lyndonkl/portfolio-drift-rebalancer

Portfolio Drift Rebalancer

Table of Contents

• Overview
• Input contract
• Workflow
• Tax-efficient rebalancing logic
• Output contract
• Guardrails

Overview

The household typically holds investments across three account types: a taxable brokerage (e.g., Fidelity), a 401k, and an HSA's invested portion. A target allocation is defined at the household level (e.g., 55/20/20/5 US/intl/bonds/cash). This skill rolls up current allocation across all three accounts, computes the drift from target, and proposes specific buys and sells to restore target — preferring trades inside tax-advantaged accounts (401k, HSA) where realized gains have no tax consequence.

It never executes trades. It produces a proposal for the user.

Input contract

The caller provides:

• accounts — array of investment accounts: {id, type, institution, holdings[]} where each holding has {symbol, shares, value_cents, cost_basis_cents, asset_class}.
• target_allocation — { asset_class: target_pct } summing to 1.0.
• drift_threshold_pct — default 5 percentage points (5.0). Below threshold, no rebalance proposed.
• cash_to_deploy_cents — optional new cash being added (e.g., a paycheck deposit).
• tax_rate_long_term — for cost-comparison estimates (default 15%).

Workflow

Drift / Rebalance Progress:
- [ ] Step 1: Aggregate holdings by asset_class across all accounts
- [ ] Step 2: Compute current allocation percentages
- [ ] Step 3: Compute drift vs target
- [ ] Step 4: Check drift threshold; if all classes within threshold, return no-op
- [ ] Step 5: If new cash, allocate it preferentially to under-target classes
- [ ] Step 6: For remaining drift, plan trades
- [ ] Step 7: Order trades by tax cost (advantaged-account first)
- [ ] Step 8: Verify proposal nets to zero cash impact (or matches new-cash deposit)
- [ ] Step 9: Emit proposal with per-trade rationale

Step 1 — Aggregate

For each asset_class in target, sum value_cents across all holdings of all accounts. Include cash positions in the brokerage (typically asset_class cash).

Step 2 — Allocation

current_pct[class] = total_value[class] / total_value_all. Keep two decimal places.

Step 3 — Drift

drift_pp[class] = current_pct[class] − target_pct[class] in percentage points.

Step 4 — Threshold

If max(|drift_pp|) < drift_threshold_pct, emit rebalance_needed: false and stop. Surface drift values for context but propose no trades.

Step 5 — Deploy new cash

If cash_to_deploy_cents > 0, allocate to under-target classes first. Each dollar of new cash to under-target classes reduces the trade volume needed in step 6.

Allocation order: largest negative drift_pp first.

Step 6 — Plan trades

After cash deployment, residual drift calls for trades. For each over-target class, compute excess_cents; for each under-target class, compute shortfall_cents. Match by greedy pairing (largest excess to largest shortfall), producing a list of (sell_class, buy_class, amount_cents) swaps.

Step 7 — Tax-efficient ordering

Within each swap, choose specific holdings:

Sell side priority (most preferred to least):

1. Holdings inside 401k or hsa (any sale is tax-free).
2. Holdings in taxable brokerage with unrealized loss (sale generates harvestable loss — coordinate with tax-loss-harvest-scanner).
3. Holdings in taxable brokerage with smallest unrealized gain (least tax cost).
4. Holdings in taxable brokerage with long-term gains (taxed at 15%).
5. Holdings in taxable brokerage with short-term gains (taxed at marginal rate).

Buy side priority:

1. Holdings inside 401k or hsa of the desired asset class (no transaction-tax friction).
2. Existing holdings in taxable brokerage of the desired asset class (avoid wash-sale by not buying back something just sold).
3. New holding (broad-market index fund preferred).

For each proposed sell from a taxable account, compute and surface the estimated tax cost: (value − cost_basis) × tax_rate_long_term. The user can decide whether the rebalance benefit exceeds the tax cost.

Step 8 — Verify

The proposal must satisfy: Σ buys = Σ sells + cash_to_deploy_cents. If not, the planning has a bug — emit a proposal_invalid warning and stop.

Step 9 — Emit

Per-trade output includes: account, symbol, action (buy/sell), shares, dollar amount, asset_class change, estimated tax cost, rationale (under-target / over-target / new cash deploy).

Tax-efficient rebalancing logic

| Account | Sale tax cost | Notes | |---|---|---| | 401k | $0 | All sales tax-free | | HSA (invested) | $0 | All sales tax-free | | Brokerage (long-term gain) | 15% × gain | Hold > 1 year | | Brokerage (short-term gain) | marginal income tax × gain | Hold < 1 year — usually avoid | | Brokerage (loss) | savings (TLH benefit) | Coordinate with tax-loss-harvest-scanner |

Wash-sale awareness. If a sell is paired with a buy of substantially identical security within 30 days (in any account, including the spouse's), the loss is disallowed. When a rebalance proposes selling an ETF in taxable for a loss, the buy side must use a different but similar ETF (e.g., sell VTI, buy ITOT). Document the wash-sale guard on each trade.

Free-money flags — emit even when no rebalance is needed:

• 401k contribution rate < employer-match cap → free_money_alert: missed_employer_match.
• HSA cash balance > investment threshold but not invested → free_money_alert: idle_hsa_cash.

Output contract

{
  "as_of": "2026-04-25",
  "totals": {
    "portfolio_value_cents": 49250000,
    "by_account": {
      "acc_inv_fid_001": 18750000,
      "acc_401k_001": 22500000,
      "acc_hsa_001": 8000000
    }
  },
  "current_allocation": {
    "us_equity": 0.625,
    "intl_equity": 0.155,
    "us_bond": 0.180,
    "cash": 0.040
  },
  "target_allocation": {
    "us_equity": 0.55,
    "intl_equity": 0.20,
    "us_bond": 0.20,
    "cash": 0.05
  },
  "drift_pp": {
    "us_equity":  7.5,
    "intl_equity": -4.5,
    "us_bond":    -2.0,
    "cash":       -1.0
  },
  "rebalance_needed": true,
  "max_drift_pp": 7.5,
  "trades": [
    {
      "account_id": "acc_401k_001",
      "action": "sell",
      "symbol": "FXAIX",
      "asset_class": "us_equity",
      "value_cents": 1850000,
      "shares": 18.5,
      "tax_cost_cents": 0,
      "rationale": "Reduce US equity overweight using tax-advantaged account"
    },
    {
      "account_id": "acc_401k_001",
      "action": "buy",
      "symbol": "FSPSX",
      "asset_class": "intl_equity",
      "value_cents": 1850000,
      "shares": 41.0,
      "tax_cost_cents": 0,
      "rationale": "Increase intl equity to target via tax-advantaged account"
    }
  ],
  "alerts": [],
  "free_money": [
    {
      "type": "missed_employer_match",
      "account_id": "acc_401k_001",
      "current_contribution_pct": 3,
      "match_cap_pct": 4,
      "annual_dollars_left_cents": 200000,
      "action": "increase 401k contribution from 3% to 4%"
    }
  ]
}

Guardrails

• Never execute. This skill produces proposals only.
• Drift threshold is meaningful. Sub-5pp drift is noise; rebalancing it generates frictional costs (taxes, transaction overhead).
• Tax-cost transparency. Every taxable sale carries an explicit estimated tax cost. The user must see it.
• Wash-sale guard. Pair sell-for-loss with a different (not substantially identical) buy. Document.
• Whole shares only when a holding doesn't support fractional shares — round to nearest whole share and surface the residual cash.
• Don't auto-deploy new cash to over-target classes. Even if it minimizes trades, it widens drift; deploy to under-target instead.
• Account-level constraints respected. A 401k might only offer specific funds; never propose buying a fund the account does not support. The agent layer reads the eligible-fund list per account.