sorted ascending by
instrument_id, each {action:"BUY", instrument_id, notional_usd_m (1 dp)}.
post_trade_metrics (MV 2dp, hy% 2dp, dur 2dp, ytm 2dp). sales_positioning:
target_segment from the client context (e.g. multi_asset_income), theme from
the dominant signal (e.g. lng_export_tailwind). data_precedence =
current_environment_over_stale_payload.
(b) International equity correlation review (e.g. PF-INT-NEXVEN)
Compute the full correlation matrix over the requested index universe (§3a).
extreme_pairs.highest_positive and .lowest per §3a; return_observations
from the window (11 for a 12-level year).
index_set = the universe sorted alphabetically.
concentration:
high_threshold_breached = any pair correlation ≥ 0.8 (true when the cluster
is tightly correlated).
china_asia_dependence_flag = true when China/Asia/EM sleeves are mutually
≥ 0.8 (a dedicated-China + Asia-beta overlap exists).
primary_code = the dominant concentration label. Choose
CHINA_ASIA_DEPENDENCE when the China/Asia block drives it, else
GLOBAL_DEVELOPED_OVERLAP when the top pair is a broad developed/global
overlap (e.g. EM↔World), else NO_MATERIAL_CONCENTRATION. (Keep
china_asia_dependence_flag and primary_code logically consistent with which
pair is actually the maximum.)
diversification_candidates = the low-correlation diversifier sleeve(s) from
the allowed set (the index whose correlations are negative / well below the
0.2 low threshold is the clearest — a LatAm-type sleeve; include an India-type
only if it is a genuine relative diversifier, exclude broad EM-ex-China which
stays ≥0.8).
sleeve_actions (ordered ascending by sleeve): trim the concentrated
sleeve (the dedicated China/over-weighted one) and add the low-correlation
diversifier. Prefer decisive trim/add over monitor/hold for a flagged
concentration.
(c) Active allocation refresh (balanced multi-asset, e.g. Q2 focus list)
For each focus opportunity_set, emit a row {opportunity_set, asset_class, view,
change, conviction, rationale_code} using §3c. Order rows by the request payload's
focus order. Lineage: policy_id = POL_ALLOCATION_MAPPING; quarters echoed
from the request; as_of_date = env as-of.
risk_overlay:
Pick overlay_code/primary_action from the dominant cross-asset tilt implied
by the views: duration support + IG quality with HY underweight →
DURATION_QUALITY_TILT / tilt_to_duration_quality; a credit-led de-risk →
CREDIT_RISK_REDUCTION / trim_credit_beta; strong cyclical equity →
EQUITY_BETA_EXTENSION; defensive FX → CURRENCY_DEFENSIVE_HEDGE; else
NO_OVERLAY.
rationale_codes = the supporting macro rationale codes in business-priority
order (duration/quality support first, then the credit-risk code), highest
priority first.
(The allocation rows are the heavily weighted, deterministic part — get the
score-mapping, conviction boundaries, and change exactly right.)
(d) Fixed-income risk rebalance (sell-and-fund, e.g. PF-FI-LUMEN)
Use POL_CREDIT_RISK_REDUCTION (target HY reduction 4.0pp, cap 20%, band [3,5]).
Recipe:
Identify held HY (rating_bucket==HY) and held watchlist names (issuer.watchlist).
Sells: always sell the held watchlist name(s) to clear watchlist
exposure to 0. Add the minimum additional HY needed to bring post-trade
HY% under the 20% cap (and to meet the ≥4pp reduction). When choosing which
extra HY to sell, prefer the lowest-carry (lowest YTM) HY name so carry is
preserved — do not dump the highest-yielding HY.
Buys: fund with non-watchlist IG candidate bonds named in the desk
shortlist, keeping total buys == total sells so MV/duration are preserved
and no duration shortfall is created. Exclude any shortlist name whose
issuer is on the watchlist (a tempting high-carry watchlist candidate must be
dropped). Favor the cross-desk LNG IG name and a sector diversifier over a
pure long-duration ballast when both are offered.
risk_metrics per §3b. exception_flags: hy_cap_pass (≤20%),
duration_band_pass (in [3,5]), target_hy_reduction_met (≥4pp),
watchlist_exposure_cleared (==0) — all true for a clean rotation.
watchlist_handling.watchlist_sell_ids = held watchlist ids you sold
(ascending). buys_avoid_watchlist = true. risk_note_code = the dominant
pressure (hy_cap_pressure when starting HY% far exceeds the cap).
rotation.trades ordered SELL before BUY, then instrument_id ascending within
each action; quantities 1 dp.
allocation_views per §3c, including prior_view (from prior-views) and
signal_score (the macro score, 3dp). Refresh stale currency views: a stale
note keeping USD overweight is superseded — recompute from the live score
(often lands at N), with change DOWN from the prior OW.
target_sleeve_actions (item order matches the request): map view direction
to action — UW→trim, OW→add (high-conviction OW may hold), and for a
currency sleeve being de-risked use hedge rather than hold.
portfolio_risk_concentration_flag = true when a pair ≥ the 0.8 high
threshold (e.g. China↔EM concentration).
rebalance_trigger = committee_review for a committee decision file
(prefer this over the more specific correlation_cap_breach in committee
framing). next_step = approve_with_monitoring when a concentration flag is
true but the rotation is actionable (escalate/defer only on ≥2 material
exceptions).
5. Pre-submission checklist
Pulled live env data; reconciled every stale local number to it.
as_of_date and policy_id taken from the environment, not the payload.
All enum values match the template's allowed sets verbatim.
